University of Southern California Dissertations and Theses

Essays on sovereign debt

(USC Thesis Other)

Essays on sovereign debt

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON SOVEREIGN DEBT by Timo B. D¨ ahler A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Political Science and International Relations) August 2021 F¨ ur meine Eltern, deren Unterst¨ utzung und Vertrauen Kontinente ¨ uberspannen. Ich hoffe, ich erf¨ ulle euch mit Stolz. ii Acknowledgements While writing a dissertation is for the most part a rewarding and instructive endeavor, disappointing and even demoralizing moments seem to be part and parcel of the writing process. Luckily, I have a number of colleagues, friends, and family members who have supported me wholeheartedly along all steps of the process. They encouraged me to give my best, sometimes actively through asking clarifying questions and providing constructive feedback, sometimes passively through forgiving my erratic mood swings after a failed coding session. My dissertation would not be the same had it not been for them and it is my great pleasure to express my gratitude to them. First and foremost, I extend my gratitude to my dissertation committee: Joshua Aizen- man, Saori Katada, Pablo Barber´ a, and Jeffrey Nugent. I am especially grateful for the intellectual freedom and trust that my committee granted me. The diverse and eye-opening perspectives that the committee members brought to the table only helped to make my dis- sertation better—thank you for your support and guidance with this dissertation, for reading earlier versions of chapters patiently and with great care, and for knowing precisely when to give a gentle steer to stay the course. In particular, I thank Joshua for the invaluable guidance and intellectually fertile en- vironment that he provided me with along all the steps of my academic journey at the University of Southern California. Additionally, as coauthor of one of my chapters, he also directly contributed to a successful peer-reviewed publication—thank you, Joshua. I am thankful to Saori Katada for her in-depth feedback on earlier versions of my dis- sertation chapters through the CIS working paper series and for her continuing support of my research thrust. In particular, my dissertation benefited immensely from her push to make my work accessible to an audience from a variety of different academic fields and to show the political and economic implications of the findings—thank you, Saori. I am grateful to Pablo Barber´ a who offered a welcome, balancing perspective as a iii rigorous computational social scientist. His teaching and his interdisciplinary work helped hone my computational research skills and allowed me to see the impact scholarship can generate outside the academy—thank you, Pablo. Lastly, I would like to extend my gratitude to Jeffrey Nugent. His talent to distill in- sights from different fields and connect them to provide constructive feedback proved in- valuable for my dissertation. Additionally, his strong support of my own research ideas, and confidence in my abilities were benefits not all doctoral students enjoy but should—thank you, Jeff. Beyond the dissertation committee, it is hard to exaggerate just how much I owe to my colleagues in the POIR department at the University of Southern California. They provided me with a home away from home and simultaneously acted as an intellectual incubator for my research ideas. In particular, I want to thank Patrick James who encouraged the thrust of my dissertation and who shaped it in the early stages through numerous rounds of feedback—thank you, Patrick. I am also indebted to my fellow PhD students and particularly my cohort who adopted me readily. Through seminars, pub crawls, and road trips, we built a team spirit and sup- ported each other on both a professional and personal level. Over the years, some of my fellow students turned from nice office colleagues into great friends and there is no doubt that my time in Los Angeles would not nearly have been as memorable without them— thank you, Edward Gonzalez, Shiming Yang, Kyle Rapp, Miriam Barnum, Samuel Bonilla Bogaert, Matthew Nelson, Alix Ziff, Fabi´ an Naranjo Gonz´ alez, and Nicol´ as Albertoni. I would also like to express my deep gratitude to Veri Chavarin who has always had a listen- ing ear and provided help, especially during times when I had doubts about my research. At the University of Southern California, I also benefited greatly from the interaction with colleagues at the economics department whose support is gratefully acknowledged. In particular, I want to thank Tal Roitberg, Grigory Franguridi, Jeong Yoo and Rashad Ahmed. All of them provided invaluable help when I was confronted with econometric difficulties iv and the exchange of ideas across department silos was incredibly useful to sharpen my arguments. Last but not least, I would like to thank my parents Clelia and Manfred and my brother Nias. For more than a quarter of a century, they have provided me with unconditional sup- port, tremendous encouragement, and unwavering trust. I know that my academic journey has come at a significant cost to them all but they never hesitated a moment to help me. Finally, I hope that this achievement will complete the dream that you had for me all those years ago when you chose to give me the best education you could. There are no words to describe how much I owe you. Viele Dank eu allne! v TABLE OF CONTENTS Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xviii Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Research focus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Contribution of the dissertation . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Outline of the dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Chapter 2: Biased Sovereign Ratings: Theory & Evidence . . . . . . . . . . . . 5 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3.1 Why would there be a home bias? . . . . . . . . . . . . . . . . . . 21 vi 2.3.2 Why would there be an in-group bias? . . . . . . . . . . . . . . . . 22 2.3.3 How and why would a home government affect ratings of its do- mestic agencies? . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3.4 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.5 Alternative explanations . . . . . . . . . . . . . . . . . . . . . . . 26 2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.4.1 Dependent variable . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.4.2 Operationalization of dyad-specific explanatory variables . . . . . . 29 2.4.3 Sovereign-specific explanatory variables . . . . . . . . . . . . . . . 30 2.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.5.1 Identification strategy . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.5.2 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.5.3 Exploration of home bias transmission . . . . . . . . . . . . . . . . 39 2.5.4 Exploration of in-group bias transmission . . . . . . . . . . . . . . 42 2.5.5 Beyond average treatment effects . . . . . . . . . . . . . . . . . . . 45 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 2.7 Paper appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Chapter 3: Implications of Biased Sovereign Ratings in the IPE . . . . . . . . . 98 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 3.3 Implications of biased ratings: theory . . . . . . . . . . . . . . . . . . . . 106 3.3.1 A simple model of default probability and capital cost interaction . . 106 vii 3.3.2 An extended model with credit ratings as perceived default proba- bilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 3.3.3 Do sovereign ratings affect perceived default probabilities? . . . . . 111 3.4 Implications of biased ratings: estimates . . . . . . . . . . . . . . . . . . . 114 3.4.1 Are all CRAs relevant for investors’ behavior? . . . . . . . . . . . . 115 3.4.2 What is the economic significance of biased ratings? . . . . . . . . 119 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Chapter 4: Emerging Markets Sovereign CDS Spreads During COVID-19: Eco- nomics versus Epidemiology News . . . . . . . . . . . . . . . . . . . 126 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 4.2 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 4.2.1 Literature on country-specific drivers of sovereign risk pricing . . . 131 4.2.2 Literature on global drivers of sovereign risk pricing . . . . . . . . 132 4.2.3 Literature on time-varying drivers of sovereign risk pricing . . . . . 133 4.3 Stylized facts about emerging markets and COVID-19 . . . . . . . . . . . . 135 4.3.1 Mortality patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 4.3.2 Fiscal responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 4.3.3 Sovereign CDS spread changes . . . . . . . . . . . . . . . . . . . . 137 4.4 Analysis of COVID-19 dominance . . . . . . . . . . . . . . . . . . . . . . 138 4.4.1 Are sovereign CDS spread drivers time-varying? . . . . . . . . . . 138 4.4.2 Which country-specific factors drove CDS spreads during COVID- 19? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 4.4.3 Did country-specific or global factors dominate CDS spreads dur- ing COVID-19? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 viii 4.4.4 Does COVID-19 mortality affect CDS spreads after controlling for immediate economic ramifications of the pandemic? . . . . . . . . 152 4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 4.6 Paper appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Chapter 5: Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 5.1 Research contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 5.2 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 ix LIST OF TABLES 2.1 Home vs. foreign rating examples. . . . . . . . . . . . . . . . . . . . . . . 8 2.2 Third-party rating examples . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3 Example of the construction of the home country dummy. . . . . . . . . . 29 2.4 Coefficients of variables of interest across different samples and estimation types. Each cell refers to a separate regression of the baseline specification Equation 2.4 without variables of interest. All regressions contain agency- fixed effects and time-fixed effects. The dependent variable is a country’s rating on a numeric scale from 1-21. Standard errors are clustered at both the agency-time and the sovereign level. Full sample: 1/1990-12/2019. GFC sample: 9/2008-12/2019. *,**,*** correspond to 10%, 5% and 1% significance, respectively. P-values are displayed in parenthesis. . . . . . . 35 2.5 Results when variables of interest are controlled for simultaneously. Each column refers to a separate regression; full results are in Table 2.19. . . . . 36 2.6 Coefficients of variables of interest across different fixed effect models. The dependent variable is a country’s sovereign rating on a numeric scale from 1-21. Each cell refers to a separate regression. The table displays only the coefficients of the respective variable of interest of each regression. All regressions contain political and economic control variables, time-fixed effects, as well as the fixed effects specified in the head row of the table. Column 1 is identical with column 1 of Table 2.4 to facilitate comparisons. The full sample is used (1990-2019). Standard errors are clustered at both the agency-time and the sovereign level. ***,**,* indicate significance at the 1%, 5% or 10% level. P-values are displayed in parentheses. . . . . . . 38 2.7 Home bias and in-group bias for Chinese agencies. Each column refers to a separate regression. Full results are provided in Table 2.24 . . . . . . . . 43 2.8 Credit rating agency names and their abbreviations used in this paper. . . . 55 2.9 Home countries of credit rating agencies used in this paper. . . . . . . . . . 56 x 2.10 Translation of alphabetic and alphanumeric ratings onto a numeric scale. Source: company websites. . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.11 Difference in the assigned rating to the countries that are the domicile of at least one credit rating agency which rates sovereigns. The biggest deviation between ratings assigned by domestic and foreign agencies are seen in the case of China, Japan, Trinidad and Tobago, and Spain. . . . . . . . . . . . 58 2.12 The descriptive statistics are for the full sample, i.e. for data from 1990 through 2019. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 2.13 Exact variable definitions and data sources. . . . . . . . . . . . . . . . . . 71 2.14 Regression results without controlling for variables of interest. . . . . . . . 77 2.15 OLS results, full sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 2.16 OLS results, GFC sample. . . . . . . . . . . . . . . . . . . . . . . . . . . 79 2.17 Probit results, full sample. . . . . . . . . . . . . . . . . . . . . . . . . . . 80 2.18 Probit results, GFC sample. . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.19 Full results when adding all variables of interest simultaneously. . . . . . . 82 2.20 Coefficients of variables of interest for agency-individual regressions. All regressions contain political and economic variables as well as the vari- ables of interest indicated above. Neither all agencies nor all variables of interest are shown due to data limitations that don’t allow running all agency-specific regressions. . . . . . . . . . . . . . . . . . . . . . . . . . . 83 2.21 Office locations of the credit rating agencies as of 2019-12-31. Source: company websites. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 2.22 Examination of transmission channels for home bias. The sample consists of pooled data from all agencies, 1990-2019. Note that data on offices in rated countries represents information as of 2019-12-31. . . . . . . . . . . . 94 2.23 Further examination of transmission channels for home bias. The sample consists of pooled data from all agencies, 1990-2019. Note that data on offices in rated countries represents information as of 2019-12-31. . . . . . 95 2.24 Home bias and in-group bias for Chinese agencies. Data ranges from 1990- 2019. Standard errors are clustered at both the agency-time and sovereign level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 xi 2.25 Biased sovereign ratings at important thresholds and by quintile. All agen- cies pooled. Full sample; 01/1990-12/2019. GFC sample; 09/2008-12/2019. The dependent variable is a country’s numeric rating. Each cell refers to a separate regression. The table shows only the coefficients on the respective variable of interest for each regression. All regressions contain the con- trol variables from the baseline specification, excluding other variables of interest, as well as time-fixed effects and agency-fixed effects. Standard errors are clustered at both the agency-time and sovereign level. To ac- count for important rating thresholds, the paper calculates the percentiles of the dependent variable (numeric rating) within the regression sample and and selects the closest percentile that is located in between the rating categories around the respective threshold. *,**,*** correspond to 10%, 5% and 1% significance, respectively. P-values are displayed in parenthe- ses. Money market fund threshold; AA+/AA. US pension fund threshold; A/A-. Investment-grade threshold; BBB-/BB+. . . . . . . . . . . . . . . . 97 3.1 Number of rated countries by agency as of 2019-12-31. . . . . . . . . . . . 123 3.2 Certifications of CRAs which rate sovereigns. The asterisk (*) denotes that an agency is ESMA-certified as opposed to EU-based and registered. ESMA information as of 2021-01-04, SEC information as of 2021-01-25. . 124 4.1 List of the thirty emerging markets in the large sample. . . . . . . . . . . . 159 4.2 Weights of countries in the global factor are in brackets. The weights are constructed by dividing a country’s 2019 GDP by all 20 countries’ com- bined 2019 GDP. Luxemburg is not included in the Eurozone as CDS data was unavailable. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 4.3 Classification of large sample into geographic groups. These groups are used to calculate the regional factors in the first-stage regression. Exam- ple for China: Indonesia’s weight is [Indonesia GDP / (GDP of Indonesia, Malaysia, Philippines, Thailand]. . . . . . . . . . . . . . . . . . . . . . . 163 4.4 First-stage regression results of model (1) over the period January 2014 to June 2019. Column 6 provides country-specific out-of-sample correla- tion coefficients between the actual CDS changes and model-implied CDS changes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 4.5 List of 20 emerging markets that constitute our reduced sample. The 20 countries were obtained after deleting 10 countries from the full sample based on low correlation coefficients between actual vs fitted values over the pre COVID-19 out-of-sample period July 2019 to February 2020. . . . . 165 xii 4.6 Analysis of daily peak COVID-19 residuals (March 2020, 20 country sam- ple). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 4.7 Analysis of daily peak COVID-19 CDS spread changes (March 2020, 20 country sample). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 4.8 Analysis of daily CDS spread changes (January 2014 to June 2020, 30 country sample). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 xiii LIST OF FIGURES 2.1 Schematic illustration of the rating process. . . . . . . . . . . . . . . . . . 51 2.2 Variables used by Standard & Poor’s to assess a sovereign issuer. Similar statements are made, if at all, by virtually all other credit rating agencies in the sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.3 Pair-wise correlation coefficients between sovereign ratings from different agencies as of 2019-12-31. . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.4 Pair-wise correlation coefficients between sovereign ratings from different agencies, 1990-2019. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 2.5 Number of rated countries by agency, 1990-2019 . . . . . . . . . . . . . . 61 2.6 Global coverage and rating distribution by ACRA . . . . . . . . . . . . . . 62 2.7 Global coverage and rating distribution by AcraEurope . . . . . . . . . . . 62 2.8 Global coverage and rating distribution by BCRA . . . . . . . . . . . . . . 62 2.9 Global coverage and rating distribution by CariCRIS . . . . . . . . . . . . 63 2.10 Global coverage and rating distribution by Chengxin . . . . . . . . . . . . 63 2.11 Global coverage and rating distribution by CI . . . . . . . . . . . . . . . . 63 2.12 Global coverage and rating distribution by Creditreform . . . . . . . . . . . 64 2.13 Global coverage and rating distribution by Dagong . . . . . . . . . . . . . 64 2.14 Global coverage and rating distribution by DBRS . . . . . . . . . . . . . . 64 2.15 Global coverage and rating distribution by Fareast . . . . . . . . . . . . . . 65 2.16 Global coverage and rating distribution by Fitch . . . . . . . . . . . . . . . 65 xiv 2.17 Global coverage and rating distribution by HR . . . . . . . . . . . . . . . . 65 2.18 Global coverage and rating distribution by JCR . . . . . . . . . . . . . . . 66 2.19 Global coverage and rating distribution by JCREurasia . . . . . . . . . . . 66 2.20 Global coverage and rating distribution by KROLL . . . . . . . . . . . . . 66 2.21 Global coverage and rating distribution by Lianhe . . . . . . . . . . . . . . 67 2.22 Global coverage and rating distribution by Moody’s . . . . . . . . . . . . . 67 2.23 Global coverage and rating distribution by RAEX . . . . . . . . . . . . . . 67 2.24 Global coverage and rating distribution by RAM . . . . . . . . . . . . . . . 68 2.25 Global coverage and rating distribution by R&I . . . . . . . . . . . . . . . 68 2.26 Global coverage and rating distribution by S&P . . . . . . . . . . . . . . . 68 2.27 Global coverage and rating distribution by Scope . . . . . . . . . . . . . . 69 2.28 Global coverage and rating distribution by ShanghaiBrilliance . . . . . . . 69 2.29 Temporal variance of ratings by different agencies for home country Bahrain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 2.30 Temporal variance of ratings by different agencies for home country Brazil 84 2.31 Temporal variance of ratings by different agencies for home country Bul- garia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 2.32 Temporal variance of ratings by different agencies for home country Canada . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 2.33 Temporal variance of ratings by different agencies for home country China 86 2.34 Temporal variance of ratings by different agencies for home country Cyprus 86 2.35 Temporal variance of ratings by different agencies for home country Ger- many . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 2.36 Temporal variance of ratings by different agencies for home country Japan 87 xv 2.37 Temporal variance of ratings by different agencies for home country Malaysia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.38 Temporal variance of ratings by different agencies for home country Mex- ico . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.39 Temporal variance of ratings by different agencies for home country Russia 89 2.40 Temporal variance of ratings by different agencies for home country Slo- vakia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 2.41 Temporal variance of ratings by different agencies for home country Spain 90 2.42 Temporal variance of ratings by different agencies for home country Thai- land . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 2.43 Temporal variance of ratings by different agencies for home country Trinidad and Tobago . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 2.44 Temporal variance of ratings by different agencies for home country Turkey 91 2.45 Temporal variance of ratings by different agencies for home country USA . 92 3.1 Multiple equilibria in the Romer model . . . . . . . . . . . . . . . . . . . . 107 3.2 Debt dynamics in the Romer model . . . . . . . . . . . . . . . . . . . . . . 109 3.3 Number of rated countries by agency, 1990-2019 . . . . . . . . . . . . . . 125 4.1 EM sovereign CDS spreads . . . . . . . . . . . . . . . . . . . . . . . . . . 159 4.2 Emerging market exchange rate dynamics in early 2020 . . . . . . . . . . . 160 4.3 Emerging market COVID-19 related fiscal stimulus . . . . . . . . . . . . . 160 4.4 COVID-19 mortality rate curves for the top and bottom five emerging markets161 4.5 COVID-19 deaths per million residents for the top and bottom five emerg- ing markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 4.6 Change in international reserve holdings of emerging markets between March and April 2020 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 xvi 4.7 CDS spreads of emerging markets between January 2020 and April 2020 . . 162 4.8 CDS spreads of advanced economies between January 2020 and April 2020 163 4.9 Emerging market spread development, July 2019- June 2020 (reduced sam- ple) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 4.10 Actual vs fitted values before COVID-19 (July 2019 to January 2020) . . . 164 4.11 Emerging market average COVID-19 residual (20 country sample). . . . . . 166 4.12 Actual vs fitted values before COVID-19 (July 2019 to January 2020) . . . 166 4.13 Actual vs fitted values before COVID-19 (July 2019 to January 2020) . . . 170 xvii ABBREVIATIONS CDS Credit Default Swap CRA Credit Rating Agency CRAs Credit Rating Agencies CSRC China Securities Regulatory Commission ECB European Central Bank EM Emerging Market EMBI Emerging Marets Bond Index EMU Economic and Monetary Union ESMA European Securities and Markets Authority FCA Financial Conduct Authority Fed Federal Reserve GFC Great Financial Crisis IPE International Political Economy ITC International Trade Center LTFC long-term foreign-currency NDRC National Development and Reform Commission NRSROs Nationally Recognized Statistical Rating Organizations PBoC People’s Bank of China SEC Securities and Exchange Commission SR Sovereign Rating xviii ABSTRACT This dissertation uses empirical methods to examine the link between sovereign credit rat- ings, COVID-19 related variables, and sovereign debt. Chapter 1 provides preliminary background information that puts my research into context. It clarifies the focus of the dissertation, highlights its value and contribution, and outlines the structure of the dissertation. Chapter 2 investigates whether sovereign credit ratings are biased. It explores whether credit rating agencies assign preferential ratings to their respective home country, its geopo- litical in-group, and to countries with strong economic ties to the home country. To this end, the article constructs a panel of monthly data between 1990-2019 from 28 credit rating agencies based in 17 countries for 147 rated sovereigns and provides estimates of the de- terminants of sovereign ratings. The paper shows that the home country gets significantly higher ratings than justified by its economic and political fundamentals, suggesting that the home bias is a common phenomenon among all credit rating agencies under analysis. Likewise, countries with strong economic ties to the home country receive preferential rat- ings from home country agencies. In contrast, there is little evidence that the geopolitical in-group of a home country (measured by agreement scores based on UN generally assem- bly votes) receives substantially higher ratings than country fundamentals would suggest. However, there is evidence that agencies from authoritarian states have a geopolitical in- group bias. Moreover, such agencies seem to exhibit a magnified home bias and it appears that they also weigh the economic interests of their home country relatively stronger than other agencies. These results are robust to the choice of estimation method (OLS, ordered probit, quantile regressions), different sample lengths (recognizing the potential structural break of the Great Financial Crisis), different operationalizations of the core variables of interest, as well as different identification strategies. xix Chapter 3 analyzes the effects of biases in sovereign ratings on the debt dynamics of af- fected governments. First, the article presents a theoretical model that shows how sovereign ratings bear the potential to affect government debt dynamics through the process of self- fulfilling prophecy. In the next step, the article probes whether ratings from all agencies are equally influential and have the capacity to trigger self-fulfilling prophecies. Based on regulatory and practical considerations, the article arrives at the conclusion that only the three US agencies Fitch, Moody’s, and S&P have the capacity to influence interest rates of a large number of countries. Moreover, the article provides examples which suggest that the impact of the home bias in ratings in particular can be substantial and exceed 1% of annual GDP on a recurring basis. Chapter 4 studies whether bad news about COVID-19 induce negative expectations on sovereign credit risks. Specifically, the article investigates the factors driving credit default swap (CDS) spreads of emerging market sovereigns around the outbreak of COVID-19. Using 2014-2019 data, a two-factor model of global and regional risks is first estimated and then extrapolated to obtain model-implied spreads for the period July 2019–June 2020. Intriguingly, the model initially predicts the realized spreads well but loses predictive accu- racy during the COVID-19 pandemic. Fiscal space and oil-revenue dependence primarily drive the differences between the realized and predicted sovereign spreads. An augmented- factor model indicates that the cumulative COVID-19 mortality rate growth is positively associated with CDS spreads. The evidence suggests that the epidemiological deteriora- tion can lower confidence in the sovereign credit markets due to the prospects of prolonged lockdowns and a slower GDP growth recovery. The results also hold for a single regression of daily spread changes during 2014-2020. Chapter 5 presents a condensed review of the central findings of the dissertation. It reflects on the scholarly contributions of the dissertation project and discusses future research av- enues. xx CHAPTER 1 INTRODUCTION “[I]t seems to me that so many writers and teachers in conventional international relations are like the orthodox theologians in Galileo’s time. They are like Flat Earthers who refuse utterly to recognise that the earth is round and revolves around the sun. Similarly, they refuse to see that the relations between states is but one aspect of the international political economy, and that in that international political economy, the producers of wealth—the transnational corporations—play a key role.” Susan Strange, “Big Business and the State” (1991) 2020 made two things crystal clear; the world is globalized and (consequently) there are limits to what government policies can achieve. A globalized world means that a virus can not only spread from one country to the rest of the world within weeks but also wreak havoc in supply chains and grind the world economy to a halt in the process. Limits to government policy manifest in that to stop a virus from spreading rapidly, many governments resort to unsustainable, draconian policies of business closures and social distancing mandates. These policies are not only extraor- dinarily costly but also nearly unenforceable against an increasingly frustrated population. Hence, a vaccine or a very effective therapy are the only ways to get the world back to where it was at the beginning of 2020. However, vaccines and treatments are developed, manufactured, and sold by transnational corporations, which highlights the extent to which entire states depend on the actions of a small number of transnational corporations. 1 As Susan Strange cautioned scholars thirty years ago, not embracing and incorporating transnational corporations into the analysis of the International Political Economy (IPE) may be ill-fated. In fact, we may never get to the bottom of some research puzzles unless we put the study of transnational corporations alongside states at the center of the analysis. While a number of scholars in IPE have made substantial progress in including transna- tional corporations in the analysis of international relations (e.g. Babic et al., 2017; Eden, 1991), the behavior and reach of Credit Rating Agencies (CRAs) as major transnational corporations remains under-explored. In the competitive environment of today’s global- ized economy and particularly in interstate conflict, access to international capital markets is a central determinant of a state’s prosperity, independence, and relative power. As such, a guiding question for this dissertation is to what extent are agency and structure factors of influence that determine or limit governments’ scope for financial decisions? 1.1 Research focus The main goal of this dissertation is to bridge the gap between the theoretical underpinnings of sovereign debt from the economics literature and the practical side of politics. In order to reach this goal, the focus is on the under-explored link between sovereign credit ratings, epidemiological variables, and sovereign debt dynamics. Specifically, this dissertation is comprised of three articles which take different perspectives on this link: • The first article (chapter 2) is an empirical exploration of the determinants of sovereign credit ratings. In particular, the article asks the question whether sovereign credit ratings of rating agencies across the globe show a home bias as well as a geopolitical in-group bias. • The second article (chapter 3) is both theoretical and practical in nature and addresses the question whether the observed biases in sovereign credit ratings are of economic rel- evance to affected countries. More specifically, the chapter outlines the scope conditions under which an agency’s ratings can affect the interest rate a government has to pay in 2 order to access international capital markets. In addition, it identifies which agencies have the capacity to influence the global allocation of capital across sovereigns. • Lastly, the third article (chapter 4) is an empirical exploration of the question whether bad news about COVID-19 can induce negative expectations on sovereign credit risk in emerging markets. In addressing this question, the article explores the effect of a simul- taneous internal and external demand shock caused by a global pandemic on particularly fragile economies. 1.2 Contribution of the dissertation By exploring the link between sovereign credit ratings and sovereign debt dynamics, this dissertation is contributing valuable insights to the scholarly debate on the importance and shortcomings of sovereign ratings in regulating the global financial system. In particular, previous studies have not looked at the global universe of credit rating agencies to explore if there are systematic differences in rating outcomes by agencies from different countries with different systems of government. Equally, in the existing literature there is no thorough exploration in regards to the possibly differential impact severity and reach of sovereign ratings from different agencies. In addition, by exploring the link between epidemiological variables and sovereign risk pricing, the research of this dissertation is somewhat urgent for understanding the economic ramifications of public health crisis caused by COVID-19 and provides policy makers and leaders with valuable insights. 1.3 Outline of the dissertation Structurally, the dissertation is written in such a way that each chapter of the main analysis (chapter 2, chapter 3 & chapter 4) is self-sufficient and can be read and understood without resorting to other chapters’ hypotheses, findings, or other subsections. Each article of the main analysis follows a similar structure. First, there is a general 3 introduction to the topic and the research question or puzzle. Next, a literature review gives an overview of the existing research carried out on a topic and highlights gaps or under-explored areas which the article aims to fill. Then, each article presents its research approach and analysis and concludes by summarizing the findings. Subsequent to the main analysis, chapter 5 presents a condensed review of the central findings of the dissertation. It reflects on the contributions of the research approach taken and discusses future research avenues. 4 CHAPTER 2 BIASED SOVEREIGN RATINGS: THEORY & EVIDENCE Abstract Are sovereign credit ratings biased? This paper explores whether credit rating agencies assign preferential ratings to their respective home country, its in-group, and to countries with strong economic ties to the home country. To this end, the paper constructs a panel of monthly data between 1990-2019 from 28 credit rating agencies based in 17 countries for 147 rated sovereigns and provides estimates of the determinants of sovereign ratings. The data shows that the home country gets significantly higher ratings than justified by its eco- nomic and political fundamentals, suggesting that the home bias is a common phenomenon among all credit rating agencies. Likewise, countries with strong economic ties to the home country receive preferential ratings. In contrast, there is little evidence that the in-group of a home country, as defined by geopolitical alignment, receives substantially higher ratings than country fundamentals would suggest. However, there is evidence that agencies from authoritarian states have a geopolitical in-group bias. Moreover, such agencies seem to ex- hibit a magnified home bias and it appears that they weigh the economic interests of their home country relatively stronger than other agencies. These results are robust to the choice of estimation method (OLS, probit, quantile regressions), different sample lengths, differ- ent operationalizations of the core variables of interest, as well as different identification strategies. Financial support by the Center of International Studies at USC is gratefully acknowledged. 5 2.1 Introduction A Sovereign Rating (SR), often used synonymously with the terms “sovereign credit rating” or “sovereign debt rating,” is an assessment of a country’s creditworthiness. The precision and impartiality of such ratings is crucial as low ratings can have several adverse effects: Firstly, they can affect the credit costs of states (Afonso et al., 2012). 1 Secondly, they can set de-facto ceilings for corporate ratings (Durbin and Ng, 2005; Borensztein et al., 2013), thereby affecting the broad economy of a state. Lastly, they can amplify economic and fiscal crisis in a pro-cyclical way and set in motion self-fulfilling prophecies (Schmukler, 1999). Most sovereign ratings are unsolicited. That is, most sovereigns do not pay for the service of a rating agency to issue a rating. This means that the principal-agent problem inherent in paid-for corporate ratings is mostly absent in sovereign ratings. Hence, in a perfect world, competition and reputation concerns should lead CRAs to publish accurate ratings. However, scholars have criticized CRAs time and time again for incomprehensible assessments, opaque rating methodologies, and poor business integrity. In fact, there is a scientific literature that uncovers serious slips on part of credit rating agencies along several dimensions: • Ferri et al. (1999) find that sovereign ratings were pro-cyclical in the Asian Financial Crisis and show that ratings aggravate economic and financial crises. • Vernazza and Nielsen (2015) find that sovereign rating downgrades were excessive during the Euro crisis and that they provoked factitious doubts about the debt sus- tainability of members of the euro zone. • Cornaggia et al. (2017) find that ratings aren’t comparable across asset classes and show that ratings are generally more favorable for corporate issuers than for sovereign issuers of the same risk category. 1 See chapter 3 for estimates of the effect of biased ratings on government budgets. 6 • Fuchs and Gehring (2017) show that sovereign ratings favor the country where the credit rating agency has its headquarters. Beyond the criticism found in the scientific literature, voices grow louder against credit rating agencies in the political sphere, too. For example, when Standard & Poor’s down- graded the US in 2011, the Obama administration challenged the credibility of the decision, stating that the rating firm relied on faulty math and acted in haste. Across the Atlantic we observe a similar picture: When Moody’s downgraded its assessment on Portugal’s debt payment capacity, then president of the European Commission, Jos´ e Manuel Barroso, crit- icized the firm for being “speculative” and “biased.” Additionally, speaking to reporters, Barroso said the EU planned to strengthen regulations overseeing CRAs and that European legislators would also look into issues of “civil liability” for incorrect judgments by agen- cies on the credit-worthiness of sovereign European nations. Moreover, Barroso indicated that there would be some developments regarding the possibility of CRAs originating in Europe which would reduce Europe’s reliance on US-based credit rating agencies. 2 In summary it can be said that both scholars and politicians question the veracity and purpose of credit ratings. This paper, too, challenges the idea that credit ratings are only based on considerations about creditworthiness. Before introducing the paper’s argument, let’s consider some examples. To begin, let’s look at Table 2.1 which shows the ratings for the US and China for a selection of American and Chinese credit rating agencies. Given that rating agencies generally look at the same data, it is a reasonable prior to assume that ratings from different agencies for a given country should be similar if not identical. And this is indeed the case for the ratings from the American agencies for the US; “AA+”, “AAA”, “Aaa.” However, enter the “BBB+” rating of Chinese agency Dagong. This rating is on average 6.67 notches lower than those provided by American agencies. If we look at the ratings for China, a similar picture emerges. American agencies give China a rating that is on average 4 notches 2 See report by Reuters, https://www.reuters.com/article/eurozone-ratings-barroso-idUSB5E7HL07O 20110706; accessed 2020-10-12. 7 Ratings for the US Dagong (CN) S&P (US) Fitch (US) Moody’s (US) BBB+ (Negative) 1/16/2018 Just 2 notches above junk status AA+ (Stable) 6/10/2013 Dagong +6 notches AAA 4/2/2019 Dagong + 7 notches Aaa (Stable) 4/25/2018 Dagong + 7 notches Ratings for China Dagong (CN) S&P (US) Fitch (US) Moody’s (US) AAA (Stable) 5/26/2017 highest possible rating A+ (Stable) 9/21/2917 Dagong - 4 notches A+ 11/19/2019 Dagong - 4 notches A1 (Stable) 7/4/2919 Dagong - 4 notches Table 2.1: Home vs foreign rating examples. lower than the rating provided by Dagong. As another example, let’s consider Table 2.2 which shows the ratings of the same four agencies for Israel and Japan. Geopolitically, Israel is a long-standing ally of the US and also has a relatively warm relationship with China. The relationship between the US and Japan is equally amicable. In contrast, the relationship between China and Japan has been strained at times. This is mostly because of Japan’s refusal to acknowledge its wartime past to the satisfaction of China as well as the aggressive actions of the People’s Liberation Army towards Japan. Interestingly, the ratings are more similar in the case of Israel— an ally of both China and the US—than in the case of Japan which holds an ambivalent relationship with China and the US. Are the discrepancies discovered in Table 2.1 & Table 2.2 merely an oddity or indeed a hint of a deeper pattern? This paper argues that such discrepancies are the result of the sus- ceptibility of the rating process both to external incentives and internal biases. Specifically, the paper proposes that favorable ratings for the home country can be caused by govern- ment interference as well as internal biases of credit rating agencies. Moreover, the paper puts forward the argument that government interference can also lead sovereign ratings to 8 Ratings for Israel Dagong (CN) S&P (US) Fitch (US) Moody’s (US) A-/Stable 6/22/2018 AA-/Stable 8/3/2018 Dagong + 3 A+ 8/29/2019 Dagong + 2 A1/Positive 7/20/2018 Dagong +2/ superior outlook Ratings for Japan JCR (JP) S&P (US) Fitch (US) Dagong (CN) AAA/Stable 8/9/2018 A+/positive 4/13/2018 JCR -4 notches A 2/2/2020 JCR -5 notches A/negative 2/11/2018 JCR -5 notches / worse outlook Table 2.2: Third-party rating examples. be affected by geopolitical considerations. Terminological accuracy matters when it comes to demonstrating a “bias.” As such, this paper uses the following definitions: • A credit rating agency has a “home bias” if it gives its home country a favorable rating. A favorable rating is defined as a rating that is higher than what would be ex- pected based on a home country’s economic and political fundamentals. The “home country” is defined as the country in which the agency’s headquarters are located. • A credit rating agency has an “in-group bias” if it tends to give higher ratings to member countries of its in-group relative to countries in its out-group. The “in- group” is defined as the countries which are geopolitically aligned with the home country and vice versa for the out-group. The definition for the home bias used in this paper deviates from its definition in previous studies. Particularly in Fuchs and Gehring (2017) a home bias is seen as the residual of the rating that cannot be explained by the political or economic fundamentals of the home country. However, this leaves open the possibility that an agency doesn’t actually have a “bias” and that the home country in fact sets incentives for home agencies to rate it higher. 9 Therefore, the distinction between a “natural” home bias and a bias caused by coercion on behalf of the home government is crucial; The former represents an actual bias, i.e. a favorable rating that is the result of an agency’s internal cognitive bias. The latter represents the result of external incentives by the home government that an agency reacts to. To test the propositions above, this paper investigates the behavior of 28 rating agencies from 17 different countries and uses monthly, dyadic data for up to 147 sovereigns for the period January 1990 to December 2019. The findings of the paper can be summarized in four points: First, there is strong evidence for a home bias in sovereign credit ratings among all agencies. That is, credit rating agencies assign their home country ratings which exceed ratings based on economic and political fundamentals on average by a full rating notch. The home bias does not go away when controlling for information asymmetries that may arise when a domestic agency has to compare its home country to a foreign country with a different language. Similar results are also found when comparing the timing of rating changes; The home agency typically reacts to up- or downgrades of the home country by foreign agencies with a lag rather than initiating the rating adjustment wave proactively. Moreover, there is evidence that the home bias is larger at the upper end of the rating distribution where it reaches as much as 3.6 rating notches. In addition, there is evidence that agencies based in authoritarian countries have a magnified home bias. Second, the paper finds that an in-group bias is not a broad phenomenon. If it exists at all, its economic relevance is relatively negligible. Third, the paper presents evidence suggesting that economic considerations related to the in-group bias have more severe implications for ratings than a potential in-group bias has on its own. Specifically, agencies assign higher ratings to countries which are major export markets for domestic companies, and to a lesser degree to countries which are ma- jor lending markets for domestic banks. Similarly, agencies assign ratings to countries to which the home country extends a swap line that are on average almost one notch above 10 what would be expected based on economic and political fundamentals of the rated country. Fourth, the paper revisits the in-group bias by taking the economic and political system of the home country into consideration. Doing so, the paper provides evidence that agencies in authoritarian regimes assign systematically higher ratings to geopolitically aligned coun- tries when compared to agencies from economically libertarian countries. In other words, authoritarian intervention can cause or magnify an in-group bias in sovereign ratings. Like- wise, the paper presents data indicating that the home bias is magnified for agencies based in authoritarian countries. Moreover, there is evidence suggesting that agencies from au- thoritarian countries have a stronger tendency to assign ratings in line with the economic interests of the home country. Importantly, the paper takes several steps to move as close as possible to a causal inter- pretation of the findings. First, the paper tests a variety of alternative mechanisms that, if not accounted for, could render the discovered biases endogenous. Second, the paper im- plements a variety of unit fixed effects to account for differences in the rating scales across agencies. Equally, the paper uses time fixed effects to rule out that the observed biases are driven by time-invariant factors such as the membership in a monetary union or benefits enjoyed by providing a major reserve currency. The paper also uses quantile regressions to better understand the heterogeneity in treatment effects and to asses if the home bias and in-group bias also affect countries at important rating thresholds. Additionally, the paper uses agency-specific regressions to allow for differential weights and assessments across agencies such as to test if the proposed mechanisms hold across agencies. Lastly, the pa- per analyses different sample time frames and different choices of estimation methods to account for the discrete nature of the dependent variable. The rest of the paper is structured as follows: Section 2.2 discusses the existing lit- erature. Section 2.3 derives a theory that predicts a home and in-group bias in sovereign ratings. Section 2.4 discusses the data and the agencies examined in this paper. Section 2.5 contains the main empirical analysis and provides a number of robustness checks. Sec- 11 tion 2.6 concludes. 2.2 Literature review Scholarship on home biases has a long history in the economics and finance literature. In- deed, there are several papers that argue for the home bias—broadly defined—as a peren- nial feature of international transactions: For example, McCallum (1995) argues for a home bias in international trade and shows that for the United States and Canada, inter-province trade is 20 times larger than international trade, holding other determinants of trade fixed. Wolf (2000) documents the same phenomenon for OECD countries. French and Poterba (1991) argue for a home bias in international finance and show that individuals and insti- tutions in most countries hold only modest amounts of foreign equity despite the empirical observation that returns on national equity portfolios suggest substantial benefits from in- ternational diversification. 