University of Southern California Dissertations and Theses

Growth and development in Africa: macroeconomic and microeconomic perspectives

(USC Thesis Other)

Growth and development in Africa: macroeconomic and microeconomic perspectives

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content GROWTH AND DEVELOPMENT IN AFRICA: MACROECONOMIC AND
MICROECONOMIC PERSPECTIVES
by
Fatou Kin´e Thioune
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 2024
Copyright 2024 Fatou Kin´e Thioune

Acknowledgments
I am deeply grateful to my advisor, Caroline Betts, for her invaluable guidance and unwavering
support throughout my entire PhD journey. A heartfelt acknowledgment is also extended to my
mentor, Illenin Kondo, who has been by my side since the PhD application process in 2017 until
my successful defense in 2024. I extend my sincere gratitude to Jeffrey Nugent, whose support
and words of encouragement consistently uplift my spirits. Special thanks are also due to Joshua
Aizenman, Monica Morlacco, Vittorio Bassi, and Jeffrey Weaver.
In every endeavor, the unwavering inspiration and motivation from my family—Mame Yacine, Fifi,
Abdou, and Papa—remain my driving force. I am thankful for being blessed with these number one
cheerleaders who believe that I can achieve anything. To my ‘badi`enes’ and my late grandmother,
whose undeniable love and prayers from Kaolack have been a constant source of strength, I express
my heartfelt thanks. I am thankful to my second family, Tonton Abdou and Tata Adou, who have
taken me as their own daughter and support and pray for me in every endeavor. I also appreciate
the support from my families in the US, who opened their homes and hearts, ensuring my smooth
transitions in the US from the beginning of my undergraduate studies in 2014 to the present: my
host parents Rosemary and Papis; my host sister Amber and her husband Damon; and my LA
family Tata Fatima, N´en´e Ndiaye, and Tonton Boll´e.
I am surrounded by a strong army of people who have carried me through my darkest times and
who have shown me that indeed “no man is a failure who has friends.” Special appreciation goes to
my partner, Saliou, whose unwavering support, both emotional, mental, and intellectual, patience,
and care have been instrumental. From his constant encouragement since the beginning of this PhD
journey to his growing interest in and enthusiasm for my research, he has been a constant source of
inspiration. I am immensely grateful to my personal and unpaid therapists, as well as my loudest
ii

cheerleaders—Gbemisola, Monira, Diarra, Sira A¨ıcha, Sokhna, and Mouhammed—for the countless
hours spent listening to me and providing revived hope in life. I extend my gratitude to Bright, a
trailblazer and a role model, as well as my ALA inner ring: Ouli, Olivier, and Nady. To my peers
and friends who made my time in LA memorable—Najim, Hiba, Amy, Bintou, and Nadia—and
to my friends back home who make home what it was and still is—Ramatoulaye, Khady, Cathy,
Colette, Arame, Oumou, Kadella, Palaye, and Petit—I am truly thankful.
This thesis is entirely dedicated to my late father, Badara, who planted the seeds and did not live
long enough to see the fruit. He has raised an ambitious and resilient daughter and has instilled
in me the values of hard work, perseverance, and integrity that have guided me throughout my life
and have brought me this far. One of his fondest wishes was to see one of his kids become a doctor,
and my biggest regret is that he did not live long enough to see that dream come true.
iii

Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Figures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
Abstract. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
1 Disparate Structural Change and Growth Among African Countries: A Comparative
Analysis of Mauritius and Senegal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter One: Introduction ....................................................................... 1
Chapter One: Data................................................................................ 8
National accounts data construction ....................................................... 8
Micro data ................................................................................... 15
Chapter One: Aggregate Growth Accounting Exercises........................................ 16
Factor shares of income ..................................................................... 17
Estimation of the capital stock series ...................................................... 19
Derivation of TFP Series.................................................................... 22
Aggregate growth accounting with baseline factor shares ................................ 23
Aggregate growth accounting with adjusted factor shares ............................... 25
Chapter One: Sectoral Growth Accounting Exercises .......................................... 27
Sectoral factor shares of income ............................................................ 27
Sectoral capital stock series................................................................. 28
Sectoral TFP series.......................................................................... 30
iv

Sectoral growth accounting with baseline factor shares .................................. 31
Chapter One: Model of Structural Transformation and Growth .............................. 34
Household .................................................................................... 35
Production ................................................................................... 35
Competitive equilibrium .................................................................... 37
Chapter One: Calibration ........................................................................ 41
Chapter One: Quantitative Results .............................................................. 43
Benchmark model ........................................................................... 43
Counterfactual exercises .................................................................... 47
Sensitivity analysis - no income effects .................................................... 51
Chapter One: Reforms and Policy Discussion................................................... 53
Chapter One: Conclusion ......................................................................... 55
2 Misallocation and Financial Constraints Among Firms in Sub-Saharan Africa . . . . . . . . . . . . . 57
Chapter Two: Introduction ....................................................................... 57
Chapter Two: Related work ...................................................................... 60
Misallocation Literature..................................................................... 60
Financial Constraints and Firm Growth Literature....................................... 62
Chapter Two: Model of Misallocation ........................................................... 63
Final Output................................................................................. 63
Industry-Level Output ...................................................................... 63
Firm-Level Output .......................................................................... 64
Chapter Two: Misallocation in Sub-Saharan Africa- An Empirical Analysis................. 65
Data .......................................................................................... 66
Calibration ................................................................................... 69
Misallocation and Firms’ Obstacles ........................................................ 70
Chapter Two: The Role of Managers’ Education and Experience ............................. 90
Chapter Two: Robustness Checks ............................................................... 93
Bootstrap Standard Errors ................................................................. 93
Excluding Size as a Control................................................................. 94
v

Chapter Two: Sources of Financing.............................................................. 94
Chapter Two: Conclusion......................................................................... 98
3 COVID-19 and Gender Inequality: Impact in Southern Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Chapter Three: Introduction ..................................................................... 99
Chapter Three: Literature Review ............................................................... 102
Chapter Three: Data.............................................................................. 105
Chapter Three: Stylized Facts.................................................................... 106
Pre-existing Gender Inequality ............................................................. 106
Impact of COVID-19 on Gender Inequality ............................................... 109
Chapter Three: Empirical Strategy: Assessing the Impact of COVID-19 on Gender Inequality ....................................................................................... 112
Chapter Three: Results ........................................................................... 114
South Africa ................................................................................. 114
Eswatini: Probit Model on 2021 COVID-19 Rapid Gender Assessment ................ 132
Lesotho: Probit Model on 2020 COVID-19 Rapid Impact Survey ....................... 135
Namibia: Panel and Probit Models on COVID-19 Household and Jobs Tracker ....... 137
South Africa: Consistency Analysis- Probit Models on NIDS ........................... 139
Chapter Three: Policy Recommendations ....................................................... 140
Policy Recommendations ................................................................... 140
Enabling Environment for Gender Equality ............................................... 142
Chapter Three: Conclusions ...................................................................... 143
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Chapter One Appendix ........................................................................... 155
Aggregate growth accounting without tourism sector .................................... 155
Aggregate growth accounting without finance sector ..................................... 156
Aggregate growth accounting without tourism and finance sectors...................... 156
Sectoral growth accounting with adjusted factor shares.................................. 158
vi

Sectoral growth accounting without tourism sector....................................... 160
Sectoral growth accounting without finance sector ....................................... 161
Sectoral growth accounting without tourism and finance sectors ........................ 162
Other Decompositions....................................................................... 163
Chapter Two Appendix ........................................................................... 172
Derivation of Cost Shares................................................................... 172
Regressions with multiple selected obstacles .............................................. 174
Robustness Checks .......................................................................... 186
Model of Financial Constraints............................................................. 192
Chapter Three: Appendix ........................................................................ 197
Robustness Checks .......................................................................... 197
Addressing Gender Inequality in Southern Africa: An Overview ........................ 202
vii

List of Tables
1.1 Sector classification by system of national accounts ...................................... 10
1.2 Labor shares of income ...................................................................... 19
1.3 Decomposition of VA-Mauritius ............................................................ 24
1.4 Decomposition of VA-Senegal .............................................................. 24
1.5 Decomposition of VA per working-age person-Mauritius ................................. 25
1.6 Decomposition of VA per working-age person-Senegal ................................... 25
1.7 Decomposition of VA-Mauritius ............................................................ 26
1.8 Decomposition of VA-Senegal .............................................................. 26
1.9 Decomposition of VA per working-age person-Mauritius ................................. 26
1.10 Decomposition of VA per working-age person-Senegal ................................... 26
1.11 Sectoral labor shares of income ............................................................. 28
1.12 Adjusted sectoral labor shares of income .................................................. 28
1.13 Decomposition of VA-Mauritius ............................................................ 32
1.14 Decomposition of VA-Senegal .............................................................. 32
1.15 Decomposition of VA per working-age person-Mauritius ................................. 33
1.16 Decomposition of VA per working-age person-Senegal ................................... 33
1.17 Non-homothetic parameters relative to final consumption expenditure- extrapolations 42
1.18 Non-homothetic parameters-in terms of GDP of base year .............................. 42
1.19 Consumption weights........................................................................ 43
1.20 Capital shares................................................................................ 43
1.21 Changes in VA and employment shares - model vs data - Mauritius.................... 46
1.22 Changes in VA and employment shares - model vs data - Senegal ...................... 47
viii

1.23 Average annual growth of total VA ........................................................ 50
2.1 List of Countries............................................................................. 67
2.2 Obstacles ..................................................................................... 68
2.3 Regressions of output distortions on all obstacles......................................... 72
2.4 Output distortions and financial constraints .............................................. 73
2.5 Output distortions and financial constraints .............................................. 74
2.6 Angola ........................................................................................ 76
2.7 Cameroon .................................................................................... 77
2.8 Ethiopia ...................................................................................... 78
2.9 Ghana ........................................................................................ 79
2.10 Guinea ........................................................................................ 81
2.11 Madagascar .................................................................................. 82
2.12 Mali........................................................................................... 83
2.13 Mozambique.................................................................................. 84
2.14 Nigeria........................................................................................ 85
2.15 Senegal ....................................................................................... 86
2.16 Uganda ....................................................................................... 87
2.17 Zambia ....................................................................................... 88
2.18 Regressions of output distortions by size .................................................. 90
2.19 Regressions of output distortions on financial constraint................................. 91
2.20 Regressions of output distortions on financial constraint................................. 92
3.1 UNDP-UN Women COVID-19 Global Gender Response Tracker ....................... 111
3.2 South Africa: NIDS-CRAM COVID-19 Effects on Employment of All Forms.......... 116
3.3 South Africa: LFS COVID-19 Effects on Employment of All Forms .................... 118
3.4 South Africa: NIDS-CRAM COVID-19 Effects on Business Employment .............. 120
3.5 South Africa: LFS COVID-19 Effects on Business Employment ........................ 121
3.6 South Africa: NIDS-CRAM COVID-19 Effects on Informal and Formal Businesses... 122
3.7 South Africa: LFS COVID-19 Effects on Informal and Formal Businesses ............. 123
3.8 South Africa: NIDS-CRAM COVID-19 Effects on Weekly Hours Worked.............. 125
ix

3.9 South Africa: LFS COVID-19 Effects on Weekly Hours Worked ........................ 126
3.10 South Africa: NIDS-CRAM COVID-19 Effects on Monthly Income .................... 128
3.11 South Africa: NIDS-CRAM Dynamic Effects on Selected Outcome Variables ......... 131
3.12 South Africa: LFS Dynamic Effects on Selected Outcome Variables.................... 132
3.13 Eswatini: Effect of Gender Inequality on Employment and Work ....................... 134
3.14 Eswatini: Effect of Gender Inequality on Domestic and Care Work..................... 135
3.15 Lesotho: Effect of Gender Inequality on Employment and Work ........................ 136
3.16 Namibia: Effect of Gender Inequality on Employment Types and Sectors.............. 137
3.17 Namibia: Effect of Gender Inequality on Work Status ................................... 138
3.18 South Africa: Effect of Gender Inequality on Employment of All Forms and Business
Employment ................................................................................. 139
19 Decomposition of VA-Mauritius ............................................................ 155
20 Decomposition of VA per working-age person-Mauritius ................................. 155
21 Decomposition of VA-Mauritius ............................................................ 156
22 Decomposition of VA per working-age person-Mauritius ................................. 156
23 Decomposition of VA-Mauritius ............................................................ 157
24 Decomposition of VA per working-age person-Mauritius ................................. 157
25 Decomposition of VA-Mauritius ............................................................ 158
26 Decomposition of VA-Senegal .............................................................. 158
27 Decomposition of VA per working-age person-Mauritius ................................. 159
28 Decomposition of VA per working-age person-Senegal ................................... 159
29 Decomposition of VA-Mauritius ............................................................ 160
30 Decomposition of VA per working-age person-Mauritius ................................. 160
31 Decomposition of VA-Mauritius ............................................................ 161
32 Decomposition of VA per working-age person-Mauritius ................................. 161
33 Decomposition of VA-Mauritius ............................................................ 162
34 Decomposition of VA per working-age person-Mauritius ................................. 162
35 Real VA per working-age person-Growth rate............................................. 164
36 Growth rate-real VA per employed person ................................................ 166
37 Labor per working-age person .............................................................. 166
x

38 TFP growth decomposition with baseline factor shares of income....................... 168
39 TFP growth decomposition with adjusted factor shares of income ...................... 169
40 Sectoral contribution to TFP growth ...................................................... 170
41 Decomposition of growth of labor productivity ........................................... 171
42 Angola: Distortions and Obstacles Regressions ........................................... 174
43 Cameroon: Distortions and Obstacles Regressions........................................ 175
44 Ethiopia: Distortions and Obstacles Regressions ......................................... 176
45 Ghana: Distortions and Obstacles Regressions............................................ 177
46 Guinea: Distortions and Obstacles Regressions ........................................... 178
47 Madagascar: Distortions and Obstacles Regressions...................................... 179
48 Mali: Distortions and Obstacles Regressions .............................................. 180
49 Mozambique: Distortions and Obstacles Regressions..................................... 181
50 Nigeria: Distortions and Obstacles Regressions ........................................... 182
51 Senegal ....................................................................................... 183
52 Uganda ....................................................................................... 184
53 Zambia ....................................................................................... 185
54 Regressions of output distortions on financial constraint with bootstrap SEs .......... 187
55 Regressions of output distortions on financial constraint with bootstrap SEs .......... 188
56 Regressions of output distortions on financial constraint with bootstrap SEs .......... 189
57 Regressions of output distortions excluding size .......................................... 190
58 Regressions of output distortions excluding size .......................................... 191
59 South Africa LFS: Clustering Standard Errors at the Province Level................... 197
60 South Africa LFS: Clustering Standard Errors at the Occupation Level................ 198
61 South Africa LFS: Clustering Standard Errors at the Sector Level ..................... 198
62 South Africa LFS: Effects of Marriage on Selected Outcome Variables ................. 199
63 South Africa NIDS: Effects of Age on Selected Outcome Variables ..................... 199
64 South Africa LFS: Effects of Age on Selected Outcome Variables ....................... 200
65 South Africa NIDS: Effects of Education on Selected Outcome Variables .............. 200
66 South Africa LFS: Effects of Education on Selected Outcome Variables................ 201
xi

List of Figures
1.1 Growth, institutions, and commodity export.............................................. 2
1.2 Real GDP per capita-log .................................................................... 3
1.3 Sectoral real value added shares............................................................ 4
1.4 Sectoral employment shares ................................................................ 4
1.5 Nominal value added from different sources-Mauritius ................................... 12
1.6 Nominal compensation of employees from different sources-Mauritius .................. 12
1.7 Nominal value added from different sources-Senegal...................................... 13
1.8 Nominal compensation of employees from different sources-Senegal..................... 14
1.9 All estimated capital stock series........................................................... 21
1.10 Total factor productivity .................................................................... 22
1.11 Sectoral capital stock series................................................................. 29
1.12 Sectoral TFP- Baseline factor shares ...................................................... 30
1.13 Value added shares-data vs baseline model-Mauritius.................................... 44
1.14 Employment shares-data vs baseline model-Mauritius ................................... 45
1.15 Value added shares-data vs baseline model-Senegal ...................................... 45
1.16 Employment shares-data vs baseline model-Senegal ...................................... 46
1.17 Value added shares-data vs baseline vs counterfactual 1-Senegal ........................ 48
1.18 Value added shares-data vs baseline vs counterfactual 2-Mauritius ..................... 49
1.19 Value added shares-data vs baseline vs counterfactual 2-Senegal ........................ 49
1.20 Value added shares-data vs baseline vs counterfactual 3-Senegal ........................ 50
1.21 Value added shares-data vs baseline vs no income effects -Mauritius ................... 52
1.22 Labor shares-data vs baseline vs no income effects - Mauritius.......................... 52
xii

1.23 Value added shares-data vs baseline vs no income effects - Senegal ..................... 53
1.24 Labor shares-data vs baseline vs no income effects-Senegal .............................. 53
2.1 Degree of access to finance obstacle ........................................................ 68
2.2 Firm size distribution ....................................................................... 69
2.3 Sources of financing working capital ....................................................... 95
2.4 Sources of financing working capital by firm size ......................................... 96
2.5 Loan application ............................................................................. 97
2.6 Existence of a line of credit ................................................................. 97
2.7 Loan application rejection- reasons ........................................................ 97
3.1 Growth Differential with ASEAN5 Countries ............................................. 101
3.2 SACU Countries: Gender Inequality Overview ........................................... 107
3.3 SACU Countries: Gender Inequality on Business and Political Activities .............. 108
3.4 SACU Countries: Gender Inequality on Financial Activities ............................ 109
3.5 SACU Countries: Gender Inequality on Various Employment Sectors .................. 110
3.6 SACU Countries: Gender Inequality in Health ........................................... 111
7 Aggregate TFP-whole economy vs economy without tourism-Mauritius................ 155
8 Aggregate TFP-whole economy vs economy without finance-Mauritius ................ 156
9 Aggregate TFP-whole economy vs economy without tourism and finance-Mauritius .. 157
10 Real VA per working-age person ........................................................... 164
11 Growth rate-real VA per working-age person ............................................. 165
12 Real VA per employed person .............................................................. 165
13 Growth rate-real VA per employed person ................................................ 166
14 Labor per working-age person .............................................................. 167
15 Decomposition real VA per WA person .................................................... 167
xiii

Abstract
This dissertation contributes to our understanding of African economies and their development
outcomes, as well as various factors affecting their macroeconomic growth. In the first chapter, I
investigate the driving forces behind divergent economic growth and structural transformation in
Mauritius and Senegal, two historically similar sub-Saharan African nations with differing development trajectories. In chapter two, I explore the extent to which financial constraints contribute
to the firm-level resource misallocation that I show is present in 12 sub-Saharan African countries.
Finally, chapter three analyzes the impact of COVID-19 on gender disparities in the Southern
African region, notably in South Africa, Eswatini, Lesotho, Botswana, and Namibia. Each chapter
addresses pivotal issues crucial for comprehending macroeconomic development in Africa, including
productivity, resource allocation inefficiencies, gender inequality, and the impact of macroeconomic
shocks. Consequently, this dissertation makes substantive contributions to policy-relevant research
in Africa.
xiv

Chapter 1
Disparate Structural Change and Growth Among African Countries: A Comparative Analysis of Mauritius and Senegal
Chapter One: Introduction
Historical data show that developed countries across the world experienced a period of structural
change, characterized by a falling value added share of the agricultural sector, a rising value added
share of the services sector, and a hump shape in the value added share of the manufacturing sector
(Herrendorf et al., 2014). This process of structural change is a key feature of long-run, sustained
economic growth in neoclassical growth theory (Kuznets (1973), Herrendorf et al. (2014)). The
extant literature has demonstrated that both income and price effects play critical roles in structural change (Herrendorf et al. (2014), Ngai and Pissarides (2007), Kongsamut et al. (2001)) and
that these forces can cause sectoral reallocation of resources to persist in the long-run (Comin et al.
(2021)). Some developing countries, however, experience an alternative pattern of structural change
in which value added moves from primary to tertiary sectors, thus affecting their development outcomes (Adhikari (2019) and Verma (2012) analyze the case of India, for example). In Africa, many
countries remain poor with large agricultural sectors, while a few have successfully industrialized
and grown richer. There is very little understanding of countries’ heterogeneous structural change
and economic growth performances.
The empirical and theoretical contribution of this paper is to better understand disparate structural
change and growth in sub-Saharan African countries and the extent to which factors such as total
factor productivity (TFP), capital, and labor explain the sectoral reallocation of their economies
1

Figure 1.1: Growth, institutions, and commodity export
Notes: Growth rate of real value added per working-age person: average of one-year growth rates
between 1990 and 2018. Political stability index in 2019. Size of the bubbles represents commodity
export intensity in 2018-2019; smallest size: the percent of total merchandise exports composed of
commodities is less than 50, medium size: it is bigger than 50 and less than 80; and largest size it
is bigger than 80.
Data sources: Political stability index from World Bank World Governance Indicators; real value
added from Economic Transformation Database; working-age population from UN national populations; percent of total merchandise exports composed of commodities, 2018-2019 from UNCTAD.
over time. The focus of the project is a comparison of the very different growth and structural
change experiences of two sub-Saharan African countries, Mauritius and Senegal. Historically,
Mauritius and Senegal are two similar African countries: they are both former French colonies,
have been politically stable, and are not major commodity exporters. However, between 1980 and
2019, Mauritius has experienced a fast and sustained growth, while Senegal has struggled to grow.
Figure 1.1 shows that only six out of a sample of 21 African countries have a positive political
stability index, and of those six, only Senegal has a growth rate that is less than one percent over
the period 1990-2018. Of the five countries with more than one percent growth, only Mauritius has a
percentage of commodity export that is less than 80 percent of total export, which is comparable to
the level of Senegal. We can therefore rule out commodity export boom as a potential explanation
of Mauritius’ fast growth, and political instability as the cause of Senegal’s slow growth. Their
2

stark growth differences must therefore have been driven by policies and reforms that have altered
the sectoral reallocation of their economies.
Mauritius has followed the quintessential standard pattern of structural transformation, where
both labor and economic activity moved from agriculture to the manufacturing sector, and after
a brief period of industrialization, are moving to the services sector. On the other hand, Senegal
has experienced little to no structural transformation demonstrated by its constant sectoral value
added shares. These puzzling facts are depicted in Figures 1.2, 1.3 and 1.4.
Figure 1.2: Real GDP per capita-log
Data source: World Bank World Development Indicators.
This project seeks to answer the following question: what accounts for the different growth and
structural change experiences in Mauritius and Senegal? These two countries constitute an important case study that will give broader implications for the prospects of growth and development
in Africa, as Mauritius is one of the fastest growing economies in sub-Saharan Africa, and Senegal
is one of the largest economies in Francophone West Africa but has experienced little growth over
the past four decades.
3

Figure 1.3: Sectoral real value added shares
Data constructed by author.
Figure 1.4: Sectoral employment shares
Data sources: The 10-Sector and the Economic Transformation
Databases.
4

To this end, I first collect historical national accounts data mostly from primary sources and harmonize them to construct historical sectoral data series in the same comparable national accounts
system from 1980 to 2019. This data is used to first conduct an empirical accounting of Mauritius
and Senegal’s economies using aggregate and sectoral growth accounting exercises, and to isolate
the contribution of capital, labor, and TFP. I then perform a structural accounting by calibrating
and solving a three-sector neoclassical growth model of structural change with both income and
price effects to match features of these two economies. I compare the model’s predictions to the
data to see the extent to which the model can account for facts of structural transformation across
the three major sectors of economic activity: agriculture, manufacturing, and services. I use the
baseline model to run some counterfactual growth exercises to capture the important factors driving
structural transformation.
From the aggregate growth accounting exercises, I find that Mauritius and Senegal had the same
growth in labor inputs and Senegal had a faster growth in capital inputs but experienced a much
slower growth in value added due to its negative TFP growth. Further analyses excluding the
tourism and finance industries in Mauritius reveal that the fast TFP growth in Mauritius is not
driven by these two sectors. The sectoral growth accounting exercises reveal that in both Mauritius
and Senegal, labor is moving to the services sector. However, in Mauritius, the services sector has
the fastest growing TFP while in Senegal it has the lowest TFP growth. The positive correlation
between sectoral relative TFP and labor growth in Mauritius is an unexpected result, as when
goods are assumed to be gross complements in preferences, the baseline structural transformation
model predicts that the labor share of a sector decreases as its relative TFP rises.
To better understand the driving forces behind structural transformation in these two countries,
I use a three-sector neoclassical growth model of structural change, carefully calibrated to match
initial-year national income and input-output data in Mauritius and in Senegal. I simulate the
model and find that it tracks very closely the time-series data on value added shares of the agricultural, manufacturing, and service sectors in both countries - and especially so for Mauritius. There
is some discrepancy between the levels of employment shares of sectors that it predicts relative
to the data, however, the model successfully matches the trends in employment shares of sectors.
Counterfactual exercises show that: 1. Senegal would have grown 2.5 times faster between 1980
5

and 2019 if it had the sectoral TFP growth of Mauritius; 2. Senegal would have grown 1.5 times
faster than Mauritius between 1992 and 2019 if it had the sectoral TFP growth of Mauritius; 3.
income effects are a stronger driver of structural transformation in both Mauritius and Senegal,
and that explains why labor moves to the services sector in both countries despite it having the
fastest growing TFP in Mauritius. Therefore, differential TFP growth is not as important as fast
aggregate income growth. If a country like Mauritius or Senegal can sustain a fast aggregate TFP
growth, regardless of which sector’s TFP increases the fastest, the country’s services sector will
experience a boom. This finding suggests that for Senegal to experience structural transformation,
it needs to enact policies and reforms to accelerate its TFP growth, so that a faster aggregate
growth induces a faster change in the sectoral composition of its economy.
Related literature and contribution. This paper contributes to two main strands of literature:
first it adds to the literature on structural transformation and economic growth, and second to
the role of TFP in the reallocation of economic activity. Existing literature has established a link
between economic growth and structural transformation (Rodrik (2016b)). Some developing countries, mostly those in Asia, have had successful structural transformation which led to significant
economic growth (Betts et al. (2017), Dekle and Vandenbroucke (2012)). Most developing countries, however, experience an alternative pattern of structural change in which resources move from
the primary to the tertiary sectors. This has been addressed by a wide range of papers (Haraguchi
et al. (2016), Rodrik (2016b), de Vries et al. (2015)).
In Africa, structural transformation has been disappointing in most countries and there is little
research on the mechanisms driving it. Few research papers analyze a cross-section of African
countries and focus mostly on quantifying the extent of structural transformation through some
decompositions (McMillan et al. (2014), McMillan et al. (2017), Kaba et al. (2022), Oehmke et al.
(2016)). These cross-sectional studies mask differences across countries and do not delve into the
mechanisms that explain the different dynamics of sectoral reallocation. The lack of research and
evidence on structural transformation and growth in African countries is partly due to data limitation. I therefore add to the literature on structural transformation and economic growth by first
constructing historical sectoral national accounts data for Mauritius and Senegal and documenting
their growth experiences and the sectoral reallocation of their economies over each decade from
6

1980 to 2019. In addition to documenting the path and dynamics of structural transformation, I
use a theoretical model to explain the mechanisms behind the observed patterns. This paper, to
my knowledge, is the first to calibrate a model of structural transformation to data from African
countries. By focusing on these two countries as a case study, this paper paints a more complete
picture of their structural change and pins down the factors driving it.
The second strand of literature this paper contributes to is the relationship between sectoral TFP
and reallocation of resources. There are two main drivers of structural transformation: income
effects and price effects. As income grows, people consume more of the services sector output thus
leading to a faster growth of that sector (Kongsamut et al. (2001)). In parallel, when goods across
sectors are assumed to be gross complements in preferences, the sector with faster TFP growth
experiences a faster decline in its relative price, and thus a decline in its consumption share (Ngai
and Pissarides (2007)). Therefore, differential sectoral TFP growth is a main driver of structural
transformation through price effects, as argued by Jeong (2020), Dekle and Vandenbroucke (2012),
Kehoe et al. (2018). However, the extent to which price effects drive structural change relative
to income effects depends on a country’s economy. Sposi et al. (2021) find that in high income
countries, economic activity moves to sectors with lower productivity growth. On the other hand,
McMillan et al. (2014) argue that growth in low-income countries is driven by reallocation of
economic activity and labor from the low-productivity traditional sector to the high-productivity
modern sectors. This is at the basis of the dual economy model, described by Rodrik (2016a)
as the fact that traditional sectors, mainly agriculture, are stagnant, while the modern sectors,
manufacturing and services, are characterized by innovation and productivity growth. Verma (2012)
shows that in India, the services sector has boomed at the expense of the manufacturing sector
due the higher growth rate of TFP in services. Amadou and Aronda (2020) and Dieme (2018)
suggest some evidence that there are great disparities in structural transformation across African
sub-regions, as in some regions, resources are being allocated towards more productive sectors,
while in others resources move to less productive sectors. In my paper, I find the specific linkages
between sectoral TFP growth and structural transformation in Mauritius and Senegal, and further
explain the apparent disparities in the relationship between TFP growth and labor reallocation
using a structural model. I show that income effects, and not price effects, are the primary drivers
7

of structural transformation in these two countries, so that to the extent that they get richer,
resources move to the services sector.
The rest of the paper is organized as follows: in section 2, I discuss my data collection and construction exercises. Then, I explain the aggregate and sectoral growth accounting exercises and discuss
their results in sections 3 and 4. Sections 5 and 6 present the model and calibration approach. I
discuss my quantitative exercises and results in section 7 and some policy implications in section
8, before concluding in section 9.
Chapter One: Data
National accounts data construction
For my growth accounting exercises and structural transformation model, I use national accounts
data for Mauritius and Senegal on their sectoral production and income allocation. There are two
main challenges in obtaining such detailed data for African countries. The first challenge is that the
system of national accounts (SNA), which is the general accounting framework used by countries
to keep and report their national accounts, changed over time. Between 1980 and 2019, countries
have used different SNAs: 1968 SNA, 1993 SNA, and 2008 SNA. For each new SNA, revisions were
made on the classification of the sectors of economic activity, the valuation and computations of
certain variables, and through the introduction of new concepts and definitions. A second challenge
is that there are different sources for national accounts data for African countries, each with its
time-period covered, its variables included, and its SNA. There is no single database reporting all
the variables needed for my study at the sectoral level, covering the whole sample period I am
interested in (from 1980 to 2019), and in the same SNA. Part of my contribution is to overcome
these challenges by constructing comparable historical data.
The first main exercise of my project was therefore to collect and construct detailed, uniform,
and consistent time series of national accounts data for Mauritius and Senegal, disaggregated at
the industry level. To this end, I use the latest national accounts data released by the countries’
statistics offices to obtain the revised series in 2008 SNA, and then use growth rates of the series in
8

earlier systems and years from the archival data to back-cast a historical harmonized series. This
same methodology is used by the Economic Transformation Database (ETD) (Kruse et al. (2022))
and the 10-sector database (10 SD) (Timmer et al. (2015)), two databases from the Groningen
Growth and Development Center that have sectoral value added and employment for a wide range
of African countries, including Mauritius and Senegal, from 1970 to 2018. However, there are
two main reasons why their series on sectoral value added differ from mine. First, I use more
recent revisions of national accounts data and thus obtain updated series. Second, I compute the
series at a more disaggregated level of industry classification with 19 industries, while the ETD
and the 10 SD disaggregate the economies into 12 and 10 industries respectively. I however use
these databases to supplement my data and most importantly I use their estimation of sectoral
active population. Furthermore, these databases are not sufficient for my study as they do not
have data on compensation of employees and gross operating surplus, two data series necessary
for my methodological approach. The other main source of sectoral income data is the EORA
input-output tables, which have low reliability due to their high-detail yearly data obtained from
extrapolations, and the quality of the data is even lower for years prior to 1990 and for developing
countries.
The main challenge when constructing these time series was harmonizing the industries. The
2008 SNA is based on the International Standard Industrial Classification (ISIC) revision 4 with
19 industries, while the 1993 SNA is based on ISIC revision 3 with only 14 industries and the
1968 SNA divides the economy into only 10 industries. I decided to maintain the 19 industries
of the ISIC revision 4 as in the 2008 SNA. To obtain the same 19 industries for the 1993 and
1968 SNAs, I split some industries in these SNAs into subindustries to match the industries in
the 2008 SNA. For the 1993 SNA, I did the following splits: “electricity, gas and water supply”
into “electricity, gas, steam and air conditioning” and “water supply; sewerage, waste management
and remediation activities”; “transportation, storage and communication” into “transportation and
storage” and “information and communication”; “real estate, renting and business activities” into
“real estate activities”, “professional, scientific and technical activities”, and “administrative and
support service activities”; and finally “other community, social and personal services” into “arts,
entertainment and recreation” and “other service activities”.
9

Similarly for the 1968 SNA, I split: “electricity, gas and water” into “electricity, gas, steam and
air conditioning” and “water supply; sewerage, waste management and remediation activities”;
“wholesale and retail trade, restaurants and hotels” into “wholesale and retail trade” and “restaurants and hotels”; “transportation, storage and communication” into “transportation and storage”
and “information and communication”; “financing, insurance, real estate, and business activities”
into “financial and insurance activities”, “real estate activities”, “professional, scientific and technical activities”, and “administrative and support service activities”; and finally “other community,
social and personal services” into “public administration and defence; compulsory social security”,
“education”, “human health and social work activities”, “arts, entertainment and recreation” and
“other service activities”. Table 1.1 details the way in which I split the 1993 and 1968 SNAs into
the 19 industries of the 2008 SNA.
Table 1.1: Sector classification by system of national accounts
2008 SNA 1993 SNA 1968 SNA
Agriculture, forestry, and fishing Agriculture, hunting, forestry, and fishing Agriculture, hunting, forestry, and fishing
Mining and quarrying Mining and quarrying Mining and quarrying
Manufacturing Manufacturing Manufacturing
Construction Construction Construction
Electricity, gas, steam, and air conditioning supply Electricity, gas, and water Electricity, gas, and water Water supply; sewerage, waste management, and remediation activities
Wholesale and retail trade; repair of motor vehicles and motorcycles Wholesale and retail trade; repair of motor vehicles and motorcycles Wholesale and retail trade, restaurants and hotels Accommodation and food service activities Hotels and restaurants
Transportation and storage Transport, storage, and communication Transport, storage, and communication Information and communication
Financial and insurance activities Financial intermediation
Financing, insurance, real estate, and business services Real estate activities
Professional, scientific, and technical activities Real estate, renting, and business activities
Administrative and support service activities
Public administration and defence; compulsory social security Public administration and defence; compulsory social security
Education Education Producers of government services
Human health and social work activities Health and social work
Arts, entertainment, and recreation Other services Community, social, and personal services Other service activities
To achieve these splits, I compute the share of each subindustry based on the data from 2008 SNA.
Then I extrapolate the shares of these subindustries for the 1993 and 1968 SNA data using the
data on the industry I want to split, and use the extrapolated shares to estimate the series for the
subindustries. For example, to split “electricity, gas and water supply” from the 1993 SNA into
“electricity, gas, steam and air conditioning” and “water supply; sewerage, waste management and
remediation activities” as in the 2008 SNA, I compute the shares of “electricity, gas, steam and air
conditioning” and “water supply; sewerage, waste management and remediation activities” out of
the total of those two industries from the data in 2008 SNA. I then use the “electricity, gas and
water supply” series from the 1993 SNA to extrapolate the shares of “electricity, gas, steam and
air conditioning” and “water supply; sewerage, waste management and remediation activities” and
10

get the 1993 SNA estimates of these two sectors. I repeat this exercise for all the subindustries in
both the 1993 and 1968 SNAs. A detailed explanation of the construction of my time series is given
in the Online Appendix of this paper. Below, I give an overview of the data sources and series of
the two countries.
Mauritius
I have five main sources for Mauritius national accounts data: the national accounts database
from Statistics Mauritius; the input-output tables from Statistics Mauritius; the national accounts
statistics: main aggregates and detailed tables from the United Nations; archival data from the
Government Information collections of the Harvard Lamont library; and the 10-sector database.
The last national accounts from Statistics Mauritius were released in December 2022 and are the
latest and most accurate estimations of the country’s economic activities. They were compiled
following the 2008 SNA and cover the period from 2006 to 2019. The input-output tables are
published every five years from 1992: 1992, 1997, 2002, 2007, 2013, 2018. The UN main aggregates
and detailed tables were published in 2019 and were based on the 2008 SNA from 2013 to 2019, and
on the 1993 SNA from 2008 to 2013 (with overlapping year 2013). The archival data from Harvard
are the longest sectoral time series data I use and are reported based on all three SNAs: 2008 SNA
from 2015 to 2019, 1993 SNA from 1998 to 2015, and 1968 SNA from 1980 to 1998. The 10-sector
database gives sectoral data on value added and employment for a wide range of African, Asian,
and Latin American countries from roughly 1960 to 2010. For Mauritius, I construct historical
and harmonized sectoral time series for: value added, compensation of employees, gross operating
surplus, and gross fixed capital formation.
11

Figure 1.5: Nominal value added from different sources-Mauritius
Data sources: ETD: Economic Transformation Database; 10
SD: 10-sector database; My data: data constructed by author.
Figure 1.6: Nominal compensation of employees from different sources-Mauritius
Data sources: EORA input-output tables; My data: data constructed by author.
Figures 1.5 and 1.6 show my data series and those from other sources for Mauritius. My sectoral
12

value added are almost identical to those reported by the ETD and the 10 SD. This validates my
computations and the small differences indicate the revisions in the national accounts since the ETD
was published. However, there is a big disparity between my series on compensation of employees
and those reported by EORA input-output tables. This confirms that the data from EORA IO
tables are of lower quality compared to those obtained directly from the national accounts. Some
of the discrepancy may however be due to that fact that the EORA IO tables report the series
in USD and I converted their series in LCU by multiplying them by the exchange rates from the
World Bank.
Senegal
Figure 1.7: Nominal value added from different sources-Senegal
Data sources: ETD: Economic Transformation Database; 10
SD: 10-sector database; My data: data constructed by author.
The data construction for Senegal follows the same methodology as the one used for Mauritius. I
have two main data sources for Senegal: the UN national accounts: Main Aggregates and Detailed
Tables, and archival data from Harvard library’s Government Information collections. I consider
the data from the UN main aggregates and detailed tables published in 2020 as the latest revisions
of Senegal’s national accounts in 2008 SNA. The data from the UN tables published in 2019 and
2018 are in 1993 SNA. The archival data from Harvard’s library are published in 1968 SNA and
13

