University of Southern California Dissertations and Theses

Three essays on heterogeneous responses to macroeconomic shocks

(USC Thesis Other)

Three essays on heterogeneous responses to macroeconomic shocks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THREE ESSAYS ON HETEROGENEOUS RESPONSES TO MACROECONOMIC SHOCKS
by
Juan Andres Espinosa-Torres
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2023
Copyright 2023 Juan Andres Espinosa-Torres
Acknowledgements
They say it takes a village to get there. Turns out that it takes more than that. Below I present
my thanks to all of the people who pushed me and helped me get through the completion of my
dissertation.
Specifically,IwanttothankProfessorPabloKurlatforhismentorshipandsupervisionthrough-
out this dissertation. I also want to thank Professor Monica Morlacco for her many helpful com-
mentsandsuggestions,aswellasforbelievinginmyideasfromanearlyphase. ThankstoProfessors
CarolineBetts,DavidArgente,RicardodelaO,andDavidZeke,andtheattendantsofthemacroe-
conomics reading group, the LACEA-LAMES 2022 conference, and the attendants of the UCSB
Fall seminar in applied microeconomics for their helpful feedback and engagement.
IalsowanttothankmyfriendsSebastianLobo,FranciscoPacheco,KarimFajury,DanielAngel,
and coauthors John Jairo Leon and Jaime Ramirez, whose continuous suggestions, encouragement,
and camaraderie helped me push through these years.
I also want to thank my therapist Amy Wiseman, whose guidance helped me grow and remain
resilient throughout this journey.
Last but not least, I want to give special thanks to my beloved mother Luisa, brother Juan
Sebastian and my dear Horus cuddling me from dog-heaven, for their immediate support and
company. This accomplishment is theirs as much as it is mine.
ii
Table of Contents
Acknowledgements ................................................................................ ii
List of Tables ...................................................................................... vi
List of Figures ..................................................................................... viii
Abstract............................................................................................ x
1 Chapter1Decodingtheinfluenceoffinancialconstraintsonretailproductstrategies
during economic cycles in the US .......................................................... 1
1.1 Introduction ......................................................................... 1
1.1.1 Related Literature ........................................................ 3
1.2 Data.................................................................................. 4
1.3 Results ............................................................................... 8
1.3.1 Relationship between financial constraints and product innovations . 8
1.3.2 Relationship between financial constraints and sales................... 13
1.4 Conclusion. .......................................................................... 21
2 Chapter 2 Redistributive Effects of Oil Shocks for Commodity Exporting Economies. 23
2.1 Introduction ......................................................................... 23
2.2 Data.................................................................................. 25
2.3 Empirical Evidence.................................................................. 27
2.3.1 Changes in Disposable Income........................................... 33
2.3.2 Changes in Energy Expenditure......................................... 39
2.3.3 The Fiscal Channel: government expenditure.......................... 43
2.4 Final Remarks....................................................................... 47
3 Chapter 3 The Effects of the Pandemic on Market Power and Profitability............ 48
iii
3.1 Introduction ......................................................................... 48
3.2 Data.................................................................................. 50
3.2.1 Outcomes of interest...................................................... 51
3.2.2 Entry and exit of stock exchange listings in Compustat............... 52
3.2.3 Historical trends in markup and profit rates ........................... 54
3.3 Methodology......................................................................... 56
3.3.1 Forecasting model......................................................... 58
3.3.2 Estimation................................................................. 59
3.3.3 Assumption testing, model fit, and quality of forecasts................ 60
3.4 Counterfactual analysis results..................................................... 62
3.4.1 Aggregate data............................................................ 62
3.4.2 Heterogeneity analysis.................................................... 65
3.4.3 Heterogeneity by industry................................................ 69
3.4.4 Business cycle analysis ................................................... 70
3.5 Robustness of findings to other modeling strategies and choices ................ 71
3.5.1 Robustnessoffindingstoothermodelingchoices: ahyper-parameter
sensitivity analysis. ....................................................... 72
3.5.2 Robustness of findings to other modeling strategies: local linear
projections................................................................. 73
3.6 Conclusion ........................................................................... 77
References.......................................................................................... 80
Appendices.................................................................................... 87
A Bias in aggregate Entry/Exit rates......................................................... 87
B Alternative specifications for Entry/Exit regressions. .................................... 89
B.1 Robustness to specification. ........................................................ 89
B.2 Robustness to crisis definition and aggregate financial distress.................. 95
C Alternative specifications for Sales regressions. ...........................................102
D Exogeneity of Oil Prices.....................................................................104
E Additional Tables and Figures..............................................................106
iv
F Data and summary statistics ...............................................................112
G Counterfactual analysis using cost-of-goods weighting....................................114
G.1 Forecast performance for BSTS and LLPs under a balanced sample............115
H Quarterly Analysis ..........................................................................115
H.1 Quarterly heterogeneity analysis...................................................118
Effect heterogeneity regressions ....................................................118
H.2 Tracking the fraction of firms with significant effects in the short-term.........121
Fraction of firms with significant effects of the pandemic, LLPs Model.........124
Fraction of firms with significant effects of the pandemic, Sensitivity Analysis 125
I Pandemic effect heterogeneity by quintiles of key firm characteristics...................127
v
List of Tables
1.1 Descriptive Statistics, Nielsen-GS1-Compustat ........................................... 6
1.2 Entry, Exit and financial constraints during the Great Recession ....................... 11
1.3 Sales dependency during the financial crisis. .............................................. 12
1.4 Decomposition of change in Sales........................................................... 16
1.5 Decomposition of change in Sales for quality groups...................................... 19
2.6 Descriptive Statistics. ....................................................................... 27
2.7 Effect of Oil Price on Government Transfers.............................................. 44
2.8 Effect of oil price shocks on Government Expenditure.................................... 46
3.9 Priors on Bayesian Structural Model ...................................................... 60
3.10 FractionofPeriodsthatObservedOutcomeFallsOutsideofBayesianCredibleIntervals 61
3.11 Forecast Quality Statistics of Bayesian Structural Time Series Model.................. 62
3.12 Heterogeneity in the effects of the pandemic on markups (Yearly Specification)....... 67
3.13 Heterogeneity in the effects of the pandemic on profit rates (Yearly Specification).... 68
3.14 Sensitivity Analysis on Hyperparameters of Bayesian Structural Model................ 72
3.15 Forecast Quality Statistics of Local-Linear Projections .................................. 76
B.16 Alternative specifications for Liq
ft− 4
....................................................... 90
B.17 Alternative specifications for Lev
ft− 4
. ..................................................... 91
B.18 Alternative specifications for Age
ft
. ....................................................... 92
B.19 Alternative specifications for Liq
ft− 1
....................................................... 93
B.20 Alternative specifications for trailing 12 month moving average of Liq
ft
. .............. 94
B.21 Alternative specifications for Lev
ft− 1
. ..................................................... 95
B.22 Relationship between Entry and Exit for Liq
ft− 4
, using different crisis measures...... 97
B.23 Relationship between Entry and Exit for Lev
ft− 4
, using different crisis measures. .... 98
B.24 Relationship between Entry and Exit for Age
ft
, using different crisis measures........ 98
B.25 Relationship between Entry and Exit for Liq
ft− 4
, using measures of financial distress. 99
B.26 Relationship between Entry and Exit for Lev
ft− 4
, using measures of financial distress.100
B.27 Relationship between Entry and Exit for Age
ft
, using measures of financial distress..101
C.28 Robustness of decomposition of change in Sales for Liq
ft− 4
. ............................103
vi
C.29 Robustness of decomposition of change in Sales for Lev
ft− 4
.............................104
D.30Oil Price event analysis......................................................................105
D.31Event dates ..................................................................................105
F.32 Summary Statistics..........................................................................113
G.33Forecast Quality Statistics Across Models, Balanced Sample. ...........................116
H.34Heterogeneous Effects of the pandemic on Markup Rates (Quarterly Specification)...119
H.35Heterogeneous Effects of the pandemic on Profit Rates (Quarterly Specification) .....120
vii
List of Figures
1.1 Average entry and exit rates. .............................................................. 7
1.2 Average entry and exit rates versus constraints. ......................................... 9
1.3 Average sale decomposition under financial constraints. ................................ 14
1.4 Relationship between Liq
ft− 4
and Sale components by Liq
ft− 4
quintiles............... 18
1.5 Relationship between Liq
ft− 4
and Sale components by quality quintiles................ 20
2.6 Mexican Oil prices........................................................................... 26
2.7 Effects of oil price shocks to consumption expenditure. .................................. 29
2.8 Effects of oil price shocks to consumption expenditure by Quartiles .................... 30
2.9 Effects of oil price shocks to consumption expenditure by sector........................ 32
2.10 Effects of oil price shocks to consumption expenditure by occupation .................. 33
2.11 Effects of oil price shocks to disposable income. .......................................... 34
2.12 Effects of oil price shocks to disposable income by Quartiles ............................ 36
2.13 Effects of oil price shocks to disposable income by sector................................ 37
2.14 Effects of oil price shocks to disposable income by occupation .......................... 38
2.15 Effects of oil price shocks to energy expenditure
E
Y− T
.................................... 39
2.16 Effects of oil price shocks to
E
Y− T
by Quartiles ........................................... 41
2.17 Pass-through of crude-oil prices to retail gasoline prices.................................. 43
3.18 Entry and Exit Rates ....................................................................... 53
3.19 Average Markup and Profit Rates for Entrant and Exiting firms........................ 53
3.20 Markup and Profit Rates ................................................................... 55
3.21 Markup and Profit Rate Volatility......................................................... 56
3.22 Comparison of Observed and Counterfactual Markup Rates using Bayesian Struc-
tural Time Series Models ................................................................... 63
3.23 Comparison of Observed and Counterfactual Profit Rates using Bayesian Structural
Time Series Models.......................................................................... 64
3.24 Effects on Markups and Profit Rates by Industry ........................................ 69
3.25 Histogram of Coefficient of the Cycle in Models (Yearly Frequency).................... 70
viii
3.26 Comparison of Observed and Counterfactual Markups and Profit Rates Under Dif-
ferent Hyperparameter choices.............................................................. 73
3.27 Comparison of Observed and Counterfactual Profit Rates using Local Linear Pro-
jections ....................................................................................... 75
3.28 Comparison of Observed and Counterfactual Profit Rates using Panel Local Linear
Projections................................................................................... 77
A.29Average entry and exit rates on each end of the sample. ................................ 87
A.30Average entry and exit rates versus constraints, industry medians. .................... 88
C.31 Average sale decomposition under financial constraints, industry medians. ...........102
E.32 Effects of oil price shocks to consumption expenditure, three binned Quartiles........107
E.33 Effects of oil price shocks to consumption expenditure, three binned Deciles...........108
E.34 Effects of oil price shocks to disposable income, three binned Quartiles ................109
E.35 Effects of oil price shocks to disposable income, three binned Deciles...................110
E.36 Effects of oil price shocks to
E
C
by Quartiles ..............................................111
G.37ComparisonofObservedandCounterfactualMarkupandProfitRates, weightingby
Cost of Goods Sold (COGS) ...............................................................115
H.38Counterfactual Analysis of Average and Quartiles of Markup Rates....................117
H.39Counterfactual Analysis of Average and Quartiles of Profit Rates ......................118
H.40FractionofFirmswithSignificanteffectsofthepandemic, BayesianStructuralTime
Series Model .................................................................................122
H.41Fraction of Firms with Significant effects of the pandemic, LLPs Model ...............124
H.42Sensitivity Analysis: Fraction of Firms with Significant effects of the pandemic
(Quarterly Specification)....................................................................126
I.43 Heterogeneity Analysis of Effect of Pandemic in Markups (Yearly Specification)......129
I.44 HeterogeneityAnalysisofEffectofPandemicinScaledMarkups(YearlySpecification)130
I.45 Heterogeneity Analysis of Effect of Pandemic on Profit Rate (Yearly Specification)...131
I.46 Heterogeneity Analysis of Effect of Pandemic in Markups (Quarterly Specification) ..132
I.47 Heterogeneity Analysis of Effect of Pandemic in Scaled Markups (Quarterly Specifi-
cation)........................................................................................133
I.48 Heterogeneity Analysis of Effect of Pandemic in Profit Rate (Quarterly Specification)134
ix
Abstract
The first essay, which I used as my job market paper, documents two adjustment margins used
by firms facing financial constraints during the Great Recession: (i) changes in product entry and
exit and (ii) the components of sales variation. Using Nielsen Homescan, GS1, and the Compustat
datasets, andfollowingpanelestimationtechniques, Ifindthefollowingstylizedfacts: (i)firmsfac-
ing lower liquidity constraints, measured as a higher liquid to total assets, introduce a new product
on average during the Great Recession, an increase of roughly 1.2%; (ii) adjustments in product
specificyearlysalespercentagechangesaredrivenbyprices, withsalesremaininvariantwithprices
increasing about 1% without statistically significant movement on the customer base or quantities.
These responses mask heterogeneity across the quality distribution: financially constrained firms
increasetheirlower-qualityproductpricesby1.8%whilereducingtheirhigh-qualityproductprices
by roughly 7% in order to capture customers trading down and exploiting locked-in customers
buying the low-quality varieties. These findings clarify the role of creative destruction on economic
activity during downturns, as well as explain the source of changes in sales by their components.
The second essay coauthored with John Jairo Leon studies how Commodity prices can have
significant adverse distributional effects on the consumption and income of households, even in
commodity-exporting economies. The discretionary income channel dominates the immediate re-
sponse of consumption to changes in oil prices: a 1 percent increase in oil prices reduces average
consumption by 0.18 percent, mainly driven by low and middle-wealth households. This result is
also partially driven by an increase in expenditure on energy-related products for all income quar-
tiles(0.05percentonaverage). Householdswhoseconsumptionisnotaffectedliewithinthesecond
and fourth quartile of the wealth distribution and see an increase in their disposable income (0.08
percent and 0.14 percent) that they seemingly use to hedge their increase in energy consumption.
However, one year after the oil price shock, consumption increased by about 0.3 percent across all
income quartiles. This is consistent with a redistributional channel driven by a procyclical fiscal
policy. Using an instrumental variable (IV) design, the analysis in this paper finds a 1.1 percent
increase in government expenditure due to a 1 percent increase in oil-related incomes produced by
an oil price shock.
Finally, the third essay coauthored with Jaime Ramirez explores firm-level markup and profit
x
rates during the COVID-19 pandemic for a panel of 3,611 publicly traded firms in Compustat
and finds increases for the average firm. We offer conditions to give markups and profit rate
forecasts a causal interpretation of what would have happened had the pandemic not happened.
Our estimations suggest that had the pandemic not happened, markups would have been 4% and
7% higher than observed in 2020 and 2021, respectively, and profit rates would have been 2.1
and 6.4 percentage points lower. We perform a battery of tests to assess the robustness of our
approach. We further show significant heterogeneity in the impact of the pandemic on firms by key
firm characteristics and industry. We find that firms with lower than forecasted markups tend to
have lower stock-exchange tenure and fewer employees.
xi
1 Chapter 1 Decoding the influence of financial constraints on
retail product strategies during economic cycles in the US
1.1 Introduction
Foralongtime,economistsidentifiedproductinnovationasamechanismthroughwhichcompanies
contribute to economic activity, with these agents also playing a key role in the propagation of
aggregate shocks (Clementi and Palazzo, 2016). In particular, firm entry is pro-cyclical, with
entrants having smaller impacts on economic activity, and the older incumbents driving most of
the variation. Interestingly, It is commonly thought that these entrants are the main responsible
force behind innovation. However, recent evidence shows that new technologies can also take place
in large firms as well (Yang, 2022). This suggests that both the firm and product entry and exit
play a key role in economic activity.
The movements on firm entry and exit have been thoroughly documented, but the product
innovation process itself remains a matter of debate among researchers. A recent paper by Argente
et al. (2018) documents the pro-cyclical behavior of aggregate product reallocation during the
Great Recession, showing that roughly 8% of products in the total economy either entered or
exitedquarterlyintheeconomy, mostlywithinincumbentfirms. Thesefindingsareconsistentwith
those of those by Hottman et al. (2016), who show that cannibalization is a prominent driver of
firm heterogeneity and plays a role in its placement within markets.
As relevant as this channel has proven to be, these studies have concerned themselves with the
aggregate dynamics, overlooking the role of financially constrained firms. As studied by multiple
authors(Gilchristetal.,2017;Kim,2021),firmsfacingfinancialconstraintsduringeconomicdown-
turns play a role in explaining some puzzling dynamics of these variables, such as the low inflation
observed during the Great Recession and the apparent pro-cyclical behavior of markups and their
relationship with their customer base (Nekarda and Ramey, 2020). In this paper, I study the role
of financial constraints during the Great Recession in product entry and exit rates and explore the
changes in product sales to reconcile these aggregate dynamics.
To achieve this, I use a combined dataset of firm balance sheet data and scanner data to
identify product entry and exit rates, sales, customer base size, and quantity and price movements
1
for each of the products these companies sell. The analysis of these components is done in two
steps. First, I study the co-movements between multiple measures of financial constraints with
those of observed product entry and exit rates at the firm level, and characterize the contribution
of entrant, incumbent, and salient products into the aggregate firm sales, denoting this as the
extensive margin. The second step focuses on analyzing the product-level data for studying the
variation of product-specific sales, and each of its components: customers, quantities per customer,
and prices, denoting this as the sales margin.
Three main stylized facts arise from following this approach. First, companies with higher
liquidity use product entry as a mechanism to generate cash flow and seldom retire products. This
results in the introduction of roughly 1 new product per quarter for the median firm. However,
because new products are difficult to place, the increase in sales remains modest, with entrant
productscontributingabout1.2%ofthetotalquarterlysales. Second,firmswithhigherliquiditysell
moreduringarecession–anaverageofroughly8%moreperproduct. Thisvariationisdrivenmainly
by preserving the customer base and increasing prices, taking advantage of locked-in customers.
Third, although sizable, price increases averaging 1% are modest in absolute terms for the average
product. Thus, these variations seem to be untraceable for customers, for they do not respond to
thesechanges, thusremainingasourceofadditionalcashflowsthatcompaniescanexploittohedge
the consequences of economic downturns.
2
1.1.1 Related Literature
My analysis intersects with three main strands of literature. First, studies have shown the close
relationship between firm financial constraints and their role in economic downturns (Gilchrist and
Zakrajˇ sek, 2012; Gilchrist et al., 2017; Cloyne et al., 2018). In the face of a recession, companies
make adjustments to their investment, employment, and production plans, usually by borrowing
money or using their liquid assets to stay in business. However, firms lacking this resource rely on
differentadjustments,suchasfiringworkersorincreasingprices,thereforetakingadvantageoftheir
sticky customer base (Chodorow-Reich, 2014; Gourio and Rudanko, 2014) to gain some short-term
cash-flows needed to hedge the shock Gilchrist et al. (2017); Kim (2021). In this regard, I go one
step ahead by decomposing the responses of each component of sales.
Second, there is a set of studies focused on characterizing multi-product firms and the implica-
tionsoftheirbehavioroneconomicactivity. SincetheseminalworkofBrodaandWeinstein(2010)
documenting that a significant share of product creation and destruction in U.S. industries occurs
within existing firms, Hottman et al. (2016) did a follow-up study showing that variation in firm
appeal and product scope explains at least four-fifths of the variation in firm sales. Moreover, Min-
nitiandTurino(2013)studiestheamplificationofbusinesscyclepropagationthroughthischannel.
In this regard, I study product entry and exit during economic downturns for these firms and show
the existence of significant responses in firms that rely on innovation to produce additional sales.
The final stream of literature is concerned with determining the role of consumer preferences in
the productive process of firms. Among others, the main findings are: household preferences drive
firms to protect and consolidate their customer base (Hopenhayn, 1992), consumer inertia drives
firms to lower their markups at first, and increase them as they consolidate their consumer base
(Bornstein, 2018), most product innovation does not come from patenting (Argente et al., 2020)
butpatentinghelpsfirmstoconsolidatetheircustomerbase. Systematicdifferencesinconsumption
acrossbothincomeandagegroupsallowforfirmstoconsolidatetheirpositioninthemarket(Faber
and Fally, 2017; Michelacci et al., 2019; Neiman and Vavra, 2019), with higher income households
demanding higher quality goods. In this paper, I characterize heterogeneous responses on the sales
variations across product quality distribution.
In summary, this paper contributes to the literature by being the first to study the intersection
3
of the role of financially constrained firms with both the product entry/exit and sales margin
jointly, studying movements in the customer base and its role within its co-movement with that
of aggregate sales. The forthcoming sections document the data, report the results of my analysis
and derive conclusions.
1.2 Data
I combine three different sources: Nielsen Homescan Panel (HMS), Compustat, and GS1. The first
one provides granular data on bar-codes sold to a representative panel of households for each of
their shopping trips within the United States, which allows me to track the introduction and exit
of products, calculate sales and construct price indices. The second source provides a quarterly
panel of balance sheet data for publicly traded firms, representing roughly 30% of US employment
(De Loecker et al., 2020). The final source (GS1) is a registry provided by the sole institution
allowed to issue bar-codes, and includes company name to bar-code linkages, providing links of the
bar-code data to the firm’s balance sheets.
The HMS data features household trip data on pricing, volume, and demographics spanned
by roughly 55,000 yearly participating households between the years 2004-2020 that cover approx-
imately $430 Million in yearly sales, representative of the following shares of the US bar-code
universe: 53% of grocery stores, 55% within drug stores, 32% of mass merchandisers, 2% in conve-
nience stores, and 1% in liquor stores in the United States. Each bar-code (upc) is organized under
four levels of aggregation: brands, product modules, groups, and departments. For example, the
ice cream ”BSK-RB Gold Medal Ribbon” is sold under the ”Baskin-Robbins” brand, belongs to the
”Ice-Cream, Bulk” module, ”Ice Cream” group and falls under the ”Frozen Foods” department
1
.
For the purpose of my analysis, a product is defined as an individual upc.
I aggregate product information to the quarterly frequency, firstly by computing quantity-
weighted averages of all the bar codes within a household, and secondly by using the population
weightsprovidedbyNielsentocalculate sales, prices, quantities, and thenumber ofcustomersthat
bought each bar code i at quarter t. In addition, product entry (exit) is defined as the first (last)
quarter in which a product was sold to at least one of the households. Because this measure will
1
The dataset spans a total of 10 departments, 118 groups, 1,133 modules, 490,213 brands, and 2,268,683 upc
bar-codes
4
incorrectly classify entry (exit) within the first (last) period of the overall sample, I exclude both
years 2004-2005 and 2019-2020
2
.
To measure firm’s conditions and financial constraints, I use the publicly available quarterly
balance sheet data from Compustat. Specifically, by selecting the variables from Gilchrist et al.
(2017); Ottonello and Winberry (2020); Cloyne et al. (2018) for my analysis, namely: sales (sale),
costofgoodssold(cogs), administrativeexpensesandadvertisement(xsga), totalassets(at), short
term and long term debts (dlc and dltt), age measured as the date when the firm was firstly
publicly traded, short-term investments (ivst) and cash (ch), deflating all the variables using the
GDP deflator, using 2010 as the base year.
I follow the standard procedure to organize and clean the data, namely: (i) removing all firms
with negative sales, cogs, and sg&a; (ii) records below the first and above the 99th quarterly
percentiles of the cogs and sg&a to sales ratios are eliminated, as well as observations without
either Compustat firm identification numbers or industry codes; (iii) I eliminate values below the
first and above the 99th percentile of two cost-shares (the ratio of cogs to the total of cogs and
capital expenditure, and the ratio of cogs to the total of cogs, capital expenditure, and sg&a);
as a final step, (iv) records that meet either of the following two conditions are eliminated: the
ratio of cash plus short term investments to total assets is outside the 0 and 1 range, or the yearly
percentage variation in sales and cogs exceeds 200% in absolute terms.
Using the GS1-Nielsen dataset, I match about 80% of the bar codes to a firm. Then, following
the fuzzy match described in Argente et al. (2018)
3
, match 1,632 Compustat firms from the total
54,254 universe of companies identified from the Nielsen-GS1 dataset. These represent about 49%
and50%ofthepanelobservationsandsales, respectively. Table1.1showssomesummarystatistics
of the resulting Nielsen-GS1-Compustat linked datasets. Clearly, non-publicly traded firms are
usuallythesmallest, aswellasthosefacingthetightestfinancialconstraints. Therefore, theresults
of my analysis provide a conservative estimate associated with the (usually) top-performing firms.
2
See Appendix A for more details.
3
I thank David Argente for his comments and for sharing the company codes (gvkey) codes he matched for
validation purposes.
5
Table 1.1: Descriptive Statistics, Nielsen-GS1-Compustat
(a) Firm level data
mean sd p10 p25 p50 p75 p90
∆ log(Sales
ft
) 0.02 0.21 -0.14 -0.05 0.02 0.09 0.19
∆ log(Cogs
ft
) 0.02 0.22 -0.15 -0.06 0.02 0.10 0.20
∆ log(Sg&a
ft
) 0.02 0.21 -0.15 -0.06 0.02 0.09 0.20
∆ log(Emp
ft
) 0.01 0.19 -0.12 -0.04 0.00 0.06 0.15
Liquidity ratio (Liq
ft
) 0.12 0.13 0.01 0.02 0.07 0.16 0.28
Leverage ratio Lev
ft
0.29 0.28 0.01 0.12 0.26 0.39 0.57
Firm age (years) 23.31 10.08 7.75 15.00 26.25 31.25 35.25
Product sold by firm ( N
ft
) 223.80 784.01 1 2 9 62 540
Product entry rate (ER
ft
) 0.09 0.21 0.00 0.00 0.00 0.08 0.25
Product exit rate (XR
ft
) 0.10 0.21 0.00 0.00 0.02 0.08 0.30
Net product entry rate (ER
ft
− XR
ft
) -0.01 0.26 -0.15 -0.02 0.00 0.00 0.11
Share of firm entry 0
Share of firm exit 0.02
Total f× t 14,646
(b) Product level data
mean sd p10 p25 p50 p75 p90
Salesit (thousands $) 573.08 1933.83 5.42 23.91 103.44 435.26 1377.42
Number of customers (Mit, thousands) 108.48 316.52 1.43 5.82 21.43 84.50 264.90
Average sales (per customer) 7.26 15.00 1.63 2.55 4.25 7.74 13.99
Quantities per customer (qit) 51.23 245.96 1.11 6.00 17.62 39.53 101.03
Product Price (pit) 3.70 48.31 0.05 0.10 0.23 0.69 4.10
Product age (years) 4.98 3.70 1.25 2.00 3.75 7.00 10.75
Quality 0.91 0.73 0.24 0.48 0.76 1.10 1.63
∆ log(Salesit) -0.15 1.47 -1.85 -0.71 -0.07 0.48 1.39
∆ log(Mit) -0.15 1.36 -1.73 -0.66 -0.07 0.41 1.26
∆ log(qit) 0.01 0.42 -0.41 -0.14 0.00 0.14 0.43
∆ log(pit) 0.00 0.27 -0.20 -0.06 0.00 0.08 0.20
Total i× t 2,994,223
Notes: Entry and exit rates are computed as ER
ft
=
E
ft
N
ft
and XR
ft
=
X
ft
N
ft
, where E
ft
is the number of products that
were introduced by the firm at time t, X
ft
is the products that exited and N
ft
is the total products offered by the firm f.
A Firm is labeled as an entrant when it first became publicly traded and exited whenever it stopped being publicly traded.
the ∆ operator denotes yearly variations and the sample covers the years 2006-2018. Quality is measured as in Jaimovich
et al. (2020), where a sales-weighted average price is computed across modules, and the index is defined as the relative price
of the product to its module average. The liquidity and leverage ratios are defined as
Cash
ft
+Short term Investments
ft
Assets
ft
and
Short term Debt
ft
+Long term Debt
ft
Assets
ft
, respectively.
6
On average, we see that product entry and exit rates are about 10%, with Net entry being a
negative 1%. This provides some stylized evidence of the cannibalization practices described by
(Hottman et al., 2016; Argente et al., 2019). As Figure 1.1 also suggests, this falling trend may be
closely related to the fall in business dynamism documented by Autor et al. (2020), as older and
most productive firms remain, these tend to innovate less. This is also shown in the average age of
the firms used in this dataset, featuring companies that have been around for more than 23 years,
since this reports time passed since the company became publicly traded.
Figure 1.1: Average entry and exit rates.
.08 .09 .1 .11 .12 .13 .14 .15 .16 .17
(%)
2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1
Entry rate Exit rate
Note: this figure plots average entry and exit rates, computed as ER
ft
=
E
ft
N
ft
and XR
ft
=
X
ft
N
ft
, where E
ft
is the
number of products that were introduced by the firm at time t, X
ft
is the products that exited and N
ft
is the total
products offered by the firm f. I compute entry and exit rates per firm and aggregate these using simple averages.
Interestingly, the average exit rate became higher than the entry rate during the crisis without
thetrendrevertingatall,withtheformerfluctuatingandthelatterremainingmoreorlessconstant
throughout this period. The financial crisis brought turmoil in markets that forced companies to
reallocatetheirresourcestostayafloatandkeepcashflows. Partoftheseadjustmentsareconsistent
with the literature on financial constraints (Gilchrist and Zakrajˇ sek, 2012; Gilchrist et al., 2017),
wherecompaniesadjustsomeoftheirmarginstoremaininbusiness. Motivatedbythis,Istudythe
7
relationship between financial constraints and the adjustment margins of the firms in the following
section.
1.3 Results
In theforthcoming sections, I studythe role played by financial constraints over two main margins:
the product creation/destruction margin and the influence of these over the sales margin and its
components.
1.3.1 Relationship between financial constraints and product innovations
Product innovation is costly, mainly because product placement is difficult as it requires the use
of resources such as advertising expenditures, R&D, and often, the cost of goods sold for new
products could be potentially high. With that being said, movements in the customer base could
drive companies to retire products if these turn out to be too costly to be produced relative to
the revenue these may generate. Usually, companies make use of disposable resources, such as
cash or they borrow money to invest in their development. This suggests a link between financial
constraints (or lack thereof) and product introduction/destruction rates.
I use the following measures of financial constraints: the liquidity ratio proposed by Gilchrist
et al. (2017), that is, Liq
ft
≡ Cash
ft
+Short term Investments
ft
Assets
ft
. This variable measures the percentage
of assets that are immediately available plus the number of short-term investments that could be
used. Larger values of this variable suggest that a firm is more liquid. Secondly, I use the leverage
ratioproposedbyOttonelloandWinberry(2020), Lev
ft
≡ Short term Debt
ft
+Long term Debt
ft
Assets
ft
, whose
largervaluessuggestthatfirmsholdcurrentlylowerassetsrelativetotheirtotaldebts. Intheshort
term,thiscouldmeanthatacompanyisfacingfinancialdifficultiesandisrelyingondebttocoverits
losses. Finally, I use the company age as in Cloyne et al. (2018) as a predetermined variable, which
proxies for the life-cycle of the firm. As they suggest, older firms face larger leverage ratios, give
out more dividends, issue more bonds, and have better credit scores, but are less liquid, because of
capital accumulation, as well as other long-term assets. Using the liquidity ratio, Figure 1.2 shows
a stylized look into the potential link between constraints and innovation
4
.
4
I conduct a robustness exercise using industry medians in Appendix A.
8
Figure 1.2: Average entry and exit rates versus constraints.
(a) Entry Rates
.06 .07 .08 .09 .1 .11 .12 .13 .14 .15 .16 .17 .18
(%)
2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1
Non−Constrained Liquidity Constrained
(b) Exit Rates
.06 .07 .08 .09 .1 .11 .12 .13 .14 .15 .16 .17 .18
(%)
2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1
Non−Constrained Liquidity Constrained
Note: To remove sources of bias, I exclude the first and last two years of the sample. Entry and exit rates are
computedasER
ft
=
E
ft
N
ft
andXR
ft
=
X
ft
N
ft
, whereE
ft
isthenumberofproductsthatwereintroducedbythefirmat
time t, X
ft
is the products that exited and N
ft
is the total products offered by the firm f. I compute entry and exit
rates per firm and aggregate using simple averages. I define a Firm as Liquidity constrained if its trailing 12-month
average of the Liquidity ratio is under the median at time t, as in Gilchrist et al. (2017).
Interestingly, publicly traded companies facing higher constraints (lower immediate resources)
face both lower entry and exit products. This suggests that their overall product reallocation rates
are lower overall. It is also worth noticing that average entry rates fell during the Great Recession
and they slowly recovered, only to keep a fluctuating pattern that remains slightly staggered,
whereas exit rates started out steady and then, they kept climbing. As such, understanding the
net allocation of products is fuzzy (at least from initial inspection). Therefore, to assess the role of
financial constraints over this margin, I propose the following panel regression design, for a firm f
at quarter t:
Y
ft
= (1− crisis
t
)× friction
ft− 4
β 0
+crisis
t
× friction
ft− 4
β 1
+Z
ft
δ +η f
+ϵ ft
(1)
whereY
ft
isanoutcomevariable. Here,Iconsiderentry