3 Tesar and Werner (1995) corroborate these findings, adding that the high volume of cross-border capital flows and the high turnover rate on foreign equity investments relative to turnover on domestic equity markets suggests that variable transactions costs are an unlikely explanation for this home bias. Similar to the vast scholarship on home biases in international economics and finance, the literature on the determinants of corporate, municipal, and sovereign credit ratings is comprehensive and a detailed discussion of it lies beyond the scope of this paper. Instead, this paper focuses on previous studies dealing with the determinants of sovereign ratings in general, and studies dealing with home biases in ratings in particular. The idea of using a structural model to establish if predicted and actual ratings converge is first developed in Cantor, Packer, et al. (1994). They model ratings as r i;t =f(e i;t ) (2.1) 3 Cf. Feldstein and Horioka (1979) for the origin of this idea. 12 wherer i;t is the rating for countryi at timet ande represents its economic fundamentals. Based on this model and statements of major CRAs, they identify eight quantitative criteria as the determinants of sovereign ratings: per capita income, GDP growth, inflation, fiscal balance, external balance, external debt, economic development, and default history. In their econometric analysis, most of these variables are closely related to the actual ratings assigned and the predictive ability of the variables combined reached more than 90% with a residual standard error of about 1.2 rating notches. Ferri et al. (1999) use this method in a dynamic context to show that sovereign ratings can be pro-cyclical. That is, there are periods during which CRAs are less systematic in their assessments and weigh discretionary data more heavily, thereby having the potential to amplify economic fluctuations. Afonso (2003) extends the method of Cantor, Packer, et al. (1994) and uses linear, logistic, and exponential transformations of the rating scales and other robustness checks. Of the large number of variables that agencies claim to look at, they find that six variables appear to be the most relevant in determining a country’s rat- ing: income per capita, external debt, level of economic development, default history, real growth rate, and inflation rate. Interestingly, while the incentives for a sovereign to default may differ from those of other obligors and may be affected by domestic political factors (Tomz and Wright, 2007; Panizza et al., 2009), Archer et al. (2007) find that the regime type and most other political factors have little effect on ratings. Instead, they confirm that trade, inflation, growth, and default history strongly affect ratings. However, while their results show that most political factors are not significant determinants for ratings, their interviews with CRAs would suggest that political factors are considered when assessing sovereigns. Importantly, scholarship in this fashion does not look at differences in ratings across agencies and instead focuses on explaining ratings based on characteristics of the rated entities. The first to break with this tradition and to include characteristics of the rating agency in the analysis of rating determinants are Ammer and Packer (2000). They examine differ- 13 ences in default rates by sector and debtor domicile and provide evidence that CRAs are imperfectly calibrated across issuer sectors. They do not find significant differences in de- fault rates between US and foreign firms even though foreign firms are rated significantly lower. Similarly, Shin and Moore (2003) compare ratings assigned to Japanese corpora- tions by two US agencies and two Japanese agencies and find that US agencies assign systematically lower ratings than their Japanese counterparts. Transferring the idea of Shin and Moore (2003) over to sovereign ratings, Fuchs and Gehring (2017) (henceforth FG) are the first to examine the role of an agency’s home country on rating outcomes. They do so by first discussing what they call the “demand” and “supply” channels that could influence ratings. Then they empirically examine ratings with a model of the form r a;j;i;t =f(x j;i;t ;e i;t ;p i;t ) (2.2) wherer a;j;i;t is the rating by agencya in home countryj for countryi at timet,e andp are economic and political fundamentals of the rated country andx j;i;t are home-sovereign pair specific variables. They find that agencies assign relatively higher ratings to 1) their home country, 2) culturally more similar countries, and 3) countries to which home-country banks have larger risk exposure but not to countries to which the home country has geopolitical ties. To ascertain why culture matters, they explore in greater detail whether information asymmetries or differences in risk perception contribute to cultural proximity’s effect and find that cultural proximity is associated with a more optimistic perception of risk rather than information asymmetries. To asses whether geopolitical ties influence ratings, they use two measures. First, they use data on bilateral voting alignment in the United Nations General Assembly as a measure for geopolitical alignment between the home country of the rating agency and the rated country. Second, they employ a country’s share of total US military aid and take it as a proxy for the strategic importance that the United States assigns to countries. Crucially, FG fall short of taking the home country seriously and do not consider that 14 different home countries vary along a number of dimensions. This means that the “home bias” that they find is the mere consequence of an agency’s location. They don’t consid- ered the possibility that different home sovereigns have different regime types (democratic, authoritarian, etc.) and varying economic systems (capitalist, socialist, etc.). As such, differences in government intervention across different home countries is not considered. A second weak spot in FG is that they criticize that “most of the literature analyzes only sovereign ratings issued by the big three U.S.-based agencies” but fail to live up to their own criticism by only only examining nine agencies from six countries. This may be prob- lematic as it is plausible that by implicitly selecting agencies based on their mainstream name recognition their results may suffer from a selection bias. As a consequence of ex- cluding most agencies outside OECD and NATO member countries, it doesn’t surprise that they don’t find geopolitical effects on ratings. Therefore, verifying the robustness of the insignificant geopolitical effects on ratings reported by FG is crucial to gain a more comprehensive view of the international political economy of credit ratings. Indeed, from the examples in Table 2.2 it appears that conclusions that disagree with FG’s findings about geopolitical biases may be drawn if the analysis is extended to take into account the political regime under which agencies operate and by extending the number of agencies considered in the analysis. Indirect support for the conjecture that an in-group or geopolitical bias should be found in ratings also comes from a specific reading of the findings in Reinhart et al. (2016). There, the authors discover the curious case of the missing defaults. That is, in spite of the drying up of global capital flows and a sharp fall in commodity prices from 2012 to 2016, sovereign defaults particularly in emerging market economies in this period did not spike higher as the record of the prior two hundred years would predict. In a follow-up paper, Reinhart (2019) argues that China may be responsible for these missing defaults in two important respects. For one, the larger global footprint of the Chinese economy, which is growing steadily at a rapid rate, stabilized global trade and hence the revenue streams of many countries. For another, Chinese ambition to become a global power led to large offi- 15 cial flows to a variety of countries. 4 As a comprehensive new data set by Horn et al. (2019) shows, China has extended many more official loans mostly to developing countries than previously known. This systematic under-reporting of Chinese loans has created a “hidden debt” situation, meaning that international institutions such as the IMF and World Bank have until recently had an incomplete picture of how much countries around the world owe to China and under which terms and conditions. However, now that the data are available, controlling for hidden debt as part of the debt/GDP ratio is important if we want to estab- lish potential biases of sovereign ratings. Additionally, the dyadic nature of the data set of hidden Chinese loans may provide leverage when it comes to identifying in-group biases as the amounts lent may not only affect the actual creditworthiness of countries but also indicate the strength of a dyadic relationship. Similar to the emergence of China as a global lender, China’s network of bilateral swap lines has gained momentum in recent years, along with swap lines between other country pairs more generally. 5 This is mainly due to the turbulence’s emanating from the Great Financial Crisis and the Euro Crisis, which led the Fed to extend swap lines in order to keep global financial markets working smoothly (cf. Obstfeld et al., 2009; Aizenman and Pasricha, 2010; Chey, 2013; Baker, 2013). In contrast, the global reach of Chinese swap lines and the speed with which this network has been expanded raises questions about the motives behind it. In fact, between 2009 and 2018 the People’s Bank of China has created 38 swap lines—more than double the number that are currently offered by the Fed—with 4 Over the past two decades, China has become a major global lender, with outstanding claims now ex- ceeding more than 5% of global GDP. Almost all of this lending is official. That is, it is coming from the gov- ernment and state-controlled entities. See report by Harvard Business Review, https://hbr.org/2020/02/how- much-money-does-the-world-owe-china; accessed 2020-10-10. 5 A (currency) swap line is an agreement between two central banks to exchange currencies. This allows a central bank to obtain foreign currency liquidity from the central bank that issues it if a need arises. Such a need may arise when a central bank should provide foreign currency to its domestic commercial banks to maintain stability in the financial sector. One example for a swap line is the agreement between the Fed and the ECB that enables the ECB and all the national central banks in the Eurosystem to receive US dollars from the Fed in exchange for an equivalent amount of euro provided to the Federal Reserve. While this has primarily positive effects for liquidity of banks in the euro zone and thus for financial stability in Europe, US banks also benefit indirectly: Because of the role that the US dollar plays in global financial markets, strains in dollar funding markets overseas can disrupt financial conditions in the United States so that the US Fed benefits from acting as a “global lender of last resort.” 16 a total volume above US$ 500 billion in Chinese renminbi. This includes agreements with countries seemingly unimportant for the financial stability in China such as Belarus, Iceland, Suriname, and others. Thus, whereas swap lines have historically been signed to safeguard against liquidity crunches, Beijing may have approached the tool with a different objective in mind, namely the internationalization of its currency and the de-dollarization of international trade. Bahaj and Reis (2020b) find that this strategy is reaching its objective so far, albeit only to a relatively small degree; they show that swap line agreements signed by the People’s Bank of China increase the likelihood of renminbi usage in the following months, signalling the effectiveness of this policy, but the increases are relatively small. Crucial to understanding the theory developed in the next section is the idea that there are two ways to think about the interplay between swap lines and sovereign debt: On the one hand, there’s the “orthodox” way of thinking about swap lines as resem- bling unsecured sovereign debt. This view implies a two-way relationship: 1) As unsecured sovereign debt, swap lines may be affected by the same considerations that affect regular sovereign borrowing. That is, factors that explain better access to sovereign borrowing— volatility, a history of low incidence of default, growth, etc.—may also explain if a country has access to a swap line. In fact, Aizenman, Jinjarak, et al. (2011) suggest that better insti- tutional quality means lower moral hazard which should imply more elastic access to larger swap lines and provide evidence that this is the case for the swap lines between the OECD countries. In other words, access to a swap line reflects a strong perceived creditworthiness, indirectly suggesting a strong credit rating. 2) In contrast, access to a swap line can also affect the creditworthiness of a country. This is particularly the case for those countries prone to be caught in a diabolic loop. 6 A diabolic loop is caused by a concentration of gov- ernment bonds held by domestic banks and government guarantees given to these banks. To understand this line of reasoning, imagine that, because of a liquidity crisis, investors raise their perceived default risk on government bonds. An increase in the interest rate of 6 See Brunnermeier and Reis (2019) who coined the phenomenon of the diabolic loop discussed here. 17 new bonds implies that older bonds held by banks are now worth less. This loss is signifi- cant and leads to cuts in lending. Less lending lowers economic activity, which lowers tax revenues and raises spending through the automatic stabilizers so that the government’s fi- nances deteriorate. At the same time, with the drop in the banks’ equity, the likelihood that the government guarantees will be triggered rises. This extra spending also worsens the fiscal balance. Both combined, the public finances become worse, which puts additional strains on the sustainability of government debt. Their prices fall further, triggering a new turn in the diabolic loop. Swap lines can alleviate the diabolic loop, at least to the extent that the government borrowed in foreign currency and domestic banks hold such foreign currency government bonds so that a swap line indirectly supports the government’s solvency. Taken together, swap lines can impact the creditworthiness of a government so that there’s a reasonable case to be made that swap lines should be controlled for when investi- gating sovereign credit ratings. On the other hand, there’s a more “heterodox” way of thinking about swap lines that suggests that extending a swap line to a country is a “signal” of goodwill and thus a means to gain diplomatic influence. This way of thinking about swap lines seems particularly relevant for the swap lines that China extends. In reality, there are only a few isolated cases in which swap lines have been drawn upon (examples include Russia, Turkey, Pakistan and Argentina) and the extension of swap lines has only increased the usage of renminbi in international trade to a small degree. However, the swap lines extended by China certainly boost China’s diplomatic influence as a short-term bank of intermediate liquidity when Western institutions are unavailable, signalling its goodwill to other countries. In fact, the signalling power of Chinese swap lines may be its key defining feature as the case of Argentina illustrates. When Argentina was faced with high inflation and was on the brink of yet another economic crisis in 2014, it was unable to obtain US dollars, which inhibited the country from importing vital consumer goods. In this situation, Argentina drew on its 18 swap line with China to obtain renminbi. However, rather than facilitating trade between Argentina and China only, the borrowed yuan were exchanged for US dollars which were then introduced into Argentina’s economy. A similar dynamic played out the year before when Pakistan tapped into its swap line with China and used it to obtain US dollars to bolster its falling international reserves. Importantly, in both cases China did not protest the use of renminbi as an intermediate currency and instead emphasized how swap lines could further international trade. The vastness of China’s swap line network in conjunction with how swap lines have been used so far makes a case for seeing swap lines as a signal of goodwill and a means of diplomatic influence. Thus, swap lines should not only be controlled for when investigating sovereign credit ratings on the grounds of their impact on government solvency but also because swap lines may be used as a geopolitical tool. For all the increased attention scholars and policy makers have paid to credit ratings and particularly to potential biases in recent years, systematic research on the role of hidden debt and swap lines on sovereign ratings is still lacking, as is the verification of the results by Fuchs and Gehring (2017). The present paper fills this gap in the literature. Specifically, it derives an explicit theory to show how a home bias as well as an in-group bias may form and defines scope conditions to test the theory empirically. By controlling for hidden debt and swap lines, the paper also tests the robustness of the results in FG. That China will define the 21st century is almost a foregone conclusion. Therefore, it behooves scholars to seek a better understanding of China and its rating agencies. As a consequence, the present paper puts emphasis on pondering the specifics of Chinese rating agencies and how China as an authoritarian regime may interfere with its own rating agencies. The paper’s primary contribution is to take the home country seriously in examining sovereign ratings. In doing so, the paper adds to the knowledge of biases and government interference in sovereign ratings. Additionally, it adds to the young and quickly growing literature on China’s role in international finance (cf. Reinhart, 2019) as well as to the liter- ature on familiarity and cultural biases in economic decision-making (cf. Huberman, 2001; 19 Guiso et al., 2006). Moreover, by providing a coherent theory to explain the home and in- group bias, the paper formulates testable hypotheses. Lastly, by assembling a new data set of ratings from 28 agencies from around the world, the paper is the first to systematically test the home and in-group bias hypotheses for smaller, less known credit rating agencies. 2.3 Theory Why would sovereign ratings deviate from objective, fair values? The answer, this paper argues, lies in the susceptibility of the rating process to external pressures as well as internal biases. To comprehend this argument, let’s look at Figure 2.1 which shows that a sovereign rating is generated in a four-step process. In step 1), a government either solicits a rating from an agency or an agency initiates a rating on its own, which is more common. 7 A team is assigned to rate the sovereign and data about the country is gathered. In step 2), meetings with representatives of the government may take place to gather additional data so that the team can begin the weighting and interpretation of the data. Importantly, this step of the rating process involves the analysis of both quantitative and qualitative data. Therefore, agencies typically use a two-stage procedure to account for the diversity of data types. In the first stage, an econometric model is used to generate a rating score based on quantitative criteria. In the second stage, the computed rating score is augmented by a qualitative overlay adjustment framework to generate the final rating. 8 In step 3), a report is drafted and a rating is proposed by the sovereign-specific team to an agency’s rating committee. The committee either approves the proposed report/rating or asks for amendments before notifying the issuer of the rating. Step 4) consists of publishing the report/rating and monitoring the political and economic development of the country. If the circumstances require it, a sovereign is reassessed and possible changes, i.e. upgrades or downgrades, are published. 7 Data on whether a rating was solicited or not is not available across agencies. Hence, the analysis in this paper cannot control for any effects that the solicitation may have on the rating. 8 Compare for example to the sovereign rating tool by Fitch at https://www.fitchratings.com/sovereigns/ interactive-sovereign-rating-model 20 What variables do agencies claim to use in the rating process, particularly in step two? Generally, agencies state that their sovereign rating criteria incorporate factors that they believe to affect a sovereign’s willingness and ability to service its financial obligations. Consequently, their ratings rest on five pillars: a sovereign’s institutional, economic, exter- nal, fiscal, and monetary standing. 9 If we consider the institutional pillar as as being made up of political variables and the other four as economic ones, agencies essentially claim to use a model that can be expressed as r i;t =f(e i;t ;p i;t ) (2.3) wherer i;t is the rating assign to countryi at timet ande i;t andp i;t are its economic and political fundamentals at that time. With this knowledge on the rating process, let’s turn to the theory that explains how biases in ratings can come about. 2.3.1 Why would there be a home bias? The essence of a credit rating agency as a company is that it is ultimately a collection of people making collective decisions. Hence, cognitive biases of individual analysts aggre- gate and translate into a bias of the group, i.e. the agency. Thus, arguing that agencies have a home bias in their risk assessment is tantamount to arguing that individuals have such a home bias. However, that the latter is the case has already been established in a number of studies: For example, work by French and Poterba (1991) shows that people have more optimistic expectations about domestic stock returns compared to foreign stocks by show- ing that people do not diversify their portfolios enough. Additionally, Shiller et al. (1991) provide direct evidence for a home bias of people, i.e. evidence that shows a more opti- mistic risk assessment of domestic assets than foreign assets; In early 1990, they surveyed professional portfolio managers in Japan and the United States. US portfolio managers expected an average return of -0.3% on the Dow Jones Industrial Average over the next 9 Compare to Figure 2.2. 21 twelve months, compared with an expected return of -9.1% on the Nikkei. In contrast, Japanese portfolio managers expected an average return of 12.6% on the Dow, and 10.8% on the Nikkei. Hence, while the Japanese portfolio managers were more optimistic than their US counterparts with respect to both markets, they were relatively more optimistic about the Tokyo market and vice versa for the US portfolio managers. Moreover, Huber- man (2001) claims that “familiarity is associated with a general sense of comfort with the known and discomfort with—even distaste for and fear of—the alien and distant.” and provides compelling evidence that people have a strong preference for the familiar in many economic settings. To summarize, given the home bias of people in other economic set- tings, it seems plausible that a home bias in ratings is due to the human default position of having a preference for the “known.” One implication of this theory is that no interference on behalf of the home government is needed for the home bias to emerge. Instead, it is deeply ingrained in the internal mecha- nisms of CRAs so that we can speak of a “true” home bias. This is not to say, however, that government interference cannot take place. Another implication is that there is no reason to ex ante expect substantial differences in the magnitude of the home bias across agencies. 2.3.2 Why would there be an in-group bias? Beyond stating that agencies “have incentives to draft ratings in a way that caters for their respective home country’s economic and geopolitical interests” FG do not explain how a geopolitical bias would emerge. What’s more, speaking of a “bias” in this context is mis- leading as they explain its rise through incentives rather than agency-internal factors. Given their moot definition and the small number of agencies they examine, it is not surprising that their conclusion is that “[g]eopolitical ties do not seem to play a decisive role.” Rather than dismissing the argument by FG altogether, the present paper argues that they were on to something and their null result may be caused by a lack of clarity in their argumentation. In fact, logical consistency demands that if a home bias is theoretically tenable in the first 22 place, then an in-group bias must be equally tenable. This is because a home bias can be seen as the manifestation of the difference in preference for the most extreme underlying reference pair. 10 Hence, if this logic is accepted, then the idea of an in-group bias must be accepted. So, the next question becomes: what is the in-group and what is the out-group of a given country? In principle, there is little that speaks against the idea that countries’ in- and out-groups are primarily defined by their bilateral geopolitical alignment. Problem- atically, however, is that there is little evidence that suggests that geopolitical alignment is the primary feature by which people of different countries define each other. Rather, it seems more plausible that mutual trust between average people of two countries is a clearer marker of in-group status and the absence of trust in turn is a clearer marker of out-group status. However, data on mutual trust is not available on a global scale whereas different measures about mutual geopolitical alignment are available on a consistent basis across time and countries. To summarize, it is plausible that the same cognitive bias that creates the home bias also creates an in-group/out-group bias among credit rating agencies. In terms of opera- tionalization, it appears that geopolitical alignment of countries is the best measure for in- group/out-group status, particularly because of its broad and consistent availability across time and countries. However, any measure of geopolitical alignment catches only one aspect of the in-group/out-group dynamic so that the effect of geopolitical alignment on sovereign ratings should be relatively minor. A crucial implication of this theory on the in-group bias is that no interference on be- half of the home government is needed for an agency to assign relatively higher ratings to geopolitically aligned countries. Hence, we can speak of a “true” in-group bias rather than a residual that is likely caused by external incentives such as the ones outlined in FG. Note that this theory does not argue that a government’s geo-strategic goals and prefer- ences do not have a role to play in ratings. Indeed, those may play a role and governments 10 To put it differently, the home bias shows the size of the “us versus them” effect whereas the in-group bias shows the size of the effect “which of them do we like more?.” 23 may try to interfere with the ratings of their domestic agencies to reach goals of geopolitical importance. Hence, the next section discusses how certain governments could “magnify” both the home and in-group bias. 2.3.3 How and why would a home government affect ratings of its domestic agen- cies? The preceding two sections outline how a favorable treatment of the home country and of countries which are geopolitically aligned with it can emerge from within agencies; no in- terference by the home government was necessary to reach this conclusion. Nevertheless, a claim repeatedly made by commentators and news stations is that home governments actively attempt to influence ratings. Essentially, this argument claims that it is not the internal bias of rating agencies but the pressure that home governments put on them that leads to a preferential treatment of the home country and its allies. As is often the case with the chattering buzz of attention-seeking journalism, however, the truth takes a back seat. Indeed, considering that rating agencies are private companies like countless others, there is little reason to assume that governments treat domestic credit rating agencies differently than other important companies. If government interference in credit rating agencies truly existed, it would most likely have been documented with hard evidence by investigative journalists or scholars. This has not been the case so far. That said, it is possible that par- ticular governments may try to intervene and “dictate” what ratings their domestic agencies should assign and to who. Specifically, it seems rational that authoritarian regimes aim to obtain influence in domestic credit rating agencies. In fact, for an authoritarian regime to have a grip on credit rating agencies is tantamount to having a propaganda tool. For ex- ample, an authoritarian regime could indirectly “assign” itself a high rating to demonstrate domestic and foreign audiences how fine a job it supposedly does at managing its finances. As another example, to gain influence abroad an authoritarian regime might also be in- clined to have its agencies assign favorable ratings to geopolitically important countries, 24 thereby weaponizing sovereign ratings. Importantly, it is hard for CRAs in authoritarian states not to follow the government’s nudge as ignoring it could lead to the government taking over the agency or to the agency’s loss of accreditation. Without accreditation, the agency can no longer rate corporations, sovereigns, or financial products and will thus go bankrupt before long. In summary, it is clear that the home country’s economic and political system needs to be taken seriously in research on credit ratings and their determinants. Equally, it seems clear that credit rating agencies as private companies are no more and no less susceptible to government interference than other private companies, at least in western-style democ- racies. However, things may be different when we consider credit rating agencies in au- thoritarian states. There, it seems plausible that a government has both an interest and the capacity to influence the ratings process of its domestic agencies. One implication of this influence is that it seems likely that an authoritarian regime asks for a favorable rating for itself. A second implication is that it seems probable that domestic agencies in authoritar- ian states are weaponized so that they give geopolitical allies of its home country favorable ratings and vice versa for adversaries. 2.3.4 Hypotheses The theory outlined above can be summarized in four falsifiable hypotheses: 1. As a natural default, agencies tend to have a home bias. 2. Authoritarian regimes put pressure on home agencies to achieve higher ratings. This results in a “magnified” home bias, i.e. a preferential rating that has “political” com- ponent. 3. An geopolitical in-group bias is the exception rather than the rule. 4. Authoritarian regimes use rating agencies as foreign policy tools so that an in-group bias should be observed for agencies under authoritarian rule. 25 2.3.5 Alternative explanations There are a number of alternative explanations why the home country and countries which are geopolitically aligned with it could receive higher ratings by domestic agencies that have neither to do with internal biases of agencies nor with interference by home gov- ernments. Not considering such alternatives would be negligent and could render results endogenous. Thus, this paper tests a variety of alternative channels that could be related to the home and in-group bias. 11 Alternative explanations for the home bias First of all, credit rating agencies are deeply intertwined with their home market and pre- sumably generate a large share of their profits from it. Hence, anything that could lead to a deterioration of the domestic economic climate is against the interest of domestic CRAs. Consequently, analysts may treat the home country with greater care than other countries so that a preferential rating for the home country could emerge independent of an actual home bias. Unfortunately, this mechanisms cannot be tested directly because there is no data across agencies on the business share in their home country. However, we can test this mechanism indirectly for the in-group bias through home country exports and home bank exposure as shown later. Secondly, it may be that analysts have better information about the home country com- pared to foreign countries because of language differences. This has two possible implica- tions. On the one hand, inferior information about foreign sovereigns could lead to less pre- cise ratings. On the other hand, inferior information about foreign sovereigns could system- atically lead to lower ratings. A theoretical explanation for this can be gained by adopting a similar argument given in Gehrig, 1993: Assume that a rating agency estimates the liquidity L of a sovereigni withE[L i ]sN(; i 2 ). Further, assume that a sovereign enters a stage 11 The alternative explanations outlined herein draw on FG who speak of “supply-” and “demand” side channels. 26 of default ifL i <z. Thus, the probability of default isP (L i <z) = ( jzj i ) =F ( jzj i ). Now, assume that two sovereignsA andB posses identical economic and political funda- mentals. That is, they have the same, but sovereignB is culturally more distant to the rating agency’s home country because of more severe language differences. As a result of higher information transmission costs, the agency collects less information about the characteristics of distant sovereignB. This has the implication that the prediction ofL B , i.e. the liquidity of the sovereign, is less precisely known compared to that of countryA. Mathematically, this translates into A 2 < B 2 , which implies thatF A ( jzj A )<F B ( jzj B ) for all z < . All this says is that the predicted default probably is seen as higher for the culturally more distant sovereign which could translate into a lower rating than would be justified in reality. That is, the home bias could result from insufficient information transmission between countries of dissimilar languages. To test this mechanism, this paper controls for differences in language between the home country-rated country pair and for the presence of offices of the home CRA in the rated country. Alternative explanations for the in-group bias To begin, geopolitically aligned countries tend to have a stronger trade relationship com- pared to unaligned countries (Laidi, 2008). This could induce home country CRAs to give geopolitically aligned countries higher ratings for two reasons. One the one hand, it may be that the home country tries to boost ratings of geopolitical partners not only because of geopolitical considerations, but also because deeper economic integration makes it a ra- tional strategy; if low ratings could destabilize trading partners, then a home government is well advised to try to avoid any instability in its major export markets. On the other hand, it could also be that corporate clients in the home country as well as in the respective foreign country could try to nudge agencies to assign favorable ratings to trading partners, expecting continuing good business from it. To test this mechanism, the paper controls for the share that a rated country has of the total exports of the home country. 27 Additionally, geopolitically aligned countries do not only tend to have stronger trade relationships but also deeper financial integration than unaligned countries. As a result, home country banks tend to have relatively stronger exposure to such markets. Because domestic banks are often major shareholders of domestic CRAs, they may try to use their leverage to boost ratings of important markets abroad to minimize exposure risks. To test this mechanism, the paper controls for the share of bank lending of home banks to rated countries. 2.4 Data 2.4.1 Dependent variable This paper uses long-term foreign-currency (LTFC) sovereign ratings of 28 credit rating agencies from 17 different home countries as the dependent variable. 12 Different agencies initiate and change ratings at different times and the highest fre- quency at which explanatory variables are available is monthly. As a result, this paper uses end-of-month ratings for the 30-year period January 1990 to December 2019. 13 To make ratings comparable across agencies, their alphabetic and alphanumeric rat- ings are translated into a common numeric scale. 14 This means that the highest ratings (“AAA”/“Aaa”) correspond to the value “21”, whereas the lowest ratings (“D”/“SD”) cor- 12 Table 2.8 summarizes the full and abbreviated names of the agencies and table 2.9 shows their respective home countries. See also Figure 2.3 and Figure 2.4 which show the pair-wise correlation coefficients of ratings at the end of the sample period as well as over the entire sample period, respectively. There is one agency (Egan-Jones Ratings Co) that could not be included in the sample because its ratings disclosure is subscription-based. The author reached out to agency several times to ask if the data would be made available for academic purposes but the agency declined. 13 Most of the literature uses end-of-year ratings rather than end-of-month ratings. However, FG point out that this results in a loss of information, most importantly because rating changes within a year are ignored. Additionally, an agency sometimes takes up and drops coverage of a sovereign again within the same year so that such observations would be lost by focusing on year-end ratings. In contrast, constructing monthly ratings alleviates this problem. 14 Refer to table 2.10 for an overview of the translation scale. For an overview of the number of countries covered per agency across time, refer to Figure 2.5. For an overview of the geographical distribution of assigned ratings per agency at the sample end date, consult Figure 2.6 through Figure 2.28. Note that some agencies don’t rate any sovereigns at the sample end date, so that no geographical overview figures are provided. 28 respond to the lowest notch “1.” 2.4.2 Operationalization of dyad-specific explanatory variables To operationalize the home bias, the paper constructs a dummy variable that activates for any observation in which the “rated country” is identical to the “home country” of the rating agency. See Table 2.3 for activation examples of the home country dummy. Agency Home country Rated country Home country dummy ... S&P USA USA 1 ... S&P USA Canada 0 ... ... ... ... ... ... Dagong China China 1 ... Dagong China Canada 0 ... ... ... ... ... ... Table 2.3: Example of the construction of the home country dummy. As explained in the previous section, language differences between an agency’s home country and a rated country could lead to favorable ratings for the home country indepen- dent of an actual “home bias.” Therefore, this paper uses a dummy variable that activates for any observation in which the rated country’s official language is identical to the home country’s official language. For the operationalization of the in-group bias, the paper uses the bilateral voting align- ment (agreement score) in roll-call votes in the UN General Assembly provided by (V oeten, 2019). UN voting data is widely used in the literature to measure bilateral alignment but work such as Barro and Lee (2005), Fuchs and Gehring (2017) and Dreher and Gassebner (2008) has been criticized for using it in a time series context. This is because the original data from V oeten does not make corrections to account for the temporal dependence that has become standard in binary time series cross-sectional regressions. However, the new data provided in (V oeten, 2019) controls for temporal dependence so that it can be used for the panel analysis in this paper. Nevertheless, this paper chooses two additional oper- ationalizations as robustness checks: First, the paper uses US military aid to capture the 29 alignment of country pairs. The drawback of this strategy is of course that observations are limited to ratings from US credit rating agencies only. To mitigate this, the paper con- structs a dummy variable that activates for any observation in which the home country of the agency provides a swap line to the rated country, thereby showing alignment. As explained earlier, the difference in the export dependence of the home country on other countries could lead to a preferential rating of important export markets of the home county. Not accounting for this could lead to an omitted-variable bias that warps the co- efficient of the in-group bias. Therefore, this paper controls for home countries’ export shares. Similarly, home countries’ bank exposure to rated countries proxies for the influ- ence that home country banks may have through their leverage as shareholders of credit rating agencies. To summarize, the effect of seven variables of interest will be tested in section 2.5; home country dummy and common official language dummy, geopolitical alignment (agreement score in UNGA), US military aid, home country swap line provision, bank exposure and export interests. 2.4.3 Sovereign-specific explanatory variables In addition to the variables of interest, the present paper controls for those factors that CRAs claim to look at for their assessments. In doing so, the paper builds on and combines the sets of explanatory variables used in Cantor, Packer, et al. (1994), Ferri et al. (1999), Archer et al. (2007) and FG. However, it extends these sets with data on swap lines and hidden debt owed to China. Specifically, the following variables are included for a country’s fiscal assessment: 15 • Government debt. Total central government debt as a percentage of GDP. • Hidden debt to China. Hidden central government debt owed to China as a percent- age of GDP. 15 For an overview of descriptive statistics, see Table 2.12. 30 • Cyclically adjusted balance. The structural budget balance refers to the general gov- ernment balance cyclically adjusted for nonstructural elements beyond the economic cycle. These include temporary financial sector and asset price movements as well as one-off, or temporary, revenue or expenditure items such as the sales of mobile phone licenses. The variable is measured as a percentage of potential GDP. • GDP growth. Annual percentage growth rate of GDP at market prices based on constant local currency. • GDP per capita. GDP per capita based on purchasing power parity (PPP) in 2019 international $. • Natural resources. Sum of oil rents, natural gas rents, coal rents (hard and soft), mineral rents, and forest rents in % of GDP. To control for a country’s payment and default history, these variables are included: • Historic default. This is a binary indicator variable that notes whether a country has ever defaulted since 1970. • Recent default. This is a binary indicator variable that notes whether, for a given country and year, that country has defaulted on its debt in the last five years. 16 The following variables are included as metrics of a country’s external economic perfor- mance: • Current account balance. Current account balance as a percentage of GDP. • Trade openness. Sum of exports and imports as a percentage of GDP. • External debt. Debt owed to nonresidents repayable in foreign currency, goods, or services as a percentage of GDP For the monetary assessment, these variables are controlled for: • Inflation. Annual inflation in percentage as measured by the consumer price index. • Swap line availability. Dummy that indicates if a country is at the receiving end of 16 Example: If a country defaults in 2008 and never after, the dummy is active for that country in the years 2008, 2009, 2010, 2011, and 2012. Thus, for that country in 2012, there was a default in the last five years but from the perspective of 2013, there was not. 31 at least one swap line. For a country’s institutional and political assessment, eight variables are considered: • Rule of law. Assessment of both the strength and impartiality of the legal system and of popular observance of the law on a 6-point scale. • Polity. Regime authority on a 21-point scale ranging from -10 (hereditary monarchy) to +10 (consolidated democracy). • Election in last 12 months dummy. 1 if presidential elections (for presidential or assembly-elected systems) or parliamentary elections (for parliamentary systems) were held in rated country during the last 12 months. • Years in office. Number of years the chief executive has been in office as of 1 January. • Government orientation. 1 if the chief executive’s party is defined as communist, socialist, social democratic or left-wing. • Absence of internal conflict. Assessment of both political violence in the country and its actual or potential impact on governance on a 12-point scale. • Absence of external conflict. Assessment of the risk to the incumbent government from foreign action on a 12-point scale. • Absence of military in politics. Assessment of the degree of military participation in politics on a 6-point scale. All time-varying control variables enter as lagged moving averages over one or three years. For the exact source and definition of the variables as well as an explanations of lagged moving average calculations, see Table 2.13 and the mathematical description in the ap- pendix. 2.5 Analysis 2.5.1 Identification strategy To identify the home and in-group bias, three sources of variation in the dependent vari- able can be utilized: 1) rating variation across sovereigns at a given point in time, 2) rating 32 variation across time for a given sovereign, 3) rating variation across agencies for a given sovereign at a given point in time. To use all three types of variation, the baseline specifi- cation in this paper is r a;j;i;t =x j;i;t + e i;t +p i;t + a;j + t + a;j;i;t (2.4) wherex j;i;t represent dyad-specific variables of interest for home countryj and rated coun- try i; e i;t and p i;t represent country-specific economic and political factors; a;j and t represent agency-fixed effects and time-fixed effects and a;j;i;t is the error term. As error terms may be correlated at both the agency-time and sovereign level, two-way clustering on both dimensions is allowed for. Importantly, agency-fixed effects rather than home- country-fixed effects are used to account for differences in the average rating level between different agencies from the same home country. 17 To mitigate concerns about unobserved time-invariant characteristics of rated countries (for example advantages from being the issuer of the global reserve currency or disadvan- tages from being in a monetary union), the paper also shows results for a specification that adds sovereign-fixed effects to Equation 2.4. This results in r a;j;i;t =x j;i + e i;t +p i;t + a;j + t + i + a;j;i;t (2.5) where i denotes sovereign-fixed effects. Crucially, this specification no longer uses vari- ation of type 1) to identify the home and in-group bias so that it can be seen as a more restrictive test for the home and in-group bias. The most restrictive way to identify the home and in-group bias is to rely purely on vari- ation across agencies for a given sovereign at a given point in time (type 3 variation). The paper does this by using agency-fixed effects a;j as well as sovereign-time-fixed effects 17 This seems particularly justified considering that ratings are an ordinal measures of default risk. 33 i;t so that r a;j;i;t =x j;i + e i;t +p i;t + a;j + i;t + a;j;i;t (2.6) This is indeed a very strict test. Essentially it rules out identifying a bias through time- inconsistencies of agencies and relies purely on cross-sectional variance in ratings. Primarily, the paper uses OLS to estimate the home and in-group bias. However, be- cause OLS treats the dependent variable as cardinal, results of ordered probit models are reported as well. In addition, because of the structural brake that happened in September 2008 with the bankruptcy of Lehman Brothers and AIG, the paper reports results for two sample periods. The full sample covers January 1990 to December 2019 whereas the “GFC sample” covers September 2008 through December 2019. 2.5.2 Main results Results for the baseline specification when all variables of interestx j;i;t are excluded are shown in Table 2.14 in the appendix. The results are reassuring in that they are broadly in line with previous studies and show that economic and political variables account for up to 80.6% of the variance in rating outcomes. The table does not tell us whether a home or in-group bias exists, however. For this, the paper adds one variable of interest at a time to the baseline specification. Hence, each cell in Table 2.4 refers to a separate regression and shows the coefficient of the added variable using either OLS or ordered probit for different sample lengths. 18 Do rating agencies have a home bias? First off, note the coefficient for the home country dummy in the first column of Table 2.4. It is highly statistically significant and its size (1.7) is of considerable economic magnitude. It suggests that agencies on average assign a rating to their home country that is almost two rating notches above the value that would be expected based on economic and political fundamentals. Focusing on the time after 18 Results for the control variables of the respective regressions are shown in the appendix in table 2.15— table 2.18. 