Figure 1.8: Nominal compensation of employees from different sources-Senegal
Data sources: EORA input-output tables; My data: data constructed by author.
go back to 1980. I link data from these different sources to obtain detailed uniform historical time
series of national accounts data for Senegal. I construct such sectoral times series for: value added,
compensation of employees, gross operating surplus, and consumption of fixed capital.
Figures 1.7 and 1.8 show the sectoral value added and compensation of employees from my construction compared to other sources. Again, my sectoral value added series are fairly close to the
ETD series, but quite different from the ones reported by the 10 SD. And again, the EORA IO
data are very different from my own data.
With this extensive exercise of data collection and detailed constructions of sectoral national accounts series, I have a database of key time series disaggregated at 19 industries, for the two
countries, making a substantive contribution to data collection and construction for African countries. In addition to providing more disaggregated sectoral value added, I have constructed reliable
sectoral income data which were not readily available before. Moreover, I have sectoral gross fixed
capital formation and consumption of fixed capital that will allow me to later compute sectoral
capital stock series, again data that are not readily available for African countries. Finally, with all
these data and the derived capital stock series, I derive TFP series at the aggregate and sectoral
levels for both countries, another major contribution in understanding these African economies.
14

Micro data
In addition to the macro data from the countries’ national accounts, I use micro-level data, specifically household and labor force surveys for Mauritius and Senegal respectively. The goal of the
usage of micro data is to derive national estimates of sectoral and aggregate average hours worked
and income from self-employment and from informal employment for the growth accounting exercises. The average hours worked, multiplied by the number of people employed, is a measure
of effective labor. In my growth accounting, I use this measure for labor inputs as a robustness
check. Additionally, there is generally an underestimation of labor share of income in developing
countries in particular, as measures of compensation of employees from national accounts do not
include the income from self-employment and informal employment. Therefore, I use estimates of
national income from self-employment and informal employment derived from the micro data to
adjust for the labor shares of income.
For Mauritius, I use their Continuous Multi-Purpose Household Survey (CMPHS) from Statistics
Mauritius and for Senegal I use the ‘Enquˆete Nationale de l’Emploi au S´en´egal’ (ENES) from the
statistics agency (‘Agence Nationale de la Statistique et de la D´emographie’). The CMPHS is a
yearly household survey covering a range of topics including the labor force, and conducted in
1999 and then from 2001-2019. The ENES is a national labor force survey conducted in 2015,
and quarterly from 2017-2019. Unfortunately, these are the longest time-series micro data covering
labor force activity there is for each country.
To obtain estimates of national average hours worked and informal income, I model after Young
(1995) methodology. Specifically, from the micro data, for each gender and age group category,
I first compute sample employment rates (total employment rate and informal employment rate)
as the ratio of the sample number of people employed to the sample working-age population as
shown in equation 1.1. Then I multiply these employment rates by the total national workingage population by gender and age group obtained from the United Nations’ population database
to derive national estimates of total employed population by gender and age group, as shown in
equation 1.2. To derive total hours worked for each gender and age group category, I multiply these
estimates of total employed populations by the sample average hours worked by gender and age
15

group from the micro data (equation 1.3). I take the sum across all the categories to get the total
number of national hours worked and further divide this by the total employed population to get
the average hours worked nationally. Similarly for income from the informal sector, I multiply the
sample average wages for workers in the informal sector by the total informally employed population
by gender and age group to obtain the national income from informal employment by gender and
age group (equation 1.4). I take the sum over all the categories to obtain the total national monthly
income from self-employment. There are two gender categories, male and female, and 11 age groups
for the working-age population: 15-19, 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64,
65-69. I do these exercises at the sectoral level as well. A more detailed explanation of the micro
data and the national estimates of hours worked and informal wages is outlined in the Online
Appendix.
sample (informal) ERtgac =
sample number of people (informally) employedtgac
sample W APtgac
(1.1)
(informal) NEtgac = sample (informal) ERtgac × national W APtgac (1.2)
average HWtc =
1
number of people employedtc
X
g,a
NEtgac × average HWtgac (1.3)
informal incometc =
X
g,a
informal NEtgac × average monthly informal incometgac (1.4)
where ER is employment rate, W AP is working-age population, NE is national employment, HW
is hours worked, at time t for country c, for each gender g age group a.
Chapter One: Aggregate Growth Accounting Exercises
The first growth accounting exercise I conduct consists of decomposing each economy’s growth in
value added into growth in its different components, namely growth in TFP, growth in capital
inputs, and growth in labor inputs. The goal of this exercise is to estimate the contribution of each
of these factors to the countries’ respective growth and understand the origins of the divergences
in the two countries’ growth performance. I first do this exercise at the aggregate level.
16

For each country, I assume their aggregate value added is produced using a Cobb-Douglas production function:
Yt = AtKα
t L
1−α
t
(1.5)
for year t, where Y is the real value added, A the total factor productivity, K and L are the amounts
of capital and labor inputs respectively, and α is the capital share of income. Because of the lack
of data on hours worked for the two countries, labor inputs here are simply the number of people
employed. I first estimate the capital shares of income, α. I then estimate the capital stock series
K for the two countries using the perpetual inventory methods. Once factor shares of income and
the capital stock are estimated, TFP can be inferred. The aggregate growth accounting exercises
will illuminate the factors that can account for the divergence in growth of the two countries’ value
added.
Factor shares of income
The standard formula to compute the labor share of income, 1−α, is the ratio between unambiguous
labor income and the sum of unambiguous labor income and unambiguous capital income, as follows:
1 − α =
unambiguous labor income
unambiguous labor income + unambiguous capital income (1.6)
However, such series are not readily available for these two countries. I therefore use their data on
compensation of employees and gross operating surplus to estimate their factor shares of income:
1 − α =
compensation of employees
compensation of employees + gross operating surplus (1.7)
For both countries, compensation of employees is defined as all payments by producers of wages and
salaries to their employees, in kind and in cash. These include commissions, overtime payments,
bonuses, cost of living allowances, housing allowances, etc. Gross operating surplus is defined as
the excess of value added over the cost of employees’ compensation, consumption of fixed capital
and indirect taxes reduced by subsidies. I take the average of 1 − α over the whole sample period.
This measure is not a perfect measure of labor share of income as gross operating surplus may give
17

an overestimate of capital income. Specifically, the measure of gross operating surplus may include
payments towards other factors than capital, such as interest on loans or dividends to shareholders,
payments to enterprise owners, and mixed income. However, this is the best estimate of capital
income I could obtain from the two countries’ national accounts.
To correct for these mismeasurements, I adjust the labor income for self-employment and informal
employment wages for Mauritius and Senegal respectively. For Mauritius, compensation of employees does not include wages paid to the employer/ the self-employed. And for Senegal, it does
not include wages to people informally employed. Thus, these incomes from self-employment and
informal employment are included in gross operating surplus. Using the annual total income from
self-employment in Mauritius derived from the CMPHS and the annual total income from informal
employment from the statistics office in Senegal, I compute an adjusted labor share of income as
follows:
1 − α =
compensation of employees + other employment income
compensation of employees + gross operating surplus (1.8)
where other employment income is self-employment income for Mauritius and informal employment
income for Senegal. However, the CMPHS goes only from 2001 to 2018. Thus, I compute the
baseline labor share for the whole sample period (average), 1980-2019, then compute the average
for the period 2001-2018. Then I take the ratio between the long baseline labor share and the
short baseline labor share, and multiply this ratio by the adjusted labor share (average) to obtain
an adjusted labor share that reflects the whole sample period. These adjustments are shown in
equation 1.9. For Senegal, I use informal income from the Senegalese statistics agency ‘Agence
Nationale de la Statistique et de la D´emographie’ (ANSD) to adjust for the labor share using the
same equation as above. This informal income is from 2014 to 2019. I then follow the same steps
described above to adjust for the whole sample period 1980 to 2019.
long adjusted labor share1980−2019 =
baseline labor share1980−2019
baseline labor share2001−2019
× adjusted labor share2001−2019
(1.9)
18

Table 1.2 shows the baseline and adjusted labor shares. We see that by adjusting for self-employment
and informal employment income, labor share increases by about 5.5 percentage points in Mauritius
and 4 percentage points in Senegal.
Table 1.2: Labor shares of income
Mauritius Senegal
Baseline 0.44 0.39
Adjusted (short) 0.46 0.38
Adjusted (long) 0.50 0.43
Notes: Baseline is the unadjusted labor share
of income. Adjusted (short) is the adjusted labor share of income over the sample period for
which informal income data is available. Adjusted (long) is the equivalent adjusted factor
share for the whole sample period.
Estimation of the capital stock series
To estimate capital stock series for Senegal and Mauritius, I employ the perpetual inventory methods. The first methods I use estimate both an initial level of K and an average depreciation rate.
I solve the following system of equations:
K1
Y1
= (Y
10
t=1
Kt
Yt
)
1
10 (1.10)
1
40
X
40
t=1
δK =
1
40
X
40
t=1
cc (1.11)
Kt+1 = It + (1 − δ)Kt (1.12)
for the capital stock Kt and the average depreciation rate δ, using data on value added Y , consumption of fixed capital cc, and investment I, all in constant 1980 local currency. I use this system
of 41 equations and 41 unknowns to solve for a series of Kt
.
I do the same exercise by replacing equation 1.10 with the arithmetic mean instead of the geometric
mean. Specifically, I solve:
19

K1
Y1
=
1
10
X
10
t=1
Kt
Yt
(1.13)
1
40
X
40
t=1
δK =
1
40
X
40
t=1
cc (1.14)
Kt+1 = It + (1 − δ)Kt (1.15)
Moreover, I account for the possibility that the countries went through a period of capital deepening
in the 1980s by replacing equation 1.11 with the following equation:
KT1
KT1
−
K1
K1
=
KT2
KT2
−
KT1
KT1
(1.16)
where T1 is year 1984 and T2 is year 1989, and solving it with equations 1.10 and 1.12 for Kt and
δ.
A relatively straightforward methodology is to guess an initial level of capital, and use investment
and depreciation rate data from the Penn World Table (PWT) (Feenstra et al. (2015)) to derive a
series of capital stock using the law of motion of capital.
Kt+1 = It + (1 − δt)Kt (1.17)
It follows from this equation that once an initial K1 is known, we can derive all the subsequent K
in the following time periods. The initial guess of capital matters less as we move away from the
initial time period. Therefore, although my sample period is from 1980 to 2019, I use the earliest
year data for investment and depreciation rate is available as my initial time period, guess a K1 for
that period, and derive a time series of capital. I have investment data from 1980 to 2019 from the
national accounts, and I obtained investment from 1976 and 1965 from the World Bank’s World
Development Indicators for Mauritius and Senegal respectively. I first convert the World Bank
investment data into 2008 SNA to make it consistent with the investment data from the national
accounts. I deflate the series using the GDP deflator to have them in constant 1980 LCU. I also
20

Figure 1.9: All estimated capital stock series
Notes: The different capital stock series are obtained from different methods. Guess is obtained from guessing an initial level of capital. PIM geo mean is obtained from the perpetual
inventory method with the geometric mean. PIM arith mean is obtained from the perpetual
inventory method with the arithmetic mean. And PIM cap deep is the perpetual inventory
method taking into account the possibility of capital deepening. See section 1 for details.
obtain depreciation rates for Mauritius and Senegal from 1976 and 1965, respectively, to 2017 from
the PWT, and used a linear forecasting method to forecast the depreciation rates for 2018 and
2019. For the initial level of capital, I use the capital level obtained from methods 1 and 2 in 1980,
and get the capital to output ratio in 1980. I get a ratio of 1.5. I then use the value added of
1965 and multiply it by 1.5 to obtain my guess of an initial capital stock in 1965. From there it is
straightforward to obtain a time-series of capital stock from 1965 to 2019.
For Mauritius, I do not have consumption of fixed capital data. I therefore use the depreciation rate
from the PWT and employ the perpetual inventory methods to derive an initial level of capital,
then use the law of motion of capital to derive the capital stock series, as shown by equations 1.10
and 1.12. Additionally, I use the guess method to estimating a capital stock series using equation
1.17.
21

For Senegal, I have data on gross fixed capital formation which is my investment data and consumption of fixed capital from their national accounts, and I also have depreciation rate from the
PWT. I therefore employ all the methodologies described above to compute capital stock series for
Senegal.
We can see from Figure 1.9 that I obtain very consistent estimates of capital stock series across
different methods. My preferred capital stock is the one derived from the perpetual inventory
method with the geometric mean.
Derivation of TFP Series
Figure 1.10: Total factor productivity
Notes: TFP residual for Senegal and Mauritius, using the baseline and adjusted factor shares
computed above, and deflating Y , K, and L by an aggregate deflator. TFP is normalized
such that TFP in 1980 is 1.
Once I derive the factor shares of income and estimate the capital stock series for each country, I
get the TFP residual from equation 1.5:
22

At =
Yt
Kα
t L
1−α
t
(1.18)
Figure 1.10 shows the TFP series using the baseline and adjusted factor shares. These series are
normalized so that the level of TFP in 1980 is 1. There is a drastic divergence between the two
countries TFPs. Mauritius’ TFP had an upward trend throughout the whole period, while Senegal’s
kept spiraling downwards.
Aggregate growth accounting with baseline factor shares
In this section, I discuss my results from the aggregate growth accounting, decomposing real value
added growth into TFP, capital, and labor input factor growth. Using equation 1.5, I decompose
the growth rate of real value added as follows:
gY = gA + αgK + (1 − α)gL (1.19)
where, for a variable X, using the Taylor series approximation of growth rates,
gX = log(Xt+1) − log(Xt) (1.20)
I take averages of the one-year growth rates of the variables for each decade, and for the whole
sample period. Tables 1.3 and 1.4 show the results for Mauritius and Senegal respectively. The
biggest contributor to the growth of real value added is capital inputs, accounting for 70% and 97%
of Mauritius and Senegal’s growth in real value added over the whole sample period, respectively.
These two countries are thus growing through capital investments rather than experiencing an
innovation-led growth. They have comparable labor inputs growth with Senegal having an edge
over Mauritius. However, the main factor that drives the difference between the two countries
growth experiences is TFP. Indeed, in Mauritius, TFP growth is positive between 1980 and 2019,
and in each decade, while in Senegal TFP growth is negative over the whole sample period and in
each decade except the last one, during which TFP did not grow.
23

Table 1.3: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
1981–1990 6.53 0.79 4.06 1.68
1991–2000 6.42 1.50 4.30 0.63
2001–2010 4.22 0.58 3.15 0.50
2011–2019 3.49 0.39 2.50 0.60
1980–2019 5.21 0.83 3.53 0.86
Table 1.4: Decomposition of VA-Senegal
Period Y A Kα L
1−α
1981–1990 2.63 -3.13 3.87 1.89
1991–2000 3.03 -0.94 3.11 0.86
2001–2010 3.59 -0.88 3.30 1.17
2011–2019 4.87 0.00 3.28 1.59
1980–2019 3.49 -1.27 3.39 1.37
Furthermore, I decompose real value added per working-age person by first dividing all the exponents in equation 1.5 by (1 − α) and dividing both sides of the equation by Y
α
1−α to obtain the
following:
Yt = A
1
1−α
t
(
Kt
Yt
)
α
1−α Lt (1.21)
I finally divide 1.21 by Nt
, the working-age population to express value added per working-age
person in terms of a TFP factor, a capital to value added ratio factor, and a labor factor as follows:
Yt
Nt
= A
1
1−α
t
(
Kt
Yt
)
α
1−α
Lt
Nt
(1.22)
Similarly, I take the averages of the one-year growth rates of each component for the whole sample
period and for each decade. Equation 1.22 is a nice representation because on a balanced growth
path, the growth of K
Y
and L
N
is 0, so that the growth of Y
N
is the growth of the TFP factor. The
results of this decomposition are reported in Tables 1.5 and 1.6.
The first observation we can make is that neither of these countries is on a balanced growth path, as
the capital output ratio is not constant. This decomposition allows us to estimate the contribution
of the TFP factor, capital deepening, and labor factor in the growth rate of the countries’ per capita
24

income. For Mauritius, the TFP factor and capital deepening accounted for most of the growth,
accounting for about 40% and 42% respectively. However in Senegal, there is a stark difference as
TFP factor growth is negative throughout the whole period, except the last decade when it was
0. Instead, Senegal capitalized off of its capital deepening and labor input factor for most of the
period. Between 1980 and 2019, Senegal’s capital to output ratio grew at more than twice the rate
of Mauritius’, its labor factor grew as fast as Mauritius’, but its negative TFP factor growth offset
the positive contribution of its capital and labor factor growth.
Table 1.5: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981–1990 4.71 1.79 0.96 1.96
1991–2000 4.86 3.38 1.63 -0.15
2001–2010 3.29 1.30 1.81 0.19
2011–2019 2.98 0.89 1.25 0.84
1980–2019 3.98 1.86 1.42 0.70
Table 1.6: Decomposition of VA per working-age person-Senegal
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981–1990 0.00 -7.92 5.76 2.17
1991–2000 0.07 -2.37 3.23 -0.79
2001–2010 0.87 -2.23 2.85 0.25
2011–2019 1.97 0.00 0.84 1.12
1980–2019 0.7 -3.21 3.23 0.68
This aggregate growth accounting suggests that in the aggregate, TFP and capital are the main
drivers of growth in the two countries, and TFP is at the core of the difference between the two
countries’ growth experiences between 1980 and 2019. Despite having labor inputs growth rates
very comparable to Mauritius’ and much faster capital factor growth, Senegal lags behind due to
its negative TFP growth over the sample period.
Aggregate growth accounting with adjusted factor shares
I conduct the same decompositions as above, but this time using the adjusted factor shares instead
of the baseline ones. Tables 1.7 and 1.8 show the results of decomposing the growth of the real value
added for Mauritius and Senegal respectively. The results are very similar to the ones obtained
25

using the baseline factor shares. Labor inputs growth rates are higher on average in both countries,
capital inputs growth lower, and TFP growth higher, compared to the results using the baseline
factor shares.
Table 1.7: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
1981–1990 6.53 0.98 3.66 1.89
1991–2000 6.42 1.84 3.87 0.71
2001–2010 4.22 0.82 2.84 0.56
2011–2019 3.49 0.56 2.25 0.67
1980–2019 5.21 1.07 3.18 0.97
Table 1.8: Decomposition of VA-Senegal
Period Y A Kα L
1−α
1981–1990 2.63 -3.07 3.61 2.08
1991–2000 3.03 -0.82 2.91 0.94
2001–2010 3.59 -0.78 3.08 1.29
2011–2019 4.87 0.06 3.07 1.74
1980–2019 3.49 -1.18 3.17 1.51
Similarly, the results reported in Tables 1.9 and 1.10 are qualitatively similar to the ones obtained
with the baseline factor shares.
Table 1.9: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981–1990 4.71 1.98 0.77 1.96
1991–2000 4.86 3.70 1.31 -0.15
2001–2010 3.29 1.65 1.45 0.19
2011–2019 2.98 1.13 1.01 0.84
1980–2019 3.98 2.14 1.14 0.7
Table 1.10: Decomposition of VA per working-age person-Senegal
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981–1990 0.00 -7.06 4.9 2.17
1991–2000 0.07 -1.89 2.75 -0.79
2001–2010 0.87 -1.81 2.43 0.25
2011–2019 1.97 0.13 0.72 1.12
1980–2019 0.7 -2.73 2.75 0.68
Given that the tourism and finance industries are very developed in Mauritius compared to Senegal,
26

the fast TFP and economic growth in Mauritius may be driven by these two industries. To control
for that, I conduct the same aggregate growth accounting exercises excluding the tourism industries,
excluding the finance sector, and finally excluding both the tourism and finance industries in
Mauritius. The results for these exercises are reported in Chapter One Appendix. These growth
accounting exercises show TFP and value added growth rates for Mauritius that are very similar
to the ones reported in this section. Therefore, although these two industries in the services sector
dominate the Mauritian economy, they do not fully explain the fast growth of the country.
Chapter One: Sectoral Growth Accounting Exercises
In addition to the aggregate growth accounting, I conduct a sectoral growth accounting to characterize the growth experience of the three major sectors of economic activity and to document
which sectors are responsible for the aggregate growth accounting results I obtain. The sectoral
growth accounting follows the same steps as the aggregate growth accounting, except that I assume that the aggregate economy comprises three main sectors of economic activity: agriculture,
manufacturing, and services. The agricultural sector includes all agricultural, forestry, and fishing
activities. The manufacturing sector includes mining and quarrying, manufacturing, construction,
and utilities. And the services sector includes all the other activities. I assume that each sector s
has a Cobb-Douglas production function:
Yst = AstK
αs
st L
1−αs
st (1.23)
Sectoral factor shares of income
I compute sectoral factor shares using the same methodology described above. Specifically, for each
sector s, I compute:
1 − αs =
compensation of employeess
compensation of employeess + gross operating surpluss
(1.24)
27

Similarly, I adjust for income from self-employment and informal employment as follows:
1 − αs =
compensation of employeess + other employment incomes
compensation of employeess + gross operating surpluss
(1.25)
For Mauritius, the sectoral self-employment income is derived from the micro data from 2001 to
2018. I do the same adjustments as the aggregate labor share of income described in section 1 to
get an adjusted labor share for the period 1980-2019. Similarly for Senegal, the sectoral informal
income is obtained from ANSD from 2014 to 2019, and I do the same adjustments to derive an
adjusted labor share of income for the period 1980-2019. Tables 1.11 and 1.12 show the baseline
and adjusted sectoral labor shares of income. We can see that in both countries agriculture has the
highest labor share, which can be explained by the fact that agriculture is less capital-intensive in
these countries. The manufacturing sector has the lowest labor share in Senegal, while in Mauritius
it is the services sector.
Table 1.11: Sectoral labor shares of income
Agriculture Manufacturing Services
Mauritius 0.50 0.46 0.42
Senegal 0.54 0.27 0.45
Table 1.12: Adjusted sectoral labor shares of income
Agriculture Industries Services
Mauritius 0.61 0.56 0.46
Senegal 0.54 0.31 0.49
Sectoral capital stock series
For Mauritius, I have sectoral investment data but I do not have data on capital consumption. I
therefore compute sectoral capital stock by using the aggregate depreciation rate and employing
the perpetual inventory methods to estimate an initial capital stock for each sector. I then use
the law of motion of capital to derive the sectoral time series of the capital stock. Unfortunately,
for Senegal, I do not have sectoral investment data, but only sectoral capital consumption data.
Because I do not know how this capital consumption data is computed and what it includes, I
28

derive my own simulated aggregate consumption of fixed capital, by multiplying the depreciation
rate and the aggregate capital stock obtained from the perpetual inventory method. I then derive
the sectoral capital consumption series by multiplying the simulated aggregate capital consumption
series by the sectoral shares of capital consumption from the raw data. These steps are shown in
equation 1.26. I finally divide these sectoral consumption of fixed capital by the depreciation rate
to obtain estimates of sectoral capital stock.
ccconstructed
it = δ × Kt ×
ccdata
it
ccdata
t
(1.26)
where cc is capital consumption.
Figure 1.11 shows the final sectoral capital stock series for Mauritius and Senegal. As noted earlier,
the agriculture sector is the least capital intensive sector in both countries with the lowest and
slowest growing capital stock. The services sector has the highest and fastest growing capital stock.
Figure 1.11: Sectoral capital stock series
29

Sectoral TFP series
Once sectoral factor shares and capital series are estimated, I derive the TFP residual for each
sector s as follows:
Ast =
Yst
K
αs
st L
1−αs
st
(1.27)
Figure 1.12 shows the derived TFP series using the baseline factor shares. We see that in Senegal,
TFP in all sectors has a downward trend, but agriculture’s TFP picks up after 2010. In Mauritius,
all sectors’ TFP has an upward trend except agriculture, whose TFP rises initially but has a
long-run downward trend.
Figure 1.12: Sectoral TFP- Baseline factor shares
30

Sectoral growth accounting with baseline factor shares
The first sectoral accounting exercise I conduct uses the baseline sectoral factor shares. As in the
aggregate growth accounting described in section 1, I conduct two separate decompositions. For
each sector s, I decompose the growth rate of Y into the growth rates of A, Kα, and L
1−α. Then
I conduct a second decomposition exercise by following equation 1.22. Tables 1.13, 1.15,1.14, and
1.16 show the results from the two decompositions for Mauritius and Senegal respectively.
In Mauritius, the three sectors have different patterns in the growth rates of their real value added.
The average growth rates of all sectors’ value added were high in the 1980s, but was highest in manufacturing and services. In the subsequent decades, services had the fastest growing value added,
followed by manufacturing, while the value added of agriculture started declining following the first
decade. This reveals that economic activity shifted significantly from agriculture to manufacturing
and services at the beginning of the period, and then moved to mostly services. For labor inputs,
services had the fastest growth in all decades. Both manufacturing and agriculture saw negative
growth rates in their labor inputs in the later decades, while services kept a relatively high and positive growth. There is not a significant difference in capital inputs growth across sectors. Linking
these observations on value added and labor inputs growth across sectors to the sectoral growth in
TFP paints a more complete picture. At the beginning of the sample in the 1980s, agriculture had
the highest TFP average growth rate. In the 1990s, when economic activity started shifting to the
manufacturing and services sectors, agriculture had a negative TFP growth rate, while the growth
rates of the TFP in the manufacturing and services were increasing significantly. This suggests that
economic activity and labor moved from the sector with lower productivity growth (agriculture)
to the ones with higher productivity growth (manufacturing and services). Over the whole sample
period, agriculture had the lowest TFP growth and at the same time the lowest growth in real
value added and labor.
In Senegal, the three sectors have similar value added growth rates during the whole period, suggesting a less dynamic reallocation of economic activity. Overall, value added has an upward trend
in all sectors. Only agriculture has a positive TFP growth overall. TFP growth was negative in
almost all decades in the manufacturing and services sectors. Capital and labor inputs growth is
31

positive in all three sectors throughout the whole period, but labor inputs grew negatively in the
last decade in agriculture. Capital inputs growth rates are highest in manufacturing and services,
but the negative TFP growth rates offset the positive contribution of capital. Overall, the sector
with the fastest TFP growth, agriculture, had the lowest labor inputs growth and the lowest value
added growth. Therefore, economic activity and labor are moving out of agriculture, the faster
TFP growth sector, to manufacturing and services, the slower TFP growth sectors.
Table 1.13: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
Agriculture
1981–1990 5.40 1.18 4.08 0.14
1991–2000 -0.37 -2.65 3.47 -1.20
2001–2010 -0.81 -1.92 2.82 -1.71
2011–2019 0.44 -1.46 1.95 -0.06
1981–2019 1.18 -1.21 3.11 -0.73
Manufacturing
1981–1990 7.53 0.45 4.34 2.75
1991–2000 5.03 1.42 3.45 0.17
2001–2010 2.80 1.32 2.22 -0.75
2011–2019 0.57 -1.23 1.92 -0.12
1981–2019 4.07 0.53 3.01 0.53
Services
1981–1990 6.23 1.15 3.62 1.46
1991–2000 8.29 2.14 4.62 1.53
2001–2010 5.20 0.16 3.52 1.51
2011–2019 4.51 0.85 2.74 0.93
1981–2019 6.10 1.08 3.65 1.37
Table 1.14: Decomposition of VA-Senegal
Period Y A Kα L
1−α
Agriculture
1981–1990 2.40 -2.22 2.38 2.25
1991–2000 2.64 0.05 2.08 0.51
2001–2010 2.91 1.33 1.00 0.58
2011–2019 4.28 7.21 0.05 -2.98
1981–2019 3.03 1.45 1.41 0.17
Manufacturing
1981–1990 3.37 -2.77 4.63 1.51
1991–2000 3.18 -1.21 3.00 1.4
2001–2010 3.96 -1.27 3.37 1.85
2011–2019 5.62 1.93 2.01 1.68
1981–2019 3.99 -0.90 3.28 1.61
Services
1981–1990 2.43 -4.73 4.29 2.86
1991–2000 3.10 -0.83 2.22 1.71
2001–2010 3.65 -1.19 2.88 1.96
2011–2019 4.71 -1.32 1.67 4.35
1981–2019 3.44 -2.03 2.79 2.68
32

Table 1.15: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981–1990 3.58 2.38 2.73 -1.53
1991–2000 -1.93 -5.35 7.37 -3.96
2001–2010 -1.75 -3.88 6.52 -4.39
2011–2019 -0.07 -2.94 3.50 -0.63
1981–2019 -0.04 -2.43 5.07 -2.68
Manufacturing
1981–1990 5.71 0.97 0.67 4.07
1991–2000 3.47 3.06 1.62 -1.21
2001–2010 1.86 2.85 1.55 -2.54
2011–2019 0.07 -2.65 3.47 -0.75
1981–2019 2.85 1.15 1.79 -0.09
Services
1981–1990 4.41 2.71 0.08 1.62
1991–2000 6.72 5.05 -0.36 2.03
2001–2010 4.26 0.38 1.26 2.63
2011–2019 4.00 2.00 0.33 1.68
1981–2019 4.87 2.55 0.33 2.00
Table 1.16: Decomposition of VA per working-age person-Senegal
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981–1990 -0.23 -4.09 2.36 1.51
1991–2000 -0.32 0.09 1.61 -2.02
2001–2010 0.20 2.45 -0.61 -1.64
2011–2019 1.38 13.25 -3.49 -8.38
1981–2019 0.23 2.66 0.06 -2.49
Manufacturing
1981–1990 0.74 -10.28 8.05 2.96
1991–2000 0.22 -4.50 2.49 2.24
2001–2010 1.24 -4.70 1.78 4.16
2011–2019 2.73 7.18 -7.81 3.35
1981–2019 1.19 -3.34 1.36 3.17
Services
1981–1990 -0.20 -10.59 6.61 3.78
1991–2000 0.14 -1.87 1.13 0.88
2001–2010 0.94 -2.66 1.92 1.68
2011–2019 1.81 -2.95 -2.10 6.86
1981–2019 0.64 -4.56 1.99 3.21
I conduct the same decompositions using the adjusted factors shares and the results are robust.
I include the tables of results in Chapter One Appendix. Additionally, I do the same exercises
in Mauritius excluding tourism, finance, and both, and the results are included in Chapter One
Appendix. Again, my results are robust to these sensitivity tests.
The growth accounting exercises have demonstrated that TFP growth is the principal driver of the
33

divergence in the two countries’ growth in economic activity, as well as the reallocation of resources
across the sectors. From the results, it is clear that in Mauritius, fast TFP growth drove its fast
growth in output per working-age person especially in the first two decades of the sample period,
and that both economic activity and labor moved to sectors with higher TFP growth. However in
Senegal, the negative TFP growth throughout the period offset the contribution of labor and capital
inputs, such that the country barely grew especially in the first three decades. At the same time,
there was little to no reallocation of economic activity, and labor was being reallocated to sectors
with lower TFP growth rate. These results explain why in Senegal, sectoral labor shares are shifting
while value added shares have remained relatively constant throughout the whole period. These
findings from the sectoral growth accounting exercises are puzzling if we consider only price effects.
The canonical structural transformation model developed in Herrendorf et al. (2014) predicts that
in a closed economy, over time and as countries grow, resources move from the sectors with faster
growing TFP to the slower TFP growing sectors. In other words, the correlation between sectoral
differences in labor and sectoral differences in TFP should be negative. This is established by
the fact that as a sector’s TFP increases relative to the other sectors’ TFP, its relative price
decreases. Assuming sectors are complements such that their elasticity of substitution is less than
one, the spending share of the sector with faster TFP growth decreases, and consequently, so does
its labor share. This is the opposite of what we observe in Mauritius, which nonetheless has had
a successful structural transformation. The reallocation resulting from differential TFP growth
rates and price effects however does not necessarily hold in the presence of income effects, different
capital intensities, and trade. In my next exercise, I use a multi-sector structural model with both
price and income effects, as well as different sectoral capital intensities, to better understand the
dynamics observed in Mauritius and Senegal.
Chapter One: Model of Structural Transformation and Growth
I employ a structural transformation model to map the dynamics of sectoral reallocation of the
two economies and pin down the factors that drive them. This three-sector growth model is
adopted from Herrendorf et al. (2014), with one main deviation: I use different sectoral factor
34

shares of income calibrated from the data. In this model, there are two forces driving structural
transformation: income effects and prices effects.
Household
The economy is comprised of a representative household with lifetime utility:
U =
X∞
t=0
β
t
log(Ct) (1.28)
where β is the discount factor of future consumption. The household allocates its income to investment and consumption across the three sectors: agriculture, a, manufacturing, m, and services, s.
Ct
is the composite consumption and is given by:
Ct = [ω
1
ϵ
a (cat − c¯a)
ϵ−1
ϵ + ω
1
ϵm(cmt)
ϵ−1
ϵ + ω
1
ϵ
s (cst + ¯cs)
ϵ−1
ϵ ]
ϵ
ϵ−1 (1.29)
where ωi
is sector i’s preference weight, ϵ governs the elasticity of substitution across the sectors,
c¯a and ¯cs are the non-homothetic terms and represent the household’s subsistence requirement in
agricultural goods and its endowment in services, respectively. The income and price effects driving
structural transformation are introduced through these non-homothetic terms and the elasticity of
substitution, respectively. With the non-homothetic terms, the income elasticity of the services sector is greater than one, so that as income grows over time, people consume more of its goods relative
to the agricultural and manufacturing sectors, whose income elasticities are less than and equal to
one, respectively. ϵ is assumed to be less than one, so that goods are gross complements. This
implies that as a sector’s relative price decreases, its consumption share declines. The household
owns capital that it rents to the firms and supplies its labor inelastically.
Production
The consumption goods are produced from capital k and labor n following production functions:
35

yit = ϕitk
θi
it(γ
t
init)
1−θi
(1.30)
for i ∈ {a, m, s}. ϕit is a sector specific productivity parameter that affects the level but not
the growth rate of TFP on a balanced growth path and γ
t
i
is the exogenous labor-augmenting
technology progress. Define:
Ait ≡ ϕitγ
t(1−θi)
i
(1.31)
as the total factor productivity for sectors i ∈ {a, m, s}. The production functions can therefore be
rewritten as:
yit = Aitk
θi
it n
1−θi
it (1.32)
Investment is derived from the manufacturing and services value added and produced by a perfectly
competitive representative firm with constant returns to scale and zero profits who sells It to
households at price pIt, following a Cobb-Douglas aggregator:
It =
1
(1 − η)
1−ηη
η
i
η
mti
1−η
st (1.33)
The aggregate capital follows the standard low of motion of capital:
Kt+1 = It + (1 − δ)Kt (1.34)
At the aggregate level, markets clear such that:
36

Kt = kat + kmt + kst (1.35)
1 = nat + nmt + nst (1.36)
yat = cat (1.37)
ymt = cmt + imt (1.38)
yst = cst + ist (1.39)
The investment good is the numeraire so that its price is normalized to be 1.
Competitive equilibrium
Household’s optimality conditions
The problem of the household is as follows:
max
Kt+1,Ct,cat,cmt,cst
X∞
t=0
β
t
log(Ct) (1.40)
subject to
Ct = [ω
1
ϵ
a (cat − c¯a)
ϵ−1
ϵ + ω
1
ϵm(cmt)
ϵ−1
ϵ + ω
1
ϵ
s (cst + ¯cs)
ϵ−1
ϵ ]
ϵ
ϵ−1 (1.41)
patcat + pmtcmt + pstcst + Kt+1 = (1 − δ + rt)Kt + wt (1.42)
The static optimality conditions from the household’s problem are as follows:
37

β
t
Ct
C
1
ϵ
t ω
1
ϵ
a (cat − c¯a)
−1
ϵ = λtpat (1.43)
β
t
Ct
C
1
ϵ
t ω
1
ϵm(cmt)
−1
ϵ = λtpmt (1.44)
β
t
Ct
C
1
ϵ
t ω
1
ϵ
s (cst + ¯cs)
1
ϵ = λtpst (1.45)
Raising equations 1.43, 1.44, and 1.45 to (1 − ϵ) and adding them yields the following:
β
t
Ct
= λt
[ωap
1−ϵ
at + ωmp
1−ϵ
mt + ωsp
1−ϵ
st ]
1
1−ϵ (1.46)
This equation equates the marginal benefit of an additional unit of composite consumption today
to its marginal cost. Because λt
is the marginal value of an additional unit of expenditure in period
t, the term in brackets is the price of a unit of composite consumption. Define:
Pct ≡ [ωap
1−ϵ
at + ωmp
1−ϵ
mt + ωsp
1−ϵ
st ]
1
1−ϵ (1.47)
as the price index.
Furthermore, if we multiply both sides of equations 1.43, 1.44, and 1.45 by (cat − c¯a), (cmt), and
(cst + ¯cs) respectively, and add them together, we get:
β
t
λt
= pat(cat − c¯a) + pmtcmt + pst(cst + ¯cs) (1.48)
The right hand side of equation 1.48 represents the total expenditure on consumption composite,
PctCt
. Therefore, the household budget constraint can be rewritten as follows:
patcat + pmtcmt + pstcst = PctCt + patc¯a − pstc¯s (1.49)
Combining equations 1.42 and 1.49, we can rewrite the household problem:
38

max
Ct,Kt+1
X∞
t=0
β
t
log(Ct) (1.50)
subject to
PctCt + Kt+1 = (1 − δ + rt)Kt + wt − patc¯a + pstc¯s (1.51)
The dynamic optimality conditions for the household’s problem:
λt = λt+1(1 − δ + rt+1) (1.52)
β
t
Ct
= λtPct (1.53)
β
t+1
Ct+1
= λt+1Pct+1 (1.54)
This yields the following Euler equation:
Pct+1Ct+1
PctCt
= β(1 − δ + rt+1) (1.55)
Taking the ratios between 1.43, 1.44, and 1.45 gives:
pat(cat − c¯a)
pmtcmt
=
ωa
ωm
(
pat
pmt
)
1−ϵ
(1.56)
pat(cat − c¯a)
pstcst + ¯cs
=
ωa
ωs
(
pat
pst
)
1−ϵ
(1.57)
pmtcmt
pst(cst + ¯cs)
=
ωm
ωs
(
pmt
pst
)
1−ϵ
(1.58)
We can see from equations 1.56 to 1.58 that if ϵ is less than 1, the consumption share of one good
relative to another, pitcit, increases as its price increases. These consumption shares are closely
related to value added shares. This shows how the elasticity of substitution across sectors, governed
by ϵ, is one factor driving sectoral reallocation through price effects.
39