E
ft
N
ft

,exit

E
ft
N
ft

andnetentry

E
ft
− X
ft
N
ft

rates. Z
ft
are firm controls, which include the yearly growth rate of firm sales, cost of goods sold,
9
and yearly variation in sg&a. The first two control for short-term changes in demand, whereas
sg&a serves as a proxy for fixed costs. crisis
t
is a dummy that takes a value of 1 for the periods in
which the Great Recession happened as determined by NBER
5
, η f
and ϵ ft
correspond to firm fixed
effectscontrollingforcompany-specificsteady-statecharacteristics, andanerrorterm, respectively.
Friction
ft
is one of the financial friction measures used before, using last-year values (except for
age), as these reflect the firm situation at the moment that innovation decisions are made. In
particular, the development process lasts as less as 6 months and up to 4 years, with an average
development time of one year for a single product (Griffin, 1993). Therefore, β 0
measures the effect
of facing one standard deviation decrease of the financial frictions over product entry and exit
during normal times, while β 1
measures this relationship during the Great Recession, respectively.
Table 1.2 reports the estimates of (1). The results are mixed: all else equal, companies with
one standard deviation higher liquidity ratios increased their entry ratio by about 1.2% during the
financial crisis (column 1), without retiring incumbent products, resulting in an average increase of
2% in the net entry rate (column 3). The median firm sells 9 products, so this increase represents
the insertion of roughly a single new product, or a 10% increase relative to the mean entry rate
of 9%
6
. On the other hand, companies with one standard deviation larger leverage ratios seem to
remove1.4%oftheproductsonlyduringnormaltimes(column5), resultinginanetentrydecrease
of 1.7% (column 6). This suggests that in the presence of a crisis, firms put additional effort into
keeping their products in the market because, as poorly as they may be performing, they may be
providing needed cash flows to hedge the shock. Finally, the relationship with age shows that this
margin is the same regardless of the period, with entry rates falling roughly 6% (column 7) and
exit rates rising 8% (column 8), with a net entry fall of about 14% (column 9), as consistent with
the ”incumbent curse phenomenon” firstly introduced in Chandy and Tellis (2000).
5
I explore alternative crises definitions, measures of financial distress, controls, fixed effects, and lags, with the
results described in Appendix B.
6
Thiscalculationis doneasfollows: thenew rate increase is defined as
E
t+1
N
t+1
− 0.09. Column 1 of Table 1.2reports
this increase to be 0.012, then Et+1 = 0.102Nt+1. Using the median firm yields Et+1 = 0.92≈ 1, whereas if using
the average firm, it will be 22 .7≈ 23.
10
Table 1.2: Entry, Exit and financial constraints during the Great Recession
Liq
ft− 4
Lev
ft− 4
age
ft
(1) (2) (3) (4) (5) (6) (7) (8) (9)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Friction× Normal -0.001 0.000 -0.002 -0.003 0.014*** -0.017** -0.057*** 0.081*** -0.137***
(0.004) (0.004) (0.005) (0.005) (0.006) (0.007) (0.007) (0.008) (0.010)
Friction× Crises 0.012** -0.008 0.020** 0.003 0.010 -0.006 -0.059*** 0.088*** -0.147***
(0.006) (0.008) (0.009) (0.006) (0.006) (0.008) (0.007) (0.009) (0.012)
∆ log(Sales
ft
) 0.036** -0.016 0.051** 0.037** -0.019 0.056** 0.028 -0.003 0.031
(0.017) (0.015) (0.023) (0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.019 0.002 -0.021 -0.019 0.005 -0.024 -0.016 -0.003 -0.013
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) 0.000 -0.014* 0.014 -0.001 -0.012 0.011 -0.003 -0.011 0.008
(0.010) (0.008) (0.015) (0.011) (0.008) (0.015) (0.010) (0.008) (0.015)
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R
2
0.28 0.32 0.10 0.28 0.32 0.10 0.28 0.33 0.13
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%, and three
at the 1%. E
ft
is the number of products that were introduced by the firm at time t, X
ft
is the products that exited, and
N
ft
is the total products offered by the firm f. Financial friction and age variables are transformed to have a standard
deviation of 1 for comparison.
The previous findings suggest that companies who faced fewer financial constraints in the pre-
vious year decided to launch products during the financial crisis, potentially because these were
already in the process of development before and thus, introducing the product, even if it performs
poorly, it could bring some extra cash flows for the companies in the short term. I explore this by
running a regression of the share of sales due to new, exiting, and incumbent products. Namely, I
estimate
sh
ft
= (1− crisis
t
)× friction
ft− 4
γ 0
+crisis
t
× friction
ft− 4
γ 1
+Z
ft
δ +η f
+ϵ ft
(2)
11
where sh
ft
is the share of sales due to Old, New, and Incumbent products, and the other variables
are specified as before. Therefore, γ 0
pins the relationship between the financial friction and its
corresponding component of firm sales during normal times, and γ 1
measures this relationship
during the Great Recession. The shares are identified from the total sales within the product level
data aggregated to the firm quarter level, with the results reported on Table 1.3.
Table 1.3: Sales dependency during the financial crisis.
Liq
ft− 4
Lev
ft− 4
Age
ft
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Old New Incumbent Old New Incumbent Old New Incumbent
Friction× Normal -0.003 -0.001 0.004 0.009* -0.000 -0.009 0.059*** -0.035*** -0.024**
(0.003) (0.004) (0.005) (0.005) (0.004) (0.006) (0.007) (0.006) (0.009)
Friction× Crises -0.008 0.012* -0.004 0.005 0.003 -0.008 0.064*** -0.036*** -0.027***
(0.006) (0.006) (0.009) (0.006) (0.005) (0.007) (0.008) (0.007) (0.010)
∆ log(Cogs
ft
) -0.001 0.003 -0.003 -0.001 0.004 -0.004 0.002 0.002 -0.004
(0.009) (0.011) (0.014) (0.009) (0.011) (0.014) (0.009) (0.011) (0.014)
∆ log(Sg&a
ft
) -0.011 0.009 0.002 -0.010 0.008 0.002 -0.008 0.006 0.002
(0.008) (0.010) (0.012) (0.008) (0.010) (0.012) (0.008) (0.010) (0.012)
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.27 0.23 0.36 0.27 0.23 0.36 0.28 0.23 0.36
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5% and three
atthe1%. Eachcolumnreportstheresultsofaregressionoftheshareoftotalsalesduetonew, oldandincumbentproducts
as a function of the covariates. Financial friction and age variables are transformed to have a standard deviation of 1 for
comparison.
ConsistentwiththefindingsofTable1.2,weseethatfirmswithonestandarddeviationincrease
in liquidity see an increase of roughly 1% (column 2) of the total sales due to these new products.
Hence, the influence of this margin is modest, most likely because product placement is costly and
during bad times even more difficult to implement.
In general, we also observe that existing products are barely influenced by these constraints.
This is most likely due to the fact that old products already have a customer base that may have
12
beendecreasingovertime,assuch,firmschoosenottoretireproductsabruptlybutrather,letthem
”die-out” until the cost of producing these outweighs the gain in term of sales. Finally, observing
the role played by age seems to be consistent with the cannibalization and characteristics of older
firms: they innovate less, therefore they rely on older products as they age (their most successful)
and eventually are retired by the end of their life-cycle.
In summary, we see that financial constraints play a role in the firm innovation margin during
the Great Recession. However, the movement in sales due to the additional products introduced
by firms facing lower financial constraints is rather modest. This is most likely due to the fact
that product placement takes considerable time and effort, and placing products during a recession
requires even more of these. In the following section, I explore the sales margin and its components
in light of these constraints.
1.3.2 Relationship between financial constraints and sales.
As discussed above, firm adjustments could be done at different margins. In particular, companies
with higher liquidity introduce products during crises, this is potentially related to the findings
of Kim (2021), where liquidating inventories during a crisis can produce a cash flow needed for a
company to keep in business.
If companies with sufficient liquidity introduce products, they have incentives to sell them,
probablyatalowerpricetoproducesaidcashflows. Thissectionanalysestherelationshipbetween
financial constraints and sales. To do so, I exploit the granular sales by product level data from
Nielsen,andcomputethefollowingexactsales(fromNielsen)decompositionforproductiatquarter
t, in the spirit of Afrouzi et al. (2020):
log(Sales
it
)= log(M
it
)+log(q
it
)+log(p
it
) (3)
13
and take yearly differences,
log(Sales
it
)− log(Sales
it− 4
)
| {z }
∆ Sales
it
= log(M
it
)− log(M
it− 4
)
| {z }
∆ M
it
+log(q
it
)− log(q
it− 4
)
| {z }
∆ q
it
+log(p
it
)− log(p
it− 4
)
| {z }
∆ p
it
(4)
where M
it
is the number of customers who buy the product, p
it
is average price paid per customer
and q
it
are average quantities bought per customer. I average these measures across firms and plot
the decomposition in Figure 1.3
7
.
Figure 1.3: Average sale decomposition under financial constraints.
(a) Non Constrained
−.25 −.2 −.15 −.1 −.05 0 .05 .1
(%) change
2007q1 2008q1 2009q1 2010q1 2011q1
date
Sales M Q P
(b) Liquidity Constrained
−.25 −.2 −.15 −.1 −.05 0 .05 .1
(%) change
2007q1 2008q1 2009q1 2010q1 2011q1
date
Sales M Q P
Note: Data is averaged across firms. I define a firm as Liquidity constrained if its trailing 12-month average of the
Liquidity ratio is under the median at time t, as in Gilchrist et al. (2017).
It is clear that during the Great Recession, firms increased their prices (Gilchrist et al., 2017),
but more so whenever they faced tighter liquidity constraints, peaking around 5% later in 2008.
This allows them to fare better in terms of sales (having negative sales growth but this loss being
smaller). It also can be seen that they lose more customers, with the quantity per customer seldom
changing. Assuch,wecanconcludethatthesecompaniesadjustbythemasses,usuallybychanging
prices and customers entering (exiting) the customer base of the firm, but always selling around
the same amount per locked customer. For example, people buying soda will always buy the
7
I conduct a robustness exercise using industry medians in Appendix C
14
same amount but will decide whether or not to buy it based on its price per the desired amount,
whether it is 10 cans of 100 ml or 6 cans of the same size. Motivated by these findings, I assess
the relationship between financial constraints and sales, and their components during the Great
Recession by estimating the following panel regression
Y
ift
= (1− crisis
t
)× friction
ft− 4
α 0
+crisis
t
× friction
ft− 4
α 1
+X
ift
δ +η i
+ϵ ift
(5)
for good i within firm f and quarter t. Y
ift
correspond to each of the components of Equation (4),
crisis
t
and friction
ft
are defined as before and η i
and ϵ ift
feature bar-code fixed effects and the
error term, respectively. X
ift
are controls, featuring firm employment yearly growth rates, product
age, and quarter dummies to control for seasonality. I also include the following product quality
index introduced by Jaimovich et al. (2020):
Quality
it
=
p
it
p
k(i)t
, (6)
where p
it
is the price across all product transactions made during the quarter t, and p
k(i)t
is the
sales-weightedaverageacrossallproductswithinmodulek. Highervaluesofthisindeximplyhigher
quality,withtheaverageequaling1. Then,α 0
andα 1
measuretherelationshipbetweenthefinancial
friction and each of the sales components during normal times and recessions, respectively. The
estimates of these coefficients are reported on Table 1.4. Unsurprisingly, higher quality products
sell more by reporting higher prices (columns 4 and 8) and selling fewer quantities (columns 3 and
7). Consistent with Argente et al. (2019), older products sell less, with a year contributing to a
sales decrease of 5%, when considering both specifications (columns 1 and 5).
15
Table 1.4: Decomposition of change in Sales.
Liq
ft− 4
Lev
ft− 4
(1) (2) (3) (4) (5) (6) (7) (8)
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
Friction× Normal 0.027** 0.024** 0.001 0.002 -0.020 -0.017 -0.004 0.000
(0.012) (0.012) (0.001) (0.001) (0.029) (0.027) (0.002) (0.002)
Friction× Crises 0.001 -0.007 -0.002 0.010*** -0.020 -0.029 -0.003 0.012***
(0.015) (0.014) (0.002) (0.003) (0.026) (0.025) (0.002) (0.002)
Quality
ift
0.084*** 0.011* -0.016*** 0.076*** 0.084*** 0.011* -0.016*** 0.076***
(0.030) (0.006) (0.005) (0.027) (0.030) (0.006) (0.005) (0.027)
age
ift
-0.054*** -0.047*** -0.003*** -0.004*** -0.052*** -0.046*** -0.003*** -0.004***
(0.005) (0.005) (0.000) (0.001) (0.006) (0.005) (0.000) (0.000)
∆ log(Emp
ft
) -0.005 0.009 0.001 -0.016* -0.002 0.009 0.001 -0.012
(0.032) (0.025) (0.002) (0.008) (0.040) (0.033) (0.002) (0.008)
Product FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807
R-squared 0.15 0.15 0.07 0.18 0.15 0.15 0.07 0.18
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5% and three
at the 1%. Each column reports the results of a regression for each of the sales components as a function of the covariates.
Financialfrictionvariablesaretransformedtohaveastandarddeviationof1forcomparison. Quarterdummiesareincluded
to control for seasonality.
In terms of financial constraints, both variables tell similar stories: during normal times, the
availabilityofimmediateresourcesdoesnotinfluencethefirmchoicesthatarerealizedintochanges
in sales, with firms experiencing higher liquidity ratios selling 2.7% more as a result of higher
customer presence. However, during the financial crisis, a one standard deviation increase in past
Liquidity does not translate into good sales growth, most of which is driven by a price increase of
roughly1% (columns4and 8). Inprinciple, companies are losing customers during downturns, but
the extra liquidity seems to be used to preserve them.
These findings suggest that companies try to exploit this base while incentivizing it to keep
buying the product at the same time, for the change in quantities per customer remains practically
16
the same (0.2% increase), with their corresponding customer base not changing. This price change
is small, for the average good is sold at a price of $3.7 (Table 1.1) per measure and customer, so
a 1% increase would result in that same good being sold at $3.73, a change that some customers
would not perceive.
However,thissmalladjustmentismultipliedbyamassivenumberofcustomers,whichseemingly
compensates for the quantity and customer losses. This suggests that locked-in customers remain
there, potentially because of the cost of finding an adequate substitute, or because firms selling
these goods may specialize in selling essential items.
Without further inspection, it seemed like these findings are at odds with those in Figure 1.3
and the findings of Gilchrist et al. (2017). The numbers reported on Table 1.4 show the average
relationship, which could be masking heterogeneity in the relationship. I further study this by
plotting the estimates of (5) for each quintile of Liq
ft− 4
8
. The results of this are reported on
Figure 1.3 and show that the relationship holds for all the quantiles, but its mainly driven by the
lower values, namely those under the median value of Liq
ft− 4
. Therefore, we observe that even
though the relationship is positive (or close to zero) across quantiles, especially for the price, this
effect in the most financially constrained firms is the one driving the relationship.
8
I reserve the results for the leverage ratio to Appendix C.
17
Figure 1.4: Relationship between Liq
ft− 4
and Sale components by Liq
ft− 4
quintiles
−0.100
−0.075
−0.050
−0.025
0.000
0.025
0.050
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(a) ∆ Sales
it
−0.20
−0.15
−0.10
−0.05
0.00
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(b) ∆ M
it
0.00
0.01
0.02
0.03
0.04
0.05
0.06
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(c) ∆ q
it
0.00
0.04
0.08
0.12
0.16
0.20
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(d) ∆ p
it
Notes: Each panel plots the coefficient of the relationship estimated in Equation 5, for each component of the
yearly variation of total sales: customer base (Mit), quantities per customer (qit) and price per unit of quantity (pit).
The x-axis shows quintiles of Liq
ft− 4
, whereas the y-axis displays values of each of the sales components.
This relationship seems to be inconsistent with that found by Gilchrist et al. (2017). One
possibilityisthatfirmshaveheterogeneousproduct-specificresponses,someofwhichmaybedriving
the result. I explore this across different values of quality. As such, I re-estimate (5) by removing
the quality control and including an additional interaction to each crisis
t
and friction
ft
pair
u
it
=