34 Full sample, OLS GFC sample, OLS Full sample, ordered probit GFC sample, ordered probit Home country dummy 1.7049*** ( 0.0000 ) 1.41153*** ( 0.0000 ) 1.38975*** ( 0.0000 ) 1.23119*** ( 0.0000 ) Export interests 0.13974*** ( 0.0000 ) 0.10421*** ( 0.0000 ) 0.14626*** ( 0.0000 ) 0.09342*** ( 0.0000 ) Bank exposure -0.0574*** ( 0.0002 ) -0.02847 ( 0.4836 ) 0.00497 ( 0.7746 ) -0.57427*** ( 0.0000 ) Geopolitical alignment -0.02274*** ( 0.0000 ) 0.02845*** ( 0.0000 ) 0.02135*** ( 0.0000 ) 0.02737*** ( 0.0000 ) US military interest 0.19747*** ( 0.0000 ) 0.69647*** ( 0.0000 ) -0.07555*** ( 0.0044 ) 0.22536*** ( 0.0000 ) Common official language 0.68616*** ( 0.0000 ) 0.2685*** ( 0.0000 ) -0.04762 ( 0.3194 ) -0.76868*** ( 0.0000 ) Home country swap line 0.39208*** ( 0.0000 ) 0.28136*** ( 0.0000 ) 1.16383*** ( 0.0000 ) 1.16383*** ( 0.0000 ) Table 2.4: Coefficients of variables of interest across different samples and estimation types. Each cell refers to a separate regression of the baseline specification Equation 2.4 without variables of interest. All regressions contain agency-fixed effects and time-fixed effects. The dependent variable is a country’s rating on a numeric scale from 1-21. Stan- dard errors are clustered at both the agency-time and the sovereign level. Full sample: 1/1990-12/2019. GFC sample: 9/2008-12/2019. *,**,*** correspond to 10%, 5% and 1% significance, respectively. P-values are displayed in parenthesis. Lehman’s bankruptcy when CRAs came under increased scrutiny, the coefficient remains highly statistically significant while its size decreases only marginally to 1.41. In other words, even after the Great Financial Crisis, agencies still assigned considerably higher ratings to their home country than would be objectively justified. Next, consider the coefficient for common official language, which is an alternative channel that could lead to a favorable rating of the home country that is unrelated to an actual bias. As the table shows, the coefficient is highly statistically significant across all regressions except one. Particularly for the sample that starts before the Great Financial Crisis the size is also considerable (0.69). This suggests that agencies may favor countries whose language they find easier to comprehend based on informational advantages. How- ever, this result should not be over-interpreted as it emerges from regressions that do not control for different variables of interest simultaneously. Nevertheless, the results so far discussed are robust to the type of estimator and sample length used, suggesting that there is strong evidence in favor of the hypothesis that there is a home bias in sovereign ratings. Do rating agencies have an in-group bias? The picture painted in Table 2.4 is mixed. On the one hand, the coefficient for geopolitical alignment switches signs between sample lengths but overall hovers around zero, suggesting that there’s no in-group bias in general. On the other hand, it looks as if an in-group bias manifests when coefficients for US mili- tary interest are examined. There, it would seem that (American) agencies on average give 35 more favorable ratings to countries that the US government supports militarily. Even if the coefficients don’t look large in absolute terms we must keep in mind that the underlying variable (countries’ different shares of the total of US military aid given) varies quite sub- stantially. Then again, the results on US military interest are not robust to the choice of the estimation method. Also, consider the other variables possibly related to the in-group bias. The coeffi- cient for export interests takes on the expected sign and while statistically significant its economic magnitude is relatively minor. For bank exposure, the coefficient even becomes slightly negative but not across all regressions so that it could also be said that bank expo- sure seems unrelated to sovereign ratings. To verify the robustness of previous results, the next step adds all variables of interest simultaneously to the model. Results are shown in Table 2.5. Dependent variable: Numeric sovereign rating OLS OLS OLS OLS OLS OLS (1) (2) (3) (4) (5) (6) Home country dummy 1:70490 1:41153 2:31882 1:06238 (0:11528) (0:10431) (0:12038) (0:11418) Geopolitical alignment 0:07306 0:03434 0:03566 0:01896 (0:00258) (0:00345) (0:00218) (0:00250) Export interest 0:16791 0:09788 (0:00599) (0:00689) Bank exposure 0:23063 0:12788 (0:01547) (0:04481) Home country swap line 0:58529 0:23534 (0:05640) (0:06837) Sample Fullsample GFCsample Fullsample GFCsample Fullsample GFCsample Observations 18,451 12,176 10,750 6,494 10,750 6,494 Adjusted R 2 0.73720 0.80891 0.76895 0.80057 0.76902 0.80080 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. Standard errors are clustered at both the agency-time and sovereign level. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency- and time-fixed effects. Standard errors are displayed in parenthesis. Table 2.5: Results when variables of interest are controlled for simultaneously. Each col- umn refers to a separate regression; full results are in Table 2.19. Let’s first focus on the alternative mechanism that could be related to the home bias (common official language) in columns 1 and 2. Estimating both coefficients simultane- ously is not possible due to singularity that arises from poor data availability. As such, only the home country dummy is reported so that the coefficients in columns 1 and 2 correspond to their respective values in Table 2.4. 36 Next, columns 3 and 4 introduce all variables related to the in-group bias simultane- ously. The results are broadly in line with the ones shown in Table 2.4. Specifically, the coefficient for geopolitical alignment is very close to zero and if anything would take on an unexpected negative sign at times. Likewise, export interest and bank exposure have nearly identical coefficients independent of whether they are estimated individually as in Table 2.4 or simultaneously as here. Lastly, the coefficient for home country swap line shows some instability. When estimated individually, it is positively associated with sovereign ratings whereas it is negatively related to ratings when controlled for alongside other variables of interest. Finally, columns 5 and 6 introduce the home country dummy and geopolitical align- ment simultaneously. The results are again in line with previous findings in this paper. In particular, there is evidence of a considerable home bias over a longer time period (2.32) as well as over the shorter sample that starts after Lehman’s chapter 11 filing (1.06). Also, introducing both bias measurement variables at the same time doesn’t change the interpre- tation about the relative absence of an in-group bias; though the post-GFC coefficient is statistically significant and takes on the expected sign, it is relatively small in economic terms. 19 To summarize, there is quite strong evidence for agencies having a home bias but an in-group bias does not find equally strong support in the data. These results are robust to the choice of the estimator (OLS and ordered probit) as well as to the sample length. Additionally, the results are relatively robust to the model specification so that the coeffi- cients are quite similar across individual regressions and regressions that include variables of interest simultaneously. Importantly, the result about the relative absence of an in-group bias is in line with the theory developed in subsection 2.3.2 as these regressions do not take into consideration the different nature of the political and economic system across home 19 As an additional robustness check, the paper tried to estimate a model in which all variables of interest are included. However, poor data availability causes singularity so that several coefficients cannot be estimated and the results are not reported. 37 countries. This point is revisited in subsection 2.5.4. So far, the paper relies on regressions of the type outlined in Equation (2.4) that utilize variation in ratings across countries because ratings for countries can be sluggish across time. However, for such countries a home and in-group bias could be reflected mostly in their initial rating level which then persists over time and which is not captured by an identification strategy that relies on variation over time. Thus, using regressions of the type Equation 2.5 and Equation 2.6 constitutes a strong robustness check. And the results shown in Table 2.6 suggest that the theory passes this check. eq. (2.4) Set of FE eq. (2.5) Set of FE eq. (2.6) Set of FE Home country dummy 1.7049*** ( 0.0000 ) Agency 0.9777*** ( 0.0000 ) Agency; Sovereign 0.8548*** ( 0.0000 ) Agency; Sovereign-Year pair Export interests 0.1397*** ( 0.0000 ) Agency 0.1021*** ( 0.0000 ) Agency-sovereign pair 0.0677* ( 0.0818 ) Agency-sovereign-year pair Bank exposure -0.0574*** ( 0.0002 ) Agency -0.1458*** ( 0.0000 ) Agency-sovereign pair -0.0586*** ( 0.003 ) Agency-sovereign-year pair Geopolitical alignment -0.0227*** ( 0.0000 ) Agency -0.0075*** ( 0.0004 ) Agency-sovereign pair 0.0023 ( 0.4856 ) Agency-sovereign-year pair US military interest 0.1975*** ( 0.0000 ) Agency 0.0355 ( 0.1166 ) Agency-sovereign pair 0.0229 ( 0.6018 ) Agency; sovereign-year pair Common official language 0.6862*** ( 0.0000 ) Agency -0.2034*** ( 0.0000 ) Agency; Sovereign -0.436*** ( 0.0000 ) Agency; sovereign-year pair Home country swap line 0.3921*** ( 0.0000 ) Agency -0.3026*** ( 0.0000 ) Agency; Sovereign 0.0728* ( 0.0725 ) Agency-sovereign-year pair Table 2.6: Coefficients of variables of interest across different fixed effect models. The dependent variable is a country’s sovereign rating on a numeric scale from 1-21. Each cell refers to a separate regression. The table displays only the coefficients of the respective variable of interest of each regression. All regressions contain political and economic con- trol variables, time-fixed effects, as well as the fixed effects specified in the head row of the table. Column 1 is identical with column 1 of Table 2.4 to facilitate comparisons. The full sample is used (1990-2019). Standard errors are clustered at both the agency-time and the sovereign level. ***,**,* indicate significance at the 1%, 5% or 10% level. P-values are displayed in parentheses. First off, note that the home country dummy remains relatively large in value (0.98 and 0.85) even in the restrictive models while the same is not the case for common official language. This may be indication that agencies do not discriminate based on differences in language/information and instead have a true home bias. Secondly, note that the coefficient for geopolitical alignment is quite stable between columns 1 and 2 and changes only slightly in column 3. However, in column 3 the coef- ficient is not statistically significant whereas it is in the other two columns. The relatively small size of the coefficient provides additional support for hypothesis 3 that states that there isn’t generally an in-group bias in sovereign ratings. Similarly, the results for the 38 in-group bias as measured by US military interest are fairly robust. In fact, the coefficient loses statistical significance in columns 2 and 3, providing evidence that US agencies don’t seem to to have an in-group bias for countries of the same language. Equally, the coeffi- cients for export interests and bank exposure are nearly unchanged across all three columns but are very close to zero. This constitutes further evidence that the in-group bias is not a broad phenomenon. Furthermore, column 3 of Table 2.6 suggests that the biases are nei- ther driven by time-invariant factors such as the advantage of being the issuer of the reserve currency, nor by sovereign-time specific factors such as short-term volatility in domestic financial market conditions. So far, the paper pooled data from all agencies together into one regression. Implicitly, this assumes that each agency weighs all sovereign-specific factors in the same way. As a robustness check, Table 2.20 shows results of agency-specific regressions so that the as- sumption of common weights for factors across agencies is relaxed. Note that due to the small amount of rated countries as well as poor data availability for certain home coun- tries, the baseline model cannot be estimated for each agency individually. However, for those agencies where the home country dummy and geopolitical alignment variable can be estimated, the coefficients are broadly in line with their counterparts in the pooled setting. Taken together, the evidence presented so far supports both hypothesis 1 and 3. Specif- ically, there is strong evidence for a home bias in sovereign ratings among all major rating agencies. In contrast, there is relatively weak evidence for the in-group bias being a broad phenomenon. 2.5.3 Exploration of home bias transmission In the previous subsection, the paper uses the home country dummy and the common offi- cial language dummy to explore if the preferential treatment of the home country is mainly caused by an actual home bias or by informational advantages stemming from easier infor- mation collection and analysis for country dyads with shared language. For the most part, it 39 appears as if the home bias accounts for a larger share of the preferential treatment of home sovereigns because its coefficients are more stable throughout different specifications and robustness checks whereas the coefficient for the common official language dummy some- times approaches zero. Nevertheless, having a common official language seems to play a role in several settings, which warrants a deeper exploration of the channels that lead to a preferential treatment of agencies’ home countries. A first approach to getting a clearer understanding of the mechanisms leading to a pref- erential treatment of the home country is to analyze the timing of rating changes between home and foreign agencies. This approach goes back to Alsakka and Ap Gwilym (2010) which analyzes lead–lag relationships in sovereign ratings across five agencies and find ev- idence of interdependence of rating actions. Importantly, if domestic agencies have better access to information about the home sovereign than foreign agencies, then they should be the first to up- or downgrade its rating whenever its fundamentals indicate so. If they instead act later than foreign agencies do, then this would suggest that the preferential treat- ment of the home country is mostly based on an actual home bias rather than information advantages. Indeed, the data in Figure 2.29 through Figure 2.45 shows that home agencies are late to the game of changing ratings, although not to the same extent across all home countries. The figures indicate that domestic agencies generally change the rating of the home sovereign only after foreign agencies have already acted and that this phenomenon is particularly evident for Bulgaria, Cyprus, and Trinidad and Tobago. This suggests that informational advantages are not likely to be the main reason for the better treatment of the home country. Secondly, comparing the ratings of smaller and larger agencies is another way to ex- plore the mechanisms leading to relatively high home country ratings. In particular, large and established agencies should have an information advantage over smaller agencies be- cause they have larger analyst teams and more resources to draw on. By implication, the coefficient for the home country dummy should be smaller for large agencies. However, 40 Table 2.20 does not provide clear evidence that this is the case. For example, while the relatively small agencies Dagong and Lianhe have coefficients which are much bigger than the ones of the major agencies S&P, Moody’s and Fitch, other small agencies such as CI, RAEX and Scope have smaller coefficients than major agencies. A third approach to understanding the mechanisms behind the home bias is developed in FG. It relates to Giannetti and Yafeh (2012) who find that the cultural bias in bank lending is mitigated when banks have a subsidiary in the foreign country. For this approach, a binary variable is constructed that takes on the value 1 whenever a rating agency has an office in the rated country and 0 otherwise. 20 Table 2.22 provides the regressions results. Interestingly, in this setting there is little evidence that having an office in a rated country is associated with a higher rating for this country. Although the coefficient of office in rated country is statistically significant, it is negligibly small (0.14). This implies that the local presence of agency staff does not mitigate the home bias which in turn suggests that information advantages are not the main reason for the preferential treatment of home countries. A last approach to compare the magnitude of the home bias with the magnitude of the preferential home country treatment caused by information advantages is shown in Table 2.23. If information is the main channel through which the difference in the language of home country and the rated country affects ratings, then the coefficient on common official language should shrink in absolute terms once we control for the informational transparency of the rated country as well as for the presence of an office of the rating agency in the rated country. However, this does not seem to be the case. In fact, the coefficient for common official language at first even increases when adding information transparency and then remains above the initial value when adding information transparency and office 20 An overview of the offices of agencies at the end of the sample period (2019-12-31) is provided in Table 2.21. The central challenge of this approach is the static nature of the data. Only at the sample end-date is it known with certainty whether an agency has an office in a given country or not, but historical time series are not available and cannot be assembled based on the information from company websites. Thus, the paper makes the assumption that if an agency has an office in a given country at the end of the sample period, then it has had this office through the entirety of the sample period. 41 in rated country simultaneously. This further increases confidence in an actual home bias rather than the idea that preferential ratings for the home country are related to information asymmetries. 2.5.4 Exploration of in-group bias transmission So far, the paper provides evidence supporting H1 and H3. That is, the analysis of sovereign ratings suggests that agencies tend to have a home bias of considerable magnitude and that the geopolitical in-group bias—as a broad phenomenon across agencies—is relatively small if not absent. In what follows, the paper sets out to test hypotheses 2 and 4. Specifically, it attempts to find out whether authoritarian regimes put pressure on home agencies to achieve higher ratings for the home country (H2) and for geopolitically aligned countries (H4) such as to weaponize credit ratings as a foreign policy tool. To test H2 and H4, the notion of an “authoritarian regime” must be clearly defined and operationalized. One way to do so is by ranking countries according to their level of democracy/authoritarianism as reported in indices such as the Economist Intelligence Unit’s Democracy Index. However, such indices tend to be everything-but-the-kitchen- sink and often aggregate many somewhat unrelated concepts into one number. As such, they are not a suitable measure for this paper. Instead, the paper constructs a new variable to measure whether a home country is authoritarian in the sense that it demonstrably inter- feres with domestic agencies’ independence. Of the 17 home countries of the agencies used in this study (cf. Table 2.9) China is the only country where government involvement in CRAs is not only conceivable but also demonstrable. For example, in April 2019 a Chinese state-owned enterprise has taken over the credit rating agency Dagong. This came after regulators suspended Dagong’s licences, arguing that Dagong suffered from lax corporate governance and conflicts of interest. Similarly, China Orient Asset Management—a Chi- nese state-owned enterprise—is the controlling shareholder of Beijing-based Golden Credit 42 Rating International. 21 Considering the way how the Chinese government has gained di- rect influence over Dagong and Golden, it doesn’t appear far-fetched that private Chinese agencies may align their rating behavior to Dagong’s and Golden’s lest the government may intervene in them too. Following this logic, the paper codes a binary “authoritarian government” variable that takes on the value 1 for Chinese agencies and 0 otherwise. Be- cause China is the only home country that is classified as authoritarian, the binary variable is called Chinese agency dummy. (1) (2) (3) (4) Home country dummy 1:15639 (0:17481) Geopolitical alignment 0:05458 0:05710 (0:00235) (0:00236) Export interest 0:13776 0:13259 (0:00448) (0:00440) Home country dummy:Chinese agency dummy 0:98124 (0:23518) Geopolitical alignment:Chinese agency dummy 0:19496 0:19187 (0:00703) (0:00929) Export interest:Chinese agency dummy 0:82661 0:80275 (0:06969) (0:06864) Table 2.7: Home bias and in-group bias for Chinese agencies. Each column refers to a separate regression. Full results are provided in Table 2.24 . To evaluate whether domestic agencies in authoritarian countries show a systematically magnified home and in-group bias, variables of interest are sequentially added to the base- line specification and results are shown in Table 2.7. For the first regression, the home country dummy is added alongside its interaction with the Chinese agency dummy. Interestingly, the home country dummy is highly statistically significant and its magnitude is comparable to previous results which corroborates the ear- lier finding that a home bias is a broad phenomenon across all agencies. More importantly, note that the interaction Home country dummy:Chinese agency dummy is not only highly 21 Golden is not part of the sample of this paper. This is because Golden only started reporting sovereign ratings in 2020 but the sample end date for this paper is 2019-12-31. 43 statistically significant but the coefficient (0.98) supports H2. Put differently, the coeffi- cient suggests that authoritarian regimes such as China seem to influence domestic rating agencies to the effect that they assign their home country a particularly favorable rating that is on average almost one rating notch higher than the home country ratings of agencies in states without authoritarian governments. Next, note that the coefficient for geopolitical alignment in column 2 is in line with the previous findings in this paper; an in-group bias as measured by alignment does not seem to be a broad phenomenon. However, the interaction Geopolitical alignment:Chinese agency dummy suggests that the in-group bias is heightened for agencies from authoritarian states. Indeed, the coefficient is highly statistically significant in spite of the relatively small sample size, and its economic magnitude, while not overwhelming, is considerable at 0.19. It implies that the difference in ratings that Chinese agencies assign to other countries can be as large as 0.38—almost half a rating notch—between a country that is politically aligned with China and a country which is opposed to China’s interest, at least as far as UN general assembly votes are concerned. The coefficient of export interest in column 3 is comparable to values found earlier in this paper. However, the interaction export interest:Chinese agency dummy shows that agencies from authoritarian countries weigh their home country’s export interest stronger than other agencies. In fact, the coefficient of 0.83 suggests that Chinese agencies assign substantially higher ratings to countries which are major export destinations for Chinese goods. A back-of-the envelope calculation illustrates this point. Take for example Japan and Germany, which are major export destinations for Chinese companies with shares of Chinese exports of 5.7% and 3.2%, respectively. The difference in their export share (2.5 percentage points) would predict a difference in assigned ratings of 2.5*0.83=2.08, which is more than two rating notches. Obviously, the predicted effect is even larger for country pairs with bigger differences in Chinese export shares. All that is to say is that there is evidence for agencies in authoritarian countries to assign discriminatory ratings based on 44 the economic interest of the home country. As a robustness check, consider column 4 in which the home country dummy and it’s interaction terms as well as the geopolitical variables and their interaction terms are simul- taneously added to the baseline specification. 22 Crucially, the coefficients of geopolitical alignment and export interest are very similar between columns 3 and 4, suggesting that the in-group bias and the preferential treatment of strong trading partners happen in parallel rather than being endogenous. The same thing happens with their interaction terms, which suggests that Chinese agencies not only have a true in-group bias but a tendency to favor certain countries for economic reasons. Taken together, there is evidence for hypotheses 2 and 4. That is, there is evidence that CRAs in authoritarian countries assign preferential ratings to geopolitically aligned countries as well as to countries with strong economic ties with the home country. The effects are highly statistically significant and particularly in the case of export interest they are also of economic significance. 2.5.5 Beyond average treatment effects All previous regressions in this paper show average “treatment effects” of the variables of interest. For example, the paper demonstrates that the average home bias is somewhere be- tween one to two rating notches, depending on the sample length and the type of variance of the dependent variable used in the estimation process. Equally, the paper shows that on average the in-group bias as measured by geopolitical alignment is close to zero. Knowing average treatment effects is essential for testing the paper’s hypotheses and demonstrat- ing that sovereign ratings are biased and influenced by business considerations of agencies. However, average treatment effects have limitations, particularly when it comes to inferring the political implications of ratings, for two reasons. On the one hand, average treatment effects do not show if the discovered phenomena of the home and in-group bias are robust 22 Note that the coefficient for the home country dummy and its interaction with the Chinese agency dummy is not reported due to singularity of the data matrix. 45 across the rating distribution. On the other hand, average treatment effects do not allow us to understand the economic implications of biased ratings on affected sovereigns. Ex- actly because average treatment effects boil down all information into a single number, they can mask some of the subtleties of the underlying distribution so that a comprehension of heterogeneity in treatment effects along the rating distribution is not feasible. As a conse- quence, we do not know if the home and in-group bias are present at regulatory thresholds that determine eligibility of financial market participants to hold certain debt instruments and thus influence the governments’ access to financial markets. 23 Quantile regressions allow us to capture the heterogeneity of treatment effects. In ad- dition to quantiles, the paper reports treatment effects at three important rating thresholds: the threshold for money market fund purchases (“AA+/AA”), the threshold for US pension fund investments (“A/A-”) and the investment-grade threshold for bank purchases (“BBB- /BB+”). Results are shown in Table 2.25. First off, note that the home country dummy is highly statistically significant at all thresholds as well as at all quintiles except the 2nd (i.e. the 40th percentile). The co- efficients are also positive across all columns, suggesting that the home bias is a broad phenomenon that exists throughout the universe of rated sovereigns and agencies. Inter- estingly, the coefficients at the lower end (0.88 for the 20th percentile, 0.56 for the 36th percentile that marks the threshold to investment grade, and 0.29 for the 40th percentile) tend to be smaller than their counter parts at the higher end (2.32 for the 60th percentile, 2.31 for the 61th percentile that marks the threshold for US pension fund purchases, and 3.48 for the 81st percentile that marks the threshold for money market fund purchases). Next, consider the coefficients of common official language which is negative across all columns. However, at the lower end of the rating distribution, the coefficients are close to 23 Jaramillo and Tejada (2011) find that sovereign investment grade status is often associated with lower spreads in international markets. Using a panel framework for 35 emerging markets between 1997 and 2010, they show that crossing the threshold to investment grade status reduces spreads by 36 percent, above and beyond what is implied by macroeconomic fundamentals. This compares to a 5-10 percent reduction in spreads following upgrades within the investment grade asset class, and no impact for movements within the speculative grade asset class, holding all else equal. 46 zero and they are not statistically significant. In contrast, at the higher end the data sug- gests that having a common language is negatively related with ratings, which is surprising given that the paper has previously mostly found the relationship to be positive (compare to Table 2.4). The finding of a negative relationship suggests that the preferential treatment of the home country is mainly due to internal biases of agencies rather than information disadvantages. This is because countries at the lower end of the rating distribution tend to be relatively less developed and to have weak institutions and low government transparency so that information disadvantages should play a particularly large role there, which is not the case. Moving on to the in-group bias and it’s related variables, consider the coefficient for geopolitical alignment. It is statistically significant across all columns and the coefficients are always positive, although not in an economically significant way. This confirms the previous finding that an in-group bias as measured by geopolitical alignment is not a broad phenomenon. Similar results are found if we use US military interest as a measure for geopolitical alignment. The coefficients for export interest and bank exposure are fairly equal at all eight per- centiles and are in line with the findings from Table 2.4. Specifically, the coefficients suggest that economic interests of the home country as measured by bank exposure and export interest are positively related to sovereign ratings throughout the rating distribution. Likewise, swap lines given by the home country to the rated country seem to be positively correlated with sovereign ratings across all percentiles, suggesting that the in-group bias as measured by swap lines is considerable and highly statistically significant. Lastly, the quantile regressions for the GFC sample generally generate coefficients that are similar in economic and statistical significance to their counterparts in the full sample. Taken together this underlines the importance of the variables of interest in explaining sovereign ratings and highlights that the home bias is a broad, robust finding whereas the in-group bias is negligibly small. However, economic interests related to the in-group bias 47 seem to be an important factor for sovereign credit ratings. 2.6 Conclusion Sovereign credit ratings are a central organizing feature of international financial markets. In light of this, the present paper investigates the behavior of credit rating agencies. Specif- ically, the paper constructs a panel of monthly data from 28 credit rating agencies based in 17 countries over the thirty year period 1990-2019 for 147 rated sovereigns to test if sovereign ratings show a home and a geopolitical in-group bias. The main results can be summarized in four points: First, there is strong evidence for a home bias in sovereign credit ratings among all agencies. That is, credit rating agencies assign their home country ratings which exceed ratings based on economic and political fundamentals on average by a full rating notch. This finding is robust to different identification strategies, sample periods, and estimation methods (OLS and ordered probit). In addition, the home bias does not go away when controlling for information asymmetries that may arise when a domestic agency has to compare its home country to a foreign country with a different language. Similar results are also found when comparing the timing of rating changes; The home agency typically reacts to up- or downgrades of the home country by foreign agencies with a lag rather than initiating the rating adjustment wave. Moreover, there is evidence that the home bias is larger at the upper end of the rating distribution where it reaches as much as 3.6 rating notches. Furthermore, there is evidence that agencies based in authoritarian countries have a magnified home bias. Second, the paper finds that an in-group bias is not a broad phenomenon. If it exists at all, its economic relevance is relatively negligible. Again this finding is robust to the choice of estimation method, sample length, and identification strategy. Third, the paper presents evidence suggesting that economic considerations related to the in-group bias have more severe implications for ratings than a potential in-group bias 48 has on its own. Specifically, agencies assign higher ratings to countries which are major export markets for domestic companies, and to a lesser degree to countries which are major lending markets for domestic banks. Similarly, domestic agencies assign ratings to coun- tries to which the home country extends a swap line that are on average almost one notch above what would be expected based on economic and political fundamentals of the rated country. Fourth, the paper revisits the in-group bias by taking the economic and political sys- tem of the home country seriously. Doing so, the paper provides evidence that agencies in authoritarian regimes assign systematically higher ratings to geopolitically aligned coun- tries in comparison to agencies from economically libertarian countries. In other words, authoritarian intervention can cause or magnify an in-group bias in sovereign ratings. Sim- ilarly, the paper presents data indicating that the home bias is magnified for agencies based in authoritarian countries. Moreover, there is evidence suggesting that agencies from au- thoritarian countries have a stronger tendency to assign ratings that reflect the economic interests of the home country. As far as regulators are concerned, the findings discussed should not be taken as in- dication of the obsolescence of credit ratings at large. Indeed, credit ratings perform an important function in the global financial system in that they provide heuristics for risk as- sessments. Nevertheless, the results presented suggest that an over-reliance of regulators on credit ratings may be counter-productive and naive. However, one way to make credit rat- ings a more objective and reliable tool for measuring risk is for regulators to require credit rating agencies to make their assessment models publicly available and reproducible. As far as governments are concerned, biases as small as one rating notch around impor- tant thresholds can influence the conditions at which international capital markets can be accessed. Thus, governments should be careful not to risk a lower rating through imprudent economic policies, independent of whether they benefit from positively biased ratings or suffer from negatively biased ratings. 49 As far as private investors are concerned, the findings suggest that credit ratings should not go unquestioned. Specifically, given that biased ratings around important threshold may falsely prevent investors from investing in certain debt instruments, investors have a vested interest in regulators overhauling their due diligence frameworks for credit rating agencies. 50 2.7 Paper appendix Sub-steps Steps of the rating process 1) Rating initiation 2) Due diligence 3) Rating assignment 4) Rating publication/monitoring Rating request/unsolicited initiation Assignment of rating team Data gathering Questionnaires/meetings with sovereign Gathering additional data Analysis Draft report/rating proposal Committee meeting Notification of issuer Finalized report/rating Rating publication Ongoing monitoring Figure 2.1: Schematic illustration of the rating process. Details about variables used by credit rating agencies Figure 2.2 uses the example of Standard & Poor’s to show what rating agencies typically state about how they rate sovereigns. A comparison of the websites of the 28 agencies examined in this paper shows that virtually all agencies list either the same five pillars or a subset thereof as the base for their sovereign ratings, if they provide such an overview at all. While most agencies remain relatively unspecific about the exact variables that they consider in forming a rating, Standard & Poor’s provides such an overview in its rating 51 Figure 2.2: Variables used by Standard & Poor’s to assess a sovereign issuer. Similar statements are made, if at all, by virtually all other credit rating agencies in the sample. methodology. An excerpt is shown in the box below. The full rating methodology can be found at https://www.standardandpoors.com/en US/web/guest/article/-/view/sourceId/ 10221157. 52 5. Our sovereign rating criteria incorporate the factors that we believe affect a sovereign government’s willingness and ability to service its financial obligations to nonofficial creditors on time and in full. The foundation of our sovereign credit analysis rests on five pillars. 6. The institutional assessment reflects our view of how a government’s institutions and policymaking affect a sovereign’s credit fundamentals by delivering sustainable public finances, promoting balanced economic growth, and responding to economic or political shocks. It also reflects our view of the transparency and accountability of data, processes, and institutions; a sovereign’s debt repayment culture; and potential external and domestic security risks. 7. The history of sovereign defaults suggests that a wealthy, diversified, resilient, and adaptable economy ultimately boosts its debt-bearing capacity. The economic assessment incorporates our view of: • The country’s income levels as measured by its GDP per capita, indicating broader potential tax and funding bases upon which to draw, which generally support creditworthiness; • growth prospects; and • Its economic diversity and volatility. 8. A country’s external assessment, which refers to the transactions and positions of all residents (public- and private-sector entities) vis-` a-vis the rest of the world, is primarily driven by our view of: 53 • The status of a sovereign’s currency in international transactions; • The country’s external liquidity, which provides an indication of the econ- omy’s ability to generate the foreign exchange necessary to meet its public- and private-sector obligations to nonresidents; and • The country’s external position, which shows residents’ assets and liabilities (in both foreign and local currency) relative to the rest of the world. 9. The fiscal assessment reflects our view of the sustainability of a sovereign’s deficits and its debt burden. This measure considers fiscal flexibility, long-term fis- cal trends and vulnerabilities, debt structure and funding access, and potential risks arising from contingent liabilities. Given the many dimensions that this assessment captures, the analysis is divided into two segments, ”fiscal performance and flexibil- ity” and ”debt burden.” 10. The monetary assessment considers our view of the monetary authority’s ability to fulfill its mandate while sustaining a balanced economy and attenuating any major economic or financial shocks. We derive the monetary assessment by analyzing: • The exchange rate regime, which influences a sovereign’s ability to coordinate monetary policy with fiscal and other economic policies to support sustainable economic growth; and • The credibility of monetary policy as measured, among other factors, by infla- tion trends over an economic cycle and the effects of market-oriented monetary mechanisms on the real economy, which is largely a function of the depth and diversification of a country’s financial system and capital markets. a a The schematic illustration is based on information from the website of Standard&Poor’s. 54 Abbreviation Agency name ACRA ACRA AcraEurope ACRA Europe Austin Austin Ratings Axesor Axesor Ratings BCRA BCRA-Credit Rating Agency AD CariCRIS CariCRIS Chengxin Chengxin CI Capital Intelligence Ratings Creditreform Creditreform Rating Dagong Dagong DBRS Dominion Bond Rating Service Fareast Far East Credit Rating Fitch Fitch HR HR ratings IIRA Islamic International Rating Agency JCR Japan Credit Rating Agency JCREurasia JCR Eurasia Rating KROLL Kroll Bond Rating Agency Lianhe Lianhe Credit Rating Moody’s Moody’s Pengyuan Pengyuan R&I R&I RAEX RAEX RAM RAM Ratings S&P Standard & Poors Scope Scope Ratings ShanghaiBrilliance Shanghai Brilliance Credit Rating & Investors Service Co.,Ltd TRIS TRIS Rating Table 2.8: Credit rating agency names and their abbreviations used in this paper. 55 Agency Home country IIRA Bahrain Austin Brasil BCRA Bulgaria DBRS Canada Chengxin China Dagong China Fareast China Lianhe China Pengyuan China Scope China CI Cyprus Creditreform Germany RAEX Germany S&P Germany JCR Japan R&I Japan RAM Malaysia HR Mexico ACRA Russia AcraEurope Slovakia Axesor Spain TRIS Thailand CariCRIS Trinidad and Tobago JCREurasia Turkey Fitch USA KROLL USA Moody’s USA ShanghaiBrilliance USA Table 2.9: Home countries of credit rating agencies used in this paper. 56 Numeric scale ACRA AcraEurope Austin Axesor BCRA CariCRIS Chengxin CI Creditreform Dagong DBRS Fareast Fitch HR IIRA JCR JCREurasia KROLL Lianhe Moody’s Pengyuan RI RAEX RAM SP Scope ShanghaiBrilliance TRIS Numeric scale 21 AAA AAA AAA AAA AAA AAA AAAg AAA AAA AAA AAA AAA AAA HR AAA (G) AAA AAA AAA AAA AAA Aaa AAA AAA AAA gAAA AAA AAA AAA AAA 21 20 AA+ AA+ AA+ AA+ AA+ AA+ AAg+ AA+ AA+ AA+ AAH AA+ AA+ HR AA+ (G) AA+ AA+ AA+ AA+ AA+ Aa1 AA+ AA+ AA+ gAA1 AA+ AA+ AA+ AA+ 20 19 AA AA AA AA AA AA AAg AA AA AA AA AA AA HR AA (G) AA AA AA AA AA Aa2 AA AA AA gAA2 AA AA AA AA 19 18 AA- AA- AA- AA- AA- AA- AAg- AA- AA- AA- AAL AA- AA- HR AA- (G) AA- AA- AA- AA- AA- Aa3 AA- AA- AA- gAA3 AA- AA- AA- AA- 18 17 A+ A+ A+ A+ A+ A+ Ag+ A+ A+ A+ AH A+ A+ HR A+ (G) A+ A+ A+ A+ A+ A1 A+ A+ A+ gA1 A+ A+ A+ A+ 17 16 A A A A A A Ag A A A A A A HR A (G) A A A A A A2 A A A gA2 A A A A 16 15 A- A- A- A- A- A- Ag- A- A- A- AL A- A- HR A- (G) A- A- A- A- A- A3 A- A- A- gA3 A- A- A- A- 15 14 BBB+ BBB+ BBB+ BBB+ BBB+ BBB+ BBBg+ BBB+ BBB+ BBB+ BBBH BBB+ BBB+ HR BBB+ (G) BBB+ BBB+ BBB+ BBB+ BBB+ Baa1 BBB+ BBB+ BBB+ gBBB1 BBB+ BBB+ BBB+ BBB+ 14 13 BBB BBB BBB BBB BBB BBB BBBg BBB BBB BBB BBB BBB BBB HR BBB (G) BBB BBB BBB BBB BBB Baa2 BBB BBB BBB gBBB2 BBB BBB BBB BBB 13 12 BBB- BBB- BBB- BBB- BBB- BBB- BBBg- BBB- BBB- BBB- BBBL BBB- BBB- HR BBB- (G) BBB- BBB- BBB- BBB- BBB- Baa3 BBB- BBB- BBB- gBBB3 BBB- BBB- BBB- BBB- 12 11 BB+ BB+ BB+ BB+ BB+ BB+ BBg+ BB+ BB+ BB+ BBH BB+ BB+ HR BB+ (G) BB+ BB+ BB+ BB+ BB+ Ba1 BB+ BB+ BB+ gBB1 BB+ BB+ BB+ BB+ 11 10 BB BB BB BB BB BB BBg BB BB BB BB BB BB HR BB (G) BB BB BB BB BB Ba2 BB BB BB gBB2 BB BB BB BB 10 9 BB- BB- BB- BB- BB- BB- BBg- BB- BB- BB- BBL BB- BB- HR BB- (G) BB- BB- BB- BB- BB- Ba3 BB- BB- BB- gBB3 BB- BB- BB- BB- 9 8 B+ B+ B+ B+ B+ B+ Bg+ B+ B+ B+ BH B+ B+ HR B+ (G) B+ B+ B+ B+ B+ B1 B+ B+ B+ gB1 B+ B+ B+ B+ 8 7 B B B B B B Bg B B B B B B HR B (G) B B B B B B2 B B B gB2 B B B B 7 6 B- B- B- B- B- B- Bg- B- B- B- BL B- B- HR B- (G) B- B- B- B- B- B3 B- B- B- gB3 B- B- B- B- 6 5 CCC+ CCC+ CCC+ CCC+ CCC+ CCCH CCC+ CCC+ HR C+ (G) CCC+ CCC+ CCC+ Caa1 CCC+ CCC+ CCC+ gC1 CCC+ CCC+ CCC+ 5 4 CCC CCC CCC CCC CCC CCC CCCg CCC CCC CCC CCC CCC CCC HR C (G) CCC CCC CCC CCC CCC Caa2 CCC CCC CCC gC2 CCC CCC CCC 4 3 CCC- CCC- CCC- CCC- CCC- CCCL CCC- CCC- HR C- (G) CCC- CCC- CCC- Caa3 CCC- CCC- CCC- gC3 CCC- CCC- CCC- 3 2 CC CC CC CC CC CC CCg CC CC CC CC CC CC CC CC CC CC CC Ca CC CC CC CC CC CC 2 1 C C C C C C Cg C C C C C C C C C C C C C C C C C C 1 1 RD SD DDD DDD DDD DDD RS SD SD 1 1 SD DD SD DD HR D+ (G) DD DD 1 1 D D D D D Dg D D D D D HR D (G) D D D D D D / SD D D gD D D D 1 1 RD HR D- (G) RD 1 Table 2.10: Translation of alphabetic and alphanumeric ratings onto a numeric scale. Source: company websites. 57 Country Average rating by home agencies Average rating by foreign agencies Home agency markup Brazil 11.00000 11.300000 -0.3000000 Bulgaria 13.00000 13.714286 -0.7142857 Canada 21.00000 20.900000 0.1000000 China 20.50000 17.111111 3.3888889 Cyprus 11.00000 12.500000 -1.5000000 Germany 21.00000 20.900000 0.1000000 Japan 20.50000 17.600000 2.9000000 Malaysia 16.00000 15.818182 0.1818182 Mexico 15.00000 14.727273 0.2727273 Russia 12.00000 13.888889 -1.8888889 Slovakia 17.00000 16.769231 0.2307692 Spain 16.00000 15.461538 0.5384615 Trinidad and Tobago 20.00000 12.000000 8.0000000 Turkey 12.00000 9.769231 2.2307692 USA 20.66667 20.000000 0.6666667 Bahrain - 11.000000 - Thailand - 14.454546 - Table 2.11: Difference in the assigned rating to the countries that are the domicile of at least one credit rating agency which rates sovereigns. The biggest deviation between ratings assigned by domestic and foreign agencies are seen in the case of China, Japan, Trinidad and Tobago, and Spain. 58 0.96 1 0.98 1 0.9 0.95 1 0.92 1 0.99 0.97 1 0.98 0.99 1 1 0.970.920.98 1 0.94 1 0.99 1 0.990.990.99 1 1 1 1 1 1 1 1 1 1 0.96 0.95 0.660.96 0.98 0.970.94 1 0.970.970.87 1 1 0.8 0.94 0.9 0.8 0.860.53 0.61 1 0.940.91 0.9 0.990.810.920.95 0.9 0.920.960.910.78 0.96 −1 0.910.980.97 0.980.940.960.92 0.920.970.920.52 0.88 0.990.960.970.96 0.980.960.96 0.890.890.770.990.79 −0.72 0.76 0.9 0.920.890.880.88 0.9 0.9 0.93 0.970.86 1 0.94 0.980.920.970.97 0.970.980.970.83 0.860.990.91 1 0.950.910.960.960.990.950.960.920.89 0.980.930.110.610.830.840.840.990.870.870.770.71 1 1 1 0.99 1 1 0.910.79 0.9 0.96 1 0.890.940.910.88 0.590.960.84 −1 0.71 1 0.820.940.97 0.970.960.86 0.930.880.950.890.920.910.92 0.970.980.950.980.980.94 0.950.960.96 0.9 1 1 0.99 0.930.990.94 0.97 0.9 0.91 ACRA AcraEurope Axesor BCRA CariCRIS Chengxin CI Creditreform Dagong DBRS Fareast Fitch HR JCR JCREurasia KROLL Lianhe Moody's R&I RAEX RAM S&P Scope AcraEurope CI Creditreform Dagong DBRS Fareast Fitch HR JCR JCREurasia KROLL Lianhe Moody's R&I RAEX RAM S&P Scope ShanghaiBrilliance −1.0 −0.5 0.0 0.5 1.0 Corr Figure 2.3: Pair-wise correlation coefficients between sovereign ratings from different agencies as of 2019-12-31. Most of the correlation coefficients range between 0.52 to 1. In some cases, the coefficients take on negative values. However, this is due to the exceedingly small overlap of rated countries between some pairs of agencies. 59 0.97 1 0.99 1 0.960.96 1 0.93 1 0.99 0.97 1 0.98 1 1 1 0.970.960.98 1 0.94 1 0.99 1 0.990.990.99 0.82 0.66 0.870.820.76 0.67 0.97 1 0.98 1 1 1 1 0.95 0.95 1 0.96 0.94 0.680.95 0.97 0.960.930.81 0.980.93 0.9 0.85 0.94 0.840.95 0.9 0.860.860.55 0.56 1 0.890.930.86 0.990.860.910.95 0.930.930.980.920.82 0.950.46 0.880.960.950.990.970.940.960.94 0.920.970.94 0.5 0.86 0.990.940.970.96 1 0.980.960.97 0.910.890.770.980.77 −0.84 0.69 0.9 0.93 0.9 0.750.880.91 0.9 0.92 0.970.770.990.93 0.980.870.970.970.990.940.970.970.88 0.860.990.91 1 0.970.920.960.950.990.950.960.930.89 0.970.940.090.520.820.870.860.830.880.890.760.71 1 1 0.940.98 0.99 1 0.990.920.780.890.94 1 0.870.91 0.9 0.87 0.370.850.92 0.1 0.720.57 0.850.980.97 0.980.970.94 0.910.850.950.86 0.9 0.890.91 0.970.980.950.970.980.94 0.980.930.960.960.93 0.99 1 0.980.97 0.920.980.930.27 0.970.93 0.93 ACRA AcraEurope Austin Axesor BCRA CariCRIS Chengxin CI Creditreform Dagong DBRS Fareast Fitch HR JCR JCREurasia KROLL Lianhe Moody's R&I RAEX RAM S&P Scope AcraEurope CI Creditreform Dagong DBRS Fareast Fitch HR JCR JCREurasia KROLL Lianhe Moody's R&I RAEX RAM S&P Scope ShanghaiBrilliance TRIS −1.0 −0.5 0.0 0.5 1.0 Corr Figure 2.4: Pair-wise correlation coefficients between sovereign ratings from different agencies, 1990-2019. Most of the correlation coefficients range between 0.8 to 1. In some cases, the coefficients take on negative values. However, this is due to the exceedingly small overlap of rated countries between some pairs of agencies, even over longer periods of time. 60 0 50 100 1990 2000 2010 2020 Agency ACRA AcraEurope Austin Axesor BCRA CariCRIS Chengxin CI Creditreform Dagong DBRS Fareast Fitch HR IIRA JCR JCREurasia KROLL Lianhe Moody's Pengyuan R&I RAEX RAM S&P Scope ShanghaiBrilliance TRIS Figure 2.5: Number of rated countries by agency, 1990-2019. 61 14 17 19 21 Figure 2.6: Global coverage and rating distribution by ACRA. Ratings as of 2019-12-31. 10 11 13 17 18 20 21 Figure 2.7: Global coverage and rating distribution by AcraEurope. Ratings as of 2019-12- 31. 9 10 12 13 14 16 Figure 2.8: Global coverage and rating distribution by BCRA. Ratings as of 2019-12-31. 62 9 10 20 Figure 2.9: Global coverage and rating distribution by CariCRIS. Ratings as of 2019-12-31. 6 7 10 11 15 20 21 Figure 2.10: Global coverage and rating distribution by Chengxin. Ratings as of 2019-12- 31. 3 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Figure 2.11: Global coverage and rating distribution by CI. Ratings as of 2019-12-31. 63 8 12 13 15 16 17 18 19 20 21 Figure 2.12: Global coverage and rating distribution by Creditreform. Ratings as of 2019- 12-31. 