Firm’s optimality conditions
For each sector i ∈ {a, m, s}, the representative firm solves the following profit maximization
problem:
max
kit,nit
pitAitk
θi
it n
1−θi
it − kitrt − nitwt (1.59)
The optimality conditions from this problem are:
rt = θipitAit(
nit
kit
)
1−θi
(1.60)
wt = (1 − θi)pitAit(
kit
nit
)
θi
(1.61)
For two different sectors i and j, taking the ratio between the rental rate yields:
θipitAit(
nit
kit
)
1−θi
θjpjtAjt(
njt
kjt
)
1−θj
= 1 (1.62)
Simplifying the above equation gives this expression for relative prices:
pit
pjt
=
θj
θi
Ajt
Ait
(
kit
nit
)
1−θi
(
njt
kjt
)
1−θj
(1.63)
Equation 1.63 shows the inverse relationship between prices and productivity, such that for sector
i, as its TFP Ait increases relative to sector j’s, it relative price decreases.
Furthermore, taking the ratio between 1.60 and 1.61 gives us the capital to labor ratio for each
sector i ∈ {a, m, s}:
kit
nit
=
θi
1 − θi
wt
rt
(1.64)
The investment-producing firm solves the following problem:
40

max
imt,ist
1
(1 − η)
1−ηη
η
i
η
mti
1−η
st − pmtimt − pstist (1.65)
The first order conditions of this problem yield the following:
η
1 − η
ist
imt
=
pmt
pst
(1.66)
Similarly, there are price effects in the investment sector such that as the price of the services
sector pst increases relative to the manufacturing sector’s, the relative investment spending from
the services sector decreases.
Competitive equilibrium definition
A competitive equilibrium is factor prices {wt
, rt}, goods prices {pit, Pct}, and allocations {cit, nit, kit, imt, ist}
for i ∈ {a, m, s} such that given an initial level of capital K0 and a series of sectoral productivities
{Ait}:
• given prices, {cit, nit, kit} solve the household’s utility maximization problem (equations 1.51,
1.55, 1.56, 1.57, 1.58)
• given prices, {cit, nit, kit, imt, ist} solve the firms’ profit optimization problems (equations 1.60,
1.61, 1.63, 1.64, 1.66)
• markets clear
Chapter One: Calibration
To calibrate the model, I use sectoral data on household consumption, labor and capital income,
and value added. I also need aggregate data on capital and GDP to calibrate the initial capital
stock in each country. I obtain data on sectoral household consumption from the countries’ inputoutput tables. For Mauritius, the earliest year for which I have an input-output table is 1992.
41

As such, I take 1992 as my base year for Mauritius. For Senegal however, I have input-output
tables from 1980, so 1980 is my base year. The set of parameters I need to calibrate are the
following: a) share parameters of the consumption and investment functions, calibrated to target
base year input-output data, and sectoral non-homothetic parameters, targeted to match baseyear consumption data; b) elasticities, taken as independent estimates from the literature; c) initial
conditions: aggregate (capital stock) and sectoral (value added and employment and capital income
shares), targeting base-year capital-output ratios and value added, employment, and capital-income
shares in the data. I calibrate the model to match base year data in Mauritius and Senegal. All
variables in the data are normalized such that base year total value added equals 100. Base year
prices are set equal to 1.
I assume the same ratios of agriculture and services consumption non-homothetic terms to their
consumption found by Herrendorf et al. (2013) for the US. They computed the values for 1947 and
2010. Following Kehoe et al. (2018), I use a linear extrapolation method to compute the values
for Mauritius and Senegal for 1992 and 1980, my base years. The results are shown in Table 1.17.
From these ratios, I find the values for ¯ca and ¯cs for the two countries by multiplying them by the
respective sectoral consumption in the specific year. I report the non-homothetic terms I find for
Mauritius and Senegal for 1980 and 1992 in Table 1.18.
Table 1.17: Non-homothetic parameters relative to final consumption expenditure- extrapolations
1980 1992
c¯a
ca
-0.71 -0.67
c¯s
cs
0.93 0.73
Table 1.18: Non-homothetic parameters-in terms of GDP of base year
Mauritius 1992 Senegal 1980
c¯a -2.49 -9.10
c¯s 40.62 35.55
For the sectoral consumption weights ωi
, I use the optimality conditions from the household’s
problem given by equations 1.56 and 1.57, the constraint that the sum of the weights equals 1, and
data on nominal sectoral consumption in the base year to solve for ωa, ωm, and ωs. Table 1.19
shows the results.
42

Table 1.19: Consumption weights
Mauritius 1992 Senegal 1980
ωa 0.05 0.17
ωm 0.18 0.25
ωs 0.77 0.58
For the sectoral capital shares, I divide the gross operating surplus by its sum with the compensation
of employees in the base year. The results are reported in Table 1.20.
Table 1.20: Capital shares
Mauritius 1992 Senegal 1980
θa 0.46 0.51
θm 0.47 0.77
θs 0.56 0.47
Finally, I estimate η for the two countries from their input-output tables, by taking the ratio of
investment that is derived from manufacturing sector. I find η to be around .8 for both countries.
For the rest of the parameters, I follow standard literature and set δ = .05 (Kehoe et al. (2018)),
ϵ = .89 (Herrendorf et al. (2013)), and β = .98 (Verma (2012)).
Chapter One: Quantitative Results
I solve a general equilibrium of the model above using sectoral TFPs from the growth accounting
exercises as exogenous shocks and by calculating a gradual convergence of the economy to a steady
state equilibrium in which TFP in each sector is constant. I first compare the value added and
employment shares from the benchmark model with the data to evaluate the goodness of fit of the
model, and then do a few counterfactual exercises to answer the question: what would the two
countries’ growth trajectory be like if they had alternate productivity and technology?
Benchmark model
The sectoral value added and employment shares obtained from the model are plotted in Figures
1.13, 1.14, 1.15, and 1.16, along with the data. In Mauritius, the model matches very closely the
43

sectoral value added shares, and it does a good job at tracking the trends in the sectoral employment
shares. However, there is a level discrepancy between the model and the data in employment shares,
as the model predicts a higher manufacturing labor share than observed in the data. In Senegal,
the model predicts a slight decline in the manufacturing value added share. Again, there is a
discrepancy between the labor shares of the model and data in Senegal, but the trends in the data
are relatively well matched by the model. To formalize the goodness of fit of the model, Tables 1.21
and 1.22 report the changes in the sectoral value added and employment shares observed in the
data and predicted by the model. In Mauritius, the model’s directions of change are in line with the
data, and the magnitudes of the changes are very close to the data.1 However, in Senegal, there is
a lesser goodness of fit of the model as the model moves resources out of the manufacturing sector
in favor of the services sector, while in the data resources move out of the agricultural sector.2
Figure 1.13: Value added shares-data vs baseline model-Mauritius
1For the employment share, because of the level discrepancy between the data and the model, I compute the
change between 1993 and 2019.
2For the employment share, because of the level discrepancy between the data and the model, I compute the
change between 1981 and 2019.
44

Figure 1.14: Employment shares-data vs baseline model-Mauritius
Figure 1.15: Value added shares-data vs baseline model-Senegal
45

Figure 1.16: Employment shares-data vs baseline model-Senegal
Table 1.21: Changes in VA and employment shares - model vs data - Mauritius
Statistic Sector Data Baseline Model
VA share
Agriculture -0.077 -0.0053
Manufacturing -0.13 -0.17
Services 0.21 0.18
Employment share
Agriculture -0.073 0.00
Manufacturing -0.16 -0.19
Services 0.23 0.19
46

Table 1.22: Changes in VA and employment shares - model vs data - Senegal
Statistic Sector Data Baseline Model
VA share
Agriculture -0.054 0.042
Manufacturing 0.052 -0.090
Services 0.0026 0.048
Employment share
Agriculture -0.022 0.046
Manufacturing 0.0019 -0.13
Services 0.020 0.087
Counterfactual exercises
I use the model to do some counterfactuals to see how the two economies would have behaved
between 1980/ 1992 and 2019 had they experienced different productivity growth or technologies.
In the first counterfactual exercise I conduct, I simulate Senegal’s structural transformation and
growth trajectory had it had the same sectoral TFP growth as Mauritius. I therefore assume
Senegal’s three sectors to have the TFP growth of Mauritius’ sectors, between 1980 and 2019.
Figure 1.17 depicts the sectoral value added share of Senegal from this counterfactual. We can
see that with faster sectoral TFP growth and the services sector having the fastest growth TFP,
resources move out of the agriculture and manufacturing sectors to the services sector. Therefore,
even in Senegal, conditional on faster TFP growth, the services sector experiences a faster rise in
its share of the economy despite having the highest TFP growth rate.
In the second counterfactual exercise, I test whether the economies would have behaved differently
if their three sectors had the same technology. I therefore assume that in each country, the three
sectors have the same capital share of income, specifically that of agriculture. In Mauritius, the
agricultural sector has the smallest capital intensity but the three sectors have very similar capital
intensities, while in Senegal the manufacturing capital share of income is much higher than that
of agriculture and services. By setting the capital share of income of each sector in each country
equal to the capital share of income in the agricultural sector, I therefore significantly lower the
manufacturing capital intensity in Senegal. Figures 1.18 and 1.19 show the value added shares
47

Figure 1.17: Value added shares-data vs baseline vs counterfactual 1-Senegal
Notes: Counterfactual 1: SEN has the same sectoral TFP growth as MUS.
from this exercise. The predictions from this counterfactual are largely similar to the ones from
the baseline model. In Senegal, we can notice a faster decline of the manufacturing sector to the
benefit of the agriculture sector. A smaller capital intensity in the manufacturing sector therefore
led to a smaller share of the sector in the economy, against a higher share of agriculture, the sector
with now a relatively higher capital intensity.
In the third and last counterfactual, I investigate how Senegal’s sectoral composition would have
looked like if it had the same sectoral technology as Mauritius. I therefore assume the sectoral
capital shares of income in Senegal to be the same as those in Mauritius. The results are depicted
in Figure 1.20. While in Senegal the manufacturing sector has, by far, the highest capital share
of income, in Mauritius, the services sector has a slightly higher capital share of income than the
manufacturing sector. We can see from the results that there is a faster decline of the manufacturing
share of the economy, to the benefit of the services sector. Again, these results suggest that higher
capital intensity leads to a higher and faster growing share of a sector in the economy.
In table 1.23, I report the average annual growth rates of Mauritius and Senegal’s total value added
from the baseline model and the different counterfactual exercises. In the baseline model, Mauritius
grew on average at 3.8 percent between 1992 and 2019, compared to an average growth of 2 percent
48

Figure 1.18: Value added shares-data vs baseline vs counterfactual 2-Mauritius
Notes: Counterfactual 2: all sectors have the same capital intensity-that of agriculture.
Figure 1.19: Value added shares-data vs baseline vs counterfactual 2-Senegal
Notes: Counterfactual 2: all sectors have the same capital intensity-that of agriculture.
for Senegal in the same time period. In the first counterfactual, when Senegal has the sectoral TFP
growth of Mauritius, the average annual growth rate of Senegal total value added is 9 percent from
1980 to 2019, and 6 percent from 1992 to 2019. If Senegal had the same sectoral TFP growth as
Mauritius, it would have grown about two times faster between 1980 and 2019 compared to the
49

Figure 1.20: Value added shares-data vs baseline vs counterfactual 3-Senegal
Notes: Counterfactual 3: SEN has the sectoral capital intensity of MUS.
baseline. Additionally, it would have grown 1.5 times faster than Mauritius between 1992 and 2019.
Table 1.23: Average annual growth of total VA
Baseline Counter. 1 Counter. 2 Counter. 3
Mauritius (1992–2019) 3.77 2.87
Senegal (1980–2019) 3.94 9.25 1.6 1.72
Senegal (1992–2019) 2.12 6.35 0.25 .26
Overall, the model’s predictions are in line with results from the empirical accounting. In the
benchmark model, resources are being reallocated to the services sector in Mauritius although the
services sector has the fastest growing TFP. Additionally, in the first counterfactual with Senegal
having the sectoral TFP growth of Mauritius, the services sector’s share of the economy increased
relatively faster. Therefore, although goods are assumed to be gross complements in preferences, the
model predicts a reallocation of resources towards the sector with faster decline in relative prices
when countries grow fast. This hints that income effects dominate price effects in the resource
reallocation across sectors and begs the question: to what extent do prices effects drive structural
transformation in these economies? To answer this question, I isolate the role that income effects
play in resource reallocation, relative to price effects, by solving the model economy without income
effects.
50

Sensitivity analysis - no income effects
In this analysis, I set the non-homothetic terms ¯ca and ¯cs to be zero such that there are no
income effects. That way, the only effects driving structural transformation in the model are prices
effects. The results are shown in Figures 1.21, 1.22, 1.23, and 1.24. We can see that despite very
differential sectoral TFP growth in Mauritius, the model with no income effects barely predicts any
reallocation at all. Similarly in Senegal, the model without income effects predicts constant sectoral
value added and employment shares. Income effects are therefore a stronger predictor of structural
transformation than price effects in these economies. In the benchmark model that matches the
data well, Mauritius’ structural transformation is driven by the fact that the country is growing
fast with its positive TFP growth. On the other hand, in Senegal, the negative TFP growth implies
an economy that is not growing as fast, and thus resources are not being reallocated across sectors
even though the sectors have differential TFP growth. Therefore, for a structural transformation in
these economies, it matters more to have an overall fast TFP growth than differential sectoral TFP
growth. Indeed, when a country is as poor as Senegal, income effects are relatively important for
structural change compared to those in a rich country. Thus, when TFP growth at the sectoral level,
and especially in the larger sectors, is low and/or negative, there is little structural change because
aggregate income growth is low. To elicit aggregate income growth, (all) sectoral TFP growth rates
need to be fast enough. This implies that for Senegal to experience structural transformation, it
needs to focus on policies and reforms that can boost its TFP growth in all sectors.
51

Figure 1.21: Value added shares-data vs baseline vs no income effects -Mauritius
Figure 1.22: Labor shares-data vs baseline vs no income effects - Mauritius
52

Figure 1.23: Value added shares-data vs baseline vs no income effects - Senegal
Figure 1.24: Labor shares-data vs baseline vs no income effects-Senegal
Chapter One: Reforms and Policy Discussion
From the baseline model and the counterfactual experiments, I have provided evidence that the
differential sectoral TFP growth rates do not matter as much as the aggregate growth of the
53

country. Through this lens, for Senegal to experience some structural transformation, it needs to
foster positive TFP growth in all sectors, without necessarily prioritizing one sector, such as the
manufacturing one, over the others.
TFP levels and growth rates off long-run paths are understood to reflect country-specific policies,
practices, and institutions. Some distinguishable policies Mauritius has implemented date back to
the 1970s, during the Washington Consensus following the debt crisis in many developing countries
(Mistry (1992)). Mauritius’ structural reforms were centered around one main strategy: private
sector expansion to promote export diversification. In the 1970s, Mauritius established export
processing zones (EPZs) to give firms producing for export markets preferential treatment, such as
duty free access for imported inputs and tax incentives (Subramanian and Roy (2001), Woldekidan
(1992)). The goal of the EPZs was to promote export of manufacturing goods such as sugar and
textiles. They led to high production and exports of these goods. These policies are reflected in
the data given that in the 1980s, Mauritius was already fairly industrialized with a manufacturing
sector accounting for almost 40 percent of total value added against only 20 percent in Senegal.
Moreover, they have contributed to the fast increase of aggregate productivity as the EPZs were
characterized by fast productivity growth (Subramanian and Roy (2001)).
On the other hand, Senegal’s effort to position itself as an exporter in the international market
failed, as the country had an import substitution policy. In the 1970s, it enacted an importsubstitution policy that was based on high tariff rates and export subsidies (Goldsbrough et al.
(1996), Ciss´e et al. (2016)). These measures limited access to intermediate inputs and therefore
had perverse effects through lost competitiveness due to low productivity, thus having a negative
impact on export performance (Annabi et al. (2005), Ciss´e et al. (2016)).
It is clear that these reforms must have played a role in the current economic landscapes of the
two countries, specifically their TFP levels and growth rates. Mauritius’ fast TFP growth must
have been driven, at least partially, by its focus on facilitating domestic firms to have access to
inputs and export markets. A major obstacle to Senegal experiencing fast TFP growth has been
the difficult access to imported inputs. Therefore, policies to boost Senegal’s TFP growth can be
centered around facilitating access to intermediate inputs, especially for exporting firms. Since
54

2010, its agricultural sector’s TFP has been rising rapidly and this is partially attributable to its
GOANA (Great Agricultural Offensive for Food and Abundance) plan. The GOANA plan, launched
in 2008, aimed to foster food production in the country through greater access to agricultural
inputs and equipment (ACET (2013)). This demonstrates that with strong policies to enhance
sectoral production through access to intermediate inputs, Senegal can boost the productivity of
a sector. Thus, Senegal can push for these policies to accelerate the TFP growth in services and
manufacturing and maintain the recent boom in TFP growth in agriculture.
Chapter One: Conclusion
Long standing economic evidence across countries suggests that structural transformation and
growth go hand in hand, and contemporary trends show different patterns of structural transformation in developing countries. In this paper, I study the disparate structural change and growth
experiences in Mauritius and Senegal, two sub-Saharan African countries that are otherwise similar.
I first conduct aggregate and sectoral growth accounting exercises and find that TFP is the main
driver of the faster growth in Mauritius. At the sector level, economic activity is reallocated to
faster growing TFP sectors in Mauritius while in Senegal it is being reallocated to slower growing
TFP sectors. In light of these results, I develop and calibrate a multi-sector structural transformation model. I solve for a general equilibrium of the model for Mauritius and Senegal taking sectoral
TFPs as exogenous shocks and assuming different sectoral factor shares calibrated from the data.
Overall, the model matches the data on sectoral value added shares well, and it matches the trends
in sectoral employment shares.
I further conduct counterfactual exercises to explore the alternate sectoral composition and growth
trajectory of these countries had they had different productivity and technology. I find that Senegal
would have grown 2.5 times faster between 1980 and 2019 if it had the sectoral TFP growth of
Mauritius. The counterfactual exercises also demonstrate that in these economies, labor is always
being reallocated to the services sector, even when the latter has the fastest TFP growth and despite
goods being gross complements. Isolating income effects from price effects reveals that the rise of
the services sector follows from income growth, such that income effects are the main drivers of
55

structural transformation in these countries. In light of these findings, countries like Senegal need
to focus on reforms that boost their aggregate TFP growth to promote structural transformation.
This project has a few limitations, all of which relate to the structural model I use. In my model, I
abstract from population growth, government, and trade, all of which are important components of
the two economies. A natural extension of this model is therefore to assume open economies with
government and population growth. This is particularly important in accounting for the role of
trade in structural transformation and growth. Moreover, while my paper underscores the centrality
of TFP in explaining divergent growth, the specific determinants of TFP in these countries remain
unexplored. My discussion in section 1 highlights the non-negligible role the structural reforms of
the 1970s around trade openness and industrialization played in the successful industrialization and
fast growth in Mauritius compared to Senegal. This emphasizes the importance of including trade
and an intermediate input structure in the model. Future research can shed light on the impact of
such reforms in these countries on fostering fast positive TFP growth using such a model.
56

Chapter 2
Misallocation and Financial Constraints Among Firms in Sub-Saharan
Africa
Chapter Two: Introduction
If factors of production such as labor, capital, and intermediate inputs are concentrated in less
productive firms at the expense of more productive ones, then these more productive firms will not
operate to their fullest potential while the less productive ones will use more resources than they
optimally would, therefore creating inefficiencies. These distortions result in lower than optimal
firm-level productivity and output, thus reducing aggregate productivity and output (Hsieh and
Klenow (2009), Restuccia and Rogerson (2008)). In Africa in particular, firms and their productivity are an important part of economic research. One of the main economic issues in most African
countries is the high level of youth unemployment, which is estimated at almost 21% in 2021 by the
International Labor Organization.1 This is due to the very young population in Africa, where in
most countries, more than 60% of the population is below the age of 25.2 Small and medium enterprises account for about 90% of all businesses in Africa and create about 80% of jobs.3 Therefore,
these small firms are an integral part of the solutions against the high levels of youth unemployment
across the continent. Misallocation however can be a great barrier to firms’ growth. As mentioned
above, if productive firms use less than optimal resources for their operation due to inefficiencies,
1
ILO: ”Global Employment Trends for Youth 2020: Africa” https://ilo.org/wcmsp5/groups/public/—
dgreports/—dcomm/documents/briefingnote/wcms737670.pdf
2
”Africa’s Youth: Action Needed Now To Support the Continent’s Greatest Asset”
https://mo.ibrahim.foundation/sites/default/files/2020-08/international-youth-day-research-brief.pdf
3
”Why SMEs are key to growth in Africa” https://www.weforum.org/agenda/2015/08/why-smes-are-key-togrowth-in-africa/
57

then they do not employ as many workers as they would have otherwise. Therefore, understanding
firms’ dynamics and productivity, and the main drivers of the distortions they face, is crucial in
the effort to tackle Africa’s high youth unemployment.
There are many factors that lead to misallocation in an economy. Some of these factors stem from
government intervention, such as taxes and subsidies, and labor regulations. Particularly, if the
government taxes (subsidizes) firms with high (low) productivity, then highly productive firms will
be constrained while low productive firms expand. Other factors are institutional challenges, such
as corruption and political instability.
The goal of this paper is to provide evidence on the link between institutional obstacles and distortions, with a closer look at financial constraints. I use a misallocation model following Hsieh
and Klenow (2009) with capital, labor, and intermediate inputs as factors of production. Calibrating the model using firm-level data from 12 sub-Saharan African countries, I derive measures of
capital, labor, and output distortions. I then empirically test the relationship between these measures of misallocation and the obstacles firms face, namely in transportation, access to land, access
to finance, corruption, tax rates, labor regulations, inadequately educated workforce, and political instability. I first use within- and cross-country firm-level OLS regressions with industry and
country fixed effects of these measures of misallocation on the obstacles, to unveil the relationship
between them. Preliminary results suggest that financial constraints are the main determinants
of distortions, especially output distortions, in most countries considered in my study. I then focus on financial constraints and run, again, within-country firm-level OLS regressions of the three
measures of misallocation on financial constraints and a vector of controls, namely size, age, imports, exports, region, whether the firm is in an export processing zone, and the percentage of the
firm owned by the government. Additionally, I investigate the role that managers’ education and
experience play in mitigating (or otherwise) distortions, by controlling for the managers’ schooling
years and years of experience, interacted with the measure of financial constraints.
I find from the cross-country analysis that financial constraints statistically significantly increase
output distortions even after controlling for some confounding factors. I further find that size is the
main channel through which financial constraints affect output misallocation. Specifically, smaller
58

firms are more financially constrained, and even after controlling for the same degree of financial
obstacles, smaller firms still face higher distortions. In other words, smaller firms have higher levels
of misallocation relative to bigger firms, holding the severity of financial constraints constant,
meaning that size exacerbates the negative effect that financial constraints have on misallocation.
Moreover, there are lower distortions in export processing zones, which are industrial areas in which
duty-free imported raw materials are processed for export. This result suggests that the policies
and regulations in these zones are efficient in lowering size distortions. The within-country analysis
suggests that financial constraints are determinants of output misallocation in three countries out
of the twelve in my sample. The weak results in most countries may be due to the small sample
sizes. In addition, there is evidence that managers’ education and experience decrease output
misallocation stemming from poor access to financing. Due to the possibility of bias of my results,
I run some robustness checks, notably by using bootstrap standard errors and excluding size from
my control variables. My baseline results are robust to such corrections, and these robustness
checks strengthen my results.
I include a suggestive model of financial constraints following Midrigan and Xu (2014), ZetlinJones and Shourideh (2017), and Buera et al. (2011) in Chapter Two Appendix to illustrate the
growth dynamics of firms when they are financially constrained. The equations characterizing the
equilibrium of the model suggest that firms that are financially constrained use less factor inputs
and therefore grow more slowly.4
The rest of the paper is structured as follows. I review the literature in section 2, outline the model
of misallocation in section 3, and describe the main empirical results in section 4. I explore the
role of managerial expertise in section 5, run robustness checks in section 6, explore the different
sources of financing of firms in section 7, and conclude in the remainder.
4Unfortunately, I was not able to calibrate the model using data from an African country because it requires
firm-level panel data and I have not found adequate data for any sub-Saharan African country.
59

Chapter Two: Related work
Misallocation Literature
There is a growing body of literature exploring misallocation of resources among firms both in
developed and developing countries. Particularly, since the seminal paper by Hsieh and Klenow
(2009), there has been an increasing interest in analyzing the role that misallocation plays in crosscountry differences in TFP and economic outcomes.
Several studies have demonstrated the significant and negative effects of misallocation on firm size,
aggregate TFP, and GDP in various countries such as China, India, Italy, Portugal, and Mexico
(Bento and Restuccia (2016), Boehm and Oberfield (2020), Calligaris (2015), Hsieh and Klenow
(2009), Dias et al. (2016), Restuccia and Rogerson (2008), Baqaee and Farhi (2020), Misch and
Saborowski (2018)). Misallocation across firms within the same industries has been argued to be
responsible for as much as 80% of TFP losses in Italy (Calligaris (2015)), 30-50% of TFP losses in
China, and 40-60% in India (Hsieh and Klenow (2009)). Similar evidence is also found in African
countries. Particularly, an optimal reallocation of resources across firms would arguably result in
aggregate productivity gains of 30% in Cˆote d’Ivoire and as much as 160% in Kenya (Cirera et al.
(2020)). The growth of misallocation over time and its effects on TFP and GDP have also been
documented in Europe, where the degree of misallocation has been increasing over the years (Dias
et al. (2016), Gopinath et al. (2017)). For instance, in Portugal, over the period 1996-2011, it
has been found that efficient allocation of resources would have increased gross output by 17%
in 1996 and 28% in 2011 (Dias et al. (2016)). Misallocation accounts for a large proportion of
TFP differences across countries and its magnitude is growing in some countries (Bartelsman et al.
(2013)).
Given the apparent importance of misallocation in aggregate productivity and GDP, it is relevant
to investigate the causes of misallocation, both within country and across industries, and across
countries. Several studies have shed light on these drivers of misallocation. Such drivers include
policies, obstacles firms face to optimally run their operations, and firm-level frictions such as adjustment costs. Particular attention has been given to capital misallocation, which stems mostly
60

from adjustment costs and informational frictions, and idiosyncratic factors that affect firms’ investment decisions such as unobserved heterogeneity in markups and production technologies (David
and Venkateswaran (2019)). For example, Bau (2020)’s analysis of India’s liberalization of foreign
capital shows that policies restricting the use of production factors contribute greatly to misallocation. Similarly, weak contract enforcement which signals weak rule of law has been argued cause
misallocation and thus lower aggregate productivity (Boehm and Oberfield (2020)). Management
practices, especially as they relate to the managers’ education and the delegation of decision making
have been documented to be main sources of low productivity among firms in developing countries
(Bloom and et al. (2010)).
In African countries, there is some evidence that points to financial frictions as the main source of
misallocation of resources across firms (Kalemli-Ozcan and Sørensen (2012), Le´on-Ledesma (2016),
Cirera et al. (2020)). Firms that face more severe access to finance obstacle have much higher
marginal products of capital suggesting inefficiently low capital use. Other important sources of
misallocation include trade regulations, the functioning of courts, crime, and corruption. Trade
openness can lead to deeper inefficiencies as highly subsidized firms export and produce even more
(Bai et al. (2020)), and institutional obstacles such as bad functioning of courts and corruption
present obstacles for firms to optimally use inputs and organize production (Boehm and Oberfield
(2020), Le´on-Ledesma (2016)). All these constraints have been documented to prevent firms from
growing in productivity and size, and this stagnation has characterized firms in Africa for the past
six decades at least (Bloom and et al. (2010)).
Little research has focused on misallocation in sub-Saharan African countries. Of the papers cited
above (Kalemli-Ozcan and Sørensen (2012), Le´on-Ledesma (2016), Cirera et al. (2020)), only one
has focused on sub-Saharan Africa (Cirera et al. (2020)), specifically on Kenya, Cˆote d’Ivoire,
Ghana, and Ethiopia. My study not only extends this analysis to a broader set of African countries,
but it also generalizes extant analyses by including intermediate inputs as a factor of production.
All studies that look at African countries ignore intermediate inputs use at the firm level, and
in my analysis, I find that intermediate inputs are substantially used by firms as factor inputs
and size distortions are significantly larger when intermediate inputs are taken into account. In
addition, I find evidence that financial constraints increase intermediate inputs distortions relative
61

to capital and labor misallocation in at least three African countries. Furthermore, I develop a
model of financial frictions following Midrigan and Xu (2014), Zetlin-Jones and Shourideh (2017),
and Buera et al. (2011) to underline the sources of limited financing and analytically suggest the
dynamics of limited financing and firm productivity growth.
Financial Constraints and Firm Growth Literature
Financial constraints have been documented to be a barrier to firms’ productivity and output
growth and thus a source of misallocation, as reviewed by Buera et al. (2015). They have also been
argued to limit firms’ ability to innovate, grow, and invest in their capital stock, and this in turn
leads to low capital levels and lower TFP (Bah and Fang (2015), Duval et al. (2019), Goyette and
Gallipoli (2015), Zetlin-Jones and Shourideh (2017), Guner et al. (2008)). In African countries, it
has been found that low financial development contributes to the poor performance of firms and
prevents them from growing, leading to a high concentration of smaller firms (Bah and Fang (2015),
Goyette and Gallipoli (2015)).
The extent to which financial constraints constitute an obstacle to firms’ growth also depends on the
firms’ own financing capacity. Firms experiencing high and positive productivity shocks can grow
out of their financial constraints by using their own funds to finance their use of capital (Midrigan
and Xu (2014), Lian and Ma (2021), Moll (2014)). The aim of this paper is not only to unveil the link
between misallocation and financial constraints, but to also study more closely the extent to which
African firms are able to generate enough revenues and relax their borrowing constraints. These
distortions related to financial constraints are most important in the manufacturing sector, where
firms’ attainment of large scale is relatively important (Buera et al. (2011)). As a result, studying
the extent to which financial constraints limit firms’ production and growth and the degree to which
these firms can relax them through enough productivity growth is essential in understanding the
potential for African countries to grow their manufacturing sector and industrialize their economies.
62

Chapter Two: Model of Misallocation
Final Output
Final output in the economy is produced combining the output Ys of S manufacturing industries
using a Cobb-Douglas production function:
Y =
Y
S
s=1
(Ys)
θs
(2.1)
for S industries, where PS
s=1 θs = 1. The final output production takes place in a perfectly
competitive market, leading to a profit maximizing problem as follows:
max
Ys
P Y −
X
S
s=1
PsYs (2.2)
where Ps is the price of industry s’ output, Ys, and P is the price of the final output, Y , which I
consider to be the numeraire, thus P = 1. Solving this profit maximization problem yields:
PsYs = θsP Y (2.3)
Industry-Level Output
The manufacturing sector comprises S industries, and each industry produces an output that is a
CES aggregate of the differentiated products produced by the firms within the industry. Specifically,
the industry-level production function is:
Ys = [X
I
i=1
Y
σ−1
σ
si ]
σ
σ−1 (2.4)
for I firms, where σ governs the elasticity of substitution between different varieties. Assuming free
entry and monopolistic competition, profit maximization at the industry level yields the following
63

inverse demand functions:
Psi = PsY
1
σ
s Y
−1
σ
si (2.5)
Firm-Level Output
Firms produce output using three inputs, namely capital K, labor L, and intermediate inputs M,
using a CES aggregator as in Atalay (2017):
Ysi = Asi[(1 − µs)
1
ϵm ((Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
)
ϵm−1
ϵm + µ
1
ϵm
s M
ϵm−1
ϵm
si ]
ϵm
ϵm−1 (2.6)
where Asi is the firm’s TFP, αs and µs govern the firms’ usage of capital, labor, and intermediate inputs, and ϵm is the elasticity of substitution between the capital and labor bundle, and
intermediate inputs. I allow for firm-level capital, labor, and output distortions, τksi, τlsi, and τysi
respectively. The output distortions affect all factors of production by increasing their marginal
revenue products by the same proportion. The capital and labor distortions, on the other hand,
increase the marginal revenue products of capital and labor respectively, relative to intermediate
inputs’. Therefore, firms that face higher capital (labor) distortions are constrained in their use
of capital (labor) relative to their use of intermediate inputs, and will have higher marginal revenue products of capital (labor) relative to intermediate inputs. Firms therefore solve this profit
maximization problem:
max
Ksi,Lsi,Msi
(1 − τysi)PsiYsi − (1 + τksi)RsKsi − (1 + τlsi)WsLsi − ZsMsi (2.7)
where Rs is the rental rate of capital, Ws the wage rate, and Zs the unit cost of intermediate
inputs in sector s. Solving this profit maximization problem yields the following expressions for the
64

distortions:
1 − τysi =
σ
σ − 1
ψZsM
1
ϵm
si
PsiYsiµ
1
ϵm
s
(2.8)
1 + τksi = ( αs
1 − αs
)
1−αs
1 − µs
µs
1
ϵm σ
σ − 1
M
1
ϵm
si Zs(
Lsi
Ksi
)
1−αs
(( Ksi
αs
)
αs (
Lsi
1−αs
)
1−αs )
1
ϵm Rs
(2.9)
1 + τlsi = (1 − αs
αs
)
αs
(
1 − µs
µs
)
1
ϵm
σ
σ − 1
M
1
ϵm
si Zs(
Ksi
Lsi
)
αs
(( Ksi
αs
)
αs (
Lsi
1−αs
)
1−αs )
1
ϵm Ws
(2.10)
where
ψ = (1 − µs)
1
ϵm ((Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
)
ϵm−1
ϵm + µ
1
ϵm
s M
ϵm−1
ϵm
si (2.11)
In my analysis, I will use the measures of distortions in equations 2.15-2.10 to predict their determinants. As mentioned above, firm-level TFPs partly predict aggregate TFP and firm-level
misallocation lowers aggregate TFP, which is the most important consequence of misallocation as
it drives cross-country differences in productivity and growth. The use of intermediate inputs and
the CES production function does not make the mapping from firm-level TFPs to aggregate TFP as
straight forward as in a Cobb-Douglas production function. In this paper, I have not explored the
consequences of firm-level misallocation on aggregate TFP as I have not derived a transformation
to link firm-level and aggregate TFPs.
Chapter Two: Misallocation in Sub-Saharan Africa- An Empirical
Analysis
The first part of my analysis consists of quantitatively and empirically analyzing the degree of misallocation in sub-Saharan African countries and estimating the extent to which financial constraints
(and other obstacles) drive these distortions. I use data from the Enterprise Survey by the World
Bank.
65

Data
The Enterprise Survey study is conducted by the World Bank and covers small, medium, and large
firms in the manufacturing, the services, the transportation, and the construction sectors. The
firms are interviewed, and the questionnaires cover questions ranging from firms’ characteristics
such as age, size, and industry group, to firms’ operations such as their sales, costs of production,
and assets. These surveys have been conducted for a couple of rounds in most African countries, in
different years, and are largely cross-sectional with some smaller panels available for some countries.
For each of the countries included in my analysis, I use data from the 2005, 2006, or 2007 surveys,
which are when most African countries were covered for the first time. I include countries for which
all the variables needed for the analysis are reported. My study includes firms in the manufacturing
sector only because capital data are not reported for firms in the services sector. To make sure the
analysis is not biased by outliers on revenue, labor costs, replacement costs of capital, and costs of
materials, I drop observations that are three standard deviations away from their means in each
country. Finally, I include only countries that have more than 100 observations. See Table 2.1 for
a list of the 12 sub-Saharan African countries in my final sample along with the years of survey
and the sample sizes.
For my analysis, I use the firms’ reported sales as their revenues (PsiYsi), the replacement value
of all machinery and equipment as their level of capital (Ksi), their total cost of labor, including
wages, salaries and bonuses, and social payments as the amount of labor hired (Lsi), and their
total spending on raw materials and intermediate inputs as the level of intermediate inputs used
(Msi). Firms are also asked to determine, on a scale of 0 to 4, the extent to which a specific factor
is an obstacle to them, higher values meaning more severe obstacles. Such factors include access to
infrastructure and to services such as transportation and financing; taxes; weak institutions such
as corruption and political instability. I use these variables in my regression analysis to estimate
the extent to which they drive firm-level distortions. Table 2.2 presents summary statistics. We
can see that access to finance constitutes the most severe obstacle with an average value of 2.4,
followed by tax rates with an average of 1.8. Labor regulations seem to be the least severe obstacles
66