Quality
it
>1
Quality
it
≤ 1
,
measuring if a product is of high or low quality. I report the results of this exercise on Table 1.5
and find that the price setting adjustment is heterogeneous: during a downturn, companies facing
18
aonepercentdecreaseintheirliquidityratioswhoselllow-qualityproductsincreasepricesby2.3%
(column 4), a response that seems to hold as well during normal times without it being translated
into meaningful changes on sales. On the other hand, the opposite response is implemented for
high-quality products, where firms increase their prices for higher liquidity ratios.
Table 1.5: Decomposition of change in Sales for quality groups.
Liq
ft− 4
Lev
ft− 4
(1) (2) (3) (4) (5) (6) (7) (8)
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
Friction x Low Quality x Normal -0.005 0.017 0.006*** -0.027*** -0.056** -0.025 0.002 -0.031***
(0.013) (0.012) (0.001) (0.004) (0.027) (0.025) (0.003) (0.003)
Friction x High Quality x Normal 0.092*** 0.038*** -0.009*** 0.058*** 0.048 -0.002 -0.014*** 0.060***
(0.016) (0.014) (0.002) (0.006) (0.032) (0.030) (0.003) (0.004)
Friction x Low Quality x Crisis -0.035** -0.010 0.005** -0.028*** -0.049** -0.032 0.002 -0.018***
(0.017) (0.015) (0.002) (0.004) (0.022) (0.022) (0.002) (0.004)
Friction x High Quality x Crisis 0.066*** -0.002 -0.014*** 0.078*** 0.035 -0.025 -0.012*** 0.070***
(0.021) (0.019) (0.003) (0.007) (0.034) (0.032) (0.003) (0.005)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807
R-squared 0.15 0.15 0.07 0.16 0.15 0.15 0.07 0.19
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%, and three
at the 1%. Each column reports the results of a regression for each of the sales components as a function of the covariates.
Financialfrictionvariablesaretransformedtohaveastandarddeviationof1forcomparison. Quarterdummiesareincluded
to control for seasonality.
Since the pass-through on sales remains for high-quality products regardless of the times, it is
clearthatthecustomerbasedemandinghigh-qualityproductsseemstobesticky,withoutsignificant
tradingdown. Assuch, firmsareabletoexploitthisgroupaccordingly. Furthermore, Iexplorethis
relationship across the quality distribution, plotting the estimated relationship for different quality
quantiles on Figure 1.5
9
. The observed relationships for all of the sales components are monotonic
across the quality space. This allows us to conclude that firms embracing the price adjustment
behavior identified by Gilchrist et al. (2017) are concentrated on low-quality products, with the
9
I reserve the results for the leverage ratio to Appendix C.
19
higher-quality counterparts being large enough to drive the average relationship.
Figure 1.5: Relationship between Liq
ft− 4
and Sale components by quality quintiles.
−0.20
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(a) ∆ Sales
it
−0.08
−0.06
−0.04
−0.02
0.00
0.02
0.04
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(b) ∆ M
it
−0.02
−0.01
0.00
0.01
0.02
0.03
0.04
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(c) ∆ q
it
−0.15
−0.10
−0.05
0.00
0.05
0.10
1 2 3 4 5
Quintile
Coefficient
Normal Times Crisis
(d) ∆ p
it
Notes: Each panel plots the coefficient of the relationship estimated in Equation 5, for each component of the
yearly variation of total sales: customer base (Mit), quantities per customer (qit) and price per unit of quantity (pit).
The x-axis shows quintiles of Liq
ft− 4
, whereas the y-axis displays values of each of the sales components.
20
In summary, there is clear heterogeneity in the adjustment of each sales component. During
the financial crisis, companies with higher levels of past liquidity and leverage ratios exploit their
customer base by doing modest increases in prices, following a scale motive that allows them to
secureadditionalcashflows. Theseresponsesarealsoheterogeneousacrossthequalitydistribution,
withfirmsincreasinglow-qualityproductpricestohedgeshort-termlosses,asdescribedbyGilchrist
et al. (2017). This evidence also illustrates the counter-cyclical markup behavior, a motive of
controversy in the business cycle literature, with different explanations being offered (Hong, 2017a;
Nekarda and Ramey, 2020; Burstein et al., 2020).
1.4 Conclusion.
In this study, I examined the relationship between financial constraints during the financial crisis
for large firms and their response across two main margins: product insertion and destruction, and
sales.
Using a unique dataset of supermarket products combined with that of publicly traded balance
sheet information, I test different measures of financial constraints, finding that the ratio of liquid
assetsistheoneco-movingthemostacrossmarginsduringtheGreatRecession. Usingthismeasure,
Ifindthatonaverage,anincreaseof13percentagepointincreaseinliquidassetsduringtherecession
produced a 1.2% in product entry rates, with a small and imprecisely estimated fall in exit rates
(less than 1%), which implies a net increase of 2% in the product entry rate. This accounts for the
introduction of a single new product for the median firm, which represents an increase in sales of
1.2%, a modest increase but one where companies are shown to adjust accordingly.
Althoughpreviousstudieshavedocumentedtheprice-adjustingbehavioroffirmsfacingfinancial
constraints. I go one step further, and exploit the richness of the dataset to decompose the change
in sales into three main components: customers, quantities per customer, and the price paid per
unit sold. Using this data, I find that the same increase in the liquidity ratio translates to a 1.6%
increase in price to sterilize movements on sales. Even though this increase is considerable, the
change in dollars for the average firm is very small, potentially to customers, most of whom seem
to not perceive it for more liquid firms seem to be better at keeping their customer base.
Thesefindingsshowthatfinanciallyconstrainedfirmsadjustdifferentmarginstostayinbusiness
during an economic downturn. Furthermore, there is heterogeneity in this adjustment behavior,
21
in particular, firms facing higher financial constraints lower the price of low-quality products while
increasing that of their high-quality products. This adjustment translates to a higher increase in
sales for the latter group because customers buying low-quality goods seem more prone to trade
down,whereascustomersshoppingforhigher-qualitygoodsseemtoremainlockedin. Then,further
inquiry into customer behavior during crises could shed some light on the nature of these findings,
in particular, to verify the movement in the customer capital channel.
As such, this set of findings suggests that policymakers interested in providing bailouts to
struggling firms could target their efforts into supporting their efforts to keep customers, but they
are to be wary of the inflation cost that they could be incurred in doing so. Consistent with
the findings of Gilchrist et al. (2017), the Great Recession did not bring a fall in prices, because
the financially constrained firms increased theirs for the low-quality products, thus breaking the
divine coincidence of monetary policy. Therefore, further research into customer capital dynamics
can have a long-lasting impact on different mechanisms used by policymakers to respond during
economic downturns.
22
2 Chapter 2 Redistributive Effects of Oil Shocks for Commodity
Exporting Economies
2.1 Introduction
Oil prices have fluctuated widely since 1970. Emerging markets in Latin America and Asia have a
historicaldependenceontheproductionofthiscommodity,sounderstandingtheeffectsofcommod-
ity price fluctuation is particularly relevant (Bornstein et al., 2017; Fern´ andez et al., 2017, 2018).
This paper explores whether commodity price shocks have redistributive effects on consumption
in commodity-exporting economies. It shows the presence of significant heterogeneous responses
across the household income distribution.
The paper explores three main mechanisms of redistribution: the discretionary income channel
(Edelstein and Kilian, 2009), a commodity-exporting channel, and a fiscal channel. The first
channel suggests that consumers can benefit from lower prices that pass through to lower prices on
gasoline by redirecting their spending on gasoline towards non-energy-related items. The second
channel suggests that commodity exporters can benefit from a boost of overall economic activity
induced by a rise in oil prices and improvement of the terms of trade. The third channel suggests
that consumers can benefit from redistribution induced by either fiscal or monetary policy. In
emerging economies, fiscal policy plays an essential role in hedging the use of commodity royalties
as insurance when there is countercyclical spending (Pieschac´ on, 2012; Ag´ enor, 2016; Martinez,
2019).
To explore these channels, this paper uses Mexico as a case study and collects survey data on
household income and expenditures from Mexico’s National Institute of Statistics and Geography
(Instituto Nacional de Estad´ ıstica, Geograf´ ıa e Inform´ atica – INEGI).
10
The paper also draws on
government finance data and international oil prices from Mexico’s Energy Information System
(Sistema de Informaci´ on Energ´ etica – SIE).
11
The case of Mexico is of particular interest for three main reasons. First, the country bene-
fits from the upswing in commodity prices because it is a commodity exporter, which allows for
exploring whether higher prices induce redistribution via the improvement in overall economic ac-
10
See the INEGI website at https://www.inegi.org.mx/programas/enigh/nc/2018/ (accessed 9 February 2023).
11
See the SIE website at http://sie.energia.gob.mx/ (accessed 9 February 2023)
23
tivity. Second, gasolineanddieselpricesaredirectlylinkedtointernationalprices, whichallowsfor
exploring the discretionary income channel. Finally, Mexico benefits from the transfers of profits
from the state-owned petroleum company (Petr´ oleos Mexicanos - PEMEX) to the government’s
budget, and those funds are then used to fund social expenditures (Unda, 2019).
The analysis conducted for this paper finds a contemporaneous negative effect on consumption
from an increase in oil prices. This effect is exacerbated in households in the bottom decile and
quartile of the wealth distribution of households.
12
The result is consistent with the discretionary
income channel: consumers are more affected by higher oil prices due to a more significant pass-
through via gasoline prices. In addition, increasing oil prices augments expenditures on energy-
related products. However, this change is particularly exacerbated in households in the lowest
decile and quartile of the income distribution.
No evidence was found that an increase in oil prices benefits specific sectors of the economy, or
that households whose members work in the formal sector benefit from an increase in oil prices. In
contrast, households whose members work in the informal sector are more likely to be negatively
affected by the price increase. Higher oil prices do not positively impact the tradable sector, which
might suggest that the benefits of higher revenues in oil production are not immediately translated
into improvements in income or consumption for workers in that sector.
Finally, the analysis finds that one year after the increase in oil prices, there is an increase
in consumption across the entire income distribution. This lagged effect is consistent with a fis-
cal policy channel response from an oil windfall. For example, in Mexico, oil windfalls occur
when PEMEX rebates a share of its revenues to the federal government in December for the next
year’s budget. Then, the federal government transfers these royalties to each state through the
Federal Fiscal Coordination Law (Ley de Coordinaci´ on Fiscal Federal) for both discretional and
non-discretional purposes. According to the findings, the rebound in consumption is consistent
with larger government expenditures transmitted through social programs, more investment, and
more hiring.
Theredistributioninducedbyfiscalandmonetarypolicyhasbecomeakeytopicofdiscussionin
12
A household is classified according to the wealth distribution in the year 2000. For example, a household lies
within the second decile of the distribution if it is at least richer than the poorest 10% but it is poorer than 70% of
the remainder of homes within a year. Similarly, a household lies in the fourth quartile if it is richer than 75% of the
other households within a given year.
24
recentliterature. SincetheseminalworkofKaplanetal.(2014,2018), agrowingbodyofliterature
has tackled the mechanisms that produce such effects for advanced economies such as the United
States, Denmark, and Norway. However, analysis for emerging markets is different because most
of these economies are exposed to different global shocks. Auclert et al. (2021); Motyovszki (2020);
Guo et al. (2020); Cugat (2019); Hong (2020); Messina and Silva (2019) discuss the implications of
such policies in emerging markets under the small open economy framework.
This paper contributes to the literature by providing empirical reduced-form evidence of these
channelsandbyexploringtheroleofinsurancethroughoptimalfiscalmanagementwhencommodity-
exporting economies face redistributive effects on consumption after commodity price shocks. Sec-
tion 2.2 describes the data used in the analysis, Section 2.3 presents the empirical evidence on the
distributional impact of oil prices on consumption and disposable income, and Section 2.4 presents
some final remarks.
2.2 Data
OilpriceandoutputdatawerecollectedfromSIE,whichorganizestheyearlyreportsfromPEMEX.
PEMEX’s oil output is available at the state level, whereas the price is competitively set (in U.S.
dollars) in the world oil futures market as a weighted average of two U.S. crude oils, West Texas
Sour (WTS) and Light Louisiana Sweet (LLS), and the British Brent (Figure 2.6). As documented
by Pieschac´ on (2012), the price series are obtained for mixes of crude oils of different densities
sold by PEMEX: Maya, Itsmo, and Olmeca. The oil price measure is then deflated using the U.S.
consumer price index.
In addition, state-level public finance data (deflated using the GDP deflator) were collected, as
werehousehold-leveldatafromMexico’sHouseholdIncomeandExpenditureSurvey(Encuesta Na-
cional de Ingresos y Gastos de los Hogares -ENIGH),whichisconductedbyINEGI
13
.TheENIGH
is a biannual repeated cross-sectional survey that has been featured in other studies (Cravino and
Levchenko, 2017; Cugat, 2019). It is representative both at the aggregate and state levels, fea-
turing rich information on households’ characteristics, such as the number of people living in the
household, municipality of residence, occupation, informality status (measured as whether or not
householdmemberscontributetosocialsecurity), age, gender, education, laborincomeandwealth,
13
Seethe“Topics”sectionoftheINEGIwebsiteathttp://en.www.inegi.org.mx/datos/(accessed9February2023).
25
Figure 2.6: Mexican Oil prices.
20 40 60 80 100
2013 USD/Barrel
2000 2005 2010 2015 2020
Source: Sistema de Informaci´ on Energ´ etica.
Note: This price series result from the averaging the West Texas Sour (WTS) and Light Louisiana Sweet (LLS),
and the British Brent oil prices, these series are obtained directly from the SIE website.
and consumption expenditure on goods defined for multiple categories such as food, energy, and
rent, among others. The period of analysis covers the years 2000 through 2018, which allows for
comparing periods of both commodity booms and busts, as can be observed in Figure 2.6. The
period of analysis yields between 4,000 and 45,000 households per year and all the variables are
reported on a quarterly basis.
The analysis follows Cugat (2019) in the data-cleaning procedure in the following ways: (i) It
removesobservationswithnegativevaluesforconsumptionexpenditureandcashonhand(measured
as disposable income plus assets) net of taxes and rent and disposable income; (ii) It removes the
top and bottom 1 percent earners of gross income per year, as well as corresponding municipalities;
(iii) It removes observations for the top and bottom 1 percent consumption expenditure, labor
income, and cash-on-hand levels; and (iv) It covers households whose head is within the 25-64
year-old range. Consumption (C) (excluding taxes and rent) is computed as the total expenditure
divided by household members in the following categories: food, apparel, energy, cleaning supplies,
healthcare, transport, education, and leisure. Finally, all of the household variables using the
Mexican consumer price index are deflated. Table 2.6 displays these variables.
26
Table 2.6: Descriptive Statistics.
Variable Mean Std. Dev. 10th Perc. 25th Perc. 50th Perc. 75th Perc. 90th Perc.
Disposable income Y − T (log) 2.04 0.76 1.078 1.527 2.022 2.544 3.043
Cons. exp. C (log) 1.67 0.75 0.745 1.190 1.653 2.149 2.644
Cash in hand X (log) 2.10 0.77 1.147 1.592 2.080 2.606 3.110
C/(Y − T) 0.78 0.77 0.389 0.521 0.696 0.921 1.214
E/C 0.06 0.05 0.015 0.029 0.051 0.083 0.125
Age of head of household 43.45 10.21 30 35 43 51 58
Weekly worked hours 45.22 25.46 0 32 48 60 74
∆ log(P
o
t
) 0.13 0.19 -0.203 -0.012 0.199 0.248 0.286
(%) workers in tradable industries 0.43
(%) formal sector workers 0.38
(%) female heads of household 0.19
(%) people living in urban areas 0.49
N 193,676
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: nominal variables are deflated using CPI index and use 2013 as the base year. All income and consumption
variables are divided by the household size and keep the information of the head of the household for the compu-
tations. Consumption expenditure excludes rent and taxes. We define tradable industries as agriculture, mining,
or manufacturing and set the rest as non-tradable. A worker is defined as formal if they make contributions to
social security and/or retirement.
A similar dispersion for income, consumption, and cash on hand is observed: a roughly 0.75
percentdispersionfromthemeanacrosshouseholds. Ontheotherhand,energyconsumption,which
includes electricity, oil, and gas, accounts for roughly 6 percent of total consumption expenditure,
with considerable variation, reaching levels of around 12 percent in some cases. This provides some
stylized evidence of the effect of commodity price shocks, which will be explored in Section 2.3:
that is, higher levels of oil prices lead to higher costs of energy for households, which would suggest
an increase in the share of this component of total expenditure. Furthermore, this suggests that
certain households will be more exposed to the direct effect of the shock.
2.3 Empirical Evidence
The exogeneity of oil prices is exploited to assess its distributional effects on consumption expendi-
ture(see,AppendixD).Inparticular,atestisconductedtodeterminewhetheroilpricesdirectlyor
indirectly affect households across several dimensions. The first dimension considered is consump-
tion, a measure that reflects monetary expenditure on food, clothing, footwear, housing, cleaning,
health, transportation, education, and transfers divided by the number of household members. For
this purpose, the following model is estimated:
27
Z
i
=δ j(i)
+δ k(i)
+
4
X
s=0
β s
∆ log(P
o
t(i)− s
)+ζ X
it(i)
+ϵ it
(7)
whereZ
i
istheoutcomeofinterestforhouseholdi,inmunicipalityj(i),statek(i)andyeart(i);δ j(i)
and δ k(i)
correspond to fixed effects at the state and municipality level, respectively; P
o
t
is oil price
measure, andX
it(i)
is a vector of controls at the state and household level. This vector includes oil
outputand its lagsfor state k, statepopulation, gross value added growth rates, real exchange rate
growth rates in year t, the United States Chicago Fed National Activity Index (CFNAI) developed
by Stock and Watson (1999) which measures the US economic activity, the global asset price
factor calculated by Miranda-Agrippino and Rey (2020), which proxies movements of the global
financial cycle, and household characteristics such as whether the head of household is an informal
worker, gender, education level, industry of occupation, three dummies grouping their age (under
35, between 35 and 50 and older than 51), type of household, and the log of energy consumption.
Figure 2.7 depicts the results for β s
in the full sample. The results show a negative average
contemporaneous effect of 0.18 percent for a 1 percent increase in oil prices, followed by a strong
subsequent recovery. The year after the shock, an increase of 1 percent in oil prices leads to
a consumption increase of roughly 0.3 percent. The shock then fades and average consumption
seemingly returns to its steady state. These results mask large heterogeneity in the responses at
different levels, which are explored below.
First, using the information available in 2000, we separate the results quartiles of the state-
specific cash-in-hand distribution. This measures the household’s wealth distribution and accounts
for potential differences in the cost of living across states.
14
Then, we estimate Equation 7 by
allowing different slopes, and report the results on Figure 2.8.
14
We use the initial sample year to avoid incurring endogeneity coming from shifts in the distribution that were
specific to other years.
28
Figure 2.7: Effects of oil price shocks to consumption expenditure.
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: This figure shows the effect of an average 1% increase in the oil price for the households’ consumption.
Shadedareascorrespondtothe95%, 90%, and80%confidenceintervalsusingthestandarderrorsfromtheestimated
Equation (7)
29
Figure 2.8: Effects of oil price shocks to consumption expenditure by Quartiles
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 2
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Quartile 3
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(d) Quartile 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ consumption for each
quartile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using
the standard errors from the estimated Equation (7).
These results show that the immediate impact of an oil price shock is non-statistically different
from zero for households in the second and fourth quartiles of the wealth distribution. At the same
30
time, it is negative for all households but stronger for the households within the first quartile, but
not monotonic regardless. In this case, an increase of 1% in oil prices represents a reduction in
consumption of around 0.3%, while it is of roughly 0.2% for households within the third quartile
and roughly 0.1% for quartiles 2 and 4.
In contrast, the effect of the shock one year later is positive at all levels of disaggregation.
Appendix E examines three groups: the top and bottom quartiles and deciles, and the remainder
ofthesample. Theexercisefindsthatsubtractingmorehouseholdsfromthetopofthedistribution
impacts the responses: there is a positive contemporaneous impact of the shock on consumption
for the top decile, as well as a stronger impact on the bottom decile. This suggests redistributional
effectsonconsumptiondrivenbychangesinoilprices. Thoseadverseeffectsaremainlyconcentrated
around households below the 25th percentile of the cash-on-hand distribution.
The differences in the response are then assessed based on the household’s economic activity.
Figure 2.9 shows that working in the formal or informal sectors shows very similar negative reac-
tions, at least contemporaneously (0.2 percent for informal workers and roughly 0.15 percent for
formal workers), to the shock. However, there is a strong increase in consumption for both sector
workersoneyearafterapricevariation: anincreasein1percentofoilpricesincreasesconsumption
forbothformalandinformalsectorworkersby0.3percent. Giventhewayinformalityisdefined, it
is possible that workers in both the first and third quartiles of the wealth distribution participate
in both sectors, mainly because both low- and middle-income business owners and self-employed
people would plausibly fall within this category.
31
Figure 2.9: Effects of oil price shocks to consumption expenditure by sector
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Informal workers
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Formal workers
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ consumption for
formal and informal workers. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using the
standard errors from the estimated Equation (7). A worker is defined as formal if they make contributions to social
security and/or retirement.
Similarly, Figure 2.10 highlights that those working in the tradable sector face higher risk, as
their contemporaneous consumption response to the shock is approximately 0.2 percent, which is
twiceasmuchasthatofworkersinthenontradablesectorwheneverthereisa1percentincreasein
prices. Thehomogeneousrecoveryayearlaterof0.3percentsuggeststhattheimmediateresponses
are potentially due to changes in the price of goods and services and income variations.
32
Figure 2.10: Effects of oil price shocks to consumption expenditure by occupation
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Non-tradable sector workers
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Tradable sector workers
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ consumption for each
occupation group. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using the standard errors
from the estimated Equation (7). We define tradable industries as agriculture, mining, or manufacturing and set the
rest as non-tradable.
What explains these heterogeneous effects on consumption? Why is there a contemporaneous
and lagged impact of oil prices? While there might be multiple potential explanations for these
results, the next section explores the relevance of three main mechanisms that could impact con-
sumption: changes in the disposable income of households, changes in the share of expenditure
devoted to energy-related products, and fiscal transmissions.
2.3.1 Changes in Disposable Income
This section evaluates how disposable income dynamics (total income minus housing rents) evolve
after an oil price increase for the groupings presented in the previous section. The idea is that
a positive change in income due to variations in oil prices can have different impacts on con-
sumption because of either differences in the marginal propensities to consume or the presence of
non-homothetic preferences. Figure 2.11 shows that considering the entire sample, an increase in
oil prices does not have a significant average effect on disposable income. Nevertheless, consump-
33
tion exhibits a significant increase one year later (as shown in Figure 2.7), which is not necessarily
matched by a significant change in disposable income. This suggests that alternative drivers may
play a role in the relationship between oil prices and consumption.
Figure 2.11: Effects of oil price shocks to disposable income.
−0.125
−0.100
−0.075
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: This figure shows the effect of an average 1% increase in the oil price for the households’ disposable income.
Shadedareascorrespondtothe95%, 90%, and80%confidenceintervalsusingthestandarderrorsfromtheestimated
Equation (7)
The analysis further explores whether there is heterogeneity in the income responses across the
wealth distribution, in the same fashion as before. Figure 2.12 presents the result for the cash-
on-hand quartiles. Interestingly, the response for disposable income is highly heterogeneous: the
poorest households see an approximate decrease of about 0.08 percent in their disposable income
followinga1percentincreaseinoilprices. Thisseemstobeconsistentwiththedropinconsumption
observed in panel (a) in Figure 2.8. On the other hand, the non-significant changes in consumption
in panels (b) and (d) for the lower middle quartile and the richest household (fourth quartile) of
Figure 3 seem to be consistent here as well: their disposable income increases by roughly 0.08 per-
cent and 0.14 percent, respectively, but their consumption does not change. This suggests a degree
34
of non-homothetic preferences for these groups. Finally, it is worth mentioning that households in
the third quartile do not adjust their income, but they do decrease their consumption. This may
be related to a different channel or a mere substitution pattern between energy expenditure and
regular consumption.
35
Figure 2.12: Effects of oil price shocks to disposable income by Quartiles
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 2
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Quartile 3
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(d) Quartile 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ disposable income
for each quartile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals
using the standard errors from the estimated Equation (7).
However, one year after the shock, the dynamics between disposable income and consumption
arenolongeraligned. Intermsoflaggedeffects,incomedoesnotdecreaseatallforanyquartile. At
36
thesametime,anincreaseinconsumptionisobservedforallgroups,whichsuggeststhatalternative
mechanisms drive the consumption response for this income quartile. To shed more light on the
nature of these responses, this exercise is then carried out as before for sector and occupation.
Figure 2.13: Effects of oil price shocks to disposable income by sector
−0.125
−0.100
−0.075
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Informal workers
−0.125
−0.100
−0.075
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Formal workers
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ disposable income
for formal and informal workers. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using the
standard errors from the estimated Equation (7). A worker is defined as formal if they make contributions to social
security and/or retirement.
Figure 2.13 shows the estimation results across the formal and informal sectors. Interestingly,
neither of the sectors exhibits a significant response to the shock. This suggests that there may
be workers employed in both the formal and informal sectors coming from all the wealth quartiles.
In addition, the response is roughly 0.025 percent for both groups contemporaneously, which could
mean that certain occupations may benefit, with a rather low pass-through to workers, from the
effects of the shocks. This is analyzed further by computing the responses across occupational
sector, with the results shown in Figure 2.14.
37
Figure 2.14: Effects of oil price shocks to disposable income by occupation
−0.12
−0.08
−0.04
0.00
0.04
0.08
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Non-tradable sector workers
−0.12
−0.08
−0.04
0.00
0.04
0.08
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Tradable sector workers
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ disposable income for
each occupation group. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using the standard
errors from the estimated Equation (7). We define tradable industries as agriculture, mining, or manufacturing and
set the rest as non-tradable.
Disposable income for workers in the nontradable sector appears to be more affected by oil
prices and is statistically significant at the 10 percent level. In response to a change of 1 percent
in prices, disposable income increased by 0.05 percent during the year of the shock. This pattern
is consistent with the Balassa-Samuelson effect, which states that the relative price of nontradable
goods increases as a country gets wealthier. Then, the results indicate some evidence of the pass-
throughofthiseffectonhouseholds. Interestingly,thisisnotconsistentwiththehouseholdresponse
to consumption, which decreases. This suggests that other forces may be moving that eliminate
these small gains in income for households whose members work in nontradable industries. On
the other hand, income does not increase for tradable sector workers, and their consumption falls.
Taking both together, it seems that consumption may be reacting to a different source.
38
2.3.2 Changes in Energy Expenditure
Oil prices are associated with increases in distribution costs because of their effect on gasoline and
diesel prices. While it is difficult to evaluate how the price level of goods purchased by different
householdsevolvesduringanoilpriceboom(forexample),anassessmentcanbemadeofchangesin
expendituresonenergyduetoaboom. Theanalysisfindsthathouseholdsdevotemoreresourcesto
energyduringapriceboom,particularlyhouseholdsatthebottomofthecash-on-handdistribution.
Thus,theredistributionaleffectscanbesizable,aspoorerhouseholdsseetheirconsumptionofother
types of goods crowded out.
Figure 2.15: Effects of oil price shocks to energy expenditure
E
Y− T
.
−0.03
−0.02
−0.01
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: This figure shows the effect of an average 1% increase in the oil price for the households’ energy expenditure,
as a share of their disposable income. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using
the standard errors from the estimated Equation (7)
Figure2.15showstheaverageresponseofconsumptionenergyexpenditureasashareofdispos-
able income. Consider that disposable income does not change and consumption decreases. Then,
a 0.05 percent increase in the energy expenditure share in response to a 1 percent contempora-
neous increase in oil prices suggests that the oil shock is crowding out consumption. However, a
39
year after the shock, there is a 1 percent decrease in energy expenditure that is not statistically
significant. Thus, it can be concluded that the recovery in consumption could partially be driven
by a reallocation of energy expenditures to consumption.
Similarly, Figure2.16examinestheeffectsoftheoilpriceshocktoenergyexpendituresbycash-
on-hand distribution by quartiles. Oil price increases are associated on impact with an increase in
energy consumption expenditure across all quartiles. The dynamics are similar to the average and
showanapproximate0.05percentincreaseforallquartiles,withanon-significantdecreasetheyear
after the shock. This suggests that the energy channel is playing a role here: an oil price increase
brings about a rise in intermediate input prices due to an increase in gas prices, which impacts all
quartiles in a similar way.
40
Figure 2.16: Effects of oil price shocks to
E
Y− T
by Quartiles
−0.02
0.00
0.02
0.04
0.06
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.02
0.00
0.02
0.04
0.06
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 2
−0.02
0.00
0.02
0.04
0.06
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Quartile 3
−0.02
0.00
0.02
0.04
0.06
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(d) Quartile 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ energy consumption
as a share of disposable income for each quartile of the wealth distribution. Shaded areas correspond to the 95%,
90%, and 80% confidence intervals using the standard errors from the estimated Equation (7).
41
However, energy expenditure is comprised of multiple items beyond gasoline consumption. Al-
though it is not possible to verify how each component moves individually, it is possible to test the
aggregate gasoline prices in Mexico for the sample period. To achieve this, the following structural
Vector Autoregressive (VAR) model is estimated:



∆ log(P
o
t+1
)
∆ log(P
G
t+1
)


 = F
1



∆ log(P
o
t
)
∆ log(P
G
t
)


+



1 0
b
21
1






ϵ o
t+1
ϵ G
t+1


 (8)



ϵ o
t+1
ϵ G
t+1


 ∼ iid
N(0,Σ)
Σ=



σ 2
o
0
0 σ 2
G


 (9)
WhereP
o
t
andP
G
t
arethecrude-oilandgasolineretailpricesandF
1
isa2× 2coefficientmatrix.
The gasoline price is deflated using the Mexican consumer price index and then converted to U.S.
dollars using the purchasing power parity exchange rates to rule out other nominal and currency
market volatility movements. Then, the impulse response functions are computed to calculate the
pass-through of crude oil prices to gasoline retail prices. The results are shown in Figure 2.17.
The results are consistent with the energy expenditure dynamics: a 1 percent increase in crude
oil prices translates contemporaneously into a 2 percent increase in gasoline prices. One year after
the shock, this increase falls to 0.4 percent but it is not statistically significant, with the dynamic
suggesting that the system goes back to its original steady state.
42
Figure 2.17: Pass-through of crude-oil prices to retail gasoline prices.
−0.01
0.00
0.01
0.02
0.03
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
Source: Author’s calculations based on retail gasoline price data from SIE.
Note: This figure shows the effect of an average 1% increase in the oil price for the retail gasoline price, Shaded
areas correspond to the 95%, 90%, and 80% confidence intervals using bootstrapped standard errors, using 5000
replications.
Taking stock of these results together, it can be concluded that changes in energy expenditure
seem to drive the contemporaneous consumption responses. However, this channel seems to fall
short of explaining the puzzling one-year recovery in consumption dynamics. This leads us to
explore the last potential channel in the next section.
2.3.3 The Fiscal Channel: government expenditure
The dynamic response in consumption suggests that oil windfalls take time to reach the economy.
Given that this transmission mechanism is rooted in a fiscal response, this section examines the
effects of oil price fluctuations on fiscal transfers and expenditures. In Mexico, the mechanism is as
follows: each December, PEMEX rebates a share of its revenues to the federal government for next
year’sbudget. Thefederalgovernmenttransferstheseroyaltiestoeachstatefollowingadistribution
43
rule detailed within the Federal Coordination Fiscal Law based on three factors: how much oil was
producedineachstate,eachstate’sgrossvalueadded,andeachstate’spopulation. Theexpenditure
destination for these royalties is coordinated between federal and state governments for healthcare,
public goods, and education, with some of these resources designated as discretionary spending
(Unda, 2019). Given the data available, this channel can be directly tested by assessing the effects
of oil price shocks on these federal transfers. Using the state-level government’s finances, the
following specification is run:
∆ log(TR
kt
)= δ k
+β TR
∆ log(P
o
t− 1
)+ζX
kt− 1
+ϵ kt
(10)
where TR
kt
are defined as the Federal Government’s transfers to the states, X
kt
are control
variables: total oil production by PEMEX on state k, state’s gross value added, state’s population,
total Mexican GDP growth, the measure of the global financial cycle from Miranda-Agrippino and
Rey(2020),thecyclicalcomponentoftheUSGDP,andδ k
,whichcorrespondstostatefixedeffects.
In this case, β TR
measures the average elasticity of the transfers to the oil shock. (10) is
estimated using the state public finance data, and the findings are reported in Table 2.7.
Table 2.7: Effect of Oil Price on Government Transfers
Simple Controls Share of income Targeted Discretionary
(1) (2) (3) (4) (5)
∆ log(P
o
t− 1
) 0.073
∗∗∗ 0.062
∗∗∗ 0.021
∗ 0.114
∗∗∗ − 0.011
(0.006) (0.005) (0.009) (0.012) (0.009)
Global financial factor 0 .009
∗∗∗ 0.001 0.011
∗∗ 0.007
∗∗ (0.002) (0.003) (0.003) (0.002)
USA business cycle 0.002 0.508
∗ − 0.835
∗∗∗ 1.329
∗∗∗ (0.096) (0.215) (0.165) (0.077)
Observations 637 637 637 637 637
Source: Author’s calculations based on Mexican state public finance data.
Note: *** (**) [*] denotes significance at the 1 (5) [10] percent level. Clustered standard errors in parenthesis. All
specifications control for oil output, gross value added and population per state.
It can be seen that, on average, state governments get a 6.2 percent increase in the amount of
44
the transfers they receive, which amounts to an increase of approximately 2 percent of their total
state income, as reported in columns 2 and 3 of Table 2.7. The government transfer account is
composed of two items: targeted (aportaciones) and discretionary (participaciones) transfers. The
resultsincolumns4and5showthatthetargeteditemmainlydrivesthisincrease. Asexplainedby
Unda(2019),thesetargetedtransfersmustbedestinedforhealthcare,publicgoods,andeducation.
Therefore, heavily cyclical government expenditure should fall short of delivering these transfers
when the oil price significantly declines. This can also be tested, since β TR
is a good predictor for
∆ log(TR
kt
), which allows for using the design fromCaselli and Michaels (2013) and running the
following second stage of an instrumental variable (IV) design where the oil price instruments the
transfers. The following regression is run:
∆ log(G
kt
)= δ k
+β EX
\
∆ log(TR
kt
)+ζX
kt− 1
+u
kt
(11)
whereG
kt
istotalgovernmentexpendituresexcludingtransferstomunicipalities,debtpayment,
subsidies, and others. and
\
∆ log(TR
kt
) are the predicted values from estimating (10), which are
reported in column (2) of Table 2.7. The coefficient β EX
measures the percentage increase of
government expenditure via transfers exclusively because of the shock. The results are shown in
Table 2.8.
Column(2)showsafalsificationtestinthespiritofDippeletal.(2019)inwhich, oncetransfers
are accounted for, the effect of the oil price disappears. This means that this shock will only affect
public expenditure through the transfer channel. The IV result in column (3) shows a strong and
significant effect of the shock. This confirms the presence of Dutch disease within the Mexican
government, which is known to have procyclical expenditures: a 1 percent increase in transfers
due to an increase in the oil price (which does not necessarily mean that PEMEX will do better)
incentivizes government expenditure by a considerable margin of 1.17 percent. This procyclical
response also highlights that the government is overindulgent in the use of its windfall.
Furthermore, we inquire into the driver of this expenditure by disaggregating it into two main
components: public capital investment and current expenditure. The former includes the acquisi-
tion of intangibles, investment in the provision of public goods, and investment in financial assets,
45
Table 2.8: Effect of oil price shocks on Government Expenditure.
OLS IV
Baseline Transfers Total Cap. Exp. Curr. Exp.
(1) (2) (3) (4) (5)
∆ log(P
o
t− 1
) 0.073
∗ 0.023
(0.028) (0.030)
∆ log(TR
kt
) 0.813
∗∗∗ 1.177
∗∗ 2.448
∗ 0.780
∗∗ (0.191) (0.394) (1.162) (0.266)
Observations 637 637 637 637 637
Source: Author’s calculations based on Mexican state public finance data.
Note: *** (**) [*] denotes significance at the 1 (5) [10] percent level. Clustered standard errors
in parenthesis. All specifications control for oil output, gross value added, population per state, the
measure of the global financial cycle, and the cyclical component of the US GDP growth.
with the second being the main account (most states do not invest in the financial markets or ac-
quire intangibles for multiple years). For its part, current expenditure is composed of the following
items: payroll and inputs such as tablets, computers, and other common-use goods.
As columns (4) and (5) confirm, both accounts increase, with a stronger capital expenditure
componentgrowth(2.4%). Inparticular,publicgoodsareincludedwithinthisaccount,andbecause
of the timing of the windfall (arriving at the end of the calendar year to be allocated within
next year’s budget), the expenditure effect is consistent within the household one year after the
shock responses shown in Figure 2.8, and inFigure 2.16, where households benefit from a public
expenditure externality and allocate their consumption and energy expenditures accordingly.
Therefore, a decline in the oil price would limit the government’s capacity to increase its public
investment during bad times. These findings may be driven by two reasons. First, there may
be evidence of Dutch disease (the government feels richer than it really is). Second, there may
be some reallocation between social and current expenditure, that is, social expenditure increases
because an increase of the targeted oil royalties allows for some of the discretionary resources to
be redirected towards current expenditure by hiring more government workers or acquiring office
equipment.
46
2.4 Final Remarks
This paper has documented empirical evidence of the redistributive effects of an oil price shock
in Mexico. The evidence seems consistent with that induced by discretionary income and fiscal
channels. After an oil price increase, households face higher costs for energy goods, which crowds
outspendingonothergoods. Thiseffectismorepronouncedforpoorerhouseholds. Thesecostsare
partially recovered a year later, which is seen in an increase in consumption that is not necessarily
matchedbyanincreaseindisposableincome. Thissuggeststhataredistributiveoilwindfallcoming
from government expenditure aids in this recovery process.
Thegrowingliteratureonthistopicshowsthatthereispotentialforfurtherresearch,sincethere
is still a great deal of research to be done to refine our understanding of the role of commodity
pricesandtheirinteractionswiththeeconomy. Traditionally,commoditypriceboomshaveinduced
temporary gains in wealth for households, with long-term losses driven by the procyclicality of
the fiscal policy that characterizes these economies. This paper contributes to the literature by
being the first to assess the heterogeneity of these gains and losses, which may be leading to
greater wealth inequality and worse social outcomes. As these effects remain to be studied from a
macroeconomic perspective, these new findings could inspire fiscal policies that may smoothen the
welfare distortions created by these shocks.
47
3 Chapter 3 The Effects of the Pandemic on Market Power and
Profitability
3.1 Introduction
The COVID-19 pandemic deeply disrupted both supply and demand in a myriad of markets. In
order to understand these disruptions, we analyze markup and profit rates from 2020 to 2021.
This analysis is relevant for two reasons: First, markup rates summarize the interactions between
supply and demand in a market and can therefore provide insight into the effects of the pandemic
on markets. Second, it is uncertain whether the pandemic exacerbated previous trends in markup
orprofitrates,orwhetheritcreatedtheconditionsforanincreaseinmarketpower. Thegoalofthis
paper is not only to document recent movements in markups and profitability during the pandemic
but also to determine whether pre-pandemic trends can predict these movements or whether the
pandemic itself triggered the conditions for an increase in market power.
This paper makes three contributions to the recent literature on the market power and prof-
itability of publicly-traded firms in the US during the COVID-19 pandemic. First, we document
both the increasing markup and profit rates, and more volatile profit rates. Specifically, after av-
erage markup rates decreased from 1.61 to 1.49 between 2016 and 2019, they rebounded to 1.54 in
2021. Additionally, the average profit rate reached its all-time high at 21.6% in 2021, at least since
the start of historical records in 1955. We present our findings using both yearly and quarterly
firm-level data.
Second, to further understand the rising trends in market power and profitability during the
pandemic, we evaluate whether previous trends in firm-level markup and profit rates predict these
increases. To do so, we use separate firm-level Bayesian structural time series models to compute
firm-level forecasts of markups and profit rates, using information up to 2019. We then calculate
the difference between the observed value and its forecast. Our results indicate that, for the
average firm, the observed markup rate was 4.3% and 6.6% lower than its forecast in 2020 and
2021, respectively. In contrast, the average firm had a profit rate of 21.6% in 2021, which is 6.4
percentage points higher than its forecast value. These findings suggest that while markup rates
were lower than expected, profit rates were higher than expected.
48
Third, we analyze the heterogeneous effects of rising market power and profitability during the
pandemic. Firms that performed poorly in terms of markups during the pandemic tend to have
become recently publicly traded and have fewer employees. As for profitability, there seems to
be low effect heterogeneity across firms with different levels of employment, tenure in the stock
exchange, or market share.
Alternatively, when analyzing effect heterogeneity in markups by industry, information, real
estate,andchemicalmanufacturingexperiencedlowermarkupsthanexpected,whereaswarehousing
and entertainment had higher markups than expected. When analyzing heterogeneity in profit
ratesbyindustry,transportation,entertainment,andhospitalitytendedtoperformpoorly,whereas
warehousing and real estate outperform. These findings suggest that the impact of the pandemic
on market power and profitability was uneven across firms.
To support these contributions, we perform a battery of statistical tests to assess the efficacy
of the Bayesian structural time series model in producing forecasts of firm-level markup and profit
rates. Specifically, we perform three exercises to evaluate the quality of our models. First, we
statisticallytestwhetheracriticalassumptionofourBayesianstructuraltimeseries(BSTS)models
holds for most firms. Second, we evaluate whether these firm-specific models accurately reflect the
observed firm’s markup and profit rates before the pandemic. Third, we confirm that our forecasts
have adequate forecast quality for a sample before the pandemic.
In addition to evaluating the quality of the BSTS models, we offer robustness checks for our
resultsbyconsideringtwoalternativemodelingchoicesandframeworks. First,wevarythemodeling
choices of the BSTS model by choosing alternative hyperparameters of the priors in the Bayesian
model. These hyperparameters attribute either more or less variation to the latent components of
the structural model. Second, we alternatively implement forecasts using the flexible approach of
local linear projections (Jord` a, 2005). Particularly, when comparing the forecast quality of BSTS
and local linear projections models, the later tend to have wide confidence intervals which suggests
that local linear projections do not have as much power for this data. Despite these changes, our
main results remain unchanged. These robustness checks help to strengthen the reliability and
robustness of our findings.
Our results inform the recent literature on the rising markups of publicly-traded firms in the
US (De Loecker and Warzynski, 2012; Hall, 2018; De Loecker et al., 2020; Autor et al., 2020; Berry
49
et al., 2019). In particular, De Loecker et al. (2020) documents the rising nature of markups from
1955 to 2016. We extend their analysis by providing evidence suggesting that previous trends in
markup and profit rates do not predict average market power behavior during the pandemic.
Additionally, we uncover the average behavior in markups and profitability and analyze its het-
erogeneity by key firm characteristics such as size, market share, and tenure on the stock exchange.
We further argue that our methods may provide a causal accounting of the effects of the pandemic
on markups and profit rates. We support our identification strategy on a similar fact noted in
Jord` aetal.(2022): pandemicsareexogenousandunpredictabletotheeconomyand, evenmoreso,
to granular firm dynamics. A similar approach in the context of country-level impact of the pan-
demic uses shocks to economic growth expectations as identifying variation (Chudik et al., 2021).
Our methodology is related to the event study method commonly used in finance and economics
(MacKinlay, 1997). However, we do not limit our analysis to estimating the impact on the stock
price of a given sample of companies.
Furthermore,agrowingliteraturedocumentstheeffectsoftheCOVID-19pandemiconmacroe-
conomic aggregates, such as inflation (Cavallo, 2020; Ball et al., 2022; Binder and Kamdar, 2022;
Dietrich et al., 2022) and economic output (Ludvigson et al., 2020; McKibbin and Fernando, 2021;
Baqaee and Farhi, 2020), and the consequential public policy response (Guerrieri et al., 2022;
Chudik et al., 2021; Woodford, 2022; Bigio et al., 2020; Auerbach et al., 2021). Brodeur et al.
(2021) provides a comprehensive survey of the rapidly growing body of literature on the economics
ofCOVID-19. Noneofthesepapersoffersevidenceofthedisruptionofthepandemiconcompetition
dynamics.
This paper has six sections including this introduction. The second section describes the data
and outcomes of interest. The third section describes our methodology including the Bayesian
structural time series model. The fourth section presents our main results. The fifth section
introduces our robustness checks. The sixth section concludes.
3.2 Data
We use yearly (and quarterly) information for 3,611 (3,192) publicly-traded companies from Com-
pustat, which provides a panel of financial statements since 1955 (1980) representing roughly 29%
of US jobs (Davis et al., 2006). To make our sample comparable to that in previous literature,
50
we follow the same sample selection steps as in De Loecker et al. (2020).
15
Furthermore, to evalu-
ate whether our sample is comparable to the sample in previous literature, we compute summary
statistics for sales, COGS, capital stock, SG&A, wage bill, and employment in both 1955–2021 and
1955–2016. Our numbers are indeed quite similar to those in De Loecker et al. (2020). Differences
arise as Compustat updates and companies revise the corresponding financial statements.
16
3.2.1 Outcomes of interest
Markup rates
We compute markup rates using the production function approach (Hall, 1988; De Loecker and
Warzynski, 2012; De Loecker et al., 2020).
17
This approach identifies the markup rate by solving
the firm’s cost minimization problem and two main assumptions: first, the firm is a price taker
in the input market. Second, there are two types of factors, a variable factor, in which firms can
flexibly decide how much to use within a given period (typically a year), and a fixed factor, such as
capital, which remains constant within the same period. We compute markup rates, µ ijt
, at firm
i, industry j, and period t (either quarter or year) as
18
µ ijt
=θ jt
Sales
ijt
COGS
ijt
,
where θ jt
is the industry-level output elasticity of variable inputs, and COGS
ijt
are costs of goods
sold.
To estimate the industry-level output elasticities, we follow the common practice of using a
Cobb-Douglas production function at the industry level with year-varying input elasticities. One
challenge of this estimation is accounting for productivity shocks, while another is converting rev-
15
For a more detailed description of the sample, please refer to Appendix F.
16
Table F.32 in the Appendix reports summary statistics for these variables for three samples: 1955–2016, 1955–
2021, and the sample in De Loecker et al. (2020). We use the same coding routine in Stata as those authors, which is
available in their published supplemental material. Although the mean and median of the cost of goods sold differs
between our sample and the sample in De Loecker et al. (2020), we find a similar series of average markup rates as
those authors, and both series have a high positive correlation (0.92).
17
Alternatively, there are two other competing methods to compute markup rates. First, the demand approach
by Berry et al. (1995), which requires information on the demand side and, consequently, is more data-intensive and
hard to carry out for an extensive set of industries simultaneously. Second, the approach assumes constant returns
of scale, which implies that the markup rate can be estimated from the ratio of sales to costs.
18
We define an industry at the 2-digit level of the North American Industry Classification System (NAICS).
51
enue and expenditure data into units of output and input.
19
Profit rates
Our measure of profitability follows standard accounting practice by calculating the percentage of
sales that remains after subtracting the costs of goods sold and selling, general, and administrative
expenses from total sales.
Profit rate
it
=
Sales
it
− COGS
it
− XSGA
it
Sales
it
,
where XSGA is selling, general and administrative expenses.
3.2.2 Entry and exit of stock exchange listings in Compustat
Wefurther analyzethecompositionoffirms inCompustat’s surveyto evaluate whether our sample
exhibits similar patterns in the share of firms entering or exiting before and during the pandemic.
Figure 3.18 depicts the yearly entry and exit rates and the total number of companies from 1955–
2021.
One potential issue with our sample of firms is that the COVID-19 pandemic may have caused
some firms to leave Compustat. However, our sample indicates that only 7% of the firms in
Compustat in 2019 left the survey in 2020, which is only slightly above 1965–2019 exit rate average
of 5.7%. This exit rate increased in the second year of the pandemic, with 17.6% of firms with data
in2020leavingthesurveyin2021. Furthermore,firmsthatenterandexitCompustat’ssurveyeach
yeararesimilartothefirmsthatremaininthesurvey,atleastintermsofmarkupsandprofitrates.
Figure 3.19 plots the average markup and profit rates for three samples: first, a sample including
all the firms in a given year; second, firms entering the stock exchange; and third, firms exiting the
stock exchange. The markup rates for all three samples have been increasing since 1980, although
entering and exiting firms’ markup rates tend to vary more relative to the average markup rate.
Likewise, entering and exiting firms tend to have the same profit rates as the average firm. These
findings suggest that the firms that leave the survey are not significantly different from the firms
that remain.
19
For more details on this methodology, please see Appendix B of De Loecker et al. (2020).
52
Figure 3.18: Entry and Exit Rates
0
2000
4000
6000
8000
Total Number of Firms
0
.1
.2
.3
.4
(%)
1960 1970 1980 1990 2000 2010 2020
Entry rate (left) Exit rate (left) Number of firms (right)
Notes: The figure displays entry and exit rates for the firms in Compustat. We classify a firm as entering at year t if
t is the first year it appeared in the survey (the firm became publicly traded), and we label a firm as exiting at time
t if it is the last period where we observe the firm in the survey.
Figure 3.19: Average Markup and Profit Rates for Entrant and Exiting firms.
1
1.5
2
2.5
3
index
1960 1970 1980 1990 2000 2010 2020
Entry Exit All
(a) Average Markups
0
10
20
30
40
(%)
1960 1970 1980 1990 2000 2010 2020
Entry Exit All
(b) Average Profit Rates
Notes: This figure plots the average markups and profit rates for three samples: firms entering the stock exchange,
firms exiting, and all firms in the stock exchange.
53
3.2.3 Historical trends in markup and profit rates
We supplement this section with a historical analysis of markup and profit rates. Figure 3.20
presents annual markup and profit rates for publicly-traded firms in the US between 1955 and
2021. The average profit rate is at its highest level in history, whereas the average markup rate is
below its highest recorded value in 2016.
The markup rate saw a significant decrease between 2016 and 2019, but this trend reversed
during the pandemic. In the first year of the pandemic, the sales-weighted average markup rate
increased from 1.49 to 1.52. In the second year, it increased further to 1.54, which is slightly lower
thanthehistoricalmaximumobservedin2016. AsdocumentedinDeLoeckeretal.(2020)through
2016, the increase in markup rates was driven by the rising markups of top firms, which continued
from 2016–2021. For example, the third quartile markup rate increased from 1.81 to 1.82 between
2016 and 2021, while the markup rate for firms in the lower quartile decreased from 1.12 to 1.04.
Unlike markup rates, profit rates decreased in 2020 before recovering in 2021. For instance, the
sales-weightedprofitrate increasedfrom 17.6% to 21.6% between 2016 and 2021, with a temporary
decrease to 16.9% in 2020. Interestingly, the most profitable firms were less affected than less
profitable firms: the third quartile firms by profit rate experienced profit rates of 19.8% and 23.4%
in 2020 and 2021, respectively, with the latter being their highest profit rate since 195. In contrast,
the lower quartile firms reduced their profit rate from 3.1% to 1.2% between 2016 and 2020 before
rebounding to 3.7% in 2021.
Inadditiontotheincreaseinmarkupandprofitratescausedbythepandemic,ourfindingsindi-
catethattheirvolatilitywashigherthanusualduringthefirsttwoyearsofthepandemic. However,
this increased volatility was still within historical levels for most firms. Figure 3.21 shows how the
volatility of sales-weighted markup and profit rates changed from 1959 to 2021.
20
Consistent with
the heterogeneous effects of the pandemic on profit rates, the lower quartile experienced increasing
volatility from 2000 to 2021, with a particularly pronounced increase during the pandemic.
20
For a given time series, The 5-year rolling coefficient of variation is calculated as the absolute value of the ratio of
the standard deviation and the average, with the average and standard deviation being determined using data from
the previous five years.
54
Figure 3.20: Markup and Profit Rates
1 1.2 1.4 1.6 1.8
index
1960 1980 2000 2020
Average p25
p50 p75
(a) Markup Rates
0 5 10 15 20 25
%
1960 1980 2000 2020
Average p25
p50 p75
(b) Profit Rates
Notes: The figure shows the sales-weighted average, the first, second, and third quartiles for the markup rates, based
on De Loecker et al. (2020), and profit rates for publicly traded firms collected in Compustat. Markup rates are
calculated as the multiplication of industry-level variable-input output elasticities and the firm-level sales to cost of
goodssoldratio. Profitratesaretotalsalesminuscostsofgoodssoldandselling,generalandadministrativeexpenses
as a share of sales.
55
Figure 3.21: Markup and Profit Rate Volatility
0 .01 .02 .03 .04
index
1960 1980 2000 2020
Average p25
p50 p75
(a) Markups
0 .1 .2 .3 .4
index
1960 1980 2000 2020
Average p25
p50 p75
(b) Profit Rates
Notes: The figure shows the 5-year rolling coefficient of variation for the sales-weighted average, the first, second,
and third quartiles for the markup rates and profit rates for publicly traded firms collected in Compustat.
3.3 Methodology
Let Y
i,t
be an observed outcome of interest, either the markup rate or the profit rate, for a firm i
in period t. Let Y
i,t
(1) and Y
i,t
(0) be the potential outcomes for a firm i in time period t when a
pandemic happens and a pandemic does not happen, respectively. We denote T as the last period
before the pandemic happens.
Our main parameter of interest is the effect of the pandemic in a firm-level outcome, namely,
τ i,T+h
=Y
i,T+h
(1)− Y
i,T+h
(0), h=1,2,....
We only observe the potential outcome when a pandemic happens after time period T for any firm,
so Y
i,T+h
= Y
i,T+h
(1) for h = 1,2,..., but we cannot observe the outcome had the pandemic not
taken place, Y
i,T+h
(0).
To approximate the outcome had the pandemic not taken place, we propose to use the forecast
based on information up to period T as a counterfactual for each firm i, Y
i,T+h
(0). Specifically, let
I
T
denote all information available up to period T; we define our counterfactual as the conditional
expectation of Y
i,T+h
(0) given all observed information up to period T, i.e.,E[Y
i,T+h
(0)|I
T
], which
56
potentially includes pre-treatment values of both outcomes and available covariates.
Usingtheconditionalexpectation, wecancompute anapproximation to thecausal effect asthe
difference between the realized value of our outcome of interest and the counterfactual
e τ i,T+h
=Y
i,T+h
(1)− E[Y
i,T+h
(0)|I
T
].
To operationalize our counterfactual, we pick a family of models to estimate the conditional
expectation of the outcome had the pandemic not taken place. We explain this family of models in
Section 3.3.1. In this sense, our firm-level identification rests on the specific parametric family we
use to estimate the counterfactual. We denote our estimator as
b
Y
i,T+h
(0)≡ b
E[Y
i,T+h
(0)|I
T
]. Our
final estimator of the firm-level causal effect is
b τ i,T+h
=Y
i,T+h
− b
Y
i,T+h
(0).
The causal effect estimator, b τ i,T+h
, is a good approximation to the true causal effect, τ i,T+h
, under
twoconditions: first,thedifferencebetweentheactualpotentialoutcomewithoutthepandemicand
its counterfactual is near zero, Y
i,T+h
(0)− E[Y
i,T+h
(0)|I
T
]≈ 0, also called conditional expectation
error. Second, the difference between the actual forecast and the counterfactual is approximately
zero, E[Y
i,T+h
(0)|I
T
]− b
Y
i,T+h
(0) ≈ 0, which we refer to as forecasting error. We make these
conditions explicit in the following decomposition
b τ i,T+h
=τ i,T+h
+(Y
i,T+h
(0)− E[Y
i,T+h
(0)|I
T
])+(E[Y
i,T+h
(0)|I
T
]− b
Y
i,T+h
(0)). (12)
In practice, it is difficult to assess these two conditions. We can, however, aim to reduce the
forecasting error as much as possible by selecting the best model within a comprehensible family
of models.
Furthermore,wecanprovideconditionstoobtainaconsistentestimatoroftheaveragetreatment
effect across firms,
1
n
P
n
i=1
E[τ i,T+h
]. In particular, let Y
1
,...,Y
n
be a random sample, and define
E[τ T+h
]=E[τ i,T+h
]. In Equation (12), summing across firms and dividing by the number of firms,
57
we obtain
1
n
n
X
i=1
b τ i,T+h
=
1
n
n
X
i=1
h
τ i,T+h
+(Y
i,T+h
(0)− E[Y
i,T+h
(0)|I
T
])+(E[Y
i,T+h
(0)|I
T
]− b
Y
i,T+h
(0))
i
.
By invoking the law of large numbers, the average conditional expectation error should be close to
zero as the number of firms in our sample increases,
1
n
n
X
i=1
(Y
i,T+h
(0)− E[Y
i,T+h
(0)|I
T
])
P
− → E[Y
i,T+h
(0)]− E[E[Y
i,T+h
(0)|I
T
]]=0,
where the equality follows from the law of iterated expectations, and
P
− → means convergence in
probability.
Finally, if our estimator of the conditional expectation is consistent in the sense that
1
n
n
X
i=1
(E[Y
i,T+h
(0)|I
T
]− b
Y
i,T+h
(0))
P
− → 0, (13)
we can conclude that
1
n
n
X
i=1
b τ i,T+h
P
− →
1
n
n
X
i=1
E[τ i,T+h
]=E[τ T+h
],
where the equality follows from the random sample assumption.
3.3.1 Forecasting model
We use a Bayesian structural time series (BSTS) model forecast to create a counterfactual at the
firm level (Brodersen et al., 2015). These models are highly flexible in the sense that they embed a
broad class of time-series models. Besides their flexibility, they are also modular: they encompass
different features of time-series data such as trend, seasonality, or holiday effects.
Within the class of BSTS models, assume that the firm-level outcome of interest, Y
i,t
, has a
stochastic trend representation (also known as a local-level trend or random walk) as follows
Y
i,t
=δ i,t
+ε
i,t
, (14)
δ i,t
=δ i,t− 1
+η δ,i,t
, (15)
58
whereδ i,t
isatrendcomponentattimet,and,ε
i,t
andη δ,i,t
aremutuallyindependentGaussianerror
components with variances σ 2
ε
and σ 2
µ , respectively. In the state-space representation literature,
Equation (14) is commonly known as the observation equation, whereas Equation (15) is the state
equation.
For the quarterly specification, the model also incorporates a seasonal component. Specifically,
letγ i,t
be an additive quarterly component such that we augment the model in Equations (14)-(15)
with the following two equations
Y
i,t
=δ i,t
+γ i,t
+ε
i,t
, (16)
γ i,t+1
=− 2
X
s=0
γ i,t− s
+η γ,i,t
. (17)
The seasonal specification in Equation (17) allows for the seasonal component to change over
time.
21
Toavoidperfectmulticollinearityintheestimationoftheseasonalcomponent,theseasonal
component is restricted to sum to zero across all seasons and time periods.
To avoid identification or endogeneity issues, we decide not to include any contemporaneous
variables on the right-hand side of Equations (14) or (16). In principle, the pandemic may have
affected all contemporaneous variables that are relevant to markup and profit rates. The choice
to exclude these variables is supported in the bad control problem literature (Angrist and Pischke,
2008).
3.3.2 Estimation
To estimate the model in Equations (14)-(15) (Equations (14)-(17) in the quarterly model), we
implement Gibbs sampling. This method jointly estimates unobservables, such as trend and sea-
sonality components, as well as the parameters of the model, such as the error term variances. To
ensure that the results are not sensitive to the initial values chosen and given the Markov-chain
nature of Gibbs sampling, we choose 10,000 Monte-Carlo iterations.
Table 3.9 shows the priors for the variances of the model and the initial values of the state
components of the model. As is common with linear models and following Brodersen et al. (2015),
21
In contrast to a specification with seasonal indicator variables, which restricts the seasonal effect to be fixed
across years, the specification in Equation 17 allows the seasonal component to change over time (see Harvey, 1990,
Section 2.3.4).
59
we set an inverse-Gamma prior for any of the variances σ 2
i
(either σ 2
ε,i
, σ 2
δ,i
, or σ 2
γ,i
) as 1/σ
2
i
∼ G(v/2,s/2) for firm i. Since the ratio of the hyperparameters s and v is equal to the expected
value of σ 2
i
, it is common to refer to s as the sum of squares and v as the degrees of freedom. A
common choice for the hyperparameters is to set s/v = 0.1s
y
i
, where s
y
i
is the sample standard
deviation of the outcome of interest Y
i
for firm i. As in Brodersen et al. (2015), we have a prior
belief that the error components are small so we set small values for both the degrees of freedom,
v, and the expected value of the variance, s/v. We use weakly informative Gaussian priors for the
initial values of the state parameters.
Table 3.9: Priors on Bayesian Structural Model
Parameter Prior density Hyperparameter 1 Hyperparameter 2
σ ε,i
Inverse-Gamma 1 0.01
σ δ,i
Inverse-Gamma 0.01 32
σ γ,i
Inverse-Gamma 0.01 0.01
δ i,1
Normal
y
0
− y
ˆ σ y
1
γ i,1
Normal 0 1
Notes: σ ϵ,i ,σ δ,i
,andσ γ,i arethevariancesoftheobservationequationerror(Equa-
tion (14)), trend, and seasonal components, respectively, for a firm i. δ i,1 and γ i,1
are the initial values of the trend and seasonal state components. When the prior
density is inverse-gamma, hyperparameters 1 and 2 can be interpreted as the de-
grees of freedom and the sum of squared errors, respectively (see Brodersen et al.,
2015). When the prior density is normal, hyperparameters 1 and 2 are the mean
and standard deviation, respectively.
3.3.3 Assumption testing, model fit, and quality of forecasts
We perform a battery of statistical tests to evaluate whether the firm-level markup and profit
rates follow the data-generating process described in Equations (14)-(15), which forms the base of
our forecasts. Under this data-generating process, the differentiated series of markup and profit
rates, Y
i,t
− Y
i,t− 1
, should be normally distributed for each of the firms in our sample. To test for
the normality of the differentiated series, we perform Jarque-Bera tests on these series. Indeed,
we do not find evidence to reject the hypothesis that a majority of these series have a normal
distribution. Specifically, for yearly (quarterly) markups, we find that 84.6% (64%) of the cases
among 3,604 (3,192) firms do not reject the null hypothesis of normality at the 5% level after
applying the Bonferroni correction for multiple testing. Similarly, for yearly (quarterly) profit
rates, we find that 82.2% (74%) of our tests among 3,611 (3,192) firms do not reject the normality
60
Table3.10: FractionofPeriodsthatObservedOutcomeFallsOutsideofBayesianCredibleIntervals
Outcome (frequency) Average Percentile 75th Number of firms
Markups (quarterly) 0.025 0.043 3,192
Markups (yearly) 0.028 0.053 3,611
Profit rate (quarterly) 0.029 0.045 3,192
Profit rate (yearly) 0.035 0.063 3,611
Notes: The table introduces model fitness measures computed for markup rates
and profit rates during the pre-pandemic period. For each firm, we compute the
fraction of observations of the pre-pandemic period that do not belong to the
estimated 95% Bayesian credible intervals. The table displays the average and
the 75th percentile of these fractions across firms.
assumption after controlling for the Bonferroni correction. We obtain similar results after using
other corrections for multiple testing, including the Holm, Hochberg, and Hommel corrections. In
summary, we do not find evidence to reject the normality assumption embedded in our forecasting
model.
Wealsoevaluatewhetherthefirm-specificmodelsaccuratelyreflecttheobservedfirm’sbehavior
before the pandemic. To this purpose, we calculate the fraction of periods for each firm when the
observed outcome, either markup or profit rate, falls outside the model’s 95% Bayesian credible
interval for each period that the firm has data before 2020. Table 3.10 shows the average of these
fractions across firms for each outcome and frequency. On average, 3% of the observations are
not within the Bayesian credible interval for both markup and profit rates. In addition, at least
75% of the firms have a model that leaves no more than 6.3% of the periods outside of the 95%
Bayesian credible intervals (see Table 3.10, column 3). These results suggest that most of the
models correctly pin down the pre-pandemic trends.
Furthermore, wefindthatourforecastshaveadequateforecastquality. Toevaluateourmodel’s
forecast quality, we compute forecasts for markup rates and profit rates for 2018–2019 using firm-
level models that are estimated with observations up to 2017. We present some common forecast
quality statistics in Table 3.11. Two of these statistics confirm the forecast quality of our models.
First, as suggested by Equation (13), the objective is to minimize the mean forecast error (ME) to
be close to zero, which seems to be the case for the markup rates. Second, for at least half of the
firms, the median absolute percentage error (MAPE) is 0.1 and 0.3 for markup and profit rates,
respectively. These metrics support the forecast quality of our Bayesian models.
61
Table 3.11: Forecast Quality Statistics of Bayesian Structural Time Series Model
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 -0.008 0.827 0.373 0.215 0.204 0.107 2858
Markup rate 2019 -0.046 0.900 0.409 0.241 0.241 0.122 2872
Profit rate 2018 19.923 147.174 28.151 3.402 2.951 0.335 2858
Profit rate 2019 18.482 151.506 30.590 4.464 3.610 0.380 2872
Notes: The table introduces forecast quality statistics computed for 2018–2019 forecasts for
markupandprofitratesestimatedusingfirm-levelmodelswithsamplesupto2017. Thestatistics
computed are: mean error (ME), root mean square error (RMSE), mean absolute error (MAE),
standardized mean absolute error (MAES), and the mean absolute percentage error (MAPE).
Median MAPE is the median MAPE across firms. n is the number of firms.
3.4 Counterfactual analysis results
We introduce the results of the counterfactual effects of the pandemic in the markups and profit
rate in this section. We first summarize the findings for the average firm. We then perform a
heterogeneity analysis to elucidate which firms fared better during the pandemic.
3.4.1 Aggregate data
Most firms had lower markup rates than what could have been observed without the pandemic
happening. To illustrate this point, we show in Figure 3.22 four aggregates of both the observed
and counterfactual markup rates, namely, its sales-weighted average and quartiles in 2000–2021.
For instance, for the average firm, the observed markup rate was 4.3% and 6.6% lower than its
counterfactual value in 2020 and 2021, respectively. These effects are significant as the observed
markup rate falls outside the 95% credible interval of the counterfactual. Furthermore, there is
considerable heterogeneity across firms in terms of quartiles of markup rates. For instance, the
firms in the third quartile of markup rates had the largest negative effects in 2020 (2021) with a
decrease of 5% (4.6%) relative to their counterfactual value. In contrast, the median markup firm
(quartile 2) experienced less dramatic effects: a 3.3% (2.6%) lower markup rate with respect to
their counterfactual in 2020 (2021).
22
22
Similar effects arise when using quarterly data (see Figure H.38b in Appendix H).
62
Figure 3.22: Comparison of Observed and Counterfactual Markup Rates using Bayesian Structural
Time Series Models
1.50
1.55
1.60
1.65
1.70
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual
(a) Average
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2.0
2010 2015 2020
p25 p50 p75
(b) Quartiles
Note: The figure shows four aggregate statistics of the yearly observed and counterfactual markup rates. The
statistics are the sales-weighted average (shown in Panel (a)), the first quartile (shown in Panel (b) in red), the
second quartile (shown in Panel (b) in black), and the third quartile (shown in Panel (b) in gray). The solid lines
represent the observed values, while the dotted lines represent the counterfactual values for 2020-2021, as calculated
using firm-level Bayesian structural time series models. The shaded areas represent the 95%, 90%, 80%, and 50%
(equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations on Panel (a). 95%
(equally-tailed) credible intervals are displayed for the statistics on Panel (b).
63
Figure 3.23: Comparison of Observed and Counterfactual Profit Rates using Bayesian Structural
Time Series Models
13
14
15
16
17
18
19
20
21
22
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual
(a) Average
0
5
10
15
20
2010 2015 2020
p25 p50 p75
(b) Quartiles
Note: The figure shows four aggregate statistics of the yearly observed and counterfactual profit rates. The statistics
are the sales-weighted average (shown in Panel (a)), the first quartile (shown in Panel (b) in red), the second quartile
(shown in Panel (b) in black), and the third quartile (shown in Panel (b) in gray). The solid lines represent the
observed values, while the dotted lines represent the counterfactual values for 2020-2021, as calculated using firm-
levelBayesianstructuraltimeseriesmodels. Theshadedareasrepresentthe95%,90%,80%,and50%(equally-tailed)
credible interval for each statistic, calculated based on 5,000 posterior simulations on Panel (a). 95% (equally-tailed)
credible intervals are displayed for the statistics on Panel (b).
Unlike the observed negative effects of markups, we find that most firms had higher profit rates
than their counterfactuals during the pandemic. Figure 3.23 shows the (sales-weighted) average
and quartiles of both the observed and counterfactual profit rates across firms. When analyzing
the counterfactual effects of the pandemic on profit rates, most firms had a positive impact. For
instance,theaveragefirmhadaprofitrateof16.9%and21.6%in2020and2021,respectively,which
are 2.1 and 6.4 percentage points higher than the counterfactuals. These effects are significant as
theobservedmarkupratefallsoutsidethe95%credibleintervalofthecounterfactual. Additionally,
firmswithhighermarkupratesexperiencedlowermarkupratesthanwhatcouldhavebeenexpected
without the pandemic. For instance, the third-quartile firm registered a 19.8% profit rate in 2020
which is 1.7 percentage points lower than the profit rate these companies would experience had the
64
pandemic not happened. In contrast, firms in the first and second quartiles experienced no effects
on profit rates in 2020 while experiencing positive effects in the second year of the pandemic.
23
Lookingatbothresultsformarkupsandprofitratesjointly,itseemsthattheuncertaintyduring
the pandemic potentially promoted the decrease in markups and the increase in profits(Altig et al.,
2020). In the next section, we explore the heterogeneity of these effects across firm characteristics
as well as the industries most affected.
3.4.2 Heterogeneity analysis
We investigate potential heterogeneity in the effects of the pandemic on markup rates and prof-
itabilitytoelucidatewhetherfirmswithdifferentlevelsofkeycharacteristicsdifferintheirresponse
to the pandemic. These characteristics include pre-pandemic levels of cost of goods sold, sales, em-
ployment, stock tenure, and market share. Specifically, to summarize the effect heterogeneity in a
convenient form, we run the following cross-sectional regression
b τ i
=β 0
+β 1
logCOGS
i
+β 2
logSales
i
+β 3
logEmployment
i
+β 4
Stock tenure
i
+Market share
i
+δ j
+ϵ i
,
whereb τ i
is the average effect in 2020–2021 for a firm i (on either markup or profit rates), the right-
handsidevariablesarefirm-levelaveragesforCOGS,sales,employment,stocktenure,marketshare
in 2015–2019, and δ j
is an industry-level fixed effect such that firm i belongs to industry j =j(i).
Tables 3.12 and 3.13 show the estimated regression for the yearly specification for markups and
profit rates, respectively. As expected, those firms with higher tenure in the stock exchange have
larger positive effects on markup rates since they are more likely to be in a consolidated industry
or have survived a merger or an acquisition. Likewise, firms with high sales or low costs have
mechanically more positive effects on profit rates. These regressions suggest that market share,
number of employees, or tenure in the stock market imply higher effect heterogeneity in profit
rates.
Wefindsimilareffectheterogeneityresultswhenusingthequarterlyspecificationforthemarkup
andprofitrates(seeTablesH.34andH.35,respectively). Likewise,wefindsimilarconclusionswhen
23
Similar effects arise when using quarterly data (see Figure H.39b in Appendix H).
65
webuildscatterplotsoffirm-leveleffectsagainstpre-pandemicvaluesofkeyfirmcharacteristics. In
particular,weplottheeffectineithermarkuporprofitratesagainstpre-pandemicvaluesofmarkup
rates,sales,stockexchangetenure,profitrate,marketshare,andemployment—thebinscatterplots
in Appendix I illustrate this analysis.
We provide further evidence of the heterogeneity of the response to the pandemic by tracking
thefractionoffirmswithnegativeandpositiveeffectsofthepandemicineachquarterin2020–2021
(see Appendix H.1). In particular, at the beginning of the pandemic, within those firms that are
significantly affected—by having an observed outcome outside the 95% credible interval for the
counterfactual outcome— most firms are negatively affected both in terms of markups and profit
rates. Bythelastquarterof2021, theshareoffirmswithpositiveeffectsincreasedbutisstilllower
than those with negative effects. This suggests that there were more negatively affected firms than
positively affected firms in most quarters.
66
Table 3.12: Heterogeneity in the effects of the pandemic on markups
(Yearly Specification)
Dependent Variable: Average markup rate pandemic effect in 2020–2021
All 2020 2021
Model: (1) (2) (3) (4) (5) (6)
Variables
COGS 0.0856 0.0659 0.0490 0.0463 0.1300 0.0937
(0.0851) (0.0912) (0.0893) (0.0965) (0.1497) (0.1591)
Sales -0.1534
∗ -0.0864 -0.1393 -0.0975 -0.1689 -0.0724
(0.0905) (0.0926) (0.0946) (0.0983) (0.1594) (0.1614)
Employment 0.0519
∗∗∗ 0.0082 0.0951
∗∗∗ 0.0568
∗∗∗ -0.0016 -0.0550
∗ (0.0156) (0.0170) (0.0183) (0.0187) (0.0260) (0.0295)
Stock-exchange tenure 0.0030
∗∗∗ 0.0031
∗∗∗ 0.0025
∗∗∗ 0.0018
∗∗ 0.0035
∗∗∗ 0.0043
∗∗∗ (0.0007) (0.0007) (0.0007) (0.0008) (0.0012) (0.0013)
Market share 0.8069
∗∗ -0.3262 -0.0722 -0.6241 1.818
∗∗ 0.0390
(0.3312) (0.4701) (0.3854) (0.5395) (0.7961) (0.7951)
Fixed-effects
2-digit NAICS industry Yes Yes Yes
Mean 1.700 1.674 1.732
Mean effect -0.069 -0.086 -0.048
Fit statistics
Observations 5,771 5,771 3,139 3,139 2,632 2,632
R
2
0.00632 0.05509 0.01250 0.05801 0.00734 0.07110
Within R
2
0.00303 0.00495 0.00561
Notes: The table presents cross-section firm-level regressions of the average pandemic effect at the firm-level
on the 2015–2019 average of the logarithm cost of goods sold, logarithm of sales, logarithm of employment,
years since publicly-traded, and market shares. NAICS = North American Industrial Classification System.
Heteroskedasticity-robust standard-errors in parentheses. Signif. Codes: ***: 0.01, **: 0.05, *: 0.1.
67
Table 3.13: Heterogeneity in the effects of the pandemic on profit rates
(Yearly Specification)
Dependent Variable: Average profit rate pandemic effect in 2020–2021
All 2020 2021
Model: (1) (2) (3) (4) (5) (6)
Variables
COGS -19.46
∗∗∗ -21.09
∗∗∗ -18.76
∗∗∗ -15.69
∗∗ -20.03
∗ -27.06
∗∗ (5.923) (7.359) (6.236) (6.714) (10.50) (13.76)
Sales 16.81
∗∗ 18.46
∗∗ 12.90 10.12 21.25 28.37
∗ (7.897) (9.153) (9.326) (9.903) (13.09) (16.04)
Employment -0.2620 -0.2221 3.092 2.458 -4.328 -3.901
(3.919) (4.421) (4.992) (5.771) (6.238) (6.861)
Stock-exchange tenure -0.0211 -0.0288 -0.0914 -0.1333 0.0539 0.0761
(0.0667) (0.0796) (0.0795) (0.0978) (0.1117) (0.1303)
Market share -5.302 -29.90 -10.50 15.34 -1.726 -84.34
(26.36) (72.81) (37.57) (84.41) (37.10) (127.3)
Fixed-effects
2-digit NAICS industry Yes Yes Yes
Mean 2.438 1.274 3.833
Mean effect 7.145 5.443 9.183
Fit statistics
Observations 5,771 5,771 3,139 3,139 2,632 2,632
R
2
0.00534 0.01288 0.00867 0.02122 0.00411 0.01382
Within R
2
0.00509 0.00664 0.00527
Notes: The table presents cross-section firm-level regressions of the average pandemic effect at the firm-
level on 2015–2019 average of the logarithm cost of goods sold, the logarithm of sales, logarithm of
employment, years since publicly-traded, and market shares. NAICS = North American Industrial Clas-
sification System. Heteroskedasticity-robust standard errors in parentheses. Signif. Codes: ***: 0.01, **:
0.05, *: 0.1
68
3.4.3 Heterogeneity by industry
We supplement our heterogeneity results by analyzing the effects of industry. Figure 3.24 shows
the industry breakdown of the effects on markup and profit rates. First, we observe that the trans-
portation industry experienced negative effects on profit rates in both 2020 and 2021. Following a
negative demand shock due to lockdowns, we also note a fall in this industry’s markups. Interest-
ingly, we observe similar behavior in the oil and gas industry; in 2020, lockdowns drove demand
and prices down, which lead the industry to report losses. However, the commodity price surge in
2021, partly in response to a myriad of public policy responses, triggered increases in profit rates
of roughly 10%.
Figure 3.24: Effects on Markups and Profit Rates by Industry
Warehousing
Entertainment
Food Manufacturing
Education
P75
Industrial Manufacturing
Agriculture
Finance
Healthcare
Construction
General Retail
Wholesale Trade
Administrative Services
Materials and Food Retailers
P25
Mean
Professional Services
Hospitality
Other Services
Mining and Oil
Other
Utilities
Transportation
Chemical Manufacturing
Real Estate
Information
−0.75 −0.50 −0.25 0.00 0.25 0.50 0.75 1.00
Effect in markups
2020 2021
(a) Markups
Warehousing
Real Estate
Industrial Manufacturing
P75
Information
Healthcare
Professional Services
Food Manufacturing
Construction
General Retail
Finance
Other Services
Administrative Services
Mean
Utilities
P25
Materials and Food Retailers
Other
Wholesale Trade
Chemical Manufacturing
Education
Agriculture
Mining and Oil
Hospitality
Entertainment
Transportation
−20 −15 −10 −5 0 5 10 15 20
Effect in profit rate
2020 2021
(b) Profit Rates
Note: The figure plots the sales-weighted average effects on markup and profit rates by the 2-digit NAICS industry.
The mean and percentiles 25 (P25) and 75 (P75) correspond to the aggregate effects for the sample of 3,611 firms
in Compustat (see also Figures 3.22 and 3.23). Industries are organized in ascending order according to their
corresponding effect in 2020.
Similarly, hospitality displays adverse effects on profit rates in 2020, with roughly no effect
on markups. In contrast, information and technology, warehousing, and real estate experienced
69
profit rates above what could have been predicted by previous trends. Likewise, professional and
administrative services, which could easily migrate to online environments, had profit rates above
their historical trends. Lastly, we see modest effects on the retail industries for both markups and
profit rates, which are known to have low markups (Philippon, 2019).
3.4.4 Business cycle analysis
Onemaywonderthatfluctuationsineconomicactivitymayaffecteithermarkuporprofitrates. To
answerthisquestionandundertheassumptionthatthepandemicdoesnotaffectthesefluctuations,
we further augment Equation (14) with a cyclical component.
Y
it
=δ it
+α cycle
t
+ε
it
, (18)
where cycle
t
is taken as the Hodrick-Prescott decomposition of the US real gross domestic product
in period t.
Figure 3.25: Histogram of Coefficient of the Cycle in Models (Yearly Frequency)
0
2
4
−1.0 −0.5 0.0 0.5 1.0
Coefficient
(%)
(a) Markup Rates
0
2
4
6
−1.0 −0.5 0.0 0.5 1.0
Coefficient
(%)
(b) Profit Rates
Note: The plots show the histograms for the average firm-specific coefficient of the cycle using a yearly specification.
The red dotted line shows the average across all of these coefficients. Data were standardized before estimating the
model, so the coefficients are included in the [ − 1,1] interval.
70
We find evidence for acyclical markups and profit rates on average across firms. In particular,
we estimate the firm-level coefficient of the GDP cycle on both markup and profit rates, α in
Equation (18). Figures 3.25a and 3.25b show histograms of these coefficients for separate models
for markup and profit rates. This distribution suggests that, on average, the cycle does not seem
to affect markups and profit rates. Furthermore, we compute the percentage of significant cycle
coefficients. In the case of markup rates, we find 63% of the coefficients to be significant with a
rather symmetrical share of positive (30%) and negative (33%) coefficients. Similarly, in the case
of profit rates 62% of the coefficients are significant (34% positively significant and 28% negative
significantly).
These wide-ranging cycle coefficients may be connected to the literature on the cyclicality of
markups, which has identified potential reasons for both pro- and counter-cyclical markups. In
particular, thebehaviorofliquidity-constrainedfirmsthatadjusttheircustomerbasetostayafloat
during recessions may explain counter-cyclical markups (Gilchrist et al., 2017), with larger firms
being able to further smooth these liquidity constraints out (Hong, 2017b), whereas Keynesian
demand effects may explain pro-cyclical markups (Nekarda and Ramey, 2020). It seems that these
threeconjecturesmayhavesomeweightinexplainingtheapparentaverageacylicalityinthemarkup
and profit rates.
3.5 Robustness of findings to other modeling strategies and choices
Inthissection, weintroducetwoapproachestoevaluatetherobustnessofourmainfindings. Inthe
firstapproach,wemodifythehyperparametersassociatedwiththepriorsoftheBayesianstructural
timeseries(BSTS)models,whichwerefertoassensitivityanalysis. Inthesecondapproach,weuse
the commonly-used methodology of local linear projections (LLPs) to evaluate if our main findings
are robust to this fairly flexible strategy. Our sensitivity analysis corroborates our main results
in Section 3.4. The LLPs approach gives similar point effects in markup rates compared to the
Bayesian structural model, and a smaller effect in profit rates; however, it seems that LLPs lack
power for this data as the confidence intervals seem to be too wide.
71
3.5.1 Robustnessoffindingstoothermodelingchoices: ahyper-parametersensitivity
analysis.
We validate the findings from the BSTS model by performing a sensitivity analysis of our choice
of hyperparameters. This analysis ensures that our results are not driven by this choice. As noted
in Section 3.3.2, the hyperparameters govern the prior densities of the variances and initial values
of the trend and seasonality structural components. We perform two main exercises: we either
increaseordecreasehyperparametersthatgovernthescaleofthepriorsdensitiesby25%compared
tothebaselinevaluesinTable3.9;bydoingso,weallowfortheincreasesinmarkupandprofitrates
to have either more or less variability (Brodersen et al., 2015). As the location hyperparameters
associated with the initial values are calibrated with summary statistics of the actual series, we do
not change those location hyperparameters. Table 3.14 shows the choice of hyperparameters for
each of these exercises.
Table 3.14: Sensitivity Analysis on Hyperparameters of Bayesian Structural Model
(a) Exercise 1: 25% Increase in Hyperparameters
Parameter Prior density Hyperparameter 1 Hyperparameter 2
σ ε,i
Inverse-Gamma 1.25 0.0125
σ δ,i
Inverse-Gamma 0.0125 40
σ γ,i
Inverse-Gamma 0.0125 0.0125
δ i,1
Normal
y
0
− y
ˆ σ y
1.25
γ i,1
Normal 0 1.25
(b) Exercise 2: 25% Decrease in Hyperparameters
Parameter Prior density Hyperparameter 1 Hyperparameter 2
σ ε,i
Inverse-Gamma 0.75 0.0075
σ δ,i
Inverse-Gamma 0.0075 24
σ γ,i
Inverse-Gamma 0.0075 0.0075
δ i,1
Normal
y
0
− y
ˆ σ y
0.75
γ i,1
Normal 0 0.75
Notes: σ ϵ,i , σ δ,i
, and σ γ,i are the variances of the observation equation error (Equation (14)), trend, and
seasonalcomponents,respectively,forafirm i. δ i,1 andγ i,1 aretheinitialvaluesofthetrendandseasonal
state components. When the prior density is inverse-gamma, hyperparameters 1 and 2 can be interpreted
as the degrees of freedom and the sum of squared errors, respectively (Brodersen et al., 2015, see). When
the prior density is normal, hyperparameters 1 and 2 are the mean and standard deviation, respectively.
Figure 3.26 shows the change in the effects on markups and profit rates under this sensitivity
analysis. After either increasing or decreasing the variability of our priors, our finding that the
72
markup rate is negatively affected for most firms does not change. Likewise, our findings for the
profit rates on the impact of the pandemic remain unchanged: on average, firms performed better
than the counterfactual without the pandemic. In general, when we increase the variability of our
hyperparameters, the effects are larger for most firms both in terms of markup and profit rates.
Figure3.26: ComparisonofObservedandCounterfactualMarkupsandProfitRatesUnderDifferent
Hyperparameter choices.
1.50
1.55
1.60
1.65
1.70
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Sales weighted average Counterfactual
Counterfactual 125% HP Counterfactual 75% HP
(a) Markup Rate
13
14
15
16
17
18
19
20
21
22
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Sales weighted average Counterfactual
Counterfactual 125% HP Counterfactual 75% HP
(b) Profit Rate
Note: The figure shows yearly sales-weighted average observed and counterfactual markup and profit rates. The
solid lines represent the observed values, while the dotted lines represent the counterfactual values for 2020-2021, as
calculated using firm-level local linear projection models. The shaded areas represent the 95%, 90%, 80%, and 50%
(equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations.
3.5.2 Robustness of findings to other modeling strategies: local linear projections.
We also validate the robustness of our findings by estimating forecasts using the flexible approach
of Local Linear Projections (LLPs) (Jord` a, 2005; Montiel Olea and Plagborg-Møller, 2021). This
method creates forecasts using linear regressions of the outcome of interest on its lags and possibly
lags of other exogenous variables, which offers a flexible linear approximation of the conditional
expectation of the outcome given exogenous variables.
We estimate LLPs with either firm-specific projection coefficients, which we called firm-specific
73
LLPs, or with projection coefficients constant across firms, which we called panel LLPs. Particu-
larly, in the case of the firm-specific LLPs, for every forecast horizon h and firm i, we consider the
linear projection of the outcome of interest, Y
i,t+h
, on a vector stacking a constant and outcome
lags up to period t, X
it
=(1,Y
i,t− 1
,Y
i,t− 2
,...,Y
i,t− p
i
), with p
i
a positive integer,
Y
i,t+h
= X
′
it
β ih
+ξ ith
, h= 1,2,...H; i=1,...,I, (19)
where β ih
is a p
i
+1 vector of projection coefficients, ξ ith
is the projection residual, and H is the
maximum forecast horizon, i.e., either two years or eight quarters. Thus, Equation (19) provides
the firm-specific LLPs forecast
b
Y
i,t+h
for either markup or profit rates. To estimate these models,
we select firm-specific lag lengths p
i
according to the Akaike information criterion, with a lag of
up to 3 years with annual data and up to 8 quarters with quarterly data. As with BSTS, we also
evaluate whether LLPs provide precise forecasts. Particularly, we find that LLPs tend to produce
similarly accurate forecasts (if not slightly more accurate) compared to BSTS models according to
most forecast quality statistics we consider (see Tables 3.11 and 3.15).
24
24
In Appendix G.1, we use the same sample of firms and periods to calculate forecast quality statistics for BSTS
and LLPs – we do not see relevant changes to the comparison across methods after using this sample.
74
Figure 3.27: Comparison of Observed and Counterfactual Profit Rates using Local Linear Projec-
tions
1.40
1.45
1.50
1.55
1.60
1.65
1.70
1.75
1.80
1.85
2010 2015 2020
Observed
Counterfactual (LP)
95% CI 90% CI
80% CI 50% CI
(a) Markup Rate
15
16
17
18
19
20
21
2010 2015 2020
Observed
Counterfactual (LP)
95% CI 90% CI
80% CI 50% CI
(b) Profit Rate
Note: The figure shows yearly sales-weighted average observed and counterfactual markup and profit rates. The
solid lines represent the observed values, while the dotted lines represent the counterfactual values for the period of
2020-2021, as calculated using firm-level local linear projection models. The shaded areas represent the 95%, 90%,
80%, and 50% (equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations.
Using LLPs, we replicate our analysis of aggregate markup and profit rates in Section 3.4.1. In
particular, Figure 3.27 shows the counterfactual markup and profit rates estimated using firm-
specific LLPs. These LLPs forecasts result in similar point estimates for markup rates when
compared to the BSTS, but with higher point estimates for profit rates, which suggests that the
firm-specific LLPs predict a smaller effect of the pandemic on profit rates.
To test for significant effects using the firm-specific LLPs, we compute aggregate confidence
intervals using the bootstrap method. Specifically, we create 10,000 resamples with replacement
by sampling firms in each year of the pandemic. We then create a 95% confidence interval by
taking the 2.5th and 97.5th percentiles of the average response across the 10,000 resamples. These
confidence intervals appear to be wide, with a larger range than the BSTS’s credible intervals,
which only allows us to reject the null hypothesis of no significant effect on the aggregate profit
rate in 2021.
75
Table 3.15: Forecast Quality Statistics of Local-Linear Projections
(a) Firm-Specific Local-Linear Projection
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 -0.019 0.697 0.264 0.152 0.150 0.058 2858
Markup rate 2019 -0.086 0.821 0.358 0.211 0.217 0.091 2751
Profit rate 2018 23.490 488.429 30.526 3.689 2.294 0.203 2858
Profit rate 2019 16.957 206.812 29.079 4.244 3.029 0.313 2762
(b) Panel Local-Linear Projection
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 0.017 0.555 0.223 0.128 0.120 0.054 2808
Markup rate 2019 -0.027 0.743 0.309 0.182 0.180 0.081 2808
Profit rate 2018 5.809 96.882 16.257 1.965 1.590 0.318 2843
Profit rate 2019 2.536 75.113 18.855 2.752 2.305 0.391 2825
Notes: Thetableintroducescommonforecastqualitymeasurescomputedformarkupratesandprofitrates
for 2018–2019 with firm-level models estimated using information up to 2017. The measures computed
are: mean error (ME), root mean square error (RMSE), mean absolute error (MAE), standardized mean
absolute error (MAES), and the mean absolute percentage error (MAPE). Median MAPE is the median
of the MAPEs across firm-level MAPEs. n is the number of firms.
Additionally, we extend our LLPs framework to analyze whether potential unobserved hetero-
geneityintheprojectioncoefficient(Equation (19))plays a role in the forecast accuracyof LLPs.
25
We refer to this extension as panel LLPs. To this purpose, we extend the linear projection in
Equation (19) to incorporate a constant projection coefficient across firms and firm fixed effects as
follows,
Y
i,t+h
= X
′
it
β h
+u
ih
+ζ ith
; h= 1,2,...,H; i=1,...,I; (20)
whereX
it
=(1,Y
i,t− 1
,Y
i,t− 2
,...,Y
i,t− p
)isap+1vectorstackingaconstantandlagsoftheoutcome
Y
it
, β h
is a panel-level projection coefficient, u
ih
is a firm fixed effect, and ζ ith
is a projection
residual.
26
Similarly to the firm-specific LLPs specification, we choose the lag length p based on
theAkaike informationcriterion.
27
Using this extended model, we compute individual forecasts for
25
When favoring either firm-specific estimation (Equation (19)) instead of pooled estimation (Equation (20)),
Pesaran et al. (2022) argues that either (i) a high level of heterogeneity in the projection coefficient, (ii) a large
number of observed periods for each firm, or (iii) models with outcome lags will favor firm-specific estimation.
26
In Equation (20), we do not include time fixed effects as we cannot forecast time fixed effects for 2020 and 2021
which makes it infeasible to compute forecasts for the outcome of interests.
27
Wealsoestimatethefollowingspecificationsnotreportedhere: (i)Equation(20)replacingfirm-fixedeffectswith
76
Figure 3.28: Comparison of Observed and Counterfactual Profit Rates using Panel Local Linear
Projections
1.35
1.40
1.45
1.50
1.55
1.60
1.65
1.70
1.75
1.80
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual (LP panel)
(a) Markup Rate
15
16
17
18
19
20
21
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual (LP panel)
(b) Profit Rate
Note: The figure shows yearly sales-weighted average observed and counterfactual markup and profit rates. The
solid lines represent the observed values, while the dotted lines represent the counterfactual values for the period of
2020-2021, as calculated using a panel data local linear projection model. The shaded areas represent the 95%, 90%,
80%, and 50% (equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations.
each firm and bootstrapped standard errors as before, and plot these estimates in Figure 3.28.
28
The panel LLPs results are quite similar to the firm-level LLPs, i.e., there are no effects in
markups and profit rates but positive effects in profit rates in 2021 (see Figures 3.27 and 3.28). In
summary, it seems that the LLPs have limited power for this data, as their confidence intervals are
relatively wide and their forecast quality statistics indicate they are not as precise as the BSTS.
3.6 Conclusion
We assess the impact of the COVID-19 pandemic on market power and profitability. Specifically,
we compare the realized series of markup and profit rates to a forecast based on information up to
the beginning of the pandemic. Essentially, we argue that this forecast represents a counterfactual
industry fixed effects, (ii) Equation (20) without the firm fixed effect u
ih
(or pooled specification), and (iii) Equation
(20) with industry-specific projection coefficients β h
. We find similar results in these specifications to our baseline
specification in Equation (20).
28
It seems that panel LLPs are more accurate than firm-specific LLPs (see Panel (a) and (b) in Table 3.15), which
suggests that there is no high unobserved heterogeneity in the projection coefficient, β ih
. In Appendix G.1, we use
the same sample of firms and periods to calculate forecast quality statistics for BSTS and LLPs – we do not see
relevant changes to the comparison across methods after using this sample.
77
of what would have happened if the pandemic had never occurred. Our counterfactual reveals
that the COVID pandemic adversely affected markup rates for the average firm in 2020 and 2021.
Likewise, we find that most firms had record profit rates that are significant in the sense that
pre-pandemic trends cannot explain those trends.
We estimate counterfactuals using the arguably flexible framework of Bayesian structural time
series (BSTS), which can accommodate multiple sources of variation, including local trends and
seasonality. On the one hand, these estimations suggest that markup rates were 4.3% and 6.6%
lower than their counterfactuals in 2020 and 2021, respectively. On the other hand, the counter-
factual suggests that had the pandemic not happened, the profit rate would have been 15.2% in
2021 which is 6.4 percentage points lower than the observed value in 2021. Furthermore, we find
that the effects of the pandemic are statistically significant.
Ourfindingsarelargelyrobusttochangesinourmodelingstrategy. Particularly,weevaluatethe
plausibility of our assumptions and the robustness of our results by implementing three strategies.
First, specification and goodness-of-fit tests support the assumptions and the quality of the model.
Second, our results are largely resilient to changes to modeling choices, such as different data
frequencies(eitherquarterlyoryearly)orchangingthehyperparametersoftheBSTSpriors. Third,
similar findings follow from counterfactuals constructed from other modeling strategies, such as
local linear projections, which is a commonly used and flexible method to compute forecasts. In
summary, specification and goodness-of-fit tests validate our assumptions, and other modeling
strategies indicate that our results are robust.
Furthermore, this paper uncovers the heterogeneity of these effects in terms of pre-pandemic
baseline characteristics such as markup rates, firm size, stock-exchange tenure, employment, prof-
itability, and market shares. Our findings show that companies with a longer history on the stock
exchange tend to have a larger impact on markup rates. Additionally, companies with high sales
or low costs tend to have a significant positive impact on profit rates.
We complement this heterogeneity analysis with a breakdown of the effects of the pandemic
by industry. Particularly, industries such as information, real estate, and chemical manufacturing
had lower markups than expected, while warehousing and entertainment had higher markups. A
similaranalysisofprofitratesbyindustryshowsthattransportation,entertainment,andhospitality
tended to have lower profit rates than expected, whereas warehousing and real estate had higher
78
profit rates.
Our results have implications for the current academic and economic policy debate on the
impacts of the pandemic on businesses. For instance, this work can help guide future research on
the impact of fiscal and monetary COVID relief policies. Likewise, understanding the changes,
and the subsequent policy response, in markup and profit rates during the pandemic can help
policymakersguidestrategiesforaddressingtheimpactsofthepandemiconmarkets. Additionally,
future research will uncover whether firms may have reacted to the increased uncertainty brought
bythepandemicbyreducingmarkupsinordertomaintainorexpandtheircustomerbase. Finally,
the pandemic-induced lockdowns may have created a reallocation within firms’ input use, such as
the rise of remote workers (Bloom et al., 2015; Bloom, 2020), which could have decreased selling,
general, and administrative expenses, while increasing the value of the costs of goods sold.
79
References
Afrouzi, H., Drenik, A., and Kim, R. (2020). Growing by the masses-revisiting the link between
firm size and market power.
Ag´ enor, P.-R. (2016). Optimal fiscal management of commodity price shocks. Journal of Develop-
ment Economics, 122:183–196.
Altig,D.,Baker,S.,Barrero,J.M.,Bloom,N.,Bunn,P.,Chen,S.,Davis,S.J.,Leather,J.,Meyer,
B., Mihaylov, E., et al. (2020). Economic uncertainty before and during the covid-19 pandemic.
Journal of Public Economics, 191:104274.
Angrist, J. D. and Pischke, J.-S. (2008). Mostly Harmless Econometrics: An Empiricist’s Com-
panion. Princeton University Press.
Argente, D., Baslandze, S., Hanley, D., and Moreira, S. (2020). Patents to products: Product
innovation and firm dynamics.
Argente, D., Lee, M., and Moreira, S. (2018). Innovation and product reallocation in the great
recession. Journal of Monetary Economics, 93:1–20.
Argente,D.,Lee,M.,andMoreira,S.(2019). Thelifecycleofproducts: Evidenceandimplications.
Available at SSRN 3163195.
Auclert, A., Rognlie, M., Souchier, M., andStraub, L.(2021). Exchangeratesandmonetarypolicy
withheterogeneousagents: Sizinguptherealincomechannel. Technicalreport,NationalBureau
of Economic Research.
Auerbach, A. J., Gorodnichenko, Y., and Murphy, D. (2021). Inequality, fiscal policy and covid19
restrictions in a demand-determined economy. European Economic Review, 137:103810.
Autor, D., Dorn, D., Katz, L. F., Patterson, C., and Van Reenen, J. (2020). The fall of the labor
share and the rise of superstar firms. The Quarterly Journal of Economics, 135(2):645–709.
Ball, L. M., Leigh, D., and Mishra, P. (2022). Understanding us inflation during the covid era.
Technical report, National Bureau of Economic Research.
Baqaee,D.andFarhi,E.(2020). Nonlinearproductionnetworkswithanapplicationtothecovid-19
crisis. Technical report, National Bureau of Economic Research.
Bermingham, C. and D’Agostino, A. (2014). Understanding and forecasting aggregate and disag-
gregate price dynamics. Empirical Economics, 46(2):765–788.
80
Berry, S., Gaynor, M., and Scott Morton, F. (2019). Do increasing markups matter? lessons from
empirical industrial organization. Journal of Economic Perspectives, 33(3):44–68.
Berry, S., Levinsohn, J., and Pakes, A. (1995). Automobile prices in market equilibrium. Econo-
metrica: Journal of the Econometric Society, pages 841–890.
Bigio, S., Zhang, M., and Zilberman, E. (2020). Transfers vs credit policy: Macroeconomic policy
trade-offs during covid-19. Working Paper 27118, National Bureau of Economic Research.
Binder, C. and Kamdar, R. (2022). Expected and realized inflation in historical perspective.
Journal of Economic Perspectives, 36(3):131–156. Date revised - 2022-07-21; Availability -
URL:http://www.aeaweb.org/jep/] Publisher’s URL; Last updated - 2022-10-08.
Bloom, N. (2020). How working from home works out. Institute for Economic Policy Research
(SIEPR). Policy Brief June, pages 1–9.
Bloom, N., Liang, J., Roberts, J., and Ying, Z. J. (2015). Does working from home work? evidence
from a chinese experiment. The Quarterly Journal of Economics, 130(1):165–218.
Bornstein, G. (2018). Entry and profits in an aging economy: The role of consumer inertia.
Technical report, mimeo.
Bornstein, G., Krusell, P., and Rebelo, S. (2017). Lags, costs, and shocks: An equilibrium model
of the oil industry. Technical report, National Bureau of Economic Research.
Broda, C. and Weinstein, D. E. (2010). Product creation and destruction: Evidence and price
implications. American Economic Review, 100(3):691–723.
Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., and Scott, S. L. (2015). Inferring causal
impact using bayesian structural time-series models. The Annals of Applied Statistics, 9(1):247–
274.
Brodeur, A., Gray, D., Islam, A., and Bhuiyan, S. (2021). A literature review of the economics of
covid-19. Journal of Economic Surveys, 35(4):1007–1044.
Burstein, A., Carvalho, V. M., and Grassi, B. (2020). Bottom-up markup fluctuations. Technical
report, National Bureau of Economic Research.
Caselli, F. and Michaels, G. (2013). Do oil windfalls improve living standards? evidence from
brazil. American Economic Journal: Applied Economics, 5(1):208–38.
Cavallo, A. (2020). Inflation with covid consumption baskets. Technical report, National Bureau
81
of Economic Research.
Chandy, R. K. and Tellis, G. J. (2000). The incumbent’s curse? incumbency, size, and radical
product innovation. Journal of marketing, 64(3):1–17.
Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: Firm-level
evidence from the 2008–9 financial crisis. The Quarterly Journal of Economics, 129(1):1–59.
Chudik, A., Mohaddes, K., Pesaran, M. H., Raissi, M., and Rebucci, A. (2021). A counterfactual
economic analysis of covid-19 using a threshold augmented multi-country model. Journal of
International Money and Finance, 119:102477.
Clementi, G. L. and Palazzo, B. (2016). Entry, exit, firm dynamics, and aggregate fluctuations.
American Economic Journal: Macroeconomics, 8(3):1–41.
Cloyne, J., Ferreira, C., Froemel, M., and Surico, P. (2018). Monetary policy, corporate finance
and investment. Technical report, National Bureau of Economic Research.
Contessi,S.,DePace,P.,andGuidolin,M.(2014). Howdidthefinancialcrisisalterthecorrelations
of us yield spreads? Journal of Empirical Finance, 28:362–385.
Cravino, J. and Levchenko, A. A. (2017). The distributional consequences of large devaluations.
American Economic Review, 107(11):3477–3509.
Cugat, G. (2019). Emerging markets, household heterogeneity, and exchange rate policy. In 2019
Meeting Papers, number 526. Society for Economic Dynamics.
Davis, S. J., Haltiwanger, J., Jarmin, R., Miranda, J., Foote, C., and Nagypal, E. (2006). Volatil-
ity and dispersion in business growth rates: Publicly traded versus privately held firms [with
comments and discussion]. NBER macroeconomics annual, 21:107–179.
De Loecker, J., Eeckhout, J., and Unger, G. (2020). The rise of market power and the macroeco-
nomic implications. The Quarterly Journal of Economics, 135(2):561–644.
DeLoecker,J.andWarzynski,F.(2012). Markupsandfirm-levelexportstatus. Americaneconomic
review, 102(6):2437–71.
Dietrich, A. M., Kuester, K., Muller, G. J., and Schoenle, R. (2022). News and
uncertainty about covid-19: Survey evidence and short-run economic impact. Jour-
nal of Monetary Economics, 129:S35–51. Date revised - 2022-09-08; Availability -
URL:http://www.sciencedirect.com/science/journal/03043932] Publisher’s URL; Last updated
- 2022-10-08.
82
Dippel, C., Gold, R., Heblich, S., and Pinto, R. (2019). Mediation analysis in iv settings with a
single instrument. Technical report, Mimeo.
Edelstein,P.andKilian,L.(2009). Howsensitiveareconsumerexpenditurestoretailenergyprices?
Journal of Monetary Economics, 56(6):766–779.
Faber, B. and Fally, T. (2017). Firm heterogeneity in consumption baskets: Evidence from home
and store scanner data. Technical report, National Bureau of Economic Research.
Fern´ andez, A., Gonz´ alez, A., and Rodriguez, D. (2018). Sharing a ride on the commodities roller
coaster: Common factors in business cycles of emerging economies. Journal of International
Economics, 111:99–121.
Fern´ andez, A., Schmitt-Groh´ e, S., and Uribe, M. (2017). World shocks, world prices, and business
cycles: An empirical investigation. Journal of International Economics, 108:S2–S14.
Gilchrist,S.,Schoenle,R.,Sim,J.,andZakrajˇ sek,E.(2017). Inflationdynamicsduringthefinancial
crisis. American Economic Review, 107(3):785–823.
Gilchrist, S. and Zakrajˇ sek, E. (2012). Credit spreads and business cycle fluctuations. American
economic review, 102(4):1692–1720.
Gourio, F. and Rudanko, L. (2014). Customer capital. Review of Economic Studies, 81(3):1102–
1136.
Griffin,A.(1993). Metricsformeasuringproductdevelopmentcycletime. Journal of Product Inno-
vation Management: An International Publication of the Product Development & Management
Association, 10(2):112–125.
Guerrieri, V., Lorenzoni, G., Straub, L., and Werning, I. (2022). Macroeconomic implications of
covid-19: Can negative supply shocks cause demand shortages? American Economic Review,
112(5):1437–74.
Guo, X., Ottonello, P., and Perez, D. J. (2020). Monetary policy and redistribution in open
economies. Technical report, National Bureau of Economic Research.
Hall, R.E.(1988). Therelationbetweenpriceandmarginalcostinusindustry. Journal of political
Economy, 96(5):921–947.
Hall, R. E. (2018). Using empirical marginal cost to measure market power in the us economy.
Technical report, National Bureau of Economic Research.
83
Harvey, A. C. (1990). Forecasting, structural time series models and the kalman filter.
Heresi, R. (2019). Efficient Reallocation and Productivity during Commodity Price Cycles . PhD
thesis.
Hong, S. (2017a). Customer capital, markup cyclicality, and amplification. Markup Cyclicality,
and Amplification (2017-04-30) .
Hong, S. (2017b). Customer capital, markup cyclicality, and amplification. Technical report, FRB
St Louis Working Paper,(2017-33).
Hong, S. (2020). Emerging market business cycles with heterogeneous agents. Technical report,
Tech. rep.
Hopenhayn, H. A. (1992). Entry, exit, and firm dynamics in long run equilibrium. Econometrica:
Journal of the Econometric Society, pages 1127–1150.
Hottman, C. J., Redding, S. J., and Weinstein, D. E. (2016). Quantifying the sources of firm
heterogeneity. The Quarterly Journal of Economics, 131(3):1291–1364.
Jaimovich, N., Rebelo, S., Wong, A., and Zhang, M. B. (2020). Trading up and the skill premium.
NBER Macroeconomics Annual, 34(1):285–316.
Jord` a,
` O. (2005). Estimation and inference of impulse responses by local projections. American
economic review, 95(1):161–182.
Jord` a,
` O.,Singh,S.R.,andTaylor,A.M.(2022). Longer-runeconomicconsequencesofpandemics.
Review of Economics and Statistics, 104(1):166–175.
Kaplan, G., Moll, B., and Violante, G. L. (2018). Monetary policy according to hank. American
Economic Review, 108(3):697–743.
Kaplan,G.,Violante,G.L.,andWeidner,J.(2014). Thewealthyhand-to-mouth. Technicalreport,
National Bureau of Economic Research.
Kim, R. (2021). The effect of the credit crunch on output price dynamics: The corporate inventory
and liquidity management channel. The Quarterly Journal of Economics, 136(1):563–619.
Ludvigson, S. C., Ma, S., and Ng, S. (2020). Covid-19 and the macroeconomic effects of costly
disasters. Technical report, National Bureau of Economic Research.
84
MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature,
35(1):13–39.
Martinez, L. R. (2019). Sources of revenue and government performance: evidence from colombia.
Available at SSRN 3273001.
McKibbin, W. and Fernando, R. (2021). The global macroeconomic impacts of covid-19: Seven
scenarios. Asian Economic Papers, 20(2):1–30.
Messina, J. and Silva, J. (2019). Twenty years of wage inequality in Latin America. The World
Bank.
Michelacci, C., Paciello, L., and Pozzi, A. (2019). The extensive margin of aggregate consumption
demand.
Minniti, A. and Turino, F. (2013). Multi-product firms and business cycle dynamics. European
Economic Review, 57:75–97.
Miranda-Agrippino, S. and Rey, H. (2020). Us monetary policy and the global financial cycle. The
Review of Economic Studies, 87(6):2754–2776.
MontielOlea, J.L.andPlagborg-Møller, M.(2021). Localprojectioninferenceissimplerandmore
robust than you think. Econometrica, 89(4):1789–1823.
Motyovszki, G. (2020). Monetary-fiscal interactions and redistribution in small open economies.
Technical report, EUI Working Papers 2020/03, ECO.
Neiman, B. and Vavra, J. S. (2019). The rise of niche consumption. Technical report, National
Bureau of Economic Research.
Nekarda, C. J. and Ramey, V. A. (2020). The cyclical behavior of the price-cost markup. Journal
of Money, Credit and Banking, 52(S2):319–353.
Ottonello, P. and Winberry, T. (2020). Financial heterogeneity and the investment channel of
monetary policy. Econometrica, 88(6):2473–2502.
Pesaran, M. H., Pick, A., and Timmermann, A. (2022). Forecasting with panel data: estimation
uncertainty versus parameter heterogeneity.
Philippon, T. (2019). The Great Reversal: How America Gave Up on Free Markets. Harvard
University Press.
85
Pieschac´ on,A.(2012). Thevalueoffiscaldisciplineforoil-exportingcountries. Journal of Monetary
Economics, 59(3):250–268.
Stock, J. H. and Watson, M. W. (1999). Forecasting inflation. Journal of Monetary Economics,
44(2):293–335.
Storey, J.D.andTibshirani, R.(2003). Statisticalsignificanceforgenomewidestudies. Proceedings
of the National Academy of Sciences, 100(16):9440–9445.
Unda, M. (2019). Finanzas municipales en m´ exico por qu´ e unos municipios recaudan m´ as y gastan
mejor. Working Paper, Lincoln Institute of Land and Policy, WP19MU1SP.
Woodford, M. (2022). Effective demand failures and the limits of monetary stabilization pol-
icy. American Economic Review, 112(5):1475–1521. Date revised - 2022-04-07; Availability -
URL:http://www.aeaweb.org/aer/] Publisher’s URL; Last updated - 2022-04-11.
Yang, S. (2022). The distribution of innovations across firms. Technical report, USC.
86
A Bias in aggregate Entry/Exit rates.
This appendix proposes a simple falsification test: it compares aggregate entry (exit) rates under
different scenarios: using the entire panel (2004-2020), if the sample started (ended) one quarter
later (earlier), and so on. The results of this exercise are shown in Figure A.29.
Figure A.29: Average entry and exit rates on each end of the sample.
(a) Entry rates.
.1 .15 .2 .25 .3 .35
2004q3 2005q1 2005q3 2006q1 2006q3
date
0 1 2 3
4 5 6
(b) Exit rates.
.15 .2 .25 .3 .35 .4 2018q1 2018q3 2019q1 2019q3 2020q1 2020q3
date
0 1 2 3
4 5 6
Note: this figure plots average entry and exit rates, computed as ER
ft
=
E
ft
N
ft
and XR
ft
=
X
ft
N
ft
, where E
ft
is the
number of products that were introduced by the firm at time t, X
ft
is the products that exited and N
ft
is the total
products offered by the firm f. I compute entry and exit rates per firm and aggregate using simple averages. The
numbers track the number of quarters used for the moving window: for example, the number zero covers the period
from 2003q4-2021q1 (full sample). The number one covers the period 2004q1-2020q4, the number 2 covers the period
2004q2-2020q3 and so on.
The bias, which is driven by false entries and exits, is computed by the difference between the
ratescomputedatwiththeentiresamplerelativetothecomputedratewithintheselectedwindow.
As confirmed by the figure, the rates seem to equalize within 2 years, with most of it disappearing
withinfour quarters. As such, I choose to remove two years on each end of the sample to guarantee
an accurately measure of both entry and exit rates.
Also, I compute the aggregate rates for liquidity constrained and unconstrained firms using
industry medians as opposed to aggregate medians. The result of this is displayed in Figure A.30
87
andconfirmsthesameresultsfromusingtheaggregatemediansasshowninFigure1.2inthemain
text.
Figure A.30: Average entry and exit rates versus constraints, industry medians.
(a) Entry Rates
.06 .07 .08 .09 .1 .11 .12 .13 .14 .15 .16 .17 .18
(%)
2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1
Non−Constrained Liquidity Constrained
(b) Exit Rates
.06 .07 .08 .09 .1 .11 .12 .13 .14 .15 .16 .17 .18
(%)
2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1
Non−Constrained Liquidity Constrained
Note: To avoid biases, I exclude the first and last two years of the sample. Entry and exit rates are computed as
ER
ft
=
E
ft
N
ft
and XR
ft
=
X
ft
N
ft
, where E
ft
is the number of products that were introduced by the firm at time t, X
ft
is the products that exited and N
ft
is the total products offered by the firm f. I compute entry and exit rates per
firm and aggregate using simple averages. I define a Firm as Liquidity constrained if its trailing 12 month average of
the Liquidity ratio is under the NAICS 2 digit industry median at time t.
88
B Alternative specifications for Entry/Exit regressions.
Thissectiondocumentstherobustnessofthefindingsonthemaintest. Iconductmultipleanalyses:
varying the specification of fixed effects, controls, using other measures of financial constraints and
I also explore different definitions of crises, also using different measures of aggregate financial
distress.
B.1 Robustness to specification.
I explore different specifications to verify the results. I include firm (quarterly) fixed effects, re-
move/add controls and use different measures of financial constraints: for Liquidity, I use the
quarter lag, as well as the trailing 12 month moving average of the Liquidity ratio used to plot the
aggregate figures; for the leverage ratio, I use its quarterly lag.
The results for the baseline variables are presented on Table B.16, B.17 and B.18, whereas the
results for the alternative variables are presented on Table B.19, B.20 and B.21. For the liquidity
ratio,theinsightremainsthesameacrossexercises: higherliquidityleadstomoreproductinsertion
duringfinancialcrises. Similarly,fortheLeverageratio,itseemsthatthesemarginsarenotadjusted
during crises and that the age co-moves accordingly regardless of the times.
89
Table B.16: Alternative specifications for Liq
ft− 4
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Friction x Normal 0.012*** 0.008* 0.004 -0.001 -0.000 -0.001 0.012*** 0.009* 0.003 -0.001 0.000 -0.002
(0.004) (0.005) (0.005) (0.004) (0.004) (0.005) (0.004) (0.005) (0.005) (0.004) (0.004) (0.005)
Friction x Crises 0.019** 0.012 0.007 0.012** -0.009 0.021** 0.018** 0.012 0.007 0.012** -0.008 0.020**
(0.008) (0.008) (0.010) (0.006) (0.008) (0.009) (0.008) (0.008) (0.010) (0.006) (0.008) (0.009)
∆ log(Sales
ft
) 0.008 -0.007 0.015 0.036** -0.016 0.051**
(0.021) (0.019) (0.023) (0.017) (0.015) (0.023)
∆ log(Cogs
ft
) 0.006 -0.007 0.012 -0.019 0.002 -0.021
(0.020) (0.018) (0.023) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.001 -0.017 0.016 0.000 -0.014* 0.014
(0.011) (0.011) (0.015) (0.010) (0.008) (0.015)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 14,624 14,624 14,624 14,493 14,493 14,493 14,624 14,624 14,624 14,493 14,493 14,493
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (10)-(12) depict the
results on the main document.
90
Table B.17: Alternative specifications for Lev
ft− 4
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Friction x Normal -0.002 0.002 -0.004 -0.003 0.015*** -0.017** -0.002 0.001 -0.003 -0.003 0.014*** -0.017**
(0.004) (0.004) (0.003) (0.005) (0.006) (0.007) (0.004) (0.004) (0.003) (0.005) (0.006) (0.007)
Friction x Crises -0.004 -0.002 -0.003 0.003 0.010 -0.007 -0.004 -0.002 -0.002 0.003 0.010 -0.006
(0.007) (0.008) (0.008) (0.006) (0.006) (0.008) (0.007) (0.008) (0.008) (0.006) (0.006) (0.008)
∆ log(Sales
ft
) 0.009 -0.006 0.015 0.037** -0.019 0.056**
(0.021) (0.019) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) 0.008 -0.004 0.013 -0.019 0.005 -0.024
(0.020) (0.018) (0.023) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.000 -0.016 0.016 -0.001 -0.012 0.011
(0.011) (0.011) (0.015) (0.011) (0.008) (0.015)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 14,624 14,624 14,624 14,493 14,493 14,493 14,624 14,624 14,624 14,493 14,493 14,493
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (10)-(12) depict the
results on the main document.
91
Table B.18: Alternative specifications for Age
ft
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Age x Normal -0.013*** -0.006 -0.007** -0.057*** 0.082*** -0.139*** -0.013*** -0.005 -0.008** -0.057*** 0.081*** -0.137***
(0.004) (0.004) (0.004) (0.006) (0.008) (0.010) (0.004) (0.004) (0.004) (0.007) (0.008) (0.010)
Age x Crises -0.009 -0.009 -0.000 -0.061*** 0.090*** -0.150*** -0.008 -0.003 -0.005 -0.059*** 0.088*** -0.147***
(0.007) (0.008) (0.009) (0.007) (0.009) (0.012) (0.007) (0.007) (0.009) (0.007) (0.009) (0.012)
∆ log(Sales
ft
) 0.002 -0.009 0.010 0.028 -0.003 0.031
(0.021) (0.018) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) 0.009 -0.004 0.013 -0.016 -0.003 -0.013
(0.020) (0.018) (0.023) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.002 -0.017 0.015 -0.003 -0.011 0.008
(0.011) (0.011) (0.015) (0.010) (0.008) (0.015)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 15,285 15,285 15,285 15,159 15,159 15,159 14,624 14,624 14,624 14,493 14,493 14,493
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. age values scaled to have standard deviation 1. Columns (10)-(12) depict the results on the
main document.
92
Table B.19: Alternative specifications for Liq
ft− 1
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Friction x Normal 0.010** 0.008 0.002 0.001 -0.004 0.005 0.010** 0.008 0.002 0.001 -0.006 0.007
(0.004) (0.006) (0.006) (0.004) (0.004) (0.006) (0.005) (0.006) (0.006) (0.004) (0.004) (0.006)
Friction x Crises 0.014* 0.027** -0.013 0.011* -0.010 0.021** 0.015* 0.021* -0.006 0.012** -0.015 0.027**
(0.008) (0.011) (0.010) (0.006) (0.009) (0.011) (0.009) (0.011) (0.009) (0.006) (0.010) (0.011)
∆ log(Sales
ft
) 0.002 -0.031 0.033 0.025 -0.018 0.043
(0.025) (0.026) (0.026) (0.017) (0.025) (0.028)
∆ log(Cogs
ft
) 0.018 0.021 -0.003 -0.010 0.013 -0.023
(0.025) (0.026) (0.029) (0.017) (0.025) (0.030)
∆ log(Sg&a
ft
) 0.002 -0.024* 0.026 0.007 -0.026** 0.033*
(0.014) (0.012) (0.017) (0.012) (0.010) (0.017)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 12,038 12,038 12,038 11,973 11,973 11,973 11,591 11,591 11,591 11,521 11,521 11,521
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (10)-(12) depict the
results on the main document.
93
Table B.20: Alternative specifications for trailing 12 month moving average of Liq
ft
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Friction x Normal 0.011*** 0.009* 0.003 -0.001 0.005 -0.006 0.012*** 0.009* 0.003 -0.001 0.002 -0.003
(0.004) (0.005) (0.005) (0.004) (0.004) (0.006) (0.004) (0.005) (0.005) (0.004) (0.004) (0.006)
Friction x Crises 0.020** 0.023*** -0.004 0.011* -0.001 0.011 0.019** 0.016** 0.002 0.011* -0.006 0.018*
(0.008) (0.008) (0.010) (0.006) (0.008) (0.010) (0.009) (0.008) (0.010) (0.007) (0.008) (0.010)
∆ log(Sales
ft
) 0.008 -0.007 0.015 0.036** -0.015 0.051**
(0.021) (0.019) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) 0.007 -0.006 0.013 -0.019 0.001 -0.020
(0.020) (0.018) (0.023) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.001 -0.017 0.016 0.000 -0.014* 0.014
(0.011) (0.011) (0.015) (0.010) (0.008) (0.015)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 15,153 15,153 15,153 15,022 15,022 15,022 14,624 14,624 14,624 14,493 14,493 14,493
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (10)-(12) depict the
results for the specification on the main document.
94
Table B.21: Alternative specifications for Lev
ft− 1
.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Eft
Nft
Xft
Nft
Eft− Xft
Nft
Friction x Normal -0.001 -0.001 -0.001 -0.003 0.014** -0.017** -0.002 -0.001 -0.000 -0.004 0.017** -0.021**
(0.004) (0.005) (0.004) (0.005) (0.007) (0.008) (0.004) (0.005) (0.004) (0.005) (0.008) (0.009)
Friction x Crises -0.000 -0.004 0.004 0.003 0.010 -0.007 -0.000 -0.002 0.002 0.002 0.013 -0.011
(0.008) (0.009) (0.010) (0.006) (0.008) (0.010) (0.009) (0.010) (0.011) (0.006) (0.009) (0.011)
∆ log(Sales
ft
) 0.004 -0.028 0.032 0.026 -0.020 0.046
(0.024) (0.025) (0.026) (0.017) (0.026) (0.029)
∆ log(Cogs
ft
) 0.017 0.020 -0.003 -0.010 0.013 -0.024
(0.025) (0.025) (0.029) (0.017) (0.025) (0.030)
∆ log(Sg&a
ft
) 0.003 -0.023* 0.026 0.006 -0.025** 0.031*
(0.014) (0.012) (0.017) (0.012) (0.010) (0.017)
Firm FE No No No Yes Yes Yes No No No Yes Yes Yes
Date FE Yes Yes Yes No No No Yes Yes Yes No No No
Observations 12,038 12,038 12,038 11,973 11,973 11,973 11,591 11,591 11,591 11,521 11,521 11,521
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (10)-(12) depict the
results on the main document.
B.2 Robustness to crisis definition and aggregate financial distress.
Although there is consensus that the financial crisis happened roughly at the end of the decade
of the 2000s, the exact dates are subject of debate: some rely on defining the crisis depending on
the timing of contraction of output, like NBER. Others focus instead on the period of observed
financial turmoil was observed (Gilchrist and Zakrajˇ sek, 2012; Chodorow-Reich, 2014).
As such, I conduct two different exercises to verify differences in the timing of the crisis: firstly,
I use the crisis dummies identified by three different sources: NBER, the dates of turmoil identified
by Contessi et al. (2014) and the observed financial turmoil originating from syndicated banks
Chodorow-Reich (2014). The other exercise consist on testing the relationship between firm finan-
cialconstraintsandanaggregatemeasureoffinancialconstraints: thecreditspreadsfromGilchrist
95
and Zakrajˇ sek (2012) (GZ) and the Moody’s Seasoned Baa Corporate Bond Yield. The former
is composed of a component capturing firm default idiosyncratic risk and a residual component
which measures investor attitudes toward corporate credit risk. The other measure captures the
yield on corporate bonds that are rated Baa, a rating that is relatively low risk, and is considered
investment grade, but it is only one grade above a junk bond rating.
The results of the first exercise are reported on Table B.22, B.23 and B.24. Interestingly, the
results for the liquidity and leverage ratios remain the same across the first specifications, but the
results using the dummy for the syndicated bank turmoil shows that firms decreased their exit rate
viatheliquidityratio. Thiscouldbemotivatedbythetiming: theperiodshownincolumns(7)-(9)
is shorter and reflects a later period of the crisis. By then, It is possible that firms with higher
liquidity already introduced new products, and they used whatever remaining resources to keep
them in the market, which leaves the same net entry rate, but this latter not showing statistically
significant differences. Lastly, the results for age merely confirm that the life-cycle of the firm
affects these rates in the same fashion, regardless of the time period selected.
The second exercise relates to direct financial stress and the results are presented on Ta-
ble B.25, B.26 and B.27.Consistent with the findings above, the relationship for Liquidity and
entry rates is negative, but as financial stress increases, this relationship has a positive effect that,
although negative in the net effect, is rather small. This could be driven by the fact that this con-
tinuous measure captures periods of financial stress non-exclusive to the crisis. Also, the findings
for the leverage ratio and age give the same conclusion as before: firms with higher leverage ratios
seem to not adjust this margin during periods of higher financial stress, and the life-cycle of the
firm seems to be independent of short term fluctuations.
96
Table B.22: Relationship between Entry and Exit for Liq
ft− 4
, using different crisis measures.
Crisis measure NBER Contessi et al. (2014) Chodorow-Reich (2014)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Liq
ft− 4
× Normal -0.001 0.000 -0.002 -0.001 0.001 -0.002 -0.001 -0.000 -0.001
(0.004) (0.004) (0.005) (0.004) (0.004) (0.005) (0.004) (0.004) (0.005)
Liq
ft− 4
× Crises 0.012** -0.008 0.020** 0.013** -0.009 0.022** -0.001 -0.013* 0.012
(0.006) (0.008) (0.009) (0.006) (0.007) (0.009) (0.008) (0.008) (0.012)
∆ log(Sales
ft
) 0.036** -0.016 0.051** 0.036** -0.016 0.052** 0.035** -0.016 0.051**
(0.017) (0.015) (0.023) (0.017) (0.015) (0.023) (0.017) (0.015) (0.023)
∆ log(Cogs ft) -0.019 0.002 -0.021 -0.019 0.002 -0.021 -0.017 0.001 -0.018
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) 0.000 -0.014* 0.014 0.000 -0.014* 0.014 -0.001 -0.014 0.013
(0.010) (0.008) (0.015) (0.010) (0.008) (0.015) (0.011) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.32 0.10 0.28 0.32 0.10 0.28 0.32 0.10
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the
5% and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict
the results on the main document. Each crisis variable consist of different time coverage: NBER measures the
financial crisis from 2007 December through 2009 June, Contessi et al. (2014) identifies financial turmoil from the
third quarter of 2007 through the second quarter of 2009. Finally, Chodorow-Reich (2014) identifies turmoil in the
syndicated loan market during the third quarter of 2008 through the second quarter of 2009.
97
Table B.23: Relationship between Entry and Exit for Lev
ft− 4
, using different crisis measures.
Crisis measure NBER Contessi et al. (2014) Chodorow-Reich (2014)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Lev
ft− 4
× Normal -0.003 0.014*** -0.017** -0.003 0.014*** -0.018** -0.003 0.014** -0.017**
(0.005) (0.006) (0.007) (0.005) (0.006) (0.007) (0.005) (0.006) (0.007)
Lev
ft− 4
× Crises 0.003 0.010 -0.006 0.004 0.009 -0.005 0.002 0.009 -0.007
(0.006) (0.006) (0.008) (0.006) (0.006) (0.009) (0.008) (0.008) (0.012)
∆ log(Sales
ft
) 0.037** -0.019 0.056** 0.036** -0.019 0.056** 0.037** -0.019 0.056**
(0.017) (0.016) (0.023) (0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.019 0.005 -0.024 -0.019 0.005 -0.024 -0.018 0.005 -0.023
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.001 -0.012 0.011 -0.001 -0.012 0.011 -0.001 -0.012 0.011
(0.011) (0.008) (0.015) (0.011) (0.008) (0.015) (0.011) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.32 0.10 0.28 0.32 0.10 0.28 0.32 0.10
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the
5% and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict
the results on the main document. Each crisis variable consist of different time coverage: NBER measures the
financial crisis from 2007 December through 2009 June, Contessi et al. (2014) identifies financial turmoil from the
third quarter of 2007 through the second quarter of 2009. Finally, Chodorow-Reich (2014) identifies turmoil in the
syndicated loan market during the third quarter of 2008 through the second quarter of 2009.
Table B.24: Relationship between Entry and Exit for Age
ft
, using different crisis measures.
Crisis measure NBER Contessi et al. (2014) Chodorow-Reich (2014)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Age
ft
× Normal -0.057*** 0.081*** -0.137*** -0.057*** 0.082*** -0.139*** -0.055*** 0.075*** -0.130***
(0.007) (0.008) (0.010) (0.007) (0.008) (0.011) (0.006) (0.008) (0.010)
Age
ft
× Crises -0.059*** 0.088*** -0.147*** -0.059*** 0.090*** -0.149*** -0.059*** 0.080*** -0.139***
(0.007) (0.009) (0.012) (0.008) (0.009) (0.012) (0.008) (0.009) (0.012)
∆ log(Sales
ft
) 0.028 -0.003 0.031 0.028 -0.003 0.031 0.028 -0.005 0.032
(0.017) (0.016) (0.023) (0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.016 -0.003 -0.013 -0.016 -0.003 -0.013 -0.017 -0.000 -0.016
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.003 -0.011 0.008 -0.002 -0.010 0.008 -0.002 -0.012 0.010
(0.010) (0.008) (0.015) (0.010) (0.008) (0.015) (0.010) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.33 0.13 0.28 0.33 0.13 0.28 0.33 0.13
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the
5% and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict
the results on the main document. Each crisis variable consist of different time coverage: NBER measures the
financial crisis from 2007 December through 2009 June, Contessi et al. (2014) identifies financial turmoil from the
third quarter of 2007 through the second quarter of 2009. Finally, Chodorow-Reich (2014) identifies turmoil in the
syndicated loan market during the third quarter of 2008 through the second quarter of 2009.
98
Table B.25: Relationship between Entry and Exit for Liq
ft− 4
, using measures of financial distress.
Financial stress measure GZ spread Moodys Baa Yield spread
(1) (2) (3) (4) (5) (6)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Financial stress 0.001 -0.002 0.003 -0.001 -0.000 -0.000
(0.002) (0.002) (0.003) (0.002) (0.002) (0.003)
Liq
ft− 4
-0.010* 0.002 -0.012 -0.013* 0.000 -0.013
(0.006) (0.005) (0.008) (0.007) (0.007) (0.011)
Liq
ft− 4
× Financial stress 0.005** -0.001 0.006** 0.004** -0.000 0.004
(0.002) (0.002) (0.003) (0.002) (0.002) (0.003)
∆ log(Sales
ft
) 0.037** -0.017 0.054** 0.036** -0.015 0.051**
(0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.018 0.002 -0.020 -0.017 0.001 -0.018
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) 0.001 -0.015* 0.016 -0.000 -0.014 0.014
(0.010) (0.008) (0.015) (0.010) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.32 0.10 0.28 0.32 0.10
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict the
results using the credit spread from Gilchrist and Zakrajˇ sek (2012), whereas columns (4)-(6) perform the same
exercise using the Moody’s Baa yield spread.
99
Table B.26: Relationship between Entry and Exit for Lev
ft− 4
, using measures of financial distress.
Financial stress measure GZ spread Moodys Baa Yield spread
(1) (2) (3) (4) (5) (6)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Financial stress 0.005* -0.002 0.007** 0.003 0.000 0.003
(0.003) (0.003) (0.004) (0.003) (0.003) (0.004)
Lev
ft− 4
ratio -0.001 0.015** -0.016* -0.000 0.015* -0.015
(0.006) (0.007) (0.009) (0.008) (0.009) (0.012)
Lev
ft− 4
× Financial stress -0.001 -0.000 -0.000 -0.001 -0.000 -0.000
(0.002) (0.002) (0.003) (0.002) (0.002) (0.003)
∆ log(Sales
ft
) 0.038** -0.020 0.058** 0.037** -0.019 0.056**
(0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.019 0.005 -0.024 -0.018 0.004 -0.023
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) 0.001 -0.013 0.014 -0.001 -0.012 0.011
(0.011) (0.008) (0.015) (0.011) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.32 0.10 0.28 0.32 0.10
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict the
results using the credit spread from Gilchrist and Zakrajˇ sek (2012), whereas columns (4)-(6) perform the same
exercise using the Moody’s Baa yield spread.
100
Table B.27: Relationship between Entry and Exit for Age
ft
, using measures of financial distress.
Financial stress measure GZ spread Moodys Baa Yield spread
(1) (2) (3) (4) (5) (6)
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
E
ft
N
ft
X
ft
N
ft
E
ft
− X
ft
N
ft
Financial stress 0.001 0.001 -0.000 0.002 0.001 0.001
(0.004) (0.005) (0.006) (0.004) (0.004) (0.007)
Age
ft
-0.054*** 0.076*** -0.130*** -0.052*** 0.073*** -0.125***
(0.007) (0.009) (0.012) (0.008) (0.010) (0.013)
Age
ft
× Financial stress -0.001 0.002 -0.002 -0.001 0.001 -0.002
(0.002) (0.002) (0.003) (0.002) (0.002) (0.003)
∆ log(Sales
ft
) 0.028* -0.003 0.031 0.029* -0.004 0.033
(0.017) (0.016) (0.023) (0.017) (0.016) (0.023)
∆ log(Cogs
ft
) -0.016 -0.001 -0.015 -0.017 -0.000 -0.016
(0.016) (0.015) (0.022) (0.016) (0.015) (0.022)
∆ log(Sg&a
ft
) -0.002 -0.011 0.009 -0.002 -0.012 0.010
(0.010) (0.008) (0.015) (0.010) (0.008) (0.015)
Observations 14,493 14,493 14,493 14,493 14,493 14,493
Adj. R-squared 0.28 0.33 0.13 0.28 0.33 0.13
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Financial friction values scaled to have standard deviation 1. Columns (1)-(3) depict the
results using the credit spread from Gilchrist and Zakrajˇ sek (2012), whereas columns (4)-(6) perform the same
exercise using the Moody’s Baa yield spread.
101
C Alternative specifications for Sales regressions.
I explore different specifications to verify the results. Firstly, I use the alternative classification of
liquidity-constrained firms relative two their industry medians, to verify that the aggregate change
in the sales margin is the same as documented on the main text. The results of this exercise are
documented on Figure C.31.
Figure C.31: Average sale decomposition under financial constraints, industry medians.
(a) Non Constrained
−.25 −.2 −.15 −.1 −.05 0 .05 .1
(%) change
2007q1 2008q1 2009q1 2010q1 2011q1
date
Sales M Q P
(b) Liquidity Constrained
−.25 −.2 −.15 −.1 −.05 0 .05 .1
(%) change
2007q1 2008q1 2009q1 2010q1 2011q1
date
Sales M Q P
Note: Data is averaged across firms. I define a Firm as Liquidity constrained if its trailing 12 month average of the
Liquidity ratio is under the NAICS 2 digit industry median at time t, as in Gilchrist et al. (2017).
Inadditiontothese,Ialsotestthattherelationshipforbothvariablesremainsunchangedtothe
inclusion of controls, which control for short-term demand changes (Cogs) and fixed costs ( Sg&a).
The results are presented on Table C.28 and C.29 and confirm the findings from the main text.
102
Table C.28: Robustness of decomposition of change in Sales for Liq
ft− 4
.
No Controls Controls
(1) (2) (3) (4) (5) (6) (7) (8)
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
Friction x Normal 0.022 0.020 0.001 0.001 0.027** 0.024** 0.001 0.002
(0.015) (0.014) (0.001) (0.002) (0.012) (0.012) (0.001) (0.001)
Friction x Crises 0.089*** 0.066*** 0.004** 0.019*** 0.001 -0.007 -0.002 0.010***
(0.021) (0.018) (0.002) (0.003) (0.015) (0.014) (0.002) (0.003)
Quality
ift
0.084*** 0.011* -0.016*** 0.076***
(0.030) (0.006) (0.005) (0.027)
age
ift
-0.054*** -0.047*** -0.003*** -0.004***
(0.005) (0.005) (0.000) (0.001)
∆ log(emp
ft
) -0.005 0.009 0.001 -0.016*
(0.032) (0.025) (0.002) (0.008)
Product Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
R-squared 0.14 0.14 0.07 0.14 0.15 0.15 0.07 0.18
Seasonal Dummies No No No No Yes Yes Yes Yes
Observations 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%
and three at the 1%. Each column reports the results of a regression for each of the sales components as a function
of the covariates. Financial friction variables are transformed to have a standard deviation of 1 for comparison.
Quarter dummies are included to control for seasonality.
103
Table C.29: Robustness of decomposition of change in Sales for Lev
ft− 4
.
No Controls Controls
(1) (2) (3) (4) (5) (6) (7) (8)
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
∆ Sales
it
∆ M
it
∆ q
it
∆ p
it
Friction x Normal -0.088** -0.076** -0.008** -0.004* -0.020 -0.017 -0.004 0.000
(0.042) (0.037) (0.003) (0.002) (0.029) (0.027) (0.002) (0.002)
Friction x Crises -0.038 -0.046 -0.004 0.012*** -0.020 -0.029 -0.003 0.012***
(0.041) (0.037) (0.003) (0.003) (0.026) (0.025) (0.002) (0.002)
Quality
ift
0.084*** 0.011* -0.016*** 0.076***
(0.030) (0.006) (0.005) (0.027)
age
ift
-0.052*** -0.046*** -0.003*** -0.004***
(0.006) (0.005) (0.000) (0.000)
∆ log(emp
ft
) -0.002 0.009 0.001 -0.012
(0.040) (0.033) (0.002) (0.008)
Product FE Yes Yes Yes Yes Yes Yes Yes Yes
R-squared 0.14 0.14 0.07 0.14 0.15 0.15 0.07 0.18
Seasonal Dummies No No No No Yes Yes Yes Yes
Observations 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807 2,331,807
Notes: Clustered standard errors at the firm level, one star denotes significance at the 10% level, two at the 5%,
and three at the 1%. Each column reports the results of a regression for each of the sales components as a function
of the covariates. Financial friction variables are transformed to have a standard deviation of 1 for comparison.
Quarter dummies are included to control for seasonality.
D Exogeneity of Oil Prices.
The analysis in this paper hinges on the exogeneity assumption of oil prices, which is based on
the notion that prices are determined by supply and demand in international oil markets. Studies
such as Pieschac´ on (2012); Fern´ andez et al. (2017); Heresi (2019) rely on this assumption as well.
Although it is difficult to show this in practice, this paper conducts a high-frequency event analysis
bygatheringdataonnewlydocumentedepisodesassociatedwithaccidentsanddisastersofPEMEX
to test whether these disturbances have an effect on oil prices within a one-week period of analysis
104
through a supply disruption. The following regression is estimated, using daily data:
∆ log(P
o
d
)=
3
X
s=− 3
γ s
Acc
d+s
+u
d
where Acc
d
takes a value of 1 if there was an accident on the day d and ∆ log(P
o
d
) is the oil price’s
daily growth rate and the sample period covers the years 2000 through 2018. In addition, we
perform a Portmanteau test that suggests that oil price growth rates follow a white noise process.
The results are presented in Table D.30. As can be seen, none of the events produce a significant
change in the oil price within the weekly window. In addition, we cannot reject the hypothesis of
white noise at the 5 percent level for the Portmanteau test. This suggests that the Mexican oil
price is not affected by shifts in domestic supply. Table D.31 presents the events and their dates.
Table D.30: Oil Price event analysis.
Acc
d− 3
Acc
d− 2
Acc
d− 1
Acc
d
Acc
d+1
Acc
d+2
Acc
d+3
-0.007 -0.001 0.006 0.003 -0.000 0.002 -0.003
(0.112) (0.867) (0.126) (0.504) (0.942) (0.527) (0.465)
Pormanteau test
QStat 53.06
degrees of freedom 40
P-val 0.081
Observations 4199
Source: Author’s calculations based on news searched-based events.
Note: P-values in parentheses.
Table D.31: Event dates
1996-11-11 2003-11-17 2005-05-03 2007-10-23 2009-06-17 2012-09-18 2015-04-03 2017-03-17
1998-11-18 2003-12-18 2005-07-08 2007-10-26 2010-09-07 2013-01-31 2015-08-11
2001-12-21 2004-12-22 2005-07-13 2008-11-17 2010-12-20 2013-08-21 2015-12-23
2002-01-04 2004-12-29 2006-10-17 2008-12-10 2011-04-12 2013-10-01 2016-02-08
2002-05-01 2005-01-26 2007-07-05 2009-01-20 2011-08-01 2014-08-08 2016-04-20
2003-06-05 2005-04-15 2007-09-10 2009-06-05 2011-12-19 2014-08-13 2017-01-12
Source: Author’s calculations based on news searched-based events.
105
E Additional Tables and Figures.
ThefiguresinthisAppendixshowthecoefficientdisaggregationfortheresultsinSection2.3. Each
panel displays the effect of a three-binned partition using the state-specific distribution of cash on
hand from 2000. For example, panel (c) of Figure E.32 shows the results for individuals between
the 25th and 75th percentiles of the cash-on-hand distribution in their corresponding state.
106
Figure E.32: Effects of oil price shocks to consumption expenditure, three binned Quartiles
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 4
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Between Quartile 1 and 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ consumption for each
quartile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using
the standard errors from the estimated Equation (7).
107
Figure E.33: Effects of oil price shocks to consumption expenditure, three binned Deciles.
−0.6
−0.4
−0.2
0.0
0.2
0.4
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Decile 1
−0.6
−0.4
−0.2
0.0
0.2
0.4
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Decile 10
−0.6
−0.4
−0.2
0.0
0.2
0.4
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Between Decile 1 and 9
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ consumption for each
decile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals using the
standard errors from the estimated Equation (7).
108
Figure E.34: Effects of oil price shocks to disposable income, three binned Quartiles
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 4
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0.20
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Between Quartile 1 and 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ disposable income
for each quartile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals
using the standard errors from the estimated Equation (7).
109
Figure E.35: Effects of oil price shocks to disposable income, three binned Deciles.
−0.20
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Decile 1
−0.20
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Decile 10
−0.20
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.15
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Between Decile 1 and 9
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ disposable income
for each decile of the wealth distribution. Shaded areas correspond to the 95%, 90%, and 80% confidence intervals
using the standard errors from the estimated Equation (7).
110
Figure E.36: Effects of oil price shocks to
E
C
by Quartiles
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(a) Quartile 1
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(b) Quartile 2
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(c) Quartile 3
−0.050
−0.025
0.000
0.025
0.050
0.075
0.100
0 1 2 3 4
Coefficient 95% CI 90% CI 80% CI
(d) Quartile 4
Source: Author’s calculations based on the final version of Mexico’s Household Income and Expenditure Survey
dataset.
Note: The figure shows the effect of an average 1% increase in the oil price for the households’ energy consumption
as a share of total consumption expenditure for each quartile of the wealth distribution. Shaded areas correspond to
the 95%, 90%, and 80% confidence intervals using the standard errors from the estimated Equation (7).
111
F Data and summary statistics
To calculate our measures of markup and profit rates, we remove all firms with negative sales, cost
of goods sold (COGS), and selling, general and administrative expenses (SG&A). We also remove
recordsbelowthefirstandabovethe99thyearlypercentilesoftheCOGSandSG&Atosalesratios
per year. We removed records without either Compustat firm identification numbers or industry
codes. Aftertheprevioussteps, wealsoremovevaluesbelowthefirstandabovethe99thpercentile
of two cost-shares (the ratio of COGS to the total of COGS and capital expenditure, and the ratio
of COGS to the total of COGS, capital expenditure, and SG&A). Before computing these ratios,
we deflate the respective series using the GDP deflator with 2010 as the base year.
In addition to these restrictions on the sample, we further impose the following conditions to
estimate counterfactuals reliably. First, we require firms with data for both the pre-pandemic and
the pandemic, so we remove 18,022 (17,003) firms that exited the panel before or in 2019, of which
79% left before 2010. Then, we eliminate 191 (431) firms that either entered the panel during
or after 2020 or had less than two years of available data before this year. Finally, to compute
counterfactuals for firms with at least 3 years of observations, we drop an additional 802 (925)
firms. After these filtering steps, our sample comprises 3,611 (3,192) publicly-traded firms with
yearly (quarterly) data, no extreme values in costs ratios, information for at least 5 years before
the pandemic and operating during it.
112
Table F.32: Summary Statistics
Variable Acronym Mean Median N. obs
Panel A: Sample 1955–2021
Sales Sale, PQ 2,125,657 159,894 269,899
Cost of goods sold COGS, V 1,414,280 94,946 269,899
Capital stock PPEGT,K 1,655,290 62,386 269,899
SG&A XSG&A,X 380,988 32,308 269,899
Wage bill CXLR,WL 1,088,658 117,408 32,982
Employment Emp,L 8,910 900 241,083
Panel B: Sample 1955–2016
Sales Sale, PQ 1,953,210 149,483 248,378
Cost of goods sold COGS, V 1,306,434 90,116 248,378
Capital stock PPEGT,K 1,474,565 58,178 248,378
SG&A XSG&A,X 349,539 30,043 248,378
Wage bill CXLR,WL 1,106,980 131,148 28,321
Employment Emp,L 8,426 866 221,685
Panel C: Sample 1955-2016 from De Loecker et al. (2020)
Sales Sale, PQ 1,922,074 147,806 247,644
Cost of goods sold COGS, V 1,016,550 55,384 247,644
Capital Stock PPEGT,K 1,454,210 57,532 247,644
SG&A XSG&A,X 342,805 29,682 247,644
Wage bill CXLR,WL 1,093,406 130,486 28,116
Employment Emp,L 8,363 863 221,121
Notes: ThousandsyearlyUS$deflatedusingtheGDPdeflatorwithbaseyear
2010. In column 2 of the table, we also report the Compustat acronym to
keep track of the specific variables with respect to that dataset.
113
G Counterfactual analysis using cost-of-goods weighting
In this section, we analyze the impact of our choice of weights on the results for the aggregate
markupandprofitrates. InFigures3.22and3.23weusesalesweights,thusgivingmoreimportance
to companies with higher sales. Alternatively, one can use costs of good sold (COGS) as weights
instead of sales as presented in Figures G.37. Therefore, we give more importance to firms or
industries with high cost structure, such as aviation, hospitality, or oil and gas. The results for
markups remain similar to those using sales-weighting, i.e., the average firm has lower realized
markups than its counterfactual markups.
On the other hand, when weighting by COGS, the results for profit rate become mixed: firms
had negative effects in 2020 and exhibited a strong recovery in 2021. This is more likely related
to the weighting: high cost firms seem to have lost profits during 2020, whereas low cost firms
increased their profits (sales-weighting case).
114
Figure G.37: Comparison of Observed and Counterfactual Markup and Profit Rates, weighting by
Cost of Goods Sold (COGS)
1.35
1.40
1.45
1.50
1.55
1.60
1.65
1.70
1.75
1.80
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual (LP panel)
(a) Markup Rate
15
16
17
18
19
20
21
2010 2015 2020
95% CI 90% CI
80% CI 50% CI
Observed
Counterfactual (LP panel)
(b) Profit Rate
Note: Note: The figure shows the cogs-weighted aggregate mean of the yearly observed and counterfactual markups
(Panel(a))andprofitrates(Panel(b)). Thesolidlinesrepresenttheobservedvalues, whilethedottedlinesrepresent
the counterfactual values for 2020-2021, as calculated using firm-level Bayesian structural time series models. The
shaded areas represent the 95%, 90%, 80%, and 50% (equally-tailed) credible interval for each statistic, calculated
based on 5,000 posterior simulations.
G.1 Forecast performance for BSTS and LLPs under a balanced sample
ThissubsectionpresentsthemeasuresonTable3.11andTable3.15,butconsideringafullybalanced
panel. The results show that the presence of unbalanced panels does not change the relative
performance of each forecasting strategy.
H Quarterly Analysis
This section expands the aggregate counterfactual analysis in Figures 3.22 and 3.23, but using
quarterly data instead of yearly data. Figures H.38 and H.39 shows similar results for the markup
rate: markupswouldhavebeenhigherhadthepandemicnothappened. Particularly,fortheaverage
firm, the annual average observed markup rate was 4.5% and 11.6% lower than its counterfactual
115
Table G.33: Forecast Quality Statistics Across Models, Balanced Sample.
(a) Firm-Specific Bayesian Time Series Models
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 -0.015 0.790 0.361 0.210 0.200 0.105 2808
Markup rate 2019 -0.061 0.844 0.384 0.230 0.233 0.120 2693
Profit rate 2018 19.632 146.502 27.645 3.224 2.957 0.333 2843
Profit rate 2019 16.809 148.274 28.480 3.878 3.608 0.367 2718
(b) Firm-Specific Local-Linear Projection
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 -0.023 0.644 0.248 0.144 0.144 0.057 2808
Markup rate 2019 -0.086 0.767 0.337 0.203 0.206 0.089 2693
Profit rate 2018 22.345 488.279 29.279 3.414 2.280 0.201 2843
Profit rate 2019 16.425 206.869 27.893 3.798 3.032 0.309 2718
(c) Panel Local-Linear Projection
Variable Year ME RMSE MAE MAES MAPE Median
MAPE
n
Markup rate 2018 0.017 0.555 0.223 0.128 0.120 0.054 2808
Markup rate 2019 -0.030 0.721 0.294 0.176 0.172 0.078 2693
Profit rate 2018 5.809 96.882 16.257 1.896 1.590 0.318 2843
Profit rate 2019 2.679 73.403 18.088 2.463 2.282 0.373 2718
Notes: The table introduces common forecast quality measures computed for markup rates and profit rates for 2018–
2019 with firm-level models estimated using information up to 2017. The measures computed are: mean error (ME),
root mean square error (RMSE), mean absolute error (MAE), standardized mean absolute error (MAES), and the
mean absolute percentage error (MAPE). Median MAPE is the median of the MAPEs across firm-level MAPEs. n
is the number of firms and the sample is balanced to consider the same observations across exercises.
value in 2020 and 2021, respectively. Interestingly, in the case of profit rates, the positive effects of
the pandemic only seem to have started in the third quarter of 2020. More precisely, the average
firm had a profit rate of 16.9% and 21.6% in 2020 and 2021, respectively, which are 0.7 and 4.75
percentage points higher than the counterfactuals.
116
Figure H.38: Counterfactual Analysis of Average and Quartiles of Markup Rates
1.45
1.50
1.55
1.60
1.65
1.70
1.75
1.80
1.85
2010 2015 2020
Sales weighted average
Counterfactual
95% CI 90% CI
80% CI 50% CI
(a) Average
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2010 2015 2020
p25 p50 p75
(b) Quartiles
Note: The figure shows four aggregate statistics of the quarterly observed and counterfactual markup rates. The
statistics are the sales-weighted average (shown in Panel (a)), the first quartile (shown in Panel (b) in red), the
second quartile (shown in Panel (b) in black), and the third quartile (shown in Panel (b) in gray). The solid lines
represent the observed values, while the dotted lines represent the counterfactual values for 2020-2021, as calculated
using firm-level Bayesian structural time series models. The shaded areas represent the 95%, 90%, 80%, and 50%
(equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations on Panel (a). 95%
(equally-tailed) credible intervals are displayed for the statistics on Panel (b).
117
Figure H.39: Counterfactual Analysis of Average and Quartiles of Profit Rates
10
12
14
16
18
20
22
2010 2015 2020
Sales weighted average
Counterfactual
95% CI 90% CI
80% CI 50% CI
(a) Average
−5
0
5
10
15
20
25
2010 2015 2020
p25 p50 p75
(b) Quartiles
Note: The figure shows four aggregate statistics of the quarterly observed and counterfactual profit rates. The
statistics are the sales-weighted average (shown in Panel (a)), the first quartile (shown in Panel (b) in red), the
second quartile (shown in Panel (b) in black), and the third quartile (shown in Panel (b) in gray). The solid lines
represent the observed values, while the dotted lines represent the counterfactual values for 2020-2021, as calculated
using firm-level Bayesian structural time series models. The shaded areas represent the 95%, 90%, 80%, and 50%
(equally-tailed) credible interval for each statistic, calculated based on 5,000 posterior simulations on Panel (a). 95%
(equally-tailed) credible intervals are displayed for the statistics on Panel (b).
H.1 Quarterly heterogeneity analysis
Thissectionanalyzestheheterogeneouseffectsofthepandemicusingquarterlydata,whichprovides
insights into the unfolding of the pandemic in 2020 and 2021.
Effect heterogeneity regressions
Finally, we validate the results in Section 3.4.2 by averaging the quarterly effects into the annual
frequencyandrepeatingtheaccountingexercise,withtheresultsdisplayedonTableH.34andH.35.
The implied yearly effects backed by the high-frequency analysis seem to be more reactive to cogs
and sales, with very similar results for employment, stock exchange tenure, and market shares. On
the other hand, the results for profit rates exhibit very similar results for all of the variables. This
118
Table H.34: Heterogeneous Effects of the pandemic on Markup Rates (Quarterly Specification)
Dependent Variable: Average markup rate pandemic effect in 2020–2021
All 2020 2021
Model: (1) (2) (3) (4) (5) (6)
Variables
COGS -0.244
∗∗∗ -0.263
∗∗∗ -0.296
∗∗∗ -0.313
∗∗∗ -0.181
∗∗ -0.197
∗∗ (0.054) (0.057) (0.066) (0.073) (0.086) (0.088)
Sales 0.197
∗∗∗ 0.248
∗∗∗ 0.230
∗∗∗ 0.278
∗∗∗ 0.160
∗ 0.212
∗∗ (0.056) (0.058) (0.067) (0.072) (0.093) (0.090)
Employment 0.037
∗∗∗ 0.008 0.065
∗∗∗ 0.033
∗∗ 0.000 -0.028
(0.013) (0.013) (0.014) (0.015) (0.022) (0.023)
Stock-exchange tenure 0.002
∗∗∗ 0.002
∗∗ 0.003
∗∗∗ 0.002
∗ 0.001 0.001
(0.001) (0.001) (0.001) (0.001) (0.002) (0.002)
Market share 0.837
∗∗∗ 0.002 0.235 -0.012 1.481
∗∗ -0.047
(0.301) (0.308) (0.253) (0.287) (0.680) (0.566)
Fixed-effects
2-digit NAICS industry Yes Yes Yes
Mean 1.599 1.590 1.610
Mean effect -0.040 -0.045 -0.033
Fit statistics
Observations 5,017 5,017 2,752 2,752 2,265 2,265
R
2
0.021 0.075 0.044 0.095 0.010 0.082
Within R
2
0.018 0.034 0.009
Notes: The table presents cross-section firm-level regressions of the average pandemic effect at the firm-
level on the 2015–2019 average of the logarithm cost of goods sold, logarithm of sales, logarithm of
employment, years since publicly-traded, and market shares. NAICS = North American Industrial Clas-
sification System. Heteroskedasticity-robust standard errors in parentheses. Signif. Codes: ***: 0.01, **:
0.05, *: 0.1
suggests that the short-term effects on markups may be more reactive to cogs and sales, in a way
that when averaged out, they still matter with respect to those calculated using the yearly data.
This could be driven by the immediate changes in very short-term demand movements Gilchrist
et al. (2017).
119
Table H.35: Heterogeneous Effects of the pandemic on Profit Rates (Quarterly Specification)
Dependent Variable: Average profit rate pandemic effect in 2020–2021
All 2020 2021
Model: (1) (2) (3) (4) (5) (6)
Variables
COGS -15.889
∗∗∗ -14.229
∗∗∗ -21.252
∗∗∗ -15.754
∗∗∗ -9.525 -12.315
(4.186) (4.621) (3.705) (3.712) (8.018) (9.062)
Sales 16.075
∗∗∗ 15.365
∗∗∗ 20.067
∗∗∗ 15.685
∗∗∗ 11.476 15.221
∗ (4.605) (4.843) (5.232) (4.780) (7.842) (8.775)
Employment 1.140 -0.076 2.610 0.796 -0.795 -1.563
(1.871) (2.122) (2.826) (2.845) (2.296) (3.126)
Stock-exchange tenure 0.149 0.005 0.140 -0.081 0.143 0.086
(0.093) (0.086) (0.115) (0.107) (0.153) (0.140)
Market share -71.887
∗ -20.439 -100.702 9.398 -43.641 -59.176
(39.196) (44.425) (70.899) (67.204) (27.539) (38.484)
Fixed-effects
2-digit NAICS industry Yes Yes Yes
Mean 2.974 -0.747 7.546
Mean effect -1.070 -4.573 3.235
Fit statistics
Observations 5,017 5,017 2,752 2,752 2,265 2,265
R
2
0.009 0.027 0.015 0.067 0.004 0.009
Within R
2
0.005 0.006 0.006
Notes: The table presents cross-section firm-level regressions of the average pandemic effect at the firm-level
on 2015–2019 average of the logarithm cost of goods sold, the logarithm of sales, logarithm of employment,
years since publicly-traded, and market shares. NAICS = North American Industrial Classification System.
Heteroskedasticity-robust standard errors in parentheses. Signif. Codes: ***: 0.01, **: 0.05, *: 0.1
120
H.2 Tracking the fraction of firms with significant effects in the short-term.
To elucidate whether these effects on markup rates are statistically significant, we calculated the
fractionoffirmswithsignificanteffects, anditsdecompositionintothefractionoffirmswithsignif-
icant and positive effects, and significant and negative effects. Figure H.40a shows this calculation
for quarterly data in 2020–2021. We find that 19% of the firms had a significant effect on their
markups at the beginning of the pandemic, with most of those firms having a negative effect and
over 7% of the firms having a positive effect. In most of 2020–2021, the number of firms with
significant, negative effects is higher than the number of firms with significant, positive, effects.
Counting the number of firms with at least one quarter with a significant effect, we find that 1,122
out of 3,192 firms have a significant effect.
Besides analyzing the statistical significance of the markup effects, we find that only a small
fractionoftheeffectsonprofitratesaresignificant. FigureH.40bshowsthiscalculationforquarterly
data in 2020–2021. Particularly, we find that 19% of the firms had significant effects on profit rates
at the beginning of the pandemic, where most of the firms have a negative effect and over 4% of
the firms had a positive effect. By the first quarter of 2021, the share of firms with positive effects
increasebutisstilllowerthanthosewithnegativeeffects. Whenwecountthenumberoffirmswith
at least one significant effect in one quarter, we find that 953 out of 3,152 firms have a significant
effect.
121
Figure H.40: Fraction of Firms with Significant effects of the pandemic, Bayesian Structural Time
Series Model
5.0
7.5
10.0
12.5
15.0
17.5
20.0
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(a) Markup Rates
2.5
5.0
7.5
10.0
12.5
15.0
17.5
20.0
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(b) Profit Rates
Note: The figure plots the fraction of firms with significant (red solid), positive and significant (blue dotted), and
negative and significant (black dotted) effects. A significant effect is defined when the corresponding 95% posterior
credible interval does not contain zero.
To avoid multiple hypothesis testing issues and given our sample has thousands of firms corre-
sponding to at least an equal number of hypothesis tests, we implement the Benjamini-Hochberg
algorithmtocontrolforthefalsediscoveryrate(FDR),whichistheexpectednumberofstatistically
significant effects that are indeed null effects.
29
As this algorithm relies on the computed p-values
for each individual test to construct a joint test for each firm for the entirety of the pandemic in
our sample, we compute the firm-level p-value for the 2020–2021 average effect on markup rates.
When controlling the FDR at 5% for the effects on markup rates, we find that there are 1,240 out
of 3,192 statistically significant effects. If we were to follow the naive approach of counting the
29
We also implement four corrections that control the family-wise error rate, i.e., the probability of detecting at
least one significant effect that is indeed null, including the Bonferroni, Hochberg, Holm, and Hommel corrections,
andfindnosignificanteffects. However,suchproceduresarequiteconservative,evenmoresowhenonehasthousands
of hypotheses such as in our case (Storey and Tibshirani, 2003). In addition to these procedures, we implement the
q-value approach of Storey and Tibshirani (2003) and found a similar number of tests that are significant while
controlling the FDR at 5%; however, the assumption on the uniform distribution of p-values in the unit interval may
not hold since the highest p-value in our sample is just below 0.5.
122
number of firms with statistically significant effects at 5% using the p-values, we would consider
1,518 significant effects. Overall, the quarterly analysis using Figure H.40 and the FDR analysis
of statistic significance are consistent with each other, with the former producing a slightly higher
fraction of firms with significant effects, i.e., 35.2% versus 38.8%.
We repeat the previous multiple hypothesis correction to evaluate the statistical significance
of the effects on profit rates.
30
Consequently, when controlling the FDR at 5% for the effects on
profit rates, we find that there are 832 out of 3,152 statistically significant effects. If we were to
followthenaiveapproachofcountingthenumberoffirmswithstatisticallysignificanteffectsat5%
using the p-values, we would consider 1,200 significant effects. Overall, the quarterly analysis using
Figure H.40b and the FDR analysis of statistic significance are consistent with each other, with
the former producing a slightly higher fraction of firms with significant effects, i.e., 29.9% versus
26.4%.
One concern when performing the Benjamini-Hochberg correction is that either markup or
profit rates may be highly correlated across firms. To alleviate this concern, we perform both
factor and correlation analyses as in Bermingham and D’Agostino (2014). To have a sense of how
correlated the series are across firms, that study suggests estimating the correlation between each
firm’smarkuporprofitratewiththeprincipalcomponentcalculatedacrossfirms. Toavoidlosinga
bigfractionofoursampleandgiventhepatternofentryandexitoffirmsinthestockexchange, we
estimate principal components in 5-year windows. We then compute the average R
2
for the firm-
level regressions of either the markup or profit rate on the corresponding first principal component.
We obtain an average R
2
of just 0.07 and 0.06 for markup rates and profit rates, respectively. This
result indicates that there is not a sizable correlation across firms for either the markup or the
profit rates. Moreover, the average pairwise correlation in the markup and profit rates across firms
is very small and equal to 0.01 and 0.016, respectively.
In addition, we conduct the corresponding robustness checks by plotting this decomposition
for both the LLPs model and using the hyperparameters in the sensitivity analysis presented in
Section 3.5.
30
Likewise, we find similar effects when using the q-value correction of Storey and Tibshirani (2003). We also
implement four extremely conservative tests correcting for the FDR (Bonferroni, Hochberg, Holm, and Hommel) and
find no significant effects.
123
Fraction of firms with significant effects of the pandemic, LLPs Model
Figure H.41: Fraction of Firms with Significant effects of the pandemic, LLPs Model
10
15
20
25
30
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(a) Markup Rates
10
15
20
25
30
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(b) Profit Rates
Note: The figure plots the fraction of firms with significant (red solid), positive and significant (blue dotted), and
negative and significant (black dotted) effects. A significant effect is defined when the corresponding 95% posterior
credible interval does not contain zero.
Using LLPs, we replicate our analysis for the aggregate markup and profit rates in Section 3.4.
Overall, both BSTS and LLPs forecasts find that there are more negatively affected firms in the
first two-quarters of the pandemic, with LLPs reporting more significantly affected firms than
BSTS. (see Figures H.40 and H.41). Additionally, both methods suggest that the share of firms
benefiting from the pandemic was initially low in 2020 and grew steadily up to the end of 2021.
LLPsseemstobemoreliberalasittendstofindmoresignificanteffectsonbothmarkupandprofit
rates compared to BSTS.
The results tell the same story as those in Figure H.40, but the LLPs model shows more
significant effects in both directions, with roughly 28% significant effects in the first half of 2020.
However,inSection3.3.3weobservethattheBayesianmodelhasbetterout-of-sampleperformance,
so the observed increase observed in the LLPs model might be driven by overestimated effects due
124
to forecasting errors.
Fraction of firms with significant effects of the pandemic, Sensitivity Analysis
Inthissection, werepeattheanalysisinFigureH.40andchangingthehyperparametersasdetailed
inSection3.5. Interestingly,theresultsarequitesimilar,withincreased(decreased)variabilitypro-
ducinghigher(fewer)significanteffects,especiallyforthoseobservingnegativeeffects. Nonetheless,
these changes seem to be modest.
After changing the variability of the hyperparameters governing the priors of the markups
models, either by increasing or decreasing that variability, we validate our finding that at the
beginning of the pandemic, there was a higher faction of negative firms that positively affected
firms (see Panels (a) and (b) of Figure H.42 and Panel (b) of Figure H.40).
125
FigureH.42: SensitivityAnalysis: FractionofFirmswithSignificanteffectsofthepandemic(Quar-
terly Specification)
4
6
8
10
12
14
16
18
20
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(a) Markups, 125% Hyperparameters
6
8
10
12
14
16
18
20
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(b) Markups, 75% Hyperparameters
2.5
5.0
7.5
10.0
12.5
15.0
17.5
20.0
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(c) Profit Rates, 125% Hyperparameters
2.5
5.0
7.5
10.0
12.5
15.0
17.5
20.0
2020−01 2020−07 2021−01 2021−07
(%)
Significant Positive Negative
(d) Profit Rates, 75% Hyperparameters
Note: The figure plots the fraction of firms with significant (red solid), positive and significant (blue dotted), and
negative and significant (black dotted) effects. A significant effect is defined when the corresponding 95% posterior
credible interval does not contain zero.
126
I Pandemic effect heterogeneity by quintiles of key firm charac-
teristics.
To scale the effect across firms with different baseline markup rates, we divide the estimate of the
causal impact for markup rates by the average value of markup rates in 2015–2019 as follows
b τ s
i,t+h
=
Y
i,t+h
− b
Y
i,t+h
Y
i,2015–2019
,
where Y
i,2015–2019
is firm’s i average markup rate in 2015–2019. We acknowledge that choosing
2015–2019 as a scaling period is somewhat arbitrary, but it has the advantage of at least taking a
period that is actually close to the pandemic years.
31
FiguresI.43andI.44reporttheresultsofthisheterogeneityanalysisforthemarkuprates, both
without and with scaling by the average markup rates in 2015–2019. Firms with low markups in
2015–2019 have experienced the strongest reductions in their markup rates during the pandemic.
Incontrast, someofthefirmsinthetop20%highestmarkupsseemtohaveincreasedtheirmarkup
rates both in the first and second years of the pandemic (Panel I.43a). Regarding firm size het-
erogeneity, there are no big differences in the effects on markups (Panel I.43b). Firms with longer
tenure as public firms seem to be better at hedging the pandemic shock since they exhibit fewer
negative effects in the first year of the pandemic (Panel I.43c). Furthermore, the top 20% firms in
the profit rate tend to have positive effects on markup rates in 2021, whereas firms in the other
quintiles tend to have negative effects on markups. The effects are quite heterogeneous for those
firms with negative profit rates in 2015–2019. On the other hand, there are no noticeable differ-
ences in the effects on markups across market shares (Panel I.43e). Finally, firms in the middle of
the distribution of the number of employees tend to have lower impacts on their markups (Panel
I.43f). We find similar patterns when using the quarterly specification of the model (Figures I.46
and I.47).
In addition to showing the heterogeneous impacts of the pandemic on firms’ markup rates,
we also explored heterogeneous effects on the profit rate. Figure I.45 shows such a heterogeneity
analysis. In terms of heterogeneity for firms with different pre-pandemic markup rates, those firms
31
Since the profit rate takes negative values for some firms in some time periods, we do not scale those rates to
avoid interpretation issues.
127
with the highest markups tend to have positive and higher profit rate effects due to the pandemic
(Panel I.45a). Smaller firms tend to be less affected, yet the differences across quintiles appear to
be not significant (Panel I.45b). Firms that became public in the last half of the 2000s and in the
early 1990s tend to have more drastic negative impacts (Panel I.45c). By profit rate, the bottom
20% of firms in the distribution of profit rate have highly heterogeneous effects, whereas the most
profitablefirmsin2015–2019tendtobebetteroffduringthepandemic(PanelI.45d). Furthermore,
firms with larger market shares tend to have larger negative effects on the profit rate (Panel I.45e).
Finally, the very large employers and the very small employers seem to have the worse profit-rate
effects of the pandemic in 2020 (Panel I.45f). When using the quarterly specification of the model,
we find similar patterns in the effects on profit rates across the 2015–2019 average levels of profit
rate, market share, and employment (Figure I.48).
128
Figure I.43: Heterogeneity Analysis of Effect of Pandemic in Markups (Yearly Specification)
−0.3
−0.2
−0.1
0.0
0.1
1 2 3 4
Index
Index
2020 2021
(a) Markup Rate
−0.2
−0.1
0.0
0.1
0.2
9 11 13 15
Thousand of 2010 dollars (log)
Index
2020 2021
(b) Sales
−0.2
−0.1
0.0
0.1
10 20 30 40 50
Y ears
Index
2020 2021
(c) Years in stock exchange
−0.2
−0.1
0.0
0.1
0.2
−60 −40 −20 0 20 40
(%)
Index
2020 2021
(d) Profit Rate
−0.3
−0.2
−0.1
0.0
0.1
0.000 0.005 0.010 0.015 0.020
Share of sales
Index
2020 2021
(e) Market share
−0.2
−0.1
0.0
0.1
0.2
0 10 20 30 40
Number of Employees (thousands)
Index
2020 2021
(f) Employment
Note: y-axis: effect on markup rates as Yit− e
Yit, where Yit is firm i’s markup rate at time period t, and
e
Yi,t is
the firm’s counterfactual markup rate built upon a Bayesian structural time series model with seasonal and local-
level stochastic trend. x-axis: firm’s i average value of either (a) markup rates, (b) sales (logarithm of constant
2010 thousand US dollars), (c) years since publicly traded, (d) profit rates, (e) market shares, and (f) employment
(thousands of employees) in 2015–2019.
129
FigureI.44: HeterogeneityAnalysisofEffectofPandemicinScaledMarkups(YearlySpecification)
−0.15
−0.10
−0.05
0.00
0.05
1 2 3 4
Index
Index
2020 2021
(a) Markup Rate
−0.10
−0.05
0.00
0.05
0.10
0.15
9 11 13 15
Thousand of 2010 dollars (log)
Index
2020 2021
(b) Sales
−0.10
−0.05
0.00
0.05
10 20 30 40 50
Y ears
Index
2020 2021
(c) Years in stock exchange
−0.1
0.0
0.1
−60 −40 −20 0 20 40
(%)
Index
2020 2021
(d) Profit Rate
−0.10
−0.05
0.00
0.05
0.000 0.005 0.010 0.015 0.020
Share of sales
Index
2020 2021
(e) Market Share
−0.1
0.0
0.1
0 10 20 30 40
Number of Employees (thousands)
Index
2020 2021
(f) Employment
Note: y-axis: effect on markup rates as ( Yit− e
Yit)/Y i,2015–2019, where Yit is firm i’s markup rate at time period t,
e
Yi,t is thefirm’s counterfactual markup rate built upon a Bayesian structural time series model with the seasonal and
local-level stochastic trend, and Y i,2015–2019 is the firm’s average markup rate in 2015–2019. x-axis: firm’s i average
value of either (a) markup rates, (b) sales (logarithm of constant 2010 thousand US dollars), (c) years since publicly
traded, (d) profit rates, (e) market shares, and (f) employment (thousands of employees) in 2015–2019.
130
Figure I.45: Heterogeneity Analysis of Effect of Pandemic on Profit Rate (Yearly Specification)
−20
0
20
40
1 2 3 4
Index
Percentage Points
2020 2021
(a) Markup rate
0
25
50
9 11 13 15
Thousand of 2010 dollars (log)
Percentage Points
2020 2021
(b) Sales
−25
0
25
10 20 30 40 50
Y ears
Percentage Points
2020 2021
(c) Years in stock exchange
0
20
40
60
−60 −40 −20 0 20 40
(%)
Percentage Points
2020 2021
(d) Profit rate
0
20
40
0.00 0.05 0.10 0.15
Share of leader's Sales
Percentage Points
2020 2021
(e) Market share
0
20
40
0 10 20 30 40
Number of Employees (thousands)
Percentage Points
2020 2021
(f) Employment
Note: Each plot shows the 5-bin binscatter of the pandemic effect on profit rates by the respective variable in each
panel. y-axis: effect on profit rates as Yit− e
Yit, where Yit is firm i’s profit rate at time period t, and
e
Yi,t is the firm’s
counterfactual profit rate built upon a Bayesian structural time series model with seasonal and local-level stochastic
trend. x-axis: firm’s i average value of either (a) markup rates, (b) sales (logarithm of constant 2010 thousand
US dollars), (c) years since publicly traded, (d) profit rates, (e) market shares, and (f) employment (thousands of
employees) in 2015–2019.
131
Figure I.46: Heterogeneity Analysis of Effect of Pandemic in Markups (Quarterly Specification)
−0.1
0.0
0.1
0.2
0.3
1.0 1.5 2.0 2.5 3.0 3.5
Index
Index
2020 2021
(a) Markup Rate
−0.1
0.0
0.1
0.2
8 10 12 14
Thousand of 2010 dollars (log)
Index
2020 2021
(b) Sales
−0.1
0.0
0.1
10 20 30 40
Y ears
Index
2020 2021
(c) Years in stock exchange
−0.1
0.0
0.1
0.2
−40 −20 0 20
(%)
Index
2020 2021
(d) Profit Rate
−0.15
−0.10
−0.05
0.00
0.05
0.10
0.000 0.005 0.010 0.015 0.020 0.025
Share of sales
Index
2020 2021
(e) Market share
−0.1
0.0
0.1
0.2
0 10 20 30 40
Number of Employees (thousands)
Index
2020 2021
(f) Employment
Note: y-axis: effect on markup rates as Yit− e
Yit, where Yit is firm i’s markup rate at time period t, and
e
Yi,t is
the firm’s counterfactual markup rate built upon a Bayesian structural time series model with seasonal and local-
level stochastic trend. x-axis: firm’s i average value of either (a) markup rates, (b) sales (logarithm of constant
2010 thousand US dollars), (c) years since publicly traded, (d) profit rates, (e) market shares, and (f) employment
(thousands of employees) in 2015–2019.
132
Figure I.47: Heterogeneity Analysis of Effect of Pandemic in Scaled Markups (Quarterly Specifica-
tion)
−0.10
−0.05
0.00
0.05
1.0 1.5 2.0 2.5 3.0 3.5
Index
Index
2020 2021
(a) Markup Rate
−0.08
−0.04
0.00
0.04
8 10 12 14
Thousand of 2010 dollars (log)
Index
2020 2021
(b) Sales
−0.10
−0.05
0.00
0.05
10 20 30 40
Y ears
Index
2020 2021
(c) Years in stock exchange
−0.10
−0.05
0.00
0.05
−40 −20 0 20
(%)
Index
2020 2021
(d) Profit Rate
−0.05
0.00
0.05
0.000 0.005 0.010 0.015 0.020 0.025
Share of sales
Index
2020 2021
(e) Market Share
−0.10
−0.05
0.00
0.05
0.10
0 10 20 30 40
Number of Employees (thousands)
Index
2020 2021
(f) Employment
Note: y-axis: effect on markup rates as ( Yit− e
Yit)/Y i,2015–2019, where Yit is firm i’s markup rate at time period t,
e
Yi,t is the firm’s counterfactual markup rate built upon a Bayesian structural time series model with seasonal and
local-level stochastic trend, and Y i,2015–2019 is the firm’s average markup rate in 2015–2019. x-axis: firm’s i average
value of either (a) markup rates, (b) sales (logarithm of constant 2010 thousand US dollars), (c) years since publicly
traded, (d) profit rates, (e) market shares, and (f) employment (thousands of employees) in 2015–2019.
133
Figure I.48: Heterogeneity Analysis of Effect of Pandemic in Profit Rate (Quarterly Specification)
−20
−10
0
10
1.0 1.5 2.0 2.5 3.0 3.5
Index
Percentage Points
2020 2021
(a) Markup rate
−20
−10
0
10
8 10 12 14
Thousand of 2010 dollars (log)
Percentage Points
2020 2021
(b) Sales
−20
−10
0
10
10 20 30 40
Y ears
Percentage Points
2020 2021
(c) Years in stock exchange
−30
−20
−10
0
10
20
−40 −20 0 20
(%)
Percentage Points
2020 2021
(d) Profit rate
−10
0
10
0.0 0.1 0.2 0.3 0.4
Share of leader's Sales
Percentage Points
2020 2021
(e) Market share
−20
−10
0
10
0 10 20 30 40
Number of Employees (thousands)
Percentage Points
2020 2021
(f) Employment
Note: Each plot shows the 5-bin binscatter of the pandemic effect on profit rates by the respective variable in each
panel. y-axis: effect on profit rates as Yit− e
Yit, where Yit is firm i’s profit rate at time period t, and
e
Yi,t is the firm’s
counterfactual profit rate built upon a Bayesian structural time series model with seasonal and local-level stochastic
trend. x-axis: firm’s i average value of either (a) markup rates, (b) sales (logarithm of constant 2010 thousand
US dollars), (c) years since publicly traded, (d) profit rates, (e) market shares, and (f) employment (thousands of
employees) in 2015–2019.
134

Abstract (if available)

Abstract The first essay, which I used as my job market paper, documents two adjustment margins used by firms facing financial constraints during the Great Recession: (i) changes in product entry and exit and (ii) the components of sales variation. Using Nielsen Homescan, GS1, and the Compustat datasets, and following panel estimation techniques, I find the following stylized facts: (i) firms facing lower liquidity constraints, measured as a higher liquid to total assets, introduce a new product on average during the Great Recession, an increase of roughly 1.2%; (ii) adjustments in product specific yearly sales percentage changes are driven by prices, with sales remain invariant with prices increasing about 1% without statistically significant movement on the customer base or quantities. These responses mask heterogeneity across the quality distribution: financially constrained firms increase their lower-quality product prices by 1.8% while reducing their high-quality product prices by roughly 7% in order to capture customers trading down and exploiting locked-in customers
buying the low-quality varieties. These findings clarify the role of creative destruction on economic activity during downturns, as well as explain the source of changes in sales by their components.

The second essay coauthored with John Jairo Leon studies how Commodity prices can have significant adverse distributional effects on the consumption and income of households, even in commodity-exporting economies. The discretionary income channel dominates the immediate response of consumption to changes in oil prices: a 1 percent increase in oil prices reduces average consumption by 0.18 percent, mainly driven by low and middle-wealth households. This result is
also partially driven by an increase in expenditure on energy-related products for all income quartiles (0.05 percent on average). Households whose consumption is not affected lie within the second
and fourth quartile of the wealth distribution and see an increase in their disposable income (0.08 percent and 0.14 percent) that they seemingly use to hedge their increase in energy consumption. However, one year after the oil price shock, consumption increased by about 0.3 percent across all income quartiles. This is consistent with a redistributional channel driven by a procyclical fiscal policy. Using an instrumental variable (IV) design, the analysis in this paper finds a 1.1 percent increase in government expenditure due to a 1 percent increase in oil-related incomes produced by an oil price shock.

Finally, the third essay coauthored with Jaime Ramirez explores firm-level markup and profit rates during the COVID-19 pandemic for a panel of 3,611 publicly traded firms in Compustat and finds increases for the average firm. We offer conditions to give markups and profit rate forecasts a causal interpretation of what would have happened had the pandemic not happened. Our estimations suggest that had the pandemic not happened, markups would have been 4% and 7% higher than observed in 2020 and 2021, respectively, and profit rates would have been 2.1 and 6.4 percentage points lower. We perform a battery of tests to assess the robustness of our approach. We further show significant heterogeneity in the impact of the pandemic on firms by key firm characteristics and industry. We find that firms with lower than forecasted markups tend to have lower stock-exchange tenure and fewer employees.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Three essays on macro and labor finance

PDF

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

PDF

Essays in macroeconomics

PDF

Essays on firm investment, innovation and productivity

PDF

Essays on Japanese macroeconomy

PDF

Essays on narrative economics, Climate macrofinance and migration

PDF

Three essays on agent’s strategic behavior on online trading market

PDF

Essays on macroeconomics of health and labor

PDF

The impact of economic shocks on firm behavior: insights from three studies

PDF

Three essays on the microeconometric analysis of the labor market

PDF

Three essays on industrial organization

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Essays on sovereign debt

PDF

Essays on the luxury fashion market

PDF

Pregnancy in the time of COVID-19: effects on perinatal mental health, birth, and infant development

PDF

Three essays on housing demographics: depressed housing access amid crisis of housing shortage

PDF

Food deserts and perceptions of food access in urban low-income areas

PDF

The impact of the COVID-19 pandemic on cancer care delivery

PDF

Essays in panel data analysis

PDF

Essays on the role of entry strategy and quality strategy in market and consumer response

Asset Metadata

Creator Espinosa Torres, Juan Andres (author)

Core Title Three essays on heterogeneous responses to macroeconomic shocks

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2023-05

Publication Date 03/06/2023

Defense Date 03/01/2023

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

Tag business cycles,commodity prices,COVID,customer base,economic crises,financial constraints,heterogeneous agents,heterogeneous firms,inflation,macroeconomics,market power,OAI-PMH Harvest,oil prices,pandemic,redistribution

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kurlat, Pablo Daniel (committee chair), De la O, Ricardo (committee member), Morlacco, Monica (committee member)

Creator Email jejespinosat@gmail.com,juanaesp@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC112764287

Unique identifier UC112764287

Identifier etd-EspinosaTo-11493.pdf (filename)

Legacy Identifier etd-EspinosaTo-11493

Document Type Dissertation

Format theses (aat)

Rights Espinosa Torres, Juan Andres

Internet Media Type application/pdf

Type texts

Source 20230313-usctheses-batch-1009 (batch), 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