1 2 4 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Figure 2.13: Global coverage and rating distribution by Dagong. Ratings as of 2019-12-31. 2 9 12 13 14 15 16 17 18 20 21 Figure 2.14: Global coverage and rating distribution by DBRS. Ratings as of 2019-12-31. 64 4 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Figure 2.15: Global coverage and rating distribution by Fareast. Ratings as of 2019-12-31. 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Figure 2.16: Global coverage and rating distribution by Fitch. Ratings as of 2019-12-31. 10 15 21 Figure 2.17: Global coverage and rating distribution by HR. Ratings as of 2019-12-31. 65 12 13 14 15 16 17 18 19 20 21 Figure 2.18: Global coverage and rating distribution by JCR. Ratings as of 2019-12-31. 6 9 11 12 13 Figure 2.19: Global coverage and rating distribution by JCREurasia. Ratings as of 2019- 12-31. 13 15 16 17 19 20 21 Figure 2.20: Global coverage and rating distribution by KROLL. Ratings as of 2019-12-31. 66 7 9 10 11 12 13 15 16 17 18 19 20 21 Figure 2.21: Global coverage and rating distribution by Lianhe. Ratings as of 2019-12-31. 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Figure 2.22: Global coverage and rating distribution by Moody’s. Ratings as of 2019-12- 31. 6 7 9 10 11 12 21 Figure 2.23: Global coverage and rating distribution by RAEX. Ratings as of 2019-12-31. 67 7 8 9 10 11 13 14 16 17 18 19 21 Figure 2.24: Global coverage and rating distribution by RAM. Ratings as of 2019-12-31. 7 9 10 12 13 14 15 16 17 18 20 21 Figure 2.25: Global coverage and rating distribution by R&I. Ratings as of 2019-12-31. 0 2 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Figure 2.26: Global coverage and rating distribution by S&P. Ratings as of 2019-12-31. 68 9 10 12 13 14 15 16 17 19 20 21 Figure 2.27: Global coverage and rating distribution by Scope. Ratings as of 2019-12-31. 6 8 9 10 11 12 13 14 15 17 18 19 20 21 Figure 2.28: Global coverage and rating distribution by ShanghaiBrilliance. Ratings as of 2019-12-31. 69 Statistic N Mean St. Dev. Min Max Numeric rating 112,901 13.58 5.06 0 21 Investment grade dummy 112,901 0.89 0.32 0 1 Home country dummy 112,901 0.02 0.14 0 1 Export interests 100,412 1.23 3.72 0.00 76.59 Bank exposure 50,823 1.77 3.77 0.0000 39.20 Geopolitical alignment 109,674 55.42 23.58 5.97 100.00 US military interest 102,024 0.73 0.84 0.00 4.82 Common official language dummy 106,209 0.17 0.37 0.00 1.00 Home country gives swap line 112,901 0.17 0.38 0 1 China debt ratio 47,317 2.74 5.63 0.00 38.36 GDP per capita (log) 112,549 9.19 1.27 5.65 12.11 GDP growth 112,525 3.51 2.90 19.28 29.30 GDP growth (squared) 112,525 20.76 33.49 0.0000 858.31 Inflation (CPI, log) 103,630 1.10 1.09 10.07 7.41 Natural resources 112,683 5.31 9.57 0.00 73.53 Cyclically adjusted balance 70,141 2.35 2.87 13.84 6.72 Government debt ratio 109,066 53.53 36.41 0.00 236.32 Default since 1970 101,954 0.28 0.45 0.00 1.00 Default in last 5 years 101,954 0.04 0.20 0.00 1.00 Trade openness 110,583 94.62 64.86 16.22 433.05 Current account balance 109,412 0.55 7.99 44.71 42.76 External debt ratio 49,658 46.99 31.04 1.19 263.82 Rule of law 105,566 4.11 1.24 1.00 6.00 Polity 104,929 5.69 5.93 10.00 10.00 Election in last 12 months dummy 112,901 0.24 0.43 0 1 Years in office 99,125 5.78 6.63 0.00 47.00 Government orientation 99,125 0.28 0.44 0.00 1.00 Absence of internal conflict 105,566 9.46 1.49 1.94 12.00 Absence of external conflict 105,566 10.11 1.20 3.06 12.00 Absence of military in politics 105,566 4.45 1.43 0.00 6.00 Office in rated country dummy 112,901 0.19 0.39 0 1 Information transparency 47,088 67.32 10.49 24.00 88.00 At least one swapline received, dummy 112,901 0.34 0.47 0 1 Table 2.12: The descriptive statistics are for the full sample, i.e. for data from 1990 through 2019. 70 Table 2.13: Exact variable definitions and data sources. Variable Definition Source Dependent variable Sovereign rating Numeric sovereign rating on a 21-point scale Eikon Refinitiv and company websites Dyad-specific explanatory variables Home country dummy 1 if rated country is the country where the headquarters of the rating agency is located Own construction Export interests Home-country exports (in % of total home- country exports), 3-year average, lag UN Comtrade via WITS Bank exposure Overall claims of home-country banks to the rated country (in % of home country’s total foreign claims; all sectors; private and public banks; guarantees extended and credit commit- ments; all on ultimate risk basis), 1-year aver- age, lag Bank for International Settlements Geopolitical alignment V oting alignment between home country and rated country in the United Nations General Assembly (in % of total votes), 3-year average, lag V oeten (2019) US military interest Military aid provided by the United States to the rated country (in % of US total military aid), 3-year average, lag USAID Common official language dummy 1 if common official or primary language in home country and rated country CEPII Home country swap line 1 if rated country receives a swap line from the home country of the rating agency Own construction 71 Fiscal variables Government debt ratio Total central government debt in % of GDP, 3-year average, lag Worldbank World Development Indicators China debt ratio Hidden central government debt owed to China in % of GDP, 3-year average, lag Horn et al. (2019) Cyclically adjusted balance The structural budget balance refers to the general government balance cyclically adjusted for non- structural elements beyond the economic cycle. These include temporary financial sector and asset price movements as well as one-off, or temporary, revenue or expenditure items such as the sales of mobile phone licenses, 3-year average, lag IMF World Economic Outlook database GDP growth Annual percentage growth rate of GDP at market prices based on constant local currency, 3-year av- erage, lag IMF World Economic Outlook database GDP per capita GDP per capita based on purchasing power parity (PPP) in 2019 international $, 1-year average, lag IMF World Economic Outlook database Natural resources Sum of oil rents, natural gas rents, coal rents (hard and soft), mineral rents, and forest rents in % of GDP, 3-year average lag Worldbank World Development Indicators Payment history variables Historic default 1 if country has experienced a sovereign debt crisis or restructuring since 1970 Own construction based on Laeven and Valencia (2020) Recent default 1 if country has experienced a sovereign debt crisis or restructuring in the last five years Own construction based on Laeven and Valencia (2020) 72 External assessment Trade openness Sum of exports and imports in % of GDP, 3-year average, lag Worldbank World Development Indicators Current account balance Current account balance in % of GDP, 3-year aver- age, lag Worldbank World Development Indicators External debt ratio Debt owed to nonresidents repayable in foreign cur- rency, goods, or services % of GDP, 3-year average lag Worldbank World Development Indicators Monetary assessment Inflation Annual inflation in % as measured by the consumer price index, 3-year average, lag Worldbank World Development Indicators Swap line availability 1 if country has at least one swap line extended to it Own construction 73 Institutional assessment Rule of law Assessment of both the strength and impartiality of the legal system and of popular observance of the law on a 6-point scale, 1-year average, lag International Country Risk Guide Polity Regime authority on a 21-point scale ranging from -10 (hereditary monarchy) to +10 (consol- idated democracy), 3-year average, lag PolityV Center for Systemic Peace Election in last 12 months dummy 1 if presidential elections (for presidential or assembly-elected systems) or parliamentary elec- tions (for parliamentary systems) were held in rated country during the last 12 months (DPI vari- ables DATEEXEC, DATELEG and SYSTEM) Database of Political Institutions 2017 Years in office Number of years the chief executive has been in office as of January 1st, lag (DPI variable YR- SOFFC; some errors corrected), 1-year average, lag Database of Political Institutions 2017 Government orientation 1 if the chief executive’s party is defined as com- munist, socialist, social democratic or left-wing, 1- year average, lag Database of Political Institutions 2017 Absence of internal conflict Assessment of both political violence in the coun- try and its actual or potential impact on gover- nance on a 12-point scale, 1-year average, lag International Country Risk Guide Absence of external conflict Assessment of the risk to the incumbent govern- ment from foreign action on a 12-point scale, 1- year average, lag International Country Risk Guide Absence of military in politics Assessment of the degree of military participation in politics on a 6-point scale, 1-year average, lag International Country Risk Guide 74 Other variables used for robustness checks Office in rated country dummy 1 if the rating agency has an office in the rated country. Own construction based on company websites Mean bilateral trust Trust level of a representative citizen towards a randomly selected individual of the rated coun- try (based on the following Eurobarometer ques- tion on a 4-point scale: “I would like to ask you a question about how much trust you have in peo- ple from various countries. For each, please tell me whether you have a lot of trust, some trust, not very much trust, or no trust at all.”) Eurobarometer Information transparency A composite indicator similar to Transparency International’s Corruption Perceptions Index that measures the informational transparency of gov- ernments Williams (2015) 75 Building on FG, time-varying explanatory variables enter the econometric models in the form of lagged moving averages. That is, the model assumes a rating methodology of agencies that suggests that they continuously update their believes by gradually incorporat- ing new information. For instance, the “1-year average lag” refers to the moving average of each variable over the previous 12 months. Mathematically, x i;t = 1 12 1 X t=12 x i;t wherex is the variable, i is the rated country, andt is time in monthly frequency. Thus, in cases where the underlying data for an explanatory variable is only available at yearly frequency, the variable becomes a weighted average of data from the previous year and the current year where the weight of the current data gradually approaches 1. As an example, take GDP per capita of countryi in July 2019 which is only available at yearly frequency. However, within a year there will be new information about GDP dynamics. Hence, it makes sense for an agency to asses GDP per capita within a year as a weighted average of the verified information in 2018 and the updated but still unverified new information. Thus, this paper computes the moving average of the GDP (per capita) variable for July 2019 as GDP 7=2019 = 1 12 6=2019 X t=7=2018 GDP i;t = 1 2 GDP 7=2018 + 1 2 GDP 6=2019 For more volatile variables such as inflation and the fiscal balance, this paper increases the lag structure to 3 years, i.e. x i;t = 1 36 1 X t=36 x i;t This procedure cancels out short-term business-cycle and volatility effects that should not influence long-term repayment capacities (cf. Block and Vaaler, 2004). 76 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) GDP per capita (log) 1:47253 3:18989 1:29717 3:15682 (0:03028) (0:04544) (0:02910) (0:05732) GDP growth 0:00010 0:16243 0:03277 0:17087 (0:01245) (0:01491) (0:01327) (0:01825) GDP growth (squared) 0:01647 0:01251 0:01012 0:01599 (0:00133) (0:00166) (0:00136) (0:00198) Inflation (log) 0:30890 0:87325 0:38071 0:28212 (0:02119) (0:03319) (0:02050) (0:03393) Natural resources 0:01496 0:08379 0:11227 0:18393 (0:00476) (0:00664) (0:00474) (0:00800) Cyclically adjusted balance 0:25675 0:32694 0:11795 0:02219 (0:00889) (0:01309) (0:00906) (0:01359) Government debt ratio 0:02615 0:01460 0:00519 0:01058 (0:00124) (0:00175) (0:00131) (0:00204) Default since 1970 1:56385 2:14613 1:60075 1:96860 (0:03485) (0:03846) (0:03844) (0:04944) Default in last 5 years 3:33810 2:19232 3:00219 2:32232 (0:07040) (0:10618) (0:08196) (0:13940) Trade openness 0:02150 0:04016 0:01156 0:02603 (0:00075) (0:00097) (0:00071) (0:00103) Current account balance 0:00125 0:04729 0:05654 0:07424 (0:00539) (0:00720) (0:00526) (0:00792) External debt 0:07198 0:08617 0:04584 0:03853 (0:00139) (0:00196) (0:00137) (0:00197) Rule of law 0:17278 0:49812 0:08834 0:11104 (0:02071) (0:03179) (0:02086) (0:03241) Polity 0:05238 0:08859 0:05960 0:08621 (0:00361) (0:00415) (0:00388) (0:00515) Election in last 12 months 0:30574 0:19299 0:37418 0:33367 (0:03035) (0:03443) (0:03201) (0:04045) Years in office 0:02723 0:04553 0:01952 0:04125 (0:00349) (0:00375) (0:00392) (0:00488) Government orientation 0:66592 0:89952 0:94207 1:00809 (0:03370) (0:03851) (0:03784) (0:04856) Absence of internal conflict 0:31648 0:83696 0:31646 1:04693 (0:01438) (0:02041) (0:01498) (0:02545) Absence of external conflict 0:31139 1:35850 0:28794 1:07868 (0:01851) (0:03061) (0:02004) (0:03619) Absence of military in politics 0:80054 1:02252 0:71020 1:24816 (0:01538) (0:02003) (0:01745) (0:02562) Sample Fullsample GFCsample Fullsample GFCsample Observations 18,451 12,176 18,451 12,176 Adjusted R 2 0.73404 0.80601 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. The GFC sample contains data from September 2008 to December 2019. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All OLS regressions contain agency- and time-fixed effects. Standard errors are displayed in paranthesis. Table 2.14: Regression results without controlling for variables of interest. 77 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) (5) (6) (7) Home country dummy 1:70490 (0:11528) Export interests 0:13974 (0:00449) Bank exposure 0:05740 (0:01543) Geopolitical alignment 0:02274 (0:00209) US military interest 0:19747 (0:02588) Common official language 0:68616 (0:04895) Home country swap line 0:39208 (0:04042) GDP per capita (log) 1:44048 1:07661 0:65156 1:56074 1:59484 1:60914 1:44865 (0:03018) (0:03154) (0:03621) (0:03144) (0:03377) (0:03360) (0:03031) GDP growth 0:01442 0:01564 0:12012 0:01618 0:02812 0:04294 0:00196 (0:01241) (0:01241) (0:01442) (0:01250) (0:01297) (0:01266) (0:01242) GDP growth (squared) 0:01775 0:01230 0:00580 0:01862 0:01448 0:02056 0:01638 (0:00133) (0:00134) (0:00154) (0:00134) (0:00135) (0:00135) (0:00133) Inflation (log) 0:31697 0:31505 0:54140 0:28882 0:37380 0:33681 0:31586 (0:02107) (0:02067) (0:02363) (0:02123) (0:02312) (0:02116) (0:02115) Natural resources 0:00834 0:00714 0:14567 0:02235 0:01339 0:00377 0:01181 (0:00475) (0:00468) (0:00777) (0:00480) (0:00551) (0:00479) (0:00476) Cyclically adjusted balance 0:25727 0:28156 0:04099 0:25388 0:23165 0:25102 0:25815 (0:00884) (0:00871) (0:01032) (0:00887) (0:00941) (0:00889) (0:00887) Government debt ratio 0:02623 0:02515 0:00660 0:02521 0:02557 0:02648 0:02675 (0:00123) (0:00121) (0:00156) (0:00124) (0:00132) (0:00124) (0:00124) Default since 1970 1:53965 1:35855 0:19984 1:56521 1:63723 1:61711 1:55976 (0:03468) (0:03476) (0:05116) (0:03483) (0:03642) (0:03589) (0:03476) Default in last 5 years 3:32383 3:48136 2:32088 3:34325 2:94819 3:19167 3:35070 (0:06999) (0:06888) (0:12332) (0:07026) (0:07263) (0:07116) (0:07024) Trade openness 0:02201 0:02167 0:04177 0:02175 0:01835 0:02188 0:02107 (0:00074) (0:00073) (0:00087) (0:00074) (0:00081) (0:00075) (0:00074) Current account balance 0:00440 0:00556 0:06941 0:00088 0:02375 0:00942 0:00114 (0:00538) (0:00527) (0:00646) (0:00539) (0:00593) (0:00539) (0:00539) External debt 0:07199 0:06671 0:08197 0:06807 0:07640 0:07041 0:07104 (0:00138) (0:00137) (0:00154) (0:00144) (0:00146) (0:00140) (0:00139) Rule of law 0:18149 0:24354 0:18321 0:18692 0:04374 0:20661 0:19068 (0:02059) (0:02042) (0:02535) (0:02070) (0:02321) (0:02084) (0:02074) Polity 0:04815 0:02243 0:09003 0:04831 0:08185 0:05184 0:04882 (0:00360) (0:00367) (0:00498) (0:00362) (0:00417) (0:00374) (0:00362) Election in last 12 months 0:30217 0:30669 0:09210 0:28775 0:29026 0:28953 0:30550 (0:03017) (0:02985) (0:03254) (0:03032) (0:03108) (0:03052) (0:03027) Years in office 0:02595 0:03450 0:06312 0:02690 0:03000 0:03143 0:02852 (0:00347) (0:00342) (0:00574) (0:00349) (0:00373) (0:00350) (0:00349) Government orientation 0:64473 0:74443 0:58788 0:65047 0:62511 0:53558 0:72130 (0:03353) (0:03344) (0:03980) (0:03366) (0:03560) (0:03427) (0:03409) Absence of internal conflict 0:31575 0:26438 0:01908 0:31586 0:34298 0:29699 0:31004 (0:01430) (0:01409) (0:01848) (0:01436) (0:01612) (0:01434) (0:01436) Absence of external conflict 0:32252 0:18022 0:24666 0:28695 0:29074 0:35155 0:28442 (0:01841) (0:01857) (0:01965) (0:01859) (0:02138) (0:01871) (0:01867) Absence of military in politics 0:80110 0:73420 0:27675 0:81429 1:00046 0:73714 0:80015 (0:01529) (0:01522) (0:02213) (0:01540) (0:01775) (0:01594) (0:01534) Observations 18,451 17,925 10,750 18,391 16,843 17,914 18,451 Adjusted R 2 0.73720 0.73417 0.72213 0.73517 0.75359 0.72266 0.73540 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency- and time-fixed effects. Standard errors are displayed in parenthesis. Table 2.15: OLS results, full sample. 78 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) (5) (6) (7) Home country dummy 1:41153 (0:10431) Export interests 0:10421 (0:00522) Bank exposure 0:02847 (0:04064) Geopolitical alignment 0:02845 (0:00229) US military interest 0:69647 (0:03707) Common official language 0:26850 (0:05712) Home country swap line 0:28136 (0:04365) GDP per capita (log) 3:16374 2:78389 1:18488 3:10554 3:36654 3:20179 3:18036 (0:04514) (0:04830) (0:08706) (0:04582) (0:04732) (0:04836) (0:04539) GDP growth 0:15740 0:12829 0:28580 0:18899 0:20977 0:13425 0:15720 (0:01480) (0:01495) (0:01905) (0:01503) (0:01499) (0:01521) (0:01490) GDP growth (squared) 0:01298 0:01114 0:00241 0:00933 0:00523 0:01556 0:01311 (0:00165) (0:00168) (0:00201) (0:00168) (0:00162) (0:00169) (0:00166) Inflation (log) 0:86483 0:80688 0:12482 0:90504 0:79739 0:85916 0:85684 (0:03295) (0:03307) (0:04411) (0:03318) (0:03531) (0:03358) (0:03324) Natural resources 0:07388 0:03825 0:13271 0:07826 0:00750 0:05838 0:07866 (0:00663) (0:00677) (0:01412) (0:00664) (0:00740) (0:00687) (0:00668) Cyclically adjusted balance 0:33216 0:37644 0:04748 0:32027 0:28246 0:34253 0:32819 (0:01300) (0:01317) (0:01695) (0:01305) (0:01289) (0:01328) (0:01307) Government debt ratio 0:01571 0:01964 0:01635 0:01424 0:01365 0:01754 0:01562 (0:00174) (0:00176) (0:00262) (0:00175) (0:00173) (0:00180) (0:00176) Default since 1970 2:12465 1:91815 0:04456 2:14082 2:45785 2:12060 2:14021 (0:03821) (0:03938) (0:08788) (0:03835) (0:03903) (0:03943) (0:03841) Default in last 5 years 2:19789 2:65381 0:12669 2:08609 1:68618 2:24578 2:22364 (0:10538) (0:10669) (0:26295) (0:10616) (0:10608) (0:10719) (0:10611) Trade openness 0:04075 0:04008 0:05587 0:04076 0:03775 0:04173 0:03990 (0:00096) (0:00096) (0:00119) (0:00096) (0:00095) (0:00098) (0:00097) Current account balance 0:05225 0:06796 0:19103 0:03880 0:01936 0:06032 0:05185 (0:00715) (0:00715) (0:01194) (0:00719) (0:00776) (0:00730) (0:00722) External debt 0:08533 0:08322 0:12775 0:09315 0:09908 0:08573 0:08618 (0:00195) (0:00195) (0:00310) (0:00204) (0:00209) (0:00198) (0:00196) Rule of law 0:51483 0:52057 0:19810 0:48096 0:37155 0:52880 0:50375 (0:03158) (0:03170) (0:04752) (0:03167) (0:03518) (0:03272) (0:03175) Polity 0:08338 0:05956 0:19446 0:09126 0:12627 0:08090 0:08509 (0:00414) (0:00436) (0:00861) (0:00415) (0:00461) (0:00433) (0:00418) Election in last 12 months 0:19013 0:19918 0:19661 0:21325 0:21616 0:18710 0:19320 (0:03417) (0:03440) (0:03596) (0:03434) (0:03446) (0:03499) (0:03437) Years in office 0:04362 0:04812 0:00502 0:05051 0:03498 0:04722 0:04631 (0:00373) (0:00372) (0:00681) (0:00376) (0:00392) (0:00378) (0:00375) Government orientation 0:88308 0:88860 0:19855 0:92659 1:01457 0:82318 0:91853 (0:03824) (0:03833) (0:06743) (0:03843) (0:04052) (0:03936) (0:03856) Absence of internal conflict 0:83425 0:76170 0:07747 0:82794 0:72834 0:81182 0:83128 (0:02026) (0:02052) (0:03419) (0:02041) (0:02216) (0:02079) (0:02040) Absence of external conflict 1:37389 1:19706 0:53932 1:37838 1:38241 1:37433 1:33061 (0:03040) (0:03160) (0:04492) (0:03073) (0:03606) (0:03109) (0:03086) Absence of military in politics 1:01476 0:93439 0:49594 1:02449 1:30047 0:97227 1:02656 (0:01988) (0:02022) (0:03464) (0:01993) (0:02486) (0:02127) (0:02000) Observations 12,176 11,680 6,494 12,116 11,501 11,669 12,176 Adjusted R 2 0.80891 0.80003 0.78569 0.80792 0.82273 0.79350 0.80667 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency- and time-fixed effects. Standard errors are displayed in parenthesis. Table 2.16: OLS results, GFC sample. 79 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) (5) (6) (7) Home country dummy 1:38975 (0:09436) Export interests 0:14626 (0:00462) Bank exposure 0:00497 (0:01736) Geopolitical alignment 0:02135 (0:00061) US military interest 0:07555 (0:02651) Common official language 0:04762 (0:04783) Home country swap line 1:23827 (0:03982) GDP per capita (log) 1:25837 1:01350 0:62695 1:13152 1:75665 1:22247 1:23491 (0:02927) (0:02945) (0:03759) (0:02961) (0:03423) (0:03134) (0:02898) GDP growth 0:02931 0:04114 0:05401 0:01488 0:03963 0:06651 0:04174 (0:01333) (0:01357) (0:02008) (0:01360) (0:01433) (0:01358) (0:01333) GDP growth (squared) 0:00995 0:00741 0:00808 0:00775 0:01018 0:01185 0:01062 (0:00137) (0:00140) (0:00202) (0:00139) (0:00144) (0:00139) (0:00137) Inflation (log) 0:37942 0:41937 1:01911 0:35399 0:55369 0:37347 0:41880 (0:02047) (0:02065) (0:03532) (0:02050) (0:02591) (0:02045) (0:02080) Natural resources 0:11033 0:08070 0:06252 0:09236 0:10292 0:10463 0:09378 (0:00474) (0:00483) (0:01038) (0:00482) (0:00574) (0:00475) (0:00478) Cyclically adjusted balance 0:11486 0:16531 0:16986 0:14087 0:02167 0:13561 0:13536 (0:00906) (0:00909) (0:01371) (0:00926) (0:00996) (0:00914) (0:00907) Government debt ratio 0:00400 0:00959 0:03665 0:00563 0:00254 0:00752 0:00900 (0:00132) (0:00133) (0:00208) (0:00134) (0:00145) (0:00134) (0:00132) Default since 1970 1:60959 1:45395 0:72750 1:61108 1:54105 1:55828 1:59935 (0:03847) (0:03913) (0:06997) (0:03866) (0:04246) (0:03943) (0:03855) Default in last 5 years 2:99461 3:27136 1:20916 2:95860 2:75632 3:08180 3:10999 (0:08197) (0:08282) (0:18659) (0:08281) (0:08709) (0:08249) (0:08247) Trade openness 0:01217 0:01360 0:04240 0:01136 0:01372 0:01195 0:01145 (0:00072) (0:00072) (0:00117) (0:00072) (0:00082) (0:00072) (0:00072) Current account balance 0:05533 0:04507 0:10193 0:05707 0:11523 0:05588 0:04582 (0:00528) (0:00527) (0:00884) (0:00536) (0:00612) (0:00527) (0:00527) External debt 0:04618 0:04361 0:07689 0:05247 0:04182 0:04846 0:04453 (0:00137) (0:00138) (0:00211) (0:00141) (0:00145) (0:00139) (0:00137) Rule of law 0:08656 0:16319 0:07633 0:04940 0:08422 0:10532 0:14338 (0:02089) (0:02093) (0:03191) (0:02096) (0:02439) (0:02121) (0:02085) Polity 0:06096 0:02817 0:16889 0:06844 0:05257 0:04956 0:05043 (0:00388) (0:00401) (0:00767) (0:00391) (0:00467) (0:00401) (0:00391) Election in last 12 months 0:38164 0:38498 0:11042 0:36746 0:37679 0:37423 0:36324 (0:03209) (0:03228) (0:04313) (0:03225) (0:03369) (0:03233) (0:03192) Years in office 0:01871 0:03179 0:08491 0:01829 0:03338 0:02206 0:02555 (0:00393) (0:00392) (0:00798) (0:00396) (0:00432) (0:00393) (0:00391) Government orientation 0:92881 1:16326 0:75993 0:87858 0:89007 0:95273 1:17593 (0:03787) (0:03905) (0:05641) (0:03789) (0:04021) (0:03860) (0:03874) Absence of internal conflict 0:33092 0:27434 0:04080 0:29687 0:47938 0:28405 0:30433 (0:01501) (0:01480) (0:02179) (0:01489) (0:01691) (0:01507) (0:01480) Absence of external conflict 0:31055 0:11379 0:18239 0:31023 0:25738 0:26502 0:18333 (0:02020) (0:02044) (0:02640) (0:02030) (0:02440) (0:02051) (0:02022) Absence of military in politics 0:71654 0:63305 0:50536 0:73433 0:88622 0:68743 0:71966 (0:01750) (0:01763) (0:03081) (0:01769) (0:02091) (0:01809) (0:01756) Observations 18,451 17,925 10,750 18,391 16,843 17,914 18,451 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. Standard errors are displayed in parenthesis. Table 2.17: Probit results, full sample. 80 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) (5) (6) (7) Home country dummy 1:23119 (0:09071) Export interests 0:09342 (0:00589) Bank exposure 0:57427 (0:05961) Geopolitical alignment 0:02737 (0:00078) US military interest 0:22536 (0:04570) Common official language 0:76868 (0:06387) Home country swap line 1:16383 (0:04925) GDP per capita (log) 3:11000 2:75020 1:31474 2:96293 3:42956 2:84228 3:07552 (0:05749) (0:06023) (0:12759) (0:05793) (0:06160) (0:05959) (0:05719) GDP growth 0:17376 0:13605 0:28949 0:20559 0:21528 0:15095 0:15025 (0:01832) (0:01846) (0:03082) (0:01866) (0:01916) (0:01870) (0:01822) GDP growth (squared) 0:01605 0:01435 0:01369 0:01232 0:01204 0:01766 0:01782 (0:00199) (0:00202) (0:00317) (0:00204) (0:00201) (0:00203) (0:00198) Inflation (log) 0:27007 0:19738 0:94837 0:25708 0:19784 0:26970 0:20279 (0:03395) (0:03407) (0:07253) (0:03424) (0:03960) (0:03405) (0:03395) Natural resources 0:17901 0:14020 0:02618 0:14928 0:17313 0:17186 0:15420 (0:00802) (0:00816) (0:02229) (0:00810) (0:00884) (0:00812) (0:00804) Cyclically adjusted balance 0:02609 0:08123 0:37879 0:05820 0:03401 0:04080 0:04780 (0:01360) (0:01395) (0:02800) (0:01386) (0:01433) (0:01385) (0:01366) Government debt ratio 0:01083 0:00254 0:05488 0:00737 0:01530 0:00777 0:00451 (0:00205) (0:00208) (0:00396) (0:00210) (0:00211) (0:00210) (0:00205) Default since 1970 1:98629 1:76693 0:58965 2:01370 2:18876 1:82707 1:94890 (0:04955) (0:05039) (0:14090) (0:04986) (0:05347) (0:05025) (0:04927) Default in last 5 years 2:34869 2:85322 1:76991 2:24382 2:12647 2:60812 2:58963 (0:13992) (0:14217) (0:61996) (0:14145) (0:14457) (0:14061) (0:14073) Trade openness 0:02660 0:02530 0:04878 0:02714 0:02423 0:02665 0:02553 (0:00103) (0:00103) (0:00170) (0:00105) (0:00109) (0:00104) (0:00103) Current account balance 0:07091 0:05771 0:25883 0:06193 0:12021 0:07328 0:04861 (0:00793) (0:00796) (0:01925) (0:00802) (0:00900) (0:00793) (0:00793) External debt 0:03848 0:03828 0:14378 0:05278 0:03624 0:04216 0:04114 (0:00197) (0:00198) (0:00518) (0:00206) (0:00211) (0:00199) (0:00197) Rule of law 0:10885 0:08486 0:69832 0:18211 0:12536 0:18857 0:07638 (0:03246) (0:03279) (0:06666) (0:03281) (0:04148) (0:03363) (0:03234) Polity 0:08821 0:05787 0:34118 0:09644 0:09959 0:06738 0:07581 (0:00515) (0:00534) (0:01551) (0:00518) (0:00611) (0:00528) (0:00518) Election in last 12 months 0:34186 0:35486 0:19102 0:34492 0:33165 0:34891 0:32444 (0:04050) (0:04098) (0:05782) (0:04077) (0:04197) (0:04103) (0:04041) Years in office 0:03969 0:04417 0:05147 0:04400 0:04140 0:03801 0:04408 (0:00489) (0:00487) (0:01110) (0:00489) (0:00526) (0:00489) (0:00486) Government orientation 1:01338 1:08574 0:01379 0:95644 1:11081 1:08690 1:12062 (0:04857) (0:04977) (0:11322) (0:04802) (0:05197) (0:04950) (0:04912) Absence of internal conflict 1:05204 0:92951 0:13764 0:96610 1:08819 1:01845 0:99197 (0:02546) (0:02569) (0:05335) (0:02565) (0:02792) (0:02593) (0:02534) Absence of external conflict 1:10471 0:86923 0:28641 1:01612 1:19819 0:97313 0:92644 (0:03636) (0:03807) (0:06741) (0:03695) (0:04336) (0:03704) (0:03650) Absence of military in politics 1:24874 1:14180 1:13126 1:31132 1:34077 1:30555 1:26640 (0:02562) (0:02611) (0:05634) (0:02591) (0:03355) (0:02746) (0:02565) Observations 12,176 11,680 6,494 12,116 11,501 11,669 12,176 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. Standard errors are displayed in parenthesis. Table 2.18: Probit results, GFC sample. 81 Dependent variable: Numeric sovereign rating OLS OLS OLS OLS OLS OLS (1) (2) (3) (4) (5) (6) Home country dummy 1:70490 1:41153 2:31882 1:06238 (0:11528) (0:10431) (0:12038) (0:11418) Geopolitical alignment 0:07306 0:03434 0:03566 0:01896 (0:00258) (0:00345) (0:00218) (0:00250) Export interest 0:16791 0:09788 (0:00599) (0:00689) Bank exposure 0:23063 0:12788 (0:01547) (0:04481) Home country swap line 0:58529 0:23534 (0:05640) (0:06837) GDP per capita (log) 1:44048 3:16374 0:77598 1:09162 1:56978 3:11297 (0:03018) (0:04514) (0:03719) (0:08596) (0:03113) (0:04566) GDP growth 0:01442 0:15740 0:14877 0:26002 0:04298 0:17602 (0:01241) (0:01480) (0:01322) (0:01852) (0:01245) (0:01504) GDP growth (squared) 0:01775 0:01298 0:00162 0:00347 0:02151 0:01077 (0:00133) (0:00165) (0:00144) (0:00197) (0:00134) (0:00168) Inflation (log) 0:31697 0:86483 0:44356 0:10160 0:28840 0:88959 (0:02107) (0:03295) (0:02166) (0:04298) (0:02101) (0:03311) Natural resources 0:00834 0:07388 0:12275 0:12515 0:01782 0:07266 (0:00475) (0:00663) (0:00711) (0:01382) (0:00476) (0:00664) Cyclically adjusted balance 0:25727 0:33216 0:01521 0:00019 0:25267 0:32626 (0:00884) (0:01300) (0:00957) (0:01660) (0:00879) (0:01302) Government debt ratio 0:02623 0:01571 0:00920 0:01269 0:02481 0:01512 (0:00123) (0:00174) (0:00146) (0:00270) (0:00123) (0:00174) Default since 1970 1:53965 2:12465 0:09920 0:13109 1:53438 2:12417 (0:03468) (0:03821) (0:04807) (0:08776) (0:03452) (0:03826) Default in last 5 years 3:32383 2:19789 1:92109 0:93964 3:32395 2:12910 (0:06999) (0:10538) (0:11465) (0:25719) (0:06956) (0:10588) Trade openness 0:02201 0:04075 0:04049 0:05135 0:02258 0:04102 (0:00074) (0:00096) (0:00080) (0:00118) (0:00074) (0:00096) Current account balance 0:00440 0:05225 0:03585 0:16985 0:00741 0:04518 (0:00538) (0:00715) (0:00603) (0:01210) (0:00535) (0:00720) External debt 0:07199 0:08533 0:06473 0:11071 0:06571 0:09015 (0:00138) (0:00195) (0:00147) (0:00312) (0:00143) (0:00206) Rule of law 0:18149 0:51483 0:31885 0:25182 0:20646 0:49862 (0:02059) (0:03158) (0:02342) (0:04601) (0:02051) (0:03161) Polity 0:04815 0:08338 0:02947 0:15398 0:04073 0:08641 (0:00360) (0:00414) (0:00482) (0:00917) (0:00361) (0:00417) Election in last 12 months 0:30217 0:19013 0:05804 0:17181 0:27392 0:20337 (0:03017) (0:03417) (0:02970) (0:03474) (0:03003) (0:03423) Years in office 0:02595 0:04362 0:05861 0:01157 0:02462 0:04747 (0:00347) (0:00373) (0:00524) (0:00662) (0:00346) (0:00376) Government orientation 0:64473 0:88308 0:82264 0:34962 0:61504 0:90419 (0:03353) (0:03824) (0:03751) (0:06945) (0:03337) (0:03837) Absence of internal conflict 0:31575 0:83425 0:00768 0:07237 0:31563 0:82920 (0:01430) (0:02026) (0:01689) (0:03308) (0:01421) (0:02034) Absence of external conflict 0:32252 1:37389 0:05041 0:38998 0:29066 1:38302 (0:01841) (0:03040) (0:01844) (0:04392) (0:01840) (0:03062) Absence of military in politics 0:80110 1:01476 0:18060 0:41618 0:82277 1:01792 (0:01529) (0:01988) (0:02049) (0:03369) (0:01525) (0:01988) Sample Fullsample GFCsample Fullsample GFCsample Fullsample GFCsample Observations 18,451 12,176 10,750 6,494 18,391 12,116 Adjusted R 2 0.73720 0.80891 0.76895 0.80057 0.74049 0.80928 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency- and time-fixed effects. Standard errors are displayed in parenthesis. Table 2.19: Full results when adding all variables of interest simultaneously. 82 Agency AcraEurope BCRA Chengxin CI Creditreform Dagong DBRS Fitch HR JCR JCREurasia KROLL Lianhe Moody’s R&I RAEX RAM S&P Scope ShanghaiBrilliance Home country dummy NA (NA) NA (NA) NA (NA) 0.8336*** (0) -0.2865 (0.4073) 9.2386*** (0) 0.1133 (0.5733) 3.5925*** (0) NA (NA) 4.7759*** (0) NA (NA) NA (NA) 4.7241*** (0) NA (NA) 3.305*** (0) 0.8897 (0.5859) -1.4027*** (0) 2.9753*** (0) -0.9364** (0.0493) 4.7341*** (0) Export interest -0.1049 (NaN) NA (NA) 1.2802** (0.0001) 0.0937** (0) 0.0454 (0.1586) 0.1146** (0) 0.0373** (0) 0.1115** (0) NA (NA) 0.1956** (0) NA (NA) NA (NA) NA (NA) NA (NA) 0.1713** (0) -1.3157** (0.0007) 0.4055** (0) 0.1693** (0) 0.1968** (0) 0.18** (0) Geopolitical alignment 0.0168 (NaN) NaN (NaN) -0.124*** (0) -0.0884*** (0) NA (NA) 0.1057*** (0) -0.0584*** (0) 0.0126*** (0) 0.0604 (0.9622) -0.06*** (0) -1.827*** (0.0001) 0.0143*** (0) 0.0001 (0.9851) -0.0142*** (0) -0.1032*** (0) -0.0403 (0.2321) -0.0649*** (0.0062) 0.001 (0.5688) NA (NA) 0.0216 (0.1751) Common official language NA (NA) NA (NA) NA (NA) 2.172*** (0) -1.0643*** (0) 7.2648*** (0) 1.9287*** (0) 1.2652*** (0) NA (NA) NA (NA) NA (NA) NA (NA) 2.4431*** (0) 1.5267*** (0) NA (NA) NA (NA) -4.0875*** (0) 1.3413*** (0) 0.5019* (0.0759) 3.4585*** (0.0012) Table 2.20: Coefficients of variables of interest for agency-individual regressions. All regressions contain political and economic variables as well as the variables of interest indicated above. Neither all agencies nor all variables of interest are shown due to data limitations that don’t allow running all agency-specific regressions. 83 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency CI Dagong Fitch JCREurasia Lianhe Moody's RAM S&P HomeCountryDummy 0 Bahrain Figure 2.29: Temporal variance of ratings by different agencies for home country Bahrain. 0 5 10 15 20 2000 2010 2020 Numeric Rating HomeCountryDummy 0 1 Agency Austin Dagong DBRS Fareast Fitch JCR Lianhe Moody's R&I S&P ShanghaiBrilliance Brazil Figure 2.30: Temporal variance of ratings by different agencies for home country Brazil. 84 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating HomeCountryDummy 0 1 Agency BCRA Dagong Fareast Fitch JCR Lianhe Moody's S&P Scope Bulgaria Figure 2.31: Temporal variance of ratings by different agencies for home country Bulgaria. 0 5 10 15 20 2000 2010 2020 Numeric Rating HomeCountryDummy 0 1 Agency Dagong DBRS Fareast Fitch HR JCR KROLL Lianhe Moody's R&I S&P ShanghaiBrilliance Canada Figure 2.32: Temporal variance of ratings by different agencies for home country Canada. 85 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency Chengxin CI Dagong DBRS Fareast Fitch JCR Lianhe Moody's Pengyuan R&I RAEX RAM S&P Scope ShanghaiBrilliance HomeCountryDummy 0 1 China Figure 2.33: Temporal variance of ratings by different agencies for home country China. 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency CI Creditreform DBRS Fitch Moody's RAEX S&P Scope HomeCountryDummy 0 1 Cyprus Figure 2.34: Temporal variance of ratings by different agencies for home country Cyprus. 86 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency AcraEurope Creditreform Dagong DBRS Fareast Fitch JCR Lianhe R&I RAEX Scope ShanghaiBrilliance HomeCountryDummy 0 1 Germany Figure 2.35: Temporal variance of ratings by different agencies for home country Germany. 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency Dagong DBRS Fareast Fitch JCR Lianhe Moody's R&I RAM S&P Scope ShanghaiBrilliance HomeCountryDummy 0 1 Japan Figure 2.36: Temporal variance of ratings by different agencies for home country Japan. 87 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency Chengxin CI Dagong Fareast Fitch JCR Lianhe Moody's R&I RAM S&P ShanghaiBrilliance HomeCountryDummy 0 1 Malaysia Figure 2.37: Temporal variance of ratings by different agencies for home country Malaysia. 0 5 10 15 20 1990 2000 2010 2020 Numeric Rating HomeCountryDummy 0 1 Agency Dagong DBRS Fareast Fitch HR JCR KROLL Lianhe Moody's R&I S&P ShanghaiBrilliance Mexico Figure 2.38: Temporal variance of ratings by different agencies for home country Mexico. 88 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency Dagong Fareast Fitch JCR Lianhe Moody's RAEX S&P Scope ShanghaiBrilliance HomeCountryDummy 0 1 Russia Figure 2.39: Temporal variance of ratings by different agencies for home country Russia. 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency ACRA AcraEurope CI Creditreform DBRS Fareast Fitch JCR Lianhe Moody's R&I S&P Scope ShanghaiBrilliance HomeCountryDummy 0 Slovakia Figure 2.40: Temporal variance of ratings by different agencies for home country Slovakia. 89 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency Axesor Chengxin Creditreform Dagong DBRS Fareast Fitch JCR KROLL Moody's R&I S&P Scope ShanghaiBrilliance HomeCountryDummy 0 Spain Figure 2.41: Temporal variance of ratings by different agencies for home country Spain. 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency CI Dagong Fareast Fitch JCR Lianhe Moody's R&I RAM S&P ShanghaiBrilliance HomeCountryDummy 0 Thailand Figure 2.42: Temporal variance of ratings by different agencies for home country Thailand. 90 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency CariCRIS Moody's S&P HomeCountryDummy 0 1 Trinidad and Tobago Figure 2.43: Temporal variance of ratings by different agencies for home country Trinidad and Tobago. 0 5 10 15 20 2000 2010 2020 Numeric Rating Agency AcraEurope CI Dagong DBRS Fareast Fitch JCR JCREurasia Lianhe Moody's R&I RAM S&P Scope ShanghaiBrilliance HomeCountryDummy 0 Turkey Figure 2.44: Temporal variance of ratings by different agencies for home country Turkey. 91 0 5 10 15 20 2000 2005 2010 2015 2020 Numeric Rating Agency Creditreform Dagong DBRS Fareast Fitch JCR KROLL Lianhe R&I RAEX S&P Scope ShanghaiBrilliance HomeCountryDummy 0 USA Figure 2.45: Temporal variance of ratings by different agencies for home country USA. 92 Agency Office locations ACRA Russia, Slovakia AcraEurope Slovakia, Russia Austin Brasil Axesor Chile, Peru, Colombia, Spain, Ireland, Portugal BCRA Bulgaria CariCRIS Trinidad and Tobago Chengxin Hong Kong, China CI India, Cyprus, Hong Kong, Germany Creditreform Germany Dagong China, Italy, Germany DBRS Canada, USA, UK, Germany, Madrid Fareast China Fitch Thailand, China, Sri Lanka, Hong Kong, Indonesia, India, South Korea, China, Singapore, Australia, Taiwan, Japan, Spain, Ger- many, Italy, Russia, France, UK, Sweden, Poland, Colombia, Guatemala, Mexico, Panama, Costa Rica, Chile, Brazil, El Sal- vador, Dominican Republic HR Mexico, USA IIRA Bahrain, Kazakhstan JCR Japan, Turkey JCREurasia Turkey, Japan KROLL USA, UK, Ireland Lianhe China, Hong Kong Moody’s USA, Argentina, Mexico, Brazil, Canada, UAE, Germany, South Africa, Cyprus, UK, Spain, Italy, Russia, France, Czech Republic, Saudi Arabia, Poland, Sweden, China, Hong Kong, India, South Korea, Singapore, Australia, Japan Pengyuan China, Hong Kong R&I Hong Kong, Japan RAEX Russia, Kazakhstan, Belarus, Germany and Hong Kong RAM Malaysia S&P Germany, UK, France, Spain, Norway, Italy, USA Scope China ShanghaiBrilliance USA, Colombia, Argentina, Mexico, Brazil, Canada, India, China, Hong Kong, Pakistan, Philippines, Australia, South Korea, Aus- tralia, Taiwan, Japan, Netherlands, UAE, Ireland, Germany, South Africa, UK, Spain, Italy, Russia, France, Israel, Sweden, Poland, Switzerland TRIS Thailand Table 2.21: Office locations of the credit rating agencies as of 2019-12-31. Source: com- pany websites. 93 Dependent variable: Numeric sovereign rating GDP per capita (log) 1:55574 (0:03434) GDP growth 0:04089 (0:01265) GDP growth (squared) 0:02019 (0:00135) Inflation (log) 0:32394 (0:02120) Natural resources 0:00065 (0:00480) Cyclically adjusted balance 0:24998 (0:00893) Government debt ratio 0:02440 (0:00126) Default since 1970 1:56776 (0:03713) Default in last 5 years 3:19512 (0:07106) Trade openness 0:02203 (0:00076) Current account balance 0:01048 (0:00539) External debt 0:06894 (0:00142) Rule of law 0:19128 (0:02115) Polity 0:05673 (0:00378) Election in last 12 months 0:29135 (0:03045) Years in office 0:03210 (0:00349) Government orientation 0:51032 (0:03464) Absence of internal conflict 0:29711 (0:01432) Absence of external conflict 0:36660 (0:01875) Absence of military in politics 0:73062 (0:01620) Common official language 0:38996 (0:06623) Office in rated country 0:14240 (0:04101) Common official language X Office in rated country 0:47086 (0:08258) Observations 17,914 Adjusted R 2 0.72392 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency fixed effects and time fixed effects. Standard errors are displayed in parenthesis. Table 2.22: Examination of transmission channels for home bias. The sample consists of pooled data from all agencies, 1990-2019. Note that data on offices in rated countries represents information as of 2019-12-31. 94 Dependent variable: Numeric sovereign rating (1) (2) (3) GDP per capita (log) 1:60914 0:90943 0:87956 (0:03360) (0:04464) (0:04484) GDP growth 0:04294 0:49282 0:47852 (0:01266) (0:02008) (0:02019) GDP growth (squared) 0:02056 0:05720 0:05577 (0:00135) (0:00189) (0:00190) Inflation (log) 0:33681 0:68439 0:68141 (0:02116) (0:02361) (0:02357) Natural resources 0:00377 0:13440 0:13098 (0:00479) (0:00593) (0:00595) Cyclically adjusted balance 0:25102 0:15199 0:14553 (0:00889) (0:01086) (0:01089) Government debt ratio 0:02648 0:03663 0:03408 (0:00124) (0:00153) (0:00159) Default since 1970 1:61711 1:38791 1:30943 (0:03589) (0:05559) (0:05709) Default in last 5 years 3:19167 2:59952 2:60222 (0:07116) (0:07609) (0:07593) Trade openness 0:02188 0:03458 0:03557 (0:00075) (0:00093) (0:00095) Current account balance 0:00942 0:05726 0:05975 (0:00539) (0:00670) (0:00670) External debt 0:07041 0:10032 0:09717 (0:00140) (0:00188) (0:00195) Rule of law 0:20661 0:45792 0:46637 (0:02084) (0:02611) (0:02610) Polity 0:05184 0:02479 0:02594 (0:00374) (0:00527) (0:00526) Election in last 12 months 0:28953 0:07360 0:08032 (0:03052) (0:03524) (0:03519) Years in office 0:03143 0:00728 0:00388 (0:00350) (0:00485) (0:00488) Government orientation 0:53558 0:01608 0:02505 (0:03427) (0:04631) (0:04675) Absence of internal conflict 0:29699 0:19077 0:20123 (0:01434) (0:01747) (0:01752) Absence of external conflict 0:35155 0:15626 0:15491 (0:01871) (0:02549) (0:02544) Absence of military in politics 0:73714 0:33965 0:32702 (0:01594) (0:02009) (0:02017) Common official language 0:68616 0:77438 0:72669 (0:04895) (0:06159) (0:06200) Information transparency 0:07125 0:06975 (0:00446) (0:00446) Office in rated country 0:25650 (0:04398) Observations 17,914 8,251 8,251 Adjusted R 2 0.72266 0.76837 0.76932 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All regressions contain agency fixed effects and time fixed effects. Standard errors are displayed in parenthesis. Table 2.23: Further examination of transmission channels for home bias. The sample consists of pooled data from all agencies, 1990-2019. Note that data on offices in rated countries represents information as of 2019-12-31. 95 Dependent variable: Numeric sovereign rating (1) (2) (3) (4) (5) Home country dummy 1:15639 (0:17481) Geopolitical alignment 0:05458 0:05710 0:02780 (0:00235) (0:00236) (0:00216) Export interest 0:13776 0:13259 (0:00448) (0:00440) GDP per capita (log) 1:43744 1:66230 1:07043 1:31146 1:57445 (0:03018) (0:03101) (0:03142) (0:03239) (0:03140) GDP growth 0:01715 0:05117 0:01184 0:05363 0:02286 (0:01242) (0:01230) (0:01237) (0:01228) (0:01249) GDP growth (squared) 0:01801 0:02420 0:01213 0:01940 0:01949 (0:00133) (0:00133) (0:00133) (0:00134) (0:00134) Inflation (log) 0:31583 0:25210 0:31126 0:25355 0:27883 (0:02106) (0:02083) (0:02059) (0:02032) (0:02120) Natural resources 0:00744 0:02868 0:00921 0:01126 0:02431 (0:00475) (0:00471) (0:00466) (0:00464) (0:00480) Cyclically adjusted balance 0:25623 0:25544 0:27670 0:27610 0:25651 (0:00884) (0:00869) (0:00868) (0:00853) (0:00886) Government debt ratio 0:02605 0:02442 0:02457 0:02343 0:02519 (0:00123) (0:00121) (0:00121) (0:00119) (0:00124) Default since 1970 1:51927 1:42285 1:33334 1:27596 1:54360 (0:03501) (0:03450) (0:03469) (0:03417) (0:03483) Default in last 5 years 3:34322 3:39126 3:50834 3:49896 3:36564 (0:07011) (0:06883) (0:06864) (0:06736) (0:07014) Trade openness 0:02221 0:02316 0:02205 0:02301 0:02174 (0:00074) (0:00073) (0:00073) (0:00071) (0:00074) Current account balance 0:00384 0:00334 0:01150 0:01507 0:00134 (0:00538) (0:00528) (0:00527) (0:00518) (0:00537) External debt 0:07236 0:06131 0:06515 0:05326 0:06705 (0:00138) (0:00143) (0:00137) (0:00143) (0:00144) Rule of law 0:18280 0:21559 0:21675 0:24692 0:19255 (0:02059) (0:02030) (0:02046) (0:02011) (0:02066) Polity 0:04554 0:03270 0:02836 0:01873 0:04619 (0:00365) (0:00359) (0:00369) (0:00365) (0:00362) Election in last 12 months 0:30500 0:27576 0:30642 0:26675 0:28481 (0:03016) (0:02970) (0:02973) (0:02922) (0:03025) Years in office 0:02669 0:02379 0:03416 0:02901 0:02592 (0:00348) (0:00342) (0:00341) (0:00335) (0:00348) Government orientation 0:63027 0:52653 0:77751 0:67640 0:62667 (0:03369) (0:03326) (0:03342) (0:03300) (0:03367) Absence of internal conflict 0:31465 0:31946 0:25948 0:26765 0:31476 (0:01429) (0:01406) (0:01404) (0:01378) (0:01432) Absence of external conflict 0:31942 0:22585 0:17912 0:10013 0:28152 (0:01842) (0:01834) (0:01850) (0:01839) (0:01855) Absence of military in politics 0:79462 0:80921 0:72773 0:76111 0:81267 (0:01536) (0:01508) (0:01517) (0:01495) (0:01536) Home country dummy:Chinese agency dummy 0:98124 (0:23518) Geopolitical alignment:Chinese agency dummy 0:19496 0:19187 (0:00703) (0:00929) Export interest:Chinese agency dummy 0:82661 0:80275 (0:06969) (0:06864) Geopolitical alignment:Chinese government owned agency 0:11387 (0:01216) Observations 18,451 18,391 17,925 17,865 18,391 Adjusted R 2 0.73744 0.74599 0.73626 0.74607 0.73644 Note: The dependent variable is a country’s sovereign rating on a numeric scale from 0-21. *,**,*** correspond to 10%, 5% and 1% significance, respectively. All OLS regressions contain agency- and time-fixed effects. Standard errors are displayed in parenthesis. Table 2.24: Home bias and in-group bias for Chinese agencies. Data ranges from 1990- 2019. Standard errors are clustered at both the agency-time and sovereign level. 96 Sample Variable 20 Percentile Investment grade threshold 40 Percentile 60 Percentile US pension fund threshold 80 Percentile Money market fund threshold Full Home country dummy 0.8804*** (0.0000) 0.5575*** (0.0000) 0.2892* (0.0500) 2.3231*** (0.0000) 2.3129*** (0.0000) 3.6416*** (0.0000) 3.4826*** (0.0000) Full Export interest -0.0017 (0.9065) -0.0192*** (0.0021) -0.0299*** (0.0001) -0.0254* (0.0993) -0.0193 (0.2939) 0.1141*** (0.0021) 0.1382*** (0.0000) Full Bank exposure 0.1497*** (0.0000) 0.1391*** (0.0000) 0.1323*** (0.0000) 0.1244*** (0.0000) 0.1216*** (0.0000) 0.1407*** (0.0000) 0.1383*** (0.0000) Full Geopolitical alignment 0.0142*** (0.0000) 0.0186*** (0.0000) 0.0196*** (0.0000) 0.0224*** (0.0000) 0.0222*** (0.0000) 0.0292*** (0.0000) 0.0305*** (0.0000) Full US military interest -0.0279 (0.2245) 0.0341 (0.1134) 0.0068 (0.8130) -0.2826*** (0.0000) -0.3214*** (0.0000) -0.3541*** (0.0000) -0.3532*** (0.0000) Full Common official language -0.02 (0.6403) -0.106* (0.0840) -0.0523 (0.3453) -0.1193*** (0.0006) -0.1909*** (0.0000) -0.4556*** (0.0000) -0.5285*** (0.0000) Full Home country swap line 1.4563*** (0.0000) 1.4004*** (0.0000) 1.3968*** (0.0000) 1.0251*** (0.0000) 0.9867*** (0.0000) 0.8956*** (0.0000) 0.7417*** (0.0000) GFC Home country dummy 0.7159*** (0.0000) 0.9741*** (0.0000) 0.9946*** (0.0000) 1.7727*** (0.0000) 1.8174*** (0.0000) 2.6722*** (0.0000) 2.6346*** (0.0000) GFC Export interest -0.4209*** (0.0000) -0.4894*** (0.0000) -0.4817*** (0.0000) -0.2013*** (0.0000) -0.1702*** (0.0000) -0.0919** (0.0127) -0.0637* (0.0856) GFC Bank exposure 0.0907*** (0.0000) 0.0759*** (0.0000) 0.0771*** (0.0000) 0.078*** (0.0000) 0.0761*** (0.0000) 0.072*** (0.0000) 0.0706*** (0.0000) GFC Geopolitical alignment 0.0193*** (0.0000) 0.0224*** (0.0000) 0.0238*** (0.0000) 0.0258*** (0.0000) 0.0264*** (0.0000) 0.0307*** (0.0000) 0.0325*** (0.0000) GFC US military interest -0.1277*** (0.0077) 0.0019 (0.9689) 0.0561 (0.1759) 0.1695*** (0.0000) 0.162*** (0.0000) 0.3855*** (0.0000) 0.1935* (0.0580) GFC Common official language -0.2342*** (0.0000) -0.9352*** (0.0000) -1.0244*** (0.0000) -0.8592*** (0.0000) -0.8256*** (0.0000) -0.8854*** (0.0000) -0.9013*** (0.0000) GFC Home country swap line 1.0354*** (0.0000) 1.1284*** (0.0000) 1.109*** (0.0000) 1.013*** (0.0000) 1.0203*** (0.0000) 0.9628*** (0.0000) 0.9518*** (0.0000) Table 2.25: Biased sovereign ratings at important thresholds and by quintile. All agencies pooled. Full sample; 01/1990-12/2019. GFC sample; 09/2008-12/2019. The dependent variable is a country’s numeric rating. Each cell refers to a separate regression. The table shows only the coefficients on the respective variable of interest for each regression. All regressions contain the control variables from the baseline specification, excluding other variables of interest, as well as time-fixed effects and agency-fixed effects. Standard errors are clustered at both the agency-time and sovereign level. To account for important rating thresholds, the paper calculates the percentiles of the dependent variable (numeric rating) within the regression sample and and selects the closest percentile that is located in between the rating categories around the respective threshold. *,**,*** correspond to 10%, 5% and 1% significance, respectively. P-values are displayed in parentheses. Money market fund threshold; AA+/AA. US pension fund threshold; A/A-. Investment-grade threshold; BBB-/BB+. 