Table 2.1: List of Countries
Country Year Sample Size
Angola 2006 189
Cameroon 2006 102
Ethiopia 2006 229
Ghana 2007 268
Guinea 2006 121
Madagascar 2005 117
Mali 2007 251
Mozambique 2007 336
Nigeria 2007 884
Senegal 2007 244
Uganda 2006 290
Zambia 2007 279
Notes: List of countries included in my
analysis, after excluding countries with
less than 100 observations and countries
with missing variables. The sample size
is the size of the data with manufacturing firms only and after eliminating outliers, specifically, observations that are 3
standard deviation away from the mean
of sales, capital, labor, intermediate inputs.
67

in these sample countries, with an average of .5. Figure 2.1 plots the firms’ ratings of access to
finance as an obstacle. Most firms, over 55%, reported access to finance to be a very severe or a
major obstacle.
Table 2.2: Obstacles
Variable Observations Mean Stand Dev. Min Max
Access to finance 3,383.00 2.36 1.50 0.00 4.00
Tax rates 3,386.00 1.84 1.38 0.00 4.00
Transportation 3,386.00 1.51 1.30 0.00 4.00
Corruption 3,387.00 1.29 1.41 0.00 4.00
Access to land 3,379.00 1.24 1.39 0.00 4.00
Inadequately educated workforce 3,386.00 0.86 1.12 0.00 4.00
Political instability 3,219.00 0.75 1.11 0.00 4.00
Labor regulations 3,387.00 0.54 0.90 0.00 4.00
Notes: Firms are asked to rank, on a scale from 0 to 4, the extent to which each of these presents
an obstacle to their operations, 0 being not an obstacle at all and 4 being very severe obstacles.
Countries: Madagascar (2005); Angola, Cameroon, Ethiopia, Guinea, and Uganda (2006); Ghana,
Mali, Mozambique, Nigeria, Senegal and Zambia (2007)
Figure 2.1: Degree of access to finance obstacle
Notes: Countries: Madagascar 2005; Angola, Cameroon,
Ethiopia, Guinea, Uganda and 2006; Ghana, Mali, Mozambique,
Nigeria, Senegal and Zambia 2007.
Data sources: World Bank Enterprise Survey
Additional variables used in my analysis are size, age, whether the firm is in an export processing
zone, the percentage of sales that are directly exported, the percentage of inputs imported, the
percentage of the firm owned by the government, and the region in which the firm is located. The
size variable ranges from 0 to 2, for small, medium, and large firms respectively. Small enterprises
are defined as enterprises with less than 10 employees, medium ones are enterprises with employees
68

between 11 and 100, and large enterprises are those with more than 100 employees. Figure 2.2
plots the size distribution of the firms, and we can see that the distribution is highly skewed to the
right. There are many small firms and very few large ones. The export processing zone variable is
a dummy that takes the value of 1 if the firm is in an export processing zone and 0 otherwise.
Figure 2.2: Firm size distribution
Notes: Bin size is 10.
Calibration
To calibrate parameters in the model, I follow the literature in setting some parameter values, and
use firms’ optimization problem to derive expressions for the parameters governing the shares of
inputs in the total costs. Specifically, I set σ = 3 and assume Rs = δs + rs, where δs = rs = .05.
Following Atalay (2017), I set ϵm = .84, which is the lower bound found by Atalay (2017) as the
interval for the values of ϵm is between .84 and .88. Moreover, I assume Ws = 1 so total labor
compensation is Lsi and is my labor inputs, and similarly, Zs = 1 so intermediate input costs=Msi.
Finally, from the firms’ optimization problem, I obtain the following:
αs =
RsKsi
RsKsi + Lsi
(2.12)
µs =
ZsMsi
RsKsi+WsLsi
ϵm
V Aϵm−1
ZsMsi
RsKsi+WsLsi
ϵm
V Aϵm−1 + Mϵm−1
si
(2.13)
69

where
V A = (Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
(2.14)
I included in the Chapter Two Appendix the derivation of αsi and µs.
Misallocation and Firms’ Obstacles
I use the calibrated model to derive measures of output, capital, and labor misallocation, namely
τysi, τksi, and τlsi respectively. I first conduct a general diagnosis of the extent to which different
obstacles that firms face contribute to the degree of misallocation in these countries. I run OLS
regressions for each country with industry fixed effects of the different measures of misallocation
on eight sources of obstacles: access to finance, transportation, access to land, tax rates, labor
regulations, corruption, workforce education, and political instability, as follows:
log(1 − τysi) = α1 + α2Oi + δs + µi (2.15)
log(1 + τksi) = β1 + β2Oi + δs + ϵi (2.16)
log(1 + τlsi) = λ1 + λ2Oi + δs + θi (2.17)
for firm i in industry s, where Oi
is the vector of obstacles and δs is industry fixed effects. I
report the results for each country in Chapter Two Appendix. Note that since the dependent
variable in equation 2.15 is log(1−τysi), a negative coefficient indicates higher output misallocation
(the dependent variable is essentially a measure of allocative efficiency). Access to finance is by
far the obstacle that contributes most significantly to output misallocation among firms in half
of the countries. Some other factors increase output distortions, such as corruption in Nigeria,
labor regulations in Ethiopia, and political instability in Zambia and Uganda. Capital and labor
distortions are not affected by these obstacles, I therefore focus on output misallocation.
Furthermore, I look at the effects of each of these obstacles separately on output misallocation
70

using the cross-section of all countries in my sample. Specifically, I run the following regression:
log(1 − τycsi) = α1 + α2Oi + α3Xi + δsc + ϵi (2.18)
for firm i in sector s in country c, where Oi
is an obstacle; Xi
is the vector of controls: size, age,
whether the firm is in an export processing zone, the percentage of sales that are directly exported,
the percentage of inputs imported, the percentage of the firm owned by the government, and the
region in which the firm is located; δsc are sector and country fixed effects.
Table 2.3 reports the results from specification 2.18. These cross-country results again show that
access to financing remains important in predicting output distortions in Africa. Specifically, firms
facing more access to financing obstacles are more distorted. Another obstacle that statistically
significantly affects output misallocation, but decreases it, is labor regulations. Given the importance of financial constraints in driving misallocation in Africa, I decided to focus on such obstacles
for the purpose of my study.
I therefore run the following regressions, for each country and at the firm level:
log(1 − τysi) = α1 + α2Fi + α3Xi + δs + µi (2.19)
log(1 + τksi) = β1 + β2Fi + λ3Xi + δs + ϵi (2.20)
log(1 + τlsi) = λ1 + λ2Fi + λ3Xi + δs + θi (2.21)
where Fi
is the degree to which the firm faces financial constraints; Xi
is my vector of controls;
and δs is industry fixed effects. Tables 2.4 to 2.17 show the results of the regressions for the 12
countries in my analysis. Models 1, 2, and 3 in tables 2.6 to 2.17 are equations 2.19, 2.20, and 2.21
respectively without the controls, Model 4 is equation 2.19 with only size and its interaction with
Fi as controls, and Model 5 is equation 2.19 with all controls, along with interactions of size with Fi
and whether the firm is in an export processing zone. Model 5 is my preferred specification given it
focuses on output distortions which seem to be more relevant and includes all the variables I deem
important in determining output distortions. I reported the results from Model 5 for all countries in
Tables 2.4 and 2.5. I focus on output misallocation since it seems to be more significantly driven by
71

Table 2.3: Regressions of output distortions on all obstacles
Financing Transportation Labor Reg Land Tax Corruption Education
Obstacle -0.089∗∗∗∗ -0.012 0.16∗∗∗∗ -0.034 0.042 -0.022 0.025
(0.023) (0.028) (0.041) (0.025) (0.025) (0.026) (0.032)
Size 1.63∗∗∗∗ 1.76∗∗∗∗ 1.79∗∗∗∗ 1.73∗∗∗∗ 1.74∗∗∗∗ 1.68∗∗∗∗ 1.68∗∗∗∗
(0.1) (0.094) (0.076) (0.084) (0.01) (0.083) (0.08)
Obstacle × size 0.066∗ 0.01 -0.041 0.045 0.027 0.092∗∗ 0.13∗∗
(0.036) (0.041) (0.061) (0.042) (0.041) (0.042) (0.049)
Age 0.0034 0.0032 0.0036 0.003 0.003 0.003 0.003
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
EPZ 0.40∗∗∗∗ 0.40∗∗∗∗ 0.38∗∗∗∗ 0.40∗∗∗∗ 0.41∗∗∗∗ 0.41∗∗∗∗ 0.40∗∗∗∗
(0.095) (0.096) (0.095) (0.096) (0.095) (0.096) (0.096)
EPZ × size 0.07 0.061 0.065 0.065 0.063 0.044 0.036
(0.13) (0.13) (0.13) (0.13) (0.13) (0.13) (0.13)
Percentage of inputs imported 0.0048∗∗∗∗ 0.0048∗∗∗∗ 0.0047∗∗∗∗ 0.0049∗∗∗∗ 0.0047∗∗∗∗ 0.0048∗∗∗∗ 0.0046∗∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Percentage of sales exported 0.010∗∗∗∗ 0.010∗∗∗∗ 0.010∗∗∗∗ 0.010∗∗∗∗ 0.010∗∗∗∗ 0.010∗∗∗∗ 0.0098∗∗∗
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
Percentage owned by state 0.0069∗∗ 0.0069∗∗ 0.0066∗∗ 0.007∗∗ 0.007∗∗ 0.0074∗∗ 0.0047
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
Region 0.016 0.019 0.027∗ 0.017 0.02 0.02 0.023∗
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Observations 3091 3091 3091 3091 3091 3091 3091
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry and country fixed effects. Dependent variable is a measure of output allocative efficiency, log(1 − τysi), so a
negative coefficient indicates higher output misallocation.
72

financial constraints relative to labor and capital misallocation. Although capital misallocation has
been documented in the literature as a main consequence of market inefficiencies and an important
source of distortions, the evidence from African countries suggests that financial constraints do not
play an important role in predicting it.
Table 2.4: Output distortions and financial constraints
Mozambique Senegal Ghana Madagascar Nigeria Zambia
Financial constraint -0.30∗∗∗∗ -0.14∗
-0.24∗∗ 0.23 0.034 0.019
(0.072) (0.074) (0.093) (0.22) (0.038) (0.098)
Size 1.75∗∗∗∗ 0.84∗∗ 1.26∗∗∗ 1.56∗∗∗ 1.57∗∗∗∗ 1.47∗∗∗∗
(0.33) (0.39) (0.44) (0.52) (0.19) (0.23)
Financial constraint × size 0.35∗∗∗∗ 0.33∗∗ 0.12 -0.32∗
-0.11 0.071
(0.10) (0.15) (0.14) (0.17) (0.069) (0.11)
Age 0.00099 -0.012 0.0042 0.043∗∗ 0.016∗∗ 0.025∗∗∗
(0.008) (0.012) (0.0097) (0.019) (0.0064) (0.0084)
EPZ 0.27 0.3 0.3 1.18 0.52∗∗ 0.16
(0.25) (0.47) (0.43) (1.19) (0.21) (0.39)
EPZ × size -0.045 0.96∗
-0.22 -0.15 -0.14 0.10
(0.33) (0.57) (0.52) (0.76) (0.32) (0.35)
Percentage of inputs imported 0.0059∗∗ 0.0055∗
-0.0049 -0.00072 0.011∗∗∗∗
(0.0029) (0.003) (0.0038) (0.0023) (0.0032)
Percentage of sales exported -0.0006 0.018∗ 0.013 0.026∗∗ 0.016∗
(0.0079) (0.0098) (0.01) (0.011) (0.0092)
Observations 336 244 268 117 884 279
Adjusted R2 0.439 0.301 0.287 0.092 0.245 0.406
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variable is a measure of output allocative efficiency, log(1 −
τysi), so a negative coefficient indicates higher output misallocation.
Looking at Tables 2.4 and 2.5, we can see that under Model 5, financial constraints increase output
misallocation in Mozambique, Senegal, and Ghana. Size is an important determinant of distortions
in many countries, with bigger firms being less distorted. In Mozambique and Senegal, bigger firms
are disproportionately less affected by financial constraints relative to smaller ones. Being in an
73

Table 2.5: Output distortions and financial constraints
Angola Cameroon Ethiopia Uganda Guinea Mali
Financial constraint 0.028 -0.0064 -0.18 -0.13 0.16 -0.058
(0.072) (0.25) (0.12) (0.086) (0.12) (0.057)
Size 0.42 1.16∗ 2.09∗∗∗∗ 1.8∗∗∗∗ 1.18 1.27∗∗∗
(0.57) (0.59) (0.39) (0.39) (1.17) (0.44)
Financial constraint × size 0.23 -0.11 0.069 0.078 0.2 0.0026
(0.17) (0.18) (0.14) (0.13) (0.48) (0.16)
Age -0.027∗∗ -0.018 0.0017 -0.0068 -0.01 -0.0036
(0.011) (0.016) (0.013) (0.012) (0.021) (0.012)
EPZ 0.28 1.48∗ 2.67∗∗∗∗ 0.22 -0.31 0.47∗
(0.18) (0.76) (0.52) (0.24) (0.51) (0.26)
EPZ × size 0.73 0.07 -1.32∗∗ -0.089 0 -0.021
(0.54) (0.53) (0.54) (0.34) (.) (0.66)
Percentage of inputs imported 0.0062∗∗ 0.025∗∗∗∗ 0.0007 0.018∗∗∗∗ 0.0085∗
-0.00065
(0.0025) (0.0052) (0.0046) (0.0036) (0.0045) (0.0028)
Percentage of sales exported 0.24 0.022∗∗∗ -0.0075 0.017∗∗ -0.012 0.0086
(0.24) (0.0076) (0.012) (0.007) (0.017) (0.0078)
Observations 189 102 229 290 121 251
Adjusted R2 0.174 0.535 0.397 0.439 0.046 0.082
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variable is a measure of output allocative efficiency,
log(1 − τysi), so a negative coefficient indicates higher output misallocation.
74

export processing zone (EPZ=1) decreases distortions in Senegal and Ethiopia, after holding size
constant. My results may not be statistically significant in some countries due to the small sample
sizes, especially in Guinea, Cameroon, and Angola. That is why the cross-country analysis gives
stronger results on the effect of financial constraints on output distortions.
Tables 2.6 to 2.17 give more granular analysis of the determinants of distortions in each country.
There is considerable cross-country heterogeneity in the sources of misallocation and importance
of financial constraints. Without controlling for size and the other firm characteristics, financial
constraints are drivers of output misallocation in Ethiopia, Ghana, Mali, Mozambique, Nigeria,
and Senegal (Model 1). But once I control for size and its interaction with Fi (Model 4), financial constraints become statistically significant in fewer countries while size becomes a statistically
significant predictor of the degree of misallocation. The differential effect of financial constraints
on misallocation between smaller and bigger firms in Model 4 is given by the coefficient on the
interaction between size and financial constraints. This coefficient is significant in Madagascar,
Mozambique, Nigeria, and Senegal. A significant and positive coefficient means that the disproportionate effect of financial constraints on distortions is larger on smaller firms relative to bigger
ones. That is the case in Mozambique and Senegal.
The sum of the coefficients on size and the interaction between size and financial constraints is the
effect of being a big firm on output misallocation relative to being smaller, conditional on being financially constrained. This sum is positive and statistically significant in Madagascar, Mozambique,
Nigeria, and Senegal, meaning that in these countries, conditional on being financially constrained,
bigger firms face lower distortions. These results suggest that size is the main channel through
which financial constraints affect output misallocation. Particularly, in Madagascar, Mozambique,
Nigeria, and Senegal, a firm is more able to overcome its financial challenges the bigger it is.
In Angola (Table 2.6), age and imports seem to be important determinants of firms’ output distortions, which decrease with higher imports but increase with age (Model 5).
Financial constraints are not drivers of misallocation in Cameroon (Table 2.7) but trade matters a
lot. Both imports and exports decrease misallocation, and being in an EPZ also decreases output
distortions.
75

Table 2.6: Angola
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint 0.0091 -0.058 -0.026 -0.015 0.028
(0.067) (0.1) (0.05) (0.069) (0.072)
Size 0.93∗ 0.42
(0.54) (0.57)
Financial constraint × size 0.18 0.23
(0.17) (0.17)
Age -0.027∗∗
(0.011)
EPZ 0.28
(0.18)
EPZ × size 0.73
(0.54)
Percentage of inputs imported 0.0062∗∗
(0.0025)
Percentage of sales exported 0.24
(0.24)
Region -0.16
(0.145)
Observations 189 189 189 189 189
Adjusted R2
-0.027 -0.025 -0.026 0.091 0.174
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is
the degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher
output misallocation. Model 2: dependent variable is capital misallocation, log(1 + τksi). Model 3:
dependent variable is labor misallocation, log(1 + τlsi).
76

Table 2.7: Cameroon
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.29 -0.21 -0.13 -0.25 -0.0064
(0.2) (0.13) (0.11) (0.31) (0.25)
Size 1.35∗ 1.16∗
(0.7) (0.59)
Financial constraint × size 0.21 -0.11
(0.22) (0.18)
Age -0.018
(0.016)
EPZ 1.48∗
(0.76)
EPZ × size 0.07
(0.53)
Percentage of inputs imported 0.025∗∗∗∗
(0.0052)
Percentage of sales exported 0.022∗∗∗
(0.0076)
Observations 103 103 103 103 102
Adjusted R2
-0.024 -0.025 -0.039 0.282 0.535
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
77

Table 2.8: Ethiopia
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.24∗ 0.067 -0.049 -0.18 -0.18
(0.13) (0.092) (0.073) (0.13) (0.12)
Size 2.24∗∗∗∗ 2.087∗∗∗∗
(0.34) (0.39)
Financial constraint × size 0.091 0.069
(0.15) (0.14)
Age 0.0017
(0.013)
EPZ 2.67∗∗∗∗
(0.52)
EPZ × size -1.32∗∗
(0.54)
Percentage of inputs imported 0.0007
(0.0046)
Percentage of sales exported -0.0075
(0.012)
Percentage owned by state 0.0053
(0.0069)
Region -0.12∗∗∗∗
(0.035)
Observations 229 229 229 229 229
Adjusted R2 0.004 -0.015 -0.014 0.325 0.397
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
78

Table 2.9: Ghana
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.27∗∗∗ 0.0036 -0.05 -0.24∗∗ -0.24∗∗
(0.087) (0.077) (0.065) (0.095) (0.093)
Size 1.42∗∗∗∗ 1.26∗∗∗
(0.42) (0.44)
Financial constraint× size 0.085 0.12
(0.14) (0.14)
Age 0.0042
(0.0097)
EPZ 0.303
(0.43)
EPZ × size -0.22
(0.52)
Percentage of inputs imported -0.00485
(0.0038)
Percentage of sales exported 0.013
(0.010)
Percentage owned by state 0.011
(0.015)
Region -0.460∗∗∗∗
(0.103)
Observations 268 268 268 268 268
Adjusted R2 0.021 -0.015 -0.013 0.234 0.287
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
79

In Ethiopia (Table 2.8), the location of the firm is a strong determinant of output misallocation.
Firms in export processing zones are less distorted, even after controlling for size.
Being in an export processing zone does not decrease the degree of output misallocation for firms
in Ghana (Table 2.9), but financial constraints and size affect misallocation. More financially
constrained and smaller firms have higher output distortions. The region in which the firm is
located also seems to predict the level of misallocation.
In Guinea, neither financial constraints nor size affect misallocation (Table 2.10). Only the region
in which the firm is located and imports play an important role in how much misallocation it
experiences. Although my analysis in Madagascar is limited due to the small number of observations, the results are nonetheless very interesting (Table 2.11). In Model 1, financial constraints
increase output distortions but once I control for size in Models 4 and 5, financial constraints are
no longer important, suggesting that size is the channel through which financial constraints affect
misallocation. Specifically, the coefficient on the interaction between financial constraints and size
is statistically significant and negative, meaning that financial constraints disproportionately affect
bigger firms more relative to small ones. Older firms experience lower distortions.
Size is a determinant of output distortions in Mali (Table 2.12), as bigger firms face lower output
distortions. I also find that financial constraints decrease capital misallocation. This suggests
that being financially constrained decreases the marginal revenue product of capital relative to
intermediate inputs, which means financial constraints affect the firm’s use of intermediate inputs
relative to capital.
Mozambique is subject to multiple factors that distort allocations, where both capital and labor
distortions decrease with financial constraints (Table 2.13). These results suggest that difficulty
accessing financing constrains the firms’ use of intermediate inputs relative to labor and capital,
thus decreasing the marginal revenue products of capital and labor relative to intermediate inputs’.
Financial constraints, size, imports, government ownership, and region all affect output distortions.
Again, firms that are more financially constrained and smaller face more output distortions, and
holding the degree of financial obstacles constant, bigger firms face lower distortions relative to
small ones.
80

Table 2.10: Guinea
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint 0.16 -0.0041 0.0016 0.16 0.16
(0.11) (0.11) (0.087) (0.11) (0.12)
Size 0.64 1.18
(1.086) (1.17)
Financial constraint × size 0.08 0.21
(0.43) (0.48)
Age -0.010
(0.021)
EPZ -0.31
(0.51)
Percentage of inputs imported 0.0085∗
(0.0045)
Percentage of sales exported -0.012
(0.017)
Percentage owned by state -0.038
(0.032)
Region -1.005∗∗
(0.49)
Observations 121 121 121 121 121
Adjusted R2
-0.024 -0.043 -0.043 -0.014 0.046
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is
the degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher
output misallocation. Model 2: dependent variable is capital misallocation, log(1 + τksi). Model 3:
dependent variable is labor misallocation, log(1 + τlsi).
81

Table 2.11: Madagascar
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.22 0.076 0.04 0.16 0.23
(0.15) (0.14) (0.11) (0.22) (0.22)
Size 1.57∗∗∗ 1.56∗∗∗
(0.51) (0.52)
Financial constraint × size -0.29∗
-0.32∗
(0.17) (0.17)
Age 0.043∗∗
(0.019)
EPZ 1.18
(1.19)
EPZ × size -0.15
(0.76)
Observations 117 117 117 117 117
Adjusted R2
-0.011 -0.032 -0.031 0.077 0.092
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is
the degree of output allocative efficiency, log(1−τysi), so a negative coefficient indicates higher
output misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model
3: dependent variable is labor misallocation, log(1 + τlsi).
82

Table 2.12: Mali
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.11∗∗ -0.076 -0.03 -0.072 -0.058
(0.054) (0.052) (0.046) (0.056) (0.057)
Size 1.29∗∗∗ 1.27∗∗∗
(0.41) (0.44)
Financial constraint × size 0.0043 0.0026
(0.16) (0.16)
Age -0.0036
(0.012)
EPZ 0.47∗
(0.26)
EPZ × size -0.021
(0.66)
Percentage of inputs imported -0.00065
(0.0028)
Percentage of sales exported 0.0086
(0.0078)
Region -0.026
(0.094)
Observations 251 251 251 251 251
Adjusted R2 0.002 -0.004 -0.012 0.078 0.082
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is
the degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher
output misallocation. Model 2: dependent variable is capital misallocation, log(1 + τksi). Model 3:
dependent variable is labor misallocation, log(1 + τlsi).
83

Table 2.13: Mozambique
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.27∗∗∗∗ -0.14∗
-0.14∗∗∗ -0.31∗∗∗∗ -0.30∗∗∗∗
(0.077) (0.081) (0.047) (0.073) (0.072)
Size 1.92∗∗∗∗ 1.75∗∗∗∗
(0.26) (0.33)
Financial constraint × size 0.29∗∗∗ 0.35∗∗∗∗
(0.10) (0.10)
Age 0.00099
(0.008)
EPZ 0.27
(0.25)
EPZ × size -0.045
(0.33)
Percentage of inputs imported 0.0059∗∗
(0.0029)
Percentage of sales exported -0.0006
(0.0079)
Percentage owned by state 0.081∗∗
(0.038)
Region -0.38∗∗∗∗
(0.10)
Observations 336 336 336 336 336
Adjusted R2 0.024 -0.003 0.013 0.410 0.439
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
84

Table 2.14: Nigeria
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.11∗∗∗ 0.032 -0.03 -0.011 0.034
(0.036) (0.037) (0.026) (0.039) (0.038)
Size 1.69∗∗∗∗ 1.57∗∗∗∗
(0.19) (0.2)
Financial constraint × size -0.12∗
-0.11
(0.071) (0.069)
Age 0.016∗∗
(0.0064)
EPZ 0.52∗∗
(0.21)
EPZ × size -0.14
(0.32)
Percentage of inputs imported -0.00072
(0.0023)
Percentage of sales exported 0.026∗∗
(0.011)
Region 0.134∗∗∗∗
(0.0153)
Observations 884 884 884 884 884
Adjusted R2 0.005 -0.005 -0.004 0.165 0.245
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
85

Table 2.15: Senegal
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.13∗
-0.098 -0.022 -0.15∗∗ -0.14∗
(0.079) (0.063) (0.051) (0.074) (0.074)
Size 1.19∗∗∗∗ 0.84∗∗
(0.34) (0.39)
Financial constraint × size 0.4∗∗∗ 0.33∗∗
(0.14) (0.15)
Age -0.012
(0.012)
EPZ 0.3
(0.47)
EPZ × size 0.96∗
(0.57)
Percentage of inputs imported 0.0055∗
(0.003)
Percentage of sales exported 0.018∗
(0.0098)
Region -0.07
(0.1)
Observations 244 244 244 244 244
Adjusted R2
-0.005 -0.006 -0.016 0.265 0.301
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is
the degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher
output misallocation. Model 2: dependent variable is capital misallocation, log(1 + τksi). Model 3:
dependent variable is labor misallocation, log(1 + τlsi).
86

Table 2.16: Uganda
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.1 0.03 -0.025 -0.15∗
-0.13
(0.096) (0.077) (0.054) (0.089) (0.086)
Size 2.31∗∗∗∗ 1.8∗∗∗∗
(0.37) (0.39)
Financial constraint × size 0.015 0.078
(0.13) (0.13)
Age -0.0068
(0.012)
EPZ 0.22
(0.24)
EPZ × size -0.089
(0.34)
Percentage of inputs imported 0.018∗∗∗∗
(0.0036)
Percentage of sales exported 0.017∗∗
(0.007)
Region -0.025
(0.09)
Observations 290 290 290 290 290
Adjusted R2
-0.014 -0.017 -0.017 0.385 0.439
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation. Model 2: dependent variable is capital misallocation, log(1+τksi). Model 3: dependent
variable is labor misallocation, log(1 + τlsi).
87

Table 2.17: Zambia
Model 1 Model 2 Model 3 Model 4 Model 5
Financial constraint -0.07 0.037 -0.07 0.037 -0.12∗∗
-0.045 0.019
(0.095) (0.071) (0.05) (0.1) (0.098)
Size 1.74∗∗∗∗ 1.47∗∗∗∗
(0.22) (0.23)
Financial constraint × size 0.14 0.071
(0.11) (0.11)
Age 0.025∗∗∗
(0.0084)
EPZ 0.16
(0.39)
EPZ × size 0.10
(0.35)
Percentage of inputs imported 0.011∗∗∗∗
(0.0032)
Percentage of sales exported 0.016∗
(0.0092)
Percentage owned by state -0.004
(0.008)
Region 0.21∗∗
(0.10)
Observations 279 279 279 279 279
Adjusted R2
-0.013 -0.014 0.006 0.346 0.406
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Models 1, 4 and 5: dependent variable is the
degree of output misallocation, log(1 − τysi), so a negative coefficient indicates higher output misallocation. Model 2: dependent variable is capital misallocation, log(1 − τksi). Model 3: dependent
variable is labor misallocation, log(1 + τksi).
88

Size, age, exporting, and region are the main determinants of firms’ output distortions in Nigeria
(Table 2.14). Bigger and older firms are less confronted with output distortions, and exporting also
reduces distortions. Moreover, being in an EPZ decreases output misallocation.
Most of the factors considered in my analysis seem to play important roles in output distortions in
Senegal, except for age and region (Table 2.15). The results in Senegal are consistent with most
of the results stated above, with tighter financial constraints leading to higher misallocation. On
the other hand, both importing and exporting as well as having larger establishments lower the
levels of misallocation. Not being in an export processing zone disproportionately increases output
distortions for smaller firms more than for bigger firms.
Only size and trade seem to matter in Uganda in predicting output distortions (Table 2.16). Consistent with results found in several countries, bigger firms and those that export and import more
are less likely to face distortions.
Finally, in Zambia, size is the strongest determinant of output misallocation, with bigger firms
facing significantly lower distortions (Table 2.17). Older firms and those that trade more also
experience lower distortions.
Overall, financial constraints are non-negligible factors driving output misallocation in some subSaharan African countries, even after controlling for other confounding factors. Size is the main
channel through which these financial constraints affect misallocation. In other words, firms that
are more financially constrained are smaller, and potentially grow more slowly, which leads to
size distortions. Even given the same degree of financial constraints, bigger firms are less distorted,
suggesting that their size allows them to overcome distortions despite the financial obstacles. Export
processing zones are generally favorable as there are significantly lower distortions in those zones
even after controlling for size. This is an important result as it suggests that the policy environment
in these zones, if expanded to more areas, have the potential to decrease misallocation in a given
country. Trade is also a favorable factor in many countries where exporting and/ or importing
lessen the degree of distortions.
Given the importance of size, I conduct my analysis above on output distortions by firm size.
Specifically, I run equation 2.19 by firm size, bundling small and medium enterprises, and big en89

terprises. The results are reported in Table 2.18, where we can clearly see that financing constraints
are drivers of output distortions for small and medium enterprises only. However, it is important
to note that the sample size for the large firms is very small, and that may be the reason why the
coefficients are not statistically significant.
Table 2.18: Regressions of output distortions by size
All Small and Medium Large
Financing constraint -0.12∗∗∗∗ -0.12∗∗∗∗ 0.024
(0.03) (0.034) (0.21)
Age 0.017∗∗∗ 0.012∗∗∗ 0.032
(0.0052) (0.0044) (0.026)
EPZ 0.61∗∗∗∗ 0.54∗∗∗∗ 0.36
(0.11) (0.11) (0.31)
Percentage of inputs imported 0.0086∗∗∗∗ 0.0079∗∗∗∗ 0.001
(0.002) (0.0019) (0.0074)
Percentage of sales exported 0.025∗∗∗∗ 0.022∗∗∗∗ 0.0011
(0.0028) (0.0047) (0.0066)
Percentage owned by state 0.022∗∗∗ 0.015∗∗ -0.015
(0.0084) (0.0066) (0.013)
Region 0.0071 0.018 -0.24
(0.045) (0.045) (0.2)
Observations 3091 2978 113
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry and country fixed effects, by size of enterprises.
Dependent variable is the degree of output allocative efficiency, log(1−τysi), so a negative
coefficient indicates higher output misallocation.
Chapter Two: The Role of Managers’ Education and Experience
Literature dating back to the 1960s has documented the importance of education and human capital
in economic growth, and particularly in stimulating production (Welch (1970) Nelson and Phelps
(1966)). Here, I investigate the role that the managers’ education and experience play in the extent
to which their firms are constrained financially and face distortions as a result. The rationale is
90

that experienced, and highly educated managers may be able to better use financial institutions to
finance their firms’ activities, innovate, and invest in technology, and as a result, are less constrained
in their use of factor inputs. In Tables 2.19 and 2.20, I run specification 2.19 using the cross-section
of all countries, with country and sector fixed effects (column 1 Table 2.19), and for each country
individually with sector fixed effects.
Table 2.19: Regressions of output distortions on financial constraint
All Angola Cameroon Ethiopia Ghana Guinea
Financial constraint -0.0019 0.18 0.21 -0.054 0.16 -0.094
(0.066) (0.29) (0.57) (0.28) (0.29) (0.29)
Size 1.80∗∗∗∗ 1.40∗∗∗∗ 2.04∗∗∗∗ 1.65∗∗∗∗ 1.45∗∗∗∗ 1.11∗∗
(0.070) (0.29) (0.32) (0.27) (0.20) (0.50)
Manager education 0.21∗∗∗∗ 0.13 0.33 0.50∗∗∗∗ 0.41∗∗∗ -0.11
(0.032) (0.15) (0.22) (0.14) (0.14) (0.13)
Financial constraint × manager education -0.0064 -0.0030 -0.071 0.027 -0.094∗∗ 0.032
(0.010) (0.048) (0.069) (0.059) (0.047) (0.042)
Manager experience 0.0053 0.050∗
-0.094 0.017 0.0038 -0.058
(0.0067) (0.030) (0.078) (0.023) (0.023) (0.042)
Financial constraint × Manager experience -0.0032 -0.021∗∗ 0.016 -0.0098 0.0038 0.0084
(0.0024) (0.0095) (0.024) (0.010) (0.0076) (0.016)
Observations 2376 189 100 229 268 121
Adjusted R2 0.316 0.118 0.318 0.392 0.276 -0.003
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects (and country fixed effects for the first column). Dependent variable is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation.
With the cross-section of countries, we can see from Table 2.19 that the manager’s education decreases output distortions, while their experience does not have a significant effect on misallocation.
The same results on manager’s education are found in Ethiopia, Ghana, Mozambique, Uganda, and
Zambia. Additionally, the sum of the coefficients on education and the interaction between education and financial constraint is the effect of manager’s education for firms that are financially
constrained. This sum is positive in all countries except in Guinea, meaning that holding the level
of financial constraints constant, managers’ education has a negative effect (decreases) output misallocation. Furthermore, in Senegal, financial constraints have a disproportionate negative effect
91

Table 2.20: Regressions of output distortions on financial constraint
Mali Mozambique Nigeria Senegal Uganda Zambia
Financial constraint -0.074 -0.024 0.18 -0.53∗∗∗∗ 0.42 0.35
(0.13) (0.19) (0.65) (0.15) (0.30) (0.31)
Size 1.12∗∗∗∗ 1.94∗∗∗∗ 1.27∗∗∗ 1.44∗∗∗∗ 2.13∗∗∗∗ 1.72∗∗∗∗
(0.29) (0.19) (0.38) (0.23) (0.18) (0.17)
Manager education 0.11 0.33∗∗∗∗ 0.072 0.064 0.39∗∗∗ 0.35∗∗∗
(0.075) (0.096) (0.26) (0.068) (0.13) (0.11)
Financial constraint × manager education 0.0035 0.00088 -0.055 0.078∗∗∗ -0.066 -0.050
(0.025) (0.031) (0.093) (0.026) (0.041) (0.048)
Manager experience 0.0024 0.027∗ 0.015 -0.018 0.040 0.0023
(0.021) (0.015) (0.047) (0.017) (0.035) (0.017)
Financial constraint × Manager experience -0.00014 -0.010∗
-0.0064 0.011∗
-0.011 0.00080
(0.0068) (0.0053) (0.021) (0.0063) (0.012) (0.0080)
Observations 251 336 69 244 290 279
Adjusted R2 0.092 0.458 0.105 0.337 0.413 0.403
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variable is the degree of output allocative efficiency, log(1 − τysi), so
a negative coefficient indicates higher output misallocation.
92

on output misallocation for firms with more highly educated managers relative to firms with less
educated managers, while in Ghana the opposite is found. On the other hand, managers’ experience lowers output distortions in Angola and Mozambique only. Holding the degree of financial
constraints constant however, managers’ experience lowers output misallocation in all countries
except in Cameroon, Guinea, Senegal, and Zambia. Therefore, there is evidence that management
practices stemming from education and experience matter when it comes to firms’ production and
distortions. In sum, common results on the effect of managers’ education on output distortions are
found in Ethiopia, Ghana, Mozambique, Uganda, and Zambia, and in Angola and Mozambique for
managers’ experience. There is cross-country heterogeneity however between Senegal and Ghana
on the disproportionate effect of financial constraints on misallocation for firms with managers of
different educational attainments.
Chapter Two: Robustness Checks
Bootstrap Standard Errors
I run some robustness checks to verify that my results are strong and insensitive to some confounding factors. First, I run the same specifications 2.18 and 2.19 but using bootstrap standard
errors to correct for the standard errors and control for any possible bias of my coefficients. The
results are reported in Tables 54, 55, and 56 in the Chapter Two Appendix. In Table 56, access
to financing remains the only obstacle that statistically significantly increases output distortions.
There is also still evidence that labor regulations decrease misallocation as found previously. Additionally, the bootstrap standard errors strengthen my results as now, financial constraints increase
output distortions in Mozambique, Senegal, and Ghana, as found in my baseline model, but also in
Madagascar. Size, being in an export processing zone, and trade all still strongly predict whether
firms face more or less output distortions in most of my sample countries. My baseline results are
therefore robust to correcting for the standard errors.
93

Excluding Size as a Control
Another concern from my analysis is the endogeneity of size. Specifically, the explanatory variable
size as measured by the number of employees in the firm is correlated with the measure of output
distortions that includes a measure of labor costs. To check whether this endogeneity issue biases
my results, I run the same regressions with the same controls excluding size. The results are
reported in Tables 57 and 58 in Chapter Two Appendix. Excluding size strengthens my baseline
results. Particularly, now financial constraints increase output misallocation in Ethiopia and Mali
in addition to Mozambique, Senegal, and Ghana. However, I find evidence that in Guinea firms
that are more financially constrained have lower output distortions.
Chapter Two: Sources of Financing
Financial constraints seem to be relatively important obstacles in the cross-section of my sample
sub-Saharan African countries. A more granular look into the different sources of financing for
firms in all the sample countries reveals that firms barely use financial institutions to finance their
working capital. As shown in Figure 2.3, the main source of financing for firms is their internal
funds and retained earnings which finance 72% of their working capital. This is followed by the
funds raised from purchases on credit from suppliers and advances from customers that cover 19.5%
of the working capital. Banks finance a very small proportion of the firms’ working capital, at 4.5%,
being the third least important source of financing after non-bank financial institutions and other,
that includes informal sources of financing such as money borrowed from family and friends.
I plot the sources of financing by firm size in Figure 2.4. Not surprisingly, bigger firms use financial
institutions to finance their working capital more relative to smaller firms. Banks finance 2.4% of
small firms’ working capital, 8.3% of medium-sized firms’ working capital, and 14.8% of big firms’
working capital. Banks finance bigger firms more probably because the latter have more assets and
so have higher collateral resources. This explains why bigger firms are less constrained financially
and face lower distortions due to financial constraints. These figures expose the weak financial
development in these countries and suggest why smaller firms are more financially constrained and
94

Figure 2.3: Sources of financing working capital
face more distortions. Given that smaller firms tend to raise less revenue, they will have less means
to finance their own working capital if they do not use financial institutions. Figures 2.5 and 2.6
further show that firms really do not use banks to fund their activities, as less than 20% of all
firms in the sample had applied for a loan in the year before they were interviewed, and less than
15% had a line of credit at the time of the interview. Again, bigger firms are more likely to have
applied for a loan and to have a line of credit. Among those that applied for a line of credit and
got rejected, most of them, over 50%, were rejected because of a lack of collateral (Figure 2.7).
Given these findings, for firms to be less constrained financially, they need to be able to generate
high revenues so they can use their funds to expand. In Chapter Two Appendix, I solve a suggestive
model of financial constraints to illustrate mechanisms by which productivity and size determine
firms’ (in)ability to overcome financial challenges and show the growth dynamics when firms are
financially constrained.
95

Figure 2.4: Sources of financing working capital by firm size
96

Figure 2.5: Loan application
Figure 2.6: Existence of a line of credit
Figure 2.7: Loan application rejection- reasons
97

Chapter Two: Conclusion
In this paper, I provided evidence that in various countries across sub-Saharan Africa, financial
obstacles are main drivers of misallocation of resources across firms. These financial obstacles
constrain smaller firms from growing and therefore they face even more distortions compared to
bigger firms. And even holding the degree of financial constraint constant, smaller firms face higher
distortions than bigger ones, suggesting that bigger firms are able to overcome their financial
obstacles more relative to smaller ones. I also find evidence that export processing zones are
favorable in lowering output distortions, as are managers’ education level and experience.
The main limitation of this study is the lack of adequate firm-level panel data in African countries.
To show the growth dynamics of firms and the ways in which productivity and size allow them to
overcome their borrowing constraint using the model of financial constraint, it is necessary to have
firm-level panel data for firms of all sizes. I leave the development of such data to future research.
98