97 CHAPTER 3 IMPLICATIONS OF BIASED SOVEREIGN RATINGS IN THE IPE Abstract What are the implications of biased sovereign ratings? This paper explores whether the documented home and in-group bias have real economic effects on countries. To this end, the paper devises a theoretical model that shows how sovereign ratings bear the potential to affect government debt dynamics through the process of self-fulfilling prophecy. In the next step, the paper probes whether ratings from all agencies have the capacity to trigger this process. Based on regulatory and practical considerations, the paper arrives at the conclusion that only the three US agencies Fitch, Moody’s, and S&P have the capacity to influence interest rates of a large number of countries. Moreover, the paper provides examples which show that the impact of biased ratings can be quite substantial and exceed 1% of GDP. 98 3.1 Introduction A Credit Rating Agency (CRA) is a corporation that assesses the riskiness and creditwor- thiness of corporations, sovereigns, and structured financial products. The final result of such an assessment comes in the form of a “rating.” It is typically expressed on either an alphabetic or alphanumeric scale that reaches from “AAA” to “D.” The former indicates a very high creditworthiness whereas the latter indicates an issuer’s default. While there are roughly 150 CRAs worldwide, most of them do not have global cover- age. That is, most CRAs limit their activity to local or regional markets and specialize par- ticularly on rating corporate issuers in those markets. Less than 30 CRAs provide sovereign ratings, which are also called “sovereign debt ratings” or “sovereign credit ratings.” Of the agencies which rate sovereigns only a subset engages in rating sovereign issuers at a global scale. The rest have relatively narrow coverage of sovereign issuers. 1 Ratings play a crucial role in financial markets. They provide private and institutional investors as well as regulators with an educated guess—but a guess nonetheless—about the default probability of a certain issuer relative to other issuers. 2 That is, ratings are an opin- ion on the relative ability of a debt issuer to meet financial commitments, such as interest payments, preferred dividends, repayment of principal, insurance claims or counter-party obligations. Importantly, credit rating agencies claim that their sovereign ratings are based on coun- tries’ economic and political variables. However, it is challenging to reverse-engineer rat- ings based on these variables alone and to see if there is evidence that internal biases of and external incentives for credit rating agencies warp objective ratings. 3 Specifically, research on credit ratings points out that: 1 See Table 3.1 for an overview of the number of rated sovereigns by agency as of 2019-12-31. 2 “[C]redit ratings express risk in relative rank order, which is to say they are ordinal measures of credit risk and are not predictive of a specific frequency of default or loss.” This definition is from Fitch Ratings (2018, p. 3). However, the definitions of the other agencies are nearly identical. 3 Refer to chapter 2 for an overview of factors driving sovereign ratings which are not related to the rated countries’ economic and political fundamentals. 99 • Credit rating agencies have a home bias and assign their home country a favorable rating which is above what would be expected based on political and economic fun- damentals. • The home bias is magnified among those agencies which experience interference by their home country’s authoritarian government. • Credit rating agencies assign favorable ratings to countries with strong economic ties to their respective home country. • Credit rating agencies, particularly in authoritarian states, have an in-group bias and assign favorable ratings to countries which are geopolitically aligned with their home country. Ostensibly, showing if and how ratings are biased is a relatively narrow remit. However, CRAs rate sovereign debt worth more than $50 trillion so that even small biases could potentially have enormous consequences for the global allocation of capital and the relative distribution of power in the global system. It is therefore crucial to understand how ratings can affect government debt dynamics. In light of this, this paper explores the implications of biased sovereign ratings at both a theoretical and empirical level. The rest of this paper is structured as follows: Section 3.2 provides a literature review on the effects of sovereign credit ratings. Section 3.3 presents a theoretical model that explores the interaction between sovereign credit ratings, default probabilities, and interest payments. Section 3.4 provides rough estimates of the fiscal implications of biased ratings for affected governments. Section 3.5 concludes. 3.2 Literature review There is an extensive literature analyzing the determinants of sovereign credit ratings and the behavior of credit rating agencies more generally. 4 As an outgrowth of this literature, there are several studies that address the effect of sovereign ratings. These studies broadly 4 For examples, see the survey by De Haan and Amtenbrink (2011) or the overview provided in section 2.2. 100 fall into three categories: 1) Studies examining the immediate impact of sovereign ratings on the interest paid by governments; 2) Studies analyzing the impact of sovereign ratings on the interest paid by companies in the rated country and on the behaviour of domestic invest- ment; 3) Studies analyzing spillover effects of sovereign ratings in sovereign debt markets. In what follows, major studies of each category are presented to gain an understanding of the existing research on the effects of sovereign ratings. Literature on the impact of sovereign ratings on sovereign yields Although sovereign ratings are clearly correlated with yields on sovereign debt, it is not obvious that ratings actually influence yields. However, Cantor, Packer, et al. (1994) pro- vide evidence that this is indeed the case. To do so, they first investigate the cross-sectional relationship between ratings and yields. Their regression of the log of countries’ bond spreads against the average ratings by Moody’s and S&P has considerable power to ex- plain sovereign yields. Indeed, the single rating variable explains 92% of the variation in spreads, with a standard error of only 20 basis points, which shows that yields tend to be inversely related with ratings. In particular, their study suggests that every one-notch decrease in sovereign ratings is associated with a 22.1 percent increase in spreads. Next, they add macroeconomic and structural variables to the model and find that the explanatory power decreases considerably. This suggests that the ability of ratings to explain relative spreads cannot be wholly attributed to a mutual correlation with standard sovereign risk indicators. Based on this they conclude that ratings provide additional information beyond that contained in standard country-specific statistics incorporated in market yields. Addi- tionally, they conduct an event study and examine the average movement in credit spreads around the time of negative and positive rating announcements. Their results suggest that rating announcements cause a change in the market’s assessment of sovereign risk so that a downgrade on average increases spreads by 0.9 percentage points and upgrades decrease spreads by 1.3 percentage points. 101 Following the approach of Cantor, Packer, et al. (1994), the study by Ferri et al. (1999) estimates the determinants of sovereign ratings to construct “objective” ratings for East Asian countries for the time before and after the outbreak of the Asian Financial crisis. In other words, Ferri et al. (1999) is the first study to investigate ratings and its effects around a particular crisis. Two interesting findings stand out in their analysis. First, the ratings assigned to the four high-growth dynamic East Asian economies before the crisis were consistently higher than the economic fundamentals would warrant. Second, over the course of the crisis, the actual ratings dropped much more sharply than the model-predicted ratings, suggesting that rating downgrades were larger than the economic fundamentals would warrant. In addition, tracking countries’ ratings, fundamentals, and credit spreads, Ferri et al. (1999) provide evidence of self-fulfilling prophecies. That is, ratings were pro- cyclical and exacerbated the crisis beyond what would be justified based on the countries’ fundamentals. Building on Ferri et al., the study by G¨ artner, Griesbach, et al. (2011) investigates whether CRAs played a passive role or were a driving force in Europe’s sovereign debt cri- sis. First, they estimate relationships between sovereign debt ratings and macroeconomic and structural variables. Then, they use these estimates to decompose actual ratings into systematic and arbitrary components that are not explained by previously observed pro- cedures of rating agencies. Finally, they find that systematic and arbitrary parts of credit ratings affect spreads. This again opens up the possibility that arbitrary rating downgrades trigger processes of self-fulfilling prophecies or at least that ratings are pro-cyclical. Literature on the impact of sovereign ratings on the domestic economy Focusing on the effects of sovereign ratings on the private sector, Almeida et al. (2017) show that downgrading a sovereign has a significant effect on its real economy. Their study exploits exogenous variation in corporate ratings due to rating agencies’ sovereign ceiling policies, which require that firms’ ratings remain at or below the sovereign rating of their 102 country of domicile, to identify the effects of sovereign rating downgrades on the domestic economy. They find that firms reduce their investment and reliance on credit markets due to rising costs of capital following the sovereign downgrade. Alsakka, Ap Gwilym, et al. (2013) analyse the effects of sovereign rating changes on the credit ratings of banks in emerging markets. They find that sovereign rating upgrades (downgrades) have strong effects on bank rating upgrades (downgrades). In addition, they find that the sensitivity of banks’ ratings to sovereign rating actions is affected by the coun- tries’ economic and financial freedom and by macroeconomic conditions. Drago and Gallo (2017) verify the effects of sovereign rating revisions on banks. Specif- ically, they investigate the activity of European banks in terms of their regulatory capital ratio, profitability, liquidity, and lending supply as a reaction to sovereign rating changes of their respective home countries. They find that a sovereign downgrade has a signifi- cant impact, primarily on capital ratios and lending supply. In contrast, upgrades do not have a significant impact, indicating an asymmetric effect of sovereign rating changes. In addition, they show that three transmission channels (assets channel, funding channel, and rating channel) explain a relevant part of the impact of a sovereign downgrade. Finally, they find strong evidence that the rating-based regulation affects all measures of the activity of domestic banks, causing negative externalities for financial institutions. In addition to the effects of sovereign rating changes on bank lending, Chen et al. (2013) find that sovereign rating changes have an influence on private investment of re-rated coun- tries. Specifically, they find significant, though transitory, increases in private investment growth following upgrades in sovereign ratings. Conversely, they also find significant, tem- porary declines in private investment growth following sovereign rating downgrades. Their results hold after accounting for re-rated countries’ growth opportunities, endogeneity, and other factors that could affect private investment and they conjecture that the irreversible nature of investment may be the explanation for the temporary changes in the growth rates of physical capital investment associated with revisions in sovereign credit ratings. 103 Literature on international spillover effects of sovereign ratings There are several papers that analyze the effects of sovereign ratings beyond the immedi- ately affected country. For example, Alsakka and Ap Gwilym (2012) analyze the reaction of the foreign exchange spot market to sovereign credit signals by Fitch, Moody’s and S&P during 1994–2010. They find that positive and negative credit news affects both the own-country exchange rate and other countries’ exchange rates. In addition, they provide evidence on unequal responses to the three agencies’ signals. Fitch signals induce the most timely market responses, and the market also reacts strongly to S&P negative out- look signals. Credit outlook and watch actions and multi-notch rating changes have more impact than one-notch rating changes. Considerable differences in the market reactions to sovereign credit events are highlighted in emerging versus developed economies, and in various geographical regions. In a similar fashion, Gande and Parsley (2005) study the effect of a sovereign credit rat- ing change of one country on the sovereign credit spreads of other countries from 1991 to 2000. They find evidence of regional spillover effects in that a rating change in one country has a significant effect on sovereign credit spreads of other countries. This effect is asym- metric: positive ratings events abroad have no discernible impact on sovereign spreads, whereas negative ratings events are associated with an increase in spreads. On average, a one-notch downgrade of a sovereign bond is associated with a 12 basis point increase in spreads of sovereign bonds of other countries. The magnitude of the spillover effect follow- ing a negative rating change is amplified by recent rating changes in other countries. They distinguish between common information and differential components of spillovers. While common information spillovers imply that sovereign spreads move in tandem, differential spillovers imply opposite effects of ratings events across countries. Despite the predomi- nance of common information spillovers, they also find evidence of differential spillovers among countries with highly negatively correlated capital flows or trade flows with the US. That is, spreads in these countries generally fall in response to a downgrade of a country 104 with highly negatively correlated capital or trade flows. Variables proxying for cultural or institutional linkages, physical proximity, and rule of law traditions across countries do not seem to affect estimated spillover effects. Focusing on the European financial markets during the period 2007-2010, Arezki et al. (2011) examine spillover effects of sovereign rating news. Their main finding is that sovereign rating downgrades have statistically and economically significant effects both across countries and financial markets. The sign and magnitude of the spillover effects de- pend both on the type of announcements, the source country experiencing the downgrade and the rating agency from which the announcements originates. However, they also find evidence that downgrades to near speculative grade ratings for relatively small economies such as Greece have a systematic spillover effects across euro zone countries. They conjec- ture that rating-based triggers used in banking regulation, CDS contracts, and investment mandates may help explain these results. To summarize, the existing literature looks at the impact of sovereign ratings through a variety of lenses. However, most of the literature does not explicitly provide a model that outlines expected impacts of sovereign ratings in a certain domain. In addition, the literature generally does not investigate the extent to which the level of a sovereign rating affects a country, domestic firms, or third parties through spillovers. Instead, it relies on utilizing changes of sovereign ratings (i.e. up- or downgrades) as the explanatory variable to explain changes in spreads, foreign exchange rates, etc. At first sight, the idea to use rating changes as opposed to levels appears to have merit on grounds of causal identification of effects. However, the central problem in macroeconomic analysis is to connect a rating action with its effect. This is because rating changes take place in a dynamic context in which many changes are afoot and it is not always evident what would have happened in the absence of the rating change. Additionally, given the often transitory results of rating changes found in the literature, it is difficult to infer what the long-term consequences of the rating action would be. Therefore, switching from levels to changes may not constitute 105 substantial progress in understanding the effect of ratings. To mitigate these difficulties, this paper first outlines a theory of how biased sovereign ratings may affect the interest rate that a government has to pay. After that it provides evidence for the plausibility of the theory by tracing different channels that connect ratings with the pool of possible investors. Finally, the paper calculates quick and rough estimates to emphasize the economic effects of biased ratings. 3.3 Implications of biased ratings: theory There are a number of models which investigate the nexus between default probability and capital cost and which may generate multiple equilibria and self-fulfilling prophecies. Although the specifics of these models differ widely and may range from game theoretic to aggregate theories about multiple equilibria, the backbone of most such models is provided by Romer’s (2012) structural adaption of Calvo’s (1988) optimizing model of sovereign debt. In what follows, subsection 3.3.1 takes a quick look at the Romer model and a graphical representation of it. In a next step, subsection 3.3.2 extends the model to show how ratings can be incorporated to generate multiple equilibria. Moreover, subsection 3.3.3 discusses different channels that lend substance to the model’s predictions. 3.3.1 A simple model of default probability and capital cost interaction The canonical model of sovereign debt dynamics consists of two equilibrium conditions. 5 The first equilibrium condition renders investors indifferent between a government bond of countryI which pays interest ratei and defaults with a non-zero default probabilityp I , and some exogenous risk-free interest ratei . Under the assumption that default is a one-time event, investors are risk neutral, and the government only issues one-period bonds, the first 5 The graphical exposition if the Romer model in this dissertation draws heavily on G¨ artner and Griesbach (2017). 106 equilibrium condition becomes (1p I ) (1 +i) = (1 +i ) (3.1) or p I = ii 1 +i (3.2) In a diagram which has the default probabilityp I on the abscissa and the interest ratei on the ordinate, Equation 3.2 takes on the form of a hyperbola-shaped function. It has the y-intersection at the risk-free interest ratei . For a graphical representation, see thei-line in Figure 3.1. Figure 3.1: Multiple equilibria in the Romer model The second equilibrium condition focuses on the government decision about whether to service its debtD or to go into default. This decision depends on the difference between its ability to pay, i.e. the tax revenueT and the required payment, i.e. iD. For a given tax revenueT I and debt levelD I , the higher the interest ratei, the higher the probability of 107 defaultp I . Mathematically, p I =F (i;D I ) F i ;F D I > 0 (3.3) Figure 3.1 displays this relationship in red as a z-shaped curve. Let’s call this curve the p-line. The two equations above, represented as thei- andp-lines in Figure 3.1, complete the model and let us analyze sovereign debt dynamics. As Figure 3.1 indicates, up to interest ratei 1 servicing debt is painless and there is little risk of default. At an interest rate above i 2 , servicing debt is no longer possible as a government would have to draw funds from so many critical policy areas that it would be political suicide. Hence, abovei 2 , the default probability p I is essentially 1. In the interval between i 1 and i 2 , the default probability increases monotonically ini. When thei- andp-lines are indeed positioned as in Figure 3.1, the model features multiple equilibria. In fact, there are three equilibria. The first equilibrium is the point of intersection between the two lines in the lower left part of the diagram (pointE1). This is a “good” equilibrium in the sense that the interest rate is low and the government is highly likely to service its debt. In other words, the probability of sovereign default is very low. The second equilibrium lies at pointE2 where thei andp lines intersect again. Com- pared to equilibriumE1, equilibriumE2 is worse for a government. In equilibriumE2, a government has to pay a higher interest rate and faces a higher default risk, which bear the potential for slower economic growth and an economic crisis, respectively. What’s also clear from the diagram is that a country in equilibriumE2 essentially faces a credit spread over the risk-neutral asset with returni . This implies that such a country is rela- tively weakened when compared to other countries as it needs to spend a larger share of its productive resources on servicing a given amount of debt. The third equilibrium cannot be identified by a point in the diagram as the market for this country’s bonds has broken down and no interest rate is determined; the interest rate 108 would rise so much that default becomes inevitable. Graphically, this is illustrated as a regionE3 instead of as a point. So far, this macroeconomic model of the relationship between default probability and interest rates is static. Therefore, it doesn’t say anything about which of the equilibria are stable and about how a country could move from one equilibrium to another. However, Romer (2012) has a solution for this. He shows that under plausible assumptions such as permitting a lagged response of interest rates to changes in default probabilities the first (E1) and the third (E3) equilibrium are stable. The second equilibrium (E2) is unstable and can be regarded as a threshold. Once this threshold has been passed, the dynamics of interest rates and concomitant default probabilities bring the country into regionE3 where default is inevitable. This is shown graphically in Figure 3.2. Figure 3.2: Debt dynamics in the Romer model 109 3.3.2 An extended model with credit ratings as perceived default probabilities The Romer model discussed in the previous section illustrates the interplay between interest rates and default probabilities but it does not incorporate sovereign credit ratings into the analysis. Therefore, the present study extends the model to understand how (biased) ratings could affect countries. In a first step, it is crucial to see that the default probabilities displayed on the abscissa represent both perceived and actual probabilities. In particular, the probabilities reflected on thep line are actual default probabilities whereas the probabilities attached to thei line can be thought of as perceived default probabilities. Next, it is important to recognize that perceived default probabilities do not arise in a vacuum. Instead, sovereign credit ratings arguably influence the perceived default proba- bility among investors of a certain government both directly and indirectly. 6 If this is the case, then credit ratings can be causal in a country’s move from one equilibrium to another. To understand this process in greater detail, consider a country which is fiscally healthy and finds itself in equilibriumE2 in Figure 3.2. For this country, a rating downgrade decreases the trust that investors have in its ability to service debt. Put differently, the downgrade increases the investor’s perceived default probability of the country compared to what it was in the initial equilibriumE2. The increase in perceived default probability among in- vestors causes them to ask for a higher interest rate whenever the government tries to tap into capital markets. In turn, the higher interest rate increases the actual probability of a default of the country, which triggers further rounds of interest rate and default probability spikes until the process comes to a halt in the new equilibriumE3. Essentially, the mechanism behind the shift fromE2 toE3 is a self-fulfilling prophecy. That is, there is a negative feedback loop between interest rates and the default probability. On the one hand, higher interest rates lead to a higher probability of default as a government has to spend more of its resources on servicing debt. On the other hand, a higher perceived 6 See subsection 3.3.3 for a discussion of and evidence for the direct and indirect channels. 110 probability of default leads to higher interest rates as investors demand higher returns for the increased risk. To summarize, the process describe above opens up the possibility that credit ratings affect the type of equilibrium in which a government finds itself. This has two implications. For one, sovereign ratings can affect how much of a country’s productive resources have to be spent on servicing debt rather than consuming. More importantly, the idea that credit ratings can influence perceived default probabilities shows the perversity of the feedback loop in the context of sovereign debt; An initially unjustified increase in the interest rate would lead to an increase in the actual default probability. Ex post, this would provide justification for the initially unjustified interest rate hike. By the same token, an initially unjustifiably low (i.e. biased) rating for a country could increase its interest payments which would again justify ex post the initially unjustifiably low rating. Of course the same mechanism also works in reverse: An initially unjustifiably high rating may end up being justified through the improvement of fundamentals over the course of adjustment between equilibriaE2 toE1. 3.3.3 Do sovereign ratings affect perceived default probabilities? A crucial assumption made in the analysis above is that ratings influence the perceived default probabilities that investors hold and thus their behavior. Does this assumption have merit? There are several pieces of evidence that support this assumption. The evidence can be categorized as 1) experimental evidence, 2) regulatory evidence, 3) statistical evidence, and 4) additional evidence. 7 Experimental evidence The gold standard evidence in social science research is achieved through a randomized experiment but often such an experiment is not feasible as it is either prohibitively expen- 7 See G¨ artner and Griesbach (2017) for an in-depth discussion of the literature on evidence of rating effects on investor behavior. 111 sive or morally reprehensible to conduct. Luckily, however, researchers can draw on a real world event that resembles a randomized experiment. The event happened on 10 November 2011 when the credit rating agency Standard & Poor’s mistakenly announced a downgrade of France from its AAA rating. Reuters (2011) reports: “The erroneous alert, which S&P said was sent to some if its subscribers... contributed to the worst day for France’s govern- ment bonds since before the euro was launched in 1999. ... French bonds were hammered... Yields on 10-year paper, which were already on the rise before S&P’s mistake, spiked about a quarter of a percentage point, their largest jump since before the euro was launched in 1999. ... When S&P’s clarification hit the wires, the euro gained against the U.S. dollar, in a sign of investors’ relief.” This event shows that a “random” downgrade can send immedi- ate shock-waves through financial markets and lead to difficulties for a sovereign to access capital markets at favorable conditions. It is clear evidence that ratings matter for investor’s perceived default probabilities and that ratings influence the interest rate that a government has to pay. Regulatory evidence Over the past thirty years, financial regulation has become more codified, institutionalized, and juridified (compare to Moran, 1990). Especially the codification of financial regula- tion has promoted the relative importance of credit rating agencies. For example, financial market accords such as Basel II & III mandate financial institutions such as banks and in- surance companies to rely on external, standardized assessments of “credit risk” provided by CRAs. Specifically, Basel II forces financial institutions to respond to the downgrad- ing of a sovereign issuer by adjusting their portfolio composition. Technically, this means that a downgrade of a certain country may cause banks to wholesale the country’s govern- ment bonds, which increases the interest rate on new debt issues almost automatically as banks typically are major market makers during the issuance of new sovereign bonds. It goes without saying that the responsibility for this technical effect lies with governments 112 themselves as they were the ones that implemented pertinent regulations in the first place. However, not all governments have the same say in such matters. Particularly the US with its preeminent position as the country with the biggest global financial market and its posi- tion as a major funding contributor to international regulatory bodies often holds the upper hand in regulatory conversations (compare to Sinclair, 2014). This implies that there’s a US-led market segmentation. For example, there’s an eligibility threshold for US pension fund investments in which the minimally required rating is an “A.” Likewise, there’s an eli- gibility threshold for money market fund purchases where the minimally required rating is a “AA+.” All that is to say: at the very least at such thresholds, financial market regulation impacts how credit ratings affect interest rates. Statistical evidence The regulatory and experimental evidence strongly suggest that ratings can affect perceived default probabilities among investors and hence trigger self-fulfilling prophecies. However, particularly the regulatory evidence only applies around certain thresholds. Nevertheless, there’s statistical evidence that ratings can affect interest rates along the whole rating scale and not just around thresholds. While the issue of causality in macroeconomics is generally a tricky matter, there are a number of studies that use pertinent causality tests to show that ratings can affect interest rates across different rating levels. For example, Kamin and V on Kleist (1999) and Eichengreen and Mody (1998) identify a relationship between sovereign bond spreads and ratings for developing countries in the 1990s. Similarly, Gomez-Puig (2006) finds a negative relationship between bond spreads and ratings across the board for members of the Economic and Monetary Union (EMU) in the late 1990s. Focusing on the Great Financial Crisis (GFC), the results of Attinasi et al. (2009) suggest that the relation- ship between ratings and interest rates also holds in more recent times. Additionally, the results of G¨ artner, Griesbach, et al. (2011) suggest that the relationship between ratings and interest rates also held during the European sovereign debt crisis as their Granger causality 113 tests “can never reject the hypothesis that the credit spread [the key component of which is the interest rate] is caused by the rating.” Additional evidence There are a number of other pieces of evidence that suggest that ratings determine the interest rates that governments have to pay. Some channels are particularly worth men- tioning. First, an argument can be made purely on the grounds of rationality that ratings affect interest rates: Even a rational investor that disagrees with the (biased) assessment of a credit rating agency and who does not believe in the the default probability implicitly contained in it must heed its rating actions unless the investor believes that other market participants don’t react to the rating actions, which is counter-factual. In addition, inde- pendent of formal regulations there are built-in incentive structures in the financial sector that render ratings effective tools to influence perceived default probabilities and interest rates. To understand these incentive structures, think of portfolio managers at wealth ad- visors, banks or mutual funds and suppose that a major credit rating agency downgrades a certain country. If the portfolio manager holds on to the country’s debt and the country indeed defaults, the manager is blamed and potentially let go or even sued for neglect and bad conduct of business. If, however, the portfolio manager adjusts the portfolio and loses some of the client’s money to excessive trading fees, the manager can point to the rating agencies and argue that he acted in good faith. As such, the asymmetry in the incentive and reward structure of portfolio mangers lends additional credibility that sovereign ratings ex- ert a causal effect on perceived default probabilities and the interest rate that governments have to pay. 3.4 Implications of biased ratings: estimates In summary, there are several pieces of evidence that point in the direction that ratings affect the confidence investors have in a given sovereign’s capacity to service debt. Incorporating 114 this effect into a theoretical model of government debt dynamics arrives at the conclusion that biased ratings by private corporations in principle have the capacity to impact the in- terest rate governments have to pay in order to access international capital markets. This raises two follow-up questions: 1) Do ratings from all agencies influence investors’ expec- tations and behavior to the same degree or are only a few major agencies relevant in this regard? 2) For those agencies with a global impact, what is the economic significance of the in-group and home bias on ratings? In what follows, each question is addressed in turn. 3.4.1 Are all CRAs relevant for investors’ behavior? Assessing the relevance of specific agencies and the causal effect of their ratings on sovereigns in an econometric model is challenging, to say the least. The main difficulty lies in the data structure. Most ratings don’t change at all for long periods of time. In fact, there are rat- ings that haven’t changed over several decades. Conversely, credit spreads of sovereign debt instruments fluctuate on a daily basis and the amplitude can be substantial even over a short time period. As a consequence, modelling spreads as a function of the ratings of different agencies to investigate which ratings explain spreads the most is not feasible due to collinearity of ratings from different agencies. This leaves us with two alternative approaches to think about the relevance of different agencies. First, we can consider the impact of certifications on the regulatory importance of agencies. Second, we can use a practical approach and infer agencies’ relevance based on their name recognition in the marketplace. In what follows, each approach is considered separately to identify the scope conditions under which an agency’s ratings are relevant for government debt dynamics. Relevance of credit rating agencies: regulatory considerations If a rating agency does not possess broad regulatory approval, then its rating actions only have limited impact on financial markets. This is because banks and other financial insti- tutions are not required to dispose of certain debt instruments of a given sovereign issuer 115 only if the rating assigned by an unapproved agency falls below regulatory thresholds. In contrast, if the rating for a given sovereign issuer by an approved agency drops, then this has the implication that financial institutions may be required to re-balance their portfolios with direct consequences for the terms on which a sovereign can access capital markets. Consequently, to find out which rating agencies have the biggest impact on financial mar- kets, we have to see which rating agencies are certified by the most important financial market regulators. The most important financial market regulators are the US Securities and Exchange Commission (SEC) and its European counterpart, the European Securities and Markets Authority (ESMA). Both of them are responsible for the certification and continuous mon- itoring of credit rating agencies in their respective jurisdictions. In the case of the SEC, rating agencies need to be approved as Nationally Recognized Statistical Rating Organi- zations (NRSROs) to provide information that financial firms may rely on for regulatory purposes. In contrast to the approval process in the US, any rating agency that is domi- ciled in one of the EU member states is automatically ESMA-approved. This simplified approval process is based on the idea that by reducing market entry barriers, competition will lead to more accurate ratings. If a non-EU CRA wants its ratings to be used for reg- ulatory purposes in the EU (i.e. by EU financial institutions), the ESMA CRA regulation provides for two alternatives; equivalence/certification or endorsement. The equivalence regime is made available for CRAs from non-EU countries with no presence or affiliation in the EU, provided they are not systemically important for the financial stability or in- tegrity of the financial markets of one or more EU member states, and in order to allow the use in the EU of credit ratings related to entities established, or financial instruments is- sued in non-EU countries. In contrast, the endorsement regime is made available for CRAs that are affiliated or work closely with EU-registered CRAs. When an EU-registered CRA (the endorsing CRA) endorses a credit rating issued by another CRA of the same group established outside the Union, the rating can be used for regulatory purposes in the EU. 116 Compared to the ESMA and the SEC, other regulators such as the British Financial Con- duct Authority (FCA) or the Chinese financial market authorities are relatively unimportant at a global level. Hence, their agency certifications are not considered in this paper. 8 Table 3.2 provides an overview of the certifications of CRAs by the ESMA and the SEC. Importantly, only a few of the roughly 30 rating agencies which rate sovereigns are certified in both jurisdictions. The intersection consists of the eight agencies DBRS, Egan-Jones, JCR, Fitch, HR Ratings, Kroll, Moody’s and S&P. Thus, based on regulatory considerations alone it seems that only these eight institutions bear the potential to substantially affect government debt dynamics indirectly by influencing investors’ beliefs and behavior and directly by forcing financial institutions to construct their portfolios so that they are in line with regulatory rating requirements. Relevance of credit rating agencies: practical considerations Being certified by major financial regulators is a necessary precondition for an agency’s ratings to affect the cost of borrowing for a sovereign and ultimately the quality of life for individual citizens and consumers. This is because institutional investors and financial institutions are only required to adjust portfolios according to ratings issued by certified CRAs, particularly when they drop below important thresholds. However, to fully under- stand how powerful and relevant a certified agency is in comparison to its certified peers, we must consider other practical aspects. On the one hand, for a rating agency to be truly relevant it must have name recognition and its ratings must be easily accessible. Of course, name recognition is less important for institutional investors who are required to only use the ratings of certified agencies than it is for individual investors. In terms of name recognition, the three US agencies 8 Moreover, for the case of China there are at least three regulatory bodies that certify agencies for different purposes (rating sovereigns, corporates, etc.): China Securities Regulatory Commission (CSRC), People’s Bank of China (PBoC), National Development and Reform Commission (NDRC) (compare to Livingston et al., 2018). The author of this paper tried to find out which Chinese agencies are in possession of domestic certification to rate sovereigns, but data appears not to be publicly available. 117 S&P, Moody’s, and Fitch particularly stand out, followed by Japanese JCR and Canadian DBRS. This conclusion is reached by looking at the market shares of different agencies. For example, the SEC reports the number of outstanding credit ratings for government securities by agency. 9 Measured this way, S&P is the clear market leader with a market share of about 53%. Next come Moody’s (34%) and Fitch (12%). Hence, together these three agencies have a market share of about 99%. Comparing the total revenue of agencies, ESMA comes up with similar market shares. 10 Specifically, they estimate S&P’s market share at 42%, the one of Moody’s is 33%, and Fitch’s is estimated at about 16%. On the fourth place is DBRS with a share of roughly 2.5%. All other agencies are estimated to have less than 1% market share. Thus, the agencies S&P, Moody’s, Fitch appear to be the most familiar to financial market participants. In addition, the ratings of these three credit rating agencies feed directly into important information tools such as Bloomberg Terminal and Eikon Refinitiv so that banks, mutual funds and individual investors have easy access to them. This makes it likely that ratings from these agencies are also used for the assessment of financial decision makers outside important thresholds. Consequently, these ratings have the potential to affect sovereigns along the entire rating scale. On the other hand, for a rating agency to be influential its ratings must be available for a broad range of sovereigns; an agency which does not rate many countries automatically has less leverage to affect governments’ access to capital markets than an agency with broad coverage. In terms of coverage, again the three US agencies S&P, Moody’s, and Fitch stand out. At the end of 2019, they cover 129, 127 and 84 sovereign issuers, respectively. In comparison, the average number of rated sovereigns across all other agencies is only 23. 11 To summarize, taking into consideration the effect of regulatory certification, the rela- 9 Data from https://www.sec.gov/files/2019-annual-report-on-nrsros.pdf, retrieved on 2021-02-25. 10 Data from https://www.esma.europa.eu/sites/default/files/library/esma 33-9-383 cra market share calculation 2020 0.pdf, retrieved on 2021-02-25. 11 See Figure 3.3 for the number of rated countries by agency. 118 tive market shares of agencies, their breadth of country coverage, and the ease of access to rating data, it appears that the tree US agencies Standard&Poor’s, Moody’s, and Fitch are the most important players. Consequently, the biases of these agencies’ should have the biggest effects on sovereigns. 12 In the next section, the paper provides rough estimates of the economic effects of biased ratings on affected countries. 3.4.2 What is the economic significance of biased ratings? So far, this paper has presented a theoretical model that explains how ratings (from any agency) could affect government debt dynamics through investor behavior. In a second step the paper clarified the scope conditions under which an agency’s ratings are expected to influence investor behavior. As a result, it became clear that primarily the three major US agencies S&P, Fitch, and Moody’s have the capacity to influence the sovereign debt dynamics of an individual country at a global level. Hence, the final step in the analysis is to estimate the economic magnitude of the effects of biased ratings on affected countries. A full blown econometric analysis of all effects (government interest rate effects, ex- change rate effects for imports/exports, corporate ceiling effects, etc.) is beyond the scope of this paper. Instead, some simple examples shall illustrate the economic impact of the biased ratings from S&P, Moody’s, and Fitch. In the first example, let’s look at the advantage that the US enjoys as a result of being the domicile of the most influential credit rating agencies. To calculate the economic benefits, parameters for the following variables are needed: 1) the level of sovereign debt, 2) the interest rate on sovereign debt, 3) the impact of the home bias on the US rating, and 4) the impact of the rating on the interest rate the US government faces when tapping into financial markets. For 1) we can use the value 106.6%, which is the US general government net debt as a 12 Friedman (1995) comes to a similar conclusion when he states that “There are two superpowers in the world today in my opinion. There’s the United States and there’s Moody’s Bond Rating Service. The United States can destroy you by dropping bombs, and Moody’s can destroy you by downgrading your bonds. And believe me, it’s not clear sometimes who is more powerful.” 119 share of GDP at the end of 2020, according to the IMF World Economic Outlook Database. As we are interested in the effect of biases in long-term ratings, we can use the yields of long-term US treasury bonds as a proxy for the interest paid by the US. The average daily yield in 2021 for long-term US bonds (20 years) lies at 1.92%. Hence, we can use this value for 2). From Table 2.5 we know that the home bias is on average 1.41 rating notches, so that we can use this value for 3). 13 For 4), we can draw on the estimates of Cantor, Packer, et al. that a one unit increase in a rating is associated with a 22.1% decrease in yields. Using these parameters, we can calculate that the annual interest payments that the US government has to pay are about (106.6*0.0192)=2.04% of GDP. In contrast, without the home bias of US agencies the US rating would be 1.41 notches lower. This in turn would increase yields by 1.41*0.221=31.2% so that the total annual interest payments would rise to ((1.41*0.221)+1)*0.0192*106.6=2.68% of GDP. Put differently, the US benefit from the home bias of its rating agencies is about 0.64% of GDP, annually. Importantly, note that this is not a one time benefit. Instead, it accrues every year. Moreover, the estimate is a lower bound of the total benefit of the home bias to the US economy for two reasons. On the one hand, the home bias is probably larger than 1.41 as indicated by the results of quantile regressions, making the direct benefits to the US government larger. On the other hand, the sovereign rating provides a high ceiling for corporate ratings, which indicates additional benefits for the economy through more competitive enterprises. Lastly, US debt is expected to rise sharply in the coming years due to structural deficits. As a consequence, interest savings due to the home bias of the important US agencies will be of even larger magnitude. As another example, let’s look at the disadvantage that an emerging economy such as Mexico’s incurs through the biases of US rating agencies. To calculate the economic disadvantage, parameters for the following variables are needed: 1) the level of sovereign 13 Note that this is a conservative value as other sample lengths have found values as large as 1.7. Indeed, quantile regressions have found the home bias to be as large as 3.5 rating notches at the upper end of the rating distribution, i.e. where the US is situated (compare to Table 2.25). However, to be conservative, let’s use 1.41. 120 debt, 2) the interest rate on sovereign debt, 3) the impact of the home bias and in-group bias of the the US rating agency on the country’s rating, and 4) the impact of the rating on the interest rate the US government faces. According to the IMF, Mexico’s sovereign debt level stands at 65.5% at the end of 2020, so this is the parameter used for 1). For 2) we take the average daily yield of 20-year Mexican bonds in 2021 which is at around 7.1%. For 3) let’s take again the conservative number of 1.41. For 4) we can again draw on the estimates of Cantor, Packer, et al. that a one unit increase in a rating is associated roughly with a 22.1% decrease in yields. Based on these parameters we can calculate that the annual interest payments that the Mexican government has to pay are about (65.5*0.071)=4.65% of GDP. In contrast, without the home bias of US agencies Mexico’s rating would be about 1.41 notches higher. This in turn would decrease yields by 1.41*0.221=31.2% so that the total annual interest payments would fall to ((1.41*-0.221)+1)*0.071*65.5=3.20% of GDP. In other words, the Mexican disadvantage from the home bias of the US rating agencies is about 1.45% of annual GDP on a recurring basis. How does the geopolitical in-group bias affect the government debt dynamics of coun- tries? Presumably almost not at all. This results from the finding that on average rating agencies in non-authoritarian states do not have a substantial in-group bias and that only US agencies are globally relevant. However, chapter 2 shows that agencies in authoritarian states tend to have a magnified in-group. Equally, such agencies tend to weigh economic incentives such as a high trade exposure to another country more heavily. In light of this, it may be that some countries not only receive a disadvantage though the home bias of agencies in authoritarian states but also a disadvantage from the pronounced in-group bias of such agencies. A necessary condition for a strong impact of these magnified biases is that the rated country does not receive a rating from other, more important agencies such as S&P, Moody’s, and Fitch. That is, for biased ratings of agencies in authoritarian countries to really hurt other countries, it must be that they are the rating “authority” which requires 121 that more influential CRAs do not rate the country. 3.5 Conclusion To summarize, the present paper presents a theoretical model that shows how sovereign ratings bear the potential to affect government debt dynamics. In this model, higher ratings lead to lower interest payments, which in turn justify higher ratings ex-post. In a next step, the paper addresses whether ratings from all agencies have the same power and relevance to influence the interest rate a government has to pay. Based on regulatory and practical considerations, the paper arrives at the conclusion that only the three US agencies Fitch, Moody’s, and S&P have the capacity to influence interest rates of a larger number of coun- tries. Finally, the paper provides examples that show that the impact of biased ratings is quite substantial for affected countries. 122 Paper appendix Agency Number of countries rated as of 2019-12-31 ACRA 4 AcraEurope 8 Austin 1 Axesor 2 BCRA 6 CariCRIS 3 Chengxin 9 CI 44 Creditreform 23 Dagong 90 DBRS 36 Fareast 58 Fitch 84 HR 4 IIRA 1 JCR 36 JCREurasia 5 KROLL 9 Lianhe 68 Moody’s 127 Pengyuan 1 R&I 43 RAEX 10 RAM 22 S&P 129 Scope 36 ShanghaiBrilliance 51 TRIS 1 Table 3.1: Number of rated countries by agency as of 2019-12-31. 123 Agency ESMA-approved SEC-approved (NRSRO) JCR * x ACRAEurope x Axesor x BCRA x Capital Intelligence Ratings x Creditreform x DBRS x x Egan-Jones * x Fitch Ratings x x HR Ratings * x Kroll x x Moody’s x x RAEX x S&P x x Scope Ratings x Table 3.2: Certifications of CRAs which rate sovereigns. The asterisk (*) denotes that an agency is ESMA-certified as opposed to EU-based and registered. ESMA information as of 2021-01-04, SEC information as of 2021-01-25. 124 0 50 100 1990 2000 2010 2020 Agency ACRA AcraEurope Austin Axesor BCRA CariCRIS Chengxin CI Creditreform Dagong DBRS Fareast Fitch HR IIRA JCR JCREurasia KROLL Lianhe Moody's Pengyuan R&I RAEX RAM S&P Scope ShanghaiBrilliance TRIS Figure 3.3: Number of rated countries by agency, 1990-2019. 125 CHAPTER 4 EMERGING MARKETS SOVEREIGN CDS SPREADS DURING COVID-19: ECONOMICS VERSUS EPIDEMIOLOGY NEWS coauthored with Joshua Aizenman & Yothin Jinjarak forthcoming in Economic Modelling Abstract Can bad news about COVID-19 induce negative expectations on sovereign credit risks? We investigate the factors driving credit default swap (CDS) spreads of emerging market sovereigns around the outbreak of COVID-19. Using 2014-2019 data, we estimate a two- factor model of global and regional risks and then extrapolate the model-implied spreads for the period July 2019–June 2020. Intriguingly, the model initially predicts the real- ized spreads well but loses predictive accuracy during the COVID-19 pandemic. Fiscal space and oil-revenue dependence primarily drive the differences between the realized and predicted sovereign spreads. Our augmented-factor model indicates that the cumulative COVID-19 mortality rate growth is positively associated with the CDS spreads. The evi- dence suggests that the epidemiological deterioration can lower confidence in the sovereign credit markets due to the prospects of prolonged lockdowns and a slower GDP growth re- covery. Our results also hold for a single regression of daily spread changes during 2014- 2020. We gratefully acknowledge the financial support by the Dockson Chair and the Center of International Studies at the University of Southern California. 126 4.1 Introduction Many Emerging Market (EM) economies depend on commodity revenues, or on manufac- turing exports, whilst they have relatively limited policy space to buffer fluctuations in these markets. Frequently their populations are comparatively exposed to economic fluctuations as social safety nets and public health systems are limited in size. Additionally, emerging market governments’ access to international capital markets is fragile, limiting fiscal space and the capacity to react to external shocks. Against this backdrop, many observers were wondering how emerging markets would deal with the global outbreak of COVID-19, an unprecedented shock affecting both sides of the fiscal equation simultaneously: On the rev- enue side, the global slump in demand caused a fall in oil prices which would diminish oil exporter revenues both directly and indirectly. Equally, a fall of domestic economic activ- ity lowered household incomes and consumer confidence, further reducing tax revenues. On the expenditure side, debt became more costly as a global spike in investors’ risk aver- sion induced capital outflows from emerging markets, an increase in their sovereign Credit Default Swap (CDS) spreads and a depreciation of their currencies (see Figure 4.1 and Figure 4.2). The increase in risk aversion, in turn, can be attributed both to economic chal- lenges in investors’ home countries and in emerging markets alike. More importantly still, the pandemic forced governments to engage in unparalleled deficit spending to support strained public health systems and households severed from their income sources due to lockdowns (see Figure 4.3). The economic challenges caused by the pandemic and the simultaneous oil price col- lapse are not the first shock that emerging markets face. Indeed, there have been six pre- vious oil price collapses since 1970. Additionally, in recent decades, the increased flow of cross-border investment and trade in a more and more deeply intertwined global econ- omy has been subject to repeated interruption. For example, in 2015, there was a relatively contained crisis in China when the Chinese stock market tumbled, the Yuan slid, and a 127 lot of foreign investment quickly left the country. A year before that, dropping prices of oil and other commodities sent a shock wave through several emerging markets. Further- more, a general slowdown in emerging market growth began in 2013 with taper tantrum of the Federal Reserve (Fed). Taken together, these examples indicate that not all crises in emerging markets are the same. Some are more country-specific while others seem to be driven by some common global factor, affecting emerging markets at large. Thus, with COVID-19 as a global pandemic and an external demand shock, two research questions reveal themselves: Are sovereign CDS spreads determined by a time-varying combina- tion of global and country-specific factors and if the latter factors play a role, which ones? Particularly COVID-19 raises a corollary question about country-specific determinants of sovereign CDS spreads: is the epidemiological situation in a country significant for CDS spread determination once the economic repercussions of lockdowns and foreign demand slump as well as stimulus responses are taken into account? To address the first question, we adopt a two-stage econometric approach, using a panel of thirty investible emerging markets. In the first stage, we estimate a global/regional factor model for changes in sovereign CDS spreads before the outbreak of COVID-19. Specifi- cally, we estimate and train the model over the period January 2014 through June 2019. In the second stage, we use the estimated coefficients to extrapolate model-implied changes in CDS spreads for the period July 2019 through June 2020. This approach is appealing for several reasons: On the one hand, it allows us to compare the model’s out-of-sample properties during normal and pandemic times. On the other hand, this approach facilitates the statistical derivation of the “COVID-19 residual,” which is the difference between actual CDS changes and model implied changes for the pandemic period. Additionally, the residuals can then be used to answer the question which exact country-specific factors were the biggest drivers of CDS adjustment during COVID-19. Specifically, we explain the residuals through a variety of country-specific factors from three different realms: epidemiological, economic, and policy factors. 128 To address the second question, we leverage the panel and introduce all control vari- ables (global/regional and country-specific epidemiological, economic and policy factors) simultaneously to test the significance of the epidemiological variables. Our paper makes two distinct contributions. First, we engage in the discussion of the relative importance and time-varying nature of global/regional and country-specific fac- tors in the determination of emerging markets sovereign risk pricing, particularly during a global pandemic. Second, our analysis contributes to an understanding of the economic effects of lockdowns on government finances by identifying if and which country-specific factors drive spreads during a global pandemic. Disentangling the effects of economic contagion and of domestic economic conditions and policy responses on sovereign CDS spreads promises to hold valuable implications for policy makers weighing the economic trade-offs of different containment strategies. Our main findings can be summarized as follows: • First, the global/regional factor model traces the realized sovereign CDS spreads well for the out-of-sample period before the pandemic. This observation suggests that emerging market CDS spreads are driven largely by global/regional factors in risk-on environments. • Second, the relationship between actual CDS spread changes and changes predicted by the global/regional factor model breaks down during the pandemic. This suggests that CDS spread determinants are of time-varying nature and that investors weigh country-specific factors more heavily in their decision-making process in risk-off environments. • Third, actual CDS spread changes experience the biggest deviations from model- implied values in March 2020, suggesting that uncertainties around the ramifications of COVID-19 vary across time and were highest at the peak of the first wave. • Fourth, residuals at peak COVID-19 were primarily driven by traditional country- 129 specific factors of sovereign solvency such as fiscal space and oil income depen- dence rather than epidemiological factors. This suggests, maybe surprisingly, that the severity of the pandemic in terms of mortalities in a specific country didn’t influ- ence investors’ behavior very much, at least during peak COVID-19. • Fifth, over the entire pandemic period the growth in the cumulative mortality rate is significantly positively related to CDS spread changes after taking the economic ramifications of lockdowns and the oil price decline into account. This suggests that a higher COVID-19 mortality rate increases the odds of more prolonged lockdowns and a slower GDP growth recovery, increasing spreads. The rest of the paper proceeds as follows: Section 4.2 reviews the literature. Section 4.3 sets the stage by presenting stylized facts about emerging markets during COVID-19. Sec- tion 4.4 contains the main analysis. Its subsections provide simple economic intuition behind every mechanism that we test and then discuss the results. Section 4.5 concludes. 130 4.2 Related literature There is an ongoing debate about the degree to which country-specific and global/regional factors are drivers of sovereign risk pricing and indeed local financial market conditions more broadly. As an in-depth examination of the existing literature is beyond the scope of this paper, we refer readers to Augustin et al. (2014) for a detailed review of existing studies and only sketch the main research avenues. Generally speaking, there are two camps on opposite sides of the spectrum. One camp proposes a bottom-up logic, arguing that sovereign risk pricing is primarily a function of a country’s fundamentals such as a fiscal space, sovereign debt levels, and local economic conditions. In contrast, the other camp argues that country fundamentals are largely negligible and instead stresses the significance of global/regional factors in explaining sovereign risk pricing. Below, we first present a selection of studies from either camp. Then, we present synthesizing studies which provide evidence that suggests that determinants of sovereign risk pricing are time-varying. That is, regional/global and country-specific factors both drive spreads but to degrees which change over time. Based on these findings, we develop simple economic intuitions to explain the mechanisms that we test in the main analysis of this paper. 4.2.1 Literature on country-specific drivers of sovereign risk pricing Covering the taper-tantrum episode of 2013 and seven other episodes of severe financial turmoil since the mid-1990s, Ahmed et al. (2017) assess the importance of country fun- damentals in the transmission of international shocks to financial markets in emerging markets. They find that 1) emerging markets with relatively better economic fundamen- tals suffered less deterioration in financial market conditions during the 2013 episode. 2) Differentiation among emerging markets set in relatively early and persisted through this episode. 3) During the taper tantrum, while controlling for economic fundamentals, finan- cial market conditions also deteriorated more in those emerging markets that had previ- 131 ously experienced larger private capital inflows and greater exchange rate appreciation. 4) During emerging market crises of the 1990s and early 2000s, they find little evidence of investor differentiation across emerging markets being explained by differences in their rel- ative vulnerabilities. 5) Differentiation does not appear to be unique to the 2013 episode; it also occurred during the global financial crisis of 2008 and, subsequently, during financial stress episodes related to the European sovereign debt crisis in 2011 and China’s financial market turmoil in 2015. Taking a more focused approach, Kocsis and Monostori (2016) investigate specifically the determinants of sovereign CDS spreads of Poland, Russia and Turkey. They use a dynamic hierarchical factor model to aggregate information in indicators of economic fun- damentals. CDS spreads are then regressed on forecasts of factors. They find that domestic fundamentals explain more of CDS spread variance than global factors which is largely due to their ability to explain differences in sovereign risk across countries. 4.2.2 Literature on global drivers of sovereign risk pricing The importance of global factors in explaining sovereign risk pricing is usually motivated by the observation that CDS spreads of different countries tend to move in lockstep over longer periods of time. While there are a number of papers pointing out this observation, the most prominent exponent of this theory is H´ el` ene Rey with the notion of a “global financial cycle.” Rey (2015) argues that “Risky asset prices around the globe, from stocks to corporate bonds, have a strong common component. So do capital flows ... Global financial cycles are associated with surges and retrenchments in capital flows, booms and busts in asset prices and crises. The picture emerging is that of a world with powerful global financial cycles characterized by large common movements in asset prices, gross flows, and leverage ... The global financial cycle can be related to monetary conditions in the center country and to changes in risk aversion and uncertainty ... capital flows, especially credit flows, are largely driven by a global factor...”. 132 Fender et al. (2012) study the determinants of daily CDS spreads of emerging mar- ket sovereigns over the period April 2002 to December 2011. Using GARCH models, they find spreads are more related to global and regional risk premia than to country-specific risk factors. This result is particularly evident during the second subsample (August 2007–De- cember 2011), where neither macroeconomic variables nor country ratings significantly explain CDS spread changes. Second, measures of US bond, equity, and CDX High Yield returns, as well as emerging market credit returns, are the most important drivers of CDS spread changes. Finally, their analysis finds that CDS spreads are more strongly influenced by international spillover effects during periods of market stress than during normal times, suggesting that CDS spread drivers may be time-varying. 4.2.3 Literature on time-varying drivers of sovereign risk pricing Attempting to synthesize the empirical evidence in favor of both local and global/regional factors, Remolona et al. (2008) decompose sovereign CDS spreads into expected losses from default and the market risk premia required by investors as compensation for default risk over the period 2002 to mid 2006. They find that country-specific fundamentals pri- marily drive sovereign risk whilst global investors’ risk aversion drives time variation in risk premia. Consistent with this, they also find that within emerging market regions the sovereign risk premia is more highly correlated than sovereign risk itself. This finding suggests that there is a third category that drives sovereign risk pricing: regional factors. By examining the heterogeneity in the sensitivity of CDS spreads to changes in the global risk factor, Cepni et al. (2017) look at the global-vs-local debate through a different lens. They find that countries with lower government debt and higher reserves tend to be less subject to the variations in global risk appetite. That is, they do not argue that only global/regional or only local factors matter but instead identify which local factors interact with and determine a country’s dependence on global risk appetite. In summary, the existing literature provides evidence that both local and global/regional 133 factors drive sovereign risk pricing. Importantly, the degree to which each of these factors matter changes over time. Our study builds on these insights and extends the existing litera- ture by evaluating sovereign CDS spreads during COVID-19. Specifically, we examine the impact of country-specific epidemiological variables such as COVID-19 cases and mortal- ities on sovereign CDS spreads. Before we move on to explaining the intuition behind the epidemiological mechanisms that we test, we provide some stylized facts about COVID-19 in emerging markets. 134 4.3 Stylized facts about emerging markets and COVID-19 4.3.1 Mortality patterns COVID-19 hit the global economy. However, mortality dynamics were heterogenous, both between developed and emerging countries as well as within each group. Figure 4.4 and Figure 4.5 illustrate the discrepancies in mortality rates and mortalities per million resi- dents. As per the end of April 2020, Peru, Brazil, Panama, Romania and Turkey were the emerging market countries with the highest daily mortality rates whereas Bahrain, Qatar, Sri Lanka, China, and Thailand were at the lower end. A similar picture emerges for cumu- lative mortalities per million residents: by the end of April 2020, Hungary, Panama, Peru, Turkey, Romania had the highest death tolls whereas Bahrain, China, Qatar, Sri Lanka, and Thailand had the lowest ones. Some of this heterogeneity can be attributed to differing reporting standards across countries. Nevertheless, Jinjarak et al. (2020) provide evidence that the government pandemic policy interventions along with initial country characteris- tics such as demographics may have influenced mortality dynamics. They explore several demographic and structural features across a large sample of both advanced and emerging economies from 1/23/2020 to 4/28/2020 and find that with a lag more stringent pandemic policies were associated with lower mortality growth rates. Moreover, the association be- tween stricter lockdowns and lower future mortality growth was more pronounced in coun- tries with a greater proportion of the elderly population, greater democratic freedom, larger international travel flows, and further distance from the equator. In addition, they document that the extent to which the peak mortality rates were explained by government pandemic policies and country-specific structural features is heterogeneous. Crucially, the difference in mortality rates and mortalities per million residents across emerging markets provides us with variation that we can use to estimate the effect of country-specific factors on sovereign CDS spreads. Before estimating these effects, let’s have a look at another source of variation: fiscal responses of emerging markets to COVID- 135 19. 4.3.2 Fiscal responses The fiscal response of emerging markets to COVID-19 has been decisive. Except for Saudi Arabia, all emerging markets in our sample engaged in stimulus (see Figure 4.3) and the stimulus packages include both budgetary and non-budgetary provisions. While the stim- uli generally look impressive, Alberola-Ila et al. (2020) point out that they are relatively modest in comparison to stimulus packages of developed countries. In fact, budgetary measures in advanced economies have reached 8.3% of GDP, 6.6 percentage points higher than in the aftermath of the great financial crisis. In contrast, for emerging markets the stimulus was on average just 2.0%, which is less than in the great financial crisis. Starker than the difference in stimulus size between developed and emerging economies is how the stimulus packages are structured: The biggest difference is that credit guarantees are 6.6% of GDP in advanced economies and only 0.4% in emerging markets. The gap for funding facilities is narrower: 4% of the GDP of advanced economies versus 1.3% in the case of emerging markets. The difference in fiscal stimulus packages between advanced and emerging markets could be a symptom of a lack of fiscal space in the case of the latter. More likely, however, a potential lack of fiscal space only explains part of the difference in stimulus size. Another plausible reason may be the difference in the prevalence and severity of the pandemic between advanced and emerging economies. COVID-19 proved more lethal in advanced economies which on average have older populations that are more prone to severe symptoms of COVID-19. Lastly, another reason for comparatively smaller fiscal stimuli in emerging markets is that fiscal stimulus is a substitute to monetary stimu- lus: emerging markets have been able to take advantage of more room to cut policy rates than their advanced economy counterparts. At the start of 2020, policy rates in emerging markets were on average 4.9% (excluding Argentina) whereas the average policy rates in advanced economies were at 0.4%. Since then, emerging markets have cut policy rates by 136 around 114 basis points, almost three times the 40 basis points cut of advanced economies. However, Figure 4.6 makes it clear that rate cuts alone are no silver bullet for emerging markets. Particularly for the oil exporting countries (except for Mexico, which largely hedges oil revenues), rate cuts necessitated interventions in foreign exchange markets and the reduction of foreign reserves to stabilize currencies. 4.3.3 Sovereign CDS spread changes Similar to the variance in mortality patterns and stimulus size across emerging markets, there is variance in sovereign CDS spreads across emerging markets (Figure 4.7). Inter- estingly, the time and cross-country variance in sovereign CDS spreads among emerging markets is more pronounced than for developed economies (Figure 4.8). Crucially, the variation in CDS spreads across emerging markets and across time allows us to estimate the effect of country-specific factors on sovereign CDS spreads. In summary, there is variation in how countries handled the pandemic in terms of public health dynamics, fiscal responses, and sovereign CDS spreads. This variation can be used to investigate how sovereign CDS spreads reacted to COVID-19, which we do in the following section. 137 4.4 Analysis of COVID-19 dominance 4.4.1 Are sovereign CDS spread drivers time-varying? In this section, our primary goal is to establish if COVID-19 lead to a time-varying re- lationship of sovereign CDS spread determinants. We propose a two-stage econometric approach. In the first stage, we estimate a heterogeneous multi-factor model using daily data for thirty investible emerging markets for the period January 2014 through June 2019. The explanatory variables in the model capture global and regional risk factors. In the second stage, using a synthetic control-type procedure to extrapolate the model-implied changes in CDS spreads given the realized values of the factors, we evaluate the model’s out-of-sample properties for the period between July 2019 and June 2020. Selecting this approach has two advantages. On the one hand, it allows us to evaluate the model’s out-of-sample properties for a period before the pandemic. On the other hand, we can also evaluate the model’s out-of-sample properties for the pandemic period. Thus, we have two distinct periods for which we can compare the model’s explanatory power. This allows us to test the first proposition that the determinants of sovereign risk pricing vary across time. The economic intuition behind testing the proposition this way is as follows: If sovereign CDS spreads are predominantly driven by global/regional risk factors at all times, we would expect the model to fare similarly well for the period before and after the global outbreak of COVID-19. However, if the model fares differently across the two periods, then this would indicate that the determinants of sovereign risk pricing are time-varying, and that the importance of country-specific factors plausibly increased after the outbreak of COVID-19. Moreover, the two-stage approach has an additional theoretical benefit: By comparing realized and predicted CDS changes, we can calculate model residuals. Doing this for the pandemic period, we can calculate “COVID-19 residuals” which can then be used to 138 analyze the significance of a number of country-specific factors for sovereign risk pricing during the pandemic. In what follows, we first outline the technical details of the two-stage approach and then present the results. Data We use the following data: • Sovereign Credit Default Swap spread (CDS spread). We use the daily 5-year CDS spreads reported by Eikon Refinitiv and convert the levels into daily log changes. • Gross domestic product (GDP). We use GDP data in current $ reported by the World Bank. Model In the first stage, we estimate a factor model with daily data for thirty investible emerg- ing market sovereigns for a period of 5.5 years from January 2014 through June 2019. Investibility is defined by a country’s representation in the reference index for emerging market sovereigns, the J.P. Morgan Emerging Marets Bond Index (EMBI). 1 Specifically, we use the following specification: cds i;t = i + i CDS i;t1 + i1 GCDS t + i2 RCDS i 0 ;t + i;t (4.1) cds i;t =ln CDS i;t CDS i;t1 and Jan 1, 2014t< July 1, 2019 Our outcome variable is the daily log change in the CDS spread of country i. Our explanatory variables include the lagged dependent variable along with two risk factors. A global factor, GCDS t and a regional factor RCDS i 0 ;t . The global factor is con- structed as the GDP-weighted average of daily log CDS changes of a group of twenty 1 See Table 4.1 for a list of the thirty countries in the sample. 139 core economies; the US, Japan, and the Eurozone member states. 2 It therefore captures the common component of sovereign default risk fluctuations at the global level. 3 The re- gional factor is constructed slightly differently. First, we sort the thirty emerging markets into seven reference groups. These groups and their constituent countries can be found in Table 4.3. The grouping criterion is geographic clustering. The justification for this crite- rion is the proposition of gravity models that countries in close proximity, i.e. geographic clusters, trade relatively more with each other than partners further apart do. Because of this, countries in close proximity often have synchronized business cycles, implying simi- lar trends for governments’ expenditures and revenues. This in turn should be reflected in the pricing of sovereign risk as argued by Remolona et al. (2008). Finally, the residual is defined as cr i;t = cds i;t [^ i + ^ i CDS i;t1 + ^ i1 GCDS t + ^ i2 RCDS i 0 ;t ] (4.2) simply comparing the realized log CDS changes to the model-implied changes, given the true realization of the factors and lagged log CDS changes. Results and interpretation After training the factor model from January 2014 through June 2019 in the first stage, we extrapolate the model based on realized values of factors from July 2019 through June 2020 in the second stage. This spans time both before and after the outbreak of COVID-19. The upper-left panel of Figure 4.9 traces the emerging market average cumulative log CDS change (solid line) against that predicted by the model (dashed line). Note that the two lines greatly overlap between July and December 2019, implying that the model does a good job in predicting actual CDS spread changes for the period before the outbreak of COVID-19. 2 While the Eurozone is made up of nineteen member states, our analysis includes eighteen states as CDS data for Luxemburg was not available. 3 The weighting of countries for the global factor is based on 2019 GDP, which renders the following weights: US 53.9%, Japan 12.8%, Eurozone 33.3%. See Table 4.2 for an overview. 140 In early 2020, the two series start to fluctuate and the lift-off in both model-implied and actual values accelerates in March 2020. However, actual values sore considerably more than predicted, suggesting that the model loses some of its explanatory power in the wake of the pandemic. CDS spread widening ceased in mid to end of March but the divergence between actual and model-implied changes persisted. The contrast between the explanatory power before and after the start of the pandemic can be taken as evidence for a time-varying relationship of the determinants of sovereign risk pricing and hints that country-specific factors generally may have become more important determinants during COVID-19. The lower charts of Figure 4.9 compare the top and bottom five mortality emerging markets by end of April 2020. The bottom-left chart indicates that the change in actual CDS values was more marked for high mortality countries than for low mortality countries. This corroborates our finding from above that country-specific risk factors are partially re- sponsible for the spike in CDS spreads during the pandemic and it suggests that mortality dynamics may have been one of the dominant country-specific factors. The bottom-right chart indicates two things: High mortality countries experienced more volatility in their CDS spreads than their low mortality counter parts. Furthermore, the residuals for low mortality countries were positive for the whole period and turned sharply positive in March 2020, which stands in contrast to high mortality countries’ residuals which turned sharply negative in March 2020 and then started to approach the low mortality residual levels by the end of June. The greater volatility for high mortality countries is additional evidence that country-specific risk factors became more important determinants of sovereign risk pric- ing during the pandemic. Lastly, the upper-right of Figure 4.9 charts the cross-sectional dispersion of CDS spreads between July 2019 and June 2020. The dispersion already rose over the second half of 2019 and experienced a marked uptick in March 2020, highlighting the sharp rise in volatility and the divergent path of CDS spreads across emerging markets amid the pandemic. The increasing dispersion of CDS spreads across countries visual- 141 izes that the model’s explanatory power decreased during the pandemic, again suggesting that country-specific determinants of sovereign risk pricing took on a more prominent role during COVID-19. While the top-left chart of Figure 4.9 indicates that the model does a good job of pre- dicting emerging markets CDS spread changes at an aggregate level before the pandemic, this is not the case for all countries individually: For some countries both the R-squared in the estimation period as well as the predictive accuracy out-of-sample are relatively low (see columns 5 and 6 in Table 4.4). This in itself does not invalidate our two-stage approach. If anything, it supports the proposition that the determinants of sovereign risk pricing are time-varying at the level of individual economies, too. However, including countries for which the model performs subpar in the period before the pandemic would lead to dis- tortions in our later analysis of residuals. The reason for this is simple: If we included countries for which the model does not work well before the pandemic, we would poten- tially attribute residuals to the pandemic even though they already existed before it (see Figure 4.10 for such cases). Therefore, we drop countries for which the global/regional model does not work well from the sample. We choose a correlation coefficient between actual and fitted values for the period between July 2019 and February 2020 below 0.25 as the cut-off value. This results in a reduced sample of twenty emerging markets, which we list in Table 4.5. A possible concern with this procedure is that by trimming the ten outlier countries from our sample we reduce the veracity of our main results. We deal with this concern in section 4.4.4, where we show that our results, including the relative importance of economics versus epidemiology news, hold for the entire sample of 30 countries in a single regression analysis of daily CDS spread during January 2014 to June 2020. 4.4.2 Which country-specific factors drove CDS spreads during COVID-19? In the previous subsection we established that the global outbreak of COVID-19 lead to a time-varying relationship of the determinants of sovereign CDS spreads and that country- 142 specific factors in general became more prominent. In this section, our primary goal is to identify exactly which country-specific factors drove CDS spreads during the pandemic. We begin by defining three COVID-19 subperiods: • early COVID-19 (January-February 2020) • peak COVID-19 (March 2020) • late COVID-19 (April-June 2020) By defining three COVID-19 subperiods, we document that March 2020 is the period during which the realized values of daily CDS spread changes diverged the most from model-implied values. In contrast, the global/regional factor model does an excellent job of tracing realized values before and after that (see Figure 4.11). Furthermore, we observe that daily CDS spread changes were remarkably higher and more volatile in March 2020 than in the early and late COVID-19 periods (see Figure 4.9). As such, we focus on the peak COVID-19 period to evaluate which country-specific factors account for the variation in CDS adjustment that is not explained by the model. Clearly, there’s a myriad of country-specific factors that could potentially affect CDS spreads during peak COVID-19. However, neither are all of these factors equally plausi- ble nor can all of them be tested due to data limitations. As such we identify a number of country-specific factors that could plausibly influence sovereign risk pricing and group them into three categories: economic variables, policy responses, and epidemiological vari- ables. The economic intuition behind including country-specific economic variables builds on insights of Kocsis and Monostori (2016) and Ahmed et al. (2017): Variables that deter- mine government revenues and expenditures help explain government solvency and thus the pricing of sovereign risk. We propose seven economic variables: • an economy’s dependence on oil prices, which determines directly and indirectly a sizeable share of government revenues for many emerging markets 143 • an economy’s access to the IMF’s Rapid Financing Instrument, which could alleviate temporary liquidity problems and avoid defaults • an economy’s access of Fed swap lines, which could alleviate temporary liquidity problems and challenges related to foreign denominated debt • sovereign wealth fund buffers and international reserve buffers, which could be used to stabilize the exchange rate • external debt ratios, which indicate an economy’s vulnerability to external funding conditions and exchange rate swings • an economy’s ratio of (hidden) debt owed to China, which indicates an economy’s vulnerability to exchange rate swings that may not be captured by official numbers on external debt ratios The economic intuition behind including policy variables builds on the idea that policy responses such as fiscal stimuli lead to higher deficits and higher future sovereign debt burdens. This in turn may increase the expected likelihood of sovereign defaults and affect sovereign risk pricing. We propose three policy variables: • dummy variables that indicate the date of country-specific key fiscal policy announce- ments • dummy variables of key policy announcements by the European Central Bank • dummy variables of key policy announcements by the US Fed The economic intuition behind including epidemiological variables builds on two ideas. On the one hand, some epidemiological variables are proxying for GDP. Thus, they cap- ture potential income falls which affect government tax revenues as well as expenditures through fiscal stimulus. On the other hand, epidemiological variables like accelerated COVID-19 mortality also capture the pandemic trend, possibly impacting the speed of 144 future recovery as well as the severity of longer-term output losses due to bankruptcies, degradation of human and physical capital, and other social costs. We propose three such variables: • a variety of variables related to mortality: the daily new mortality rate per 1,000,000 population and the daily new mortality growth rate, the cumulative mortality rate per 1,000,000 population and the cumulative mortality growth rate. • daily growth rates of policy stringency indices, which indicates to what extent gov- ernments hindered the free functioning of the economy and thus proxies for economic activity Data Specifically, we use the following data and variable definitions: • Deaths. We use daily deaths per country reported by the Center for Systems Sci- ence and Engineering at Johns Hopkins University (JHU CSSE). Counts include confirmed and probable where reported. • Stringency Index SI. We use the Oxford COVID-19 Government Response Tracker which measures the strictness of lockdown policies that primarily restrict people’s behavior. • European Central Bank (ECB) Policy Dummy. Fed Policy Dummy. Fiscal Policy Dummy. Daily fiscal and monetary policy announcements. These were collected for individual countries, for the European Central Bank, and for the Federal Reserve. These columns capture whether or not an action or proposal was made by a given institution/country on a specific date. A row is coded as “1” if the date corresponded with the announcement of at least one key policy. With the exception of the Federal Reserve (whose major announcements related to reductions in the interest rate along 145 with fiscal spending), we restricted our analysis of key fiscal policies to those which provided “millions” or “billions” of local currency units in spending. The drawback is that this measure does not show the size or number of policies on any given day. However, such a measure is not available on a consistent basis across countries. 4 • External Debt. We use the total external and private sector debt stock as a share of GDP as reported by the World Bank. The data is yearly. • Debt owed to China. We use the estimated total external debt stock owed to China in current USD as a share of the debtor GDP as reported by Horn et al. (2019). The data is reported yearly up to 2018 so that we used the most recent yearly values for cross-country variance. • Oil price income effect. This is a compound variable that is calculated daily as: Daily Oil price * [(Oil share of exports * Export share of GDP) – (Oil share of imports * Import share of GDP)]. The daily oil price data is for Brent as reported by Eikon Refinitiv. The data for export and import shares of GDP are from the World Bank. The data for oil export and import shares are from the International Trade Center (ITC). • International reserves. International reserves in current USD as reported by the IMF. Data is monthly. • Sovereign Wealth Fund Volume. Sovereign wealth fund volume in current USD as reported by PWC. Data is yearly. • Rapid Financing Instrument. Approved rapid financing instrument from the IMF as share of recipient country’s GDP. 5 4 The primary data sources used to construct these policy announcement variables are: Yale COVID-19 Financial Response Tracker; Harvard Global Policy Tracker; Bruegel COVID-19 National Dataset; IMF Policy Responses to COVID-19; and the St. Louis Federal Reserve. 5 Because no RFI was announced/approved in March 2020, this variable will not show up in the regression tables. 146 • Swap line activation. Measures the activation of a swap line with the Fed. Data is from Bahaj and Reis (2020a). 6 Model Including these variables and using the peak COVID-19 residual from Equation 4.2 as the dependent variable results in the following model where# i and t represent country and time fixed effects, respectively: cr i;t =# i + t + X COVID i;t + X economy i;t +X policy i;t + i;t (4.3) March 1, 2020t March 31, 2020 Results and interpretation The results of Equation 4.3 are reported in Table 4.6. Daily new mortality rates and new mortality growth rates are positively associated with COVID-19 residuals across all three specifications, although not statistically significantly so. Specifically, countries that saw higher new mortality rates or higher new mortality growth rates were likely to see wider divergence in realized CDS spreads from model-implied values. The cumulative mortality rate and its growth rate are negatively related to the COVID-19 residuals and the growth rate becomes statistically significant when adding the economic and policy variables. While the mortality and mortality growth rates together only explain a small share of the variation in COVID-19 residuals (R-squared of 1.24%), adding the other COVID-19 specific variable (growth of policy stringency index) increases the R-squared to 1.37%. This suggests that the economic ramifications directly related to lockdowns account for a relatively minor 6 Facing visible strain in dollar funding markets during the COVID-19 pandemic, the Fed lowered the rate on the swap lines it had with five other central banks, and opened new ones with nine other countries on 19 March 2020: Australia, Brazil, Mexico, Denmark, South Korea, Norway, New Zealand, Singapore, and Sweden. As of 31 March 2020, only the central banks of Sweden, Norway, Denmark and Singapore were have completed a dollar swap line operation. Thus, no emerging market country has accessed a swap line so this variable will not show up in the panel regression. 147 share of CDS adjustment and is smaller than the mere thought about additional future lockdowns that could be indicated by mortality dynamics. In contrast, a bigger jump occurs when we include policy responses and economic variables so that we reach an R-squared of roughly 6.47%. In specifications 2 and 3 we also see that the SI growth coefficient is not statistically significant. The interpretation is that the growth of policy stringency—as an ad-hoc measure of daily economic activity—has little effect on CDS residuals. This suggests that sovereign risk pricing during peak COVID-19 has been predominantly driven by traditional country-specific economic variables. 7 While two out of three policy measures, i.e. Fed policy dummy and as well as in- teraction terms with country-specific fiscal policies, don’t have statistically significant as- sociations with COVID-19 residuals, the interaction of the fiscal policy dummy with the external debt level is nevertheless positive, indicating that countries that increased their debt burdens through stimuli or countries that already had relatively high debt burdens to begin with were likely to see larger spread increases. Equally, more (hidden) debt owed to China is also positively associated with CDS residuals, albeit again not statistically signif- icantly so. However, the interaction of debt owed to China with the fiscal policy dummy is negatively correlated with residuals. This seems to indicate a somewhat indifferent view of investors about the fiscal sustainability of announced stimuli, which can probably be partially explained through the fact that some financial market participants are not aware of the (hidden) debt that countries owe to China. Lastly, the coefficient of the oil income effect indicates that reductions in oil prices are positively correlated with peak COVID-19 7 In a previous version of the paper, we estimated additional specifications as a robustness check and re- sults are available upon request. A brief summary: We augmented specification 3 with mobility data as a proxy for economic activity and mobility’s positive coefficient was highly statistically significant and the model’s R-squared jumped to 19.25%. However, economic intuition would suggest that higher mobility (i.e. more economic activity) would generally be associated with with smaller residuals during peak COVID-19, not larger ones. The unexpected sign may be related to non-linearities and threshold effects in the relationship between mobility and economic activity. In fact, Nouvellet et al. (2021) show that virus transmission signifi- cantly decreased with the initial reduction in mobility in 73% of the countries they analysed, suggesting that mobility and economic activity were highly correlated at the very beginning of the pandemic. However, they found evidence of decoupling of transmission and mobility following the relaxation of strict control measures for 80% of countries. This suggest that over the course of the pandemic the relationship between mobility and economic activity broke down, which justifies removing mobility from the model. 148 residuals. A finding worth pointing out is that the F statistic of all three specifications is not statistically significant. Thus, we cannot reject the null that any of the groups of COVID- 19 variables (economic, epidemiological or policy) are jointly insignificant. However, the statistical insignificance of many of the coefficients individually as well jointly are not con- tradictory to our theoretical expectations. Indeed it is important to acknowledge that most coefficients take on the expected sign. This implies that the statistical insignificance of the results mainly comes from the relatively low degrees of freedom which in turn is the result of the short estimation window (March 2020), data unavailability for some country- variable combinations, and the use of time and country fixed effects. However, we show similar results over a longer period of time in section 4.4.4. Figure 4.12 provides a graphically summary by showing the economic significance of variables based on specification 3 of Table 4.6. The length of each bar is the product of the variable’s sample standard deviation with its coefficient estimate. The figure confirms that it is mostly traditional economic determinants of sovereign risk pricing that influenced residuals in March 2020. 4.4.3 Did country-specific or global factors dominate CDS spreads during COVID- 19? So far, the analysis uncovered that during COVID-19 global/regional risk drivers only partly explained CDS adjustments. We also showed that traditional factors such as the oil price effect and existing levels of sovereign debt are the country-specific factors which particularly help explain CDS adjustments in March 2020. However, we can extend the analysis by asking to which degree global/regional factors and country-specific ones more generally played a role in the CDS adjustment during peak COVID-19. To address this question, we compare the explanatory power of the global/regional factors to the explana- tory power of country-specific variables. 149 Model Specifically, we treat realized log changes in CDS spreads as the outcome variable and explain it through the model-implied values from Equation 4.1 along with country-specific explanatory variables as in Equation 4.3. This results in the following model: cds i;t =# i + t + \ cds i;t + X COVID i;t + X economy i;t +X policy i;t + i;t (4.4) March 1, 2020t March 31, 2020 where \ cds i;t =b i + b i + c i1 GCDS t + c i2 RCDS i 0 ;t are the model-implied values of the daily CDS spread changes. Essentially, in Equation 4.4 we take apart the two components that make up the residual. This way, it becomes obvious that the regression in Equation 4.3 with the residual as an outcome variable is equivalent to the regression in Equation 4.4 when we restrict = 0. Crucially, we don’t impose this restriction, which lends to a richer analysis of the relative contribution of CDS adjustments made by global/regional and country-specific drivers. Results and interpretation The results of Equation 4.4 are reported in Table 4.7. First, note that in specification 1 the coefficient for the fitted daily CDS spread change is highly statistically significant and of considerable magnitude. Specifically, a one percentage point increase in the CDS spread changes as predicted by the global/regional factor model alone is associated with approx- imately a 0.57 percentage point change of actual CDS spreads. This makes intuitively sense as we would expect global/regional factors to continue driving spreads during peak COVID-19. After including economic, policy, and epidemiological variables in specifi- 150 cation 3, the coefficient of model-implied values remains statistically significant and its magnitude increases to approximately 0.6. This implies that once we control for country- specific drivers of spreads, there is a relatively strong association of global/regional spread changes and the changes of an individual country, meriting the model selection in the first stage. Second, adding mortality variables in specification 2 only increases the explanatory power of the model from 4% to 6.9%. Intuitively, we would expect that higher mortality growth rates would indicate higher CDS spreads, all else equal, but the coefficient on the cumulative mortality growth rate is negative. However, none of the mortality variables are statistically significant. Presumably, this is the result of confounding variables that are not controlled for in specification 2. Lastly, interaction terms with announcements of fiscal responses are insignificant, both statistically and in their economic magnitude. This indicates that neither the presence of such announcements nor the existence of higher external debt burdens significantly in- creased the perceived risk. This result is interesting: even governments with higher exter- nal debt levels or large and potentially unsustainable stimuli were able to get away with it for now. This suggests a form of “COVID-19 dominance” at the height of the pandemic in that investors primarily worried about the effects of an immediate economic drop rather than the longer-term effects of higher debt burdens. In summary, the results above suggest that country-specific variables generally were the more dominant drivers of CDS spread adjustments during peak COVID-19 than global/regional factors. Including country-specific factors increases the explanatory power of the model from less than 4% to more than 10.5%, which is a relatively high value for high-frequency macro-financial regressions. Of the country-specific variables, economic factors seem to be the more important drivers of spread changes than epidemiological ones. 8 8 We also estimated specification 3 of Table 7 with an added mobility variable. Including it increases the R-squared to 21.16%. While all coefficients except the one for SWF/GDP keep its signs, many become statistically significant. However, the highly statistically significant coefficient of mobility again takes on a positive sign, which isn’t in line with a priori expectations. 