Chapter 3
COVID-19 and Gender Inequality: Impact in Southern Africa
With Giorgia Albertin, Romina Kazandijan, and Tianyuan Wang1
Chapter Three: Introduction
The COVID-19 pandemic is disproportionately impacting women worldwide and likely even more
severely in developing countries. This is due to heavier unpaid care work burdens, occupational
gender segregation in affected sectors, increased gender-based violence (GBV), and possibly permanent loss of human capital. Recent research has pointed out that unlike previous recessions, which
have affected traditionally male occupations in cyclical sectors like construction and manufacturing,
the current crisis may be a “she-cession,” having damaged the services industry, i.e., retail, tourism,
and hospitality, where female employment is concentrated (Alon et al. (2020)). The UN warns that
the negative effects on women are even worse in developing countries, where 70 percent of female
employment is in the informal sector, which is characterized by low wages and high job insecurity
(United Nations (2020)). Moreover, school and daycare closures increase the already higher burden
of unpaid care work that women perform in the household – 2.4 hours more per day than men even
before the pandemic. It could lead to missed years of schooling as millions of girls around the world
are forced to permanently drop out of school (United Nations (2020)). As a result of restrictions on
movement due to lockdown measures, GBV is “increasing exponentially” around the world (United
Nations (2020), p.2). Therefore, as with many other structural vulnerabilities, the pandemic and
1The views expressed in this article are those of the author(s) and do not necessarily represent the views of the
IMF, its Executive Board, or IMF management.
99

associated lockdown mitigation measures are exposing and exacerbating pre-existing macro-critical
gender inequalities.
Given the macro-criticality of gender equality in supporting growth, addressing gaps in the labor market, education, health, and finance is key to support the post-pandemic recovery. Recent
IMF work has established that gender inequality in opportunities and outcomes is associated with
lower growth, productivity, economic diversification, financial sector stability, and higher income
inequality (IMF (2018)). Furthermore, replicating the growth decomposition analysis in Hakura
et al. (2016), associating the growth of real GDP per capita across 117 countries with income,
gender inequality, and other growth drivers, suggests that reducing high gender inequality could
significantly improve growth in SSA (Figure 3.1).2
In particular, reducing the two gender inequality
indicators (female legal equity and gender inequality) to the level currently observed in the fastgrowing ASEAN-5 countries could boost the region’s average annual per capita GDP growth by 0.7
percentage points. This is the most significant growth determinant after the impact of closing income inequality. Replicating this exercise for the Southern Africa Custom Union (SACU) highlights
the macro-criticality of gender equality for the region. Notably, eliminating gender inequality and
ensuring female legal equity has the potential to boost per capita GDP growth by 0.6 percentage
points cumulatively.
The global factors exacerbating existing gender inequalities are also at play in South Africa, while
less is known about the impact in the other Southern African countries, Botswana, Eswatini,
Lesotho, and Namibia (Casale and Posel (2021), Parry and Gordon (2020), Dlamini (2020)). Relative to other sub-regions in Africa, less analytical and operational work has been conducted at
the IMF on addressing the issue of macro-critical gender inequality specifically in Southern Africa.
Broader sub-regional research on this topic can establish common themes of the existing gender
inequalities in the region, e.g., the COVID-19 response may interact negatively with the disproportionately high HIV prevalence among women and girls due to a necessity to access services and
medications, and competition for prioritization in health spending.
The goal of this paper is to analyze gender disparities and empirically quantify the disproportionate
impact of the COVID-19 pandemic on women in the Southern African region. Using macro-level
2See Hakura et al. (2016) for details on the methodology of this growth decomposition exercise.
100

Figure 3.1: Growth Differential with ASEAN5 Countries
Sources: IMF, World Economic Outlook; PRS Group; World Bank, World Development Indicators
database; and staff estimates. Notes: A bar with a negative value denotes what share of the growth
shortfall in SACU countries is explained by a particular variable. For example, annual economic
growth of SACU countries could be higher by approximately 0.27 percentage point if gender inequality
was reduced to the level observed in ASEAN5 countries. The ASEAN5 countries are Indonesia,
Malaysia, the Philippines, Thailand, and Vietnam, to keep consistency with the original analysis.
For more details on the methodology, please see Hakura et al. (2016).
data and micro-level data from surveys on individuals, we derive stylized facts on pre-existing
gender disparities in the region. We then use pre- and post-COVID-19 survey data to empirically
estimate the disproportionate effect of the pandemic on women’s economic outcomes, specifically
in South Africa, Eswatini, Lesotho, and Namibia.
We find that there are still existing gender inequalities in the region, especially in employment,
political representation, and health, and that the COVID-19 pandemic exacerbated these gender
disparities. Women’s labor force participation and income are much lower in all five countries,
their unemployment is higher in all countries except Namibia, and they hold less than 50 percent
of the seats in their national parliaments. They are also more affected by HIV, and maternity
mortality rates are still high in the region. We find evidence from the empirical exercise that the
COVID-19 pandemic has widened the gap between male and female economic outcomes in South
Africa, Eswatini, Lesotho, and Namibia, as it has decreased women’s employment and income
more relative to men and has disproportionately increased the unpaid work burden on women. The
101

disproportionate negative effects of the pandemic on women are due to heavier unpaid work burden
and occupational gender segregation. Given these findings, recommended policies should subsidize
unpaid care work and promote girls’ education in STEM.
Our paper is the first one, to our knowledge, to empirically investigate the effect of the COVID19 pandemic on gender disparities in these countries. Bluedorn et al. (2021) explore gender gaps
during the pandemic in South Africa but do not empirically estimate the impact of the pandemic
on gender disparities. Some evidence has also been provided in Lesotho on a “she-cession”, but it
is just statistical evidence without empirical estimations (Government of Lesotho (2020)). There
is also some research on COVID-19 and gender in some other African countries, like Ethiopia and
Nigeria, but overall, less research has been done on Southern African countries, and this paper fills
this gap in the literature.
The paper proceeds as follows. Section 2 provides a brief overview of the literature. Section
3 describes the data, while section 4 presents stylized facts. Section 5 describes the empirical
strategy, and section 6 highlights the paper’s main results. Section 7 offers policy recommendations
for country authorities, and section 8 concludes.
Chapter Three: Literature Review
Gender equality is macro-critical. Existing research has established the macro-critical impact of
gender inequality on economic growth through lower productivity and diversification and higher
income inequality. Gender inequality is associated with lower productivity, as the lack of equal
participation of women in the workforce means about half of the population is unable to contribute
equally, which translates into aggregate productivity losses due to suboptimal allocation of talent
(Bertay et al. (2020)). In addition, discrimination against women in the labor market, including
the gender wage gap, leads to a misallocation of resources going to men that would have otherwise
gone to potentially more talented women (Ranasinghe (2020), Cuberes and Teignier (2016), Loko
and Diouf (2009), Gonzales et al. (2015)). Higher participation of women in the labor force is
also favorable for diversification, including export diversification, as women bring to the table
102

different skills and perspectives that complement those of men (Brixiov´a Schwidrowski et al. (2021),
Kazandjian et al. (2019)). Gender inequality increases within-country income inequality, which has
been shown to impact growth negatively. Better gender equality also alleviates both within- and
cross-country income disparities (IMF (2016), Gonzales et al. (2015), Chantreuil and Lebon (2015),
Sever (2022)). Furthermore, gender inequality is associated with lower profitability of firms and
lower bank stability, as a higher presence of women on banks’ boards leads to greater financial
resilience (Ranasinghe (2020), Sahay and Cihak (2018)).
Drivers of gender inequality include legal restrictions, as well as social and cultural norms. Many
factors causing gender inequality have been identified in the literature. Discriminatory laws and
policies can set barriers for women in accessing critical services, such as education, healthcare,
and financial services. Removal of these legal barriers promotes women’s economic participation
(Razavi (2003), Quisumbing and Pandolfelli (2010), Deere et al. (2013), Demirguc-Kunt et al.
(2013), Demirguc-Kunt et al. (2014), Gonzales et al. (2015), Christopherson et al. (2022), Hyland
et al. (2020)). In many African countries, there are still laws limiting women’s ownership of land
and other productive assets, and laws on marriage and inheritance giving preference to men and
increasing their bargaining power in the household. There are also discriminatory cultural and
social norms that limit women’s agency and opportunities, such as preference for high fertility,
unequal gender roles, and early marriage (Galor and Weil (1996), Alesina et al. (2013), Gonzales
et al. (2015), Mitra et al. (2020)). Cultural norms that give preference to boys (“son bias”) also
limit girls’ access to education (Hill and Elizabeth (1995), Forbes (2000), Dollar and Gatti (1999),
Knowles et al. (1999), Klasen and Lamanna (2009)). In most countries, the unpaid care and
household work burden disproportionately fall on women, preventing them from fully participating
in and contributing to the economy (Elson (1998), Braunstein et al. (2011), Brixiov´a Schwidrowski
et al. (2021)). Women also face greater health challenges, being more exposed to HIV and facing
much higher rates of GBV globally (Dlamini (2020), Ouedraogo and Stenzel (2021)).
COVID-19 further exacerbated gender inequality. There has already been a body of literature
exploring the disproportionate effects of COVID-19 on women in different countries around the
world, and the evidence is mixed. In most countries like the U.S., UK, Israel, South Africa, and
Lesotho, the evidence points to a “she-cession” where women experienced larger employment losses
103

relative to men. Less-educated women with young children were particularly impacted (Alon et al.
(2020), Bluedorn et al. (2021), Fabrizio et al. (2021), Casale and Posel (2021), Government of
Lesotho (2020)). There were also declines in women’s employment shares relative to men’s in some
sectorssuch as agriculture, health, and real estatewhich is the main driver of the “she-cession” in
these countries (Bluedorn et al. (2021)). Some other countries experienced a “man-cession” because
women were less likely to experience job losses than men (Bluedorn et al. (2021)). Early evidence
from Ethiopia and Nigeria suggests this is because women are mostly self-employed and/or working
in the informal sector and in agriculture (Aoyagi (2021)). Although some research has been done
on gender gaps during the pandemic in South Africa (Bluedorn et al. (2021)), to our knowledge,
there is no empirical evidence of the direct impact of the pandemic on women’s employment and
wages yet.
Three main channels have been identified in the literature through which COVID-19 has disproportionately affected women. First, a heavier childcare and unpaid care work burden typically falls
on women. Before COVID-19, women were doing three times as much unpaid care and domestic
work as men (United Nations (2020)). COVID-19 led to school closures which exacerbated the
childcare burden, mostly falling on women. This factor is particularly critical for single mothers. Second, occupational gender segregation drives the higher likelihood of women losing their
jobs. The COVID-19 pandemic affected sectors and occupations in the services sector (hospitality, restaurants, travel, education, etc.). At the same time, women’s employment dominates the
services sector in many countries around the world. In Sub-Saharan Africa (SSA) in particular, a
sizeable majority of women, about 74 percent, work in the informal sector and women own most of
the small and medium enterprises (SMEs) (Parry and Gordon 2020, United Nations 2020). These
jobs often depend on public space and social interactions, which were severely restricted during the
COVID-19-related lockdowns. Finally, health is a major channel of transmission as well. Women
are faced with more critical demand for health services which were harder to access during the
pandemic. Finally, the pandemic has significantly increased the occurrence of GBV, which has
been shown to decrease economic activity (Ouedraogo and Stenzel (2021)).
104

Chapter Three: Data
We use both macro- and micro-level data to derive some stylized facts. At the macro level, we
obtained the countries’ gender parity index, unemployment rate, labor force participation, employment in services by gender, seats held by women in national parliaments, and health indicators
from the World Development Indicators. We use the World Bank Global Findex to obtain the
percentage of men and women with an account at a financial institution, a mobile money account,
and their borrowing. We also use the Word Bank’s Women, Business and the Law indicators, as
well as the World Economic Forum’s Global Gender Gap. From UNDP, we obtained the Gender
Inequality Index, Gender Development Index, as well as data from the COVID-19 Global Gender
Response Tracker. These aggregate indicators are used to establish pre-existing gender disparities
in the region, as shown by the stylized facts. We also use micro-level data from different sources
for the different countries to derive monthly income and business revenue by gender and sectoral
employment. For Botswana, we obtained the 2005 labor force survey and the 2007 informal sector
survey from the Statistics Botswana database. We use Eswatini’s 2018 Household Income and
Expenditure survey obtained from their Central Statistical Office. For Lesotho, we use the 2019
Labor Force Survey and the 2017 Household Budget Survey, obtained from their Bureau of Statistics. Similarly, Namibia’s Statistics Agency provided their 2018 Labor Force Survey. Finally, for
South Africa, we use data from the National Income Dynamics Survey database and the UNDP
COVID survey.
Furthermore, we use micro-level data to conduct the empirical analysis for Eswatini, Lesotho,
Namibia, and South Africa. For South Africa, we use two panel datasets: the National Income
Dynamics Survey database - specifically their COVID Rapid Mobile Survey (NIDS-CRAM) and the
2020 and 2021 quarterly Labor Force Survey (LFS) from Stats South Africa. The outcome variables
of interest from these two datasets are the employment status of the individual, including all forms
of employment and whether the respondent runs a business, their monthly income, and weekly
hours worked. For Eswatini, we use its Rapid Gender Assessment data from UN Women, collected
between January and February 2021. The dependent variables of interest are the employment status
of the individual, whether the respondent lost some income and their job due to the pandemic, and
105

whether they spent more time on unpaid care work due to the pandemic. For Lesotho, we use the
2020 Rapid Impact COVID survey to obtain data on individuals’ employment status and whether
they stopped working due to the pandemic. Finally, for Namibia, we use its COVID-19 Household
and Jobs Tracker from its National Statistics Agency. COVID-19-period data for Botswana are
currently unavailable.
Chapter Three: Stylized Facts
Pre-existing Gender Inequality
While gender gaps may not seem too striking in the SACU region, aggregate indices mask significant
gender inequalities in several dimensions, such as health, labor markets, and legal empowerment.
At the aggregate level, the SACU countries seem to be doing relatively well in terms of gender
parity with Gender Development Indices (GDI)3
close to 1 for all of them and above the SubSaharan Africa (SSA) and Emerging Markets and Developing Economies (EMDE) averages (see
Figure 3.2). Per the GDI classification, they all belong to group 1, which includes countries that
have high equality in Human Development Index (HDI) achievements between women and men.
Their Global Gender Gap (GGG)4
scores are lower but still relatively high compared to the other
EMDEs. Yet, the SACU region does worse in terms of the Gender Inequality Index (GII)5
compared
to the EMDEs with Botswana, Eswatini, and Lesotho scoring higher values, which indicate worse
inequalities. The education gap seems to have been largely closed, with girls’ enrollment even
higher than boys’ in some countries. However, it is also important to look into disparities in areas
of study, disparities across income quintiles, and across regions and areas of residence (rural vs
urban).
3The GDI is computed as the female-to-male HDI ratio, and HDI captures health measured as life expectancy,
education, and gross national income.
4The GGG captures national gender gaps on economic participation, educational attainment, health, and political
criteria.
5The GII captures gender inequality across areas of health (maternal mortality ratios and adolescent fertility
rates), empowerment (share of parliamentary seats and education attainment at the secondary level for both males
and females), and labor force participation (rates by sex). While the GII has drawbacks (such as a complicated
functional form and a combination of indicators that compare men and women with indicators that pertain only to
women), it is preferable to alternatives such as the GDI in which one of the main components is not observed and is
imputed.
106

Figure 3.2: SACU Countries: Gender Inequality Overview
Notes: 1) Higher scores for Gender Development Index and Global Gender Gap mean better gender
equality, while for Gender Inequality Index they represent worse gender disparities. Global Gender
Gap index are based on Global Gender Gap Reports which reflect values collected in the previous
year (t-1). 2) SACU index is calculated in simple average of all five SACU countries. Index of
World, SSA and developing countries are referred directly from the data source.
Gender gaps persist in the labor market, as there are still major gaps in labor force participation
and unemployment. The female labor force participation rate is lower than the male in all countries
and female unemployment is higher in all SACU countries, except for Namibia (see Figure 3.3).
Women earn less and are more likely to be poor. Micro-level data on monthly income earned reveals
that men earn more than women, and this gap is particularly striking in South Africa. In 2017,
women earned on average about 25 percent of men’s income (see Figure 3.3). Not surprisingly,
women are more likely to live in extreme poverty in all countries, except for Lesotho where extreme
poverty widely affects men and women about equally.
Political representation and empowerment lag behind, in part due to legal restrictions, except for
Namibia and South Africa. Research has shown that legal restrictions in various dimensions are a
driver of gender gaps (Gonzales et al. (2015)). In Lesotho, there was a gender quota system that
improved women’s representation at the local level, but it was revoked before 2011. Thus, there
has been a decline in women’s representation at the local level ever since. There is a quota system
at the national level in Eswatini (but not at the local government); however, it is difficult to reach
107

the percent quota, as there are still barriers to voting and being a candidate due to lack of assets
and property. South Africa, Namibia, and Lesotho seem to be doing better in terms of gender
equality than other countries according to their World Bank Women, Business and the Law (WBL)
index scores (see Figure 3.3). Eswatini has a very low score in the marriage indicator because of
the strong patriarchal culture driven by the political system. Botswana’s and Eswatini’s scores are
very low on the workplace dimension, and Botswana’s, Eswatini’s, and Lesotho’s scores are also
low on the parenthood indicator. The dual customary and civil legal systems present a problem for
some countries in the region, e.g., Lesotho and Eswatini but are not considered in the compilation
of the WBL indices (see Chapter 3 Appendix).
Figure 3.3: SACU Countries: Gender Inequality on Business and Political Activities
The gender gap in access to finance persists in most countries and women face more difficulties
in running profitable businesses. Although women are as likely as men to have bank or mobile
money accounts overall, there are fewer women who borrow money than men (see Figure 3.4). For
108

Figure 3.4: SACU Countries: Gender Inequality on Financial Activities
instance, in Botswana and South Africa respectively, 4 and 8 percent of women borrowed from a
financial institution in 2017, as opposed to 7 and 11 percent of men, respectively. Micro-level data
reveals that women generate less revenue and profit from running businesses in both Namibia and
South Africa, and that may be due to their limited access to financing or to the specific sectors of
operation of their businesses.
Impact of COVID-19 on Gender Inequality
Women are predominantly employed in services, the informal, and the contact-intensive sectors
that were most negatively affected by lockdown policies across the region. Women’s share of
employment in services is higher than men’s in all countries, and it is particularly high in South
Africa at 84 percent. Informal employment is also significant. In all countries except South Africa,
more than 50 percent of women are employed in the informal sector. This may explain why recent
studies on the impact of COVID-19 on labor markets do not find a significant drop in women’s
employment—women cannot lose (formal) jobs they never had (Aoyagi (2021)). A more granular
look, based on microdata (see Figure 3.5) shows that women are employed in sectors such as
education and health, hospitality and private households— all sectors that are contact-intensive,
109

likely to be impacted by lockdown policies, and to recover slowly even after economies reopen.
Most of the unpaid domestic work burden falls on women, keeping them out of the labor force. From
micro-level data in all SACU countries except for Botswana, out of the unemployed respondents
whose main reason for not working is unpaid domestic work, the overwhelming majority, over 70
percent, are women.
Figure 3.5: SACU Countries: Gender Inequality on Various Employment Sectors
Pre-existing health challenges and GBV complicate the COVID-19 response and may exacerbate
its impact. For all five countries, about 60 percent or more of people above 15 who live with HIV
are women (see Figure 3.6). The maternal mortality rate is also above regional and economic peers.
These health disparities may complicate the COVID-19 response, which may also have exacerbated
access to healthcare for chronic diseases other than COVID, such as HIV/AIDS, tuberculosis, as
well as maternal complications.
110

Figure 3.6: SACU Countries: Gender Inequality in Health
GBV is prevalent in all SACU countries, and as a result of restrictions on movement, GBV is
exponentially on the rise globally (United Nations (2020)). Some striking statistics point to the
high prevalence of GBV in these countries. In Botswana, 67 percent of women have experienced
abuse (UNFPA). In Eswatini, 48 percent of women have experienced sexual violence in their lifetime
(UNFPA). In Lesotho, 33 percent of women have been abused by an intimate partner (UNAIDS).
In Namibia, 33 percent of married women aged 15-49 years have experienced violence from their
partner (Namibia DHS 2013). In South Africa 25 to 40 percent of women have experienced violence
from their partner (SaferSpaces).
Table 3.1: UNDP-UN Women COVID-19 Global Gender Response Tracker
Despite the fact that women were observed to be more vulnerable to COVID-19, countries implemented few policies that directly supported women’s economic activity and health in response to
111

this shock. According to the UNDP-UN Women COVID-19 Global Gender Response Tracker (Table 3.1), only Botswana and South Africa implemented policies to address violence against women
by providing hotline services to facilitate reporting of GBV cases. Lesotho addressed women’s
economic security in their COVID-19 policy measures through their gender-sensitive grants to micro, small and medium businesses targeting accommodation and food services activities, in which
women employment is heavily concentrated. South Africa also targeted women’s economic security
through its debt relief that prioritizes women owned MSMEs (see Chapter 3 Appendix).
Chapter Three: Empirical Strategy: Assessing the Impact of COVID19 on Gender Inequality
We model our empirical analysis following Alon et al. (2021) as follows:
yit = β0 + β1Fi + β2Dt + β3Fi × Dt + β4Xit + ϕi + ϵit (3.1)
for individual i at time t, where y is our outcome variable of interest, namely a binary employment
indicator, weekly hours worked, and nominal income in local currency; F is a dummy equaling 1 if
the respondent is female and 0 otherwise; D is a treatment dummy for post-COVID time period and
equals 0 before COVID (quarter 1 of 2020) and 1 after the onset of the pandemic; ϕi are individual
fixed-effects; and X is vector of controls: age, marital status, education, area of residence (rural vs
urban), province, ethnicity. Standard errors are clustered at the individual level. Since our control
population is the male population, the impact of COVID-19 on any outcome variable of interest for
the male population is captured by β2. For the female population, it is captured by the sum of β2
and β3. So, β3 captures the disproportionate impact of COVID-19 on women relative to men and
is, therefore, our main parameter of interest. This empirical analysis can only be done in countries
with pre- and post-COVID-19 data.
We have additional specifications to capture the channels of transmission of the COVID-19 shock.
We control for the industry in which individual i is employed and their occupation to assess the
role of occupational gender segregation in the differential effects of COVID on women compared
112

to men. We classify the industries and occupations into different major categories and include a
dummy for each group of industry or occupation, holding one group constant. The specification
controlling for the employment industries is as follows:
yit = γ0 + γ1Fi + γ2Dt + γ3Fi × Dt + γ4industry1it + γ5Fi × industry1it + γ6industry1it × Dt +
γ7Fi × Dt × industry1it + γ8industry2it + γ9Fi × industry2it + γ10industry2it × Dt + γ11Fi × Dt ×
industry2it + γ12Xit + µit
where industry0 is the control industry group. Similarly, we obtain the following specification for
occupation:
yit = δ0 + δ1Fi + δ2Dt + δ3Fi × Dt + δ4occupation1it + δ5Fi × occupation1it + δ6occupation1it ×
Dt + δ7Fi × Dt × occupation1it + δ8occupation2it + δ9Fi × occupation2it + δ10occupation2it × Dt +
δ11Fi × Dt × occupation2it + δ12Xit + µit
where occupation0 is the control occupation group. We also control for the kind of employment
respondents have, specifically whether they are own-account workers, or working for a wage, or are
employers, as follows:
yit = η0+η1Fi+η2Dt+η3Fi×Dt+η4jobkind1it+η5Fi×jobkind1it+η6jobkind1it×Dt+η7Fi×Dt×
jobkind1it+η8jobkind2it+η9Fi×jobkind2it+η10jobkind2it×Dt+η11Fi×Dt×jobkind2it+η12Xit+µit
To estimate the effect that school closures and increased childcare burdens had on the differential
impact of the pandemic on women, we control for the number of children under seven years old
that the individual has and the number of children attending school living in the household, as
shown in the equation below. This data is only available in the NIDS dataset.
yit = θ0 + θ1Fi + θ2Dt + θ3childrenit + θ4Fi × Dt + θ5Fi × childrenit + θ6childrenit × Dt + θ7Fi ×
Dt × childrenit + θ8Xit + µit
Finally, we control for the increased household work burden by including time spent fetching wood
and water, which is available in the LFS data.
yit = λ0 + lambda1Fi + lambda2Dt + lambda3timeit + lambda4Fi × Dt + lambda5Fi × timeit +
lambda6timeit × Dt + lambda7Fi × Dt × timeit + lambda8Xit + µit
113

We run panel regressions following equation 3.1 for South Africa and Namibia, and we complement our analysis with probit models for Eswatini, Lesotho, and Namibia due to data limitations.
Specifically, for Eswatini, Lesotho, and Namibia, we run the following specification:
Yi = P r(Yi = 1Xi) = ϕ(Xiβ) (3.2)
for each individual i, where Yi
is a binary outcome variable, such as whether the respondent is
employed or not, whether they lost income or their job, or whether they spent more time doing
unpaid work; Xi
is a vector of explanatory variables, the main one of interest being gender, but
also including controls such as age,marital status, education, and area of residence. Coefficients β
measure the effects of the corresponding explanatory variables on Yi
.
Chapter Three: Results
South Africa
Our analysis for South Africa shows that women were more affected than men by the COVID-19
pandemic, experienced a larger loss in employment, reduction in weekly hours worked, and loss in
income. Using the two data sources, the NIDS-CRAM and the Quarterly LFS, we examine the
outcome variables of interest available in both surveys, the employment status of the respondent,
including all forms of employment, whether the respondent runs a business, and the weekly hours
worked. Additionally, the NIDS-CRAM has data on monthly income, which we also exploit in our
analysis. The sample size varies between 45,411 and 9,006 respondents for the NIDS-CRAM survey
and between 255,619 and 59,389 respondents for the LFS, depending on the different questions.
In this section, we present the results for South Africa for each dependent variable and for both
surveys.
1. Impact of the COVID-19 pandemic on employment
Our first outcome variable of interest is employment of all forms. In both surveys, respondents
are asked about their employment status for specific types of employment. In the NIDS-CRAM,
114

the types of employment are paid employment, business, and any other job (including casual labor
and/or unpaid work). In the LFS, the types of employment are paid employment, business, and
unpaid work. Our first outcome variable of interest includes employment of any kind and equals
1 if the respondent is employed and 0 otherwise. We run a linear probability panel regression
model following equation 3.1 in our baseline specification, then control for the different channels of
transmission in the additional specifications. The results using the NIDS-CRAM and the LFS are
reported in Tables 3.2 and 3.3, respectively.
a. NIDS-CRAM
The COVID-19 shock reduced women’s employment significantly more than men’s (Table 3.2,
baseline specification in column (1)). The analysis shows that the COVID-19 shock decreased the
probability of being employed for men by 12.7 percentage points, for women by 14.6 percentage
points, and thus decreased the probability of being employed for women by 1.94 percentage points
more than men.
Controlling for the channels of transmission shows that the kind of employment, occupation, children, and education explain the disproportionate negative effect of the pandemic on women. This
granular look at the differential effects of the pandemic on men and women based on different
characteristics shows not only that women’s employment overall was more affected than men’s by
COVID-19, but also that the effect is significant even after holding constant the specificities of
their employment, children, and education level. For each channel, we are mainly interested in
the disproportionate effect of the pandemic on women within each group relative to their male
counterparts, which is given by the sum of the coefficients on the double interaction between the
gender and COVID time dummies and on the triple interaction between these two variables and the
channel. Although the coefficient estimates of some of the channels have relatively large standard
errors, we confirm the joint significance of the sums of these coefficients with separate statistical
tests on each sum. We also note that once we control for some channels of transmission, such as
children and education, our main coefficient of interest on the double interaction between gender
and the COVID dummies is no longer significant.
This suggests that although the causal channel of the disproportionate effect of the pandemic is not
115

gender, women are still more negatively affected by the pandemic through these different channels.
Table 3.2: South Africa: NIDS-CRAM COVID-19 Effects on Employment of All Forms
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Dependent variable: employment of all forms. Gender=0 if male and 1 if female. Controls are age,
education level, marital status and province. Column (1): baseline specification. Columns (2)-(7): control for
different channels of transmission. Column (2): kind of employment, regular job is the control group; columns
(3) and (4): occupation for wage employment and business employment, technicians and professionals is the
control group; columns (5)-(7): number of children below 7, number of school children living in the household,
and education level. R-squared is the within one.
The disproportionately negative effect of the pandemic on women’s employment is the largest
116

among the self-employed (column (2) Table 3.2). First, we include the respondents’ different types
of employment, namely regular job, casual work, self-employment, and business employment.6
In
column (2), we see that the employment of women having regular jobs and casual work, and selfemployed women were more negatively affected than their male counterparts. There is no evidence
that women running businesses were less or more affected relative to their male counterparts.
Furthermore, women with regular jobs in low-skilled occupations were more affected than their
male counterparts. In columns (3) and (4) of Table 3.2, we see that women with regular jobs
in low-skilled and medium-skilled occupations were more affected than their male counterparts.
Furthermore, women with children below seven years were also more affected than men in the same
situations (columns (5) and (6)). Similarly, more educated women were more affected than more
educated men (column (7)).
b. LFS
The analysis using the LFS does not provide significant evidence, at the aggregate level, that
women’s employment was disproportionately more affected by COVID-19 (Table 3.3). These results
do not contradict the previous results from the NIDS-CRAM dataset, as the sign of the estimate
is still negative (consistent with the previous findings), but there is large uncertainty around the
coefficient estimate.
However, once we take a more granular look at how men and women were affected differently across
different categories, there is evidence that the kind and sector of employment and occupation are
channels of transmission. In all kinds of employment (except unpaid work), women’s employment
was disproportionately more affected by the pandemic (column (2) Table 3.3). The disproportionate negative effect on women is largest among self-employed women, followed by employers, and
then employed women (women working for someone else for pay). The LFS also has information
on the sectors of employment, which include agriculture, services, mining, manufacturing, and others. Drawing on this, our analysis suggests that women’s employment was more disproportionally
6For occupation (in the case of a regular job) and business occupation (in the case of a business job or selfemployment), the classification follows the International Standard Classification of Occupation (ISCO) and we classify
the different categories based on the skill level. Managers, professionals, technicians, and associate professionals
constitute the high-skilled occupations; clerical support workers, service and sales workers, craft and related trades
workers, armed forces occupations, plant and machine operators, and assembly, and skilled agricultural, forestry and
fishing workers are the medium-skilled ones; and elementary occupations are the low-skilled occupations.
117

Table 3.3: South Africa: LFS COVID-19 Effects on Employment of All Forms
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Dependent variable: employment of all forms. Gender=0 if male and 1 if female. Controls are age,
education level, marital status and ethnicity. Column (1): baseline specification. Columns (2)-(6): control for
different channels of transmission. Column (2): the kind of employment, working for someone else for pay is
the control group; column (3): sector of employment, agriculture is the control group; column (4): occupation,
managers, professionals and technicians is the control group; columns (5) and (6): time spent fetching water and
wood, and education level. R-squared is the within one.
118

affected than men’s in all sectors, except mining (column (3) Table 3.3). Women working in agriculture were the most disproportionately affected by the pandemic relative to women in manufacturing
and services. In terms of type of occupation, women were disproportionately more affected by the
pandemic in jobs at all skill levels, with this negative effect being the largest in low- and then in
medium-skilled occupations (column (4) Table 3.3).
Finally, our analysis points out that women that spend more time on domestic chores are disproportionally affected compared to their male counterparts (column (5) Table 3.3). Another interesting
addition of the LFS data to our analysis is the availability of time-use data, specifically the time
respondents spend fetching water and wood. Women who spend more time fetching water and
wood were disproportionately affected by the pandemic compared to men in the same situations.
However, the disproportionate negative impact of the pandemic on women relative to men is larger
for women who do not spend time fetching water and wood. The latter result can be explained by
the fact that women who spend more time fetching water and wood are less likely to be employed
even before the pandemic, thus less likely to lose employment because of the pandemic.
2. Impact of the COVID-19 pandemic on business employment
a. NIDS-CRAM
Our analysis suggests that women running businesses were not disproportionately affected by
COVID-19, even after controlling for different channels of transmission (Table 3.4). These businesses are mostly informal and these day-to-day activities, although very vulnerable to the pandemic
shock, can still be carried out despite COVID-related restrictions, by nature of being informal. For
instance, women selling vegetables at the market may be able to still carry out these activities
despite the lockdown, albeit they might see a decrease in their hours worked or income from these
activities due to reduced mobility and economic activity. Our analyses controlling for the different channels of transmission in columns (2)-(5) suggest that even after controlling for occupation,
children, and education, women’s business employment was still not disproportionately affected
relative to men’s.
b. LFS
119

Table 3.4: South Africa: NIDS-CRAM COVID-19 Effects on Business Employment
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Dependent variable: business (self ) employment. Gender=0 if male and 1 if female. Controls are age,
education level, marital status and province. Column (1): baseline specification. Columns (2)-(5): control for
different channels of transmission. Column (2): occupation business employment, technicians and professionals is
the control group; columns (3)-(5): number of children below 7, number of school children living in the household,
and education level. R-squared is the within one.
The results obtained from the LFS are similar to those from the NIDS-CRAM survey. In the
baseline specification in Table 3.5, women running businesses were not more affected than men by
the pandemic.
However, the sector of business employment and occupation seem to be channels through which
the pandemic affected women’s businesses more relative to men’s (columns (3) and (4) Table 3.5).
Controlling for sector and occupation reveals that women running businesses in manufacturing and
other sectors, as well as those in low-skilled occupations, were disproportionately more affected by
120

Table 3.5: South Africa: LFS COVID-19 Effects on Business Employment
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Dependent variable: business employment. Gender=0 if male and 1 if female. Controls are age, education
level, marital status and ethnicity. Column (1): baseline specification. Columns (2)-(6): control for different
channels of transmission. Column (2): the kind of employment, working for someone else for pay is the control
group; column (3): sector of employment, agriculture is the control group; column (4): occupation, managers,
professionals and technicians is the control group; columns (5) and (6): time spent fetching water and wood, and
education level. R-squared is the within one.
121

the pandemic than their male counterparts.
3. The role of informal versus formal businesses
A more detailed analysis points to informal businesses acting as a shock absorber of the impact of
the pandemic on women’s employment. To paint a more complete picture, we took a closer look
at the disproportionate impact of the pandemic on women running formal vs. informal businesses.
First, the informal businesses far outnumber the formal ones in the South African data: in the
NIDS-CRAM, there are 16 formal businesses and 398 informal ones, while in the LFS, there are
1,759 formal ones and 7,250 informal ones. We run our baseline specification for informal and
formal businesses separately, where the binary business employment variable equals 1 if the person
runs an informal and formal business, respectively, and 0 otherwise. The results are reported in
Tables 3.6 and 3.7 for the NIDS-CRAM and LFS datasets, respectively.
Table 3.6: South Africa: NIDS-CRAM COVID-19 Effects on Informal and Formal Businesses
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Dependent variable: business employment.
Gender=0 if male and 1 if female. Controls are age, education level, marital status and ethnicity. Columns
(1), (3) and (5): businesses not registered for VAT. Columns (2), (4) and (6): businesses registed for VAT.
R-squared is the within one. Columns (1)-(2): standard errors are clustered at the individual level. Columns
(3)-(4): standard errors are clustered at the occupation level. Columns (5)-(6): standard errors are clustered at
the province level.
From the NIDS-CRAM data, there is significant evidence that women running informal businesses
are less disproportionately affected by the pandemic relative to men running informal businesses
(columns (1) and (2) Table 3.6). However, there is no disproportionate impact of the pandemic
122

Table 3.7: South Africa: LFS COVID-19 Effects on Informal and Formal Businesses
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Dependent variable: business employment.
Gender=0 if male and 1 if female. Controls are age, education level, marital status and ethnicity. Columns (1),
(3), (5) and (7): businesses not registered for VAT. Columns (2), (4), (6) and (8): businesses registed for VAT.
R-squared is the within one. Columns (1)-(2): standard errors are clustered at the individual level. Columns
(3)-(4): standard errors are clustered at the occupation level. Columns (5)-(6): standard errors are clustered at
the province level. Columns (7)-(8): standard errors are clustered at the sector level.
on women running formal businesses. Clustering the standard errors at different levels, namely
occupation and province strengthens our results (Table 3.6). Clustering standard errors at these
different levels allows us to control for any correlation between employment outcomes of individuals
in the same occupation, province, or sector. Clustering at the occupation level gives a significant
negative effect of the pandemic on women’s relative to men’s formal businesses, which provides
evidence of the disproportionate negative effect of the pandemic on women’s formal businesses
(column (4) Table 3.6).
The LFS data does not offer significant evidence of the disproportionate effects of the pandemic on
either women-run formal or informal businesses (Table 3.7). However, our results from the NIDSCRAM offer some evidence to suggest that informal businesses served as a cushion for women
against the pandemic shock.
This evidence suggests that informal (but not formal) businesses serve as a cushion for the pandemic shock, particularly for women. This finding does not necessarily contradict recent research
on Northern Africa suggesting that informal employment did not increase and thus, it was not
countercyclical during the COVID-19 pandemic (IMF (2021b)). Our findings are specifically on
123

women’s business employment vis-`a-vis men’s, which did not decrease in the case of informal businesses relative to men’s business employment. The results provide evidence supporting the finding
of Lambert et al. (2020) that informality plays an important role as a shock absorber, limiting
unemployment and demonstrate that this effect is stronger for women than men in South Africa.
4. Impact of the COVID-19 pandemic on weekly hours worked
In addition to employment, we also investigate the impact of the pandemic on women’s hours
worked, in order to look at the intensive margin of the effects of COVID-19 on employment.
a. NIDS-CRAM
We find that the pandemic decreased women’s relative to men’s weekly hours worked by 1.2 hours
in the baseline specification (column (1) of Table 3.8). Therefore, not only did the pandemic drive
women out of work to a greater extent compared to men, but it also disproportionately decreased
their aggregate hours worked, relative to men.
This disproportionate negative effect is driven by two main channels, occupation and children.
Indeed, women in low-skilled occupations were more affected than their male counterparts and
women with children under seven years were more affected than men in the same circumstance
(columns (3) and (4) Table 3.8). This suggests that the occupational gender differences and the
childcare burden that falls mostly on women explain the fact that women worked fewer hours due
to the pandemic relative to men.
b. LFS
Similarly, the LFS also provides evidence that the pandemic affected women’s hours worked more
than men’s. Our results in Table 9 suggest that the COVID-19 pandemic decreased women’s weekly
hours worked by about 1 hour relative to men, which is of a similar magnitude to the findings from
the NIDS-CRAM above.
Data from the LFS provides more evidence on the channels through which the pandemic affected
women’s hours worked, with the kind and sector of employment, occupation, domestic work, and
education as the significant ones. When we consider the kind of employment, women doing regular
124

Table 3.8: South Africa: NIDS-CRAM COVID-19 Effects on Weekly Hours Worked
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Dependent variable: weekly hours worked. Gender=0 if male and 1 if female. Controls are age, education
level, marital status and province. Column (1): baseline specification. Columns (2)-(6): control for different
channels of transmission. Column (2): kind of employment, regular job is the control group; column (3): occupation for wage employment, technicians and professionals is the control group; columns (4)-(6): number of
children below 7, number of school children living in the household, and education level. R-squared is the within
one.
jobs were particularly disproportionately negatively affected by the pandemic (column (2) Table
3.9). Another expected result is the fact that women in services were disproportionately more
affected by the pandemic than their male counterparts (column (3) Table 3.9). Women in mediumand low-skilled occupations were also disproportionately more affected by the pandemic, and the
125

Table 3.9: South Africa: LFS COVID-19 Effects on Weekly Hours Worked
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Dependent variable: weekly hours worked. Gender=0 if male and 1 if female. Controls are age, education
level, marital status and ethnicity. Column (1): baseline specification. Columns (2)-(6): control for different
channels of transmission. Column (2): the kind of employment, working for someone else for pay is the control
group; column (3): sector of employment, agriculture is the control group; column (4): occupation, managers,
professionals and technicians is the control group; columns (5) and (6): time spent fetching water and wood, and
education level. R-squared is the within one.
126

disproportionate negative effect on women is biggest in medium-skilled occupations, which is consistent with what was found using the NIDS-CRAM (column (4) Table 3.9). Again, consistent
with what was found on employment, women who spend more time fetching water and wood were
more affected than men in the same situation (column (5) Table 3.9). More educated women were
less disproportionately affected by the pandemic relative to less educated women, as higher levels
of education decrease the significant disproportionate negative effect of the pandemic on women
(column (6) Table 3.9).
5. Impact of the COVID-19 pandemic on monthly income (data only available from NIDS-CRAM)
Women’s monthly net income decreased by 226.3 Rands beyond men’s income because of the
pandemic, which is about $13 (Table 3.10 column (1)). This might not seem like a big relative
decrease; however, given that the average monthly income for women in the sample is R3780 (about
$239), this is equivalent to an additional decrease of 5 percent of their income beyond that of men.
Even though self-employed women were less affected than self-employed men, the overall negative
effect on women is driven by the other kinds of employment and occupations, as women in lowskilled occupations were more affected than their male counterparts (columns (2) and (3) Table
3.10). Similarly, women with children under seven years lost more income than men in a similar
circumstance (column (4) Table 3.10). The disproportionate effect of the pandemic on women is
accentuated by the number of children in school, as having more children attending school increases
the disproportionate negative effect of the pandemic on women’s income (column (5) Table 3.10).
Overall, these results on South Africa suggest that women suffered more from the pandemic in
terms of their employment status, hours worked, and income earned. There is enough evidence to
suggest that these disproportionate effects are due to sectoral and occupational gender differences
and childcare and unpaid work burdens. In other words, there is evidence of a “she-cession” in
South Africa. However, there is some evidence to suggest that business employment, especially in
informal businesses, acted as a counter-cyclical shock-absorber for women. Generally, women in
low- and medium-skilled occupations, and women with children were particularly affected. On the
other hand, education reduces the disproportionate negative effect of the pandemic on women.
127

Table 3.10: South Africa: NIDS-CRAM COVID-19 Effects on Monthly Income
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Dependent variable: weekly hours worked. Gender=0 if male and 1 if female. Controls are age, education
level, marital status and province. Column (1): baseline specification. Columns (2)-(6): control for different
channels of transmission. Column (2): kind of employment, regular job is the control group; column (3): occupation for wage employment, technicians and professionals is the control group; columns (4)-(6): number of
children below 7, number of school children living in the household, and education level. R-squared is the within
one.
128

6. Robustness Checks
We run a few additional specifications as robustness checks for the results found on South Africa,
first by clustering standard errors at different levels, and we obtain similar results.7 Specifically,
we cluster the standard errors at the province, occupation, and sector levels for the LFS data. The
results are reported in Chapter 3 Appendix, Tables 59-61. Generally, clustering standard errors at
different levels does not change our results qualitatively for most dependent variables. Therefore,
these robustness checks largely support the main results.
We further look into the role that marriage plays in the impact of the pandemic on women, by
controlling for the interaction term of marital status, gender, and the post-COVID time dummies.
We find that being married decreases the disproportionate negative effect of COVID on women for
business employment (Table 62). This may be due to the fact that being married can decrease the
unpaid care and domestic work burden if these responsibilities are shared with a partner.
Moreover, controlling for age squared allows us to look at the non-linear effect of age on our outcome
variables of interest. Our main results do not change qualitatively, and we find further evidence
that younger people are less likely to be employed, earn less, and work fewer hours, but there is
some evidence from the NIDS data that as people get older, the marginal effect of age decreases
(Tables 63 and 64). In other words, as one gets older, they work and earn more, but the additional
positive effect of age on their employment outcomes decreases.
Finally, to verify that the disproportionate negative effect of the pandemic on women is not solely
coming from other observables such as education, we run the same specification as in equation (1)
but include an interaction term of the COVID dummy and these observables, namely education,
age, province, marital status, and ethnicity. We report the results for education for the NIDSCRAM and LFS datasets in Tables 65 and 66, respectively. These additional controls do not
change our results, as the disproportionate negative effect of the pandemic on women’s weekly
hours worked, and on women’s employment and monthly income (for the NIDS-CRAM) are still
significant. The other results are not reported in the paper but are available upon request from
the authors. However, we found that controlling for the interaction between the COVID dummy
7Clustering standard errors at these different levels allows us to control for any correlation between employment
outcomes of individuals in the same occupation, province, or sector.
129

and age, marital status, province, and ethnicity does not substantially alter our results, except
when we control for the effect of age using the NIDS-CRAM dataset. In fact, after including
the interaction term of the COVID dummy and age, the disproportionate effect of COVID-19 on
women’s employment and weekly hours worked disappears, and age becomes the main channel of
the disproportionate effect of the pandemic. This, however, does not suggest that there is no gender
effect of the pandemic, but rather that the effect is also driven by age.
7. Dynamic Effects of the Pandemic
We conduct additional regression analyses using the same outcome variables of interest but treating
the COVID-19 dummy as a categorical variable, instead of a continuous one. The results in Tables
3.11 and 3.12 show the effects of the pandemic on women across different waves of the surveys.
The NIDS-CRAM was conducted according to these waves: wave 0 is February 2020 (the baseline,
pre-COVID-19 period), wave 1 is April 2020, wave 2 is June 2020, wave 3 is October 2020, wave
4 is January 2021, and wave 5 is March 2021. For the LFS, time 0 is quarter 1 of 2020, the
benchmark pre-COVID period. Times 1, 2, 3, and 4 are quarters 2, 3, 4 of 2020 and quarter 1 of
2021, respectively.
a. NIDS-CRAM
Women’s employment and income take longer to recover from the pandemic shock than men’s
(Table 3.11). Women’s employment of all forms is not significantly more affected by the pandemic
in the first few months after the onset of the pandemic but is in the last waves of the survey (column
(1) Table 3.11). There is no disproportionate effect of the pandemic on women’s businesses across
all waves (column (2) Table 3.11). Just like employment, women’s income also takes longer to
recover as the disproportionate effect of the pandemic becomes significant and negative in the last
two waves (column (3) Table 3.11). Specifically, women significantly lost R941.1 and 506.4 of
monthly income more than men in January and March 2021 respectively. Women’s weekly hours
worked significantly decreased relative to men’s in the first and second-to-last waves, by 4 and 2
hours, respectively (column (4) Table 3.11).
b. LFS
130

Table 3.11: South Africa: NIDS-CRAM Dynamic Effects on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. The time variable here is treated as a categorical variable instead of a continuous one, and the categories
are the different waves of the survey (wave 0 in February 2020 (that’s the baseline, pre-COVID period), wave 1 in
April 2020, wave 2 in June 2020, wave 3 in October 2020, wave 4 in January 2021, and wave 5 in March 2021).
Columns (1)-(5) dependent variables: employment of all forms, formal wage employment, business employment,
monthly income and weekly hours worked. Gender=0 if male and 1 if female. Controls are age, education level,
marital status and province. R-squared is the within one.
Women’s weekly hours worked decreased more than men’s from the onset of the pandemic and
persisted, while their employment was negatively affected only in the third quarter of 2020 and
the first quarter of 2021 (Table 3.12). Specifically, the pandemic decreased women’s weekly hours
worked by 1.6 hours more than men’s, and this effect dropped to 0.5 hours by the fourth quarter
of 2020; however, it increased again to 1.4 hours in the first quarter of 2021 (column (3) Table
3.12). Women’s business employment was not disproportionately affected by the pandemic in any
quarters, in contrary, they were significantly less affected that men’s in the second quarter of 2020
(column (2) Table 3.12). The results for employment of all forms, business employment, and weekly
131

hours worked are consistent with the ones found using the NIDS-CRAM data.
Table 3.12: South Africa: LFS Dynamic Effects on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. The time variable here is treated as a categorical variable instead of a continuous one, and the categories
are the different quarters of the survey. Columns (1)-(4) dependent variables: employment of all forms, formal
wage employment, business employment, and weekly hours worked. Gender=0 if male and 1 if female. Controls
are age, education level, marital status and ethnicity. R-squared is the within one.
Eswatini: Probit Model on 2021 COVID-19 Rapid Gender Assessment
We conduct empirical analysis on data from UN Women’s 2021 COVID-19 Rapid Gender Assessment, collected in Eswatini between January and February 2021. Unfortunately, due to data
limitations, we are unable to conduct our preferred empirical analysis as applied to the data from
South Africa above. Instead, we estimate probit models on binary response questions data, as
specified in equation 3.2, and obtain results on the probabilities that gender may have had a dif132

ferential effect on the outcome of interest. The dependent variables in Table 3.13 are dummies
for whether the respondent is employed (columns 1-3), has lost income (column 4-6), and has lost
their job (columns 7-9) due to the pandemic. Along with gender, we also control for age, marital
status, education level, and urban-rural area. The sample size varies between 20,957 and 10,218
respondents for the different questions. For ease of interpretation, we present in the results tables
the coefficients obtained from the probit regressions, along with the corresponding odds ratios and
marginal effects.
The results in Table 3.13 suggest that the COVID-19 pandemic disproportionately affected women
relative to men in Eswatini in terms of employment; however, men were significantly more impacted
in terms of income loss. Specifically, the odds ratios imply that during the pandemic, men were
1.1 times8 more likely to be employed, while women were 1.3 times more likely to have lost their
jobs (columns (2) and (8) Table 3.13). The marginal effects suggest that being a woman decreases
the probability of being employed by 1.6 percentage points and increases the probability of losing
employment due to COVID-19 by 5.8 percentage points (columns (3) and (9) Table 3.13). However,
men in Eswatini were 1.1 times more likely to lose income as a result of the pandemic (column
(5) Table 3.13). In fact, being a woman decreases the probability of losing income due to COVID19 by 4 percentage points (column (6) Table 3.13). Since women in Eswatini are more likely to
be unemployed or out of the labor force, relative to men, these findings imply that the pandemic
further exacerbated these preexisting labor market inequalities. However, COVID-19 may have also
had a positive mechanical effect on the preexisting gender inequality in incomes, whereby men’s
incomes seem to have been more likely to be negatively impacted due to the pandemic.
In addition, the other controls have largely the expected effects on employment and income. Age
increased the probability of being employed and decreased the probability of a job loss, while
being married had a positive effect on the probability of being employed, a negative effect on the
probability of losing a job, yet a positive effect on the probability of income loss. Higher education
levels also decreased the probabilities of both income and job loss, while living in an urban area
decreased the probability of being employed.
8When the odds ratio is ¡1, we take the reciprocal, so here 1/0.915=1.09, and interpret it as lesser probability of
women to have the outcome in question. When the odds ratio is ¿1, we take the odds ratio as is, implying women
are more likely to have the outcome in question.
133

Table 3.13: Eswatini: Effect of Gender Inequality on Employment and Work
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Probit models. Dependent variables: columns (1)-(3): whether the person is employed; columns (4)-(6):
whether the person has lost some income due to the pandemic; columns (7)-(9): whether the person has lost their
job due to the pandemic. Gender=0 if male and 1 if female. Marital status=0 if the person is single, divorced of
widowed, and 1 if married. Urban/ rural=0 if rural and 1 if urban.
Second, the 2021 COVID-19 Rapid Gender Assessment data allows us to estimate similar probit
regressions on outcomes of domestic and care work. This is one of the main hypothesized channels
of the disproportionate effect of the pandemic on women, so the availability of these data is a
major advantage of this survey. The dependent variables in Table 3.14 are dummies for whether
the respondent spent more time on domestic work (columns 1-3), on childcare (column 4-6), and
on adult care (columns 7-9) due to the pandemic.
As expected, the results suggest that as a result of COVID-19, women’s time spent doing domestic,
childcare and adult care work increased relative to men’s. Specifically, women were about 1.1 times
more likely to have spent more time doing domestic, childcare, and adult care work than men.
In other words, being a woman increased a respondent’s probability of having spent more time
doing domestic, childcare, and adult care work due to the pandemic by about 3.8, 1.8, and 2.9
percentage points respectively. These results confirm the hypothesis that one of the ways in which
the pandemic disproportionately affected women in Eswatini is by increasing their unpaid care work
burden, including domestic and care work.
The other controls have positive effects on domestic and care work. Being married decreased
134

the probability of spending more time on both domestic and adult care work. Moreover, higher
education had a positive effect on the probabilities of spending more time on all three types of
unpaid work. Living in an urban area likewise had a positive effect on the probability of more time
spent doing domestic work.
Table 3.14: Eswatini: Effect of Gender Inequality on Domestic and Care Work
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Probit models. Dependent variables: whether the person spent more time on domestic work (columns
(1)-(3)), childcare (columns (1)-(3)) and adult care work (columns (1)-(3)) due to the pandemic. Gender=0 if
male and 1 if female. Marital status=0 if the person is single, divorced of widowed, and 1 if married. Urban/
rural=0 if rural and 1 if urban.
Lesotho: Probit Model on 2020 COVID-19 Rapid Impact Survey
Similarly to the empirical analysis conducted on the Eswatini data, for Lesotho we estimate simple
probit models using the 2020 Rapid Impact COVID survey data. Again, we use binary response
questions data to obtain results on the probabilities that gender may have had a differential effect
on three labor market outcomes of interest. Specifically, in Table 3.15, the dependent variables are
dummies for whether the respondent is formally employed (columns 1-3), runs a business (column
4-6), and stopped working due to the pandemic (columns 7-9). Along with gender, we also control
for age, education level, and district. The sample size varies between 3,149-1,993 respondents for
the different questions. For ease of interpretation, we again present the coefficients obtained from
the probit regressions, along with the corresponding odds ratios and marginal effects.
135

The results suggest that during the pandemic, women were less likely to be formally employed than
men, more likely to run a business, and less likely to have stopped working due to COVID-19.
Specifically, men were 1.5 times more like to be formally employed but also 1.3 times more likely
to have stopped working due to COVID-19 (columns (2) and (8) Table 3.15). On the other hand,
women were 2.2 times more likely to run a business (column (5) Table 3.15). In other words,
being a woman decreases the probability of having a formal job by 13.8 percent but increases
the probability of running a business by 28.4 percentage points (columns (3) and (6) Table 3.15).
Being a woman also decreases the probability of having stopped working due to the pandemic by
3.7 percent (column (9) Table 3.15).
As expected, considering Lesotho’s public-sector-driven economic model, higher education levels
increase the probability of formal employment and decrease the probability of running a business.
Higher education levels also increase the probability of having stopped working due to the pandemic.
Age has a small negative effect on the probability of formal employment and a small positive effect
on the probability of being a business owner.
Table 3.15: Lesotho: Effect of Gender Inequality on Employment and Work
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Probit models. Dependent variables: columns (1)-(3): whether the person is formally employed; columns
(4)-(6): whether the person runs a business; columns (7)-(9): whether the person stopped working due to the
COVID-19 outbreak. Gender=0 if male and 1 if female.
136

Table 3.16: Namibia: Effect of Gender Inequality on Employment Types and Sectors
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects. Standard errors clustered at the individual level. Dependent
variable: employment of any kind. Gender=0 if male and 1 if female. Controls are age and locality. Column
(1): baseline specification. Columns (2)-(3): control for different channels of transmission. Column (2): kind of
employment, business ownership is the control group; column (3): sector of employment, agriculture is the control
group. R-squared is the within one.
Namibia: Panel and Probit Models on COVID-19 Household and Jobs Tracker
We estimate both panel and probit models for Namibia using its COVID-19 Household and Jobs
Tracker survey data. The survey collects information on individuals’ employment status before and
during the pandemic. We extracted their employment status from before the pandemic and ran
a panel regression as specified by equation 3.1. We further used binary response questions to run
probit models and estimate the differential likelihood of being affected by the pandemic for women
relative to men. Table 3.16 shows the results from the panel regression, with the dependent variable
137

Table 3.17: Namibia: Effect of Gender Inequality on Work Status
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Probit models. Dependent variables: columns (1)-(3): whether the person is employed or not; columns (4):conditional that the respondent stopped working, whether they did so due to COVID-19; columns (5): conditional that the
respondent changed jobs, whether they did so due to COVID-19. Gender=0 if male and 1 if female. Urban/ rural=0 if the
respondent lives in a rural area.
being a binary employment status variable that takes the value of 1 if the individual is employed and
0 otherwise. Table 3.17 presents the results from the probit models with three dependent variables,
whether the respondent works or not (columns 1-3), whether the respondent stopped working due
to COVID (column 4), and whether the respondent changed jobs due to COVID (column 5). The
sample size varies between 16,882 and 413 respondents for the different questions.
The results from Table 3.16 suggest that women were more likely to have lost their employment due
to the pandemic, and Table 3.17 shows that women were less likely to work during the pandemic.
Specifically, women were 6.7 percentage points more likely to have lost their employment and 6.9
percentage points less likely to work. Women were not significantly more likely than men to have
stopped working or changed jobs due to the pandemic. However, conditional on changing jobs
or stopping work, women are not significantly more likely to have done so directly due to the
pandemic. Other reasons people have stopped working include furlough, being laid off, adult care
work, and lack of transportation, which may be indirect consequences of the pandemic and which,
138

combined, could have disproportionate negative effects on women.
South Africa: Consistency Analysis- Probit Models on NIDS
We run probit models for South Africa to establish comparability between all countries in our
analysis. The two binary dependent variables are employment of all forms and business employment
status of the individual during COVID. Pre-COVID data is not included in the probit analysis for
consistency. The analysis is done using both the NIDS-CRAM and the LFS datasets, but we only
report the results from the NIDS dataset in Table 3.18. The results are consistent with the previous
findings from the data on Eswatini, Lesotho, and Namibia. We found that women were less likely
to be employed, but more likely to run businesses. Specifically, women were 1.44 times less likely
to be employed (Table 3.18 column (2)) and 1.13 times more likely to run businesses (Table 3.18
column (5)).9
Table 3.18: South Africa: Effect of Gender Inequality on Employment of All Forms and Business
Employment
9The LFS probit models give similar results as the NIDS ones for employment of all forms, but we found that
women were less likely to run businesses during COVID, which is the opposite of what was found using the NIDS
dataset and in Lesotho. These results are available upon request from the authors.
139

Chapter Three: Policy Recommendations
Policy Recommendations
Reducing women’s occupational segregation and unpaid work burden would strengthen women’s
employment and income resilience to shocks. This paper provides strong evidence that occupational
gender segregation and unequal unpaid work burdens are the two main channels of the disproportionate impact of the COVID-19 pandemic on women’s employment and income, relative to men, in
the Southern African region. To strengthen women’s resilience to such shocks, country authorities
should thus implement policies that tackle these root causes of gender inequality. By doing so,
they will catalyze a stronger, more sustainable, and inclusive post-pandemic recovery and private
sector-driven development. Notably, we focus on policy recommendations that can be implemented
in the context of constrained fiscal space – where many countries find themselves in the aftermath
of the COVID-19 pandemic, the Russian war in Ukraine, and the global monetary and financial
tightening.
Encouraging girls’ education in science, technology, engineering, and mathematics (STEM) fields
will reduce occupational gender segregation by ensuring that women have more access to diversified
job opportunities, thus becoming more resilient to shocks. Even though the education gaps have
largely been closed in the Southern African countries, there is still scope to increase and diversify
girls’ educational attainment. In particular, there are still disparities in fields of study, with girls
more likely to opt in for non-STEM fields, characterized by lower pay, job security, and digitalization. As a result, women’s employment is more concentrated in the tertiary sector activities
that are more vulnerable to shocks like the COVID-19 pandemic, notably hospitality and tourism.
Therefore, encouraging women to pursue careers in STEM fields will support women’s representation in all sectors of the economy, thus attenuating the disproportionate negative effects of shocks
on women due to occupational gender segregation.
Fostering the provision of care services, such as childcare and adult care, and parental leave would
give women more flexibility to pursue full-time jobs. Enhancing the availability and affordability of
care services and flexible work arrangements would mitigate the effect of the pandemic on women’s
140

employment status and income, as many women had to drop out of the labor force to stay at home
and care for their children following school closures, or more at-risk elderly relatives. Research has
shown that there is significant heterogeneity in how the pandemic affected countries’ gender gaps,
with some countries being more resilient to the shock than others. One of the main reasons some
countries’ gender gaps were not worsened by the pandemic is the availability of affordable childcare
(Bluedorn et al. (2021)). Thus, providing affordable and accessible care services would strengthen
countries’ resilience to such shocks.
In light of the evidence on the importance of business employment, country authorities should complement their efforts of formalizing the informal sector with ensuring there is a robust social safety
net. The informal sector is a large component of these Southern African economies. Although it is
vulnerable in terms of pay, job security, and working conditions, it has served as a cushion against
the pandemic shock for informal-sector employees, the majority of whom are women. Recent IMF
research (IMF (2021a)) suggests that countries aiming to formalize their informal sectors should set
up permanent mechanisms to increase the coverage and adequacy of social protection systems and
provide adequate incentives for formalization. These may include cash transfers for unemployed
and informal-sector workers through digital technologies, e.g., sustainable government-to-person
mobile transfer programs. In addition to supportive social protection systems, tax policy and administration measures can decrease the incentives for informality and increase the incentives to join
the taxpayer registry, e.g., adequate minimal threshold for VAT, simpler value-added and corporate
tax systems, low payroll taxes, and simplification of tax payment procedures. Furthermore, steady
and gradual structural policy reforms, including simplifying labor market regulations, competition
policy, elimination of excessive regulations and bureaucratic requirements, and digitalization would
further support the formalization process.
One specific way to enhance business employment in these countries is by strengthening MSMEs’
financing. Access to more financing would allow small business owners, the majority of whom are
women, to scale up their businesses, so they become more resilient to economic shocks. Support
for MSMEs could include setting up incubators and facilitating business networking, as well as
increasing financial literacy and inclusion by providing access to formal financial services, preparation of financial accounts, etc. Digitalization and fintech could also play a catalytic role in scaling
141

up MSMEs. For instance, the expansion of the use of mobile money in some countries has been
an accelerator for many businesses, as it facilitates transactions with customers. Systems such as
these should be further developed and MSMEs should be facilitated in accessing them in order to
ensure sustainable private sector-led growth.
Enabling Environment for Gender Equality
Policies addressing the macro-critical gender inequalities above require a conducive enabling environment, with legal systems that ensure women’s equal economic participation and social freedom.
The restrictive legal system in some Southern African countries remains an impediment to women’s
agency, thus limiting their economic participation and contributions. For instance, in Eswatini and
Lesotho, the misalignment between and parallel application of civil and customary legal systems
allows for persistent discrimination against women. Even though constitutions in these countries
promote gender equality de jure, customary law remains exempted from these provisions. As a
result, women are still faced with discrimination in accessing financial services, such as loans, and
in property ownership and inheritance rights, which may obstruct them from owning collateral.
Our paper provides strong evidence for the importance of businesses in providing a social safety
net for women in these countries. Access to financial services and the ability to acquire assets are
crucial in giving women more agency to invest and grow their businesses, thereby enhancing private
sector-led growth. Enshrining gender equality in legal frameworks may also contribute to combating GBV and child marriage and may over time make discriminatory social norms and unequal
opportunities obsolete.
Gender responsive budgeting (GRB) could support the implementation of these recommendations
and help foster a conducive enabling environment for gender equality. GRB allows governments to
take into account gender inequality concerns when enacting policies and implementing programs
in national, state, and local budgets. Given that women’s participation is lagging in many aspects
of the economy, including the labor market, as well as in health, education, and political representation, gender equality goals should be incorporated into laws, policies, and budgets that aim to
promote the wellbeing of the population.
142

Chapter Three: Conclusions
This paper explores whether the COVID-19 pandemic disproportionately affected women in Southern Africa and finds significant evidence for a “she-cession” in South Africa, Eswatini, Lesotho, and
Namibia. Specifically, using the baseline linear probability panel regression model, we find that
women’s employment, hours worked, and incomes were disproportionately more negatively affected
during the pandemic than those of men in South Africa. Likewise, in Namibia, women were more
likely to have lost employment due to the pandemic than men. Furthermore, our analysis of the
dynamic effects of the pandemic suggests that women’s employment, hours worked, and income
also took longer to recover after the initial shock of COVID-19 in South Africa. Using alternative
standard probit models, due to more limited data availability, we complement these findings with
results for Eswatini, Lesotho, and Namibia. In Eswatini, women were less likely to be employed
and more likely to have lost their jobs due to the pandemic than men. Furthermore, in Lesotho,
women were less likely to be formally employed than men. In Namibia, women were less likely than
men to work during the pandemic. However, the findings also suggest that men in Eswatini were
relatively more likely to have lost income due to COVID-19, while men in Lesotho were more likely
to have stopped working due to the pandemic.
Our findings provide evidence supporting the role of women’s heavier unpaid work burden and
occupational gender segregation as transmission channels of the COVID-19 pandemic shock on
gender inequality. Having young children and children in school and spending more time fetching
water and wood contributed to the disproportionately negative impact of the pandemic on women’s
employment, hours worked, and income in South Africa. Furthermore, the results from the analysis
in Eswatini further suggest that women’s already disproportionate unpaid domestic, childcare,
and adult care work burdens were exacerbated by the pandemic. Moreover, women employed
in low- and medium-skilled occupations and certain sectors, such as services, were particularly
impacted by the pandemic in terms of employment, hours worked, and income, while more education
counterbalanced some of this negative effect.
We also find evidence that business employment, especially in the informal sector, acted as a
143

countercyclical shock absorber for women in South Africa and Lesotho. Unlike other forms of
employment, women’s relative to men’s business employment was not disproportionately affected
by the pandemic. Specifically, women running informal (but not formal) businesses were less
negatively affected by the COVID-19 shock, relative to their male counterparts. Furthermore,
women in Lesotho were more likely to run a business during the pandemic, as compared to men.
This evidence suggests that women in these countries may have used informal business employment
to cushion the negative impact of the pandemic.
The observed trends may be at work in the broader Southern and even sub-Saharan African region.
Due to limited data availability, we are unable to apply our baseline model to all countries in the
region. Our analysis on data from Eswatini, Lesotho, and Namibia, albeit not as granular, confirms
some of the findings about the disproportionate impact of the COVID-19 pandemic on women in
the region. For cross-country comparability reasons, this methodology was also applied to the data
from South Africa, broadly confirming the results from the other countries in this study. More
detailed, gender-disaggregated, and uniform across countries data would greatly facilitate future
research aiming to apply this empirical analysis to the broader Southern and even sub-Saharan
African region.
144

References
ACET (2013). Agriculture Dialogue Case Study Choice of Optimal Instruments for Boosting Food
Sufficiency: The case of agricultural input subsidies in Senegal. African Centre for Economic
Transformation.
Adhikari, S. (2019). Structural Transformation of India: A Quantitative Analysis. The Indian
Economic Journal.
Alesina, A., Giuliano, P., and Nunn, N. (2013). On the Origins of Gender Roles: Women and the
Plough. The Quarterly Journal of Economics 128, page 469–530.
Alon, T., Coskun, S., Doepke, M., Koll, D., and Tertilt, M. (2021). FROM MANCESSION TO
SHECESSION: WOMEN’S EMPLOYMENT IN REGULAR AND PANDEMIC RECESSIONS.
NBER WORKING PAPER.
Alon, T., Doepke, M., Olmstead-Rumsey, J., and Tertilt, M. (2020). THE IMPACT OF COVID-19
ON GENDER EQUALITY. NBER WORKING PAPER.
Amadou, A. and Aronda, T. (2020). Structural transformation in sub-Saharan Africa: A comparative analysis of sub-regions Performance. African Journal of Economic and Management Studies,
11(2):233–252.
Annabi, N., Ciss´e, F., Cockburn, J., and Decaluw´e, B. (2005). Trade Liberalisation, Growth and
Poverty in Senegal: A Dynamic Microsimulation CGE Model Analysis. CEPII, Working Paper
No 2005-07.
Aoyagi, C. (2021). Effects of COVID-19 on Regional and Gender Equality in Sub-Saharan Africa:
Evidence from Nigeria and Ethiopia. IMF Working Paper.
Atalay, E. (2017). How important are sectoral shocks? American Economic Journal: Macroeconomics.
Bah, E.-h. and Fang, L. (2015). Impact of the business environment on output and productivity in
Africa. Journal of Development Economics.
145

Bai, Y., Jin, K., and Lu, D. (2020). MISALLOCATION UNDER TRADE LIBERALIZATION.
NBER Working Paper 26188.
Baqaee, D. R. and Farhi, E. (2020). PRODUCTIVITY AND MISALLOCATION IN GENERAL
EQUILIBRIUM. The Quarterly Journal of Economics.
Bartelsman, E. J., Haltiwanger, J. C., and Scarpetta, S. (2013). CROSS-COUNTRY DIFFERENCES IN PRODUCTIVITY: THE ROLE OF ALLOCATION AND SELECTION. The American Economic Review.
Bau, N. (2020). MISALLOCATION AND CAPITAL MARKET INTEGRATION: EVIDENCE
FROM INDIA. NBER Working Paper 27955.
Bento, P. and Restuccia, D. (2016). MISALLOCATION, ESTABLISHMENT SIZE, AND PRODUCTIVITY. NBER WORKING PAPER.
Bertay, A. C., Dordevic, L., and Sever, C. (2020). Gender Inequality and Economic Growth:
Evidence from Industry-Level Data. IMF Working Paper 20/119. Washington: International
Monetary Fund.
Betts, C., Giri, R., and Verma, R. (2017). Trade, Reform, and Structural Transformation in South
Korea. IMF Economic Review, 65(4):745–791.
Bloom, N. and et al. (2010). Why Do Firms in Developing Countries Have Low Productivity?
American Economic Review.
Bluedorn, J., Caselli, F., Hansen, N.-J., Shibata, I., and Tavares, M. M. (2021). Gender and
Employment in the COVID-19 Recession: Evidence on “She-cessions”. IMF Working Paper.
Boehm, J. and Oberfield, E. (2020). MISALLOCATION IN THE MARKET FOR INPUTS: ENFORCEMENT AND THE ORGANIZATION OF PRODUCTION. The Quarterly Journal of
Economics.
Braunstein, E., van Staveren, I., and Tavani, D. (2011). Embedding Care and Unpaid Work in
Macroeconomic Modeling: A Structuralist Approach. Feminist Economics 17(4), pages 5–31.
Brixiov´a Schwidrowski, Z., Imai, S., Kangoye, T., and Yameogo, N. D. (2021). Assessing gender
gaps in employment and earnings in Africa: the case of Eswatini. GLO Discussion Paper.
Buera, F. J., Kaboski, J. P., and Shin, Y. (2011). Finance and Development: A Tale of Two Sectors.
The American Economic Review.
146

Buera, F. J., Kaboski, J. P., and Shin, Y. (2015). Entrepreneurship and Financial Frictions: A
Macrodevelopment Perspective. The Annual Review of Economics.
Calligaris, S. (2015). Misallocation and Total Factor Productivity in Italy: Evidence from FirmLevel Data. Labour.
Casale, D. and Posel, D. (2021). Gender inequality and the COVID-19 crisis: Evidence from a large
national survey during South Africa’s lockdown. Research in Social Stratification and Mobility.
Chantreuil, F. and Lebon, I. (2015). Gender contribution to income inequality. Economics Letters.
Christopherson, K. M., Yiadom, A., Johnson, J., Fernando, F., Yazid, H., and Thiemann, C. (2022).
Tackling Legal Impediments to Women’s Economic Empowerment. IMF Working Paper 22/37.
Washington: International Monetary Fund.
Cirera, X., Fattal-Jaef, R., and Maemir, H. (2020). Taxing the Good? Distortions, Misallocation,
and Productivity in Sub-Saharan Africa. The World Bank Economic Review.
Ciss´e, F., Choi, J. E., and Maurel, M. (2016). Industrial Policy in Senegal: Then and Now. In
Manufacturing Transformation: Comparative Studies of Industrial Development in Africa and
Emerging Asia.
Comin, D. A., Lashkari, D., and Mestieri, M. (2021). Structural Change with Long-Run Income
and Price Effects. Econometrica, 89(1):311–374.
Cuberes, D. and Teignier, M. (2016). Aggregate Effects of Gender Gaps in the Labor Market: A
Quantitative Estimate. Journal of Human Capital.
David, J. M. and Venkateswaran, V. (2019). THE SOURCES OF CAPITAL MISALLOCATION.
American Economic Review.
de Vries, G., Timmer, M., and de Vries, K. (2015). Structural Transformation in Africa: Static
Gains, Dynamic Losses. The Journal of Development Studies, 51(6):674–688.
Deere, C. D., D., O. A., Swaminathan, H., and Doss, C. (2013). Property Rights and the Gender
Distribution of Wealth in Ecuador, Ghana and India. Journal of Economic Inequality 11, page
249–65.
Dekle, R. and Vandenbroucke, G. (2012). A quantitative analysis of China’s structural transformation. Journal of Economic Dynamics and Control, 36:119–135.
147

Demirguc-Kunt, A., Klapper, L., and Singer, D. (2013). Financial Inclusion and Legal Discrimination against Women: Evidence from Developing Countries. World Bank Policy Research Working
Paper 6416. Washington, DC: World Bank.
Demirguc-Kunt, A., Klapper, L., and Singer, D. (2014). Are There More Female Managers in
the Retail Sector? Evidence from Survey Data in Developing Countries. Journal of Applied
Economics 17, page 213–28.
Dias, D. A., Marques, C. R., and Richmond, C. (2016). Misallocation and Productivity in the Lead
Up to the Eurozone Crisis. Journal of Macroeconomics.
Dieme, M. (2018). Transformation structurelle et comp´etitivit´e: Transformation comp´etitive de
l’´economie s´en´egalaise.
Dlamini, J. (2020). Gender-Based Violence, Twin Pandemic to COVID-19. Critical Sociology.
Dollar, D. and Gatti, R. (1999). Gender Inequality, Income, and Growth: Are Good Times Good
for Women? Washington, DC: The World Bank Development Research Group, vol. 1.
Duval, R., Hong, G. H., and Timmer, Y. (2019). Financial Frictions and the Great Productivity
Slowdown. The Society for Financial Studies.
Elson, D. (1998). Talking to the Boys: Gender and Economic Growth Models. Feminist Visions
of Development: Gender Analysis and Policy, London: Routledge, pages 155–70.
Fabrizio, S., Gomes, D. B. P., and Tavares, M. M. (2021). COVID-19 She-Cession: The Employment
Penalty of Taking Care of Young Children. IMF Working Paper.
Feenstra, R. C., Inklar, R., and Timmer, M. (2015). The Next Generation of the Penn World Table.
American Economic Review, 105(10):3150–3182.
Forbes, K. J. (2000). A Reassessment of the Relationship between Inequality and Growth. American
Economic Review 90, pages 869–87.
Galor, O. and Weil, D. N. (1996). The Gender Gap, Fertility, and Growth. American Economic
Review 85, page 374–87.
Goldsbrough, D., Coorey, S., Dicks-Mireaux, L., Horvath, B., Kochhar, K., Mecagni, M., Offerdal,
E., and Zhou, J. (1996). Reinvigorating Growth in Developing Countries: Lessons from Adjustment Policies in Eight Economies. International Monetary Fund.
148

Gonzales, C., Jain-Chandra, S., Kochhar, K., and Newiak, M. (2015). Fair Play: More Equal Laws
Noost Female Labor Force Participation. IMF Discussion Note.
Gopinath, G., Kalemli-Ozcan, S., KARABARBOUNIS, L., and VILLEGAS-SANCHEZ, C. (2017).
CAPITAL ALLOCATION AND PRODUCTIVITY IN SOUTH EUROPE. The Quarterly Journal of Economics.
Government of Lesotho (2020). Covid-19 Socio-Economic Impact on Households Survey: Report.
Bureau of Statistics.
Goyette, J. and Gallipoli, G. (2015). Distortions, efficiency and the size distribution of firms.
Journal of Macroeconomics.
Guner, N., Ventura, G., and Xu, Y. (2008). MACROECONOMIC IMPLICATIONS OF SIZEDEPENDENT POLICIES. Review of Economic Dynamics.
Hakura, D., Hussain, M., Newiak, M., Thakoor, V., and F., Y. (2016). Inequality, Gender Gaps and
Economic Growth: Comparative Evidence for Sub-Saharan Africa. IMF Working Paper 16/111.
Washington: International Monetary Fund.
Haraguchi, N., Cheng, C. F. C., and Smeets, E. (2016). The importance of manufacturing in
economic development: Has this changed? Inclusive and Sustainable Industrial Development
Working Paper Series WP 1.
Herrendorf, B., Rogerson, R., and Valentinyi, (2013). Two Perspectives on Preferences and Structural Transformation. American Economic Review, 103(7):2752–2789.
Herrendorf, B., Rogerson, R., and Valentinyi, (2014). Growth and Structural Transformation. In
Handbook of Economic Growth, volume 2B. Elsevier.
Hill, M. A. and Elizabeth, K. (1995). Women’s Education and Economic Well-being. Feminist
Economics 1, pages 21–46.
Hsieh, C.-T. and Klenow, P. J. (2009). MISALLOCATION AND MANUFACTURING TFP IN
CHINA AND INDIA. Quaterly Journal of Economics.
Hyland, M., Djankov, S., and Goldberg, P. K. (2020). Gendered Laws and Women in the Workforce.
American Economic Review: Insights, page 275–90.
IMF (2016). Time for a Policy Reset. Chapter 1, Regional Economic Outlook: Sub-Saharan Africa,
April 2016. Washington, DC: International Monetary Fund.
149

IMF (2018). How to Operationalize Gender Issues in Country Work. Strategy, Policy, and Review
Department. Washington, DC: International Monetary Fund.
IMF (2021a). The Global Informal Workforce: Priorities for Inclusive Growth. Washington, DC:
International Monetary Fund.
IMF (2021b). Labor market challenges during the pandemic, the role of informality, and the road
ahead. Chapter 2, Regional Economic Outlook: Middle East and Central Asia: Trade-Offs Today
for Transformation Tomorrow. Washington, DC: International Monetary Fund.
Jeong, H. (2020). Productivity Growth and Efficiency Dynamics of Korean Structural Transformation. Policy Research Working Paper 9285.
Kaba, K., Lin, J. Y., and Renard, M.-F. (2022). Structural change and trade openness in subSaharan African countries. The World Economy, pages 1–34.
Kalemli-Ozcan, S. and Sørensen, B. E. (2012). Misallocation, Property Rights, and Access to
Finance Evidence from within and across Africa. NBER Working Paper 18030.
Kazandjian, R., Kolovich, L., Kochhar, K., and Newiak, M. (2019). Gender Equality and Economic
Diversification. Social Sciences.
Kehoe, T., Ruhl, K., and Steinberg, J. (2018). Global Imbalances and Structural Change in the
United States. Journal of Political Economy, 126(2):761–796.
Klasen, S. and Lamanna, F. (2009). The Impact of Gender Inequality in Education and Employment
on Economic Growth: New Evidence for a Panel of Countries. Feminist Economics 15, pages
91–132.
Knowles, S., Lorgelly, P. K., and Owen, P. D. (1999). Are Educational Gender Gaps a Brake on
Economic Development? Some Cross-country Empirical Evidence. Oxford Economic Papers 54,
pages 118–49.
Kongsamut, P., Rebelo, S., and Xie, D. (2001). Beyond Balanced Growth. Review of Economic
Studies, 68:869–882.
Kruse, H., Mensah, E., Sen, K., and de Vries, G. (2022). A Manufacturing (Re)Naissance? Industrialization in the Developing World. IMF Economic Review.
Kuznets, S. (1973). Modern Economic Growth: Findings and Reflections. The American Economic
Review, 63(3):247–258.
150

Lambert, F., Pescatori, A., and Toscani, F. G. (2020). Labor Market Informality and the Business
Cycle. IMF Working Paper 20/256. Washington: International Monetary Fund.
Le´on-Ledesma, M. (2016). POtENtIAL GrOwth, MIsALLOcAtION, AND INstItutIONAL OBstAcLEs: FIrM-LEvEL EvIDENcE. adb economics working paper series.
Lian, C. and Ma, Y. (2021). ANATOMY OF CORPORATE BORROWING CONSTRAINTS.
Quaterly Journal of Economics.
Loko, B. and Diouf, M. A. (2009). Revisiting the Determinants of Productivity Growth: What’s
New? IMF Working Paper.
McMillan, M., Rodrik, D., and Sepulveda, C. (2017). Structural Change, Fundamentals and
Growth: A Framework and Case Studies. NBER Working Paper 23378.
McMillan, M., Rodrik, D., and Verduzco-Gallo, (2014). Globalization, Structural Change, and
Productivity Growth, with an Update on Africa. World Development, 63:11–32.
Midrigan, V. and Xu, D. Y. (2014). FINANCE AND MISALLOCATION: EVIDENCE FROM
PLANT-LEVEL DATA. American Economic Review.
Misch, F. and Saborowski, C. (2018). Resource Misallocation and Productivity: Evidence from
Mexico. IMF Working Paper.
Mistry, P. S. (1992). Background to the African Debt Crisis. In African Debt Revisited: Procrastination or Progress? FONDAD.
Mitra, P., Endengle, E. M. P., Pant, M., and F., A. L. (2020). Does Child Marriage Matter for
Growth? IMF Working Paper 20/27. Washington: International Monetary Fund.
Moll, B. (2014). Productivity Losses from Financial Frictions: Can Self-Financing Undo Capital
Misallocation? The American Economic Review.
Nelson, R. R. and Phelps, E. S. (1966). Investment in humans, technological diffusion, and economic
growth. American Economic Review, 56(1/2):6975.
Ngai, R. and Pissarides, C. (2007). Structural Change in a Multisector Model of Growth. The
American Economic Review, 97(1):429–443.
Oehmke, J. F., Naseem, A., Anderson, J., Pray, C., Moss, C., and Post, L. A. (2016). Contemporary
African Structural Transformation.
151

Ouedraogo, R. and Stenzel, D. (2021). The Heavy Economic Toll of Violence against Women:
Evidence from Sub-Saharan Africa. IMF Working Paper 21/277. Washington: International
Monetary Fund.
Parry, B. R. and Gordon, E. (2020). The shadow pandemic: Inequitable gendered impacts of
COVID-19 in South Africa. Gender, Work and Organisation.
Quisumbing, A. R. and Pandolfelli, L. (2010). Promising Approaches to Address the Needs of
Poor Female Farmers: Resources, Constraints, and Interventions. World Development 38, page
581–92.
Ranasinghe (2020). Misallocation across Establishment Gender. Working Paper.
Razavi, S. (2003). Introduction: Agrarian Change, Gender and Land Rights. Journal of Agrarian
Change, Vol. 3 Nos. 1 and 2.
Restuccia, D. and Rogerson, R. (2008). POLICY DISTORTIONS AND AGGREGATE PRODUCTIVITY WITH HETEROGENEOUS PLANTS. Review of Economic Dynamics.
Rodrik, D. (2016a). An African Growth Miracle? Journal of African Economies, pages 1–18.
Rodrik, D. (2016b). Premature Deindustrialization. Journal of Economic Growth, 21:1–33.
Sahay, R. and Cihak, M. (2018). Women in Finance: A Case for Closing Gaps. International
Monetary Fund Staff Discussion Note 18/005. Washington: International Monetary Fund.
Sever, C. (2022). Legal Gender Equality as a Catalyst for Convergence. IMF Working Paper
22/155. Washington: International Monetary Fund.
Sposi, M., Yi, K.-M., and Zhang, J. (2021). Deindustrialization and Industry Polarization. Working
Paper 29483.
Subramanian, A. and Roy, D. (2001). Who Can Explain The Mauritian Miracle: Meade, Romer,
Sachs, or Rodrik? IMF Working Paper 01/116.
Timmer, M., de Vries, G., and de Vries, K. (2015). Patterns of Structural Change in Developing
Countries. Routledge Handbook of Industry and Development, pages 65–83.
United Nations (2020). Policy Brief: The Impact of COVID-19 on Women. United Nations.
152

Verma, R. (2012). Can total productivity explain value added growth in services. Journal of
Development Economics, 99:163–177.
Welch, F. (1970). Education in Production. Journal of Political Economy.
Woldekidan, B. (1992). Mauritius: an export-led economic success.
Young, A. (1995). The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian
Growth Experience. The Quarterly Journal of Economics.
Zetlin-Jones, A. and Shourideh, A. (2017). External Financing and the Role of Financial Frictions
over the Business Cycle: Measurement and Theory. Journal of Monetary Economics.
153

Appendices
154

Chapter One Appendix
Aggregate growth accounting without tourism sector
Table 19: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
1981-1990 6.48 0.81 3.99 1.68
1991-2000 6.35 1.64 4.14 0.57
2001-2010 4.03 0.74 2.90 0.39
2011-2019 3.45 0.39 2.47 0.59
1980-2019 5.12 0.91 3.40 0.81
Table 20: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981-1990 4.66 1.80 0.94 1.92
1991-2000 4.79 3.65 1.44 -0.31
2001-2010 3.10 1.65 1.52 -0.08
2011-2019 2.94 0.87 1.28 0.80
1980-2019 3.90 2.02 1.30 0.58
Figure 7: Aggregate TFP-whole economy vs economy without tourism-Mauritius
155

Aggregate growth accounting without finance sector
Table 21: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
1981-1990 6.35 0.65 3.92 1.78
1991-2000 5.99 1.18 4.17 0.64
2001-2010 3.96 0.43 3.06 0.47
2011-2019 3.39 0.36 2.41 0.61
1980-2019 4.96 0.66 3.42 0.88
Table 22: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981-1990 4.53 1.40 1.10 2.04
1991-2000 4.79 2.55 2.05 -0.18
2001-2010 3.02 0.08 2.02 -0.08
2011-2019 2.88 0.79 1.27 0.82
1980-2019 3.74 1.43 1.62 0.60
Figure 8: Aggregate TFP-whole economy vs economy without finance-Mauritius
Aggregate growth accounting without tourism and finance sectors
156

Table 23: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
1981-1990 6.29 0.66 3.84 1.79
1991-2000 5.89 1.31 4.00 0.58
2001-2010 3.72 0.57 2.81 0.35
2011-2019 3.34 0.35 2.38 0.60
1980-2019 4.85 0.73 3.38 0.84
Table 24: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
1981-1990 4.47 1.40 1.07 2.00
1991-2000 4.32 2.78 1.88 -0.33
2001-2010 2.79 1.21 1.77 -0.20
2011-2019 2.83 0.75 1.30 0.78
1980-2019 3.62 1.56 1.51 0.56
Figure 9: Aggregate TFP-whole economy vs economy without tourism and finance-Mauritius
157

Sectoral growth accounting with adjusted factor shares
.
Table 25: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
Agriculture
1981-1990 5.4 2.10 3.13 0.18
1991-2000 -0.37 -1.56 2.66 -1.47
2001-2010 -0.81 -0.86 2.16 -2.12
2011-2019 0.44 -0.99 1.50 -0.08
1981-2019 1.18 -0.31 2.38 -0.89
Manufacturing
1981-1990 7.53 0.66 3.59 3.28
1991-2000 5.03 1.98 2.85 0.20
2001-2010 2.80 1.85 1.83 -0.89
2011-2019 0.57 -0.88 1.59 -0.14
1981-2019 4.07 0.95 2.49 0.63
Services
1981-1990 6.23 1.26 3.37 1.60
1991-2000 8.29 2.32 4.30 1.67
2001-2010 5.20 0.26 3.28 1.65
2011-2019 4.51 0.95 2.55 1.01
1981-2019 6.1 1.21 3.39 1.50
Table 26: Decomposition of VA-Senegal
Period Y A Kα L
1−α
Agriculture
1981-1990 2.40 -2.22 2.38 2.25
1991-2000 2.64 0.05 2.08 0.51
2001-2010 2.91 1.33 1.00 0.58
2011-2019 4.28 7.21 0.05 -2.99
1981-2019 3.03 1.45 1.41 0.17
Manufacturing
1981-1990 3.37 -2.74 4.36 1.74
1991-2000 3.18 -1.26 2.82 1.62
2001-2010 3.96 -1.36 3.18 2.14
2011-2019 5.62 1.79 1.89 1.94
1981-2019 3.99 -0.96 3.09 1.86
Services
1981-1990 2.43 -4.67 3.98 3.12
1991-2000 3.10 -0.83 2.06 1.87
2001-2010 3.65 -1.16 2.67 2.14
2011-2019 4.71 -1.59 1.55 4.75
1981-2019 3.44 -2.07 2.59 2.92
158

Table 27: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981-1990 3.58 3.42 1.69 -1.53
1991-2000 -1.93 -2.54 4.57 -3.96
2001-2010 -1.75 -1.40 4.04 -4.39
2011-2019 -0.07 -1.61 2.17 -0.63
1981-2019 -0.04 -0.5 3.14 -2.68
Manufacturing
1981-1990 5.71 1.18 0.46 4.07
1991-2000 3.47 3.56 1.12 -1.21
2001-2010 1.86 3.33 1.07 -2.54
2011-2019 0.07 -1.57 2.39 -0.75
1981-2019 2.85 1.71 1.23 -0.09
Services
1981-1990 4.41 2.72 0.07 1.62
1991-2000 6.72 4.99 -0.30 2.03
2001-2010 4.26 0.57 1.07 2.63
2011-2019 4.00 2.05 0.28 1.68
1981-2019 4.87 2.60 0.28 2.00
Table 28: Decomposition of VA per working-age person-Senegal
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981-1990 -0.23 -4.08 2.35 1.51
1991-2000 -0.32 0.09 1.61 -2.02
2001-2010 0.20 2.44 -0.61 -1.64
2011-2019 1.38 13.25 -3.48 -8.38
1981-2019 0.23 2.66 0.06 -2.49
Manufacturing
1981-1990 0.74 -8.79 6.56 2.96
1991-2000 0.22 -4.04 2.03 2.24
2001-2010 1.24 -4.37 1.45 4.16
2011-2019 2.73 5.74 -6.37 3.35
1981-2019 1.19 -3.09 1.11 3.17
Services
1981-1990 -0.2 -9.59 5.61 3.78
1991-2000 0.14 -1.70 0.96 0.88
2001-2010 0.94 -2.38 1.63 1.68
2011-2019 1.81 -3.26 -1.78 6.86
1981-2019 0.64 -4.26 1.69 3.21
159

Sectoral growth accounting without tourism sector
Table 29: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
Agriculture
1981-1990 5.40 1.18 4.08 0.14
1991-2000 -0.37 -2.65 3.47 -1.20
2001-2010 -0.81 -1.92 2.82 -1.71
2011-2019 0.44 -1.46 1.95 -0.06
1981-2019 1.18 -1.21 3.11 -0.73
Manufacturing
1981-1990 7.53 0.45 4.34 2.75
1991-2000 5.03 1.42 3.45 0.17
2001-2010 2.80 1.32 2.22 -0.75
2011-2019 0.57 -1.23 1.92 -0.12
1981-2019 4.07 0.53 3.01 0.53
Services
1981-1990 6.11 1.19 3.48 1.44
1991-2000 8.33 2.47 4.39 1.46
2001-2010 4.99 0.39 3.18 1.43
2011-2019 4.56 0.89 2.72 0.95
1981-2019 6.04 1.24 3.46 1.33
Table 30: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981-1990 3.58 2.38 2.73 -1.53
1991-2000 -1.93 -5.35 7.37 -3.96
2001-2010 -1.75 -3.88 6.52 -4.39
2011-2019 -0.07 -2.94 3.50 -0.63
1981-2019 -0.04 -2.43 5.07 -2.68
Manufacturing
1981-1990 5.71 0.97 0.67 4.07
1991-2000 3.47 3.06 1.62 -1.21
2001-2010 1.86 2.85 1.55 -2.54
2011-2019 0.07 -2.65 3.47 -0.75
1981-2019 2.85 1.15 1.79 -0.09
Services
1981-1990 4.30 2.75 0.04 1.51
1991-2000 6.76 5.71 -0.77 1.82
2001-2010 4.06 0.90 0.80 2.36
2011-2019 4.06 2.06 0.30 1.70
1981-2019 4.81 2.87 0.09 1.85
160

Sectoral growth accounting without finance sector
Table 31: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
Agriculture
1981-1990 5.40 1.18 4.08 0.14
1991-2000 -0.37 -2.65 3.47 -1.2
2001-2010 -0.81 -1.92 2.82 -1.71
2011-2019 0.44 -1.46 1.95 -0.06
1981-2019 1.18 -1.21 3.11 -0.73
Manufacturing
1981-1990 7.53 0.45 4.34 2.75
1991-2000 5.03 1.42 3.45 0.17
2001-2010 2.80 1.32 2.22 -0.75
2011-2019 0.57 -1.23 1.92 -0.12
1981-2019 4.07 0.53 3.01 0.53
Services
1981-1990 5.87 0.84 3.39 1.64
1991-2000 7.76 1.72 4.41 1.64
2001-2010 4.96 0.014 3.38 1.56
2011-2019 4.54 0.97 2.60 0.99
1981-2019 5.82 0.88 3.47 1.47
Table 32: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981-1990 3.58 2.38 2.73 -1.53
1991-2000 -1.93 -5.35 7.37 -3.96
2001-2010 -1.75 -3.88 6.52 -4.39
2011-2019 -0.07 -2.94 3.50 -0.63
1981-2019 -0.04 -2.43 5.07 -2.68
Manufacturing
1981-1990 5.71 0.97 0.67 4.07
1991-2000 3.47 3.06 1.62 -1.21
2001-2010 1.86 2.85 1.55 -2.54
2011-2019 0.07 -2.65 3.47 -0.75
1981-2019 2.85 1.15 1.79 -0.09
Services
1981-1990 4.05 1.85 0.39 1.81
1991-2000 6.20 3.80 0.34 2.06
2001-2010 4.02 0.03 1.47 2.52
2011-2019 4.06 2.15 0.22 1.68
1981-2019 4.60 1.95 0.62 2.03
161

Sectoral growth accounting without tourism and finance sectors
Table 33: Decomposition of VA-Mauritius
Period Y A Kα L
1−α
Agriculture
1981-1990 5.40 1.18 4.08 0.14
1991-2000 -0.37 -2.65 3.47 -1.2
2001-2010 -0.81 -1.92 2.82 -1.71
2011-2019 0.44 -1.46 1.95 -0.06
1981-2019 1.18 -1.21 3.11 -0.73
Manufacturing
1981-1990 7.53 0.45 4.34 2.75
1991-2000 5.03 1.42 3.45 0.17
2001-2010 2.80 1.32 2.22 -0.75
2011-2019 0.57 -1.23 1.92 -0.12
1981-2019 4.07 0.53 3.01 0.53
Services
1981-1990 5.70 0.85 3.22 1.64
1991-2000 7.75 2.03 4.15 1.58
2001-2010 4.69 0.20 3.02 1.47
2011-2019 4.64 1.05 2.57 1.03
1981-2019 5.72 1.03 3.25 1.44
Table 34: Decomposition of VA per working-age person-Mauritius
Period Y
N A
1
1−α K
Y
α
1−α L
N
Agriculture
1981-1990 3.58 2.38 2.73 -1.53
1991-2000 -1.93 -5.35 7.37 -3.96
2001-2010 -1.75 -3.88 6.52 -4.39
2011-2019 -0.07 -2.94 3.50 -0.63
1981-2019 -0.04 -2.43 5.07 -2.68
Manufacturing
1981-1990 5.71 0.97 0.67 4.07
1991-2000 3.47 3.06 1.62 -1.21
2001-2010 1.86 2.85 1.55 -2.54
2011-2019 0.07 -2.65 3.47 -0.75
1981-2019 2.85 1.15 1.79 -0.09
Services
1981-1990 3.89 1.82 0.36 1.71
1991-2000 6.19 4.36 -0.01 1.84
2001-2010 3.75 0.43 1.09 2.24
2011-2019 4.13 2.25 0.18 1.70
1981-2019 4.50 2.22 0.41 1.88
162

Other Decompositions
Decomposition of Real VA per Working-Age Person
I decompose the growth of real value added per working-age person into growth rates of real value
added per employed person and labor per working-age person. Data on active population are
obtained from the 10SD and the ETD, and the working-age population is obtained from the UN
population database. I use my own value added series. This decomposes the average year-over-year
growth in real per capita income into labor productivity growth, and labor inputs growth, as shown
in equation 3. This decomposition gives us some stylized facts on the countries’ growth experiences.
Yit
Nit
=
Yit
Lit
×
Lit
Nit
(3)
I use the Taylor series approximation of growth rates to derive:
g Y
N
= g Y
L
+ g L
N
(4)
where, for a variable X,
gX = log(Xt+1) − log(Xt) (5)
I conduct this decomposition for the two countries for the whole sample period, 1980-2019, and then
for each decade, by taking the average yearly growth rates of each component over the given period.
Tables 35, 36, and 37 show the results of my decomposition. Throughout the period 1980-2019,
Mauritius experienced a per capita income growth of about 4% on average, while Senegal’s per
capita income growth was less than 1%. The large gap between the two countries’ growth occurred
mostly in the first two decades of the period, from 1981 to 2000, where Senegal’s growth was close
to 0 and Mauritius was growing at almost 5%. Senegal started catching up a little bit in the 2000s,
when it experienced an almost 1% growth rate at the beginning of the period and a slightly lower
than 2% growth rate between 2011 and 2019. By that last decade, Mauritius and Senegal have
comparable growth rates at 3 and 2% respectively. These stylized facts are also depicted by Figures
10 and 11.
163

Figure 10: Real VA per working-age person
Table 35: Real VA per working-age person-Growth rate
Period Mauritius Senegal
1981-1990 4.71 0.00
1991-2000 4.86 0.07
2001-2010 3.29 0.87
2011-2019 2.98 1.97
1980-2019 3.98 0.70
Table 36 reveals that the main source of the divergence in the countries’ growth is their divergence
in the growth rates of their labor productivity. Labor productivity grew at an average rate of 3.3%
in Mauritius against .02% in Senegal. Throughout the four decades, Mauritius maintained a labor
productivity growth above 2%, while Senegal never reached a 1% growth. The decade between
1981 and 1990 was particularly bad for Senegal with a negative growth rate in labor productivity.
Figures 12 and 13 depict the consistently higher growth in labor productity in Mauritius.
The two countries, however, have very similar labor inputs growth. They both experienced negative
labor input growth in the 1990s, as shown in Table 37. Overall, labor productivity accounted for
82% of Mauritius’ per capita growth, while it accounted for only 2.9% of Senegal’s. Figure 14
164

Figure 11: Growth rate-real VA per working-age person
Figure 12: Real VA per employed person
165

Table 36: Growth rate-real VA per employed person
Period Mauritius Senegal
1981-1990 2.75 -2.17
1991-2000 5.01 0.86
2001-2010 3.10 0.62
2011-2019 2.14 0.85
1980-2019 3.28 0.02
Figure 13: Growth rate-real VA per employed person
shows the similar trends in the two countries’ labor per working-age person. Therefore, at the
aggregate level, although the two countries have comparable labor inputs growth, Mauritius’ rapid
labor productivity growth made the difference and that’s what created a divergence in their per
capita income growth.
Table 37: Labor per working-age person
Period Mauritius Senegal
1981-1990 1.96 2.17
1991-2000 -0.15 -0.79
2001-2010 0.19 0.25
2011-2019 0.84 1.12
1980-2019 0.70 0.68
166

Figure 14: Labor per working-age person
Figure 15: Decomposition real VA per WA person
167

TFP Growth Decomposition
Given the critical role that TFP growth plays in the differential growth experiences of Senegal and
Mauritius, I decompose it into three components as shown in equation 6, where Syst is the value
added share of sector s. TFP growth is driven by the sectoral growth in TFP, a capital growth
component, and a labor growth component. For each component, I take the average of the yearly
growth rates over each decade and the whole sample period. The first component tells us whether
TFP growth is driven by growth within each sector. Else, it is driven by reallocation of economic
activity to sectors with higher TFP growth. The results are reported in Tables 38 and 39. We see
that consistent with the results from the growth accounting exercises, TFP growth in Mauritius is
driven by sectoral TFP growth, but the effect of the contribution of sectoral TFP growth decreased
over time. We can see that the 0.83% growth in TFP in Mauritius is entirely driven by the sectoral
TFP growth. In Senegal, the negative TFP growth (-1.27%) is partly driven by the negative TFP
growth in each sector (-1.11%), which is about 87%. This means the other 13% is driven by labor
moving to sectors with lower TFP growth. This is consistent with the results from the growth
accounting exercises.
T F Pt+1 − T F Pt
T F Pt
=
X
s
T F Pst+1 − T F Pst
T F Pst
Syst+
X
s
(αs−α)
Kst+1
Kst
Syst+
X
s
(α−αs)
Lst+1
Lst
Syst (6)
Table 38: TFP growth decomposition with baseline factor shares of income
Within Residual 1 Residual 2 Total
Mauritius Senegal Mauritius Senegal Mauritius Senegal Mauritius Senegal
1980-1990 1.06 -3.66 -0.55 -3.68 0.50 3.60 1.01 -3.74
1990-2000 1.33 -1.10 -0.06 -2.92 0.02 2.82 1.29 -1.20
2000-2010 0.88 -0.78 0.49 -2.54 -0.51 2.40 0.86 -0.92
2010-2019 0.26 0.79 0.81 -2.18 -0.80 2.15 0.27 -0.82
1980-2019 0.82 -1.11 0.17 -2.82 -0.20 2.72 0.79 -1.21
168

Table 39: TFP growth decomposition with adjusted factor shares of income
Within Residual 1 Residual 2 Total
Mauritius Senegal Mauritius Senegal Mauritius Senegal Mauritius Senegal
1980-1990 1.34 -3.63 -2.19 -2.94 2.06 2.87 1.31 -3.70
1990-2000 1.70 -1.11 -1.16 -2.28 1.03 2.20 1.57 -1.19
2000-2010 1.17 -0.78 0.06 -2.03 -0.15 1.90 1.08 -0.91
2010-2019 0.44 0.60 0.78 -1.71 -0.88 1.71 0.34 -0.60
1980-2019 1.11 -1.15 -0.63 -2.23 0.54 2.16 1.02 -1.22
169

Sectoral TFP Growth Contribution
To further look into the extent to which each sector’s TFP growth contributes to aggregate TFP
growth, I compute the sectoral contribution to TFP growth, given by equation 7. This computes,
for each sector s, the percentage points growth it contributed towards the average annual growth
rate of aggregate TFP. Table 40 shows the results for each sector, each country, over each decade.
In Mauritius, each sector contributed positively to the aggregate TFP growth, except agriculture,
and the services sector contributed the most, about 80%. However in Senegal, only agriculture
contributed positively to the aggregate TFP growth, while the manufacturing and services sectors
contributed negatively, driving the negative growth rate of the aggregate TFP.
Cst+1 =
T F Pst+1 − T F Pst
T F Pst
× Syst (7)
Table 40: Sectoral contribution to TFP growth
Agriculture Manufacturing Services
Mauritius Senegal Mauritius Senegal Mauritius Senegal
1980-1990 0.24 -2.22 0.25 -0.59 0.58 -2.60
1990-2000 -0.19 0.05 0.32 -0.21 1.19 -0.79
2000-2010 0.04 1.33 0.47 -0.39 0.38 -0.63
2010-2019 -0.06 7.21 -0.27 0.39 0.60 -0.66
1980-2019 -0.04 1.45 0.21 -0.18 0.66 -1.13
Labor Productivity Growth Decomposition
In addition to TFP, I also look at labor productivity dynamics in Mauritius and Senegal. To that
end, I conduct a growth decomposition as in McMillan et al. (2014). The decomposition is given by
equation 8. The first component, the within effect, is the growth in labor productivity that is due
to growth in sectoral labor productivity. The second component, the static effect, is growth in labor
productivity that is due to labor moving from low productivity sectors to high productivity sectors.
And the last component, the dynamic effect, is the growth in labor productivity that is due to labor
moving from low productivity growth sectors to high productivity growth sectors. In Mauritius,
labor productivity is increasing within each sector while it is decling in Senegal. This makes sense
170

as sectoral employment in manufacturing and services are increasing much faster than their value
added. In both Mauritius and Senegal, labor is moving to more productive sectors (static effect),
but this effect is much higher in Senegal. However, in Senegal, the growth in labor productivity that
is due to labor moving to higher productivity sectors is undone by the decline in labor productivity
within each sector. In addition, labor is moving to sectors with lower productivity growth in both
countries. In sum, there are positive static effects but dynamic losses, both of which are higher in
Senegal.
LPt+1−LPt =
X
s
(LPst+1−LPst)×Syst+
X
s
(Syst+1−Syst)×LPst+
X
s
(LPst+1−LPst)(Syst+1−Syst)
(8)
Table 41: Decomposition of growth of labor productivity
Within Static Dynamic Total
Mauritius Senegal Mauritius Senegal Mauritius Senegal Mauritius Senegal
1980-1990 3.01 -2.98 0.16 0.94 -0.22 -0.05 2.95 -2.09
1990-2000 4.19 -0.56 0.56 1.47 0.04 -0.03 4.79 0.88
2000-2010 3.16 -0.79 0.68 1.51 -0.02 -0.04 3.82 0.68
2010-2019 1.91 -1.35 0.20 2.65 0.01 -0.40 2.12 0.90
1980-2019 3.04 -1.42 0.41 1.62 -0.05 -0.13 3.40 0.07
171

Chapter Two Appendix
Derivation of Cost Shares
From the prodution function:
Ysi = Asi[(1 − µs)
1
ϵm ((Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
)
ϵm−1
ϵm + µ
1
ϵm
s M
ϵm−1
ϵm
si ]
ϵm
ϵm−1 (9)
I derive:
∂Ysi
∂Ksi
=
Ysi
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs−1
(
Lsi
1 − αs
)
1−αs
(10)
∂Ysi
∂Lsi
=
Ysi
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs
(
Lsi
1 − αs
)
−αs
(11)
∂Ysi
∂Msi
=
Ysi
ψ
µ
1
ϵm
s M
−1
ϵm
si (12)
where
ψ = (1 − µs)
1
ϵm ((Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
)
ϵm−1
ϵm + µ
1
ϵm
s M
ϵm−1
ϵm
si (13)
V A = (Ksi
αs
)
αs
(
Lsi
1 − αs
)
1−αs
(14)
From the profit function, assuming a frictionless environment:
Ysi =
π
Psi
+
RsKsi
Psi
+
WsLsi
Psi
+
ZsMsi
Psi
(15)
I have:
∂Ysi
∂Ksi
=
Rs
Psi
(16)
∂Ysi
∂Lsi
=
Ws
Psi
(17)
∂Ysi
∂Msi
=
Zs
Psi
(18)
172

Equating equations 15, 16, 17 to equations 20, 21, 22 respectively, I get:
Rs
Psi
=
Ysi
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs−1
(
Lsi
1 − αs
)
1−αs
(19)
Ws
Psi
=
Ysi
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs
(
Lsi
1 − αs
)
−αs
(20)
Zs
Psi
=
Ysi
ψ
µ
1
ϵm
s M
−1
ϵm
si (21)
Denote by A, B, C the capital, labor and intermediate inputs cost shares respectively. I obtain:
A ≡
Rs
Psi
Ksi
Ysi
=
1
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs−1
(
Lsi
1 − αs
)
1−αsKsi (22)
B ≡
Ws
Psi
Lsi
Ysi
=
1
ψ
(1 − µs)
1
ϵm V A
−1
ϵm (
Ksi
αs
)
αs
(
Lsi
1 − αs
)
−αs
(23)
C ≡
Zs
PsiMsi
Ysi
=
1
ψ
µ
1
ϵm
s M
−1
ϵm
si (24)
Now, I take the ratio between A and B and get:
A
B
=
αs
1 − αs
(25)
Which yields:
αs =
A
A + B
≡
RsKsi
RsKsi + WsLsi
(26)
And to derive µs, I take the ratio between A and C and simplify to obtain:
(
1 − µs
µs
)
1
ϵm = ( V A
Msi
)
(
1 − ϵm
ϵm
)
A
C
(27)
Simplifying this equation gives:
µs =
M1−ϵm
si C
ϵm
M1−ϵm
si Cϵm + V A1−ϵmAϵm
(28)
Note that if I set ϵm = 1, the expression breaks down to:
µs =
ZsMsi
ZsMsi + RsKsi
(29)
173

Regressions with multiple selected obstacles
Table 42: Angola: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance (availability and cost) -0.0206 0.000339 0.00207
(0.0715) (0.104) (0.0524)
Transportation 0.0155 -0.159 -0.0484
(0.0817) (0.119) (0.0598)
Access to land 0.0919 0.273∗∗ 0.152∗∗∗
(0.0773) (0.113) (0.0566)
Tax rates 0.00962 -0.204 -0.0973
(0.0915) (0.133) (0.0670)
Labor Regulations -0.254∗
-0.526∗∗ -0.163
(0.141) (0.206) (0.103)
Corruption 0.0315 0.335∗∗∗ -0.0279
(0.0849) (0.124) (0.0621)
Workforce education 0.0596 -0.122 0.0322
(0.123) (0.179) (0.0898)
Political instability 0.258∗∗ 0.137 0.259∗∗∗
(0.111) (0.161) (0.0810)
Constant 18.12∗∗∗∗ 1.573∗∗∗ 0.243
(0.350) (0.511) (0.257)
Observations 189 189 189
Adjusted R2
-0.003 0.077 0.049
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the degree of output allocative
efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation; Capital distortions is log(1 + τksi);
Labor distortions is log(1 + τlsi). Independent variables: on a scale of 0 to 4, degree to which each factor constitutes an
obstacle to the firm, higher values meaning more severe obstacles.
174

Table 43: Cameroon: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.582∗∗ -0.314∗∗ -0.246∗
(0.225) (0.148) (0.127)
Transportation 0.568∗∗∗ 0.265∗ 0.224∗
(0.216) (0.142) (0.122)
Access to land -0.348 -0.160 -0.111
(0.237) (0.156) (0.134)
Tax rates 0.192 0.0867 0.283∗∗
(0.239) (0.157) (0.135)
Labor regulations -0.289 0.108 -0.157
(0.300) (0.197) (0.169)
Corruption 0.00575 0.0130 0.0104
(0.0271) (0.0179) (0.0153)
Workforce education 0.0192 -0.238 -0.0401
(0.256) (0.169) (0.145)
Political instability 0.302 -0.0433 -0.0844
(0.224) (0.147) (0.126)
Constant 22.97∗∗∗∗ 1.416∗ 0.412
(1.266) (0.833) (0.714)
Observations 102 102 102
Adjusted R2 0.033 -0.019 0.015
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
175

Table 44: Ethiopia: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.151 -0.00273 -0.0800
(0.124) (0.0936) (0.0728)
Transportation 0.200 -0.0941 -0.0884
(0.164) (0.124) (0.0964)
Access to land -0.360∗∗∗ 0.169∗
-0.0332
(0.119) (0.0896) (0.0696)
Tax rates 0.198 0.100 0.147∗
(0.129) (0.0977) (0.0759)
Labor regulations 0.763∗∗∗ -0.141 0.205
(0.236) (0.178) (0.138)
Corruption -0.115 -0.0209 0.0666
(0.141) (0.107) (0.0829)
Workforce education 0.311∗∗ -0.157 -0.138∗
(0.136) (0.103) (0.0799)
Political instability -0.00460 0.0178 0.0167
(0.146) (0.110) (0.0856)
Constant 7.990∗∗∗∗ 0.653∗∗ 0.653∗∗∗
(0.409) (0.309) (0.240)
Observations 229 229 229
Adjusted R2 0.097 -0.004 0.002
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
176

Table 45: Ghana: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.219∗∗ 0.0259 -0.0142
(0.0888) (0.0762) (0.0643)
Transportation 0.188∗ 0.263∗∗∗ 0.302∗∗∗∗
(0.0995) (0.0854) (0.0721)
Access to land -0.0421 -0.0561 -0.0539
(0.0884) (0.0759) (0.0640)
Tax rates 0.105 -0.117 0.0147
(0.0909) (0.0780) (0.0659)
Labor regulations 0.456∗∗ 0.0147 0.241∗
(0.195) (0.168) (0.142)
Corruption -0.169 -0.298∗∗∗ -0.229∗∗∗
(0.120) (0.103) (0.0870)
Workforce education -0.312∗∗ -0.307∗∗ -0.106
(0.149) (0.128) (0.108)
Political instability 0.0889 0.111 0.0120
(0.208) (0.178) (0.151)
Constant 23.20∗∗∗∗ 0.998∗∗∗∗ 0.300
(0.340) (0.292) (0.246)
Observations 268 268 268
Adjusted R2 0.055 0.073 0.068
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
177

Table 46: Guinea: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance (availability and cost) 0.0735 -0.0909 -0.0979
(0.124) (0.124) (0.0978)
Transportation 0.146 0.227∗ 0.114
(0.130) (0.131) (0.103)
Access to land 0.0662 0.141 0.0914
(0.136) (0.137) (0.108)
Tax rates -0.00463 -0.0639 -0.0188
(0.122) (0.123) (0.0969)
Labor Regulations 0.319 0.145 0.226
(0.228) (0.230) (0.181)
Corruption 0.0639 0.305∗∗ 0.154
(0.128) (0.129) (0.102)
Inadequately educated workforce 0.0316 -0.327∗∗ -0.0706
(0.161) (0.163) (0.128)
Political instability -0.0881 -0.0138 0.0353
(0.135) (0.136) (0.107)
Constant 19.64∗∗∗∗ 0.0118 -0.316
(0.670) (0.675) (0.531)
Observations 121 121 121
Adjusted R2
-0.047 -0.012 -0.041
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the degree of output allocative
efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation; Capital distortions is log(1 + τksi);
Labor distortions is log(1 + τlsi). Independent variables: on a scale of 0 to 4, degree to which each factor constitutes an
obstacle to the firm, higher values meaning more severe obstacles.
178

Table 47: Madagascar: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.286∗∗ -0.0737 -0.102
(0.142) (0.129) (0.116)
Transportation 0.204 0.247 0.206
(0.176) (0.160) (0.144)
Access to land 0.0856 0.110 0.135
(0.159) (0.144) (0.129)
Tax rates 0.155 0.206 0.233∗
(0.163) (0.148) (0.133)
Labor regulations 0.0819 0.225 -0.0793
(0.189) (0.172) (0.154)
Corruption -0.428∗∗∗ -0.462∗∗∗∗ -0.244∗∗
(0.143) (0.130) (0.117)
Workforce education 0.239 0.138 0.100
(0.153) (0.139) (0.125)
Constant 24.44∗∗∗∗ 0.540 0.315
(0.539) (0.490) (0.439)
Observations 117 117 117
Adjusted R2 0.038 0.063 0.012
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
179

Table 48: Mali: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.0920∗
-0.0650 -0.0317
(0.0550) (0.0541) (0.0464)
Transportation -0.0176 -0.0138 0.0596
(0.0735) (0.0724) (0.0620)
Access to land -0.0501 -0.0438 -0.0679
(0.0718) (0.0707) (0.0606)
Tax rates -0.0415 -0.0375 0.000116
(0.0579) (0.0570) (0.0489)
Labor regulations 0.0737 -0.0485 0.0158
(0.142) (0.139) (0.119)
Corruption 0.123∗ 0.0339 0.0326
(0.0747) (0.0735) (0.0630)
Workforce education 0.0684 0.0276 0.0786
(0.0985) (0.0970) (0.0831)
Political instability -0.0461 -0.000835 0.0113
(0.159) (0.156) (0.134)
Constant 19.96∗∗∗∗ 0.968∗∗∗∗ 0.454∗∗
(0.212) (0.209) (0.179)
Observations 251 251 251
Adjusted R2
-0.007 -0.026 -0.027
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
180

Table 49: Mozambique: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.283∗∗∗∗ -0.150∗
-0.150∗∗∗
(0.0779) (0.0847) (0.0495)
Transportation -0.103 0.0417 -0.0310
(0.0953) (0.104) (0.0606)
Access to land -0.0190 -0.228∗∗ -0.0302
(0.105) (0.114) (0.0666)
Tax rates 0.00108 0.0146 0.0408
(0.0915) (0.0994) (0.0582)
Labor regulations 0.374∗∗∗ -0.109 0.00390
(0.138) (0.150) (0.0876)
Corruption 0.259∗∗∗ 0.0598 0.0945
(0.0936) (0.102) (0.0595)
Workforce education 0.150 0.0879 0.0105
(0.106) (0.115) (0.0671)
Political instability -0.341∗∗ -0.307∗∗ -0.118
(0.142) (0.155) (0.0905)
Constant 17.25∗∗∗∗ 1.232∗∗∗∗ 0.727∗∗∗∗
(0.265) (0.288) (0.169)
Observations 336 336 336
Adjusted R2 0.081 0.006 0.009
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
181

Table 50: Nigeria: Distortions and Obstacles Regressions
Output Distortions Capital Distortions Labor Distortions
Access to finance (e.g. collateral) -0.0296 0.0502 -0.0158
(0.0399) (0.0411) (0.0291)
Transportation -0.0553 -0.0304 -0.0195
(0.0467) (0.0482) (0.0341)
Access to land -0.0809∗
-0.0236 -0.00413
(0.0416) (0.0429) (0.0304)
Tax rates 0.00283 0.0410 0.0585∗
(0.0439) (0.0452) (0.0320)
Labor regulations 0.120∗ 0.107 0.0330
(0.0690) (0.0711) (0.0504)
Corruption -0.150∗∗∗∗ 0.0352 -0.0266
(0.0446) (0.0460) (0.0326)
Workforce education -0.0262 -0.168∗∗ -0.0436
(0.0648) (0.0668) (0.0473)
Political instability -0.0169 -0.0938∗
-0.0410
(0.0527) (0.0543) (0.0384)
Constant 18.74∗∗∗∗ 0.910∗∗∗∗ 0.546∗∗∗∗
(0.133) (0.137) (0.0970)
Observations 884 884 884
Adjusted R2 0.027 0.001 -0.004
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the degree of output
allocative efficiency, log(1−τysi), so a negative coefficient indicates higher output misallocation; Capital distortions
is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent variables: on a scale of 0 to 4, degree to which each
factor constitutes an obstacle to the firm, higher values meaning more severe obstacles.
182

Table 51: Senegal
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.172∗∗ -0.0993 -0.0519
(0.0837) (0.0669) (0.0534)
Transportation -0.0540 0.0335 0.0616
(0.0939) (0.0751) (0.0599)
Access to land -0.110 -0.0720 0.0140
(0.0969) (0.0774) (0.0618)
Tax rates 0.162∗ 0.0539 0.0749
(0.0959) (0.0767) (0.0612)
Labor regulations -0.233 -0.0369 -0.155
(0.150) (0.120) (0.0959)
Corruption 0.0906 0.0177 0.0836
(0.111) (0.0884) (0.0705)
Workforce education 0.182 -0.0906 0.0976
(0.129) (0.103) (0.0826)
Political instability 0.253 0.116 0.0360
(0.158) (0.126) (0.101)
Constant 20.98∗∗∗∗ 1.079∗∗∗∗ 0.387∗∗
(0.245) (0.196) (0.156)
Observations 244 244 244
Adjusted R2 0.010 -0.024 -0.005
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
183

Table 52: Uganda
Output Distortions Capital Distortions Labor Distortions
Access to finance (availability and cost) -0.0738 0.0565 -0.0171
(0.0935) (0.0762) (0.0550)
Transportation 0.0445 0.180∗∗ 0.0276
(0.0986) (0.0803) (0.0580)
Access to land -0.275∗∗∗ 0.000322 -0.0219
(0.104) (0.0843) (0.0609)
Tax rates -0.0589 -0.0247 -0.0108
(0.107) (0.0869) (0.0628)
Labor Regulations 0.616∗∗∗ -0.0718 0.172
(0.229) (0.187) (0.135)
Corruption 0.379∗∗∗∗ 0.0750 0.0772
(0.101) (0.0823) (0.0594)
Workforce education -0.161 -0.274∗∗∗ -0.0677
(0.129) (0.105) (0.0759)
Political instability -0.306∗∗ -0.266∗∗∗ -0.152∗∗
(0.118) (0.0962) (0.0695)
Constant 22.39∗∗∗∗ 1.000∗∗ 0.572
(0.595) (0.485) (0.350)
Observations 290 290 290
Adjusted R2 0.066 0.026 -0.015
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the degree of output allocative
efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation; Capital distortions is log(1 + τksi);
Labor distortions is log(1 + τlsi). Independent variables: on a scale of 0 to 4, degree to which each factor constitutes an
obstacle to the firm, higher values meaning more severe obstacles.
184

Table 53: Zambia
Output Distortions Capital Distortions Labor Distortions
Access to finance -0.117 0.0729 -0.0889∗
(0.0996) (0.0768) (0.0532)
Transportation -0.0380 -0.0145 -0.0148
(0.127) (0.0981) (0.0679)
Access to land -0.0187 0.0808 0.0565
(0.116) (0.0895) (0.0620)
Tax rates 0.164 -0.0205 -0.0136
(0.114) (0.0876) (0.0607)
Labor regulations 0.443∗∗ 0.151 0.0763
(0.181) (0.139) (0.0964)
Corruption 0.00748 -0.157∗
-0.0813
(0.121) (0.0935) (0.0648)
Workforce education 0.381∗∗∗ 0.134 0.0932
(0.139) (0.107) (0.0740)
Political instability -0.587∗∗ -0.322 -0.317∗∗
(0.273) (0.211) (0.146)
Constant 23.69∗∗∗∗ 0.756∗∗∗∗ 0.745∗∗∗∗
(0.239) (0.184) (0.128)
Observations 279 279 279
Adjusted R2 0.046 -0.004 0.019
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variables: Output distortions is the
degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output
misallocation; Capital distortions is log(1 + τksi); Labor distortions is log(1 + τlsi). Independent
variables: on a scale of 0 to 4, degree to which each factor constitutes an obstacle to the firm, higher
values meaning more severe obstacles.
185

Robustness Checks
186

Table 54: Regressions of output distortions on financial constraint with bootstrap SEs
Mozambique Senegal Ghana Madagascar Nigeria Zambia
Financial constraint -0.3∗∗∗ -0.14∗∗ -0.24∗∗∗∗ 0.23∗ 0.034 0.019
(0.1) (0.07) (0.072) (0.13) (0.042) (0.11)
Size 1.75∗∗∗∗ 0.84 1.26 1.56∗∗∗∗ 1.57∗∗∗∗ 1.47∗∗∗
(0.46) (0.86) (0.81) (0.37) (0.32) (0.5)
Financial constraint × size 0.35∗∗∗ 0.33 0.12 -0.32∗∗∗∗ -0.11∗ 0.071
(0.11) (0.66) (0.21) (0.067) (0.057) (0.13)
Age 0.00099 -0.0120 0.0042 0.043∗∗∗ 0.016 0.025∗∗∗∗
(0.0064) (0.01) (0.0064) (0.016) (0.016) (0.0076)
EPZ 0.27 0.3 0.3 1.18∗∗∗∗ 0.52∗∗ 0.16
(0.2) (0.66) (0.27) (0.28) (0.21) (0.58)
EPZ × size -0.045 0.960 -0.22 -0.15 -0.14 0.1
(0.099) (1.49) (0.71) (0.2) (0.32) (0.53)
Percentage of inputs imported 0.0059∗∗ 0.0055 -0.0049 -0.00072 0.011∗∗∗∗
(0.0028) (0.0039) (0.0064) (0.0031) (0.0014)
Percentage of sales exported -0.0006 0.018 0.013 0.026 0.016∗∗
(0.0085) (0.019) (0.011) (0.016) (0.0062)
Percentage owned by state 0.081 0.011∗
-0.004∗∗∗∗
(0.14) (0.0068) (0.00073)
Observations 336 244 268 117 884 279
Adjusted R2 0.439 0.301 0.287 0.105 0.245 0.406
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry and country fixed effects. Dependent variable is the degree of output allocative efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation. Standard errors are bootstrap SEs.
187

Table 55: Regressions of output distortions on financial constraint with bootstrap SEs
Angola Cameroon Ethiopia Uganda Guinea Mali
Financial constraint 0.028 -0.0064 -0.18 -0.13 0.16∗∗∗∗ -0.058
(0.022) (0.27) (0.16) (0.12) (0.032) (0.08)
Size 0.42 1.16 2.087∗∗∗∗ 1.8∗∗∗∗ 1.18 1.27∗∗∗
(0.61) (1.21) (0.24) (0.35) (1.53) (0.44)
Financial constraint × size 0.23∗∗∗ -0.11 0.069 0.078 0.21 0.0026
(0.072) (0.27) (0.13) (0.13) (0.54) (0.068)
Age -0.027∗∗∗∗ -0.018 0.0017 -0.0068 -0.01 -0.0036
(0.0062) (0.012) (0.0044) (0.0073) (0.0073) (0.014)
EPZ 0.28∗∗ 1.48∗ 2.67∗∗∗∗ 0.22∗
-0.31 0.47∗∗∗∗
(0.12) (0.77) (0.55) (0.12) (0.55) (0.093)
EPZ × size 0.73∗ 0.07 -1.32∗
-0.089 0 -0.021
(0.4) (0.53) (0.69) (0.1) (0) (0.7)
Percentage of inputs imported 0.0062∗∗∗∗ 0.025∗∗∗∗ 0.0007 0.018∗∗∗∗ 0.0085∗∗∗ -0.00065
(0.0017) (0.0071) (0.0028) (0.0037) (0.003) (0.0022)
Percentage of sales exported 0.24∗∗∗∗ 0.022∗∗∗∗ -0.0075 0.017 -0.012 0.0086
(0.02) (0.0038) (0.013) (0.022) (0.025) (0.022)
Region -0.16∗∗ -0.12∗∗∗∗ -0.025 -1.005∗∗ -0.026
(0.072) (0.017) (0.15) (0.49) (0.2)
Percentage owned by state 0 0.0053∗∗ -0.038∗∗∗∗
(0) (0.0023) (0.011)
Observations 189 102 229 290 121 251
Adjusted R2 0.174 0.539 0.386 0.439 0.046 0.077
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry and country fixed effects. Dependent variable is the degree of output allocative
efficiency, log(1 − τysi), so a negative coefficient indicates higher output misallocation. Standard errors are bootstrap
SEs.
188

Table 56: Regressions of output distortions on financial constraint with bootstrap SEs
Financing Transportation Labor Reg Land Tax Corruption Education
Obstacle -0.089∗∗∗ -0.012 0.16∗∗∗∗ -0.034 0.042 -0.022 0.025
(0.0323) (0.027) (0.042) (0.039) (0.043) (0.029) (0.033)
Size 1.63∗∗∗∗ 1.76∗∗∗∗ 1.79∗∗∗∗ 1.73∗∗∗∗ 1.74∗∗∗∗ 1.68∗∗∗∗ 1.68∗∗∗∗
(0.19) (0.16) (0.15) (0.15) (0.22) (0.13) (0.17)
Obstacle
× size 0.066 0.01 -0.041 0.045 0.027 0.092
∗ 0.13∗∗
(0.054) (0.062) (0.063) (0.052) (0.065) (0.051) (0.063)
Age 0.0034 0.0032 0.0036 0.0031 0.0031 0.0028 0.003
(0.0045) (0.0049) (0.0041) (0.0042) (0.0045) (0.0042) (0.0041)
EPZ 0.4∗∗∗ 0.4∗∗∗ 0.38∗∗ 0.4∗∗∗ 0.41∗∗∗ 0.41∗∗∗ 0.4∗∗∗
(0.13) (0.13) (0.16) (0.13) (0.14) (0.15) (0.15)
EPZ
× size 0.07 0.061 0.065 0.065 0.063 0.044 0.036
(0.15) (0.16) (0.19) (0.18) (0.18) (0.17) (0.18)
Percentage of inputs imported 0.0048∗∗∗∗ 0.0048∗∗∗ 0.0047∗∗ 0.0049∗∗ 0.0047∗∗ 0.0048∗∗∗ 0.0046∗∗
(0.0015) (0.0017) (0.002) (0.002) (0.0019) (0.0016) (0.0019)
Percentage of sales exported 0.01∗∗∗ 0.01∗∗∗ 0.01∗∗∗∗ 0.01∗∗∗∗ 0.01∗∗∗∗ 0.01∗∗∗∗ 0.0098∗∗∗∗
(0.0032) (0.0032) (0.003) (0.0031) (0.0031) (0.0023) (0.0027)
Percentage owned by state 0.0069
∗ 0.0069 0.0066 0.007 0.007 0.0074 0.0047
(0.0042) (0.0056) (0.0062) (0.0079) (0.0055) (0.007) (0.0047)
Region 0.016 0.019 0.027 0.017 0.02 0.019 0.023
(0.039) (0.039) (0.034) (0.041) (0.048) (0.034) (0.031)
Observations 3091 3091 3091 3091 3091 3091 3091
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry and country fixed effects. Dependent variable is the degree of output allocative efficiency, log(1
− τysi), so
a negative coefficient indicates higher output misallocation. Standard errors are bootstrap SEs.
189

Table 57: Regressions of output distortions excluding size
Mozambique Senegal Ghana Madagascar Nigeria Zambia
Financial constraint -0.21∗∗∗ -0.13∗
-0.27∗∗∗ -0.16 -0.05 -0.018
(0.074) (0.075) (0.086) (0.15) (0.035) (0.086)
Age 0.013 -0.0019 0.01 0.042∗∗ 0.023∗∗∗∗ 0.045∗∗∗∗
(0.0099) (0.013) (0.011) (0.02) (0.0068) (0.0094)
EPZ 0.33 0.96∗∗ 0.41 1.46∗∗ 0.65∗∗∗∗ 0.56∗∗
(0.26) (0.41) (0.38) (0.63) (0.18) (0.28)
Percentage of inputs imported 0.0127∗∗∗∗ 0.0093∗∗∗ -0.00024 -0.0013 0.017∗∗∗∗
(0.0035) (0.0032) (0.0042) (0.0025) (0.0036)
Percentage of sales exported 0.031∗∗∗∗ 0.042∗∗∗∗ 0.026∗∗ 0.032∗∗∗ 0.03∗∗∗
(0.0092) (0.0099) (0.011) (0.011) (0.01)
Region -0.27∗∗ -0.16 -0.53∗∗∗∗ 0.15∗∗∗∗ 0.17
(0.13) (0.11) (0.11) (0.017) (0.12)
Observations 336 244 268 117 884 279
Adjusted R2 0.124 0.153 0.116 0.032 0.114 0.208
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variable is the degree of output allocative efficiency, log(1−τysi),
so a negative coefficient indicates higher output misallocation.
190

Table 58: Regressions of output distortions excluding size
Angola Cameroon Ethiopia Uganda Guinea Mali
Financial constraint 0.035 -0.23 -0.2∗
-0.047 0.2∗
-0.091∗
(0.07) (0.14) (0.11) (0.086) (0.12) (0.055)
Age -0.023∗∗ -0.0078 0.01 0.018 0.0066 0.0039
(0.012) (0.016) (0.014) (0.014) (0.021) (0.012)
EPZ 0.36∗ 1.8∗∗∗∗ 2.52∗∗∗∗ 0.21 -0.47 0.44∗
(0.19) (0.45) (0.44) (0.24) (0.52) (0.25)
Percentage of inputs imported 0.0089∗∗∗∗ 0.029∗∗∗∗ 0.001 0.03∗∗∗∗ 0.0082∗ 0.00089
(0.0025) (0.0051) (0.005) (0.004) (0.0046) (0.0028)
Percentage of sales exported 0.22 0.029∗∗∗∗ -0.0068 0.034∗∗∗∗ -0.011 0.011
(0.25) (0.0073) (0.013) (0.0079) (0.017) (0.008)
Region -0.091 -0.19∗∗∗∗ -0.075 -0.42 -0.047
(0.15) (0.037) (0.11) (0.45) (0.096)
Observations 189 102 229 290 121 251
Adjusted R2 0.068 0.503 0.257 0.211 -0.017 0.011
Standard errors in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
Notes: OLS regressions with industry fixed effects. Dependent variable is the degree of output allocative efficiency,
log(1 − τysi), so a negative coefficient indicates higher output misallocation.
191

Model of Financial Constraints
In light of the findings above, I solve a suggestive model of financial constraints following Midrigan
and Xu (2014), Zetlin-Jones and Shourideh (2017), and Buera et al. (2011). The goal of this exercise
is to suggest mechanisms by which productivity and size determine firms’ (in)ability to overcome
financial challenges and show the growth dynamics when firms are financially constrained.
There is a final output and an industry-level output produced in the economy, as in the misallocation
model above, except this time the model has a time dimension. From the profit maximization
problem at the industry-level in the misallocation model, I had:
PsitY
1/σ
sit = PstY
1/σ
st (30)
Following the assumption made by Hsieh and Klenow (2009) that PstY
1/σ
st = 1 and setting −σ = ˜η,
I obtain:
Psit = Y
1/η˜
sit (31)
The economy is inhabited by individuals who decide at the beginning of each period to work for a
wage or become entrepreneurs. Their choice is based on the quality of their entrepreneurial ideas,
−→Q. The vector of entrepreneurial ideas is drawn from a distribution µ(
−→Q) and they die with a
constant hazard rate of (1 − γ) in which case a new vector of ideas is independently drawn from
µ(
−→Q). Individuals maximize their lifetime utility:
U(C) = E[
X∞
t
β
tu(ct)] (32)
where ct
is the consumption at time t for workers, and for entrepreneurs, they maximise their utility
over their dividends, Dt
.
Entrepreneurs own plants that produce the following output, as in the misallocation model:
Ysit = Asit[(1 − µs)
1
ϵm ((Ksit
αs
)
αs
(
Lsit
1 − αs
)
1−αs
)
ϵm−1
ϵm + µ
1
ϵm
s M
ϵm−1
ϵm
sit ]
ϵm
ϵm−1 (33)
192

Asit grows at rate g, which is deterministic and endogenous. Entrepreneurs own their capital stock,
buy intermediate inputs and hire labor. Their profits are given by:
πsit = PsitQsitYsit − Ksit − LsitWst − MsitZst − fs (34)
where Psit is the price of firm i’s output, Qsit is the quality of the entrepreneur’s idea, Ksit is
capital owned by the entrepreneur, Lsit is labor inputs, Wst is the wage paid to workers, Msit is
intermediate inputs, Zst is the price of intermediate inputs, and fs is the fixed cost of entry to each
sector s. The profit function can be rewritten as:
πsit = QsitY
η
sit − Ksit − LsitWst − MsitZst − fs (35)
where η =
σ−1
σ =
η˜+1
η˜
Financial side of model: There are competitive financial intermediaries that receive deposits
from workers and lend the money to entrepreneurs in the form of bonds. Entrepreneurs use that
money to pay for factors of production and the fixed cost, and produce. They receive proceeds
from sales at the beginning of next period and pay back their debt.
Suppose Nsit−1 is the plant’s net worth at the beginning of period t and after the entrepreneur
repays their debt:
Nsit−1 = Qsit−1Y
η
sit−1 + (1 − δ)Ksit−1 − Bsit−1(1 + r) (36)
where Bsit−1 is firm’s borrowing at time t − 1. The plant’s borrowing constraint at time t is:
Bt ≤ λ[QsitY
η
sit + (1 − δ)Ksit] = λ(Nsit + Bsit) (37)
and its dividends are given by:
Dsit = Qsit−1Y
η
sit−1 + (1 − δ)Ksit−1 − Bsit−1(1 + r) − Ksit − LsitWst − MsitZst + Bsit − Psitfs (38)
Recursive problems: An Individual’s state depends on their net worth N, or assets for workers,
193

a, and their ideas, −→Q.
Value function of being a worker:
V
w(N, Q) = maxC,a′U(C) + β[γV (N
′
, Q) + (1 − γ)EQ′[V (N
′
, Q′
)]] (39)
s.t.
P C + a
′ ≤ W + (1 + r)a (40)
where a is asset of the individual worker saved with the financial intermediaries.
Value function of being an entrepreneur:
V
e
(N, Q) = maxK,N′
,M,LU(D) + β[γV (N
′
, Q) + (1 − γ)EQ′[V (N
′
, Q′
)]] (41)
s.t.
D ≤ QY η + (1 − δ)K − (1 + r)B − K − W LMZ + B
′ − P f (42)
B ≤ λ[QY η + (1 − δ)K] (43)
Rewriting the constraints on the entrepreneur’s problem, I obtain:
V
e
(N, Q) = maxK,N′
,M,LU(D) + β[γV (N
′
, Q) + (1 − γ)EQ′[V (N
′
, Q′
)]] (44)
s.t.
D = N − K − W L − ZM +
1
1 + r
(QY η + (1 − δ)K − N
′
) (45)
N
′ ≥ (1 − (1 + r)λ)(QY η + (1 − δ)K) (46)
Definition of a recursive competitive equilibrium: A recursive competitive equilibrium is
a distribution of −→Q denoted with µ(
−→Q); policy functions K(N, Q), N’(N, Q), M(N, Q), L(N, Q),
C(N, Q), a’(N, Q); and prices W, Z, R P such that:
194

1. given prices, the policy functions K(N, Q), N’(N, Q), M(N, Q), L(N, Q), C(N, Q), a’(N, Q)
solve (29), (30), (31), (32) and (33);
2. financial market clears: a
′ = B′
Solving the entrepreneur’s recursive problem, I get the following first order conditions:
U
′
(D) = (1 + r)β(1 − γ)EQ′[V
e
1
(N
′
, Q′
)] + (1 + r)θ (47)
RK(1 + r)λ + (1 − (1 + r)λ)
1
U′(D)
β(1 − γ)(1 + r)EQ′(V
e
1
(N
′
, Q′
)) = 1 + r (48)
RL(1 + r)λ + (1 − (1 + r)λ)
1
U′(D)
β(1 − γ)(1 + r)EQ′(V
e
1
(N
′
, Q′
)) = W(1 + r) (49)
RM(1 + r)λ + (1 − (1 + r)λ)
1
U′(D)
β(1 − γ)(1 + r)EQ′(V
e
1
(N
′
, Q′
)) = Z(1 + r) (50)
where θ is the multiplier on the borrowing constraint, and RK, RL and RM are as follows:
RK = η(1 − µ)
1/ϵmY
ηQ
1
ψ
V A
ϵm−1
ϵm (
K
α
)
−1 + 1 − δ (51)
RL = η(1 − µ)
1/ϵmY
ηQ
1
ψ
V A
ϵm−1
ϵm (
L
1 − α
)
−1
(52)
RM = ηµ1/ϵmY
ηQ
1
ψ
M−1/ϵ
s
(53)
and
V A = (K
α
)
α
(
L
1 − α
)
1−α
(54)
Comparative statics: Equations (37) to (40) summarize the dynamics of firms’ growth under
borrowing constraints. Equation (37) relates the marginal utility of today’s consumption for the
entrepreneur and the discounted marginal utility of its future value function. If the entrepreneur
is constrained and θ ̸= 0, then marginal utility of consuming today is higher than the discounted
marginal utility of its future value function. This implies entrepreneurs will invest less in input
factors and shows the ways in which financial constraint can limit firms’ use of factor inputs and
create distortions.
195

Equation (38), (39), and (40) show the relationship between the returns to the factors or production
RK, RL, and RM and the interest rate. If the borrowing constraint does not bind and θ = 0, then
the returns to the factors of production equal to (1+r). However, if the entrepreneur is constrained
and θ ̸= 0, then (1 + r) ≤ the returns to the factors of production, meaning that the entrepreneur
is using below optimal levels of the input factors. As λ increases, meaning that the entrepreneur’s
borrowing constraint is loosening, then (1+r) increases and thus the gap between (1+r) and RK,
RL, and RM decreases and the equilibrium outcome gets closer to optimality.
The model can be calibrated using firm-level panel data. Unfortunately, I have not found adequate
panel data for African countries. The Orbis database has very comprehensive and rich firm-level
data on most countries, but for African countries, it is mostly very large firms that report their
financial accounts information and so Orbis only has data on very large firms in Africa. Therefore,
using that data would not be adequate for my analysis given that I am mostly interested in showing
how small firms in Africa are constrained from expanding.
196

Chapter Three: Appendix
Robustness Checks
Table 59: South Africa LFS: Clustering Standard Errors at the Province Level
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the province level.
Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours worked.
Gender=0 if male and 1 if female. Controls are age, age squared, education level, marital status and ethnicity.
R-squared is the within one.
197

Table 60: South Africa LFS: Clustering Standard Errors at the Occupation Level
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the occupation
level. Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours
worked. Gender=0 if male and 1 if female. Controls are age, education level, marital status and ethnicity. Rsquared is the within one.
Table 61: South Africa LFS: Clustering Standard Errors at the Sector Level
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the sector level.
Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours worked.
Gender=0 if male and 1 if female. Controls are age, education level, marital status and ethnicity. R-squared is
the within one.
198

Table 62: South Africa LFS: Effects of Marriage on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours
worked. Gender=0 if male and 1 if female. Married=0 if not married and 1 if married. Controls are age,
education level, marital status and province. R-squared is the within one.
Table 63: South Africa NIDS: Effects of Age on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Columns (1)-(4) dependent variables: employment of all forms, business employment, monthly income and
weekly hours worked. Gender=0 if male and 1 if female. Controls are age, age squared, education level, marital
status and province. R-squared is the within one.
199

Table 64: South Africa LFS: Effects of Age on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours
worked. Gender=0 if male and 1 if female. Controls are age, age squared, education level, marital status and
ethnicity. R-squared is the within one.
Table 65: South Africa NIDS: Effects of Education on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, NIDS dataset. Standard errors clustered at the individual
level. Columns (1)-(4) dependent variables: employment of all forms, business employment, monthly income and
weekly hours worked. Gender=0 if male and 1 if female. Controls are age, age squared, education level, marital
status and province. R-squared is the within one.
200

Table 66: South Africa LFS: Effects of Education on Selected Outcome Variables
Standard errors in parentheses
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗
p < 0.1
Notes: Panel regressions with individual fixed-effects, LFS dataset. Standard errors clustered at the individual
level. Columns (1)-(3) dependent variables: employment of all forms, business employment, and weekly hours
worked. Gender=0 if male and 1 if female. Controls are age, age squared, education level, marital status and
ethnicity. R-squared is the within one.
201

Addressing Gender Inequality in Southern Africa: An Overview
Legal and Strategic Framework on Gender Equality
South Africa has a long history of combating gender inequality. The Commission for Gender Equality was established in 1996 and the National Policy Framework for Women’s Empowerment and
Gender Equality (National Gender Policy Framework, 2014)10 was adopted, focusing on fostering
women’s employment, mainstreaming gender policies and legislation, and empowering women in
rural areas. In addition, gender equality is one of the cardinal objectives of the National Development Plan (NDP) 2030, aiming to eliminate poverty and reduce inequality by 2030. With the NDP,
the authorities have promoted gender equality by fostering a fair environment for young women to
attain higher education and participation in economic activities (e.g., Women Empowerment Fund
of the National Empowerment Fund11), ruling out barriers for underdeveloped regions, increasing
food security, expanding health treatments, and protecting women from crime. Furthermore, building on the 1996 Constitution, South Africa has consistently improved the legislation on women,
especially on disallowing gender discrimination (the Employment Equity Act 1998; National Empowerment Fund Act 1998; the Promotion of Equality and Prevention of Unfair Discrimination
Act 4 of 2000), eliminating GBV and consolidating the justice system for GBV survivors (Women
empowerment and gender equality bill 2013; Criminal and Related Matters Amendment Act 12
of 2021; the Criminal Law (Sexual Offences and Related Matters) Amendment Act 13 of 2021;
the Domestic Violence Amendment Act 14 of 2021). Additional policy responses on GBV have
been taken since 2019, including the Emergency Response Action Plan in 2019 and the National
Strategic Plan on GBV and Femicide12 published in 2020, which sourced funding through budget
reprioritization and targeted to be completed before the first half of 2024.
Addressing gender inequality is also an undisputed element in Eswatini’s national strategy since
the mid-2000s. The new Eswatini Constitution (2005)13 underscored that gender inequality is
10The implementation of the National Gender Policy Framework is centralized by the Ministry of Women in the
Presidency.
11It provides a range of funding instruments for business whose ownership is at least 51 percent made up of black
women.
12The Emergency Response Action Plan and GBV and Femicide underpin the existing implementation aimed at
increasing women’s participation in economic activities.
13The Constitution also stipulates a quota on the number of female persons in the Parliament.
202

one of the aspects of discrimination. A National Gender Policy adopted in 2010 identifies key
areas to tackle gender inequality, including improving social awareness, economic empowerment,
women’s leadership, healthcare inclusiveness, and legal and education development. Drawing from
the National Gender Policy, Eswatini’s authorities implemented two initiatives, the Lower Usuthu
Smallholder Irrigation Project and the Rural Finance Enterprise Development Programme, that are
targeted toward reducing poverty in rural areas and redressing social norms and cultural elements
which hinder gender equality. These two initiatives are supported by the Government of Eswatini
and the International Fund for Agricultural Development. Furthermore, to mainstream gendersensitive policies, the Eswatini authorities received adequate training from UNDP and UNFPA to
ensure gender-responsive budgeting can be institutionalized. A further important step was taken
in 2018, when the Sexual Offences and Domestic Violence Act (2018) was introduced to criminalize
sexual and gender-based violence (SGBV). Supplemented with the National Strategy and Action
Plan to End Violence (2017-2022), a workflow against SGBV has been established across regions and
various ministries. As SGBV is also closely related to HIV/AIDS infection, especially for adolescent
girls and young women, the authorities developed a National Strategic Framework 2018-2023 to
end HIV/AIDS.
Lesotho has significantly strengthened its legal and institutional framework to foster gender equality. Lesotho’s authorities embarked on initiatives to tackle gender inequality much later than
those in South Africa, and the main focus is on improving the legal framework of gender equality. Women have been legally equal to men since 2008, after laws strengthening married women’s
rights14, women’s leadership15 and financial inclusiveness.16 A department of Gender was created
in the Lesotho Ministry of Gender and Youth, Sports and Recreation to address gender issues, especially GBV. In addition, the Department of Gender initiated an anti-GBV coordination forum to
strengthen the ongoing efforts on tackling GBV. A key legislation on protecting women from GBV,
14Lesotho Legal Capacity of Married Persons Act (LCMPA) enacted in 2006 removes marital power and allows
married women to have equal say in all joint matrimonial properties and issues. The Land Act of 2010 further ensures
equal access to land by men and women, which could potentially improve female access to collateral, and therefore,
access to finance.
15Lesotho Companies’ Amendment Act (2008) allows women to be promoted to the management level, to operate
their own business, and to be granted loans from commercial financial institutions. According to the Lesotho Local
Government Elections Act amended in 2011, women reserve 30 percent of seats in municipal, urban, and community
councils.
16Lesotho Bank Saving and Development Amendment Act (2008) legally enable women to open bank accounts
without permission from their husbands.
203

the Domestic Violence Bill 2021, was passed by the National Assembly in April 2022, reflecting a
historical progress for Lesotho on GBV issues.
Since its independence in 1990, gender equality has been integrated into the development of
Namibia. Improving gender discrimination as a step toward equality for all was well established
in Namibia’s Constitution, supplemented by the Married Persons Equality Act introduced in 1996
to ensure the rights of married women. Legal reforms that helped law enforcement on gender
inequality also include the Local Authorities Act (1992),17 Traditional Authorities Act (1995),
and Combating of Domestic Violence Act (2003). Besides the development of a legal framework,
progress is also notable in national planning. The first National Gender Policy was formulated in
1997, along with the National Gender Plan of Action. The Namibian government mainstreamed
its commitment to gender equality into Gender Responsive Budgeting (GRB), in response to the
GRB guidance constructed in 2015 by the Ministry of Gender Equality and Child Welfare. With
GRB, gender-oriented expenditure is planned within the Medium-Term Expenditure Framework.
It includes budget allocation on increasing income and employment, and improving health, GBV,
and education. Attributed to years of efforts in promoting gender equality, Namibia has been
highlighted as the second top-placed country in SSA based on its narrowing gender gap (World
Economic Forum, 2022).
Gender-sensitive Policies during the Pandemic
South Africa provided targeted initiatives and financial support to tackle gender inequality during
the pandemic. GBV emergency help was reinforced during the national lockdown and the stay-athome strategy against the COVID-19 pandemic (UNDP Gender Tracker). This includes a national
24/7 call center facility18 that operates across various platforms. In terms of financial assistance
and cash transfer, the Early Childhood Development Stimulus Relief Fund19 was established in
June 2020 to ease households’ financial burden from the loss of income caused by COVID-19.
In addition, the Presidential Employment Stimulus also introduced grant support in early 2020.
17The Act introduced a quota of female persons on elections of any local authority council.
18It is operated by the GBV Command Centre under The Department of Social Development.
19It is a relief fund for employees of Early Childhood Development services business who were registered, conditionally registered or unregistered, aiming to make up the loss of income due to COVID-19 by sending out a maximum
of R4,470 one-time payment to eligible people. This program is operated by the Department of Social Development.
204

Under the Presidential Employment Stimulus, the department of Women, Youth and Persons with
Disabilities received R119 million budget to support youth-owned informal SMEs, a third of which
are owned by female entrepreneurs (Building a Society that Works - Presidential Employment
Stimulus, 2020). Other financial support programs explicitly targeted women during the pandemic,
including support for SMEs through a debt relief finance scheme, a Business Growth/Resilience
Facility, the Tourism Equity Fund,20 the Township and Rural Entrepreneurship Programme, and
the introduction of agro-processing and agribusiness programs for women entrepreneurs.
Gender-sensitive policies were implemented during the pandemic also in Eswatini. The MSME
Revolving Relief Fund and the Eswatini Economic Recovery Plan include gender-sensitive policies.
Since more than half of employees in the MSMEs are women and female employment in the manufacturing sector is larger than male employment, female employees can be expected to benefit more
from these policies.
During the pandemic, financial support through social transfers specifically supported women in
Lesotho. Given that about 80 percent of workers in the textile sector are women, providing a threemonth salary subsidy to each employee in the industry supported female workers. A three-month
disbursing grant scheme was also guaranteed by the government’s Private Sector Competitiveness and Economic Diversification Project and was targeted the tourism sector where women are
also predominantly employed. In addition, female-led enterprises in Lesotho also benefit from the
government’s tax relief plan (IMF Financial Access COVID-19 Policy Tracker).
Multiple social programs in Namibia supported gender equality during the COVID-19 pandemic.
In response to the immediate need for medical and health services, UNFPA Namibia partnered
with Namibia Planned Parenthood Association and Society for Family Health to provide digital
platform and medical care for youth who need tests on sexual and reproductive health and HIV.
Online facilities and shelters were also offered to women and adolescent girls who suffered SGBV. In
addition, Namibia’s Build Back Better program provided food support and interventions to household food production in collaboration with the Government of Japan, UNDP Namibia, Ministry
of Agriculture, Works and Land Reform and four municipalities. The Namibia authorities also
20All three financial support programs were set to give priorities to businesses owned by women, youth, and people
with disabilities.
205

implemented cash transfers to help the recovery of women in informal employment (UN Women –
ILO Policy Tool, 2021).
Continuing efforts to tackle gender inequality are still needed, though progress has been made in
the past decades through legislative reforms and economic policies. In the past twenty years, South
Africa, Eswatini, Lesotho and Namibia have all set up ministries or departments on gender issues
and established national strategies or legislative reforms on gender. Anti-SGBV is also a priority
area. During the pandemic, countries launched several programs or financial packages targeted
at women, or industries and occupations that may benefit women more than men. Nevertheless,
Southern African countries should ensure enacted legislative frameworks are implemented to ensure
equal rights for women. Governments in Southern African countries should also extend such support
programs for women even after the pandemic. In addition, given the relatively limited gendersensitive policies implemented, based on the UNDP-UNW COVID-19 Global Gender Response
Tracker (see Table 1), more economic support and new forms of gender-sensitive policies are needed.
206

Abstract (if available)

Abstract This dissertation contributes to our understanding of African economies and their development outcomes, as well as various factors affecting their macroeconomic growth. In the first chapter, I investigate the driving forces behind divergent economic growth and structural transformation in Mauritius and Senegal, two historically similar sub-Saharan African nations with differing devel- opment trajectories. In chapter two, I explore the extent to which financial constraints contribute to the firm-level resource misallocation that I show is present in 12 sub-Saharan African countries. Finally, chapter three analyzes the impact of COVID-19 on gender disparities in the Southern African region, notably in South Africa, Eswatini, Lesotho, Botswana, and Namibia. Each chapter addresses pivotal issues crucial for comprehending macroeconomic development in Africa, including productivity, resource allocation inefficiencies, gender inequality, and the impact of macroeconomic shocks. Consequently, this dissertation makes substantive contributions to policy-relevant research in Africa.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays in macroeconomics and macro-finance

PDF

Essays in macroeconomics

PDF

Three essays in international macroeconomics and finance

PDF

Growth, trade and structural change in low income industrializing economies

PDF

Essays in development economics

PDF

Essays on applied microeconomics

PDF

Inter-temporal allocation of human capital and economic performance

PDF

Essays on development economics and adolescent behavior

PDF

Structural transformation and globalization

PDF

Essays on the dual urban-rural system and economic development in China

PDF

Bijection between reduced words and balanced tableaux

PDF

Early-warning systems for crisis risk

PDF

Three essays on the microeconometric analysis of the labor market

PDF

Essay on monetary policy, macroprudential policy, and financial integration

PDF

Essays in uncertainty and aggregate economic activity

PDF

Essays in international economics

PDF

Entanglement parity effects in quantum spin chains

PDF

Essays on sovereign debt

PDF

Essays on labor and development economics

PDF

The impact of digital transformation on urban communities, welfare, and distributional outcomes

Asset Metadata

Creator Thioune, Fatou Kiné (author)

Core Title Growth and development in Africa: macroeconomic and microeconomic perspectives

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2024-05

Publication Date 03/01/2024

Defense Date 02/20/2024

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

Tag Africa,COVID-19,Development,gender inequality,Growth,macroeconomics,misallocation,structural transformation

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Betts, Caroline (committee chair), Aizenman, Joshua (committee member), Kondo, Illenin (committee member), Nugent, Jeffrey (committee member)

Creator Email thioune@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113842316

Unique identifier UC113842316

Identifier etd-ThiouneFat-12678.pdf (filename)

Legacy Identifier etd-ThiouneFat-12678

Document Type Dissertation

Format theses (aat)

Rights Thioune, Fatou Kiné

Internet Media Type application/pdf

Type texts

Source 20240304-usctheses-batch-1127 (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.

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