151 4.4.4 Does COVID-19 mortality affect CDS spreads after controlling for immediate economic ramifications of the pandemic? So far, our analysis has only tangentially addressed the question whether the epidemio- logical severity affected CDS spreads once the economic challenges through lockdowns as well as declining oil prices are taken into account. To tackle this question, we use a panel which is consistent with Equation 4.4 but over a longer stretch of time. Hence, it also uses global/regional and country-specific factors to explain CDS spreads but it covers the entire period between January 2014 and June 2020. The analytical appeal behind the proposed empirical strategy is evident: Choosing a panel estimator that explains CDS spread changes through global/regional and country- specific factors—particularly epidemiological factors—allows us to test the significance of the independent effect of mortality dynamics on spreads. If there is no independent effect of mortality on spreads, then this would suggest that there is no “pandemic effect.” In other words, the entire adjustment of sovereign risk pricing over the course of the pandemic would be attributed to traditional economic drivers. However, if we observe a significant effect of mortalities on spreads, then this opens up two interpretations, depending on the sign of the coefficient. One interpretation of a positive association of mortalities and spread changes could be that investors are concerned about intensifications or extensions of lockdowns. The eco- nomic intuition: Higher mortality numbers may engender longer or more stringent lock- downs which would increase the need for additional stimulus and the debt burden and thus the likelihood of sovereign default. In contrast, a negative association could be interpreted as a positive productivity shock in the eyes of financial market participants. Specifically, economies with relatively high numbers of mortalities will need to spend relatively less of their productive resources on the elderly in the medium to longer term. This would alleviate strains on government finances caused by health and retirement expenses, lowering the expected probably of sovereign 152 default. While this may seem like a morally reprehensible interpretation, the economic argument draws on a paper by Young (2005) with the provocative title “The Gift of the Dying: The Tragedy of AIDS and the Welfare of Future African Generations.“ This paper identifies two competing effects of the AIDS epidemic. On the one hand, it finds that the epidemic has a detrimental impact on human capital accumulation of orphaned children as their education is permanently interrupted at the time of their parents’ death. On the other hand, it finds that widespread community infection lowers fertility through both a reduction in the willingness to engage in unprotected sexual activity and by increasing the scarcity of labor and hence the value of a woman’s time. The study concludes that the fer- tility effect dominates even under the most pessimistic assumptions concerning reductions in educational attainment. Thus, AIDS is an example of a viral disease that despite bring- ing immeasurable horror upon people may have had some positive long-term economic effects on future generations. As COVID-19 primarily affects older generations, a some- what similar mechanism may unfold: Specifically, it may be that the relative shrinking of older generations may lead to reduced strains on government budgets and lowered expected probabilities of sovereign default. Model We use the following model that combines global/regional and country-specific factors to address the question whether and how mortalities affect CDS spreads: cds i;t = i cds i;t1 + i1 GCDS t + i2 RCDS i 0 ;t +X COVID i;t + X economy i;t +X policy i;t + i;t (4.5) January 1,2014 t< July 1,2020 153 Data modifications While we can generally use the same data and variables that we introduced in section 4.4.2 and used to estimate Equation 4.4, some modifications are necessary. This is to account for the fact that the reporting of some variables only begins in 2020 while we need them for January 2014 to June 2020. Specifically, we make the following modifications for COVID- 19 case and mortality numbers: COVID-19 epidemiological measures such as daily cases and mortalities and their growth rates are back-filled with the number “0” for all observa- tions before January 22, 2020. The intuition behind this is that the JHU CSSE COVID-19 reporting begins on January 22, 2020. Thus, it is innocuous to assume that prior to that date there were no (known) COVID-19 cases or mortalities. Results and interpretation The results of Equation 4.5 are reported in Table 4.8. Let’s first look at specification 1 which is essentially model (1) but over a longer period that covers the pandemic. All three coefficients are highly statistically significant, their signs are as expected, and the model has notable explanatory power with an R-squared of 20.26%. Specifically, the lagged CDS change has a negative sign, which indicates a mean reverting process and the positive signs of the global and regional factor indicate positive relationships of CDS spreads across the globe. Also note that a one percentage point increase in CDS spreads of the core countries (USA, Japan, and Eurozone countries) is associated with a 2.3 percentage point increase in emerging markets. This suggests that increases in CDS spreads of the global economy’s core countries have a magnifying effect on emerging markets (along the line of “when the core sneezes the rest of world catches a cold”). There is also a positive relationship between an emerging market’s CDS spread and the spreads of its regional peers which may indicate some form of regional economic contagion and dependence. This effect also remains significant when we construct the regional factor differently in specification 2. There, we investigate how CDS spreads co-move with the spreads of all emerging market 154 peers when we use bilateral trade shares as the weights to construct the factor instead of using GDPs of local regional peers as the weights. Now that we have established that the simple global/regional factor model does a good job at modeling CDS spreads over this longer period, let’s look at specification 3 which shows the results of a panel estimator consistent with Equation 4.4. First, note that the first three coefficients all remain highly statistically significant and barely change in magnitude when compared to the global/regional model. Also, consider Figure 4.13 that shows the products of the estimated coefficients and the standard deviations of the variables and suggests that the global and regional factors remain dominant factors after adding country-specific factors. Interestingly, note that both ratios of international reserves/GDP and external debt/GDP are not significant over the duration of the longer panel. This may surprise at first as one would expect that higher debt ratios would lead to higher CDS spreads and vice versa for reserve ratios. However, this finding is in line with the relative insignificance of these variables for the peak COVID-19 period and it may be reconciled further by pointing out that both these variables are relatively slow moving so that the country fixed-effects already swallow some of the cross-country variance. Second, note that SI growth is statistically significant and shows the expected sign: Economic intuition would suggest that to fight off the pandemic, more sever restrictions needed to be put in place in some countries. This then hinders public life, afflicts GDP and increases spreads. However, the magnitude of the coefficients is relatively small. For example, a one percentage point increase in the growth of SI is associated with an increase in CDS spreads of 0.46 percentage points. Third, the coefficient for the fiscal response dummy interacted with stimulus size as a share of GDP is positive and statistically significant. This is consistent with economic intuition in two ways: In the medium to long run, comparatively bigger stimulus packages add to the future debt burden. This increases the expected likelihood of default and pushes CDS spreads up. In the short run, investors may consider higher stimulus packages as a 155 signal of a relatively more severe economic crisis and as such start asking questions about debt sustainability, which pushes up CDS spreads. At the same time, the magnitude of this coefficient is considerable: it suggests that a one percentage point increase in stimulus is associated with approximately a 2.3% percentage point increase in spreads. However, we should keep in mind the relatively narrow distribution of the size of fiscal packages which becomes visible when we multiply the coefficient with the standard deviation of the variable in the sample (see Figure 4.13). This relatively small product shows that the stimulus packages on their own are not a main driver of spreads over a longer period. Next, consider the coefficient for the growth of the cumulative mortality rate. It is statistically significant and shows that a one percentage point increase in the growth rate of the cumulative mortality rate is associated with an increase in spreads of about 0.84 percentage points. Considering that it shows the detrimental effect on spreads after the immediate effects of the economic downturn resulting from lockdowns and the effects of lower oil prices have been taken into account, the effect is of considerable magnitude and suggests that investors closely monitor COVID-19 mortality numbers, making mortalities a temporary focal point during the pandemic. There are a number of complementary in- terpretations of the positive association: One interpretation is that investors are concerned about imminent intensification of lockdowns, which would strain sovereign budgets and increase spreads. Another interpretation of the positive relationship is that investors are worried about longer-term consequences of the pandemic; either they worry that it leads to sub-trend growth for a longer time because of prolonged lockdowns or they worry that the pandemic will ultimately lead to increased bankruptcies, degradation of human and physical capital, and other social costs, all slowing GDP recovery. Ultimately, why and how exactly mortalities drive spreads once economic drivers are accounted for remains an empirical question for now and only time and the future availability of additional data will allow a more nuanced interpretation. Lastly, a comparison of Table 4.6/Figure 4.12 with Table 4.8/Figure 4.13 shows that the 156 main results of the paper, including the relative importance of economic variables versus epidemiological variables, hold for the entire sample of 30 countries in a single regression analysis of daily CDS spread changes during January 2014 to June 2020. 4.5 Conclusion Understanding the effect of COVID-19 on emerging market sovereigns is critical given pre- vious debt crises and the enormous pressure that the pandemic puts on sovereign finances. In light of this, this paper analyses the drivers of sovereign CDS spreads of thirty emerging market sovereigns. Our main findings can be summarized in five points: First, a model that uses global/regional factors to model spread changes traces the re- alized spread changes well for the period before the pandemic. This observation suggests that emerging market spreads are driven largely by global/regional factors in risk-on envi- ronments. Second, the relationship between actual spread changes and their changes predicted by the factor model breaks down during the pandemic. This suggests that spread determinants are of time-varying nature and that investors weigh country-specific factors more heavily in their decision-making process in risk-off environments. Third, actual spread changes experience the biggest deviations from model-implied values in March 2020, suggesting that uncertainties around ramifications of the pandemic were highest at the peak of the first wave. Fourth, spreads at peak COVID-19 (March 2020) were primarily driven by traditional country-specific factors of sovereign solvency such as fiscal space and oil income depen- dence rather than epidemiological policies and dynamics. This suggests, maybe surpris- ingly, that the severity of the pandemic in terms of mortalities in a specific country didn’t influence investors’ behavior much, at least not during peak COVID-19. Fifth, over the entire pandemic period, however, the growth in the cumulative mortality rate is significantly positively associated with CDS spread changes after taking the eco- 157 nomic ramifications of lockdowns and the oil price decline into account. This suggests that investors either feared a near-term intensification of lockdowns or a longer-term continua- tion of lockdowns, and possibly a slower GDP growth recovery. 158 4.6 Paper appendix 0 5000 10000 2014201620182020 Argentina 200 300 400 500 600 2014 2016 2018 2020 Bahrain 100 200 300 400 500 2014 2016 2018 2020 Brazil 50 100 150 2014 2016 2018 2020 Chile 25 50 75 100 125 150 2014 2016 2018 2020 China 100 200 300 2014 2016 2018 2020 Colombia 300 350 400 450 2014 2016 2018 2020 Dominican Republic 300 400 500 600 2014 2016 2018 2020 Egypt 500 1000 2014 2016 2018 2020 Ghana 100 200 2014 2016 2018 2020 Hungary 50 100 150 200 250 2014 2016 2018 2020 Indonesia 100 200 300 2014 2016 2018 2020 Kazakhstan 50 100 150 200 250 2014 2016 2018 2020 Malaysia 100 150 200 250 300 2014 2016 2018 2020 Mexico 50 100 150 200 2014 2016 2018 2020 Panama 50 100 150 200 2014 2016 2018 2020 Peru 60 100 140 2014 2016 2018 2020 Philippines 60 80 100 2014 2016 2018 2020 Poland 50 100 150 2014 2016 2018 2020 Qatar 100 150 2014 2016 2018 2020 Romania 200 400 600 2014 2016 2018 2020 Russia 50 100 150 200 2014 2016 2018 2020 Saudi Arabia 200 300 400 500 2014 2016 2018 2020 South Africa 300 350 400 450 500 2014 2016 2018 2020 Sri Lanka 200 300 400 500 600 2014 2016 2018 2020 Turkey 0 5000 10000 15000 2014201620182020 Ukraine 100 200 300 400 500 2014 2016 2018 2020 Uruguay 75 100 125 150 2014 2016 2018 2020 India 50 100 150 2014 2016 2018 2020 Thailand 40 50 60 2014 2016 2018 2020 Czechia Figure 4.1: 5-year sovereign CDS spreads of emerging markets, January 2014-June 2020. Data source: Eikon Refinitiv. Argentina Bahrain Brazil Chile China Colombia Czechia Dominican Republic Egypt Ghana Hungary India Indonesia Kazakhstan Malaysia Mexico Panama Peru Philippines Poland Qatar Romania Russia Saudi Arabia South Africa Sri Lanka Thailand Turkey Ukraine Uruguay Table 4.1: List of the thirty emerging markets in the large sample. 159 0 10 20 30 Brazil South Africa Russia Mexico Colombia Indonesia Uruguay Kazakhstan Ukraine Chile Turkey Hungary Thailand Czechia Poland Argentina India Malaysia Peru Sri Lanka Romania Dominican Republic China Ghana Philippines Bahrain Saudi Arabia Panama Qatar Egypt (%) Figure 4.2: Changes in exchange rates of emerging markets against the US Dollar between January and March 2020. Positive values indicate a weakening of the EM currency. Data source: Eikon Refinitiv. 0 5 10 Qatar Brazil Poland Thailand Peru Chile Panama Bahrain Kazakhstan Indonesia Argentina India Turkey Hungary China Philippines Malaysia Russia Colombia Romania Egypt Mexico Dominican Republic Uruguay Sri Lanka South Africa Ukraine Ghana Fiscal Stimulus Announced (% of GDP) Figure 4.3: Announced 2020 COVID-19 related fiscal stimulus of emerging market coun- tries. COVID-19 fiscal stimulus data taken from the IMF COVID policy tracker. Japan (12.8%), 2019 GDP in million USD: 5,081,770 US (53.9%), 2019 GDP in million USD: 21,427,700 Eurozone (33.3%), 2019 GDP in Million USD: 13,264,737 Germany, France, Greece, Ireland, Belgium, Spain, Netherlands, Austria, Cyprus, Estonia, Italy, Latvia, Lithuania, Malta, Portugal, Slovenia, Slovakia, Finland Table 4.2: Weights of countries in the global factor are in brackets. The weights are con- structed by dividing a country’s 2019 GDP by all 20 countries’ combined 2019 GDP. Lux- emburg is not included in the Eurozone as CDS data was unavailable. 160 0.0000 0.0002 0.0004 0.0006 02−2020 03−2020 04−2020 05−2020 06−2020 07−2020 New Mortality Rate (%) Bahrain Brazil China Panama Peru Qatar Romania Sri Lanka Thailand Turkey Figure 4.4: COVID-19 mortality rate curves for the top and bottom five emerging markets as per the end of April 2020 (based on 30 country sample). New mortality rate as 7-day rolling averages. Data source: JHU CSSE. 0 100 200 300 01−2020 04−2020 07−2020 Total Deaths Per Million Bahrain China Hungary Panama Peru Qatar Romania Sri Lanka Thailand Turkey Figure 4.5: COVID-19 deaths per million residents for the top and bottom five emerging markets as per the end of April 2020 (based on 30 country sample). Data source: JHU CSSE. 161 −5.0 −2.5 0.0 2.5 Panama Hungary Ghana Uruguay Thailand Philippines Mexico Chile Czechia Kazakhstan Colombia India China Indonesia Malaysia Argentina Qatar South Africa Russia Ukraine Sri Lanka Brazil Dominican Republic Romania Poland Egypt Turkey Saudi Arabia Bahrain Change in int. reserves in March/April 2020 % of 2019 GDP Figure 4.6: Change in international reserve holdings of emerging markets between March and April 2020 as a share of 2019 GDP. Data source: IMF. Peru not pictured due to data unavailablility. 7500 10000 12500 Jan Feb Mar Apr May Argentina 200 300 400 500 Jan Feb Mar Apr May Bahrain 100 200 300 Jan Feb Mar Apr May Brazil 40 80 120 160 Jan Feb Mar Apr May Chile 30 40 50 60 70 80 Jan Feb Mar Apr May China 100 200 300 Jan Feb Mar Apr May Colombia 300 330 360 390 Jan Feb Mar Apr May Dominican Republic 300 400 500 600 Jan Feb Mar Apr May Egypt 300 400 500 600 700 Jan Feb Mar Apr May Ghana 50 60 70 80 Jan Feb Mar Apr May Hungary 50 100 150 200 250 Jan Feb Mar Apr May Indonesia 60 80 100 120 Jan Feb Mar Apr May Kazakhstan 50 100 150 Jan Feb Mar Apr May Malaysia 100 150 200 250 300 Jan Feb Mar Apr May Mexico 60 90 120 150 Jan Feb Mar Apr May Panama 40 80 120 160 Jan Feb Mar Apr May Peru 60 100 140 Jan Feb Mar Apr May Philippines 50 55 60 Jan Feb Mar Apr May Poland 50 100 150 Jan Feb Mar Apr May Qatar 90 120 150 Jan Feb Mar Apr May Romania 100 200 300 Jan Feb Mar Apr May Russia 50 100 150 200 Jan Feb Mar Apr May Saudi Arabia 200 300 400 500 Jan Feb Mar Apr May South Africa 420 450 480 Jan Feb Mar Apr May Sri Lanka 300 400 500 600 Jan Feb Mar Apr May Turkey 500 750 1000 Jan Feb Mar Apr May Ukraine 100 150 200 250 300 Jan Feb Mar Apr May Uruguay 74.48 74.52 74.56 74.60 Jan Feb Mar Apr May India 20 40 60 80 100 Jan Feb Mar Apr May Thailand 35 36 37 38 39 40 Jan Feb Mar Apr May Czechia Figure 4.7: Development of sovereign 5-year CDS spreads of emerging market economies between January 2020 and April 2020. Data source: Eikon Refinitiv. 162 10 15 20 25 Jan Feb Mar Apr May Germany 15 25 35 45 55 Jan Feb Mar Apr May France 100 200 300 400 Jan Feb Mar Apr May Greece 20 30 40 50 60 Jan Feb Mar Apr May Ireland 20 30 40 50 Jan Feb Mar Apr May Belgium 40 80 120 160 Jan Feb Mar Apr May Spain 10.0 12.5 15.0 17.5 20.0 Jan Feb Mar Apr May Netherlands 10 15 20 25 Jan Feb Mar Apr May Austria 99.950 99.975 100.000 100.025 100.050 Jan Feb Mar Apr May Cyprus 56 60 64 Jan Feb Mar Apr May Estonia 100 150 200 250 Jan Feb Mar Apr May Italy 56 60 64 Jan Feb Mar Apr May Latvia 58 60 62 64 Jan Feb Mar Apr May Lithuania 208.72 208.74 208.76 208.78 208.80 208.82 Jan Feb Mar Apr May Malta 40 80 120 160 Jan Feb Mar Apr May Portugal 55 60 65 Jan Feb Mar Apr May Slovenia 35 40 45 50 Jan Feb Mar Apr May Slovakia 12 15 18 21 Jan Feb Mar Apr May Finland 14 16 18 20 22 24 Jan Feb Mar Apr May US 20 30 40 Jan Feb Mar Apr May Japan Figure 4.8: Development of sovereign 5-year CDS spreads of advanced economies between January 2020 and April 2020. Data source: Eikon Refinitiv. Africa Central Asia East Asia Europe LATAM Middle East South Asia Egypt Kazakhstan China Czechia Argentina Bahrain Sri Lanka Ghana Russia Indonesia Hungary Brazil Qatar India South Africa Malaysia Poland Chile Saudi Arabia Philippines Romania Colombia Thailand Turkey Dominican Republic Ukraine Mexico Panama Peru Uruguay Table 4.3: Classification of large sample into geographic groups. These groups are used to calculate the regional factors in the first-stage regression. Example for China: Indonesia’s weight is [Indonesia GDP / (GDP of Indonesia, Malaysia, Philippines, Thailand]. 163 Actual Predicted −0.25 0.00 0.25 0.50 0.75 Jul 2019 Oct 2019 Jan 2020 Apr 2020 Jul 2020 Cumulative Change (Log CDS) EM Average CDS Spreads, Out−of−Sample Period 0.0 0.2 0.4 0.6 Jul 2019 Oct 2019 Jan 2020 Apr 2020 Jul 2020 Standard Deviation EM CDS Spreads Dispersion Low Mortality High Mortality −0.2 0.0 0.2 0.4 0.6 Jul 2019 Oct 2019 Jan 2020 Apr 2020 Jul 2020 Cumulative Change (Log CDS) High Vs. Low COVID Mortalities: Actual Low Mortality High Mortality −0.2 −0.1 0.0 0.1 Jul 2019 Oct 2019 Jan 2020 Apr 2020 Jul 2020 Cumulative Residuals High Vs. Low COVID Mortalities: Actual−Fitted Figure 4.9: Emerging market spread development, July 2019- June 2020 (reduced sample). 0 1 2 Jul Oct Jan Argentina −0.3 −0.2 −0.1 0.0 Jul Oct Jan Ghana −0.6 −0.4 −0.2 0.0 Jul Oct Jan Hungary −0.2 −0.1 0.0 Jul Oct Jan Kazakhstan −0.3 −0.2 −0.1 0.0 Jul Oct Jan Poland −0.3 −0.2 −0.1 0.0 Jul Oct Jan Romania −0.100 −0.075 −0.050 −0.025 0.000 0.025 Jul Oct Jan Sri.Lanka −0.4 −0.3 −0.2 −0.1 0.0 Jul Oct Jan Ukraine −0.2 −0.1 0.0 Jul Oct Jan India −0.20 −0.15 −0.10 −0.05 0.00 0.05 Jul Oct Jan Czechia Figure 4.10: Actual (solid) vs fitted values (dashed) before COVID-19 (July 2019 to Jan- uary 2020). These are the ten countries for which the correlation coefficient between actual vs fitted values are below 0.25. That is, the model does not do a good job at predicting the actual values out-of-sample. Therefore, these countries are removed from the full sample and won’t be used for the second-stage regressions. 164 Dependent Variable: CDS i;t CDS i;t1 GCDS i 0 ;t RCDS i 0 ;t R-squared CDSi;t; \ CDSi;t Out-of-sample Remains in sample Kazakhstan 0.058** 1.605** 0.67*** 0.208 0.027 No Sri Lanka 0.101*** -0.07 0.066** 0.016 0.049 No Argentina -0.034 -5.067*** 1.355*** 0.078 0.075 No Ghana -0.079*** -0.094 0.179*** 0.013 0.109 No India 0 0.006 0.007 0 0.139 No Hungary 0.025 2.615*** 0.353*** 0.16 0.141 No Romania -0.151*** 2.695*** 0.164*** 0.098 0.153 No Czechia -0.07*** 1.827*** 0.207*** 0.104 0.154 No Ukraine -0.002 -1.709 1.033*** 0.024 0.230 No Poland -0.023 2.582*** 0.298*** 0.091 0.232 No Uruguay -0.171*** -0.786 0.673*** 0.061 0.260 Yes Bahrain -0.157*** 0.996 0.26*** 0.044 0.284 Yes Egypt -0.007 0.437 0.139*** 0.011 0.426 Yes Turkey 0.007 -0.115 1.204*** 0.305 0.582 Yes Saudi Arabia 0.021 2.278*** 0.652*** 0.117 0.599 Yes Thailand -0.013 2.899*** 0.78*** 0.271 0.613 Yes Dominican Republic -0.126*** 0.02 0.252*** 0.043 0.63081716 Yes Qatar 0.096*** 0.738 0.537*** 0.129 0.663 Yes China -0.004 4.128*** 1.094*** 0.335 0.673 Yes Russia -0.039* -0.33 1.338*** 0.262 0.682 Yes Philippines -0.002 3.222*** 1.008*** 0.325 0.694 Yes Malaysia 0.001 3.999*** 1.05*** 0.303 0.704 Yes South Africa -0.066*** 0.233 1.352*** 0.441 0.726 Yes Chile -0.091*** -0.456 1.729*** 0.486 0.732 Yes Indonesia 0.01 3.041*** 1.065*** 0.344 0.741 Yes Mexico -0.12*** -0.69 1.701*** 0.46 0.789 Yes Brazil -0.101*** -2.219*** 1.483*** 0.37 0.800 Yes Panama -0.1*** 0.597 1.534*** 0.516 0.812 Yes Peru -0.085*** -0.283 1.635*** 0.507 0.826 Yes Colombia -0.081*** -0.413 1.759*** 0.543 0.838 Yes Note: Country-specific time-series regression estimates from Equation 4.1. *, **, *** correspond to 5%, 1% and 0.1% significance, respectively. Number of daily observations per country, T, equal to 2,005. Out-of-sample statistic shows the correlation coefficient between actual CDS spread changes and model-predicted changes for the period outside the estimation sample but before the global outbreak of COVID-19 (July 2019 to February 2020). Table 4.4: First-stage regression results of model (1) over the period January 2014 to June 2019. Column 6 provides country-specific out-of-sample correlation coefficients between the actual CDS changes and model-implied CDS changes. Countries dropped from full sample Countries in reduced sample Argentina, Ghana, Hungary, Kazakhstan, Poland, Romania, Sri Lanka, Ukraine, India, Czechia Bahrain, Brazil, Chile, China, Colombia, Dominican Republic, Egypt, Indonesia, Malaysia, Mexico, Panama, Peru, Philippines, Qatar, Russia, Saudi Arabia, South Africa, Turkey, Uruguay, Thailand Table 4.5: List of 20 emerging markets that constitute our reduced sample. The 20 countries were obtained after deleting 10 countries from the full sample based on low correlation coefficients between actual vs fitted values over the pre COVID-19 out-of-sample period July 2019 to February 2020. 165 −0.1 0.0 0.1 0.2 Jan 01 Jan 15 Feb 01 Feb 15 Mar 01 Daily Change (Log CDS) Actual Fitted Emerging Markets Average CDS Spreads, Jan 2020 − Feb 2020 −0.1 0.0 0.1 0.2 Mar 02 Mar 09 Mar 16 Mar 23 Mar 30 Daily Change (Log CDS) Actual Fitted Emerging Markets Average CDS Spreads, March 2020 −0.1 0.0 0.1 0.2 Apr May Jun Jul Daily Change (Log CDS) Actual Fitted Emerging Markets Average CDS Spreads, Apr 2020 − June 2020 Figure 4.11: Emerging market average COVID-19 residual (20 country sample). Sovereign Wealth Fund volume/GDP Oil Income Price Effect Ext. Debt/GDP Cumulative Mortality Rate ECB Policy Dummy ** International Reserves/GDP New Mortality Rate Cumulative Mortality Rate Growth * Debt owed to China/GDP Fed Policy Dummy Ext. Debt/GDP X Fiscal Response Dummy Debt owed to China/GDP X Fiscal Policy Dummy New Mortality Rate Growth SI Growth 0.00 0.05 0.10 0.15 Sign of coefficient negative positive Figure 4.12: The x-axis shows the product of the coefficient estimate and the standard deviation of the variable as absolute values over the sample period. The coefficients are from Table 4.6 model (3). Significance levels are attached next to variable names. 166 Dependent variable: CDS spread COVID Residual (1) (2) (3) New Mortality Rate 0.0285 0.0287 0.1388 (0.0391) (0.0393) (0.1704) New Mortality Rate Growth 0.0027 0.0025 0.0020 (0.0051) (0.0051) (0.0055) Cumulative Mortality Rate 0.0071 0.0073 0.0307 (0.0092) (0.0091) (0.0394) Cumulative Mortality Rate Growth 0.0278 0.0275 0.0407 (0.0202) (0.0208) (0.0245) SI Growth 0.0092 0.0063 (0.0194) (0.0224) ECB Policy Dummy 0.0882 (0.0388) Fed Policy Dummy 0.0183 (0.0261) Ext. Debt/GDP 0.0018 (0.0018) Ext. Debt/GDP x Fiscal Policy Dummy 0.0006 (0.0006) Debt owed to China/GDP 0.0077 (0.0403) Debt owed to China/GDP x Fiscal Policy Dummy 0.0103 (0.0117) Oil income effect 0.0175 (0.0302) International Reserves/GDP 0.1521 (0.2914) Sovereign Wealth Fund volume/GDP 0.5109 (1.1034) Fixed effects? Y Y Y Observations 173 171 153 R 2 0.0124 0.0137 0.0647 F Statistic 0.3781 0.3298 0.4644 Note: *,**,*** correspond to 10%, 5% and 1% significance, respectively. HAC robust standard errors, clustered by country. Time and Country FEs. Table 4.6: Analysis of daily peak COVID-19 residuals (March 2020, 20 country sample). 167 Dependent variable: Daily CDS Spread Change (1) (2) (3) Fitted Daily CDS Spread Change 0.5684 0.6485 0.6049 (0.2698) (0.2620) (0.3461) New Mortality Rate 0.0298 0.1499 (0.0419) (0.1623) New Mortality Rate Growth 0.0020 0.0010 (0.0052) (0.0057) Cumulative Mortality Rate 0.0069 0.0310 (0.0098) (0.0367) Cumulative Mortality Rate Growth 0.0267 0.0393 (0.0196) (0.0234) SI Growth 0.0029 (0.0249) ECB Policy Dummy 0.0884 (0.0359) Fed Policy Dummy 0.0211 (0.0287) Ext. Debt/GDP 0.0019 (0.0016) Ext. Debt/GDP x Fiscal Policy Dummy 0.0004 (0.0006) Debt owed to China/GDP 0.0076 (0.0391) Debt owed to China/GDP x Fiscal Policy Dummy 0.0086 (0.0120) Oil income effect 0.0182 (0.0292) International Reserves/GDP 0.1813 (0.2874) Sovereign Wealth Fund volume/GDP 0.5380 (1.0528) Fixed effects? Y Y Y Observations 620 173 153 R 2 0.0399 0.0687 0.1054 F Statistic 23.6398 1.7567 0.7303 Note: *,**,*** correspond to 10%, 5% and 1% significance, respectively. HAC robust standard errors, clustered by country. Time and Country FEs. Table 4.7: Analysis of daily peak COVID-19 CDS spread changes (March 2020, 20 country sample). 168 Dependent variable: CDS (1) (2) (3) (4) (5) CDS, lagged 0:0114 0:0354 0:0116 0:0144 0:0100 (0:0034) (0:0034) (0:0035) (0:0042) (0:0040) GCDS, weighted by core countries’ GDP 2:3390 1:3723 2:3277 2:5620 2:4820 (0:1357) (0:1392) (0:1412) (0:1799) (0:1779) RCDS, weighted by regional peers’ GDP 0:5635 0:5776 0:5628 0:6052 (0:0045) (0:0048) (0:0062) (0:0059) RCDS, weighted by 29 peers’ trade shares 0:7356 (0:0061) Cumulative Mortality Rate Growth 0:0084 0:0097 0:0090 (0:0034) (0:0042) (0:0043) SI Growth 0:0046 0:0095 0:0050 (0:0026) (0:0033) (0:0030) Fiscal Response Dummy X stimulus as % of GDP 0:0228 0:0288 0:0226 (0:0109) (0:0126) (0:0127) ECB Policy Dummy 0:0079 0:0068 0:0121 (0:0015) (0:0019) (0:0019) Fed Policy Dummy 0:0001 0:0005 0:0004 (0:0008) (0:0010) (0:0010) Oil Income Price Effect 0:1527 0:0967 0:0259 (0:0492) (0:0720) (0:0723) External debt as % of GDP 0:00001 (0:00001) Foreign currency debt as % of GDP 0:000000 (0:000000) Debt to China as % of GDP 0:0004 (0:0002) International reserves as % of GDP 0:0007 0:0018 0:0009 (0:0021) (0:0033) (0:0050) Fixed effects? Y Y Y Y Y Observations 70,742 70,742 66,040 47,194 49,529 R 2 0.2026 0.1889 0.2084 0.1703 0.1995 F Statistic 5,987.3740 5,488.3030 1,579.6010 880.0748 1,121.1270 Note: *,**,*** correspond to 10%, 5% and 1% significance, respectively. HAC robust standard errors, clustered by country. Country FEs. Table 4.8: Analysis of daily CDS spread changes (January 2014 to June 2020, 30 country sample). 169 Delta RCDS, weighted by regional peers' GDP *** Delta GCDS, weighted by core countries' GDP *** ECB Policy Dummy *** Oil Income Price Effect *** Delta CDS, lagged *** External debt as % of GDP Cumulative Mortality Rate Growth ** Fiscal Response Dummy X stimulus as % of GDP ** SI Growth * International Reserves as % of GDP Fed Policy Dummy 0.000 0.004 0.008 0.012 Sign of coefficient negative positive Figure 4.13: The x-axis shows the product of the coefficient estimate and the standard deviation of the variable as absolute values over the sample period. The coefficients are from Table 4.8 model (3). Significance levels are attached next to variable names. 170 CHAPTER 5 CONCLUSION This dissertation project represents a systematic approach to examine how country-specific decisions, structural forces in the international political economy, and credit rating agencies as transnational corporations affect a country’s financial standing. The project is made up of three papers that illuminate these links from different perspectives. The papers in chapter 2 & chapter 3 highlight how crucial it is for IPE scholars to put credit rating agencies at the center of the analysis when studying the political economy of government’s access to international capital markets. The paper in chapter 4, on the other hand, points out how path-dependence can render a country’s previous decisions either detrimental or conducive to its economic resilience to external shocks. In the rest of this concluding chapter, I reflect on the scholarly contributions which this dissertation project provides and outline avenues for future research. 5.1 Research contributions By looking at credit rating agencies as important players in the global economy, my dis- sertation project contributes to a deeper understanding of the determinants of countries’ financial standing. In particular, my project compiles a vast database that includes ratings from more agencies than any previous study. This extends the analysis of credit rating agencies to a global scale. At the same time, this extension mitigates uncertainty around a potential selection bias. The existing studies possibly suffer from such a bias as they focus only on agencies whose data is easily available or agencies which are well known. In addition, the project evaluates the effect of previously unexplored but increasingly prominent features of today’s global economy on ratings: swap lines. Investigating the effects of this variable on government’s financial standing constitutes an important step 171 towards a fuller understanding of global default dynamics. In particular, controlling for this variable might help to explain the relative absence of sovereign defaults in recent years despite numerous crises and dislocations in the global economy. Lastly, the article in chapter 4 is the first of its kind in analyzing the effect of a si- multaneous internal and external demand shock caused by a global pandemic on emerging markets. Thus, this article is a contemporary contribution to understanding the economic effects of COVID-19 which may help policy makers in their decision process. 5.2 Future research I hope that this dissertation provides the foundation for a sustained research agenda into the biases of credit rating agencies and the effects of ratings in the global economy. There are several extensions that could build on this project and that particularly stand out for attention in future research: The first potential extension is an adaption of my research design to the analysis of quasi-sovereign as well as corporate debt issuers. To the best of my knowledge, there are no articles that look at the rating determinants for such entities and at whether ratings for such issuers show home and in-group biases. Therefore, it appears as a promising avenue of research to analyse how ratings of quasi-sovereign entities and corporate issuers differ between agencies from different home countries. Are the same home and in-group biases visible in the data? Under what scope conditions? And what are possible implications? A second potential extension could focus on credit rating agencies in China. Specifi- cally, it would be fascinating to see if there are systematic differences between how foreign and Chinese agencies rate domestic Chinese debt issuers. A third extension of the project could focus on what it means for a sovereign not to be rated. Does the absence of a rating contain any information about the financial standing of the underlying sovereign? The matching method seems to be a fruitful research method to investigate this question. However, given the restrictions on the number of existing 172 countries (N 200) and the high-dimensionality of variables to match on, this method may not be applicable in this context. Instead, a qualitative paper on the effects of a rating withdrawal or sudden rating coverage could be developed. Specifically, a natural question would emerge in this area of research: why have CRAs begun to rate some sovereigns in the first place, given that the direct financial incentive structures for this choice are not apparent? Lastly, my dissertation project mostly focuses on the role of CRAs in allocating capital to sovereigns and the geopolitical determinants of sovereign ratings. Thus, it does not ex- amine why CRAs have emerged in some countries and not in others. After all, if sovereign credit ratings can be weaponized as tools of foreign policy—particularly by authoritarian countries—there is little reason not to expect government support and subsidies for this industry. Given that there are only few agencies globally, and even fewer ones that are of considerable prominence, a historical analysis of the emergence and survival of agencies could shed light on the market structure under which credit rating agencies operate. 173 Bibliography Afonso (2003). “Understanding the determinants of sovereign debt ratings: Evidence for the two leading agencies”. In: Journal of Economics and Finance 27.1, pp. 56–74. Afonso, Davide Furceri, and Pedro Gomes (2012). “Sovereign credit ratings and financial markets linkages: application to European data”. In: Journal of International Money and Finance 31.3, pp. 606–638. Ahmed, Shaghil, Brahima Coulibaly, and Andrei Zlate (2017). “International financial spillovers to emerging market economies: How important are economic fundamen- tals?” In: Journal of International Money and Finance 76, pp. 133–152. Aizenman, Joshua, Yothin Jinjarak, and Donghyun Park (2011). “International reserves and swap lines: Substitutes or complements?” In: International Review of Economics & Finance 20.1, pp. 5–18. Aizenman, Joshua and Gurnain Kaur Pasricha (2010). “Selective swap arrangements and the global financial crisis: Analysis and interpretation”. In: International Review of Economics & Finance 19.3, pp. 353–365. Alberola-Ila, Enrique, Yavuz Arslan, Gong Cheng, Richhild Moessner, et al. (2020). The fiscal response to the Covid-19 crisis in advanced and emerging market economies. Tech. rep. Bank for International Settlements. Almeida, Heitor, Igor Cunha, Miguel A Ferreira, and Felipe Restrepo (2017). “The real effects of credit ratings: The sovereign ceiling channel”. In: The Journal of Finance 72.1, pp. 249–290. 174 Alsakka and Owain Ap Gwilym (2010). “Leads and lags in sovereign credit ratings”. In: Journal of Banking & Finance 34.11, pp. 2614–2626. — (2012). “Foreign exchange market reactions to sovereign credit news”. In: Journal of International Money and Finance 31.4, pp. 845–864. Alsakka, Owain Ap Gwilym, et al. (2013). “The impact of sovereign rating actions on bank ratings in emerging markets”. In: Journal of Banking & Finance 37.2, pp. 563–577. Ammer, John and Frank Packer (2000). “How consistent are credit ratings? A geographic and sectoral analysis of default risk”. In: The Journal of Fixed Income 10.3, pp. 24–30. Archer, Candace C, Glen Biglaiser, and Karl DeRouen (2007). “Sovereign bonds and the “democratic advantage”: Does regime type affect credit rating agency ratings in the developing world?” In: International organization 61.2, pp. 341–365. Arezki, Rabah, Bertrand Candelon, and Amadou Nicolas Racine Sy (2011). “Sovereign rating news and financial markets spillovers: Evidence from the European debt crisis”. In: Attinasi, Maria Grazia, Cristina D Checherita-Westphal, and Christiane Nickel (2009). “What explains the surge in euro area sovereign spreads during the financial crisis of 2007-09?” In: Augustin, Patrick, Marti G Subrahmanyam, Dragon Yongjun Tang, and Sarah Qian Wang (2014). “Credit default swaps–A survey”. In: Foundations and Trends® in Finance 9.1-2, pp. 1–196. Babic, Milan, Jan Fichtner, and Eelke M Heemskerk (2017). “States versus corporations: Rethinking the power of business in international politics”. In: The International Spec- tator 52.4, pp. 20–43. Bahaj, Saleem and Ricardo Reis (2020a). “Central bank swap lines during the Covid-19 pandemic”. In: Covid Economics 2, pp. 1–12. — (2020b). “Jumpstarting an international currency”. In: Baker, Colleen (2013). “The Federal Reserve’s Use of International Swap Lines”. In: Ariz. 175 L. rev. 55, p. 603. Barro, Robert J and Jong-Wha Lee (2005). “IMF programs: Who is chosen and what are the effects?” In: Journal of monetary Economics 52.7, pp. 1245–1269. Block, Steven A and Paul M Vaaler (2004). “The price of democracy: sovereign risk ratings, bond spreads and political business cycles in developing countries”. In: Journal of International Money and Finance 23.6, pp. 917–946. Borensztein, Eduardo, Kevin Cowan, and Patricio Valenzuela (2013). “Sovereign ceilings “lite”? The impact of sovereign ratings on corporate ratings”. In: Journal of Banking & Finance 37.11, pp. 4014–4024. Brunnermeier, Markus K and Ricardo Reis (2019). A crash course on the euro crisis. Tech. rep. National Bureau of Economic Research. Calvo, Guillermo A (1988). “Servicing the public debt: The role of expectations”. In: The American Economic Review, pp. 647–661. Cantor, Richard, Frank Packer, et al. (1994). “The credit rating industry”. In: Quarterly Review, pp. 1–26. Cepni, Oguzhan, Doruk Kucuksarac, and M Hasan Yilmaz (2017). “The sensitivity of credit default swap premium to global risk factor: Evidence from emerging markets”. In: Economics Letters 159, pp. 74–77. Chen, Sheng-Syan, Hsien-Yi Chen, Chong-Chuo Chang, and Shu-Ling Yang (2013). “How do sovereign credit rating changes affect private investment?” In: Journal of Banking & Finance 37.12, pp. 4820–4833. Chey, Hyoung-kyu (2013). “The Fed Swap Lines and the Global Lender of Last Resort: The Politics of International Monetary Relations”. In: APSA 2013 Annual Meeting Paper. Cornaggia, Jess N, Kimberly J Cornaggia, and John E Hund (2017). “Credit ratings across asset classes: A long-term perspective”. In: Review of Finance 21.2, pp. 465–509. De Haan, Jakob and Fabian Amtenbrink (2011). “Credit rating agencies”. In: Handbook 176 of Central Banking, Financial Regulation and Supervision. Edward Elgar Publishing. Drago, Danilo and Raffaele Gallo (2017). “The impact of sovereign rating changes on the activity of European banks”. In: Journal of Banking & Finance 85, pp. 99–112. Dreher, Axel and Martin Gassebner (2008). “Does political proximity to the US cause terror?” In: Economics Letters 99.1, pp. 27–29. Durbin, Erik and David Ng (2005). “The sovereign ceiling and emerging market corporate bond spreads”. In: Journal of international Money and Finance 24.4, pp. 631–649. Eden, Lorraine (1991). “Bringing the firm back in: Multinationals in international political economy”. In: Millennium 20.2, pp. 197–224. Eichengreen, Barry and Ashoka Mody (1998). What explains changing spreads on emerging- market debt: fundamentals or market sentiment? Tech. rep. National Bureau of Eco- nomic Research. Feldstein, Martin and Charles Horioka (1979). Domestic savings and international capital flows. Tech. rep. National Bureau of Economic Research. Fender, Ingo, Bernd Hayo, and Matthias Neuenkirch (2012). “Daily pricing of emerging market sovereign CDS before and during the global financial crisis”. In: Journal of Banking & Finance 36.10, pp. 2786–2794. Ferri, Giovanni, L-G Liu, and Joseph E Stiglitz (1999). “The procyclical role of rating agencies: Evidence from the East Asian crisis”. In: Economic Notes 28.3, pp. 335– 355. Fitch Ratings, F (2018). Fitch Ratings–Rating Definitions. From www.fitchratings.com. French, Kenneth R and James M Poterba (1991). Investor diversification and international equity markets. Tech. rep. National Bureau of Economic Research. Friedman, Thomas L (1995). “Don’t mess with Moodys”. In: New York Times 22.February, A15. Fuchs, Andreas and Kai Gehring (2017). “The home bias in sovereign ratings”. In: Journal of the European Economic Association 15.6, pp. 1386–1423. 177 Gande, Amar and David C Parsley (2005). “News spillovers in the sovereign debt market”. In: Journal of Financial Economics 75.3, pp. 691–734. G¨ artner, Manfred, Bjorn Griesbach, et al. (2011). “PIGS or lambs? The European sovereign debt crisis and the role of rating agencies”. In: International advances in economic re- search 17.3, pp. 288–299. G¨ artner, Manfred and Bjorn Griesbach (2017). “Rating Agencies, Self-Fulfilling Prophecy and Multiple Equilibria? An Empirical Model of the European Sovereign Debt Crisis 2009-2011”. In: Business and Economic Research 7.1, pp. 199–226. Gehrig, Thomas (1993). “An information based explanation of the domestic bias in in- ternational equity investment”. In: The Scandinavian Journal of Economics, pp. 97– 109. Giannetti, Mariassunta and Yishay Yafeh (2012). “Do cultural differences between con- tracting parties matter? Evidence from syndicated bank loans”. In: Management Sci- ence 58.2, pp. 365–383. Gomez-Puig, Marta (2006). “Size matters for liquidity: Evidence from EMU sovereign yield spreads”. In: Economics Letters 90.2, pp. 156–162. Guiso, Luigi, Paola Sapienza, and Luigi Zingales (2006). “Does culture affect economic outcomes?” In: Journal of Economic perspectives 20.2, pp. 23–48. Horn, Sebastian, Carmen M Reinhart, and Christoph Trebesch (2019). China’s overseas lending. Tech. rep. National Bureau of Economic Research. Huberman, Gur (2001). “Familiarity breeds investment”. In: The Review of Financial Studies 14.3, pp. 659–680. Jaramillo, Laura and Michelle Tejada (2011). “Sovereign credit ratings and spreads in emerging markets: does investment grade matter?” In: IMF Working Papers, pp. 1–17. Jinjarak, Yothin, Rashad Ahmed, Sameer Nair-Desai, Weining Xin, and Joshua Aizenman (2020). “Accounting for global COVID-19 diffusion patterns, January–April 2020”. In: Economics of disasters and climate change 4.3, pp. 515–559. 178 Kamin, Steven B and Karsten V on Kleist (1999). “The evolution and determinants of emerging markets credit spreads in the 1990s”. In: Kocsis, Zalan and Zoltan Monostori (2016). “The role of country-specific fundamentals in sovereign CDS spreads: Eastern European experiences”. In: Emerging Markets Review 27, pp. 140–168. Laeven, Luc and Fabian Valencia (2020). “Systemic banking crises database II”. in: IMF Economic Review, pp. 1–55. Laidi, Zaki (2008). “How trade became geopolitics”. In: World policy journal 25.2, pp. 55– 61. Livingston, Miles, Winnie PH Poon, and Lei Zhou (2018). “Are Chinese credit ratings relevant? A study of the Chinese bond market and credit rating industry”. In: Journal of Banking & Finance 87, pp. 216–232. McCallum, John (1995). “National borders matter: Canada-US regional trade patterns”. In: The American Economic Review 85.3, pp. 615–623. Moran, Michael (1990). The politics of the financial services revolution: The USA, UK and Japan. Springer. Nouvellet, Pierre et al. (2021). “Reduction in mobility and COVID-19 transmission”. In: Nature communications 12.1, pp. 1–9. Obstfeld, Maurice, Jay C Shambaugh, and Alan M Taylor (2009). “Financial instability, reserves, and central bank swap lines in the panic of 2008”. In: american Economic review 99.2, pp. 480–86. Panizza, Ugo, Federico Sturzenegger, and Jeromin Zettelmeyer (2009). “The economics and law of sovereign debt and default”. In: Journal of economic literature 47.3, pp. 651–98. Reinhart, Carmen (2019). “The curious case of the missing defaults”. In: HKS Working Paper No. RWP19-004. Reinhart, Carmen, Vincent Reinhart, and Christoph Trebesch (2016). “Global cycles: Cap- 179 ital flows, commodities, and sovereign defaults, 1815-2015”. In: American Economic Review 106.5, pp. 574–80. Remolona, Eli M, Michela Scatigna, and Eliza Wu (2008). “The dynamic pricing of sovereign risk in emerging markets: Fundamentals and risk aversion”. In: The Journal of Fixed Income 17.4, pp. 57–71. Reuters (2011). France shocked by S&P downgrade error. URL: https://www.reuters.com/ (visited on 03/15/2021). Rey, H´ el` ene (2015). Dilemma not trilemma: the global financial cycle and monetary policy independence. Tech. rep. National Bureau of Economic Research. Romer, David (2012). Advanced Macroeconomics, 4th edn. McGraw-Hil Irwin. Schmukler, Sergio (1999). Emerging markets instability: do sovereign ratings affect coun- try risk and stock returns? The World Bank. Shiller, Robert J, Fumiko Kon-Ya, and Shunichi Tsutsui (1991). “Speculative behavior in the stock markets: evidence from the United States and Japan”. In: NBER working paper w3613. Shin, Yoon S and William T Moore (2003). “Explaining credit rating differences between Japanese and US agencies”. In: Review of Financial Economics 12.4, pp. 327–344. Sinclair, Timothy J (2014). The new masters of capital: American bond rating agencies and the politics of creditworthiness. Cornell University Press. Tesar, Linda L and Ingrid M Werner (1995). “Home bias and high turnover”. In: Journal of international money and finance 14.4, pp. 467–492. Tomz, Michael and Mark LJ Wright (2007). “Do countries default in “bad times”?” In: Journal of the European Economic association 5.2-3, pp. 352–360. Vernazza, Daniel R and Erik F Nielsen (2015). “The damaging bias of sovereign rat- ings”. In: Economic Notes: Review of Banking, Finance and Monetary Economics 44.2, pp. 361–408. V oeten, Erik (2019). “Data and analyses of voting in the United Nations General Assem- 180 bly”. In: Routledge handbook of international organization. Routledge, pp. 80–92. Williams, Andrew (2015). “A global index of information transparency and accountabil- ity”. In: Journal of Comparative Economics 43.3, pp. 804–824. Wolf, Holger C (2000). “Intranational home bias in trade”. In: Review of economics and statistics 82.4, pp. 555–563. 181

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on Japanese macroeconomy

PDF

Essays on monetary policy and international spillovers

PDF

Emerging market financial crises, investors and monetary policy

PDF

Growth and development in Africa: macroeconomic and microeconomic perspectives

PDF

The murky risk of trade protectionism in an interconnected and uncertain global economy: a state, industrial, and regional analysis

PDF

Economic policy-making in a changing environment: the politics of success in Peru vis-à-vis China

PDF

Three essays in international macroeconomics and finance

PDF

Essay on monetary policy, macroprudential policy, and financial integration

PDF

Inter-temporal allocation of human capital and economic performance

PDF

Examining global financial regulatory reform in the G20 and the FSB

PDF

Essays in macroeconomics and macro-finance

PDF

Anti-foreign boycotts as a tool of economic coercion: the case of China

PDF

Essays on innovation, human capital, and COVID-19 related policies

PDF

Fear or greed? The balance of relative interest under the dual hierarchical order in Asia-Pacific

PDF

Essays on political economy of privatization

PDF

Essays in international economics

PDF

Financial crises and trade policy in developing countries

PDF

Three essays on heterogeneous responses to macroeconomic shocks

PDF

Essays in political economy and mechanism design

PDF

Essays on political economy and corruption

Asset Metadata

Creator Dähler, Timo Basil (author)

Core Title Essays on sovereign debt

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Political Science and International Relations

Degree Conferral Date 2021-08

Publication Date 07/23/2021

Defense Date 04/16/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag CDS,COVID-19,credit tatings,Economics,international economics,International finance,international political economy,IPE,macroeconomics,OAI-PMH Harvest,sovereign credit ratings,sovereign debt

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Aizenman, Joshua (committee chair), Barberá, Pablo (committee member), Katada, Saori (committee member), Nugent, Jeffrey (committee member)

Creator Email daehler@usc.edu,timo.daehler@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15618741

Unique identifier UC15618741

Legacy Identifier etd-DhlerTimoB-9849

Document Type Dissertation

Format application/pdf (imt)

Rights Dähler, Timo Basil

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu