SUSU COLLECTION in GHANA by Kweku A. Opoku-Agyemang

THE POLITICAL AND BEHAVIORAL FOUNDATIONS OF INCLUSIVE

ECONOMIC INSTITUTIONS: SUSU COLLECTION IN GHANA

Kweku A. Opoku-Agyemang

A dissertation submitted in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

(Development)

at the

UNIVERSITY OF WISCONSIN-MADISON

2012

Date of final oral examination: 5/15/12

This dissertation is approved by the following members of the Final Oral Committee:

Jeremy Foltz, Associate Professor, Agricultural and Applied Economics Aili Tripp, Professor, Political Science Scott Straus, Professor, Political Science Gay Seidman, Professor, Sociology Jennifer Alix-Garcia, Assistant Professor, Agricultural and Applied Economics

For my parents, and for Kwabena and Maame Adwoa. Ewuradze yε da w’ase.

ACKNOWLEDGMENTS

I am grateful for so much from so many. Thank God for wonderful experiences and friends, old and new, all treasures.

I am very grateful to my dissertation supervisor, Jeremy Foltz for his help in refining my ideas, his warmth and patience, and most of all, his generosity. I also thank my dissertation committee members, Aili Tripp, Scott Straus, Gay Seidman and Jennifer Alix-Garcia for their helpful suggestions, advice and support. I thank my undergraduate advisor, D.A. Laryea, and my secondary school teachers Robert Akpalu and Mr. Asare for much inspiration.

My dissertation field work in Ghana went much smoother than I dared hope thanks to too many people to list here. Yet I must thank John V. Mensah for hosting me at the Institute for

Development Studies of the University of Cape Coast and his helpful advice. I also thank Samuel

Sackey and the financial and susu officers of Kakum Rural Bank for excellent assistance. I thank

Ebo Sam for amazing logistical support. I thank Bertha Ansah-Djan and Enoch Donkoh of the

Microfinance and Small Loans Center; The Honourable Seth Terkper of the Ministry of Finance and Economic Planning; Mr. Pinkrah, George Tokpo, Alex Asmah, Araba Kwansema Brown,

Frances Adu-Mante, Robert Brown, Kojo Mbir, Rose Acquah, Elizabeth Agbemashior, Atta

Britwum, Paul Appiah, Elaine Kwami, David Odoi, and Kofi Awosabo Asare for being so generous with their time. I am especially grateful to all of the people of the Central and Greater

Accra Regions that participated in the study.

I received generous funding from an M.E.O. Fellowship, a Scott-Kloeck Jensen

Fellowship, a Center for World Affairs and Global Economy Fellowship, an A. Eugene Havens

iii

Award, a Raymond J. Penn Fellowship and a William Thiesenhusen Memorial Award: I sincerely thank everyone who gave so kindly. The work and care of the University of Wisconsin

Development Studies Program Coordinator, Chris Elholm, has been integral to my success. I also greatly appreciate the efforts of Nancy Carlisle, Cindy Munn, Vernetta Reid, and John Ackah of the Agricultural and Applied Economics Department.

Many graduate students and friends helped and encouraged me during my research journey. I would like to thank Victor Okorie, Theresa Ennin, Kwaku Addae-Mununkum, Kobina

Otoo, Kobina Amuah, Orville Harris, Gergens Polynice, Ekow Ewoodzie, Suzanne Reis, Linda

Park, Adam Auerbach, Charles Taylor, Meina Cai, Barry Driscoll, Linda Vakunta, Hillary

Caruthers, Nancy Rydberg, Thaís Passos Fonseca, Sonia Ares, Jack Buchanan, Rebecca Cleary,

Stephanie Prellwitz, Maria Rodrigo, Katherine Zipp, Dylan Fitz, Patricia Yanez-Pagans, Rachid

Laajaj, Trevor Young-Hyman, Stella Kim, Debby Sumwalt, Shoshana Griffith, George Acquah,

Horacio Aguirre, Julia Collins, Ya-Ting Chuang, Jhin Han, Dong Tam Le, Wilson Law, John

Chung-En Liu, Kyal Berends, Emily Sellars, Jared Gars, Daniel Kanyam, Iddisah Sulemana,

Stan Anamuah-Mensah, Edwin Amonoo, Maythiwan Kiatgrajai, Hannah Lanser, Wilson Cong,

Xiang Zhong, Madhumanti Sardar, Xiaodong Wang, Michael Berkowitz and Chi-Alpha, Jon

Dahl and IV, Paul and Donna Bell, Phil and Linda Malsack, Zach Stolfus, Linda Pankow, Mike

Crossley, Elzan Godlewski, and Hannah Doku. Faith Community Bible Church and House

Group have been immense places of support for me. Special thanks to my parents, Kwadwo and

Naana Opoku-Agyemang for being my role models. I also thank my brother and sister, Kwabena and Maame Adwoa for motivating me to do my very best.

Thank you all. iv

Contents

Contents iv

List of Tables viii

List of Figures xiv

Abstract xv

1 Introduction 1

2 Field and Data Environment, Survey Methodologies, and Project Overview 11

2.1 Introduction ...... 11

2.2 Field Environment and Setting: Ghana’s Central Region ...... 13

2.2.1 Community Banking in Central Ghana: Kakum Rural Bank . 19

2.3 Survey Design, Sample Selection and Methodology ...... 21

2.4 Sample Selection and Size ...... 21

2.4.1 Susu Collectors as Enumerators and Sampling Methodology . 22

2.4.2 Interviews ...... 24

2.4.2.1 Susu Collector and Client Questionnaires ...... 25

2.5 Selected Descriptive Statistics and Project Overview ...... 26

2.5.0.2 Deposit Susu Collectors ...... 26 v

2.5.1 Susu Clients ...... 31

2.6 Conclusion ...... 42

3 Can State Capacity be built out of the Informal Economy? Financial Re-

forms in Ghana 49

3.1 Introduction ...... 49

3.2 State Capacity, Politics, and Susu in Ghana ...... 58

3.2.1 Introduction ...... 58

3.2.1.1 Origins and Structure of Licensing Frameworks in

Ghana ...... 59

3.2.1.2 Legal and Fiscal Capacities: Microﬁnance Regula-

tion and Formal-Informal Financial Markets in Ghana 60

3.2.1.3 Fiscal Capacities: Informal Taxation and State Ca-

pacity ...... 63

3.3 Theoretical Preliminaries for State Capacity ...... 65

3.4 The Ghana Poverty Reduction Strategies and MASLOC as Microeco-

nomic Policy ...... 67

3.4.1 The Poverty Reduction Strategies and the Evolution of MASLOC

in Ghana ...... 67

3.4.2 The Microﬁnance and Small Loans Center (MASLOC) and the

goals of microeconomic policy in Ghana ...... 69

3.4.2.1 Current Target Beneﬁciaries of MASLOC Facilities 70

3.4.3 Direct and Indirect Modules of MASLOC ...... 71

3.4.3.1 Direct Module: MASLOC Module 1 ...... 71

3.4.3.2 Indirect Module: MASLOC Module 2 ...... 72

3.5 Inclusiveness versus Adverse Selection: Confronting the Theory with

Qualitative Data ...... 73 vi

3.6 A Basic Signaling Model: Susu Collection as a Mechanism of Building

State Capacity ...... 77

3.6.1 Informal Taxation and State Capacity: The Implications of

MASLOC and Susu Collection ...... 78

3.6.2 Summary ...... 79

3.7 Concluding Comments and Policy Implications ...... 80

3.8 A Model of State Capacity and Inclusive Economic Institutions . . . . 83

3.8.1 A Basic Model ...... 84

4 Commitment Savings subject to Personal Rules: Ghanaian Susu Collec-

tion 96

4.1 Introduction ...... 96

4.2 Susu Collectors and Savings Mobilization in Ghana ...... 102

4.3 Model and Predictions ...... 104

4.4 Survey Data, Savings Schedules Design and Measuring Internal Pref-

erences for Savings Commitments ...... 107

4.5 Empirical Strategies: OLS and Propensity Score Matching Estimators 110

4.6 Savings Mobilization Results ...... 113

4.6.1 Savings Schedules Impacts On Contributions: OLS Results . . 114

4.6.2 Savings Schedules as Personal Rules and Savings Contribu-

tions: Using all other savings schedules as comparison groups 116

4.6.3 Savings Schedules Impacts On Contributions: Using a single

savings schedule as a comparison group ...... 120

4.6.4 Susu Savings Rates Results ...... 124

4.6.5 Schedule Impacts On Savings Rates: OLS and Propensity Score

Matching ...... 126

4.6.6 Schedule Impacts On Savings Rates: Using a single savings

schedule as a comparison group ...... 132 vii

4.7 Discussion of Results ...... 136

4.8 Conclusions and Policy Implications ...... 137

5 Heterogenous Signaling at the Convergence of Formal and Informal Fi-

nance in Ghana 140

5.1 Introduction ...... 140

5.2 The Susu Collection Institution at the Convergence of Formal and In-

formal Finance in Ghana ...... 144

5.2.1 Susu Collectors and The Need for Credit among Saving En-

trepreneurs ...... 146

5.3 A Model of Susu Savings and Creditworthiness Signaling in Ghana . . 148

5.3.1 The Basic Setup ...... 149

5.3.2 Model Environment given Several Levels of Signaling Efforts 155

5.4 Data Description and Estimation Strategies ...... 163

5.4.1 Data Description ...... 163

5.4.2 Estimation Strategy ...... 166

5.5 Conclusions ...... 186

6 Women Empowerment, Gender Bias and Susu Collection in Ghana 188

6.1 Introduction ...... 188

6.2 Literature ...... 193

6.3 Matrilineality, Gender and Entrepreneurship in Central Ghana . . . . 194

6.3.1 Susu Collection, Rural Banks and the Political Economy of

Women’s Enfranchisement in Central Ghana ...... 195

6.4 Theoretical Models ...... 197

6.4.1 Signaling and Cross-Gender Bias in Susu Savings Mobilization

and Credit Provision ...... 198 viii

6.4.2 Gender Matching and Susu Savings Mobilization and Credit

Provision ...... 201

6.4.2.1 Hypotheses ...... 202

6.5 Empirical Tests: Female Susu Collectors and Gender-Matched networks 203

6.5.1 Speciﬁcations ...... 208

6.5.2 Empirical Strategy: Propensity Score Matching ...... 210

6.6 Results ...... 211

6.6.1 Susu Collector-Susu Client Gender Networks ...... 213

6.6.2 Alternative Explanations: Education and Effort in Gender-Bias,

Susu Savings and Credit Provision ...... 222

6.6.3 Survey Evidence and Mechanisms of Education and Economic

Effort ...... 223

6.6.3.1 Education and Gender Bias ...... 223

6.6.3.2 Economic Effort and Gender Bias ...... 228

6.6.4 Discussion of Education and Economic Effort Results . . . . 237

6.7 Conclusion and Policy Implications ...... 237

7 Conclusions and Future Research 245

7.1 Matching Protocol for Propensity Score Estimations ...... 249

7.1.1 Logit Regressions: Determinants of Susu Savings Schedules

(Savings) ...... 250

7.1.2 Logit Regressions: Determinants of Susu Savings Schedules

(Credit) ...... 300

7.1.3 Logit Regressions: Determinants of Susu Savings Schedules

(Gender) ...... 318

Bibliography 328

LIST OF TABLES

Table 2.1 Descriptive Statistics of Susu Collectors 28 2.2 Descriptive Statistics of Susu Clients 32 2.3 Stated Occupations of Susu Clients 34 2.4. Clients’ Stated Reasons for Using Susu Collection 36 2.5 Clients’ Reasons for Using Susu Collection by Gender 38 2.6 Client Challenges in Using Susu Collection 40 2.7. Susu Clients’ Relationships with Formal Banking 41 4.1 Summary Statistics of Susu Savings Schedules 109 4.2 Susu Savings Schedules and OLS Baseline Estimates 115 4.3 Susu Schedule Impacts on Savings Contributions (One-to-one 117 Matching without Replacement Using All Other Savings Schedules for Comparison Groups) 4.4 Susu Schedule Impacts on Savings Contributions (Kernel 119 Matching Using a Normal Density Function Using All Other Savings Schedules for Comparison Groups) 4.5 Susu Schedule Impacts on Savings Contributions (One-to-one 121 Matching without Replacement Using Individual Savings Schedules as Comparison Groups) 4.6 Susu Schedule Impacts on Savings Contributions (Kernel 123 Matching Using a Normal Density Function Using Individual Savings Schedules for Comparison Groups)

Table

4.7 Summary Statistics of Susu Savings Rates by Savings Schedules 125 4.8 Savings Schedules and OLS Estimates (Susu Savings Rates) 127 4.9 Susu Schedule Impacts on Savings Rates (One-to-one 129 Matching without Replacement Using All Other Savings Schedules for Comparison Groups) 4.10 Susu Schedule Impacts on Savings Contributions (One-to-one 131 Matching without Replacement Using Individual Savings Schedules as Comparison Groups) 4.11 Susu Schedule Impacts on Savings Rates (One-to-one 133 Matching without Replacement Using Individual Savings Schedules as Comparison Groups) 4.12 Susu Schedule Impacts on Savings Contributions (Kernel 135 Matching Using a Normal Density Function Using Individual Savings Schedules for Comparison Groups) 5.1 Savings Schedules and Traits: Summary Statistics 165 5.2 Savings Schedules and Characteristics: OLS Baseline Estimates 169 5.3 Savings Schedules and Characteristics: Tobit Baseline Estimates 170 5.4 Savings Schedules and Credit: Average Treatment Effects: 175 One-to-One Matching without Replacement (All Schedules As Comparison Group)

Table

5.5 Savings Schedules and Credit: Average Treatment Effects: 177 Kernel Matching Using Normal Density Function (All Schedules As Comparison Group)

5.6 Savings Schedules and Credit: Average Treatment Effects: 179 One-to-One Matching without Replacement (All Schedules As Comparison Group)

5.7 Savings Schedules and Credit: Average Treatment Effects: 181 Kernel Matching Using Normal Density Function (All Schedules As Comparison Group)

5.8 Missed Savings and Credit: Average Treatment Effects 183 Both One-to-One and Kernel Matching With Normal Density Function) 5.9 Robustness Checks: Effects of Credit on Savings Mobilization: 185 Average Treatment Effects (Both One-to-One and Kernel Matching)

xii

Table

6.1A Summary Statistics for Male and Female 204 Collectors 6.1B Summary Statistics for Male and Female 206 Collectors’ Clients 6.2A OLS Baseline Estimates: Savings and Credit Effects 209 Female Susu Collectors 6.2B One-to-One and Kernel Propensity Score Matching 212 Savings and Credit Effects of Female Collectors 6.3A Summary Statistics of Susu Collector-Susu Client 214 Gender-Matched Networks 6.3B Summary Statistics of Outcome Variables for 215 Susu Collector-Susu Client Gender-Matched Networks 6.4 OLS Estimates of Gender Networks: Savings and Credit 217 Outcomes 6.5 Gender Matched Networks’ Impacts on Savings and Credit 219 (One-to-one Matching without Replacement) 6.6 Gender Matched Networks’ Impacts on Savings and Credit 221 (Kernel Matching with Normal Density Function)

6.7 The Impact of Highly Educated Female Susu Collectors on 225 Savings and Credit Outcomes 6.8 The Impact of Educated Gender Matched Networks’ Impacts 227 on Savings and Credit : One-to-One and Kernel Matching

xiii

6.9 OLS Female Collector Impacts on Savings and Credit 230 (By Savings Schedules) 6.10 Female Collector-Female Client Match Network Impacts on 232 6.11 OLS Female Collector-Male Client Match Network Impacts on 234 Savings and Credit (By Savings Schedules) 6.12 OLS Male Collector-Female Client Match Network Impacts on 235 Savings and Credit (By Savings Schedules) 6.13 OLS Male Collector-Male Client Match Network Impacts on 236 Savings and Credit (By Savings Schedules) 7.1.1 Logit Regression Tables Determinants of Susu Savings 250 Schedules (Savings)

7.1.2 Logit Regression Tables Determinants of Susu Savings 300 Schedules (Credit)

7.1.1 Logit Regression Tables Determinants of Susu Savings 318 Schedules (Gender)

xiv

LIST OF FIGURES

Figure 1 Central Region of Ghana Map 43 5.1 Susu Savings and Creditworthiness Signaling 153 6.1 A Separating Equilibrium with the Absence of Gender Bias 199 6.2. A Separating Equilibrium with the Presence of Gender Bias 200

THE POLITICAL AND BEHAVIORAL FOUNDATIONS OF INCLUSIVE

ECONOMIC INSTITUTIONS: SUSU COLLECTION IN GHANA

Kweku A. Opoku-Agyemang

Under the supervision of Professor Jeremy Foltz At the University of Wisconsin-Madison

Dominant in developing countries, informal financial institutions represent weaknesses for political states and limited economic agency among the productive poor. Policy makers have long deliberated merging informal finance to formal systems and the state using inclusive economic institutions. This dissertation consists of six chapters on merging formal systems with informal financial institutions known as susu collection in Ghana. Chapter One introduces the dissertation. Chapter Two describes the survey methodologies, as well as the field and data environments of the dissertation.

What are the origins of state capacity given a dominant informal financial sector? Chapter Three finds that informal financial systems can be important contributors to state capacity when merged with the formal sector. Information asymmetries for state capacity are improved by harnessing susu collection as a policy mechanism in Ghana. This national policy initiative has political and behavioral fundamentals at the micro level.

Chapter Four studies institutional change in susu collection from a behavioral perspective. Susu collection is an external commitment savings device that is subjected to personal and behavioral schedules at the individual level. These rules vary in terms of required internal commitment and hence, savings outcomes. Chapter Five studies how the savings schedules of credit-constrained

xvi

individuals signal creditworthiness to susu deposit collectors in the formal sector using a signaling model. The system appears to work: credit responses accurately target the productive poor who exert more effort in mobilizing savings.

Gender discrimination against women is a common political phenomenon in merged formal and informal susu institutions. Chapter Six uses signaling and network models to show that enfranchising women can reduce gender-bias in economic outcomes. The incidence of political bias is significantly lessened by the individual behaviors that are a focus of this dissertation.

Chapter Seven concludes the study.

Chapter 1

Introduction

Relationships between institutions and human progress are fundamental issues spanning the entire spectrum of social science. According to Douglass North, these entities are the “set of rules, compliance procedures and moral and ethical behavioral norms designed to constrain the behavior of individuals in the interests of maximizing the wealth or utility of principals.”1. An emerging literature argues for the importance of institutions in comparative development, relative to other factors such as geography, culture, and even policy-making capabilities. This case motivates the need for understanding institutional evolution as well as its consequences2.

The evolving literature in political economics has argued that economic institutions which are inclusive are those able to secure and support markets, yielding economic opportunity for the citizenry. The key to understanding this process, is analyzing factors that uphold organized markets, and extend market-related beneﬁts to the disenfranchised. Inclusive economic institutions are often associated in the literature with inclu- 1See pp. 201-202 of Structure and Change in Economic History (1981). Other leading theoretical studies on the New Institutionalism are North (1990, 1997). The relevant literature is discussed further in Chapter Three. 2The leading statistical institutional studies in this area are Acemoglu Johnson and Robinson (2001, 2002), with a review in Acemoglu and Robinson (2012). For the factors institutionalists contend with, consider, among many others, Fisher (1989), Diamond (1997), Sachs (2006), Williamson (1990), Aghion and Howitt (2009). 2 sive political institutions: systems that allow pluralism and broad participation. This outcome is intertwined with the foundation of state capacity: the ability of a political state to mobilize revenue and underpin contracts. This link persists in the literature because of a positive relationship between state capabilities and economic development.

Otherwise equivalent institutions that concentrate enfranchisement in a minority are extractive in character. Inclusive economic institutions are considered rare (relative to extractive entities) in this literature for political and other behavioral reasons.

Existing notions of inclusiveness, however, have a conceptual tension with an overlapping but signiﬁcant body of literature on formal vis-à-vis informal institutions.

Scholars have emphasized informal institutions as various tangible or intangible systems or patterns of behavior that conceptually depart from ofﬁcial, authorized or formal institutions. Illustrations span much of social science. For example, institutions representing informal social control are sustained through customs, norms and mores (such as gender discrimination)–which are not necessarily a characteristic of formal social controls (for example, policy initiatives empowering women). In a democracy, informal norms of unregulated private presidential control (such as patrimonialism) may represent state dominance that diverges from the recognized constitution. On a related note, the informal economic system is often beyond the control of formal institutions.

Several bodies of empirical research across the social science disciplines seem to support what has nearly become conventional wisdom among many economists, political scientists and sociologists alike: informal institutions are, to some extent, conceptually seperate from formal systems even though they may interact in practice3.

Yet, there is also an important observation in the literature of merging formal and informal arrangements and institutions, particularly within the economies of developing countries. For example, African policy makers have studied mechanisms of improving economic inclusiveness by merging formal and informal arrangements at least since

3For a survey on the complex relationships between formal and informal political and other institutions, see Helmke and Levitsky (2003). 3 the explosion of informal arrangments in the 1980s4. However, as a country where the majority of National Poverty Reduction Strategies have relied on merging formal and informal arrangements, Ghana is perhaps one of the more extreme examples of this phenomenon. Several tiers of ﬁscal and legal institutions have emerged and coexist to support this harmonization as has a centralized state policy institution. Although increasing attention is being paid to informal economic arrangements, the occurrence of merged formal and informal ﬁnancial systems remain understudied5. This relative inat- tention is with particular respect to institutional change within merged formal-informal institutions. The situation persists inspite of possible additions to our knowledge on economic inclusion and its consequences.

The central contribution of this dissertation is to develop theories that reconcile existing notions of inclusive entities with informal institutions, focusing on the Ghana- ian ﬁnancial institution known as susu collection. According to Aryeetey and Udry

(1995: 38): “A substantial basis of theoretical knowledge exists, but new theory can be inspired by the outcome of research into local institutions. For example, we know of no compelling theoretical model which provides a coherent explanation of the institutional arrangements surrounding susu collectors in Ghana.” There is still scarce

“coherent explanation of the institutional arrangements” that deﬁne and are deﬁned by the Ghanaian susu institution. Yet there is a wealth of theory on inclusive institutions

(only brieﬂy outlined above) that may assist such research. Information economics and game theoretical models can be useful for studying inclusive economic institutions out of the informal ﬁnancial sector as has been the case for the previously inclusive literature (that focuses mainly on the formal economy).

There has been signiﬁcant interest (since that point in time) in studying informal

ﬁnancial arrangements and their co-existence with formal ﬁnance in several literatures

(across the vast majority of the developing world). However, relatively less progress

4Similar iniatives have been attempted in Asia and Latin America (See Chapter Four). Note Tripp (2001) for related discussions. 5See Chapter Four for further discussion. 4 has been made in outlining theories to explain how institutional change affects the functioning of such ﬁnancial mechanisms. Given that institutional evolution may occur following the merging of formal and informal ﬁnancial institutions, such research attention is potentially important. The goal of this study is to highlight such institutional dynamism. I develop and test frameworks to explain why, in Ghana, susu collection currently functions as an inclusive institution, and provide some consequences of its institutional evolution. I also provide policy implications of what I learn on susu collection in Ghana for the future integration of formal and informal institutions.

In the dissertation I propose different theories for analyzing the occurrence of merged formal and informal ﬁnancial arrangements. I then use these frameworks to provide some initial answers to the phenomenon.

I focus on four basic and general foundations that provide some parameters for the study:

1. I focus on information ﬂows: I view improved information ﬂows as having im-

plications for economic, political, and behavioral outcomes that relate to state

capacity. I assume that inclusive economic institutions beneﬁt from reduced in-

formation asymmetries. These motivations have consequences for state capacity.

2. I emphasize the fundamental importance of commitment as a behavioral foun-

dation of inclusive economic institutions. Different agents have various levels

of internal commitment, and these translate into varying outcomes of inclusive

institutions when such systems are contextualized as commitment devices.

3. Economic effort plays a central role in mitigating micro-level problems of infor-

mation assymmetry. When economic effort is duely acknowledged within inclu-

sive institutions, the targeting of inclusiveness to the productive poor is relatively

efﬁcient. 5

4. Political foundations play a signiﬁcant role in the economic and behavioral as-

pects of inclusive economic institutions by empowering women. Economic in-

stitutions are more inclusive when the beneﬁts of merged formal and informal

institutions also reduce gender-bias in economic outcomes.

I introduce the main ﬁeld and data environments of the study in Chapter Two. I conducted ﬁeld surveys and interviews to contextually support the research.

What are the origins of state capacity given a dominant informal financial economy? In Chapter Three, I present a game-theoretic model founded on a national case study of Ghanaian finance and state capacity. I motivate inclusive economic institutions from the perspective of a democratic and developmental state. The Microfinance and

Small Loans Center (MASLOC) was instituted by the Ghanaian Government in 2006 to provides credit to poor but productive entrepreneurs under two broad modules. One track is an independent, direct module. In this component, the government attempts to identify low-risk entrepreneurs with its own ofﬁcers. Alternatively, an indirect module may be used. Under this system, the credit decision is taken by susu deposit collectors

(working with rural community banks). Understanding the relative credit repayment rates for these modules have signiﬁcant implications for state capacity. I show that the direct module has had little success in identifying low-risk entrepreneurs. On the other hand, the indirect track has had a relatively consistent record of targeting productive but poor entrepreneurs. Information asymmetries are important for the results. In particular, low-risk entrepreneurs signal community banks’ deposit collectors with savings contributions to distinguish themselves from their high-risk counterparts.

There are several key implications of this analysis. Although economic comparisons between the state and the private sector are foundational to social science in general, this chapter tries to extend the discussion to informal financial institutions and their emerging role in economic development. Although the existence of inclusive eco- 6 nomic institutions are necessary for extending official markets to the disenfranchised, they are not sufficient for that outcome. While this may be attributable to different conditions, I focus on information asymmetries in this study following different studies. Unlike other state capacity literatures that focus on the information asymmetries of voters, I argue that state-level information asymmetries may be mitigated by the policy use of otherwise informal financial institutions. The durability of this sub-study will depend on behavioral, economic and political foundations pursued at the micro level.

For this cause, Chapters Four, Five and Six are devoted to assessing such empirical evidence.

The ﬁrst important task which I take up in Chapter Four, is to revisit the susu savings collection mechanism and its implications for merging formal and informal

ﬁnancial arrangements. The goal of Chapter Four is to isolate behavioral foundations of commitment savings in this context. From the perspective of the recent behavioral economics literature, external commitment devices such as susu savings are related to a lack of self-control on the part of individual agents, or, inabilities to follow through on personal saving initiatives. Individuals have long been found to generally yield to present or short-term impulses at the expense of long-term goals according to the literature (e.g. Strotz (1955), Phelps and Pollak (1968)). In the recent literatures across economics, psychology and sociology that have followed, an individual responds to inadequate self-control in one of two ways. An individual may facilitate personal rules of commitment using their limited internal willpower, or depend on an external device such as susu.

A natural question relates to the impulse (internal or external) which is relatively signiﬁcant in the maximization problem of an individual agent with inadequate self- control. Instead of resolving the above dilemma in the literature, Chapter Four focuses on merging internal and external mechanisms, by studying external commitments subject to personalized rules of saving. The literature has noted that evolution in informal 7 economic institutions is a very gradual process (see Ashraf, Karlan and Yin (2006a)).

Fortunately, I observe quasi-random variation in susu savings schedules (the frequency at which a client chooses to meet their susu collector). This phenomenon represents institutional change in the susu system. Using the quasi-random variation matching clients to their susu collectors, I ﬁnd that clients that chose to meet their collector on a daily basis made the most savings both in the aggregate, and in per capita terms.

Other savings schedules that were on relatively relaxed timetables had negative impacts on savings mobilization (relative to the daily schedule). Therefore, the daily schedule represents the highest level of internal commitment, when external commitment is constrained by personal savings rules.

To be sure, I am unable to compare internal to external commitments within the above context. However, given some possibility that our economic decisions may be amalgamations of internal and external incentives, this new evidence suggests the usefulness of a focus on merged internal and external commitments. The discussion also has implications for the creation and maintenance of social mores and behavioral norms within formal-informal economic hybrids. The fact that different non-daily savings schedules arose out of the formal-informal ﬁnancial market implies that the typical daily schedule was not entirely satisfactory. However, since the daily schedule yields the best outcomes, further institutional evolution may need to occur for the susu institution to be more inclusive. In this vein, the inclusion of community bank credit to the susu savings model may be a mechanism that renders the susu institution more inclusive with respect to the relatively disenfranchised.

Further micro analyses are motivated by the subject of Chapter Two: the credit link between state capacity, and susu collection. In Chapters Five, and Six, I turn to analyses that integrate susu savings and credit. Chapter Five studies how the savings contributions of credit-constrained individuals reveal creditworthiness to deposit collectors in the formal sector. The major goal of this analysis is to assess the extent to which “sig- 8 naling” (Spence (1975, 1976)) and targeting the poor using social community-based metrics such as economic effort lead to relatively satisfactory outcomes (e.g. Alatas,

Banerjee, Hanna, Olken and Tobias 2012). To what extent do credit responses (to economic effort and internal commitment) target the productive poor who exert more effort in mobilizing savings? Entrepreneurs who contributed savings on a daily schedule received signicantly more credit than others who deposited savings on relatively relaxed schedules. This seems a natural consequence of the previous discussion on commitments.

The relationship between commitments (from Chapter Two), and the current discussion on signaling and effort are intuitive: institutions which are relatively inclusive may be those that appeal to agents’ efforts to participate in the beneﬁts of supported markets. Simultaneously, the results also have implications for the literature arguing that inclusive institutions may become extractive. The daily susu savings schedule that requires the most effort and signals the highest degree of creditworthiness is also the most traditional. On the other hand, the innovative and newer susu savings institutions yield negative credit outcomes (relative to the daily schedule). These results may re-

flect decreased benefits of the initiative merging formal and informal institutions. The main difference between this finding and the literature on inclusive institutions is that this particular finding appears to have a behavioral motivation.

Yet, negative political foundations (in particular, gender-bias against women) are unfortunately common in both formal and informal institutions. In Africa, there is relatively low female representation across formal institutions, be they economic, political or social. Yet, the literatures are unambiguous on the role of the informal ﬁnancial economy in perpetuating gender-bias against women. On the one hand, the informal economic sector is overwhelmingly female in Ghana. On the other, credit outcomes are often biased against women often irrespective of inherent economic ability. This conundrum has similarly uncertain implications for advocating the merging of formal 9 and informal institutions. In Ghana, susu collection has been an entirely male endeavor that has been biased (in terms of credit) against women client entrepreneurs. Similar to the above observation, this is inspite of the fact that women are the majority of clients in the susu system. Women enfranchisement has been applied to microﬁnance in non- governmental institutions in the literature. Yet, the impacts of empowering women on gender-bias remain undestudied within this context.

The susu institution represents an interesting case also because, as we have already argued, credit outcomes relate to savings contributions. To this, I add the observation that credit outcomes may be gender-biased independently or as a result of biases in savings contributions. What are the implications of hiring women susu collectors on gender-bias in economic outcomes? In Chapter Six, I turn to an analysis of women empowerment and gender-bias in susu savings mobilization and credit provision. If susu collectors and clients share the same gender, a relatively better understanding of clients’ circumstances may improve economic outcomes. On the other, same-gender biases may adversely affect the efficiency of the entire institution of merged formal and informal financial arrangements. The study uses a simple social network framework based on undirected flows of funds (savings and credit).

Using quasi-random variation in matching collectors with clients, I show that female collectors do not replicate the gender-bias historically attributed to their male counterparts. Yet the simple network analysis shows that female clients are less likely to contribute savings to collectors of the same gender, while male collectors show a higher propensity to save with a female collector. Social distance (educational attainment) and contributing savings on different schedules help explain these results.

Networks consisting of susu collectors and clients of similar educational attainment mimic the gender-biases of their counterparts in terms of savings and credit outcomes.

On the other hand, the observed economic effort and internal commitment (the foci of

Chapters Four and Five) do not appear to have a signiﬁcant relationship with gender 10 or gender-bias. Although the merging of formal and informal ﬁnancial arrangements is important, I argue that more attention must be paid to the gender dynamics that are both a cause and consequence of inclusive economic institutions. The results imply that the social control of gender discrimination may be internalized in susu collection, and politically empowering women (by raising their status in the susu institution) may be one way of mitigating the impacts of gender bias. Chapter Seven closes the study.

This study argues for some revision to our understanding of economic institutions, as well as their foundations and outcomes. Recent work has substantially expanded our theories of the links between state capacity and inclusive institutions. Yet, they often take the political and behavioral norms that underline informal economic institutions, as given, although these rules may have implications for the political economics of development. Analyzing the conditions under which ofﬁcial markets can be extended to the disenfranchised in informal ﬁnance can complement our existing understanding of behavioral, political and economic institutions. I argue that merged economic institutions out of the informal sector are also important for state capacity and economic inclusiveness at large. At the same time, institutional operation may contradict ex-ante intentions. These consequences draw on overlapping political and behavioral foundations that improve or harm the goal of inclusiveness in ways that, I believe, are informative for social science research and policy. 11

Chapter 2

Field and Data Environment,

Survey Methodologies, and

Project Overview

2.1 Introduction

Linking the poor to formal financial systems via the informal savings mechanisms prevailing in developing countries has recently gained attention in policy circles. How- ever, empirical research on this policy innovation is still forthcoming. This chapter is driven by fieldwork conducted from June 2010 to September 2010 to study the evolution of informal financial markets. The field survey was implemented in twenty-seven urban, peri-urban, semi-rural and rural areas in the southern Central Region of Ghana.

Two main factors inﬂuenced the choice of Ghana as the study area. The country represents one of the few cases where banks, non-governmental organizations, and even the state have promoted linkages with indigenous deposit collectors as a devel- 12 opment strategy. At the national level, the Ghana government has focused policy on formalizing the informal ﬁnancial sector to build state capacity1. Also, a recent global initiative (funded by the World Bank, the International Fund for Agricultural Develop- ment and the African Development Bank) called the Rural Financial Services Project

(2001-2008) has focused on strengthening operational linkages between micro-ﬁnance institutions and informal savings initiatives as well as general capacity building in rural

ﬁnancial operations (World Bank, 2011). Therefore, an empirical understanding of the factors motivating the performance of these inclusive institutions has important policy implications. Second, institutional collaboration with the University of Cape Coast

Institute for Development Studies and the Kakum Rural Bank provided additional support that facilitated the data collection process, including detailed information about the ﬁeld and data environments.

The fieldwork component of the project improves on previous projects that focus on either the savings or credit side of economic transactions between two people. Fo- cusing on respondents that operate on exclusively on the savings or credit aspect would limit my ability to make inferences on how savings-credit interactions occur, for example. To mitigate this concern, the project accesses persons who make savings, as well as others who provide credit to these individuals. Specifically, the study samples susu deposit collectors, as well as a sample of susu clients, since the interactions from both sides (collectors and clients) fully define the operations of the susu institution. For example, if a deposit collector mobilized funds for a specific market area, I gained data on a sample of that particular collector’s clients. Similarly, if an itinerant client interviewed in the random sample interacted with available deposit collector, then I gained data on that collector. This survey methodology allows the collection of information from both individuals involved in two important exchanges: savings contributions from clients and loans from collectors. This approach is important partly because of poten-

1See Chapter Three for an empirical discussion of merging formal and informal ﬁnance to build state capacity. 13 tial information on the economic and other implications of these interactions. I began the project with extensive semi-structured individual interviews to help develop the parameters of the study. Iterations of these interviews (on both collectors and clients) evolved the survey questionnaire into a very context-supporting, participatory statistical instrument2.

The rest of this chapter is organized as follows. Section 2.2 gives an overview of the ﬁeld environment and setting. Section 2.3 explains the sampling methodologies used. Section 2.4 describes the susu collector and client rosters. Section 2.5 introduces descriptive statistics on collectors and clients. Section 2.6 closes the chapter.

2.2 Field Environment and Setting: Ghana’s Central

Region

Figure 1 represents the Central Region (see Chapter Appendix). The Ghana Gov- ernment carved this area out of the Western Region just before the 1970 Population

Census. It occupies an area of 9,826 square kilometres or 4.1 percent of Ghana’s land area, making it the third smallest in the nation. It shares common boundaries with

Western Region on the west, Ashanti and Eastern Regions on the north, and Greater

Accra Region on the east. Central Ghana is mainly Fante, (a subset of the nationally dominant Akan ethnic group) accounting for 2 million Ghanaians3. Ethnic diversity of the area has been entrenched since the region experienced the ﬁrst West African contact with Europeans in the 15th Century, through its hosting of the ﬁrst capital of the colonial Gold Coast, until 1877, when the capital was moved to Accra4.

2Vijayendra Rao (2002), Jean Dreze (2002), and Ravi Kanbur (2003) are the definitive literature on the broad approach defined as participatory econometrics. According to Vijayendra Rao for instance, “when respondents are directly allowed to participate in the research process, the econometrician’s work will avoid stereotypical depictions of their reality. This could result in unexpected findings that may prove to be important” (2002: 1889). 3Akan is a generic term identifying certain linguistically related peoples in the southern parts of West Africa. There are an additional 20,000 Akan-Fantes in the Baule communities of Southeastern Côte d’Ivoire, as well as a supplementary 43,000 in Liberia. 4A political and economic history of the Central Region is presented in an extensive literature on the role of its castles and forts in the trade of gold and human beings during the Trans-Atlantic Slave Trade. 14

The vast majority of the population in the Central Region inherits kinship identities matrilineally. This principle of social organization has importance for the political economics of informal ﬁnance. In matrilineal cultures, kin membership traces through the uterine line so that children belong to their mother’s kinship and not to that of their father. In this scenario, a man’s heirs are not his own children, but those of his sister5.This cultural phenomenon has inﬂuenced several social norms, including employment.

Ghanaian women have traditionally occupied a key position alongside men in the production of goods and services for the market, but as is the case in most countries, the actual contributions of women are biased downwards (or reported as missing) in national accounts for various reasons (Boserup 1970, Beneria 1981). Although the allocation of traditional entitlements and modern assets are male-biased, matrilineal kinship systems in Ghana encourage entrepreneurship since they “signify windows of opportunity for women to acquire wealth by encouraging their ownership of self- acquired property” (Bortei-Doku Aryeetey 2000: 321).

The Cape Coast Municipal District hosts the regional capital Cape Coast (population 82, 291), but it is the smallest metropolis in Ghana at 122 square kilometers. The district is bordered to the north by the Twifo-Hemang-Lower Denkyira district, east by

Abura-Asebu-Kwamankese, and west by Komenda-Edina-Eguafo-Abirem. The southern border is the Gulf of Guinea. The district is nationally famous for its quality Senior

Secondary Schools and the University of Cape Coast. The eight randomly chosen districts (out of seventeen), sampled from the Central Region are listed in the Appendix, each of which hosts at least one surveyed local market center. The commercial capital

Ethnographies emphasizing political economy of the region from a historical perspective include Thomas Edward Bowdich (1873), while Jan Hinderink and J. Sterkenburg (1975) provide a socio-economic study of Central Ghana. Detailed descriptions of the Central Region during this period are available in Joseph Casely-Hayford (1903) and Kwamina Dickson (1969). 5This system is less common than patrilineal systems (where children belong to their father’s kin group and inherit the father’s property at his death). Matrilineal systems persist in a variety of cultures in Africa, East Asia, and the Americas. David Schneider and Kathleen Gough (1961) provide details on matrilineal kinship. 15 of Central Ghana is Cape Coast, also called Oguaa (literally “market”). Similar to other communities within the region, the Queen Mother of Oguaa has meaningful political authority, representing the economic interests of businesswomen in the regional and national Houses of Chiefs6. While market trading tends to be concentrated in markets, it is not exclusive to such areas.

The Central Region was relatively prosperous after Ghana gained independence in

1957, but declined during the national political and economic crises of the 1970s and

1980s. The area is currently classiﬁed among the four poorest regions in the country

(Central Region 2011). Although Ghanaian poverty is decreasing according to the

World Bank (2011), the incidence of poverty in the Central Region remains higher than the national average, and has been worsening since the late 1990s (International

Labor Organization 2004).

The larger Central Region is also becoming an important tourist destination. The

Elmina and Cape Coast former slave castles (built by Portugese traders in 1482 and

Swedish traders in 1653 respectively) are World Heritage Sites, designated by the

United Nations Educational Scientiﬁc and Cultural Organization (UNESCO). Some economically relevant cultural attractions of the surveyed areas include annual cele- bratory festivals such as Fetu Afahye (Cape Coast), Bakatue (Elmina), Ahoba Kuma

(Abura) as well as the Pan-African Historical Theater Festival (PANAFEST). This art festival is held every two years to host people of African descent from the Americas, the Caribbean, and other locations. The Kakum National Park is the only national park offering an aerial canopy walkway at the forest canopy level in Africa (Conser- vation International, 2011). The isolated rainforest area is 33 kilometers from Cape

Coast. Low-interest loans from the Social Security National Insurance Trust (SSNIT) and the Central Region Development Commission (CEDECOM) are encouraging hotel construction and development. However, the loan-providing commercial banks focus

6Fiona Araba Gibson (2010) provides a detailed historical and contemporary case study of Queen Moth- ers and the political economy of the Akan. 16 almost exclusively on a minority of the economic actors in the region (Gartner, 2001).

Around 80% of the working population in the Central Region is employed in the private informal sector, with the exception of Cape Coast, where informal entrepreneurship represents 63% of the labor force (Central Region 2011). The region also hosts some of the ﬁnest secondary schools in the country including Wesley Girls’ High

School, Mfantsipim School, St. Augustine’s College, as well as the University of Cape

Coast, meaning that Central Ghana has a small but consistent cluster of professionals (Central Region 2011). For most workers, economic activities often derive from agricultural activities and trading in market spaces.

Owned, rented, or inherited market stalls (or kiosks, or smaller desks) often delineate individual business spaces within a market area. Market sections are typically organized by the type of commodity sold (fruits; vegetables; fish; meat; consumer goods such as cell-phones; herbal medicines; and several others). These arrangements allow traders to coordinate local prices or supply from providers who may be based beyond the boundaries of their district or region. Stalls near a market entrance may attract a significant amount of customer traffic owing to convenience, conditional on the level of expected market activity on a given time of day. Stalls of concrete have been the most attractive since they can withstand the rainy season and protect most goods, although wooden and metallic stalls are increasingly common. Recent migrant workers (in both rural and urban cases) may be more likely to sleep at their place of business to accumulate more customers earlier the next day.

It is common for market traders to manage the ﬂow of various goods into their markets, and one mechanism used is a market queen. Market queens are often wealthier, itinerant traders who serve as a link between farmer producers in an area and commer- cialization of their produce. Wives of farmers and traders may arrange with farmers to produce for them, occasionally providing credit in return for exclusive access. It is common for broader market trading to coalesce around market queens by commodity. 17

Each trader maintains their individual stall or kiosk for trading. Beyond the limits of each deﬁned market, vendors who are unable to gain a stall may trade on the street on the outskirts of an urban market area, or along a main road near the enterance to a rural area. Thus, a surveyed client may be with or without a stall in the studied rural and urban areas. While much market activity correlates with the climatic season, a large variety of (surveyed) non-agricultural occupations also exist in both rural and urban areas. These include the provision of transportation services (via tro-tros, taxis, or bicycles), health and education services, carpentry, house construction, tailoring, and hairdressing.

Cape Coast is a conduit that links every of the seven randomly chosen surveyed districts and areas, in three directions. Tro-Tros (mini-vans) maneuver either in the east- ward direction towards Accra, north towards Kumasi in the Ashanti Region, or west toward Sekondi-Takoradi in the Western Region. Tro-Tros which frequent the route towards Accra (serving Mfantseman, Mfantseman West, Ajumako-Enyan-Essiam and

Gomoa West districts), are based at the main Abura-Pedu station. This location is no- table for connecting Cape Coast to Abura-Pedu and Kakum National Park. The same station serves the surveyed districts en route to Kumasi (Assin North Municipal, and

Twifo-Hemang-Lower Denkyira). Tro-tro transportation toward the Komenda-Edina-

Eguafo-Abirem district, Sekondi-Takoradi, and the greater Western Region may be comandeered from Kotokuraba station. In all of the above cases, tro-tros depart whenever a van is ﬁlled to full capacity, or several times every hour. Alternatively, a tro-tro may depart when the total fare of the boarded passengers exceeds a threshold decided and coordinated by the tro-tro drivers’ union (the Ghana Private Road Transportation

Union). The Kotokuraba (“creek of crabs”) market is the economic heart of Cape

Coast, coordinating the bulk of commodity ﬂows of agricultural produce, wax-print and batik cloth, as well as other goods and services. Agricultural commodities emanate from northern farming communities, the coastal surroundings as well as Cape Coast it- 18 self. Two markets in Anafo and Cape Coast control the ﬁsh trade. Other surveyed satellite markets within the municipal area are the urban and peri-urban Abura-Pedu,

Amanful, Apewosika, Anaafo, Bakaano, Siwudu, and Nkanfoa.

Beyond Cape Coast Municipal area, the districts are mainly rural, in either forest or coastal areas. Rural areas within a district may appear geographically close to each other, but are politically recognized as separate and distinct areas. The Ajumako-

Enyan-Essiam district is a forest area. Agricultural produce include cassava, kenkey

(a meal made from corn dough), bush-meat and yams. These goods sold at market areas in Aburankwanta, Essiam, Enyan-Abassa, Asasan, Adjumako-Owane, Mande, and

Breman-Essiam. These market areas are located within a 40 kilometer radius of Cape

Coast. The rural district has a population of 91, 965, and covers 5% of the regional land area at 541.3 square kilometers. Assin Fosu is the district capital of Assin North

Municipal with 22,837 residents, west of the Ajumako-Enyan-Essiam district, and 100 kilometers from Cape Coast. The Twifo-Hemang-Lower Denkyira district (sharing a border with Ajumako-Enyan-Essiam) has a land area of 1,199 square kilometers, and hosts the Pra River. Surveyed areas included Jukwa (Amaase, Mfuom and Frani sub- areas), Afeaso, Twifo Mampong, and Abeka Nkwanta, all about 27 kilometers from

Cape Coast.

The capital town of Komenda-Edina-Eguafo-Abirem is Elmina (which has a population of 25,560). Elmina is 11 kilometers from Cape Coast. Surveys within the town accessed the following areas: Bantuma, the Dutch Cemetary Street area, and Boatase, all 11 kilometers from Cape Coast. Areas beyond Elmina in the survey include the

ﬁshing market areas of Akotobinsin, Bantuma, and Bronyibima. The surveys covered

Gomoa Odumase (population 149, 792) in the Gomoa West district. The Gomoa Odu- mase area is 26 kilometers from Cape Coast. The surveys accessed the following areas in Mfantseman: Abura Nkwanta, Anomabu, Egyaa, Esaman, Mankessim, Baiﬁkrom and Saltpond, all about 19 kilometers from Cape Coast. The surveys covered Jedu (39 19 kilometers from Cape Coast) in Mfantseman West.

In all of the areas cited above, individuals often save a portion of revenue with a deposit collector on a regular basis in the process known as susu collection. A client nominates a collector who collects a ﬁnancial amount every day. At the end of the month, the collector returns the total, save a day’s contribution which is kept as a commission. Loans are extremely rare in this traditional method of susu collection.

This system is popular thoughout Central (and most of) Ghana. Formal services are limited to commercial and state-motivated community banks. I discuss community banking in the ﬁeld region in the next section focusing on Kakum Rural Bank. 2.2.1 Community Banking in Central Ghana: Kakum Rural Bank

“Like a baby, we handle you with care”

–Slogan, Kakum Rural Bank.

In Ghana, rural banks7 dominate the provision of formal financial services, particularly in smaller towns and villages. This situation contrasts with non-governmental organizations (NGOs) which have offered most formal microfinancial services in most of sub-Saharan Africa (Steel and Tornyie 2010). The RCB presence in Ghana benefitted from special regulations inclined toward community-focused banking (low minimum capital requirements and limited operations). These rules were implemented by the

Bank of Ghana in 1976. RCBs in Ghana must be owned by shareholders from the local community they operate in (as limited companies under Act 179 of the Companies

Code of 1963).

Rapid initial growth in rural bank institutions was followed by decline in the 1970s following poor management and the depreciation of the capital base. Although data from the period is rare, a limited proportion closed down (Steel and Andah 2003).

7Rural bank is shorthand for “Rural and Community Bank”. The term is abbreviated RCB in the text). 20

RCBs have seen a revival following support from government and international agencies, including the Rural Financial Services Project (2001-2008), which supported the creation of the ARB Apex Bank owned by RCBs to better connect them to the formal ﬁnancial system. The current supporting system is the Microﬁnance and Small

Loans Center (or MASLOC), a government- and donor-funded institution tasked with coordinating and merging formal and informal ﬁnance in the country8.

Kakum Rural Bank (established in February 1980) was initially registered with the Bank of Ghana as Edinaman Rural Bank Limited, but the institution quickly tran- scended the Edinaman area. The name of the Kakum River was adopted by local shareholders since the body ﬂows throughout much of the Central Region of Ghana, and hence the operational area of the bank. Kakum Sika Fie (Home of Finance) is headquar- tered at Elmina, with the eight satellite agencies operating in Abura-Pedu, Kotokuraba

(in Cape Coast), Mankessim, Jukwa, Abakrampa, Moree, Mpoben, and Abrem-Agona.

RCBs such as Kakum Rural Bank offer special products for speciﬁc target groups on behalf of government and donor-funded programs, including the MASLOC, as well as

NGOs on occasion. The bank also supports social development activities in the Cen- tral Region as part of social responsibility. Activities include ﬁnancially supporting the building of schools, local libraries, roads, and scholarships for girl students. Such provisions have enhanced the credibility of the institution in the Central Region.

Like most RCBs, Kakum has introduced Susu Savings and Credit (SSC) schemes to mobilize savings by adapting the daily savings methods of susu collectors. The main improvement over private, independent deposit mobilization is an added credit incentive. This system involves regular collection of savings by collectors who may be independent collectors on commission or internally-trained staff. This deposit mobilization technology builds on the traditional culture of susu collection in Ghana. Kakum Ru- ral Bank covers all of the randomly chosen market areas and towns mentioned in the previous section.

8MASLOC is discussed at length in Chapter 3. 21 2.3 Survey Design, Sample Selection and Methodology

Fieldwork began in June 2010. A collaboration with the University of Cape Coast

Institute for Development Studies led to the creation of the survey based on my initial interviews with susu clients in Kotokuraba market and the general tro-tro station area. I conducted my first ten interviews with independent susu clients in Kotokuraba market, as well as a currently retired susu collector whose experience is widely respected in the region. Iterating the survey on several occasions allowed its length to expand into larger versions. The final instrument was implemented with susu deposit collectors employed by Kakum rural bank. Fieldwork ended in September 2010. The relatively brief three-month survey period proved adequate partly because deposit collectors typically collect data and organize sustained engagements with hundreds of customers on a regular basis. By having susu collectors conduct surveys and interviews, I was able to gather data on several financial and other outcomes. The length of the survey allowed the collection of various data related to economic and other behavior. The sample selection, size and methodologies are reported in the next sub-section.

2.4 Sample Selection and Size

The unique character of the research inﬂuenced the study design and methods. The initial phase involved my interviewing 15 Kakum deposit collectors, and administering a comprehensive survey to each of them. This instrument was similar in scope to the survey eventually administered to bank clients (details below). I ﬁrst interviewed the management to obtain an overview of institutional activity.

I also met with Kakum rural bank ofﬁcials at the Elmina bank headquarters, Kakum

Sika Fie, on August 19, 2010. After extensive discussions with the present deposit collectors, I decided the order of survey implementation should follow geographically- based routes collectors may use while mobilizing funds across market areas. I origi- nally intended to survey about 200 respondents, but the ﬁnal sample consisted of 384 22 respondents in seven districts, and twenty-seven rural, peri-urban, and urban areas.

This development occurred after a meeting I attended with rural bank ofﬁcials and susu collectors who became my enumerators9. 2.4.1 Susu Collectors as Enumerators and Sampling Methodology

Each of fifteen susu collectors (ten males and five females) visited several rural, peri-urban and/or urban areas during the survey. Each enumerator was responsible for interviewing at least 20 of their customers (on average), and administering a detailed questionnaire. My presentation to the collectors included a Question and Answer ses- sion which allowed the collectors some familiarity with the reasoning behind each question. Random sampling was stratified by geography (market area), since this factor relates to income, a potentially important variable for savings and credit outcomes, which are primary dependent variables in the study. Although the sampling is representative of the broader Central Region, it may reflect the larger Ghana only to the extent that compared rural banks have relatively similar clientele and performance levels. For instance, the sampling would not adequately reflect the Greater Accra region (where rural banks’ clientele are relatively similar to that of commercial banks on average).

The sampling and surveying methodology is mainly based on randomly chosen meetings between a susu collector and susu clients, in which the client engages in a one hour survey and interviewed by their susu collectors. The survey made it clear that there were no financial or other benefits or costs to participating in the survey and interview. Although I did not observe the implementation of every single survey, susu collectors are likely to have sufficiently internalized the need for survey administration to be random 10. Also, I designed the study to gain a representative sample within a

9This meeting consisted of a long presentation and a very active and lively discussion where I provided the motivation, feasibility and need behind every single proposed question. While not of all my questions were approved as appropriate, other questions were presented as possible alternatives in most instances. 10A concern may be that susu collectors would prefer to interview and survey clients who are creditworthy, yet creditworthy individuals account for only about 44% of the sample. Before administering every survey, a statement read by every susu collector (in Fante) stated that participation in the study was without any financial benefit. Given the nature of the survey implementation, some individuals present in the final survey were not actually Kakum Rural Bank clients, but clients of other rural bank institutions. 23 stratified survey application of market areas. A simple random sample consisted of susu clients chosen by chance. Each susu collector surveyed and interviewed around

20 clients in total.

Several clients answered interview and survey questions while operating within market enclaves in all of the sampled towns and villages. Others responded to interview and survey requests at their places of businesses located at market entrances and exits. Such clients gain relatively easier access to customers, and are usually wealthier on average as a result. Yet other surveyed clients were itinerary, conducting business along paved roads and footpaths. Some clients were not at their own places of business encountered, but interviewed in market spaces (en route to a destination for example) and surveyed if the client granted permission 11.Surveyed areas included “chop-bars”

(make-shift restaurants), drinking bars, tailor and carpentry shops, construction workers’ gathering areas, hairdressing salons, barbershops, and surrounding areas in the general vicinity of the identiﬁed markets. Similarly, some surveyed clients agreed to be interviewed while trading in small market enclaves alongside main roads, such as the Cape Coast-Accra main road, and the Elmina-Sekondi-Takoradi road.

The main merits of the design is that it yields a random sample stratified by geography (market area), and provides data on economic and other decisions made in a relevant setting. Susu savings is a well-established arrangement in Ghana, with a collector handling hundreds of clients on a daily basis in various market areas equivalent to those sampled. A potential shortcoming of the study may be that questions are being asked of susu clients by their collectors. This issue is much less of a concern than would be the case if collectors were independent (with no rural bank affiliation) because of the community-owned character of Ghanaian rural banking (that is absent in private susu collection). For example, concerns of clients being harassed by susu collectors (with potentially adverse impacts on data collection) are significantly less of

11Such interviewees include professionals such as doctors, teachers and lawyers who use susu (and agreed to be interviewed), as well as itinerant market traders. This breakdown represents a sample of the totality of clients engaging in susu collection, yielding a representative sample of individuals in Ghana’s central region. 24 a concern in this particular setting mainly because of the participatory and communual aspect of Ghanaian rural banking. Speciﬁcally, bank-operated and formalized susu collection tends to function at the behest of clients (not necessarily their collectors), which partially mitigates such concerns12.

Collaborating with Kakum Rural Bank is also useful for the study since this approach allows for survey data whose validity extends beyond the Central Region of

Ghana. The rural bank was rated as a “satisfactory” bank in 2010: the average ranking of Ghanaian rural banks in a national study (Nair and Fissha 2010). The performance rating scale of rural and community banks in Ghana ranges from “mediocre”, “satisfactory”, “fair”, and “strong”, with “satisfactory” being the performance level of Kakum

Rural Bank (in the above study, only one rural bank, Kintampo Rural Bank in the Brong

Ahafo Region, was rated as “strong”). The ﬁeld sampling methodology and interviews may therefore have implications for the generally productive poor of Ghana. 2.4.2 Interviews

Within each market area, an enumerator would interview one client at a time. In the event that a business or personal partner was present, each client would be interviewed separately. Most clients in Ghanaian market areas are women, and with female deposit collection being a relatively recent phenomenon, most deposit collectors are men (only a third of the susu collectors in the sample were female). I had expected that the interviews by female and male susu collectors would provide an important check on the accuracy of responses from clients13. However, in most cases, responses were limited to an individual’s own dealings.

Deposit collectors have daily experience in negotiating gender spaces and sensitiv-

12My pre-survey interviews showed clients to be very comfortable with answering personal questions administered by susu collectors (all of whom in the data, spend time with clients beyond their official duties). In the dissertation, Chapters Three and Four discuss changes in savings schedules which were a direct result of client assertiveness in susu collection. Chapter Six has a discussion on how the accountability of rural finance in Ghana has had ramifications on community bank hiring policies. 13Huddy, Billig, Bracciodieta, Hoeffler, Moynihan and Pugliani (1997) show that interviewer gender can be important in eliciting responses. 25 ities in their daily roles as data collectors and financial counselors, and this practice should minimize any deficiencies in the qualities of interviews. Clients mainly provided interview and survey responses in Fante. As stated previously, the accuracy of interviews were reviewed by deposit collectors of both genders, (based on their prior experience collecting data), as well as my own pilot test. Culturally speaking, age is very relevant to the expected depth of various economic discussions in the region. An unfortunate outcome of being a relatively young researcher was that my discussions

(particularly with elderly clients during interviews) were inevitably limited. Mainly for this socio-cultural reason, I believe my presence in certain interviews would have affected the accuracy of personal responses negatively, not positively.The susu collector and client questionnaires are described in the next sub-section. 2.4.2.1 Susu Collector and Client Questionnaires

I created both questionnaires to show informal ﬁnancial participation; a record of savings and credit mobilization in terms of amounts and frequencies; social interactions; education and health expenditures; remittances sent and/or received; as well as background and demographic information. Each questionnaire consisted of thirteen sections, brieﬂy summarized below.

1. Identiﬁcation Particulars which lists the area location, district, geographical and

basic information.

2. A demographic data component listing information on personal education and

occupational outcomes.

3.A Susu Savings section listing information on the quantities and frequencies of

savings over the last month; social factors; reported factors affecting susu savings

participation.

4. A section on education and health, as well as relevant ﬁnancial information on 26

expenditures over the last month.

5. An asset component listing livestock ownership and investment where relevant.

This section includes quantities of goats, sheep, cattle, pigs, chickens, and guineafowl,

as well as prices for each listing.

6. A section focusing on the respondent’s area of origin, as well as recent remit-

tances sent to, and received from relatives.

7. A component reporting recent major shocks experienced (over the last month)

and coping strategies.

8. A section on land ownership (for farmers); equipment ownership (for workers in

the ﬁshing industry); and kiosk ownership (for workers in the trading industry).

9. A Formal Savings section which lists formal ﬁnancial information, and informa-

tion detailing applications, approval, and granting of credit.

10. A component detailing mobile phone adoption, and use, as relevant to ﬁnancial

behavior.

2.5 Selected Descriptive Statistics and Project Overview

2.5.0.2 Deposit Susu Collectors

The section which follows focuses on susu collector characteristics. The traits of susu collectors are important since they yield information on their proﬁles, which allows for a comparative analysis with client proﬁles (discussed in the next section).

To address these issues, I analyze susu collectors’ demographic information, including genders, ages, marital status, education as well as their susu operations. The qualitative key for coding open-ended questions is provided in the Appendix. 27

Demographics and Main Characteristics of Susu Collectors

Table 2.1 shows demographic and economic information of susu collectors in the sample. Formalized susu collectors (employed by Kakum Rural Bank), are relatively youthful on average14 (33 years). The minimum age for a susu collector is 22 years while the maximum age is 50 years in the sample. Susu collectors must access clients on foot within market areas, and often use tro-tros across signiﬁcant distances. Due to the physical exertion required, younger susu collectors may be signiﬁcantly more able to access larger numbers of clients regularly. At the national level, susu collectors average 200-300 customers per day, according to Nair and Fisha (2010). In the sample, the average number of clients was 213 clients served (on the day of the interview). The male collectors averaged 33 years, while the female collectors averaged 30 years.

Susu collectors are typically male (10 out of 15 in the sample). In interviews, clients attributed this situation to prior gender biases in education outcomes, which respondents felt may affect perceived accounting capabilities. Another hinderance noted was the potential physical danger associated with mobilizing large quantities of funds, often in crowded spaces. This concern may be less signiﬁcant because of large social support associated with formalized susu collection. Interviewees believed that the very recent phenomenon of female susu collectors in rural banks such as Kakum could be attributed to the large gains in Ghanaian girl education, and a relative ease of trusting susu collectors of the same gender.

14Susu collectors in the informal ﬁnancial sector tend to be relatively older on average. For instance, a seminal study by Aryeetey (1994) found independent susu collectors to average 41 years. 28 29

Susu collectors have relatively high educational attainments (Senior Secondary levels) in the data, contrary to perceptions that ﬁnancial agents may pursue informal arrangements because of relatively low levels of formal education15. The average educational attainment of some Senior Secondary education is similar to the educational attainments of independent susu collectors (who had the equivalent under the previous

Secondary School system from the 1990s).

Most (69%) collectors were married, with three collectors being single, and only one collector identifying as a widow/widower. While some rural banks hire independent collectors on commission, a concern has been fraudulent behavior. Most banks

(such as Kakum) have transitioned to, and currently rely on internally-trained staff who have built up the requisite levels of trust among market women entrepreneurs in the ﬁeld. With the minimum education level in the sample being Senior Secondary

Schooling (45% of the sample), these agents are a niche above independent susu collectors, few of whom have Secondary Schooling (Aryeetey and Gockel 1991).

The speed of clearing a transaction is an indicator of how efficient a collector is, since this affects the number of clients served per period. When asked about the amount of time needed to “clear” their last recorded client, the time ranged from 3 minutes to an hour. All collectors indicated that they voluntarily socialize with clients after working hours, and 60% of the interactions involved “help with work”. The (self-reported) time spent with clients (while not on duty) averaged 37 minutes (across collectors). It is feasible that in some instances, this time spent with clients influences the time spent with clients doing official business hours, although more specific details are difficult to isolate. Kakum does not have a specific policy about associating with clients without a professional context, but it is culturally acceptable for professional and personal relationships to intertwine in southern Ghana. These relationships are important to the traditional form of susu, and it may have been necessary to incorporate (or merely

15(Informal moneylenders have shown the highest level of educational attainment in samples of informal ﬁnancial institutions, comparable to that of Ghanaian bank clerks in some instances, with independent susu collectors’ levels being slightly lower (see Aryeetey 1994:21)). 30 simulate) this in the context of formalized rural banking.

Beyond income, collectors indicated that a relationship with the larger public was important in choosing their profession, and this is viable given the large numbers of clients served regularly. The main difﬁculties reported by collectors were occasional arguments and misunderstandings with clients. For instance, if a “member” (client of a particular collector) did not honor a meeting on time, the collector’s lateness could affect him or her (and possibly their reputation) as far as meeting other clients was concerned. Collectors also complained of having to walk long distances, particularly during the rainy season. While collectors walk for relatively shorter distances, they may use public transportation to travel across towns and villages on occasion.

Most susu collectors (85%) believed that clients were aware that their services were being provided on behalf of Kakum Rural Bank, with a similar proportion (80%) estimating that “most” clients were aware. Kakum Rural Bank organizes in-house training programs to improve the efﬁciency of their collectors, and 77% believed that they were successful (every surveyed collector had attended such a program). Most clients (84% of the sample) believed that the training programs had led to a larger client base in the

ﬁeld.

These training projects are mutually enforced by updated information gleaned from experience in the field. An example relates to the frequency of meetings a collector has with a client. Traditional susu collection limits these interactions to a daily schedule, usually later in the day after most sales have been accrued. One innovation employed by Kakum Rural Bank susu collectors is to vary the regularity of collection: the bank officials and susu collectors judge whether a collector should visit a client daily, bi- weekly, weekly, or monthly. The most common initative associated with susu collectors was the flagship Enyidado Susu Scheme. Some collectors were also involved in a recent “Credit with Education” initiative (adapted from the NGO Freedom from

Hunger) to deliver ﬁnancial education services. This process involves a ﬁnancial edu- 31 cation program for clients on basic book-keeping to reduce risk associated with credit.

Surveyed collectors were primarily engaged with the Enyidado Scheme on a full-time basis. I next discuss ﬁndings on susu clients or customers in the next section. 2.5.1 Susu Clients

The present section focuses on susu client characteristics. Table 2.2 susu clients’ demographic information, including gender, ages, marital status, education as well as their susu operations. Basic Characteristics of Susu Clients

Data on clients of Kakum rural bank show that customers are mostly female (60% of the sampled 384), with an average age of 35 years (the youngest and oldest in the sample are 18 and 59 years respectively). Client age ranges are similar to those of susu collectors. Slightly more than half of the susu clients were married, with about 33% being single during the time of the study. Senior Secondary school attainment was not signiﬁcant among clients.

Clients officially join the program by expressing interest to a bank official (for instance, a susu collector in the field) or by visiting a RCB in person. From the data,

89% were aware that their collector saved their funds with a bank. While it is difﬁcult to make inferences from this phenomenon, some aspects of the relationships between clients and collectors may or may not necessarily extend to the rural bank as an institution. Only 6.25% (or 18) of the respondents indicated that they were not actually aware that their funds were saved in a bank by their collector. 32 33

Client respondents also volunteered personal employment details. Table 2.3 shows the labels clients presented when asked to name their occupations. “Market trader” was the most frequently-cited profession (accounting for 55% of the sample). Other client respondents included the following (in decreasing order of representation): farmers; tailor/seamstress; fisherman/fishmonger; teacher; hairdresser; barkeeper; doctor/nurse; baker; carpenter; driver; accountant; lawyer; caterer; repairer; house cleaner; church minister/pastor; pharmacist; shoe cobbler; local bank officer; blacksmith apprentice; construction worker; lottery agent; nursing mother/midwife; palm oil processor; film- maker; NGO sanitation agent; consultant. The diversity in employment is important since susu collection cuts across all social status levels in the studied region.

From interviews, customers are often sole-proprietors who typically use the program to save funds for regular business expenses or emergencies such as an illness.

Other cited uses included personal or child education, as well as minor investments in business (such as ﬁshing nets for ﬁshermen, farming tools for farmers and others).

These customers are a niche below salaried workers such as doctors, nurses, teachers, church ministers, and lawyers who also patronize the program (see table 2.3). The next section presents the impressions that the sample of clients had of formalized susu collection. 34 35

Attitudes toward Kakum Rural Bank’s Susu Savings and Credit Programs

When asked about speciﬁc reasons why they used susu savings, clients were generally interested in having an established and regular pattern of personal and business saving, adding up to 34% of clients. Almost of quarter of clients noted that funds may provide support against unforeseen circumstances. Having secure and convenient access to funding, via easy access to mobilized savings was almost as important to clients as having an approved loan – suggesting that while Kakum promised an improvement over private collectors by providing loans, clients were interested in increasing their savings base for outcomes transcending a credit line. A minority of clients (10%) aimed their savings at supporting their own health and education expenditures over time, (or the schooling and health costs of their children where necessary)16.

On the other hand, saving clients were mainly interested in protecting themselves against unforeseen circumstances (about 38% of the sample), as well as health and education outcomes. The convenience provided by susu collection was another important factor motivating the adoption of susu savings. The issue of banking institutions visiting clients at places of business is important because businesswomen who consistently leave their wares to visit banks (to deposit savings) may face negative reputational costs. The finding that investing savings into business in of itself is not a significant factor (even for clients motivated primarily by savings) shows that savings compete with other client commitments. 16While primary education in Ghana has reduced drastically in costs following the implementation of the FCUBE (Free Compulsory Universal Basic Education) program, textbook, uniform and related costs remain significant for many Ghanaians (Akyeampong 2009). 36 37

While access to credit was an important factor inﬂuencing the participation of clients, increased savings was also fundamentally important for clients, especially those who had saved consistently over time and qualiﬁed for a loan (see table 2.4). This also suggests that where consistent savings occur, their incentives may transcend initial credit. The savers were more strongly affected by security concerns to lessen the effects of emergencies–their highest motivation. Clients also responded to the convenience of the Kakum susu program, with 12% citing the accessibility of susu collectors as an important factor. Many market businesswomen in particular are unable to leave their stores to visit a rural bank given the competitiveness of most markets.

Although credit may not have been a noted as a major motivation for some respondents (as noted above), the majority (65%) of clients had applied for a loan in the last month. The large share of credit clients may be due in part to the high proportion interested in savings. Since the ability to receive a loan depends on consistently accumulated savings (over at least 3 months), it is plausible that the manifested proportion of credited clients is a function of the initial drive to save more regularly–that inﬂuenced adoption of the Kakum program.

Female clients were more likely to apply for credit (104 out of 184 applicants) in the last month. This may be related to the ﬁnding that female clients were more predis- posed to long-term savings (to hedge against unexpected circumstances in the future), while male clients were motivated primarily by the prospect of increased personal petty savings in the short-term. While male clients were motivated more by the prospect of personal savings, more female clients’ savings were focused on investment. More female clients (than male clients) were interested in credit access at the time of engaging in susu savings. Table 2.5 shows reasons for saving partitioned by gender. 38 39

Susu clients answered questions on the challenges posed by susu collection (see table 2.6). The main issue mentioned was “difficulties in sustaining savings contributions due to business challenges”. Economic set-backs may influence the ability of make consistent savings with a susu collector. A concern reported by only about 12% of respondents was the fear of susu collectors running off with funds. This finding is consistent with pre-survey interviews that showed this concern to be the dominant factor leading to clients disfavoring independent susu operations. Challenges in accessing funds were problems for about 7% of clients, while the lateness of susu collectors for savings appointments was another important problem.

Table 2.6 also shows information on credit outcomes. Clients must contribute susu savings for at least three months before applying for a loan. The majority of susu clients (72%) did not consider the interest rate (22%) to be a discouraging factor affecting the decision to apply for susu credit. Susu clients felt more “comfortable” with bank susu collectors, relative to independent and private collectors, with only 1.64% being unsure. Loan awareness levels were signiﬁcantly high, with only about 3% being unaware that susu collectors with Kakum rural bank were capable of providing credit.

Most susu clients (about 65%) had applied for a recent susu loan (in the last month).

The high demand for credit warrants a study of client savings and other economic behavior (shown in table 2.7).

Average income levels of clients were about 229 Ghana cedis, lower than GDP per capital ﬁgures at the national level (about 3,500 Ghana cedis). The maximum income level in the sample was $5,000 Ghana cedis. Savings contributions averaged about 6

Ghana cedis (around 3 US dollars), with a minimum of 1 Ghana cedi and a maximum of 30 Ghana cedis. Missed savings payment appointments with susu collectors may also be important in credit-gaining capabilities, and were relatively low, with only two missed payments being the average. Credit amounts were much higher than savings contributions, with loans received averaging 694 Ghana cedis. 40 41 42 2.6 Conclusion

In this chapter I motivated a study on financial behavior among clients of a rural banking program in Ghana. The process through which individuals mobilize savings in the susu collection institution and receive credit from formal channels is an important fron- tier for empirical research. Information on the supply and demand of informal finance is particularly useful to overcoming some of the challenges of inadequate data, and a survey was developed and implemented in the field. The research is an attempt to provide a rigorous case study of rural banking in Ghana.

Through fieldwork, I collect and use detailed data on economic activities focusing on clients of Kakum Rural Bank in Central Ghana. A key advantage of the dataset discussed in this chapter is the relative precision with which it defines financial network connections. It is possible for example, to distinguish between interactions focused on savings contributions from those associated with credit exchange. One weakness in the collection process was that interviews could not be observed in some instances, although these were mainly due to practical issues. Descriptive statistics provide an overview of savings mobilization and credit provision in the Central Region of Ghana. 43

Chapter 2 Appendix: Map

Map

Figure 1: Central Region, Ghana17

The following districts and areas were covered in the survey.

CAPE COAST MUNICIPAL

Cape Coast (Abura-Pedu; Amanful; Apewosika; Anaafo; Bakaano; Kotokuraba;

Kingsway; Siwudu) Nkanfoa

ASSIN NORTH MUNICIPAL DISTRICT Assin Fosu 17Credit: Wikimedia Commons. The map below does not reﬂect the creation of Mfantseman West and Gomoa West districts. 44

ADJUMAKO-ENYAN-ESSIAM DISTRICT Esiam; Enyan Abassa; Assassan; Adjumako-

Owane; Mando; Breman Essiam

TWIFO-HEMANG-LOWER DENKYIRA DISTRICT Afeaso; Jukwa Mfuom; Jukwa

Frani; Twifo Mampong; Abeka Nkwanta; Jukwa Amaase

KOMENDA-EDINA-EGUAFO-ABIREM DISTRICT Akotobinsin; Bantuma; Brony- ibima Estate; Elmina (Tertekesim/Botoase/Dutch Cemetary/Lime Street/Nana Gyan

Square/Chapel Square street areas); (Awenee; Council Lane; King of Kings; Nkon- trodo; Teterkesim; Pershie; Damabodo)

GOMOA WEST DISTRICT Gomoa Odumase

MFANTSEMAN DISTRICT Abura Nkwanta, Anomabu, Egyaa, Esaman (Eguase)

Mankessim (Old Nkusukum, New Nkusukum, Edumadzie, Estates, Sanﬁkrom,

Skuul Kesem, Twafo, Kokwado, Krofu, New Anaafo, Nkwanta);

Saltpond (Korankyekrom, Prabew, Equabaido, Estates, Nankesedo, Zongo); Baiﬁkrom

MFANTSEMAN WEST DISTRICT (Jedu)

KAKUM RURAL BANK BRANCHES

The following areas host Kakum Rural Bank branches.

Elmina (main branch), Abura-Pedu, Kotokuraba (in Cape Coast), Mankessim, Jukwa,

Abakrampa, Moree, Mpoben, and Abrem-Agona. 45 46 47 48 49

Chapter 3

Can State Capacity be built out of the Informal Economy?

Financial Reforms in Ghana

“(Governments) should also carry out some fundamental reforms. On the demand side of the equation, entrepreneurs in developing economies need to be able to signal more easily that they are creditworthy.” –Justin Lin, Walk, don’t run, The Economist (2009)

3.1 Introduction

State capacity1 (or the lack thereof) is a policy concern that has often motivated reforms in economic institutions and systems. Policy approaches to inadequate state capacity may generally vary with national levels of economic stability and development.

In developed countries for example, the state may easily ban or ignore an informal

1State capacity is deﬁned as the ability for a state to mobilize revenue, enforce contracts and support markets. 50

financial sector2 partly because of this relatively low threat to its ability to collect revenue and uphold official markets. In developing countries however, informal financial institutions have long dominated financial systems with little sign of abatement, even when formal institutions and systems are increasingly apparent. The commonality of relatively weak states in Africa gives evidence that creating economic institutions that are inclusive is not an easy task. Under what conditions may a democratic government invest in the creation of an inclusive economic institution to integrate informal finance with formal systems? What mechanisms can states rely on to build state capacity out of the informal financial sector’s creditworthy entrepreneurs?

Are state capacity benefits yielded by linking informal financial institutions to the formal sector? Policy makers in developing countries have long considered connecting the productive poor to the state through informal financial systems although empirical research is still inconclusive. I present a model to analyze the driving forces of institutional change to merge informal finance with formal banking and build state capacity.

My analysis will suggest that where the state has not developed a strong capacity, the policies regarding public finance provision may not be independent from the mechanisms of mobilizing revenue. The case study is based on the creation of an inclusive institution called the Microfinance and Small Loans Center (MASLOC) and its relationship with informal finance participants in Ghana.

The Ghana government has used MASLOC to commit hundreds of millions of dollars to otherwise informally financed women entrepreneurs since 2006. The Cen- ter was established under the Office of the President to integrate informal financial arrangements with formal banking. A central goal of the inclusive institution is to strengthen state capacity by generating government revenue. Infrastructure to mobilize tax revenue from the businesses that are typically informally financed exists. A relatively significant problem for Ghanaian governments is the scaling of small-scale

2The informal financial sector is defined as the range of financial institutions that are unregulated or unobserved by the state. 51 businesses (since more successful businesses may provide more revenue).

On the other hand, the scaling of businesses from credit assumes an ability to target creditworthy entrepreneurs that are credit constrained. To achieve this goal, the

MASLOC program operates within two modules. Speciﬁcally, MASLOC approves funds for lending both directly and indirectly. Under the direct module, state ofﬁcers must personally identify creditworthy entrepreneurs. The indirect module is a collaboration with community banks, institutions that use susu collectors. Susu collectors already mobilize savings based on principles already familiar to the entrepreneurs tar- geted by the MASLOC initiative. While the direct module has had little success in identifying low-risk entrepreneurs, the indirect track has had a relatively consistent record of targeting productive but poor entrepreneurs.

Why do states invest in building state capacity out of an informal financial sector that limits its functionality? The theoretical analysis argues that inclusive economic institutions that integrate the formal and informal financial capacities are motivated by the political and economic need to improve state capacity. Mechanisms that improve state capacity depend positively on the expected value of a common-interest finance program. By means of a finance program that represents societal interests (informal

finance participants inclusive), the state may improve its fiscal and legal capacities over time. If the expected value or demand of such a public program is expected to be high, then any political group has a large incentive to finance such common-interest spending while in power.

Yet, the existence of inclusive economic institutions may not be sufﬁcient for building state capacities. The presence of inclusive economic institutions does not guarantee successful targeting of high-performing entrepreneurs in the informal ﬁnancial sector.

Relative to low-performing ones, poor but superior entrepreneurs are more likely to repay state-sponsored loans with interest, and hence engage sustained contributions to state capacity. This outcome is associated with susu collectors–layers of intermediaries 52 with better information on entrepreneurial characteristics.

When inclusive economic institutions are provided (in my case, the state-controlled

Microﬁnance and Small Loans Center) – susu collectors may serve as devices to commit contributors to repay loans, and in fact yield Pareto improving outcomes. The marginal cost of contributions depend on entrepreneurial type in my analysis. Cred- itworthy and high-performing entrepreneurs are willing to engage in regular contributions that signal their type, distinguish them from low-performing entrepreneurs, and allow them to receive higher loans. The state obtains revenue through the intermediary mechanism of susu collection, and institutions can reasonably come to regard contribution level as a signal of entrepreneurial quality.

Qualitative evidence from Ghana supports the theoretical predictions. The Growth and Poverty Reduction Strategies (GPRS) depended heavily on bridging formal and informal finance. The Microfinance and Small Loans Center is GPRS policy instrument aimed directly at the informal financial arrangements in the country. State policy upgraded rural and community banks’ financial capacities very close to that of larger commercial banks, to enable them (with their frontline deposit collectors) to lend com- petitively. However, information constraints are problematic for state capacity. Evi- dence on the results of the program support the claim that its success (in the direct and independent sense) is limited, while its linkage with rural and community banks, (and hence susu collectors) succeeds to a much higher degree. In the theoretical predictions and analysis, intermediaries with better information are more profitable lenders to entrepreneurs since they are privy to their signaling via susu contributions.

The chapter adds to a large literature of institutions as major factors inﬂuenc- ing economic and political development outcomes, credited mainly to the work of

Douglass North3. In contrast, other theories have emphasized geography (Diamond

1997, Sachs 2006), culture (Landes 1999, La Porta, Lopez-de-Silanes, and Shleifer,

3See North and Thomas (1973), North (1982), North and Weingast (1989), North, Wallis and Weingast (1989). See also Olson (1984). These works proved inﬂuential in the studies of political economics such as Acemoglu, Johnson and Robinson (2001). 53

2008), and well-informed policy-making (Williamson 1990, Perkins, Radelet, and Lin- dauer (2006); Aghion and Howitt (2009); Besley, Montalvo and Reynal-Querol 2011a,

2011b).

The present chapter follows a recent discussion on state-supported inclusive economic institutions, albeit with a focus on informal economic arrangements. I apply a useful dichotomy from Acemoglu and Johnson (2005) and Acemoglu and Robinson

(2012) to the present discussion on merging formal and informal institutions and state capacity. Inclusive economic institutions secure and support markets, providing economic opportunity. These systems contrast with extractive institutions that concentrate economic benefit to relatively few in society. Economic institutions that are exclusive tend to be more representative than inclusive entities for political reasons, specifically, the fear of elites of losing power4. For this reason, inclusive economic institutions are often associated with inclusive political institutions that are more representative. On the other hand, extractive institutions and the impact of overall state weaknesses have been relatively common on the African continent (Herbst 2000, Bates 2001). Inclusive institutions are theoretically associated with state capacity because of a positive relationship between state capacity and economic and political development (e.g. Besley and Persson 2009, 2010, 2011) that relies on fiscal extraction motivated by conflict

(Tilly 1975, 1993). Although such military and economic power inﬂuenced the his- toric rise of states in Europe (Mann 1993), another important contribution has observed that the success of state intervention in Brazilian, Korean and Indian industrialization requires an understanding of developmental states’ limits (Evans 1995).

Although I have a similar focus on “developmental states”5 my concern is their relationship with informal economic systems and public ﬁnance institutions. In China,

4See Besley and Coate (1998); Bourguignon and Verdier (2000); Acemoglu and Robinson 2000a, 2006b for illustrations) 5The term “Developmental state” refers to “the seamless web of political, bureaucratic, and moneyed inﬂuences that structures economic life...” (Woo-Cummings 1999:1), with an extensive literature focused mainly on Asia in a wide literature; see for e.g. Johnson (1982, 1999) for important treatments of Japan, and See Wade (2003) for the role of government in East Asian Industrialization. The concept has been applied to emerging economies in Botswana (e.g. Leftwich 1994; Acemoglu 2001). 54

Tsai (2004), link limits in state capacity to inadequate formal credit and expansive informal ﬁnancial arrangements. On the other hand, Mukherjee and Zhang (2007) connect differences in Chinese and Indian rural economies to institutional differences such as political and credit institutions. Our discussion on state-supported inclusive economic institutions explains a mechanism by which economies may transition to become more formal, focusing on rural and community banking structures.

Local banking schemes have been important in inﬂuencing the development tra- jectories of many developed countries. Banks become important (relative to securities markets) at lower levels of economic activity (e.g. See Lin (2009) for a survey on the heavy reliance of China, Japan, South Korea, and the United States on local community banks in development; Demirguc-Kunt, Feyen and Levine (2011). Rajan and Zingales

(2003) propose an interest group theory that shows that negative reactions to ﬁnancial development by political incumbents throughout the twentieth century were motivated by a fear of competition. Other reasons in the literature for ﬁnancial development are inadequate social capital (Guiso, Sapienza and Zingales (2004), or the absence of political and legal institutions such as the Common Law (La Porta, Lopez-de-Silanes and

Shleifer 1997, 1998).

The state’s ability to fruitfully expand access to credit and savings facilities has attracted a large literature (see Burgess and Pande (2005) for a survey). Individuals may exit poverty through production and employment channels supported by the state

(e.g., Aghion and Bolton (1997), Banerjee and Newman (1993), Banerjee 2004). The belief that public policy may better poverty and development outcomes by increasing

ﬁnancial access led to many state-led rural credit programs aided by the government ownership of banks (see Burgess and Pande (2005)). Some researchers have shown how elite capture can limit credit access to minorities (Adams et al (1984), Braverman and Guasch (1986), Sapienza (2004)).

A primary pathway through which a state-controlled institution could adversely af- 55 fect state capacity is clientelism6. Although clientelism effects have been notoriously dominant in African politics7, the consensus is that clientelism effects are ﬁltered and moderated to some extent when interest group politics contain elements of organized competition Wantchekon (2003:400), and voting is evaluative (Lindberg and Morrison

2008). As organized competition increases, material support contingent on political support is less feasible. In my study, the potency of clientelism would have been enhanced if it were the case that unmerited support (rather than merited support and investments in state capacity) was at stake.

Clientelism effects could have played out in two ways in this scenario of state- sponsored loans. First, borrowers (from the political opposition) would have been almost immune to any prosecution on defaulting state loans. In this case, the reasoning would be that prosecuting loan defaulters has adverse political effects for a currently incumbent political party. However, simply facing the risk of losing elections, (even if that outcome did not actually materialize), would have imposed direct costs on an incumbent government. Studies focused on clientelism and electoral change in Africa and elsewhere have documented the hesitation to continue development projects started by previous regimes since the credit for any future progress would be contestable. Thus, if the MASLOC was motivated by clientelism and not standard interest group politics, the negative consequences of running an “opposition’s program” would outweigh any beneﬁts to state capacity. The institution’s contributions to state capacity would almost surely not survive a democratic government transition.

Empirical policy evidence on whether state-led credit expansion can improve the ability of African states to support markets remains limited. The central reason for this may be the dominance of informal ﬁnancial arrangements over ﬁnancial environments.

There has long been consternation among governments and international agencies on how to approach these phenomena in rural lending (note Adams and Meyer 1992).

6Clientelism refers to discriminatory material favors offered by politicians in return for voting support. 7For thorough discussions, see Lemarchand (1972) Bates (1982), Bayart (1989), Bratton and van de Walle (1994), Robinson and Verdier (2002). 56

Arnott and Stiglitz (1988) noted that non-market institutions which develop to solve market failures may be harmful.

Other policy responses to the issue have been the attempted merging of informal and formal ﬁnancial arrangements in Malaysia (Wells, 1983) the Dominican Re- public (Ladman 1985), Sri Lanka (Sanderatne 1984), and the Phillippines (Esguerra

1989, Floro and Ray 1997), using informal merchant-lenders. These creditors tend to be wealthy individuals who would on-lend credit to poorer neighbors. While these projects have not been empirically evaluated, Adams and Meyer (1992) argue that a major theoretical weakness of these initiatives has been an absence of savings deposit mobilization, a vital ﬁnancial function which is excluded from merchant-lending activities.

The nature of these merchant-lending schemes may not permit effective loan targeting at priority groups and activities. The central reason for this is the disability to consistently identify entrepreneurs within priority groups who are able to repay loans.

Speciﬁcally, by only loaning funds, and not collecting deposits, informal merchant- lenders typically focus on or converge to wealthier individuals, whereas state-led banking programs tend to target poorer people. This makes the sustained matching of poorer

(but productive) entrepreneurs with state loans problematic. In the absence of observed deposit levels, merchant lenders have an incentive to screen out such “poorest-of- the-poor” productive people (for their lending operations to persist). In this chapter,

I consider how such inclusive economic institutions can reach such productive entrepreneurs. The susu collectors in the program, by both mobilizing savings deposits and providing loans, provide a credible source of ﬁnancial intermediation8.

The recent challenges facing microﬁnance institutions are also relevant to the chapter (e.g. Armendáriz and Morduch 2005). The beneﬁts of microcredit are now being questioned in an intense debate following rising interest rates, and most recently fol-

8For an account on the near-elimination of moneylenders following the emergence of susu collectors and other mechanisms in the Ghanaian post-reforms informal financial economy, see Steel and Andah (2003). 57 lowing reports of debtor client suicides in Andhra Pradesh, India (Arunachalam 2011) and new findings showing that the impacts of microfinance institutions on business outcomes may be less significant than initially thought (see Banerjee, Duflo, Glennester,

Kinnan 2010 for a discussion). The Union government has since developed a new draft microfinance legislation in collaboration with the Reserve Bank of India aimed at providing more regulation to the microfinance industry (Srinivasan 2011), although the informal financial sector in India has not received similar attention (see Eeckhout and

Munshi 2010 for a discussion of informal ﬁnance in India).

This chapter, by studying how inclusive economic institutions affect state capacity, is a step towards understanding the impact of state institutions on informal economic development. This analysis contributes a uniﬁed model that motivates inclusive economic institutions and explains why high-performing entrepreneurs signal information to susu collectors and rural banks. Furthermore, the results suggest a link between state capacity and the availability of such information. If the optimal provision of public goods needs to rely on state administrations with improved information, such provision will be observed if inclusive economic institutions cooperate with intermediaries with better information. Information ﬂows are important for credit repayments

(with interest) in the short run, and taxation in the longer-run, given abilities to extract revenue.

I proceed as follows. In Section 3.2, I provide briefly summarize state capacity, informal finance and the motivation for the Microfinance and Small Loans Center in

Ghana. In Section 3.3, I give a theoretical motivation of inclusive economic institutions. In Section 3.4, I confront the theory with qualitative data. Sections 3.5 and 3.6 focus on adverse selection as a problem, and signaling as a possible solution to building state capacity. Section 3.7 concludes with policy implications of the study. 58 3.2 State Capacity, Politics, and Susu in Ghana

3.2.1 Introduction

This section brieﬂy explains an evolving relationship between state capacity, politics, and informal ﬁnance in Ghana. The present case study and discussions motivate the theoretical models in Sections 3 and 5.

Ghana (former Gold Coast) gained independence in 1957. The ﬁrst two interest blocs in the country’s political history consisted of rural peasants or urban workers on the one hand and educated, business, and indigenous elites on the other. The former became known as Nkrumahists after (eventual Prime Minister Kwame Nkrumah), with the latter generally identifying as followers of Joseph Boakye Danquah, Nkrumah’s political rival (Austin 1964).

Military regimes from 1966 through the late 1970s replicated the above patterns of competition with an emerging ethnic dimension (see Oquaye 1995, Morrison 1990,

2004) before the Provisional National Defense Council (PNDC) led by Jerry Rawl- ings gained power in 19829. The country was in an economic crisis for much of that decade. The reforms that followed the ﬁnancial crisis led to an expansion of informal

ﬁnancial markets relative to large-scale ﬁnancial institutions, contrary to expectations

(e.g. Aryeetey (1994) Steel, Aryeetey, Hettige and Nissake (1997)).

The National Democratic Congress (NDC) renewed and won multi-party elections in 1992 and 1996 before losing to the New Patriotic Party (NPP) in 2000 (with 42.6% and 57.4% of the Presidential votes respectively). The NPP won elections in 2004, before losing to the NDC in 2008. Each election can been seen as an improvement on the previous one as turn out has risen and the margins of defeat have increasingly narrowed. Most analysts considered the 2004 elections to be even more transparent and open than the previous one partly because of high voter turn-out (80%). The NPP

9For a discussion on the negotiations of political interests in Ghana (in both civilian and military regimes), see Lindberg and Morrison (2008) 59 retained power with 52% of the votes cast, and without much controversy, 128 seats in parliament (as opposed to NDC receiving 45% of the Presidential votes and 94 seats.)

In the 2008 elections, a new NPP ﬂag-bearer, Nana Akuffo-Addo lost at the closest margin since the return to democracy in 1992: 50.23% against 49.77%. In the years both before and since democratic elections, economic growth and the ﬁscal position of the Ghana government have generally been on the rise, with brief exceptions limited to election periods in 1992 and 200010.

Limiting political campaigns to the nascent formal sector has become politically unrealistic. Although data is scarce, the informal ﬁnancial market is considered a “vote bank” for Ghanaian politicians (Joshi and Ayee 2002: 1), which has afforded market women entrepreneurs much attention and even inﬂuence in Ghanaian politics and policy 11.Yet, with the country being equally divided between political parties, policy actions to improve state capacity are not strictly political, but also economic.

The two dimensions of state capacity this chapter focuses on are legal capacity and fiscal capacity. Several aspects of fiscal and legal capacity have been instituted for small-scale entrepreneurs in Ghana. I first discuss the origins and structures of licensing frameworks in the country. I then outline the various institutions of informal

finance as well as legal capacity: the regulatory frameworks that coordinate their activities. The next subsection discusses fiscal capacity, primarily innovative policy that allows the government to tax the participants of informal financial arrangements. I then discuss the most important policy instrument aimed at integrating informal and formal

ﬁnance, MASLOC. 3.2.1.1 Origins and Structure of Licensing Frameworks in Ghana

In colonial Ghana (the erstwhile Gold Coast), the ﬁrst branch of the Bank of British

West Africa was set up in 1896, followed by a Barclays Bank branch in 1917. Both

10Ghanaian elections are discussed at length in Prichard and Bentum (2009). 11For the political roles of market women during the Nkrumah regime (1957-1966), note Gore (1978). Also see Lyon (2003) for more recent examples. 60 institutions were run as satellites of their head branches in London. The ﬁrst local state contemporary was the Gold Coast Cooperative Bank, established in 1945 under the De- partment of Cooperatives. Their registration was cancelled in 1961 and its operations absorbed into the Ghana Commercial Bank (Republic of Ghana 1970).

In 1953 the Bank of the Gold Coast was established as the ﬁrst indigenous commercial bank with central bank functions. These capabilities would be separated between the (central) Bank of Ghana and the Ghana Commercial Bank at independence. The banking system was subsequently regulated under the Bank of Ghana Act of 1963

(updated in 1992) and the Banking Act of 1970 (updated in 1989). A new Ghana Co- operative Bank was registered in 1970 with the Department of Cooperatives and started operations in 1974.

The remainder of this section discusses the various financial institutions of the formal and informal financial sectors, with attention paid to their interdependence, and the regulatory frameworks that guide their activities. 3.2.1.2 Legal and Fiscal Capacities: Microfinance Regulation and Formal-Informal

Financial Markets in Ghana

I briefly outline the strengths and weaknesses of legal and fiscal capacities in Ghana- ian finance. Legislation governing Ghanaian microfinance have evolved with the Ghana- ian financial market. Regulation has opened up possibilities for new types of institutions while tightening up to restrain excessive entry and weak performance given the

Central Bank’s inadequate supervision capacity (Steel and Andah 2003). The result is several tiers of different types of institutions with a strong savings orientation.

Small unit Rural and Community Banks (RCBs) are supervised by the Banking

Supervision Department of the Bank of Ghana and accomodated in the Banking Act of

1970. RCBs are not permitted to undertake foreign exchange operations. Part of their policy appeal has been that their clientele and shareholders are drawn from their local catchment area (they are legally required to organize as community-based limited lia- 61 bility companies). While restrictions include limiting individual ownership to 30% and that of groups to 50%, there is no explicity exclusion of foreigners. RCBs have enjoyed lower minimum capital requirements than commercial banks, who tend to target urban middle-income and generally high net worth clients with their superior assets12.

The informal financial sector represents much of the weaknesses in Ghanaian fiscal capacity. Informal finance in Ghana is dominated by a variety of savings-based methodologies called susu, operating at both the individual and group levels. Group- based susu operations, or rotating savings and credit associations (ROSCAs), are savings clubs united by occupation, family, ethnicity, or some common interest. Members

(mostly women) contribute an amount into a “pot” whose total is received by a different member at every meeting. Since the fund rotates, every member can expect to receive the total in the future (note Ardener 1962; Geertz 1964; and Low 1995 for a survey). The ﬁnancial hardships in the 1980s led to a heightened need for cash in hand, affecting the savings mechanism.

This urgency inﬂuenced the popular substitution of the ROSCA mechanism with an individual model, often involving a susu collector (note Bortei-Doku and Aryeetey

1995). These savings devices are the most visible and extensive form of informal savings in Ghana today (Steel and Andah 2003). A client invites a susu collector to save an agreed-upon amount on her behalf on a daily basis. At the end of the month, the susu client receives the total minus a day’s deposit (which the collector keeps as a commission). The national Ghana Co-Operative Susu Collectors Association (GCSCA) has served as the coordinating body for susu collectors since 199413.

The ﬂexible, multi-tiered system has fostered linkages across segments. First, licensed RCBs use informal agents to mobilize and lend funds. Secondly, NGOs use

12Assets of RCBs were less than 4% of commercial banks in 2003 (see Steel and Andah 2003). 13Even though they mobilize savings, the Bank of Ghana has refrained from regulating susu collectors, instead allowing them to improve the quality of their industry through self-regulation. In 2003, the GCSCA collected an average of $15 per month from about 200,000 clients (GCSCA 2003). Individual moneylenders had been required to be licensed by the Ghana Police Service under the 1957 Moneylenders Ordinance, but have become less relevant following the popularization of susu collectors. 62

RCBs to handle the funds for their microﬁnance programs. Finally, susu collectors and clubs use RCBs to deposit funds14. These individuals are either previously-independent collectors who serve as employees, or rural banks’ own internally-trained staff.

Supervision of a large number of rural microﬁnance institutions (or RMFIs) may be costly relative to their potential impact on the ﬁnancial assets (Steel and Andah 2003).

The Bank of Ghana has adopted a number of strategies to cope with its limited supervision capacity: raising reserve requirement for RCBs to as high as 62%; drastically raising the minimum capital requirement for NBFIs; and permitting self-regulation of credit unions by their apex body. The Association of Rural Banks (ARB) Apex Bank serves the RBCs, linking them more effectively to the commercial banking system, and take the lead in building their capacity (Nair and Fissha 2010).

Ghana has sustained a wide range of institutions with the potential for substantial access as far as the productive poor is concerned. Over time however, governments have also permitted the easy entry of institutions with weak management. This outcome demonstrates the challenge of balancing innovation on the one hand with adequate supervision. The Bank of Ghana has exercised considerable regulatory moderation in allowing weak units time to comply with stricter regulations. Such signals allowed the number of rural banks ranked “satisfactory” to improve by more than 60% between

1999 and 2001 (see Steel and Andah (2003) for a discussion).

Still, the system has failed to achieve impressive outreach, especially to the rural poor, motivating further policy initiatives to formalize informal ﬁnance. Although the high reserve requirements imposed on Rural Banks (62% on primary and secondary) has sometimes been blamed for the small amount of credit they provide, they have actually (in general) maintained more than the required amounts of reserves (e.g. 64-

14Similar linkages with susu savings have been attempted by larger commercial banks including Barclays Bank and Ghana Commercial Bank. The Barclays Bank program was cancelled in 2010 after a successful pilot in Accra in 2006, and expansion to the national level. Cited causes for its discontinuation have been differences in economic cultures between the susu collectors and clientele, and Barclays. The Ghana Com- mercial Bank faced similar hurdles and is being phased out for similar reasons, and both institutions are expected to continue their original focus on middle-to-high earning clients (Interviews of bank ofﬁcials in possession of author). 63

65% between 2000 and 2001 against the required 62; see Steel and Andah (2003) for more examples).

In any case, the situation has also prevented them from making substantial contribution to the availability of credit in rural areas. Since 2002, steps have been taken to lower reserve requirements of RCBs according to their loan recovery performance– striking a compromise between encouraging more active lending and maintaining performance incentives.

More recently, additional inclusive economic institutions have been created that up- grade the credit capacities of rural banks. Such policies are important not only because they may yield state capacity via credit repayments in the short-term, but also tax revenue to the state in the long-term. I discuss the existing infrastructure for mobilizing revenue from informally-ﬁnanced businesses in the next section. 3.2.1.3 Fiscal Capacities: Informal Taxation and State Capacity

The Ghana government has shown an increasing desire to tax expand its ﬁscal extraction into the informal sector. The main method of informal taxation are presumptive.

These pre-determined taxes are paid based on easily identifiable business characteristics such as the location and size. The absence of effective book-keeping prevents relatively effective income estimates. These taxes should be supplemented with traditional income returns but in practice they generally function as final taxes due to significant formal sector shortcomings (Prichard and Bentum 2009).

Ghana ﬁrst introduced a lump-sum presumptive taxation in 1963 varying by industry. Due to low compliance, however, implementation had already declined in the

1970s before the ﬁnancial crises in the early 1980s. In 1987, the government introduced a system of Identiﬁable (Occupational) Groups taxation, extending across the country through the transport sector. It operated on the same principle of collecting pre-determined payments but decentralized collection to industry associations. This system was initially more successful, as it was negotiated with small business associ- 64 ations prior to implementation. For instance, taxes were collected daily, then weekly.

Payments were made relatively affordable and unlike the previous case, were only payable when drivers were actually working. The changes signiﬁcantly improved tax revenues from the sector and the unions retain a 2.5% share of revenue collected.

There are two main constraints to informal taxation in most developing countries: logistical issues, and political barriers (since the informal economy forms an important aspect of political support in single party states). However, the experience in Ghana, where the Ghana Private Road Transport Union (GPRTU) has been collecting taxes from informal workers on behalf of governments since 1987, suggests that, in certain democratic contexts, these problems can be mitigated. The Ghana Government has enjoyed increasing reach into the informal ﬁnancial sector over time.

Gaining and retaining support of the GPRTU was critical for the PNDC government after Ghana returned to democracy in 1992 (Joshi and Ayee 2003). The government allowed the GPRTU to dominate other unions via control of all tro-tro stations and stops in the country. In return, the GPRTU gave the political support (sometimes in opposition to the Trades Union Congress (or TUC)); provided transport for political rallies and elections; and disseminated government propaganda. Unfortunately, these initiatives may have been ineffective over time; the PNDC government lost elections in

2000.

The change of government (to the NPP) at the 2000 elections reduced the hold of the GPRTU in the transport industry, but the tax collection arrangements have been maintained with one main exception. The new government attempted to replace the identifiable groups system with a Tax Stamp system. This called for all informal sector operators to purchase a visible Stamp, that would be displayed on the premises of any business. The price of a Stamp would depend on the sector and the observable size of the firm. While this initiative has brought some improvement to revenue yields, but the overall collection rates not changed significantly since the last change in government 65 in 2008. The Tax Stamp system currently coexists with the GPRTU system.

Contrary to views that political interference is associated with public inefficiency, success of the GPRTU tax mobilization programs were linked to politics (Prichard and Bentum 2009). The arrangement may be of further policy use if a significant proportion of businesses are able to scale via credit access. The next section outlines theoretical motivations for state building, motivating the creation of inclusive economic institutions tasked with merging informally-financed business with the formal sector.

3.3 Theoretical Preliminaries for State Capacity

A practical starting point for a discussion on gaining state capacity out of informal

financial arrangements may be to suppose that when the majority of citizens of a country prefer institutions that are inclusive. This implies that inclusive institutions (ones that merge informal arrangements with formal systems and formalize informal finance) are a valence issue. Yet, measuring state capacity and inclusiveness holistically may be difficult since they transcend a measurable indicator. However, I will argue that a common-interest institution such as MASLOC (discussed in detail in the next section) can be motivated from existing models of state capacity, adjusted to focus on informal

ﬁnancial sectors (that represent losses to state capacity).

One way in which political institutions could affect the creation of inclusive economic institutions is through democratic politics. But that does not mean that elites are guaranteed to choose policies that favor the majority. Even when relatively free-and- fair elections are given, elites face strong incentives to keep the status-quo of extractive institutions that limit prosperity to the minority. However, without a dominant political majority, certain political and economic conditions may incentivize politicians to expand access to economic development.

A related important factor may be a dominantly informal financial and economic 66 sector. For an incumbent government in a two-party system, a largely informal system transcends the political opposition when there exists no clear political majority. The nature of informality itself implies a difficulty in imposing order (for the sake of policy favoritism). Therefore, incumbents have an incentive to expand access to public finance, and show little political discrimination in raising revenue to do so. Competitive politics also give incumbents incentives to create inclusive economic institutions that may facilitate this process of expanding official markets to the disenfranchised, and building further state capacity. The framework is based on Besley and Persson (2008,

2009, 2010, 2011a, 2011b) with a main difference: instead of an exogenous shock of conﬂict, I study state capacity given endogenous shocks of economic instability, which are a function of public ﬁnance, and state capacity. For a developmental state that is democratic, the creation of common-interest programs or inclusive economic institutions is expected to raise state capacity. The full conceptual framework is in Appendix

Pulling this discussion together, there are good theoretical grounds for thinking that Ghana is a good case to test the implications of this theory. As discussed earlier,

Ghanaian democratic elections are increasingly competitive with each political party represented in about half of the electorate. Also, the Ghanaian state has shown an increasing desire (and some limited ability) to extend fiscal extraction (taxation) into the informal financial economy to gain state capacity (Prichard and Bentum 2009). The economic shocks of financial crises in the 1980s and 1990s led to structural reforms and a growing informal financial sector, although the motivations and implications of formalizing informal finance remains understudied. However, there are reasons to be wary of arguing that the positive relationship between democracies with dominant informal finance and inclusive economic institutions is unambiguous. I therefore present a case study. Table 1 shows a corresponding timeline with further details in the Ap- pendix. 67 3.4 The Ghana Poverty Reduction Strategies and MASLOC

as Microeconomic Policy

This section outlines microfinance policy in Ghana very briefly. I first outline the Ghana government’s Growth and Poverty Reduction Strategies or GPRS (formerly Ghana

Poverty Reduction Strategies) which have increasingly relied on microfinance as a major tool to increase access to formal financial services with the support of international development and private institutions. I then discuss the mainstreaming of informal financial arrangements into the formal financial sector, culminating in the establishment of the Microfinance and Small Loans Center.

3.4.1 The Poverty Reduction Strategies and the Evolution of MASLOC in Ghana

The imbalance between formal and informal ﬁnancial and other activity in Ghana15 has been at the core of the Ghana Poverty Reduction Strategies (GPRS I, 2003-2005) and the Growth and Poverty Strategy (GPRS II, 2006-2009) (Asiedu-Mante 2005, Na- tional Development Planning Commission 2005, African Development Bank 2010).

Following the ﬁnancial crisis and reforms of the 1980s, the Government has attempted a number of special credit schemes which have had very low recovery rates and reached very few people. A partial exception has been the Enhancing Opportunities for Women in Development (ENOWID) initiative, which made over 3,500 small loans with a recovery rate of 96% using funds from the Programme of Action to Mitigate the Social

Costs of Adjustment (PAMSCAD) (note Quainoo 1997). This is the only program administered by the National Board for Small-Scale Industries to reach a 70% recovery

15By 2006, Ghana’s workforce was estimated at about 9 million (Government of Ghana 2010) with only 2 million operating in formal employment. 68 rate or as many as 200 clients16.

The Government has also tried enabling microcredit through poverty alleviation programs and the District Assembly Common Funds (Banful 2008). Before the ﬁrst

Poverty Reduction Strategy started in 2003, a temporary Emergency Social Relief

Project (ESRP) was created in 2001 to provide business loans to the productive poor at a 20% rate of interest over 2002-2004. Disbursements would be made through rural banks and NGOs, who would evaluate the beneﬁciaries. While an early goal was to provide funds to rural banks for on-lending to clients, a challenge was a perception of state funds as “political loans”, with negative implications for repayment. While the

ESRP was intended to be a temporary measure and an initial foray into incorporating informal financial arrangements into the formal sector, its perceived potential within the informal financial sector led to the initiative being subsumed into the overall GPRS when it was launched in 2003. This influenced the creation of an overall Microcredit

Scheme: a platform that would communicate state seriousness by upgrading Ghanaian rural banks to credit levels normally associated with urban commercial banks.

The government established the Microcredit Scheme in 2001 to address the wide credit gaps between commercial and rural and community banks. This was coordinated by various Ministries, Departments, and Agencies on the basis of their respective pro-poor programs. This Microcredit Scheme was labelled the Microcredit Credit Pro- gramme, the Microcredit and Small Loans Scheme (MASLOS), before the Microcredit and Small Loans Centre (MASLOC) was proposed to government.

By a Cabinet decision taken at its meeting on May 25, 2005, government approval was given for the creation of a seperate body to streamline and administer the Govern- ment’s various microcredit programs previously scattered across the Ministries, De- partments and Agencies. Since these institutions generally lacked the necessary logis-

16The ENOWID program was eventually spun-off into an NGO, ENOWID Foundation, but with an arrears rate of 35%. Other programs were the Developing Cottage Enterprise Project (1989), the NBSSI (National Board of Small-Scale Industries) Revolving Fund Scheme (1992), NBSSI/DED Credit Scheme (1993), and the NBSSI/NFEED-Dev. Assistance project (1994). 69 tical capacity to appraise projects and operate lending, the Ghana government eventually launched MASLOC on September 6, 2006 under the ofﬁce of the President to minimize bureaucracy, provide intermediation and attempt to harmonize the informal

ﬁnancial economy with the nascent formal sector.

3.4.2 The Microﬁnance and Small Loans Center (MASLOC) and the goals of microeconomic policy in Ghana

“Government has therefore prioritised micro-ﬁnance interventions under

its GPRS (Ghana Poverty Reduction Strategy) I and II, hence the launch

today of the $50 million Micro-credit and Small Loans Fund which is

unprecedented in the country’s history. This is only the seed money. Gov-

ernment will continue to provide the resources to expand the scheme.”

–President John Kufour, quoted in Atafori (2006)17

Under the ofﬁce of the President, MASLOC is mandated to hold in trust Government of Ghana and/or Development Partners’ funds for the sole purpose of administering micro and small-scale programmes. The Center was charged with managing and regulating the approved funds. Although the NPP lost power in 2008, the functioning of

MASLOC has been consistent under President John Atta-Mills’ presidency. Adminis- tratively, MASLOC is a microfinance apex body responsible for: (i) The co-ordination and faciliation of the activities of institutions and organizations in the micro-finance subsector of the economy; (ii) Promoting and enhancing the development of a decentralized microfinancial system; and (iii) The Co-operation, collaboration and comple- mentaries with other non-bank finance institutions in the operations of microfinance

17“$50m boost for average Ghanaian”, The Statesman, September 7, 2006. 70 services.18

In the short term, the goal of the Center is to support the Government’s program of a sustainable reduction in poverty by providing credit to the productive poor of the population. The long-term objective of MASLOC is to promote the emergence, development and growth of a sustainable and decentralized and formalized microﬁnancial sector, with grassroots participation in ownership, management, and control. The Cen- ter is legally empowered to undertake reforms and development measures consistent with strengthening microﬁnance operations as viable strategies for poverty reduction.

3.4.2.1 Current Target Beneﬁciaries of MASLOC Facilities

The economic activities that qualify to be funded by MASLOC fall under the following sub-sectors which are target beneﬁciaries of the Center (MASLOC 2010).

FOODCROPS: Production of root-crops (e.g. cassava, yams), cereals (e.g. maize, rice, millet, sorghum), legumes (e.g. pepper, garden eggs, okro, tomatoes), etc. Agri- cultural crops with long gestation period are not supported under the scheme (e.g. co- coa).

AGRO-PROCESSING: Extraction of palm-oil, palm kernel oil, groundnut oil, copra oil, gari processing, and fruit drink production.

POULTRY: Production of broilers, layers, turkeys, guinea fowls, ducks, ostriches, among others.

LIVESTOCK/PIGLETS: Production of pigs, goats, sheep; fattening of young bulls/steers, among others, for sale.

MICROENTERPRISE: Petty-trading including retail of provisions, foodstuffs/staples, fruits, vegetables, selling of secondhand clothing, household utensils, stationery etc.

AGRO-MARKETING: Marketing of foodstuffs such as maize, yams, tomatoes, local

18Microfinance and Small Loans Center (2010) “Profile of MASLOC” Official Document, In possession of author. 71 rice, cold-storage (selling of frozen meat and fish), and livestock. Alternative livelihood

: Bee-Keeping, mushroom cultivation, snails, grass cutter and rabbit rearing.

FISHMONGERING: Smoking and selling of ﬁsh, cold storage.

FISHING: Offshore and inland ﬁshing.

AQUACULTURE: Construction of ﬁsh pond, and ﬁsh farming.

VOCATIONS: Vocational enterprise such as dressmaking, hairdressing, batik, tie and dye production, carpentry, beads production.

HANDICRAFTS: Carving, basketry, drums production.

AGRICULTURAL MACHINERY: Farm implements, farm machinery (e.g. Tractor, power tiller, bullock ploughs), Delivery vans for carting foodstuffs and raw materials)

IMPLEMENTSANDTOOLS: Agro-processing.

FARMINPUTS: Fertilizers, herbicides, fungicides, seeds.

The funds are to be approved in the two modules outlined in the next sub-section. 3.4.3 Direct and Indirect Modules of MASLOC

MASLOC directly approves funds through two modules: (1) directly through its internal appraisal, disbursement and recovery process, and (2) indirectly through several poverty reduction programs, including rural and community banks. In these subsec- tions, I outline both the direct and indirect mechanisms of MASLOC. 3.4.3.1 Direct Module: MASLOC Module 1

MASLOC Loans are disbursed on either an individual small loan scheme or group based model. In both cases, the interest rate is 20%. In the case of the small loan scheme, an individual may access a loan between GHC 1,000 and GHC 10,000 (between $662 and $6,624), on providing a guarantor who would redeem the loan in the event of default. The group based model involves lending to savings and credit associations consisting of 5 to 25 members. Any member within the group may access any amount between GHC 100 and GHC 500, with the entire group being held responsible for the repayment of the loan (i.e. the loan is not considered “repaid” until every mem- 72 ber has settled all outstanding debts. All MASLOC loans must be repaid with interest within a year. Accessing a MASLOC Loan Facility and Criteria under the Direct Module

A prospective client can obtain, ﬁll out, and submit a form at any District Ofﬁce in the country. Successful applications are invited for a brief interview by MASLOC

Program Ofﬁcers at the District Ofﬁce on the activity or project motivating the loan application. Advice is given on guidelines and best practices.

Any Ghanaian between ages 18 and 65 years (who is mentally sound and not in- capacitated) qualiﬁes to apply. Priority is given to women, youth, and people with physical and other disabilities. Of interest among these are the productive poor who have no access to any form of credit to start or expand their business vocations-who may otherwise be susceptible to informal loans with high interest rates. Both individual and groups must open bank accounts with a MASLOC-participating bank as part of the process before the loan can be accessed, and on approval, the loan is disbursed through the bank. However, MASLOC does not require customers to provide collateral security, but relies mainly on the provisions of persons of good standing to guarantee for the repayment of loans. Under this module, MASLOC appraises applicants independently of the banks which only disburse the funds to successful applicants. 3.4.3.2 Indirect Module: MASLOC Module 2

The indirect module is through the provision of ﬁnancial support indirectly through several poverty reduction programs, including the main Community-Based Rural De- velopment Project. In this module, MASLOC disburses funds to rural and community banks through the central ARB Apex Bank, for on-lending to entrepreneurs and other customers, using susu collection. In this module, it is understood that the coordination and faciliation of the loan duties are the sole responsibility of the rural bank.

The disaggregated mechanism of MASLOC fund approval and disbursement represents a policy compromise which is indicative of intense political debates within 73 government on the most efﬁcient way to intervene in the often controversial informal

financial sector. MASLOC is considered a “revolving fund” by government officials meaning that although it has been reliant on state and international aid funds at its in- ception, it is hoped that it can be self-generating in the future. MASLOC may raise state capacity levels in Ghana via repayment of state-sponsored loans with interest in the short run, and the provision of a larger tax base in the long term. The next section provides theoretical justification for inclusive economic institutions, while the subsequent section studies how they may be effective.

3.5 Inclusiveness versus Adverse Selection: Confronting

the Theory with Qualitative Data

The existence of inclusive economic institutions may not be sufficient to assume their ability to contribute to state capacity. In this section, we show that when the state has worse information than the informal financial sector, difficulties may arise in raising revenue to contribute to state capacity.

Speciﬁcally, we investigate the targeting of creditworthy individuals by the MASLOC program. As noted in the previous section, these loans contribute to state revenue and capacity by supporting markets (integrating informal with formal ﬁnance). The repayment of loans builds state capacity through raising state revenues. This factor is important because of implications for future tax revenue (discussed in Section 3.6.1).

Since the goal is for the Center to be a revolving fund, consistent repayments should help the Government expand access to the service.

According to President John Kufour19,

“Let me state very clearly that a loan from the fund must not be viewed as

a gift from government; neither must it be seen as a compensation due to

19quoted in Atafori (2009). 74

any community for disparities in urban and rural economic development.

Government expects prompt and full repayment so that others can enjoy

the service.”

A former Minister of Finance and Economic Planning noted that the success of the

MASLOC would largely depend on the ability of the its Board to address poverty:20

"Your strategy should therefore be able to identify who the poor are, where

they live, and which aspects of activities needs ﬁnancial support. The

Board must also initiate medium and long term targets for poverty reduc-

tion using access to ﬁnance".

Therefore, the repayment of MASLOC loans do have signiﬁcant implications for state capacity in the Ghanaian case. However, qualitative evidence from the direct module of MASLOC shows that the existence of inclusive economic institutions (as motivated by the above theory) may be a necessary but not sufﬁcient condition for state capacity to improve.

According to initial media reports, MASLOC was forced to seek the assistance of the Serious Fraud Ofﬁce (Ghana’s anti-graft agency) to help the institution recover more than GHC 80 million owed it by defaulting beneﬁciaries, with the legal unit of

MASLOC pressing charges against certain defaulters21.

In the Volta Region for instance, only GHC 308,005.29 out of GHC 2,389,991 disbursed under MASLOC was recovered by July 29, 2010, with the difference outstanding since 2008. Only 11 beneficiaries (out of 112) had fully repaid loans, with an official calling the scheme a “serious crisis”, saying that the “integrity of the party must be realized to the benefit of the nation.” Similarly, only GHC 75,115 was col-

20quoted in Ghana News Agency (2009) 21Ghana Business News (2010) 75 lected out of GHC 250,056 by the Eastern Regional Ofﬁce of MASLOC. According to one government ofﬁcial “when the letters were served on them, they quickly went undercover to pay their moneys whilst others paid after the court had given judgement.

As another example, 40 entrepreneurs in the Woodwork Association of Ghana had not been able to repay their MASLOC loans. Out of GHC 16,000 lent, as much as GHC

15,003 has been outstanding since 2007. According to one government ofﬁcial, attempts at retrieving the funds had not been successful, although charges were being pressed22.The Northern Regional Secretariat of the MASLOC also took 22 clients to a

Tamale District court over non-payment of loans (ranging from GHC 1000 and GHC

5000) received in 2007 and 200823.

While there will inevitably be several reasons for the difficulties of inclusive economic institutions in general, the above discussion motivates our discussion on the need for better information (on entrepreneurs in the informal financial sector) as an important factor driving the efficiency of inclusive economic institutions such as MASLOC.

There may be other institutional aspects inﬂuencing efﬁciency, including corruption.

The Ministry of Information and National Orientation has protested media allegations of graft with much success24 following the prosecution of more than 200 defaulters in the Volta Region25, as well as MASLOC ﬁeld staff for allegedly misappropriating state funds26. Similarly, an NPP Member of Parliament faced prosecution before the 2008 elections (while in the NPP Administration) for allegedly facilitating MASLOC funds to entrepreneurs for political gain27.

22Ghana News Agency (2010) 23Ghanaian Chronicle (2001) 24 25These prosecutions occurred despite the fact that the Volta Region has historically supplied the incumbent National Democratic Congress with the majority of its votes (See Gyimah-Boadi (2001). 26Agbewode (2011) 27Alhassan (2009) 76

Qualitative Motivation

My main argument, that women have incentives to make susu contributions to signal their performance levels as entrepreneurs grew out of discussions with entrepreneurs and bank ofﬁcials in the Central and Greater Accra Regions of Ghana. I collected such information by conducting semi-open interviews with ten bank ofﬁcials (representing

five financial institutions), fifteen susu collectors, and 380 entrepreneurs with the aid of Kakum Rural Bank.

Bank ofﬁcials repeatedly complained about administering the direct module of the

MASLOC project. One ofﬁcial noted that “they (MASLOC Module 1) have good intentions, but that is not the way to run the program.” Some was concerned that the motivations of loans may be unclear: “They might be asking our bank to make a political loan since the applicant does not qualify (in terms of susu contributions).”

Entrepreneurs cite the susu loan aspect of rural banks’ operations as an essential factor motivating their desire to do business with them. Some businesswomen believed that their rural bank “helps to obtain loans easily” and is “credit-accessible”. In addition, some entrepreneurs noted the appeal of the susu loan program “people have a heavy interest in joining” while others believed “more people are coming to join” since

“they help us with loans”. Observed susu deposits also created a sense of fairness, according to some entrepreneurs. Some believe that “they attend to each customer equally.”

The majority (75%) of the surveyed entrepreneurs believed that the “government fully appreciates the usefulness of susu savings.” Some businesswomen noted that “the government accepted the bank’s operation system” and that concerning the rural banks,

“the government even adviced that we save with them.” Others believed that a reason why government did not entrust MASLOC entirely to the susu system was because of the “rampant attitudes of fraudsters”. A commonly recurring observation was that

“most susu collectors are unregistered and engage in fraud”. 77

Businesswomen seemed more conﬁdent in the indirect module, noting that unlike the direct module of MASLOC “susu savings could not become intertwined with politics” because “saving with the bank and susu collection have nothing to do with politics.” These statements imply that entrepreneurs may have conﬁdence in signaling as a way of showing their abilities to inclusive economic institutions.

The next section discusses signaling and adverse selection to discuss on how better information (as provided by the susu collectors employed by rural banks) improves the inclusiveness of MASLOC.

3.6 A Basic Signaling Model: Susu Collection as a Mech-

anism of Building State Capacity

This section develops a basic model of signaling (Spence 1973, 1974) that will analyze the impacts of improved information ﬂows on state capacity from the informal

ﬁnancial sector28. As noted earlier, the MASLOC initiative consists of two modules.

MASLOC Module 1 provides funding to entrepreneurs independently, or without the aid of rural banks or susu collectors. MASLOC Module 2 uses rural banks and susu collectors to provide credit, with an added beneﬁt of mobilizing susu savings.

The empirical case mimics a basic signaling framework because creditworthy individuals contribute savings to susu collectors to distinguish themselves from non- creditworthy entrepreneurs. The information ﬂows predicted by signaling creditworthiness through savings are important for their indication of high-performing entrepreneurs in MASLOC Module 2. This action allows individuals to signal their ability and build state capacity in three ways. In the short term, individuals show that they are creditworthy, or able to contribute to state revenue via credit repayments. As rural banks and

28The basic signaling framework is borrowed from the labor economics and human capital literatures. The basic set-up consists of workers (who want employment) and employers (who want a capable employee). In the basic framework, capable workers gain education to distinguish themselves from incapable workers, and signal their abilities to employers. Employers respond to the observed signal (education) by providing employment. 78 susu collectors collaborate with MASLOC, the susu savings signal yields information that is itself an addition to state capacity. In the longer-term, clients who were successfully identiﬁed as creditworthy may also contribute to state capacity via the existing systems of taxation as their businesses scale. I discuss the role of informal taxation mechanisms in Ghana, which may have implications for longer-term contributions to state capacity given the above discussion on information ﬂows via susu collection. 3.6.1 Informal Taxation and State Capacity: The Implications of MASLOC and Susu Collection

The existing infrastructure for informal taxation may help tax mobilization from successful enterprises over time. Since an entity independent of susu collection (the

GPRTU) mobilize taxes, clients in both MASLOC Modules are accessed by tax agents.

However, if sustained access to credit is only exclusive to the second module of MASLOC

(where susu collection is present), then susu clients may be make more contributions to tax revenue, since credit is a signiﬁcant barrier to scaling businesses in the informal sector29. The Ghana government is sustaining its collaboration with GPRTU workers to mobilize taxation from small-scale enterprises.

Government ofﬁcials have advocated the continued taxing of informal ﬁnancial arrangements to help the nation raise more revenue for development (e.g. Ghana News

Agency (2010)). According to ISD (2011), Mpohor Wassa East District Chief Execu- tive (DCE), Anthony Bassaw toured communities such as Ebukrom, Abrodzewuram,

Heman, Brofoyedur, Senchem, Prato, Mamponso, Enyinabrim, Essaman, Kakabo,

Sekyere Nsuta, Sekyere Krobo, Sekyere Aboaboso, Dompim, Krofofrom and Daboase in the Western Region30, encouraging entrepreneurs to engage in timely tax payments and avoid antagonizing ofﬁcers from the Ghana Private Road Transport Union:

29See Chapter Four for a discussion on the relationship between susu savings mobilization and credit outcomes. 30See ISD (2011) 79

The DCE also used the occasion to educate the people on the need to

pay their taxes, saying that their development needs could only be met

when they pay their taxes promptly without engaging in petty quarrels

with revenue ofﬁcers.

Credit repayments are therefore important for state capacity in the short run and tax revenues in the longer run because of existing infrastructure to tax the informal sector to a relatively successful degree. Therefore, the shown variation in susu savings, information ﬂows and credit repayment behavior (by MASLOC module) may have important implications for signiﬁcant increases in tax revenues over time.

This observation is important because clients of MASLOC Module 2 (who make susu savings contributions) are able to signal their creditworthiness. The ability to repay loans motivates the receipt of more credit and the relative ability to improve business outcomes. Improved enterprises may contribute more to taxation revenue over time. On the other hand, clients of MASLOC Module 1 may be less likely to receive future installments of credit for reasons explained in the previous section.

3.6.2 Summary

The paper takes a novel perspective on building state capacity in developing countries by focusing on economic institutions that are inclusive. Given dominant informal

financial economies, the political and economic motivations of developmental states may induce the creation of inclusive economic institutions. The need for state capacity may motivate inclusive economic institutions to formalize informal financial economies. The paper is based on a case study in Ghana where policy initiatives have focused on merging formal and informal financial institutions to build state capacity.

However, the existence of inclusive economic institutions may not be sufficient to assume their ability to contribute to state capacity. When the state has worse infor- 80 mation than the informal financial sector, difficulties may arise in raising revenue to contribute to state capacity. Also, information flows, which are themselves additions to state capacity are relatively low, which may have negative consequences for the ability to extract revenue via taxation over time.

On the other hand, susu savings collection provides improved information ﬂows by identifying creditworthy entrepreneurs who are better able to contribute to state capacity through the repayment of credit. The susu savings signal yields information that is itself an addition to state capacity, and over time may yield taxation revenue as businesses scale.

The existence and information available to inclusive economic institutions are significant factors influencing state capacity, given dominant informal financial sectors.

The qualitative discussions are in line with the model, where economic institutions must have improved information to be efficiently inclusive. While the political and economic justifications of inclusive economic institutions are important, this chapter also shows a first step at investigating at the mechanisms by which such institutions function.

Even though the preliminary theory is encouraging, much remains to be done before I can claim to have identified thorough empirical links in line with the theory’s predictions. There are many caveats from both the demand and supply side of inclusive economic institutions, related to the financial institutions in Ghana. For instance, important behavioral, economic, and political factors may influence the actual targeting of loans by susu collectors to their clients. The political economics of susu mobilization at the micro level may shed light on the underlying incentives driving the sustainability of inclusive economic institutions.

3.7 Concluding Comments and Policy Implications

What are the origins of state capacity given a dominant informal financial econ- 81 omy? Maintaining economic stability remains a priority for states, since this cannot be divorced from the economic and political functioning of governments. Investments in state capacity as purposeful decisions reflecting both circumstance and institutional structure. My theoretical analysis highlights the factors that shape these decisions, and points to inclusive economic institutions as improving fiscal and legal capacity. A first inspection at qualitative and quantitative data suggests that the common determinants proposed by the theory may indeed associate improved state capacity with information constraints.

I highlight inclusive economic institutions. The theory observes that the inclusiveness of economic institutions (with respect to incorporating informal financial arrangements) is a key factor shaping investments in state capacity. The Ghanaian Mi- crofinance and Small Loans Center administers funds sourced from government and international agencies to formalize informal finance and improve state capacity. The repayment of state loans by entrepreneurs is an important aspect of state capacity with potential implications for future tax revenues. My analysis shows that the first module of MASLOC, which singularly approves funds, is not as efficient as the second module, where this task is performed by rural bank susu collectors. This result is because the latter module has higher information thresholds, which are a function of the regular susu savings deposit mobilization it allows. State capacities and information thresholds may help to explain how informal finance can be incorporated into the formal sector.

Although information is an important factor, other issues (absent from the study) may be important. For example, in addition to the state, formalizing informal ﬁnance is an initiative that has been attempted by the private commercial sector in Ghana. In 2006 for example, commercial banks in Ghana introduced initiatives (similar to MASLOC’s susu collection module) that focused on using susu collectors to mobilize savings and provide credit to individuals in the informal sector. The programs could not be sustained (ending in 2010) partly because of inadequate trust and communication issues 82 between formal banking and informal susu clients. These problems were followed by drastic reductions in susu savings contributions while the rural bank programs have had no such issues (to my knowledge). This outcome may imply that the social capital

(between informal entrepreneurs and state rural banks) may be high relative to other actors. These issues are worthy of further study.

Discussions comparing the ability of the state to identify winners has mainly focused on the formal sector, although informal arrangements dominate developing countries. It may be possible to merge formal and informal institutions to yield state capacity in different contexts, although a rural banking system is empowered by susu collection in the Ghanaian case. Future research should add to this discussion on formalizing informal institutions particularly in weak and fragile states and regions. Information resources on the informal sector has been difficult to access for states (mainly in developing countries), which is problematic given their insufficient state capacity. Microfi- nance policy initiatives may also be channeled through existing local support systems.

These ﬁnancial arrangements are relatively capable of identifying the economic capabilities and limits of the productive poor. 83

Appendix A: State Capacity

Framework

3.8 A Model of State Capacity and Inclusive Economic

Institutions

In this section, I develop a formal model to study the theoretical justiﬁcation of inclusive economic institutions to the state. We are interested in studying the conditions that motivate the state investment in such institutions, and factors inﬂuencing their success.

The following model is based on Besley and Persson (2009, 2010), but with a main difference. Instead of political institutions and conflict, I focus on economic institutions motivated by a politically-active informal financial sector within a democratic context. This serves two purposes. First, it motivates inclusive economic institutions that integrate informal financial arrangements with formal finance as a factor positively affecting state capacity. Second, I isolate a mechanism by which this may happen when the inclusive economic institutions have limited information on informal finance participants’ ability to contribute to state capacity. Both of these features inform our understanding on the motivations for inclusive economic institutions. 84

3.8.1 A Basic Model

MODEL ENVIRONMENT

Let total population size be normalized to 1. Assume two groups {Is,Os}, where each group denotes half of the population in every time period s. In terms of notation, I signiﬁes the incumbent (the political group in power) while O refers to the opposition group.

I assume that there are only two periods s = 1,2, and that the world ends after the second period. I start the discussion from period s = 2.

At the start of period 2, we consider the group that holds power to be the incumbent government, I1. The opposition is the other group, O1. Power may be peacefully transferred to the opposition, and this occurs with an exogenous probability denoted by parameter γ. Whoever wins becomes the new incumbent I2, and the loser becomes the opposition, O2.

At the end of period s, the current incumbent, Is, sets a tax on the income of each

Js group member, denoted by ts . It also chooses a level of legal support for each group

Js p , and spends on general public goods, Gs. When period 1 ends, the incumbent I1 also invests in the state capacity of the next period.

In addition to tax income, the government earns natural resource rents Rs. These are stochastic and drawn from a two-point distribution {RL,RH } where Rs = RH in each period with probability ρ. No resource rents accrue directly to the private sector.

INDIVIDUAL MARKET INCOMESAND UTILITY

In period s individuals consume and produce, so that members of group Js earn a market income wJs such that market incomes are function of the extent to which the state supports ﬁnancial markets:

wJs = wpJs .

The variable pJs measures how well the state supports financial markets. The vari- 85 able w is an increasing and concave. I define the generic “financial market” as a system encompassing formal and informal financial sectors. This allows me to argue that support exclusive to the formal sector is necessarily limited or extractive by definition.

Note that this implementation of the capacity pJs is distinct from the nominal capacity of said policy.

Let legal capacity πs, be enjoyed costlessly by both groups. However, whether such capacity is actually extended by a government is a policy decision to be made, so that

J 0 ≤ p s ≤ πs. The government can introduce property right-protecting legislation of the two groups to varying degrees, depending on its legal capacity.

I now deﬁne the consumer of the individual as follows: cJs = 1 −tJs wJs = 1 −tJs wpJs

Individual utility in period s is linear:

Js Js Js u(G,c) = αsGs + c = αsGs + 1 −t w p

where Gs is the level of public goods with parameter αs reﬂecting the value of public goods. We assume that αs is distributed on [αL,αH ] with a c.d.f of H (α) and density h(α). To make the discussion relatively simple, I assume that there are no savings between periods.

STATE CAPACITIES AS CONSTRAINTSONGOVERNMENT

Policies are constrained by state capacity levels. Current levels of ﬁscal capacity τs, and legal capacity πs are inherited from the previous period. The incumbent group in period 1 chooses these levels for period 2 subject to the institutions in place. We discuss the institutions below with an eye toward the descriptions in the previous section.

More speciﬁcally, τ represents ﬁscal infrastructure such as the institutions necessary to tax income at source. These institutions would essentially reduce the share of market income (1 − τ) an individual can earn in the informal sector.

Fiscal capacity does not depreciate, but can be augmented by I1 through non- negative investments which cost F (τ2 − τ1), where F (.) is an increasing convex func- 86 tion with F (0) = F (0) = Fτ = 0. A higher τs allows the incumbent Is to charge higher J tax rates, such that t s < τs. To allow for redistribution in a simple way, we allow negative tax rates.

To be specific, π represents legal infrastructure investments such as legal systems governing financial institutions and arrangements. Like fiscal capacity, legal capacity does not depreciate, but can be augmented with non-negative investments at cost

L(π2 − π1), where L(.) is an increasing convex function where L(0) = Lπ (0) = 0. A J higher πs allows the government Is to better support private markets so that 0 ≤ p s ≤

πs. The government budget constraint in period s can be written:

(2)

  tJs wJs L(π2 − π1) − F (τ2 − τ1), if s = 1 0 ≤ ∑ − Gs + Rs − J ∈{I ,O } 2  s s s 0, if s = 2

where the first term on the Right-Hand Side refers to tax revenue, the second term signifies the cost of public goods, the third term refers to resource rents, and the final term denotes the costs of investments in fiscal and legal capacity.

THE NEEDFOR INCLUSIVE ECONOMIC INSTITUTIONS

Investments in economic institutions are motivated as responses to economic instability in our model for two reasons. First, the formal sector reforms that followed the

ﬁnancial crises in developing countries have been followed by an expansion of informal

financial markets relative to formal ones. Second, the clientele of the informal financial market are representative in developing countries and may be a possible “vote-bank” in democratic settings. Political incentives to reform the financial sector may be strong where neither political group represents a majority, and the formal financial sector only caters to a minority. Perceived economic instability may motivate financial reform 87 aimed at merging both formal and informal sectors for political reasons.31. I assume that political power can shift between political groups as a direct result of inadequate economic stability.

The opposition wins democratic elections with probability γ. I assume that this parameter is a measure of economic instability, and endogenously determined by investments in public goods that are inclusive. Stability has strictly positive effects on election outcomes for a government in the framework. Another assumption is that institutions of the electoral system make incumbent governments internalize the preferences of the opposition to some extent. Speciﬁcally, any incumbent attaches weights θ to the

1 opposition group and (1 − θ) to its own group, where θ ∈ 0, 2 . This parametrization captures, albeit in a simple way, the representativeness of institutions through checks and balances or electoral systems. When checks and balances are very strong (past a certain threshold) the incumbent behaves like a Utilitarian planner - treating both

1 groups equally - in which case θ = 2 . The scenario where an omnipotent autocrat faces no such constraints and behaves as if θ = 0 is not important for the discussion since I am only considering a democratic political situation.

SEQUENCE.

Each period has the following sequence:

1. The initial conditions are {τs,πs} and the identity of last period’s incumbent Is−1.

2. Public goods’ valued at αs and natural resource rents Rs are realized from a random distribution.

3. Public investments affect the probability 1 − γ , which signiﬁes the chances of

Group Is−1 staying in ofﬁce.

31In Ghana and much of Africa, informal finance has dominated formal systems, proving robust to financial reforms in the 1980s. (e.g. Steel, Aryeetey, Hettige and Nissanke 1997). The political appeal of agents in informal financial arrangements in Ghana is motivated in the main essay, based mainly on Joshi and Ayee (2002). Please see the main essay for details. 88

4. The new incumbent Is determines a vector of tax rates, legal support, and spend- n J J o ing on public goods: t s , p s , ,Gs . The period-1 incumbent also Js∈{Is,Os}

chooses state capacities for the next period τ2,π2.

5. Payoffs for period s are realized and consumption takes place.

1.2 EQUILIBRIUM POLICY

I begin with the policy choices at stage 4 of period s. The utility function’s linearity allows me to study these separately from the choices of state capacity for period 2.

With the assumed policy weights, the objective of incumbent Is is:

" # tIs wpIs +tOs wpOs V Is = (1 − θ)wpIs 1 −tIs +θwpOs 1 −tOs +α + z , 2 s

where we have replaced Gs via the government budget constraint (2), and where residual revenue zs, is deﬁned by:

  L(π2 − π1) − F (τ2 − τ1), if s = 1 zs = R −  0, if s = 2

This objective is maximized subject to the following constraints:

Js Gs ≥ 0 ,t ≤ τs and p ≤ πs.

That is, the level of public goods have to be positive, while ﬁscal and legal capacities is limited to the stock of existing state capacity.

TAXATIONANDSPENDINGONPUBLICGOODS

I now derive the equilibrium ﬁscal policy. Whenever αs ≥ 2(1 − θ), it is optimal

I for Is (the incumbent in period s) to tax its own group maximally, t s = τs, and use the 89 revenue to expand Gs . Because Is puts weight θ ≤ 1 − θ on the opposition group, it

O also sets t s = τs. When αs < 2(1 − θ), it becomes optimal to switch to a redistribu-

O tive policy, where the opposition is still taxed fully, t s = τs, but no public goods are provided to anybody.

Intuitively, if α is high enough, the incumbent taxes both groups at ﬁscal capacity and spends all available revenue (less investment costs if s = 1) on public goods. This is the equivalent of what Besley and Persson (2009) call a common-interest state. If an α is drawn below the critical limit, 2(1 − θ), no public goods are provided and all available revenue is transferred to the incumbent group (through a negative tax rate).

We will refer to this as a redistributive state.

The realized value of government funds in period s, which is obtained by differen-

I tiating V s with regard to zs in the two cases, is therefore given by

λs = Max[αs,2(1 − θ)].

1 Unless θ = 2 , the political equilibrium underprovides public goods relative to a

Utilitarian planner, who would provide public goods as long as their social value of αs exceeded their social cost (private value of goods) of 1.

PROPERTY RIGHTS

It is easy to see that (4) is increasing in the legal protection of each group. Thus, it becomes optimal to exploit any existing legal capacity fully and set

Os Is p = p = πs.

Intuitively, the incumbent group can only gain from improving property rights to both groups, either directly via a higher wage, or indirectly via a higher tax base. This production efﬁciency result agrees with Diamond and Mirrlees (1971) which argues that production efﬁciency is desirable, although a full Pareto Optimum is not always 90 necessarily achieved. The result does not mean that property rights are well protected everywhere, however, since this hinges on the chosen value of πs. Even though the setup is a bit different, the results on policy are similar to those in

BP 2008a. Collecting all the results, we have:

PROPOSITION 1 In all states,

Js Os p = πs for J ∈ {Is,Os}and t = τs.

In common-interest states,

Gs = τw(πs) + zs and t = τs,

While in redistributive states,

Is zs Gs = 0 and −t = τs + 2 . w(πs)

In all states, investments in ﬁscal capacity and legal capacity for an incumbent reﬂects available state capacities. In common-interest states, government spending re-

ﬂects state capacities, subject to constraints. On the other hand, there is no government spending and negative tax rates by the incumbent in redistributive states.

EQUILIBRIUM STATE CAPACITY Preliminaries

We can use the equilibrium policies in Proposition 1 to write the expected future payoff to the incumbent at stage 4 of period 1, when state capacity for period 2 is chosen: 91

I1 E V (π2,τ2) = w(π2)(1 − τ2) + E (λ2)[τ2w(π2) + E (z2)].

The expression E (λ2) = φαH + (1 − φ)(1 − γ)2 is the expected value of government funds in period 2 viewed from the perspective of period 1 and is a key magnitude that determines investment incentives. It depends on three underlying parameters. With probability φ, the value of public goods is high, αH , the future is a common-interest state, and all revenue is used to supply private goods. With probability (1 − φ), the future is a redistributive state and the incumbent captures a marginal return with probability (1 − γ), namely when it stays in power.

STATE CAPACITY CHOICES

The choice by incumbent group I1 of state capacity for period 2 maximizes I E V 1 (π2,τ2) − λ1 [L(π2 − π1) + F (τ2 − τ1)],

subject toπ2 ≥ π1 and τ2 ≥ τ1. Thus the choice of I1 trades off the period-2 expected beneﬁts against the period-1 costs of investment, given the realized value of public funds. When doing so, it takes into account the uncertainties about the future values of public goods and resource rents, as well as the prospects of government turnover.

Carrying out the maximization and using (4), we can write the ﬁrst-order (complementary slackness) conditions as

(6)

wp (π2){1 + τ2 [E (λ2) − 1]} ≤ λ1Lπ (π2 − π1)

and

(7)

w(π2)[E (λ2) − 1] ≤ λ1Fτ (τ2 − τ1),

where (6) concerns legal capacity and (7) concerns ﬁscal capacity.

Conditions (6) and (7) reproduce the central results of Besley and Persson (2009a, 92

2010). Since Lπ (0) = Fτ (0) = 0, if E (λ2) > 1, there is always positive investment in both kinds of state capacity. Moreover, in this case, ﬁscal and legal capacity are complements. Positive investments will continue as long as the probability of the common interest state is large enough or γ is low enough: sufﬁcient conditions are either

1 1 φ > 2 or γ < 2 . Although Besley and Persson consider also instances where legal and ﬁscal capacity are not complements, those cases are beyond the boundaries of our discussion. We now turn our attention to determinants of state capacity within the above framework.

DETERMINANTSOF STATE CAPACITY

When E (λ2) > 1, the left-hand side of (6) is increasing in τ2, while the left-hand side of (7) is increasing in π2. The resulting complementarity is interesting in its own right. However, it also simpliﬁes the analysis since it implies that the payoff function

(5) is supermodular. This means that we can use standard results on monotone comparative statics (see, e.g. Milgrom and Shannon (1994)). Thus, any factor that increases

(decreases) the expected value of government funds E (λ2), for any λ1, will increase (decrease) investment in both legal and ﬁscal capacity. The same is true for any factor that weakly decreases (increases) the right-hand side of the two expressions for given

E (λ2).

Using (6) and (7) together with the deﬁnition of E (λ2), we establish the following result motivating state capacity:

PROPOSITION 2: Investments in both legal and ﬁscal capacity increase with the following factors.

1. The expected value of public goods (Examples include a national common-

interest ﬁnance program. If the demand for such public goods are expected to

be high, any political group has a large incentive to invest in ﬁscal capacity to

ﬁnance future common-interest spending). 93

2. More inclusive economic institutions (where economic stability is relatively low,

signiﬁed by dominant informal ﬁnancial arrangements): More representative

economic institutions lower the value of redistribution such that more public

goods are provided in more states of the world. As the state becomes more about

common interests, the value of ﬁscal capacity increases and by complimentarity,

so does the value of legal capacity. 94 95 96

Chapter 4

Commitment Savings subject to

Personal Rules: Ghanaian Susu

Collection

4.1 Introduction

The great and exalted virtue of magnanimity undoubtedly demands much more than that degree of self-command, which the weakest of mortals is capable of exerting. Adam Smith, The Theory of Moral Sentiments (1759: 33).

Self-control (or the lack thereof) remains one of the foundational concepts in economic science. According to Adam Smith (1759: 282), “we are enabled to abstain from present pleasure or to endure present pain, in order to obtain a greater pleasure or avoid a greater pain in some future time.” At the same time, research investigations observe the limitations we experience in allowing these long-term goals to manifest, showing 97 that individuals generally yield to sentiments in the present or short-term (Strotz 1955).

In the literatures that have since followed these (and other) seminal works, the agent responds to perceived inadequate self-control in either of two basic ways. An agent may choose to coordinate personal rules for commitment using their limited internal willpower, or depend on an external device to engage and sustain commitments. These reactions are generally analogous to mammals’ biological response to acute stress (called the ﬁght or ﬂight response1). Yet this compartmentalization avoids a third dimension of commitment also necessary to understanding self-control: the interdependence of internal and external commitments.

The central motivation of this paper is to comparatively analyze personal savings rules within external commitment devices. In practice, the choices and outcomes of external commitments are inﬂuenced by personal rules that are relative. For instance, instead of saving a proportion of every monthly paycheck for winter holiday gifts,

Americans may use “Christmas Clubs” to save on weekly bases. Similarly, diets, reso- lutions to minimize alcohol intake, and other activities use commitment devices (such as timetables) that respect internal rules. These individualized rules may, in turn, affect the ability of any device to affect economic and other commitment outcomes. How agents react to commitments that have both internal and external dimensions is important to the empirical analyses of commitments as well as our theories of individual willpower and self-control.

This paper analyses informal susu savings deposit collectors, the most frequently- used informal savings mechanism in African economies2. Deposit collectors mobilize savings for individuals on a regular basis, returning them at the end of an agreed-upon period. These informal savings institutions exist all over the world (Ashraf Karlan and

1In my analogy, “ﬁght” refers to an agent engaging impulses that require self-control with his or her limited willpower. “Flight” denotes the agent escaping the personal burden of commitment, by relying on an external commitment device. 2see Bortei-Doku and Aryeetey (1995). For discussions on alternative mechanisms in informal ﬁnance such as relatives or friends, see Anderson and Baland 2002, Hoff and Sen (2006), Baland Guirkinger and Mali (2012). 98

Yin 2006b). While these agents traditionally operate independently, they are gradually being merged with rural banks, that often co-exist with microﬁnance non-governmental programs in developing countries.

The high marginal returns to capital in developing country micro-enterprises 3 have motivated several microcredit programs, although these have been found to be less beneﬁcial than donors initially believed (see Banerjee, Duﬂo, Glennerster and Kinnan

2010, Karlan and Zinman 2010, Kaboski and Townsend 2012). Comparative ﬁndings have that microcredit programs have lower adoption rates than microsavings programs, leading to calls for more focus on formal savings mechanisms (Robinson 2001, Dupas and Robinson 2010, Brune, Giné, Goldberg and Yang 2011). On the other hand, part- nerships between non-governmental microcredit organizations and informal savings commitments remain rare.

While informal savings programs are more common than their formal counterparts, institutional evolution in informal savings institutions tends to be very gradual and therefore, rarely observed in practice (e.g. Laibson 1997, Gugerty 2007). Yet, these sluggish adjustments may reveal unmet market demand for financial arrangements that are unmet by current formal or informal savings arrangements. The merging of formal and informal financial arrangements have been important for expanding financial access (Aryeetey 1994). From a policy standpoint, whether or not a commitment device is provided depends on whether clients are “sophisticated” and recognize their need for a commitment device that can help improve savings outcomes (e.g. Thaler and Benartzi

2004, Ashraf Karlan and Yin 2006a: 638, Duﬂo Kremer and Robinson 2009)).

At the same time, the agency or initiative of sophisticated agents has been limited to a single choice variable in the above studies on commitment savings. There are reasons why a menu of external commitments (that vary in terms of internal commitment) may be better suited to studies of commitment. First, participants who refuse a commitment treatment of a single choice may have simply preferred another better suited to their

3see de Mel, McKenzie and Woodruff (2008); Fafchamps, McKenzie, Quinn and Woodruff (2011). 99 tastes and personal preferences. Coding them as “untreated” may or may not reﬂect the state of an agent (particularly if they seek some different treatment that only varies in terms of their personal commitment). Second, providing a single treatment (on the assumption of agent sophistication) assumes that individual sophistication does not allow an agent to make right choices relative to other options. However, sophistication may manifest in personal rules and outcomes that reﬂect varying degrees of internal commitment.

This study examines the outcomes for an evolving commitment savings product that was modiﬁed to reﬂect and satisfy personal savings rules. By discussing sophistication in commitment savings as a continuum, I hypothesize that the impact of personal rules on savings outcomes is strictly increasing in the degree of commitment

(relative to personal rules that require less commitment). Although most experimental behavioral economists emphasize commitment devices take willpower and commitment as given, the social psychology literatures on willpower and commitment are often balanced by social and demographic factors (American Psychological Associ- ation 2011:3). Newer behavioral economics studies such as Spears (2011) argue that poverty in of itself makes economic decision-making difﬁcult, depleting cognitive control. Other psychology work by LeBoeuf, Shaﬁr, Belyavsky, and Bayuk (2010) show that social selves within social roles are important in decision-making. If this is the case, then pre-treatment factors may affect self-selection into savings schedules, and must be conditioned on, before I can claim to isolate impacts of commitments.

The data was collected with Kakum Rural Bank, a community bank in the Cen- tral Region of Ghana. With the aid of susu deposit collectors, I randomly surveyed entrepreneurial and other clients of the bank, which adopted individualized savings schedules to reﬂect their personal savings motivations. Clients chose to save funds on either a daily, twice-weekly (two times every week4), weekly, fortnightly (every

4This savings arrangement was implemented in collaboration with Microsfere Fund for People and Na- ture, a non-governmental microsavings organization. 100 two weeks) or monthly timetable. Susu collectors varied the times they visited various clients (to mobilize funds) according to the savings schedule the clients elected to utilize. I ﬁnd the impact of each savings schedule relative to all other schedules, before

ﬁnding the effect of each savings schedule relative to individual alternative schedules.

All savings amounts are normalized to daily contributions. My main prediction states that savings outcomes should increasing in the order of monthly, fortnightly, weekly, twice-weekly, and daily savings schedules.

I ﬁnd that the commitment differences between individuals who use different personal rules have implications for savings contributions and savings per capita. Clients whose internal personal rules reﬂected the most commitment (saving on a daily basis), yielded the largest savings contributions impacts relative to the other schedules.

The daily savings schedule also had a significantly positive effect on savings rates per capita. I found that the least commitment-intensive schedule (the personal rule reflecting the commitment of saving on a monthly basis) had a significant negative effect on both savings contributions and savings rates relative to other schedules. The schedules whose commitment levels are between the two extremes do not generally show significant effects on savings or savings rates. The outcomes of savings schedules (relative to individual alternative schedules) show correlations slightly closer to the prediction and similar to the initial results.

The recent theoretical literature on willpower focuses on general degrees and thresholds of willpower. For instance, signiﬁcantly different levels of “self-regulation” (Ben- abou and Tirole 2004: 1) would be expected to yield a corresponding path of outcomes.

In related work, Noor (2011) shows that the distance of an agent from the consequences of choices separates the normative choice he “should” make from tempting alternatives.

Fudenberg and Levine (2012) provide a theory that the cost function of self-control in different periods may be dependent on self-control used in the recent past since short- run selves are not entirely myopic. 101

Previous literature has studied internal and external sources of commitment indi- vidually. Personal rules, or self-imposed internal commitments, are a growing part of the behavioral microeconomics literature. The agent with little self-control exercises their limited willpower to follow self-appointed rules for him or her self. Illustrations include Schelling (1984), and Palacios-Huerta (2003). Much of our theoretical frameworks of the self-generated overcoming of impulses derive from Ainslie (1992, 2001), as well as a whole research agenda by Benabou and Tirole (2002, 2003, 2004) on the cognitive foundations of personal rules. More recently, Ozdenoren, Salant and Silver- man (2011) have motivated domain-speciﬁc time preferences for agents with limited willpower, as well as conditions inﬂuencing relaxed or tightened control of consumption over time.

On the other hand, the broader self-control problem is also associated with an extensive literature that focuses on external (instead of internal) commitments. This concern has inﬂuenced studies on quasi-hyperbolic time discounting (Laibson 1997;

O’Donoghue and Rabin 1999); cross-self conﬂict (Thaler and Shefrin 1981,1988; Bern- heim and Rangel 2004; Fudenberg and Levine 2006) as well as temptation (Gul and

Pesendorfer 2001, 2004). Within this second scenario, insufﬁcient self-control inﬂu- ences the adoption of relevant commitment devices. Unlike the internal commitments outlined previously, these instruments tend to be external relative to the individual5.

The present paper studies personal rules interacting with external commitments.

Studies have engaged related discussions similar to mine in the developed world. For instance, Ariely and Wertenbroch (2002) study an external commitment (paper deadlines) where deadlines reﬂected students’ preferences. Also, Trope and Fishbach (2000) investigate self-imposed penalties in the context of a medical test. In economic development studies, Gugerty (2007) as well as Ambec and Treich (2007) show how rotating savings and credit associations (ROSCAs) provide avenues for individual self-control.

5Examples include software obstructing internet access, as well as gym memberships. In developing countries, these include participation in microsavings metal-box programs, or rotating savings and credit associations (or ROSCAs). 102

As far as I am aware, this paper is the ﬁrst to apply this discussion to the use of deposit collectors or individualized savings commitments in general. The data allow me to consider several personal rules of savings attached to an external savings commitment device. The ﬁndings have policy reach for savings outcomes in the developing world, with the main proposal being that personal individualized rules may mimic optimal outcomes when they serve as constraints to an external savings commitment device.

Instead of forcing participants into a the choice of a single, main commitment device, clients choose from a menu of self-created savings commitments that vary strictly in terms of required internal commitment. While such frameworks do not allow an empirical distinction between internal and external commitments, I extend the discussion of external commitments constrained by relative personal rules in a developing country setting focusing on savings.

The rest of the paper proceeds in the following order. Section 4.2 describes susu deposit collection savings commitment in southern Ghana, and the quasi-experimental design subject to internal personal rules. Section 4.3 examines the empirical strategy.

Section 4.4 describes the data. Section 4.5 shows empirical results for estimating the impacts of adopting savings schedules with increasing levels of commitment. Section

4.6 closes the paper.

4.2 Susu Collectors and Savings Mobilization in Ghana

In Africa and much of the developing world, several ﬁnancial institutions, such as rotating savings and credit associations (roscas), savings and credit cooperatives, mu- tual assistance groups in addition to friends, relatives, and landlords et cetera, provide

ﬁnancial services to the poor and unbanked given the difﬁculty of accessing formal banks (e.g. Adams and Pischke 1992). In Ghana, informal susu deposit collectors make arrangements with clients (mainly entrepreneurs) to engage in daily savings, col- 103 lected at places of business or residence. In this traditional arrangement, at the end of the month, the total is returned to the client minus a day’s contribution, which the collector keeps as a commission6.

The ﬁrst rural bank was built in the Central Region in 1976. Rural and community bank shares, by law, must be entirely owned by residents of the traditional area hosting the institution (Steel and Andah 2003). Community members elect the Boards of Di- rectors, based on professional experience and community reputation. Following large responses to community ownership, more than a hundred were instituted in less than a decade after the late 1970s (Nair and Fissha 2010).

Following the broad Ghanaian (and African) ﬁnancial decline of the 1980s, several reforms led to rural and community banks dominating formal rural ﬁnancial activities in the country (IFAD 2008). On the other hand, susu collectors dominated the informal

ﬁnancial space following their expansion after the ﬁnancial crises (Bortei-Doku and

Aryeetey 1995, Aryeetey, Hettige, Nissanke and Steel 1997).

The increased susu participation was robust to increased formal ﬁnancial institutions and other instruments in Ghana7. Collectors are currently the most signiﬁcant informal economic group, in terms of the quantities of savings mobilized (Aryeetey

1994, Aryeetey and Udry 1995). The cultural and economic dominance of susu collection in Ghana extends to individuals who favor strictly formal means of mobilizing

ﬁnance. For example, in addition to traditionally informally-ﬁnanced entrepreneurs, our random survey of susu clients include doctors, nurses, and teachers. A national cooperative (the Ghana Cooperative Susu Collectors’ Association) was set up in 1994.

Interestingly, susu collectors have become occasional intermediaries for commercial banks (e.g. Osei 2007) as banks attempted to expand beyond the urban elite.

Rural and community banks (such as my partner, Kakum Rural Bank) are much less restricted to urban areas in Ghana, and rely on a much greater extent on susu col-

6Such commissions are not present in the model focused on in this paper. 7Aryeetey and Gockel (1991:41) 104 lectors. Local susu collectors employed by rural and community banks cooperate with bank staff to streamline bank services at more fundamental levels of bank operations.

Interconnections between rural banks, susu collectors and microfinance NGOs are increasingly common in Ghana, although implementation (and the motives influencing such applications) tend to be led by rural bank officials. Over time, some banks have gradually shifted to relying on their own internally trained collectors, drawing on their regular research and fieldwork to update their services for the market, including the design of susu savings deposit schedules.

A collector may cover several towns and hundreds of individuals every day (collectors in the sample cover 213 clients per day, on average). The arrangement represents a

ﬁnancial loss to clients (e.g. Besley 1995, Chamlee Wright 1997) and the need for individuals to protect themselves from their own impulses (e.g. Bryan, Karlan and Nelson

2010). However, formal banking also has costs that are signiﬁcant to would-be clients.

Although financial and transactional costs are short-term, reputational costs (incurred by leaving market wares and potential customers unattended for extended periods of time) may be relatively difficult to recover from. During interviews, clients observed the convenience (of having collectors come to them at their places of business and saving them the trouble of losing potential customers) as a major factor influencing their choosing to contribute savings using susu collection.

4.3 Model and Predictions

I rely on external and internal behavioral incentives that already exist in the literature.

As noted earlier, the goal of the study is to integrate these two perspectives. I do by assuming that individuals are sophisticated hyperbolic discounters who also have varying abilities to self-monitor.

Consider a set of individuals Ai(i = i...I). These individuals are hyperbolic discounters with time-inconsistent preferences, which means that for any individual, optimal 105 plans to save from today’s perspective will not be implemented by her future selves with probabilities p = 0 < π ≤ 1. Let there be a negative relationship between the independent ability to save by self-monitoring, and p. Individuals who have a lower- ability to self-monitor have likely to be more time-inconsitent (or less likely to follow through on their plans.) These preferences motivate her value of susu collection arrangements D j( j = 1...J), where ( j = 1 > ... > j)in terms of their required internal commitment. In the mould of the literature, individuals who recognize a need for susu collection and commitment are sophisticated. However, I depart from this convention by considering individuals who recognize their relative need to be comparatively sophisticated, so that individuals have varying degrees of awareness of their needs for commitment. Although susu collection is primarily externally focused (on a collector), the absence of force in a susu collection arrangement (and the presence of choice in my case) imply that personal rules within external commitment are potentially important.

Assume a sophisticated individual Ai recognizes her relative need for a susu collection arrangement D j. Individual Ai also develops personal rules Rk(k = 1...K) that are based on self-reputation over her will-power and ability to self-monitor while using susu collection. For these personal rules, k represents the timetable or schedule on which Ai decides to meet with collector D j( j = 1 > ... > J) and make savings contributions, so that Rk = 1 > ... > k. To link external commitments (based on hyperbolic discounting and sophistication) to willpower within an external commitment (personal rules), I assume via a reduced-form argument (that I do not model here) that individual sophistication is associated with personal rules made within external commitments.

Specifically, sophistication positively correlates with the level of internal commitment reflected in the personal rules an individual chooses Rk that are bound by an external commitment. Each personal rule reflects a level of personal commitment that reflects the degree of sophistication, and hence the value of commitment devices. For simplicity, the outcomes of personal rules (within an external commitment) is a random 106 variable with an identical distribution across individuals, but there is a positive correlation for individuals with similar personal rules within external commitments.

RESULT: Individuals who are relatively sophisticated select personal rules that reﬂect a higher ability to self-monitor within commitment devices.

The above result is based on intuition. Based on the above discussion, awareness

(sophistication) prior to adoption affects internal motivation within the scenario of external commitment. On the other hand, individuals who are less sophisticated select personal rules that show a lower ability to self-monitor with commitment devices. This is because I assume the less awareness (sophistication) to play out as less internal motivation within the scenario of external commitment.

PREDICTION 1: Personal rules within external commitments have positive impacts on savings outcomes.

The corresponding null hypothesis is that personal rules do not have signiﬁcant effects on savings outcomes.

PREDICTION 2: Within external commitments, the impact of personal rules on savings outcomes is strictly increasing in the degree of internal commitment (relative to personal rules that require less commitment). Given hyperbolic discounting, a higher level of sophistication correlates positively with the ability to self-monitor Rk while using an external commitment D j. As noted, positive correlation exists for individuals with similar personal rules.

Let Sl (l = 1,2,3,4,5) represent the set of savings schedules an individual may choose from, where l = 1 represents the highest level of constrained personal commitment. Prediction 2 is summarized as follows: S1 > S2 > S3 > S4 > S5. 107 4.4 Survey Data, Savings Schedules Design and Mea-

suring Internal Preferences for Savings Commitments

The author collected data over three months (June to August 2010), in a collaborative effort with susu collectors employed by Kakum Rural Bank, a rural community bank in Elmina serving the broader Ghanaian Central Region. An initial primary survey randomly selected 15 susu collectors employed at the community bank. The client survey covered 384 randomly selected susu clients. I discuss critical aspects of individualized susu savings collection.

For individuals who chose to commit to daily, weekly, fortnightly (every two weeks), or monthly savings schedules with their deposit collectors, Kakum Rural Bank created the Kakum Enyidado Susu Scheme. With feedback from susu clients, the community bank also collaborated with the Microsfere Fund for People and Nature NGO in 2009 to create a schedule whereby customers could save two times every week (Microsfere

Fund for Nature 2011). All schedules are implemented by Kakum Rural Bank. Susu clients may choose and save on any one of the above ﬁve schedules, as consistent with their perceived needs for commitment savings.

Our measures of the degree of internal commitment of an agent to save are the savings schedules or frequencies at which a client decides to meet with a susu collector and contribute savings. Saving on a daily basis, (setting aside business funds every day to honor daily meetings with a susu collector) represents the highest degree of commitment. In decreasing order of personal commitments, are twice weekly (on two days every week), weekly, fortnightly and monthly savings schedules or timetables.

Clients may choose how much to save based on their internal preferences in saving, and their commitment to the deposit collection process. As noted earlier, clients in our study are mostly sufﬁciently sophisticated to adopt a commitment savings device. The more internally committed they are to their savings device, the more they should save, 108 and the less they should save when they show less commitment to do so.

Table 4.1 shows descriptive statistics for susu clients. We ﬁnd that 25.8% of the client sample chose to save on a daily basis, 24% two days per week, 28.3% every week, 17.4% every two weeks, and 4.4% clients chose to save every month. All savings schedules yielded savings contributions averaging under 10 Ghana cedis per day

(across all savings schedules). We find that each schedule is almost split by gender, and the majority of clients self-identified as sole proprietors of micro enterprises. With the exception of the clients who saved every two weeks, household heads were minorities in every savings schedule. We also account for monthly incomes (earned on the month before survey implementation), finding them broadly comparable across savings schedules. 109 110 4.5 Empirical Strategies: OLS and Propensity Score

Matching Estimators

OLS Regression

The methodologies are discussed in this section. Although data on susu collection at the national level are rare, studies agree that the savings mechanism has been representative in the informal sector since the 1990s (e.g. Bortei-Doku and Aryeetey 1995).

This observation is signiﬁcant since informal economic systems account for 90% of employment in Ghana (Heintz 2005). At the same time, susu collection now extends into the formal ﬁnancial sector, with almost every rural bank, commercial bank, and savings and loans company in Ghana having a susu department (e.g. Adusei and Ap- piah 2012). Given the popularity of susu collection in Ghanaian formal and informal

finance, finding and surveying a group of individuals that had no observed exposure to susu collection whatsoever was simply not a realistic assumption or approach. How- ever, the creation of separate susu collection modules to better reflect customers’ personal rules provide clear comparison groups via a quasi-experimental identification strategy.8.

We allow each savings schedule, in turn, to serve as a treatment, with the savings contributions per day serving as dependent variable. The hypothesis that savings schedules representing more commitment should yield more savings contributions is tested in the following equation, where yi refers to savings outcomes for individual i:

8 A group unexposed to my susu savings treatment may be exposed to a different susu program, leading to potential bias. To the extent that susu collection unobserved during ﬁeld research could affect savings outcomes, the behavorial and economic distortions emphasized by forcing a controlled experiment approach would lead to bias in any comparative exercise. Also, some of the clients in my sample are doctors, teachers– professions typically associated with formal ﬁnance. 111

yi = α + Tkβ + Xiγ + εi

where Tk is an indicator for either a daily, twice-weekly, weekly, fortnightly, or monthly savings schedule while Xi refers to individual covariates.

Propensity Score Matching

Since the individuals chose savings schedules based on personal rules, I study the differences in savings between savings clients who choose different rules. To determine the effects of the varying individuals’ savings commitments on savings outcomes, I discuss propensity score matching (Rosenbaum and Rubin 1983) in this section. Using a probit or logit equation for the probability of using a savings schedule, a comparison group is allowed to emerge from the non-treated group, for which the distribution of covariates are similar. I use a logit model to estimate every propensity score of the probability that an individual entered a particular savings schedule as a function of the following surveyed factors: age, gender, entrepreneur, household head, and monthly income. The logit regressions for the propensity scores are provided in Appendix A. I focus on one-to-one (nearest neighbor) matching (following Dehejia and Wahba (2002) and Nunn (2007)), as well as kernel matching (using a normal density function) with the STATA software developed by Leuvena and Sianesi (2003). In both cases, the analyses are limited to regions of common support.

Since propensity score matching estimates replicate experimental ﬁndings (assuming no selection on unobservables9), this process provides Average Treatment Effects

(for the treated). Since no experimental results would be possible in areas where nearly all participants engage in a practice, I compare different savings schedules. Also, an experiment that artiﬁcially manipulated savings schedules chosen by participants would

9see Dehejia (2005). 112 likely make them worse off and so would not be ethical. My study is concerned with personal rules, which are self-selected choices by deﬁnition.

First, I use the basic nearest neighbor (one-to-one) matching with replacement. For treated individual i with propensity score Pi, the non-treated individual j ∈ I0 with propensity score Pj that is most similar to Pi is selected in the comparison algorithm as a match. For all treatments, a counterfactual is created using controls most similar on a unidimensional propensity score. The comparison of clients with similar covariates is facilitated by comparing outcomes for clients with similar propensity scores. This approach implies a reduction in bias and a rise in the average quality of matches (relative to matching without replacement10). Restricting all estimates to the region of common support, I estimate:

N (Pi) = min k Pi − Pj k, j ∈ I0 j

I also use a kernel matching estimator which is yields even smaller variance (developed by Heckman, Hidehiko, and Todd 1997, 1998a, 1998b). For treated individual i with propensity score Pi, a weighted average is calculated using multiple clients in the comparison group. In this case, the average treatment estimates are calculated on kernel weights so that although control cases closest to the treatment receive greater weights (Smith and Todd 2005). A possible shortcoming is the possibility of adverse matches, which is lowered by imposing the common support. 11

For savings outcomes si of individual i, exposed to savings schedule treatment T, we use a normal kernel, N. Let pi and p j be propensity scores for i and j respectively,

10 See Zhao (2004) and Caliendo and Kopening (2008).

11(Caliendo and Kopening (2008) note a trade-off between high bandwidth values (which correspond to better ﬁt and lower variance between the estimated and true density functions), and bias. 113 while b is the (default) bandwidth12 We are interested in the following:

pi−p j ∑ j∈{T=0} N b si sˆi = pi−p j ∑ j∈{T=0} N b

pi−p j N b ψ = pi−p j ∑ j∈{T=0} N b

where the weights of the outcomes for control j is given by ψ. All logit equations are presented in the Dissertation Annex.

4.6 Savings Mobilization Results

In this section I analyze the effects of savings schedule choices (that reﬂect internal personal rules) on savings mobilization. The results show Ordinary Least Square estimates, and average treatment effects of both nearest neighbor (one-to-one) matching without replacement and kernel matching. The treatment schedules are daily, twice- weekly, weekly, fortnightly, and monthly savings plans. I ﬁrst perform the impact analysis for each treatment schedule in turn, with all other schedules serving as a comparison group (as far as similar propensity scores were concerned), to test whether savings mobilization responses to adopted savings schedules are increasing in the degree of internal commitment. I then perform the same nearest neighbor and kernel-based propensity score matches with a single savings schedule acting as a comparison group.

12See Leuven and Sianesi (2003). 114

4.6.1 Savings Schedules Impacts On Contributions: OLS Results

OLS results are presented in table 4.2. Each column represents a different regression, with Column 1 comparing daily, twice-weekly, weekly, and fortnightly schedules to the monthly schedule. The daily schedule shows a significant positive effect on savings contributions (relative to the monthly schedule), while the fortnightly shows a similar but slightly less positive impact on savings contributions relative to the monthly schedule. None of the other schedules have significant impacts on the mobilization of savings. The daily schedule impact of a 17.57 Ghana cedis increase in savings is equivalent to about $9 while the positive fortnightly schedule’s impact (8.24) represents a $4 rise in savings. Although age, and being a household head did not have significant effects on savings mobilization, I find a negative gender effect (women in the sample save less), while people with higher incomes and entrepreneurs save more. 115 116

4.6.2 Savings Schedules as Personal Rules and Savings Contribu- tions: Using all other savings schedules as comparison groups

Individuals chose susu savings schedules based on personal rules that represent intrinsic differences at the individual level. In this section, I show differences in savings outcomes between savings clients who choose different rules. Table 4.3 shows average treatment effects of individuals who adopted schedules on savings mobilization. The

ﬁrst row presents results in which the treatment is an indicator variable. The second to

ﬁfth rows show treatments of individuals saving twice a week, ever week, every two weeks (fortnightly), and monthly, respectively. In every treatment case, all alternative schedules serve as a comparison group.

Choosing to save on a daily schedule, the treatment requiring the highest level of commitment, has large positive and statistically significant effects on savings mobilization (13.19 Ghana cedis or $7.10). On the other hand, the average treatment effects for individuals who choose the savings schedules that require less commitment do not have significant effects. Saving two days every week has a positive average treatment effect, less than the daily savings effect, but the effect is not significant. Saving on a weekly, fortnightly, or monthly basis have negative effects on savings, but not significant ones. The effects become increasingly negative (as less commitment is required), showing a consistent pattern, albeit a statistically insignificant one. These findings suggest that the average treatment effects of the daily schedule are relatively important in interpreting the importance of self-selected commitments, relative to all alternatives. 117 118

In Table 4.4, we repeat the same exercise from Table 4.3, but showing average treatment effects for individuals who chose savings schedules with kernel matching using a normal density function. The first column of the table shows a relatively similar average treatment effect for the daily schedule: individuals who chose the most commitment- intensive (daily) savings schedule have a large, positive, significant impact on savings contributions. Saving every two days, or every week, fortnight or month have negative effect on savings, but not significant impacts. On the other hand, the average treatment effect of saving on a monthly basis (the schedule requiring the least degree of commitment) has a negative (relative to the non-daily schedules) effect. 119 120

Following the second prediction, individuals’ schedules choices that require the most commitment, and the least commitment have the most positive and negative impacts on savings respectively. The schedule choices whose levels of personal commitments are between the two extremes decrease in impact but not signiﬁcantly. The results are broadly similar to the argument on increasing outcomes with higher levels internal commitment. The negative savings impact of the monthly schedule choice is relative to all other schedules (that require more internal commitment), and thus broadly consistent with the main argument. I now compare each treatment savings schedule with single alternative savings schedules, using the same matching techniques discussed above.

4.6.3 Savings Schedules Impacts On Contributions: Using a single savings schedule as a comparison group

We now turn to effects of choosing the savings schedules, using every single savings schedule as a comparison group in tables 4.5 and 4.6.

Table 4.5 presents average treatment effects using one-to-one nearest neighbor matching. Choosing the daily schedule has a positive significant savings effect, although the effects are only significant for the fortnightly and monthly comparison groups. Results from the twice-weekly treatment schedule are only slightly significant. None of the savings effects are significant when the weekly schedule serves as a treatment. Choosing the fortnightly schedule has a effect impact on savings when the daily schedule serves as a comparison group. Choosing the monthly schedule also has a statistically significant negative impact on savings relative to a daily comparison group. The monthly treatment has a negative effect relative to a fortnightly comparison group. 121

Table 4.6 shows show kernel-matched results for each susu savings schedule relative to individual alternative schedules. In the second column of Table 4.6, we ﬁnd that relative to saving twice every week, individuals who choose the daily treatment 122 schedule show a large, positive effect on savings. The daily schedule has a positive signiﬁcant effect on savings, relative to the fortnightly schedule, as well as the monthly savings schedule. Saving twice every week has a negative impact on savings contributions, relative to the daily schedule comparison group. However, the twice weekly schedule has a positive effect on savings relative to the monthly schedule. Similar to the twice-weekly schedule, individuals who choose the weekly schedule have a significantly negative effect on savings, when the individuals who choose the daily schedule serve as a comparison group. Saving every two weeks has a strong and negative effect when the daily savings comparison group is used. The monthly schedule has negative effect on savings contributions relative to daily as well as fortnightly comparison groups. 123 124

The one-to-one and kernel matching results for savings contributions are broadly similar. Individuals whose personal rules reﬂected the highest commitment (the daily schedule) relative to all other schedules had the highest impacts on savings contributions. The same ﬁnding is repeated when the impact of the daily schedule is studied relative to individual alternative schedules. On the other hand, clients who chose savings schedules with that required relatively less commitment had lower savings contributions relative to all other schedules, as well as in comparison with individual schedules.

I discuss the effects of individuals who chose various savings schedules on savings rates in the next section. While the above finding is consistent with the prediction, I find that the twice-weekly, weekly and fortnightly savings schedules do not have significant savings impacts when the comparison group consists of all other schedules. On the other hand, I find that savings schedules between the extremes (of daily and monthly) have negative effects relative to individual schedules that require more commitment.

4.6.4 Susu Savings Rates Results

In this section, we consider the effects of individuals choosing savings schedule commitments on susu contributions per monthly earnings, or savings rates. Although we already consider earnings in our matching algorithms, this measure may more robustly account for the inherent economic differences among clients. We observe that differences across schedules are less signiﬁcant in the descriptive statistics are presented in table 4.7. 125 126

Again, our treatment schedules are daily, twice-weekly, weekly, fortnightly, and monthly savings plans. We now present Ordinary Least Square estimates, and Average

Treatment Effects of both nearest neighbor and kernel-based propensity score matching estimators. We perform the matches consistently with the empirics presented earlier in the paper.

4.6.5 Schedule Impacts On Savings Rates: OLS and Propensity Score Matching

Table 4.8 shows the schedule effects on rates of saving based on OLS. In the OLS regressions, the daily schedule shows a positive effect on savings rates. In addition, saving every two weeks, or on a fortnightly basis have strong impacts on savings rates in table 4.8. The OLS results show increasing savings rate impacts of schedules as the schedules chosen require increasing internal commitment. The only exception is the twice-weekly schedule, which does not show a signiﬁcant savings rate effect. 127 128

Table 4.9 shows the results of one-to-one matching with replacement (with all other schedules serving as comparison group). The analysis did not yield signiﬁcant effects for individuals who chose any of the savings schedules. Economic impacts are broadly similar to previous discussions and in qualitative agreement with the theoretical prediction that savings schedules which require higher internal commitment yield higher savings outcomes. 129 130

On the other hand, we find the daily schedule and the monthly schedule showing negative significant impacts on savings rates when we use kernel matching. Table 4.10 shows these results. Individuals whose personal rules reflected the highest commitment

(the daily schedule) relative to all other schedules had the highest impacts on savings rates. When the comparison group consists of all other schedules, I found that the twice-weekly, weekly and fortnightly savings schedules do not have significant impacts on savings contributions (in tables 4.3 and 4.4). However, there are significant schedule impacts on savings rates. These findings imply that the savings schedules’ relative impacts become more visible when the outcome variable is relative as well. For the purpose of further disaggregation in the analysis, I use single savings schedules as comparison groups in the next section. 131 132

4.6.6 Schedule Impacts On Savings Rates: Using a single savings schedule as a comparison group

In tables 4.11 and 4.12, I find the savings rate effects of choosing the savings schedules relative to individual alternative schedules, using one-to-one and kernel matching respectively. Table 4.11 shows the results of savings rate impacts using one- to-one matching. The daily and twice-weekly schedules have significant effects on savings rates when the weekly, fortnightly, and monthly schedules (representing relatively less commitment) serve as comparison groups (using one-to-one matching). The weekly and fortnightly schedules have significant effects on savings rates relative to the monthly schedule. The latter has a negative effect on savings rates relative to the fortnightly schedule, which requires relatively more commitment. 133 134

Table 4.12 shows the results of savings rate impacts using kernel matching. In table 4.12, the clients who saved on a daily basis had significant impacts on savings rates relative to weekly, fortnightly and monthly savings schedules. The weekly and fortnightly savers had significantly lower savings rates relative to daily savers. The monthly savers had significantly lower savings rates than their counterparts who saved every day or two days every week. 135 136 4.7 Discussion of Results

Individuals whose personal rules reflected the highest commitment (the daily schedule) relative to all other schedules had the highest savings contributions. The same finding is repeated when the impact of the daily schedule is studied relative to individual alternative schedules. Where the use of external commitments (deposit collectors) are subject to internalized personal rules (savings schedules), the rules which reflect the highest degree of commitment may yield the most positive and significant outcomes, irrespective of comparison group. On the other hand, clients who chose the savings schedule that required the least commitment (the monthly schedule) had lower savings contributions relative to all other schedules, as well as in comparison with individual schedules. I find that the twice-weekly, weekly and fortnightly savings schedules do not have significant savings contributions when the comparison group consists of all other schedules. This shows that the positive relationship between internal commitment

(driven by sophistication in the model) and savings outcomes may not be strictly increasing, but be subject to disruptions or breaks. These breaks may tell us more about the comparison group than treatment groups, since I ﬁnd that savings schedules between the extremes (of daily and monthly) have negative impacts relative to individual schedules that require more commitment.

Relative to our analysis on savings contributions, I find a closer relationship between internal commitment and savings rates when we find the outcomes of a savings schedule relative to individual alternatives. For instance, when we compare daily savers with comparison groups whose savings schedules require increasingly lower levels of personal commitment, we find positive and significant effects on savings rates. The

“breaks” in significance (where there are no significant effects of schedules on savings rates) are generally rarer. On the other hand, the impacts of saving schedules on savings rates were not significant when we used the one-to-one matching estimator

(and a larger comparison group), although the economic impacts (coefﬁcient signs) are 137 consistent with our discussion of internal commitments.

The ﬁnding of occasional breaks in the relationships between individual willpower that is constrained and savings outcomes relates to theoretical literatures. For example,

Benabou and Tirole’s conceptual frameworks (2003, 2004) argue that the presence of external commitments may have negative impacts on internal commitments. Another theoretical discussion that has bearing on our empirics is conveniently summarized in

Ozdenoren, Salant and Silverman (2011) which considers agents to be endowed with a stock of willpower that cleanly diminishes as the agent restrains consumption. Within this setting, an agent with limited willpower may relax control of consumption over time while using a commitment savings device. Although I ﬁnd much evidence that is in agreement with either framework, the individuals who choose savings schedules that vary in terms of commitments did not always have strictly increasing (or decreasing) gains in savings outcomes, but also occasional breaks in savings effects. I ﬁnd that the degree of commitment is mainly important in impacts of the daily schedule

(which requires the most commitment) and the monthly schedule (which needs the least commitment). However, I do not ﬁnd an entirely strict adherence to a hierar- chy of commitment levels in most of our results, and this may be worthy of further theory-based discussion.

4.8 Conclusions and Policy Implications

Does the utility from an external (economic or other) commitment device depend on the internal personal rules manifesting in the frequency of its use? This issue draws on the self-control problem: the conflicting tendency to yield to short-term needs instead of waiting for long-term interests to manifest. The economic literature has seen a surge in applications of models to topics such as savings mobilization, and motivated external commitment devices such as microsavings devices. This focus is partially attributable 138 to the slow institutional evolution in informal financial arrangements. However, a relatively recent economics literature influenced by sociology and psychology has made the case for self-imposed rules (e.g. swim three days a week, consume alcohol only on special occasions, etc.) in commitments, and these can be applied to savings behavior.

The present paper focuses on applying this dialogue on preferences for commitment to empirical observations on savings mobilization. Using a quasi-experimental design, I apply this dialogue to new data on susu deposit collection, a commitment savings device in Ghana. Relative to other savings schedules that require lower degrees of commitment, I ﬁnd that choosing a daily savings schedule (the personal rule that represents the highest level of commitment) is correlated with savings contributions, as well as savings rates. I also ﬁnd that commitments to saving on a monthly schedule have negative impacts on savings contributions and savings rates. Where the use of external commitments (deposit collectors) are subject to internalized personal rules

(savings schedules), the rules which reflect the highest degree of commitment may yield the most positive and significant outcomes. On the other hand, commitments may yield negative outcomes if internal personal rules do not require a high-enough commitment. The findings imply that other settings whereby commitments are subject to personal rules would benefit from similar study, and could yield similar results or related models.

Although external commitment devices and internal personal rules are individu- ally studied in much of the behavorial microeconomics literature, less is known about external commitment devices that are subject to internalized personal rules. Yet, the interdependence of personal rules and external commitments are important for our conceptual frameworks and empirical investigations on willpower and self-control. The main contribution of this paper to the above conversation is that personal rules may re-

ﬂect optimal levels of commitments and individuals’ intrinsic differences when agents are presented with a menu of commitments, instead of a single choice. Observations of 139 willpower may not be entirely exogenous, but may depend on pre-treatment variables

(American Psychological Association 2011:3). Similarly, Spears (2011) argues that poverty makes economic decision-making difﬁcult, depleting cognitive control. If this is the case, then pre-treatment factors may affect self-selection into savings schedules, and must be conditioned on, before I can credibly isolate the impact of personal rules on savings outcomes.

Providing participants with a menu of commitment devices that vary in terms of the internal motivation they represent has some policy implications. The provision of agents with a single variable to choose from (instead of a menu) inherently assumes the agent’s inability to choose correctly. An individual who chooses a particular treatment may have chosen another, were it available to them–making it difﬁcult to isolate the treatment effect attributable to one treatment relative to others. This is a rare counterfactual that this study provides, allowing a comparative exercise. Similar initiatives could be attempted in varying contexts such as retirement savings, and smoking ces- sation initiatives that rely on both internal and external commitments. Since I do not observe economic outcomes of gained savings outcomes (such as business investment), the current welfare effects are not clear, although this question may be engaged with future research.

This understanding of willpower may be most important in economic initiatives aimed at combating poverty, since some of the most mental stress of the poor is in areas of ﬁnance. The recent challenges facing microcredit are instructive that newer

finance programs may benefit from more initiatives that take the economic limits of the poor into account. In this respect, a policy implication is that the poor may be relied upon to choose optimal commitment programs (in areas such as microinsurance, and business education) out of their own initiative–out of personally-modifiable menus. 140

Chapter 5

Heterogenous Signaling at the

Convergence of Formal and

Informal Finance in Ghana

5.1 Introduction

The relationship of informal finance1 with state-sanctioned, formal financial systems and policies is receiving new research attention from economists. Advocates of linking informal and formal arrangements argue that formal financial institutions have relatively little relevant information on potential borrowers in the informal sector. Infor- mation flows (especially in developing countries) may be improved by merging formal and informal financial sectors (e.g. Anon and Stiglitz 1991, Madestam 2010).

Critics, on the other hand, note the general historical experience in ﬁnancial development. Formal ﬁnancial development tends to be associated with improved economic

1Informal finance, for this paper, refers to as institutions and systems encompassing all non-market financial movements that are beyond the control of a political state. 141 outcomes (see Rajan and Zingales 1998, Levine 2006). Therefore, the policy role of informal financial institutions may be relegated to smaller subsets of the economy.

Given the many policy and academic debates on the ability for informal finance to be linked with, (or even coexist with) formal finance, this issue has significantly broad implications for economic development2.

Can informal savings arrangements improve the information ﬂows to, and from, a formal banking system? We present a creditworthiness signaling model and study a Ghanaian program to investigate this issue. This approach allows us to link lending rural banks in the formal sector to credit-constrained (but saving) borrowers in the informal sector. Our intermediary mechanisms are savings deposit collectors that transcend both sectors.

I consider a scenario where (1) deposit collectors are employed by rural and community banks, and (2) customers’ savings contributions credibly signal creditworthiness. We are interested in the following: (a) the relative extent to which different savings mobilization schedules act a signaling devices, (b) whether banks respond to these signals with credit, and (c) the ability of these credit responses to target the poor who are exert more effort in mobilizing savings3.

I worked in partnership with Kakum Rural Bank, a community-owned ﬁnancial institution focused on operationally linking formal and informal ﬁnance in Ghana. With

15 susu collector employees, we randomly surveyed 384 individuals who contribute savings on a daily, twice-weekly (two days every week), weekly, or fortnightly basis.

Owing to the varying frequency in submitting savings, (as well as barriers to successful saving), each savings schedule represents (1) a different degree of ﬁnancial effort by

2The academic discussions on formal versus informal finance extend through much of social science, spanning several literatures and sub-disciplines within the fields of economics, political science, anthropol- ogy, as well as sociology. The World Bank Development Research Group’s Finance and Private Sector Development Team has engaged and sustained a policy debate on the evolving role of informal finance in development. See http://econ.worldbank.org. 3New research indicate that targeting the poor using social community-based methods lead to higher satisfaction than proxy-means tests. See Vivi Alatas, Abhijit Banerjee, Rema Hanna, Benjamin Olken and Julia Tobias (forthcoming). 142 an individual, and (2) a different signal to a susu collector in terms of strength. We also collected data on credit outcomes for individuals who contributed savings for at least three months and applied for loans.

In a separating equilibrium, individuals whose entrepreneurial ability exceeds an arbitrary threshold will submit to a test that conﬁrms their capabilities (e.g Spence

1974, 1975). Given asymmetric information, individuals may make an indirect reliable signal of creditworthiness by providing effort that is correlated with their creditworthiness. We argue that credit-constrained individuals of increasing creditworthiness will submit to increasingly challenging tests to provide a credible signal to gain credit.

Accounting for schedules of varying difﬁculty allows us to study quantitative credit impacts of different saving efforts.

I find that individuals who exert the most effort in submitting savings (on a daily basis) showed the strongest positive credit impacts of around 300 Ghana cedis. I also control for self-selection into savings schedules. Our main outcome holds both when the comparison group incorporates all other schedules as well as when a single alternative schedule acts as a comparison group. In addition, we find a negative effect for individuals using the least rigorous fortnightly savings schedule. The signaling effects of schedules in between these schedules confirm our main result, although these impacts are not always significant. Our main findings are robust to considering impacts of missed savings payments to deposit collectors or possible reverse impacts of credit received on savings amounts.

The results have implications for several literatures. First, the convergence of informal and formal ﬁnance, may be helpful for eliminating information asymmetries in ﬁnance (note Stiglitz 1990). The role of personal relationships in related outcomes is well documented (e.g. Udry (1990, 1994), Steel, Aryeetey, Hettige and Nissanke

(1997) and La Ferrara (2003)) 4.The ﬁndings may have even more inferences for the

4Other proponents of informal institutions’ information capabilities (broadly speaking) include Stiglitz (1990), Arnott and Stiglitz (1991), Besley (1995), and Fafchamps (2003). Social capital is understood to heavily inﬂuence the information-related advantages of informal ﬁnancial arrangements in a large literature. 143 developed world. For instance, Fenn, Liang and Prowse (1997) as well as Ayyagari,

Demirgüc¸-Kunt and Maksimovic (2010: 3049) observe similarities between informal

ﬁnanciers in developing countries and wealthy “angel investors” in the United States.

These entities may rely on metrics such as the proportion of net worth invested in a ven- ture, or the proportion of equity retained by an entrepreneur as an investment decision signaling device (Prasad, Bruton and Vozikis 2000).

Although our focus is on African ﬁnancial markets, the analysis is also germane to a recent debate in the policy literature that has coincided with the recent fast growth of

Asian economies. For example, Tsai (2004), Allen, Qian and Qian (2005) and Linton

(2006) have argued that informal ﬁnancial arrangements played a important role in

Chinese ﬁnancial development. These ﬁndings contrast with Meghana Ayyagari, Asli

Demirgüç-Kunt and Vojislav Maksimovic (2010) who make a competing case that formal finance may more strongly associated with Chinese firm-level growth.

Other studies have conceptually studied vertically linking both ﬁnancial sectors, using moneylenders to mitigate information asymmetries through through peer-monitoring.

Moneylenders are often traders who provide subsidized formal credit to entrepreneurs and may improve information ﬂows to rural banks. The following studies are examples:

Hoff and Stiglitz (1997), Floro and Ray (1997), Bose (1998), and Madestam (2010).

The growing body of empirical work including Bell, Srinivasan and Udry (1997) as well as Jain (1999) and Varghese (2005) frame informal and formal ﬁnancial arrangements as competitors: they are cooperative in Ghana (Aryeetey and Steel 1995). While moneylenders provide credit, they cannot provide certain important measures of degrees of entrepreneurial risk, such as savings levels.

This function of savings mobilization is often provided by formalized savings deposit collectors. A closely-related paper to this chapter is Ashraf, Karlan and Yin

(2006b). The authors study the factors inﬂuencing the adoption of a deposit collection

Although social dynamics may be less significant in the economics literatures of formal finance, Guiso, Sapienza and Zingales (2004) observe roles of social capital in financial development. 144 program organized by the Green Bank of Caraga in the Philippines. The present study differs in two ways. One distinction is that I model a signaling process through which savings frequencies indicate and accordingly determine the merit and availability of credit. This approach helps us to understand how the economic relationships between clients and collectors are incentivized and maintained. Also, the data successfully iden- tifies a sample of clients for more than one savings timetable, motivating comparisons of savings and their inherent efforts. Another relevant study by Eeckhout and Munshi

(2010) considers matching in group-based Indian rotating savings and credit associations (ROSCAs). Since ROSCAs primarily mobilize savings, another contribution of this paper is an introduction of credit to matching discussions.

This paper reads as follows. Section 5.2 discusses the institution of susu collection and the convergence of informal and formal ﬁnance in Ghana. Section 5.3 describes the model and empirics. Section 5.4 provides the analyses and results. Section 5.5 closes the paper.

5.2 The Susu Collection Institution at the Convergence

of Formal and Informal Finance in Ghana

“Susu” (translated from the Akan language as to “measure” or “estimate”) has most often been linguistically associated with rotating savings and credit associations, abbreviated ROSCAs. These savings groups are usually united by ethnicity or occupation or some other deﬁning characteristic5.The foremost research on these institutions have been mainly anthropological with newer economic treatments6.

5A typical ROSCA operates as follows: at regular meetings, each member contributes a ﬁxed amount into a “pot”, and a single member receives the total amount. Since the fund rotates, each member can expect to receive the sum at some time in the future. ROSCAs are popular in developing countries as well as in developed countries, particularly among immigrant communities (e.g. Bonnett 1981). Savings and loan associations in the United States emerged from ROSCAs (e.g. White 1984 and Grossman 1992). 6Some initial forays into the ROSCA literature are Ardener (1962) and Geertz (1964). Economic treatments of ROSCAs generally stem from Besley Coate and Loury (1993). 145

In Ghana, the financial crises of the 1980s coincided with a heightened need for cash on hand for individual obligations. This situation influenced a relatively sudden shift in susu savings participation from group-based ROSCAs to individual susu col- lection7: the main focus of the present paper. The prevasive boom in susu collection has been related to an inertia in formal financial services (e.g. Aryeetey and Steel

1995). A survey by Aryeetey and Gockel (1991) estimated that 77% of Ghanaian market women saved their funds in this manner8.Rising numbers of individuals who may have saved collectively in the past, would nominate an reputable individual, who would visit market stalls, farms, or other areas of commerce and mobilize funds on the behalf of clients on a daily basis. The total for each saving client would be returned at the end of the month minus a single day’s deposit which the susu collector would keep as a commission.

Economic and other incentives motivate such commitment savings devices in Ghana

(such as the relative time and ﬁnancial costs of leaving places of business to deposit savings, or the difﬁculty of honoring plans to save9). However, the observed willing- ness to pay to save is counterintuitive. Using independent susu collectors is often a

ﬁnancial loss for entrepreneurs (e.g. Rutherford (2000); Ashraf, Gons, Karlan and Yin

(2003)). Another advantage, therefore, of linking susu collectors with formal institutions is the elimination of commissions that independent collectors must charge for their service. In ﬁeld interviews, it was noted that beyond the assured safety of funds, an incentive for independent susu collectors to cooperate with the formal sector is their low lending ability.

Unlike other parts of Africa where microﬁnance is dominated by NGOs, formal

ﬁnancial services are supplied in rural Ghana mainly by community banks that rely on

7The institution of susu collection in Ghana predates the 1980s and 1990s, but rose in prominence during and after this period. See Bortei-Doku and Aryeetey (1995) for details. 8This relatively new focus on susu collection has not been exclusive to Ghana. For instance, Steel, Aryeetey, Hettige and Nissanke, (1998) shows a rise in savings deposit collection following the ﬁnancial crises in Tanzania, Nigeria, and Malawi. 9See O’Donoghue and Rabin (1999) for a general discussion. 146 susu collectors (Steel and Andah 2003). These institutions are required by law to be majority-owned by the local community10. Susu collectors are important to banks because of their information advantages: they mobilize funds on a consistent basis, as discussed above. However, bank susu collectors are also important to credit-constrained clients since they allow individuals to credibly indicate their creditworthiness to community bank. We provide some necessary details on need for credit to supplement susu savings in Ghana in the next section.

5.2.1 Susu Collectors and The Need for Credit among Saving En- trepreneurs

Rural banks have increasingly relied on susu deposit collectors to mobilze funds traditionally lost to the informal sector, and often employ susu collectors to mobilize funds as permanent employees (Nair and Fisha 2010). Although rural bank susu collectors tend to mobilize funds on a daily basis consistent with their independent counterparts, they do not retain commissions from customers, being bank employees. These susu collectors may access clients in different towns or rural areas on foot or using public transportation. Before outlining our conceptual framework of savings inherently signaling creditworthiness, we motivate the need for credit in Ghanaian ﬁnancial markets.

After saving with a collector for a minimum of three months, a client may apply for a lump sum susu loan. A borrowing client must repay the amount (with a fixed rate of interest of ) in full six months after the date of approval. Our main argument, that there is a significant financial need for funds beyond independent saving thresholds is briefly motivated before we present our conceptual framework. The difficulties inherent in mobilizing savings in Africa and other regions of the developing world are summarized in Ashraf Karlan and Yin (2006a). Although a significant proportion of Ghanaians

10Act 29 of the Companies Code 63. 147 mobilize savings in the informal sector, investments and scaling businesses remain a challenge.

Consider an especially entrepreneurial migrant in peri-urban Ghana who has, over time, invested signiﬁcant proportions of personal savings into renting an investment.

This investment is a used portable photocopying machine (whose wholesale price is the equivalent of around 1000 Ghana cedis or US $1000). As is often the case, our entrepreneur makes impromptu copies for students who cannot afford to buy whole books, or (much less commonly) social science researchers in the ﬁeld. While the entrepreneur from our illustration may use their scarce savings to rent a machine from a relatively well-to-do businessperson, some difﬁculties would prevent them from investing in a machine of their own. The scenario is true of several investments of minimum scale. For example, the poorest women entrepreneurs (who carry wares on their heads in markets) may otherwise be unable to invest in market desks (costing the equivalent of $500). At the next “rung”, marketwomen who currently have desks, may be unable to invest in kiosks (also costing about $1000) with their current savings.

A signiﬁcant amount of credit would be needed to secure investments in every case mentioned. The above (and other) projects would require a lump-sum that exceeds the amount the entrepreneur in question can currently save for. A savings rate of around

8 Ghana cedis per period (the average in our sample) would mean a savings period of almost 5 years to reach the investment target of a desk in the above example of an itinerant entrepreneur. The lowest savings in the data are about 3 Ghana pesewas

(about 2 US cents). Therefore, for some individuals, this scenario may be similar to a temporary savings poverty trap situation, where consistently low savings may delay improving economic outcomes.

Bouts of inflation are another concern for Ghanaian entrepreneurs11. Given that significant periods of time are needed to accrue savings, inflation rates (which are rela-

11According to the International Monetary Fund and the World Bank, inflation in Ghana was 11.5% on average between 2006-2010, down from a higher average of 20.4% between 2001-2005 (World Bank 2012). 148 tively high in developing countries) may artificially drive up the price of an investment over time. These fluctuations may continually postpone a date of attaining an investment. Within this context, credit would be necessary to support savings and act as a hedge against inflation. The credit outcomes in the data range from 17 Ghana cedis to

2,000 Ghana cedis, with an average of 583 Ghana cedis; much larger than the personal savings (less than 10 Ghana cedis in the data). More than half (52% of susu clients) surveyed applied for a susu loan in the month of the survey. The next section presents a model whereby clients indicate their creditworthiness to susu collectors by making savings contributions.

5.3 A Model of Susu Savings and Creditworthiness Sig-

naling in Ghana

I present the fundamental dichotomous framework of signaling as introduced by Spence

(1973, 1974a, 1974b) and expounded by Mas-Collel, Whinston and Green (1995) in the ﬁrst sub-section. I develop a model with multiple degrees of signaling effort in the following sub-section.

An agent may reliably signal their creditworthiness indirectly, by submitting to a relatively costless test that requires effort. The understanding is that in any sub- game perfect Nash equilibrium, all workers whose creditworthiness exceed an arbitrary threshold in quality will yield to this test. The test in our setting refers to regular savings contributions made to a deposit collector. These amounts credibly signal the type of businessperson to a deposit collector, as explained in the present framework. 149

5.3.1 The Basic Setup

Consider two types of entrepreneurs with creditworthiness QH and QL, such that QH >

QL > 0 and λ = Prob (Q = QH )∈ (0,1). Before applying from a loan from a rural bank representative (a susu deposit collector) an entrepreneur can make some susu contributions, facilitated by the deposit collectors. These quantities (s) are observable to susu collectors. As noted in our earlier discussion, an individual’s own savings are not as economically signiﬁcant as a lump sum that can be gained via credit.

The (opportunity and psychological) cost of contributing these amounts to a susu collector for a creditworthy of type Q entrepreneur is given by the twice continu- ously differentiable function c(s,Q), so that its derivatives are as follows: c(0,Q) = 0, cs (s,Q) > 0, css (s,Q) > 0, cQ (s,Q) < 0 for all s> 0, and ssQ (s,Q) < 0. This means that the cost and the marginal cost of susu contributions are assumed to be lower for high-ability entrepreneurs; in applying this to susu lending, the contribution of saving required for a rural bank loan would easier for a highly-creditworthy individual. These clients who are able to make consistent savings (given the difﬁculties noted above) should be able to make regular credit payments as well.

We can allow u(l,s | Q) to denote the utility of an entrepreneur of creditworthiness

Q who chooses susu contribution level s and receives loan l. Thus, u(l,s | Q) = l − c(s,Q): the utility is simpliﬁed as the difference between the loan given and the cost of contributions made before it was received. An entrepreneur of type Q may earn r(Q) by working with informal susu collectors, who mobilize savings funds but do not provide loans. Alternatively, one can think of this construct as independent savings.

For simplicity, we assume throughout this section that r(QH ) = r (QL) = 0. The basic susu contribution signaling game is summarized below.

1. A random move of nature determines the entrepreneurial creditworthiness.

2. Conditional on her type (high or low creditworthiness), the entrepreneur chooses 150

how much susu contributions to make (we simplify this decision in terms of

effort the individual exerts).

3. Once the contribution level is chosen, the entrepreneur enters the market for

loans.

4. Conditional on the observed susu contributions, deposit collectors simultane-

ously make loan offers.

Equilibrium

I use the Perfect Bayesian Equilibrium. This can be deﬁned as a set of strategies and a belief function ν(s) ∈ [0,1] giving the institutions’ common probability assessment that the entrepreneur is of high creditworthiness after observing susu contribution level s. The requirements for a perfect Bayesian equilibrium are as follows:

• The entrepreneur’s strategy is optimal given the deposit collector’s strategy.

• The belief function ν(s) is derived from the entrepreneur’s strategy using Bayes’ rule where possible.

• The institutions’ loan offers following each choice constitute a Nash equilibrium

of the simultaneous-move loan offer game in which the probability that the en-

trepreneur is of high creditworthiness is ν(s).

We initiate our discussion from the end of the game. Suppose that after observing some contribution level s, the institutions attach a probability of ν(s) that the entrepreneur is of type QH . If so, the expected creditworthiness of the entrepreneur is ν(s)QH +

(1 − ν(s))QL. In a simultaneous-move loan offer game, the collectors’ (pure strategy) Nash equilibrium loan offers equal the entrepreneur’s expected creditworthiness. 151

Separating Equilibria and Susu Contributions

Suppose that s∗ (Q) be the entrepreneur’s equilibrium susu contribution choice as a function of her creditworthiness, and let l∗ (s) be the economic institution’s equilibrium loan offer as a function of the entrepreneur’s susu contribution level. Proposition 1

In any separating perfect Bayesian equilibrium, each entrepreneur type receives a

∗ ∗ ∗ ∗ loan equal to their creditworthiness level: l (s (QH )) = QH and l (s (QL)) = QL Proof: In any perfect Bayesian equilibrium, beliefs are correctly derived from the equilibrium strategies using Bayes’ rule to be on the equilibrium path. Thus, on

∗ seeing susu contribution level s (QL), institutions must assign probability one to the entrepreneur having creditworthiness of QL. Likewise, institutions assign probability one that an entrepreneur has creditworthiness QH upon seeing susu contribution level

∗ s (QH ). Note that the corresponding loans reﬂect QL and QH , respectively. Lemma 1

∗ In any separating Bayesian equilibrium, s (QL) = 0; that is, a entrepreneur having low levels of creditworthiness chooses to make no susu contributions.

Proof: This is proved by contradiction. Suppose the opposite scenario is true: that when a worker is type QL, she chooses some strictly positive savings levels ˆ > 0.

According to proposition 1, by doing so, the entrepreneur receives a loan equal to QL.

However, she would receive a credit outcome of at least QL if she instead chose s = 0. Choosing s = 0 would save her the cost of susu contributions, and therefore she would be strictly better off by doing so. This contradictions the assumption thats ˆ > 0 is her equilibrium level.

From the above discussion, type QL’s indifference curve through her equilibrium level of susu contributions and loans is shown in Figure 1 in a separating equilibrium. Using the diagram, we can construct a separating equilibrium as follows: Let 152

∗ ∗ ∗ s (QH ) =s˜, let s (QL) = 0, and we present the schedule for l (s). The economic insti-

∗ ∗ tutions’ equilbrium beliefs following susu contribution choice s are µ (e) = (l (s) − QL)/(QH − QL).

∗ Since l (e) ∈ [QL,QH ], they satisfy µ ∗ (e) ∈ [0,1] for all s ≥ 0. To verify that this is a perfect Bayesian equilibrium, institutions may have beliefs where s is neither 0 nors ˜. However, it must be the case that µ(0) = 0 and µ(1) = 1.

∗ ∗ ∗ The loan offers drawn, which have l (0) = QL, and l (s ) = QH reﬂect exactly these beliefs. Entrepreneur’s Strategy

The equilibrium path allows for many loan schedules that can arise to support susu contribution choices. In the below perfect Bayesian equilibrium, institutions believe that the entrepreneur is certain to be of high type if s ≥ s˜ and certain to be of a low

∗ ∗ ∗ ∗ type if s

The susu collectors employed by rural banks observe varying deposit mobilization levels. Rural banks’ susu collectors believe that an entrepreneur is certain to be of high type if s ≥ s˜ and certain to be of a low type if s

Entrepreneurs worthy of high credit are willing to engage in susu contributions simply because it allows them to distinguish themselves from entrepreneurs worthy of low credit and receive higher loans. The basic reason why these contributions can serve as a signal here is that the marginal cost of susu contributions depends on the type of creditworthiness (high or low) an entrepreneur merits. Because the marginal cost of contribution is higher for an entrepreneur worthy of low credit, since (csQ (s,Q) < 0), 0 a type QH entrepreneur may ﬁnd it worthwhile to contribute some positive level s > 0 to raise her loans by some amount 4l > 0, whereas a type QL entrepreneur would be unwilling to make these same levels of contributions in return for the same loan increase. As a result, institutions can reasonably come to regard contribution levels as 153 signals of entrepreneur creditworthiness.

We note that the contribution level of the high-ability entrepreneur cannot be below s˜ in a separating equilibrium because if it were, the low-ability entrepreneur would deviate and pretend to be highly creditworthy by choosing the high-credit contribution level. On the other hand, the contribution level of the high-credit entrepreneur cannot be above s1 because if it were, the high-credit entrepreneur would prefer to contribute no funds, even if this resulted in her being perceived as of low creditworthiness. Figure 5.1: Susu Savings and Creditworthiness Signaling 154

~~Kakum Rural Bank and the Set-up of Multiple Susu Savings Sched- ules~~

~~The case that clients are motivated to mobilize savings (and hence credit) to different degrees derived from discussions with collectors employed by Kakum Rural Bank~~

~~(established in February 1980) in the Central Region of Ghana. The Kakum Enyidado~~

Susu Scheme is used to provide credit to individuals who make daily, weekly, or fortnightly savings with the bank. A monthly schedule has also been instituted to allow savings mobilization every thirty days. While monthly savings tend to be limited to individuals earning monthly paychecks, some relatively poorer clients use the schedule as well, citing its relative convenience12.

A twice-weekly schedule was developed by Kakum Rural Bank in conjunction with the Microsfere Fund for People and Nature NGO in the Central Region, and received by the populations in tandem with the other schedules by early 200913. The establishment of protected areas throughout the region has simultaneously increased tourism-related revenues, controlled soil erosion, and dwindled economic livelihoods once reliant on land degradation. The twice-weekly schedule was established and promoted at the suggestion of the NGO throughout Central Ghana as a savings commitment device to encourage alternative and sustainable livelihoods, such as micro-enterprise (Microsfere

~~Fund for Nature 2011).~~

Susu collectors frequent the outdoor markets and other areas of commerce, often serving about 300 clients every day. The surveyed collectors reported serving an average of around 200 on a daily basis. Each served client is on one of the schedules mentioned above. If either the collector or the client feels the current schedule is not 12Only 17 (out of 384) clients use the monthly schedule in the data. This schedule is dropped from our analysis for relatively insignificant representation, although the descriptive statistics are comparable to other schedules (see Table 1). 13Interconnections between rural banks, susu collectors and microfinance NGOs are increasingly common in Ghana, although implementation (and the motives influencing it) tend to be led by rural bank officials. Field research by collectors or NGOs working in certain areas may lead to the introduction of new financial products thought to be helpful to the entire population, as in our case. 155 satisfactory, the client may be moved to another schedule on a case-by-case basis. In discussions and interviews, this seems to be relatively uncommon, although we are unable to explicitly verify any schedule switching in the data. For our comparison groups however, the number and proportions of missed savings payments is therefore a potentially important covariates as are pre-loan demographic factors) owing to this concern.

A related concern is that the number of missed payments may be endogenous to a particular schedule choice: our understanding is that the client considers their own behavior (as far as missing savings commitments is concerned) before choosing a treatment (savings schedule). However, a person who picks a daily schedule could miss more chances to save per month than a person who picks a weekly schedule. While our model assumes an unusually high level of self-awareness in treatment choice, we believe that this assumption is actually consistent with the absence of schedule-switching in the data, as noted above. If the schedule choice inﬂuenced the number of missed payments a priori, we would see at least some schedule-switching, which we do not observe. We extend our basic framework to account for multiple levels of effort exertion and savings. 5.3.2 Model Environment given Several Levels of Signaling Efforts

~~The basic framework above presumes that the signaling effort of savings is ﬁxed.~~

This may be because the discussion only slightly varies the degree of effort inherent in savings and thus, the intensity of signaling. We therefore extend the levels of signaling efforts by considering more than two frequencies of savings contributions in this section. We assume a strictly positive relationship between effort exertion and signaling precision of creditworthiness.

~~To investigate the relative strength of savings signals, we introduce a formal-informal~~

ﬁnancial market where the interaction of clients (and thus collectors) may or may not deviate from prescribed actions. (The implausibility of deviation in the basic scenario 156 is discussed in Figure 2b of the previous section.) Because the credit outcomes depend on the underlying fundamentals, the possibility of deviation will convey information that is useful in the coordination game. We begin by providing the setup, constructing separating equilibria, before offering our hypotheses. Setup

As before, a privately informed client’s choice of action (on the choice of savings) has two distinct effects on her credit payoff. The ﬁrst is the noninferential impact while the second is through the inferences the deposit collectors draw about the private information. In equilibrium, an informed client optimally chooses a level of savings, understanding that uninformed collectors will be taking optimal actions conditional on their inferences from that action; and these inferences will be correct. If these inferences leave no uncertainty about the saving client’s private information, then the equilibrium is separating for that agent. We generalize the above discussion to four levels of quality with respect to creditworthiness, QA where Ai = {a1,a2,a3,a4}, and

Qa1 > Qa2 > Qa3 > Qa4 > 0. These levels of creditworthiness are associated with savings schedules which are similarly decreasing in terms of efforts required, and hence, signal strength. Consistent with the above explanation, the corresponding loans or credit amounts are notationally l1,l2,l3,l4. We expound on our earlier discussion of saving effort, focusing on the multiple savings schedules discussed above. In this iteration of our model, we focus on the different savings schedules noted in the previous section. The daily schedule requires the most effort to commit to for two main reasons. The decision to adopt and utilize a daily schedule means that an entrepreneur must put aside funds aside every single day

(the smallest time unit possible) to meet savings obligations. In the survey, this is the highest savings frequency that can be chosen. A related point is the potential psychological cost or personal embarrassment of not fulﬁlling a chosen savings schedule, to 157 which a daily saver is the most exposed14. Therefore, the daily schedule requires more effort to commit to, relative to the weekly schedule for example. By committing to a daily savings timetable, an entrepreneur must put aside funds every day, whereas her counterpart on a weekly basis need do so every seven days15. As noted above, a client who chooses a daily schedule may be under more social pressure to contribute savings than a client on a more relaxed schedule.

We slightly extend our basic signaling game and suppress the subscripts for ease of explanation. We have two players which we call C (for client) and D (for deposit susu collector). C knows the value of a random variable q, whose support is a given

~~ﬁnite set Q, so that Q = {0,..N}. The prior beliefs of D are a probability distribution~~

~~π(.) over Q; these beliefs are common knowledge. Since Q is ﬁnite, π(q) is the prior probability that the client’s type is q.~~

~~1. Player C learns q, sends a savings signal s (drawn from a set S). This savings~~

~~signal is a savings schedule.~~

~~2. Player D receives the signal s, and then takes an action l (drawn from a set L).~~

~~We allow L to depend on s and C to depend on q, (to merely repeat our discussion with slightly more appropriate notation). Stage 2 ends the game: The payoff to i is given by~~

i u : Q ×S × L → R. The separating equilibrium discussed in the prior section is the benchmark for the signaling games discussed in this section. We adopt the suitable and convenient single- crossing condition introduced by Karlin (1968), and expanded by Athey (2001). This provision guarantees that indifference curves from a given family of preferences cross

14Note Bortei-Doku and Aryeetey (1995). 15To illustrate, individuals committed to the weekly schedules have smaller time constraints in contributing funds, relative to individuals committed to a daily schedule. At the extreme, weekly customer may contribute savings gained the day before the scheduled meeting with her susu collector. 158 at most once16. As noted before, we focus on classes of signaling games where a monotonic relationship exists between types and signals.

~~Deﬁnition 1:~~

~~Single Crossing Property: Let Q,S,L be real intervals.~~

~~ 0 0 0 UC (.) supports the single-crossing condition if UC (q,s,l) ≤UC q,s ,l =⇒UC q ,s,l ≤~~

~~ 0 0 0 0 UC q ,s ,l for all q > q.~~

~~In our scenario with saving clients and crediting deposit collectors, UC (.) is strictly decreasing in the second argument (the signal), and increasing in the third argument~~

~~(D’s loan response) for all types. Therefore, indifference curves are well-deﬁned in~~

S × L for all q. The single-crossing condition implies that the indifference curves of different client types will cross only once. If a lower-type is indifference between type signal-action pairs, then a higher type strictly prefers to send the higher signal. In this manner, the single crossing property links signals to types such that higher types send weakly higher signals in equilibrium. Construction of Separating Equilibria with Multiple Signals

~~We consider and construct separating equilibria for the multiple signals scenario, following our discussion in the previous section. Proposition 2a.~~

∗ ∗ The standard signaling game (li ,si ) has a separating equilibrium. Proof: Assume that the best response function of D be DR(q) be uniquely deﬁned, and strictly increasing in q. We will prove the proposition by constructing a possible equilibrium path and conﬁrming that this path is part of a separating equilibrium. We

16Mirlees (1976), Matthews and Moore (1987) offer scenarios where this may not be the case, while Mailath (1987) observes that it is almost impossible to verify that the single-crossing property will always hold. Although it should hold for a tractable subset of Q in our case, it is a suitable simplifying assumption since we have relatively few numbers and types of agents with ordered preferences (note Athey 2001). 159 walk through the following steps:

~~∗ C 1. q0 selects the signal s0 that maximizes U (q0,s,DR(q0)).~~

~~∗ ∗ C ∗ 2. Suppose that si have been speciﬁed for i =0,...,n−1 and let U (ti) =U (qi,s ,DR(qi)).~~

~~∗ 3. Deﬁne sn to solve the optimization problem:~~

C C Maximize U (qn,s,DR(qn)) subject to U (qn−1,s,DR(qn)) ≤ U (qn−1). Assume solutions for the maximization problems of Step 1 and Step 2. The process inductively produces a signaling strategy for the client and a response rule for the de-

∗ ∗ posit collector defined on s0,...,sN . When the DR function is strictly increasing, the single-crossing condition implies that the signaling strategy is strictly increasing. We complete the strategy by observing that in each game, the deposit collector takes the action DR(qn) in response to signals in the interval sn ≤ s < sn+1; the action DR(q0) for ∗ ∗ s < s0; and the action DR(qN) for s > sN. By the definition of the best response function, the deposit collector is best responding to the client’s strategy. This confirmes the existence of a separating equilibrium.

~~Proposition 2b.~~

In any separating perfect Bayesian equilibrium, each entrepreneur type receives a loan equal to their quality of creditworthiness. In particular, if we restrict creditworthiness Q to three savings schedules, the separating perfect Bayesian equilibrium implies that l ∗ (s ∗ (Qa1)) = Qa1, l ∗ (s ∗ (Qa2)) = Qa2, l ∗ (s ∗ (Qa3)) = Qa3 . Proof: In any perfect Bayesian equilibrium, the beliefs are correctly derived from the equilibrium strategies using Bayes’ rule to be on the equilibrium path. Thus, on observing the susu schedule corresponding to Qa1, the deposit collector assigns a probability of one to the client being type Qa1, and offers the corresponding loan. On 160 observing the schedule corresponding to Qa2, the deposit collector assigns a probability of one to the client being type Qa2, since that schedule requires more effort to adopt as a signal. As a result, the client receives the relevant loan. Finally, the deposit assigns a probability of 1 to the client being type Qa3 on observing the adoption of the relevant savings schedule. Following the discussion above, high-ability entrepreneurs are willing to adopt savings schedules which (to various degrees) require more effort to distinguish themselves from their low-ability counterparts. Once more, the marginal cost of contributions is a function of entrepreneurial type. The entrepreneurial type indicates creditworthiness to a deposit collector.

Contrary to our initial assumptions, no theoretical or empirical relationship may exist between savings schedule choice and credit outcomes. Also, we have thus far assumed that there may be no deviations whatsoever for any (arbitrary) savings candi- date. In particular, strategies motivated by deviations from the expected outcomes may inﬂuence credit outcomes, and are therefore important. We brieﬂy apply the arguments of Mailath (1989) and Angeletos et al (2006) to show that informed clients may deviate from prescribed actions. Finally, we note the possibility that the behavior of clients may be important for credit outcomes without a separating equilibrium being present.

~~These observations help motivate the hypotheses we will test in the data in Section 4. Deviations in Signaling and Separating Equilibria~~

Formally, we let the deviating client’s type remain q and let U (q,qˆ,sp) be her expected payoff when her true type is q. In this scenario, the uninformed deposit collector’s inference about the client’s type isq ˆ and sp is the resultant savings action. We assume that U is the reduced form payoff function incorporating the optimal actions of deposit collectors who make the subsequent move (with respect to their given beliefs).

We pick an arbitrary separating equilibrium and make the following adjustment to our prior discussion. In deviation, a client of type q deviates from her prescribed 161 action and plays sp, (which is associated with a different typeq ˆ). Rather than detecting the deviation, the deposit collectors may infer that the agent is of typeq ˆ, and make the corresponding loan. We assume that deviating clients choose to deviate to gain higher credit outcomes from collectors, than they would expect if they did not deviate.

~~However, deposit collectors may observe the deviation, and adjust their credit behavior.~~

We infer from Angeletos, Hellwig and Pavan (2006), that the presence of endogenous information (the clients are aware of varying loans) mean that the potential loans themselves yield information relevant for the coordination game. While the authors’ derivation of multiple equilibria is not directly important for our work, we simply repeat their observation that with bounded noise, agents (deposit collectors) may sometimes detect deviations (2006: 12). Within our own context, certain savings signals may be indicative of a deviation (from a clients’ actual creditworthiness) from the perspective of a susu deposit collector. However, deviations from expected actions may occur without separating equilibrium conditions, which may not hold in every empirical scenario. For this reason, we also consider deviations within a broader context of missed savings contributions in the next section. Hypothesized Impacts of Savings Schedules Adoption and Signaling on Credit

The above analyses imply that savings schedules may have impacts on credit outcomes. In addition, if savings schedules signal creditworthiness, the choice of a particular schedule will have distinct impacts. Hypothesis 1: Savings schedule choice matters (positively) for credit outcomes

Savings schedules’ adoption signals creditworthiness to a deposit collector, yielding credit outcomes via separating equilibria in the manner described above (as noted in Proposition 2a). The corresponding null hypothesis is that there are no credit impacts from adopting savings schedules. In this scenario, the candidates’ savings schedules do not matter at all for credit outcomes. If the clients’ signals from savings schedules do not establish credit outcomes, the assignment of credit will not vary by savings 162 schedule. Hypothesis 2: The effect of savings contributions on credit is strictly increasing in the effort required to make regular contributions

Given multiple available schedules, a client who adopts a schedule that requires more frequent savings contributions signals to their deposit collector that they are creditworthy (relative to other clients who adopt schedules that require less effort). For this reason, clients that adopt more effort-intensive savings schedules are hypothesized to consistently receive higher credit. Missed payments (within savings schedules) and creditworthiness (Detected Devi- ations in Signaling)

We have so far considered one measure of creditworthiness: signaling via the context of a separating equilibrium. In a separating equilibrium, creditworthy clients signal their creditworthiness through a equivalent savings schedule. In this sub-section, we focus on deviations from prescribed behavior. In particular, we focus on the regularity of actual savings, somewhat similar to a formal credit bureau: the rate at which clients are unable to make appointed savings. Missed savings contributions may be important for a client (assigned to a particular savings schedule) receiving a credit outcome different from what they would observe in a separating equilibrium. Hypothesis 3: Deviations (within savings schedules) matter (negatively) to credit provision.

Deviations in savings mobilization may be observed by a deposit collector, but may or may not be apparent to the researcher. One aspect of deviations from prescribed savings behavior within a savings schedule that would be apparent to both deposit collectors and the author are the rate of missed savings payments within a savings schedule. We therefore hypothesize that the effect of savings contributions on credit occurs through missed payments of savings within adopted savings schedule(s). In the 163 scenario of a separating equilibrium, we assume that savers within a particular schedule miss savings contributions within similar bounds. However, beyond the confines of a separating equilibrium, significant missed payments (within a chosen savings schedule) may lead to deviations in creditworthiness, and hence credit outcomes. Missed savings contributions may be important for a client (assigned to a particular savings schedule) receiving a credit outcome different from what they would observe in a separating equilibrium. The null hypothesis is that signaling (as discussed above) is sufficient to explain credit impacts from savings schedules.

~~We next discuss the empirics, and revisit the hypotheses in more detail after de- scribing the data and empirical strategy.~~

~~5.4 Data Description and Estimation Strategies~~

~~5.4.1 Data Description~~

~~The data used in the analysis were collected betweem June 2010 and September~~

2010 in the Central Region of Ghana. The survey was undertaken in twenty-seven urban, peri-urban, semi-rural and rural areas. At 9,826 square kilometres or 4.1 per cent of Ghana’s land area, the Central Region is the third smallest in the nation. Although the capital Cape Coast was the initial seat of government during the British colonial era, the national capital was moved to Accra in 1877. The Central Region has had an incidence of poverty higher than the national average (International Labor Organization

~~2004).~~

Ghanaian women have traditionally occupied a key position alongside men in the production of goods and services for the markets. Owned, rented, or inherited market stalls (or kiosks, or smaller desks) often delineate individual business spaces within a market area. Market sections are typically organized by the type of commodity sold

(fruits; vegetables; fish; meat; consumer goods such as cell-phones; herbal medicines; and several others). These arrangements allow traders to coordinate local prices or 164 supply from providers who may be based beyond the boundaries of their region. Stalls near a market entrance may attract a significant amount of customer traffic owing to convenience, conditional on the level of expected market activity on a given time of day. Stalls of concrete have been the most attractive since they can withstand the rainy season and protect most goods, although wooden and metallic stalls are increasingly common. Recent migrant workers (in both rural and urban cases) may be more likely to sleep at their place of business to accumulate more customers earlier the next day.

The relatively brief three-month survey period proved adequate partly because deposit collectors typically collect data and organize sustained engagements with hundreds of customers on a regular basis. By having susu collectors conduct surveys and interviews, I was able to gather data on personal ﬁnancial and other outcomes. Iterat- ing the survey on various occasions allowed its length to expand into larger versions.

~~The length of the survey allowed the collection of various data related to economic and other behavior.~~

~~We interviewed 384 individuals all living and working within the Central Region.~~

The initial phase involved the interviewing of 15 Kakum deposit collectors, and administering a comprehensive survey to each of them. After extensive discussions with the susu deposit collectors, we decided the order of survey implementation should follow the quasi-random geographically-based routes collectors may use while mobilizing funds across market areas. We now provide some description of our data, as well as our general estimation strategy in the next section, while we apply it in the subsequent one. Descriptive statistics are in Table 5.1. 165 166

~~5.4.2 Estimation Strategy~~

Table 5.1 shows summary statistics for the individuals who adopted the daily, twice- weekly, weekly, fortnightly, and monthly schedules. The client characteristics (age, gender, entrepreneur, household head, income, and the number of missed payments) are broadly similar across the various schedules. The credit outcomes ﬂuctuate both intuitively (the daily savers receive the largest amount), and counterintuitively (the monthly savers receive more than individuals who save twice every week for example).

The savings amounts similarly ﬂuctuate, and given the proposed signaling theory, it is possible that the savings timetables or schedules may yield even more consistent information than actual savings amounts. The monthly schedules are never used in the paper, due to relatively low representation in the sample (Nmonthly = 17). I now present the basic estimation strategy.

Our credit observations (dependent variable) refers to the amount offerred and accepted, which is some equilibrium between supply and demand. Given the discussion in Section 1 (showing the context in which we assume that demand for credit is high), we consider the credit offer observed to be more about supply than demand. We brieﬂy outline the basic strategy in this section before applying the strategy to particular methodologies in the next section.

~~We estimate the basic equation:~~

~~L = α + γkTk + θD + Xβ + ε~~

~~where L is the credit amount, Tk is an adopted saving schedule, and D is the frequency of past missed appointments to save. The Hypotheses (from the previous section) are:~~

~~H1 : γk > 0, for all k~~

~~H2 : γ1 > γ2 > γ3 > γ4.~~

~~H3 : θ < 0. 167~~

The error term, ε, is the residual capturing omitted variables as well as measurement error. Deviations captured within this error term are correlated with the rate of missed payment D variable. However, it is possible that this may not account for all deviations of savings behavior and credit impacts. Aspects of deviations uncorrelated with D would appear in ε, and if sufﬁciently large, we may reject H1, even where a separating equilibrium exists. Where deviations are not correlated with D, they may be observed by collectors, but not the researcher. Given the diverse explanations we have presented

~~(separating equilibria and missed payments), aspects of deviations uncorrelated with~~

D would explain situations where we have counterintuitive signaling behavior. For example, a savings schedule that represents more effort yields less credit than a savings timetable that requires less effort.

X is a set of client characteristics, consisting of age, gender, entrepreneur, household head, income, number of missed payments. We very brieﬂy outline some hypotheses below, relating to how each independent variable should affect creditworthiness:

~~• AGE. Younger entrepreneurs should be more economically active, and thus cred-~~

~~itworthy.~~

~~• GENDER. Female entrepreneurs may be more creditworthy, since women are~~

~~relatively economically active in markets (as explained in previously).~~

~~• ENTREPRENEUR. Sole-proprietors should be economically active, and credit-~~

~~worthy.~~

~~• HOUSEHOLDHEAD. Household heads (breadwinners) should be economically~~

~~active and creditworthy.~~

~~• INCOMES. Individuals who are more productive should earn more and be more~~

~~creditworthy.~~

~~• MISSED SAVINGS PAYMENTS. Individuals who missed more savings payments 168~~

~~in the past should be less creditworthy.~~

~~Table 5.2 shows OLS estimates, which are corroborated with Tobit estimates (in table~~

5.3). The covariates are dummy variables for age, gender, entrepreneur (sole propri- etor), income, household head and number of pre-loan missed savings (as motivated in the previous section). The savings schedules are our treatment variables. In our main regression, we ﬁnd that the daily savers received the most credit, followed by the weekly schedule. Overall, these estimates show that the amount of credit is generally increasing in the effort exerted in signaling. The past missed savings payment variable never yields signiﬁcant effects on credit outcomes. Although our study is not a controlled experiment, our initial results show seemingly close to random assignment that we believe is actually related to the unobserved propensity to exert efforts in saving contributions. 169 170 171

~~Propensity Score Estimation Methodologies~~

Although our OLS results (note column 1 of table 5.2) nearly mimic a random assignment, people who exert more effort in saving may not always be a random sample of individuals due to self-selection: they may have had lower propensities to be poor or miss daily savings payments in the past, higher propensities to shoulder responsibil- ities such as household heads, or showed initiative in areas such as entrepreneurship.

In this section, we consider that schedule choice may be quasi-random and conditional on observable factors inﬂuencing it. Propensity score matching limits the possible confounding effects in quasi-experimental scenarios such as the one discussed here, allowing differences of responses to be attributed strictly to the differences of treatments

~~(Rosenbaum and Rubin 1983). This is done by yielding a credible comparison group out of a generic control group. We outline our methodology in this section.~~

The basic matching exercise is driven as follows. Relevant differences between both treatment and control groups are adequately captured by the observables X. To account for possible confounding factors. a comparison group is assembled from the non- treated (control) group. For this comparison group the distribution of observed variables must be as similar as possible to the distribution in the group receiving treatment.

~~Rosenbaum and Rubin (1983) proves that matching treated and control groups using~~

X corresponds to matching them using a propensity score p(x). The basic propensity score yields the conditional probability of receiving treatment given X. We use a logit model to estimate every propensity score using a logit model of the probability that an individual entered a particular savings schedule as a function of relevant demographic and economic factors.

We use two kinds of propensity score balancing that are substantially different: (1) one-to-one matching with replacement and (2) kernel matching using a normal density function (note Caliendo and Kopeinig 2008 for details). We do not use other versions 172 of propensity score matching17. Matching protocols are presented in the Dissertation

~~Annex. One-to-One Matching (With Replacement)~~

This technique matches treated individuals with comparison individuals based on the similarity of their propensity score. Mapping with replacement means that the average quality of matches increases and bias lessens, since an untreated individual can be used more than once in creating a match 18. The difference in outcomes for each matched pair is calculated. Covariates are reliably balanced since comparing outcomes for individuals with similar propensity scores facilitates the comparison of outcomes of individuals with similar covariates. For each treatment savings schedule (relative to a comparison group), the matching procedure was derived from the logit regression used to balance the two subgroups, and the average treatment effects are restricted to the region of common support.

~~The one-to-one matching estimator is M, such that for individuals i with propensity score pi and individuals j with propensity score p j,~~

~~0 M (pi) = j :| pi − p j |= mink∈{T=0} {| pi − pk |}~~

~~ωik = 1(k = j)~~

~~Kernel Matching (With Normal Density Function)~~

~~This method yields a comparison group by using weighted averages of almost all individuals in the control group to create the comparison group. Kernel-matching esti-~~

17We do not use alternative methods such as subclassiﬁcation since this technique represents a special case of weighting estimators (Zhao 2004), of which kernel matching is a type. The one-to-one matching exercise is similar to Dehejia and Wahba (2002) as well as a sub-study of Nunn (2007). Similarly, we do not explore Mahalanobis-metric mapping (as done by Rubin 1980 or Lechner, Miquel and Wunsch 2011) since this method yields similar results (note Zhao 2004 and Michalopoulos, Bloom and Hill 2004). 18See Caliendo and Kopening (2008) for details of the advantages of one-to-one matching with replacement, relative to the same method without replacement. The difference between these two sub-techniques is not signiﬁcant overall (Zhao 2004). 173 mates derive from a weighted regression of the counterfactual outcome on an intercept with weights given by the kernel weights (Smith and Todd 2005). The weights depend on the distance between each individual in the control group and the participant observation for which the counterfactual is estimated.

We use weights from a normal kernel, and derive our matches from a logit regression used to balance the subgroups, limiting average treatment effects to the region of common support. An advantage of kernel matching is a lower variance since relatively more data is used in the estimation process. A demerit is the possibility of adverse matches, although this is rectiﬁed by imposing the common support, consistent with the applied literature.

~~Since there is no strict rule for kernel bandwidth selection, we use the default kernel bandwidth of b =0.06 as provided by Leuven and Sianesi (2003). 19~~

For credit outcomes li of individual i, who receive treatment T, we are interested in matched outcomes gained by a kernel-weighted average of credit outcomes for all comparison individuals. Our kernel is normal, N, and pi and p j are propensity scores for i and j respectively, while b is the (default) bandwidth. The weight given to comparison individual j is a percentage of the closeness between i and j,so that:

~~ pi−p j ∑ j∈{T=0} N b li lˆi = pi−p j ∑ j∈{T=0} N b~~

~~The weights of the outcomes for control j is given by ψ where:~~

~~ pi−p j N b ψ = pi−p j ∑ j∈{T=0} N b~~

~~19(Caliendo and Kopening (2008) note a trade-off between high bandwidth values (which correspond to better ﬁt and lower variance between the estimated and true density functions), and bias. 174~~

~~Main Results~~

~~Comparisons with Multiple Alternative Schedules as Controls: Kernel Matching and One-to-One Matching With Replacement~~

Tables 5.4 and 5.5 present the average treatment effects (differences) for the treatment schedules relative to comparison groups. For each savings schedule, we present all other savings schedules as control groups, with the comparison groups gained through the propensity score matching techniques discussed above. These results provide indi- cations of the savings efforts shown by treatment groups relative to controls. In table

~~5.4, the daily schedule is shown to have a large signiﬁcant credit impact (approaching~~

~~300 Ghana cedis). None of the other schedules yield signiﬁcant credit impacts (relative to all alternatives) when we use nearest neighbor (one-to-one) matching. 175 176~~

When we use kernel matching (in table 5.5), we find that individuals who exert more effort in signaling their ability to deposit collectors receive much more credit on average (exceeding 300 Ghana cedis). More relaxed schedules yield progressively less significant impacts on credit. Saving on a twice-weekly schedule had a less significant negative credit impact relative to all other schedules. Saving on a fortnightly schedule had the most negative credit impact (almost 200 Ghana cedis). However, individuals who use weekly savings schedules do not show statistically significant credit impacts. 177 178

~~Comparisons with Single Alternative Schedules as Comparison Groups: Kernel and One-to-One Matching With Replacement~~

In this subsection, I present average treatment effects of treatment schedules relative to single alternative schedules. The results are generally consistent with the prior discussion on using multiple alternative schedules as comparison groups.

I ﬁnd that table 5.6 (using one-to-one matching) yields generally the same outcomes to table 5.4. Each cell in table 5.6 shows average treatment effects of various savings schedules on credit. Each entry represents a seperate propensity score match using one-to-one matching with replacement. Committing to the most effort-intensive (daily) schedule has a large and signiﬁcant credit effect (more than 400 Ghana ceds) relative to the least effort-intensive (fortnightly) schedule. Using the twice-weekly schedule

(instead of the daily schedule) leads to a very significant credit decrease. On the other hand, the twice-weekly schedule has significantly better credit outcomes relative to the fortnightly schedule. Although the twice-weekly schedule has negative credit outcomes relative to the weekly schedule, the average treatment effect is not significant. The least effort-intensive fortnightly schedule has large and significantly negative effects relative to the daily schedule. 179 180

~~Table 5.7 shows average treatment effects of various savings schedules on credit~~

(with each savings schedule serving as a comparison group). Each entry represents a seperate propensity score match using kernel matching with a normal density function. Using the daily schedule has a positive effect that is generally increasing as the effort-intensity of the comparison savings schedule decreases. The daily schedule has a positive credit effect (of about 280) relative to the weekly schedule. The credit effect of the daily schedule is more pronounced relative to the fortnightly schedule. Similarly, the twice-weekly schedule has a significantly negative credit effect relative to the daily schedule, while the weekly schedule has a positive and significant credit effect relative to the fortnightly schedule. Finally, the fortnightly schedule has a negatively significant average treatment effect relative to the daily schedule, although the effects relative to other comparison schedules are not significant. 181 182

~~Robustness Checks~~

The effects of the weekly schedule do not appear to be significantly different from the alternative schedules. However, a departure in the findings (in table 5.6) is that the weekly schedule has significantly better credit outcomes than the twice-weekly schedule, despite our initial theory that it represents a signal of weaker effort. We observed that the twice-weekly schedule was instituted with an NGO. For this reason alone, it is plausible that the schedule would not yield the same credit result.

However, a plausible explanation is that the deviations (of weekly savers) are correlated with the missed payments term. If this were the case, we would expect deviations from prescribed behavior to be observed both by deposit collectors and ourselves in the data. As noted earlier, if this aspect of missed payments is signiﬁcant, it would manifest in some cases whereby savings schedules requiring less effort (e.g. weekly) receiving more credit than schedules requiring more effort (twice-weekly in this case). We ran separate propensity scores where we considered the treatment to be having missed at least savings payments while in their savings schedule (so that comparison groups were yielded from individuals never missed a savings payments). The results are columns

~~1-2 in table 5.8. The impacts on credit outcomes were negative, but not signiﬁcant.~~

~~Thus, we feel comfortable asserting that deviations from prescribed actions were generally uncorrelated with missed payments, and unobserved by the authors where they occurred. 183 184~~

Thus far, we have considered credit availability from deposit collectors to be significantly affected by savings mobilization. An alternative explanation may be that credit instead affects savings. The argument is as follows. If individuals mobilize more or less savings on receiving credit from a collector, we would observe a signiﬁcant credit impact on savings. However, the results we already have should lower concerns of reverse causality. The treatment variables we use, savings schedules are associated with

ﬁnancial effort and self-selection which should be less correlated with substituting savings with received credit. To check for robustness, we examine the possibility that the parameters of interest (savings schedules) are correlated with the error term once we control for observables in our matching algorithms. Since we already have controls which did not receive any credit, we are able to test whether reverse causality is at play: that is, whether receiving credit affects savings outcomes. We conduct similar propensity score matching exercises to our approach above, with the results are shown in table 5.9. We use one-to-one matching without replacement and kernel matching with a normal density function in the ﬁrst and second columns respectively. We did not

~~ﬁnd that credit receipt to signiﬁcantly affect savings contributions. 185 186 5.5 Conclusions~~

This paper studies an informal ﬁnancial arrangement (susu deposit collection) which has been adopted by the formal ﬁnancial economy. Contributions of the paper include a model with new data in a scenario where the savings of clients indicate creditworthiness to their deposit collectors. We develop a framework for analyzing varying degrees of signaling effort. The model predicts that savings schedules requiring more effort signal greater creditworthiness, and that this signal should lead to higher offers of credit to people who make stronger signals.

The savings schedules used to signal entrepreneurial ability and creditworthiness in the Central Region of Ghana provides a unique quasi-experimental ﬁeld environment. This study has predicted and estimated the corresponding credit gains of savings schedules which vary in savings frequency, and hence, the effort present in a particular signal. We estimate that the most effort-requiring savings schedule–where individuals save on a daily basis–to consistently result in higher credit outcomes, relative to the other schedules, which require less effort. This is in broad agreement with the theory.

~~We also motivate a scenario where deviations in prescribed behavior may be observed~~

~~(via missed savings appointments) or unobserved, ﬁnding in one instance that not all deviations may be observed by collectors.~~

Although this work is a first step in modeling and evaluating the factors influencing microcredit within the context of merging formal and informal finance, additional research is necessary. Studying the issues motivating savings within this setting could exploit issues relevant to behaviorial microeconomics. For example, intrinsic and ex- trinsic motivations and incentives would permit the study of some issues ignored here such as the tension between inadequate self-control and setting of personal rules20. For example, I found the (external) commitment savings device of susu collection, and the

20The literature on self-control (or the lack thereof) stems from Strotz (1956), Anslie (2001) O’Donoghue and Rabin (1999), Ashraf, Karlan and Yin (2006b) to mention only a few. A summary of the personal rules literature is summarized in Benabou and Tirole (2003, 2004). 187

~~(internal) setting of savings schedules to mutually reinforce savings mobilization in parallel work (see Chapter Four).~~

Susu collection and mobilization is a complicated process, and harmonizing this institution with the formal ﬁnance is at least as challenging. The savings contributions of themselves are as useful a signal to deposit collectors (and their employer rural banks) as are more precise details, such as the performance of savings schedules relative to other competing savings timetables. It is interesting that clients’ economic behavior and their unobservable and observable characteristics are broadly consistent within the framework and empirics. The alternative situation presented (investigating deviations in prescribed behavior) was only slightly important in the results. The ﬁndings are also robust to reverse-causality concerns.

From a policy perspective, it is remarkable that the daily schedule (which is the traditional schedule), seems to be the most useful signal. Since the normal practice is the best practice for loans, a greater integration of banks with susu could have great beneﬁts for susu clients without the clients engaging in signiﬁcant behavioral change. 188

~~Chapter 6~~

~~Women Empowerment, Gender~~

~~Bias and Susu Collection in~~

~~Ghana~~

~~6.1 Introduction~~

Reading the scribbled words, my heart jerked as if hit by a lightning bolt. The note showed my salary, listed next to the three male managers’ salaries: I was earning $44,724 while the highest-paid man earned $59,028 and the other two followed close behind, earning $58,464 and $58,226. Maybe I was seeing things. Maybe this note was a serious mistake or a bad joke, though I knew in my gut it wasn’t.

~~–Lilly Ledbetter, quoted in Why I ﬁght for equal pay for women (Ledbetter~~

~~and Isom, 2012).~~

What are the mechanisms and consequences of gender-discrimination? Theoretical and empirical studies of gender-based prejudice as a social norm tend to be based 189 on an agent (the recipient of discrimination) subject to the biases of a principal (the discriminator). The assumption is that an agent is embedded in a “one-sided” setting of bias. This focus is often appropriate because of our research and policy interest in the causal agents of bias and discrimination. However, women who encounter bias in different walks of life are not necessarily passive in the face of bias from males (or other females). Expected female discrimination (the anticipation of gender-bias against women) may lead to economic behavior that is reactional. For example, women work more than men do for equal compensation in formal work—perhaps when the incidence of gender bias would not surprise them. Therefore, there are reasons to believe that bias occurs in “two-sided” (or in network terminology, undirected) environments in formalized markets (such as the one in represented in the above quote).

However, understanding the mechanisms and consequences of gender-bias in the informal economic sector is arguably even more important. Gender bias in this area has implications for the women who dominate economic activity in developing countries.

Although understudied, informal ﬁnance is an important avenue for discussing gender dynamics (e.g. Ardener and Burman, 1995). For example, gender-bias persists in traditional and informal economic institutions with signiﬁcant impacts on female-headed businesses (see for e.g., Field, Jayachandran and Pande (2010) for a study of the Hindu caste system’s gender-biased economic constraints in India). However, institutional change may have implications on gender-bias as a norm, both within a gender, and across genders. No study, to my knowledge, empirically distinguishes between cross- gender discrimination and same-gender matching within such settings (for e.g. Eeckout and Munshi (2010) provide a discussion on group-based (and therefore gender-neutral) matching in Indian informal institutions).

Yet, there is an important distinction between cross-gender bias (e.g. males discriminating against females), and same-gender matching (homophily by gender (meaning people gravitating towards or being matched with others of the same gender). The 190 difference is that since homophily restricts to a specific gender, this concept of matched agents implies a weak notion of gender-bias (relative to discriminating against some of the opposite gender). The sole empirical focus on matching by gender has been on unilateral or directed gender-bias from principals to agents in microfinance (Beck, Behr and Madestam 2011). Yet, undirected networks1are significant in models of two-sided networks and their best-response processes (e.g. Golub and Jackson 2009). To my knowledge, no study has applied this theory to the phenomenon of merged formal and informal institutions. On the other hand, such financial arrangements have exploded in popularity in much of Africa since the 1990s (e.g. Steel, Aryeetey, Nissanke and

~~Hettige, 1997).~~

The chapter focuses on gender dynamics and institutional change in the Ghanaian susu ﬁnancial system. I ask two related questions: (i) is there gender bias or discrimination in susu credit outcomes? (ii) Does the gender of the susu collector create or mitigate biases in credit provision? I study a program that uses male and female savings deposit collectors (called susu collectors in Ghana) to merge formal and informal

ﬁnancial institutions. Since susu collection has historically been a male-dominated phenomenon (Aryeetey 1994), a new initiative for hiring female susu collectors makes this comparison of bias and own-gender-preferences possible.

~~If gender-bias is a function of both cross-gender bias and being matched with other agents of the same gender, then each individual perspective is by deﬁnition, limited.~~

~~However, both perspectives can be reconciled if we consider that gender-bias is an out of equilibrium belief (or an outcome that is not tied down by the deﬁnition of~~

Bayesian Nash equilibrium according to Cho and Kreps (1987). When no agent should conform to a norm, a Nash equilibrium allows that anything could be inferred about a person who does not conform. Although I do not explicitly model gender-bias in the present chapter, signaling (Spence 1973, 1974) provides a useful mechanism to

1I only study networks at the simplest level in this chapter, between two agents, or “nodes” that are joined by links. A network (or graph) is “undirected” if all links are bilateral, so that there is no order to the direction of a (unilateral) relationship between nodes. 191 explain observed gender-bias when rooted in same-gender matching based on social networks. This approach is particularly useful because there are prima facie reasons why gender-bias as well as reactions to such discrimination within networks are potentially important for our understanding of reverse gender-bias, or policy initiatives supporting female empowerment.

I use a signaling model reﬂecting that gender-bias represents an out of equilibrium belief as well as a matching framework to connect gender bias with gender-matched networks in the empirics. In the results, I therefore ﬁnd evidence of both cross-gender bias (against women) and homophily in gender-based matching of women and credit, although women provide less savings to collectors of the same gender. These results are in broad agreement with the signaling model showing that clients of the gender opposite to their collector may contribute more savings to signal their creditworthiness.

Comparative information on male and female agents interacting with male and female clients are relatively rare when the context represents the merging of formal and informal ﬁnancial institutions. Although data on informal ﬁnance is hard to access

(almost by definition), there are reasons why discrimination in informal systems may exceed formal institutions, such as amorphous legal boundaries. On the other hand, it is not obvious that gender-bias in the informal should disappear on contact with the formal sector (since gender discrimination is nearly universal). Although I cannot answer this question directly, a gender-based study on the Ghanaian susu collection can help further the discussion on gender and institutional adaptation in informal financial systems. The phenomenon of merging formal and informal finance within the susu institution has occurred in a milieu of significant institutional change.

Since susu involves both savings and credit, I can also test whether gender-bias is reciprocated, using the signaling model mentioned above. Own-preferences have been investigated in phenomena such as law enforcement (Donohue and Levitt 2001, Welch,

~~Combs and Gruhl 1988) as well as credit markets in the United States (e.g. Munnell, 192~~

~~Tootell, Browne and McEneaney 1996, Berkovec, Canner, Gabriel and Hannan 1998;~~

Ross and Yinger 2002), but gender remains understudied in two-sided ﬁnancial markets. To my knowledge, this is the ﬁrst study to consider own-gender preferences as a possibly reciprocated phenomenon.

The data I collected in 2010 include detailed information on economic networks linking 15 susu collectors to 384 clients in the Central Region of Ghana. The information represents the ﬂow of funds in both directions (between susu collectors and a sample of their clients) in undirected networks of nodes. Clients contribute business savings to susu collectors. Susu collectors provide clients perceived to be creditworthy with loans. Yet, susu collection has traditionally been a male-dominated affair, with the vast majority of collectors being male. Although clients are mainly female in the

Ghanaian informal ﬁnancial sector, they are not necessarily privileged since they tend to receive less credit (Ekumah and Essel (2001)). The availability of records in the data on within-gender and cross-gender susu matched networks is unique. The data on savings mobilization is matched with clients’corresponding credit outcomes.

Individuals who associate with others based on gender may associate more with sub-groups within a particular gender-based network. I therefore compare homophily by gender with homophily by both gender and educational attainment. There are similarly positive credit impacts of female collectors (with male and female clients) when the analysis is limited to collectors and clients who are relatively educated. On the other hand, the gender-bias of male collectors’ credit outcomes (against female clients) disappears when education levels exceed the mean. This implies that factors that correlate with homophily (e.g. status or class) are potentially important for policy. Overall economic outcomes may positively impact gender bias, although this is an empirical question.

An interesting starting point may be to consider gender-based homophily within the context of economic effort within the sample. I therefore investigate the impact of 193 homophily by both gender and economic effort. For female clients, there are positive and signiﬁcant credit impacts of having a male collector when the female client saves on a schedule representing the highest degree of economic effort. The general bias against women in the main results is mitigated for females who signal creditworthiness to a relatively high degree. This result implies that economic effort has consequences for gender-bias against women. The paper proceeds as follows. Section 6.2 summa- rizes relevant literature. Section 6.3 provides some background to the susu institution and the importance of gender. Section 6.4 explains the frameworks. Section 4 gives the empirical methodology, while Section 6.5 present the main results. Section 6.6 concludes with policy implications.

~~6.2 Literature~~

Several empirical findings support political and economic cases for women empowerment (see World Bank (2001), (2012)) for comprehensive summaries). For example, researchers have argued that using quotas to empower women economically, politically or socially may make institutions and public policy more inclusive (see, e.g. Chattopad- hyay and Duflo (2003)) Economic, political and social institutions that have never had female leadership may require first-hand experience on female ability to update current social and cultural biases against women (note Beaman, Chattopadhyay, Duflo, Pande and Topalova (2008, 2009)).

Noticeably, a signiﬁcant proportion of female empowerment at the micro-level (especially in the developing world) have occurred independently of quotas: the phenomenon of increasing female economic representation is mainly a corollary of general economic development (World Bank 2001). Notwithstanding this point, the complica- tions of experiencing unequal gender access to markets may actually blunt the gender- parity in observed outcomes attributable to women empowerment. For related reasons, 194 cash loans to female entrepreneurs did not yield signiﬁcantly better economic outcomes in the Philippines (Karlan and Zinman (2011)), Sri Lanka (De Mel, McKenzie and Woodruff (2009)) or Ghana (Fafchamps, McKenzie, Quinn and Woodruff (2011)).

As noted earlier, Beck, Behr and Madestam (2011) study the impact of microfinance officers’ genders on credit outcomes finding males to be discriminatory relative to female officers. Although the study is a natural experiment that considers the incidence of gender discrimination to be randomized, qualitative gender studies literatures in sociology have shown that gender-bias is a function of prior social and demographic characteristics (e.g. Desai 1994, Correll 2004).

Systems of inheritance are also important for gender-related outcomes. Another important related work is Gneezy, Leonard and List (2009) which compares matrilineal and patrilineal societies in terms of gender-based selection into competitive environments. Their main result is that signiﬁcantly more females choose more competitive environments in strictly matrilineal societies while males favor patrilineal societies.

~~The above result also has implications for self-selection into gender arrangements.~~

Yet it is not clear how the merging of formal and informal ﬁnancial arrangements functions in gender-biased environments. In the next section, I discuss the matrilineal Ghanaian Central Region and formal-informal ﬁnancial markets, focusing on the susu institution.

~~6.3 Matrilineality, Gender and Entrepreneurship in Cen-~~

~~tral Ghana~~

As discussed in Chapter 2, the vast majority of the Central Region’s inhabitants inherits kinship identities matrilineally. In such cultural institutions, kin membership is traced such that children belong to their mother’s kinship instead of their father’s.

~~This cultural phenomenon has inﬂuenced several social norms, including employment. 195~~

~~Although matrilineal kinship may yield positive implications on female ownership of property (Bortei-Doku Aryeetey 2000), women have often had less collateral and hence~~

ﬁnancial access. At the same time, Ghanaian women occupy a key position in the production of agricultural and other goods and services mainly for the Ghanaian informal market. In the next sub-section, I discuss the recent phenomenon of female empowerment in Ghanaian ﬁnancial markets. 6.3.1 Susu Collection, Rural Banks and the Political Economy of Women’s Enfranchisement in Central Ghana

I briefly review the discussion on rural and community banks and their relationship with informal financial arrangements in Central Ghana, the third smallest region in the erstwhile Gold Coast. State-motivated rural and community banks are the main providers of formal finance, modeled on rural bank institutions in the Philippines in the

1960s. Such institutions must be majority-owned by the hosting community, and have significantly lower minimum capital requirements than commercial entities that favor the urban elite. The share of supervision costs incurred to the Bank of Ghana by rural banks is disproportionately high relative to commercial banks, and government officials hope that positive externalities from expanding formal rural finance to the productive poor, including women entrepreneurs will justify these costs in the future (see Nair and

~~Fissha 2010).~~

Due to inadequate access to capital and other factors, a signiﬁcant proportion of women and men in Ghana mobilize savings through an institution known as susu collection (Aryeetey 1994). Deposit collectors mobilize funds for clients on a daily basis, to be returned at the end of the month (sans a commission equal to a day’s contribution). Susu collection is primarily a savings institution with very scarce and low credit provision. Rural banks, including the collaborative institution for this study, Kakum

Rural Bank (established in February 1980) typically rely on employed susu collectors to mobilize funds on their behalf, although they have recently transitioned to training 196 and using internal susu collector staff over time. Advantages bank susu collectors have over their independent counterparts are their ability to provide bank loans, and the absence of commissions. Credit is available to entrepreneurs after mobilizing savings for at least 3 months.

Although clientele skews female, susu collection has traditionally and historically been male-dominated. After the boom in informal ﬁnance in 1990s, severe gender-bias in credit distribution in the Ghanaian Central Region by the early 2000s. Using male susu deposit collectors to reverse credit discrimination against women entrepreneurs have proved difﬁcult in a number of initiatives (Essel 1996, Ekumah and Essel 2001).

During interviews in the region, a cross-section of susu entrepreneurs and rural bank ofﬁcers noted that reasons for the near-nonexistent female participation in past susu collection may have been the personal security hazards of mobilizing large sums of money on one’s person, and skewed educational outcomes (against females), which have signiﬁcantly improved as a function of policy and economic development (Sackey

(2005)). Currently, the Region hosts a disproportionate number of the best secondary schools in Ghana, ranking third in pass-rates of Mathematics and English in the national Senior Secondary School examination and overall access to educational institutions (see Overseas Development Institute 2009 for a discussion on education policy reform). Kakum rural bank has provided school scholarships for girl students in the

Central Region since the early 1990s. However, table 6.1A shows that female susu collectors are in their early 30s on average in the data, and therefore a signiﬁcant proportion may not have beneﬁtted from such direct educational support. Neverthe- less, such initiatives have helped the organization retain some legitimacy in the area as community-oriented institution with an social responsibility for education.

In meetings with local shareholders as well as current and potential entrepreneurial clients, bank hiring ofﬁcials have decided against using quotas for hiring women susu collectors. This decision was taken partly to engage the trust of local entrepreneurs in 197 the competency of female collectors to identify creditworthy clients. The rural bank still emphasizes gender in their vacancy announcements (in radio announcements for example), given the current status of women collectors as underrepresented and the overarching history of low female representation in susu collection. During the interview time of the study (in August 2010), the author had access to 10 male and 5 female susu collectors.

If susu collectors and clients share the same gender, a relatively better understanding of clients’ circumstances may improve economic outcomes. On the other, same- gender biases may adversely affect the efﬁciency of the susu institution. I now present a theoretical discussion to isolate the gender-bias impacts of susu collection on savings and credit outcomes, given the current context of both male and female collectors. I also discuss the matched networks generated by female collectors having clients of the same and opposite genders, as well as their impacts.

~~6.4 Theoretical Models~~

In this section, I model interactions between susu collectors and clients within two complementary and related frameworks made possible by the data. I ﬁrst consider cross-gender bias, using discrimination against female clients by male collectors to illustrate. The use of a separating equilibrium draws on work elsewhere in the dissertation (Chapters 3 and 5), but applied to gender bias. This approach is important because it allows a discussion on possible reactions to gender-bias. I complement this with a discussion on same-gender bias, or bias from homophily: matches of clients and collectors based on gender. 198

~~6.4.1 Signaling and Cross-Gender Bias in Susu Savings Mobiliza- tion and Credit Provision~~

The economic interactions between susu collector and client in the framework of separating equilibria are the main arguments of Chapters 3 amd 5. Savings contributions may signal creditworthiness to susu collectors, who respond by providing credit. If I assume no gender discrimination in economic outcomes, then there exists a separating equilibrium shown in Figure 6.1 (below). The variable Q represents creditworthiness, where λ =Prob (Q = QH ) ∈ (0,1). An individual is either creditworthy QH or not creditworthy QL. Without gender-bias, the perceived creditworthiness (of clients) should be equivalent to a susu collector irrespective of the client gender. Both female and male

∗ individuals who are creditworthy contribute an amount of savings s (QH ), and receive loan amount QH . Individuals who are not creditworthy do not make any savings, and receive QL. The discussion relied on perfect Bayesian equilibria, and the belief function for the susu collector ν(s) ∈ [0,1] was derived using Bayes’ Rule to respond to creditworthiness based on savings contributions (see Chapters 3 and 5 for details). We now discuss the case where gender-bias does exist in the next sub-section. 199

~~In this sub-section, I modify the model to reﬂect the possible role of gender-bias.~~

I present a similar signaling model to explain gender-bias in the interactions between susu clients and their collectors. The discussion is in the vein of above discussion where the variable Q represents creditworthiness, where λ =Prob (Q = QH ) ∈ (0,1).

An individual is either creditworthy QH or not creditworthy QL. The main addition to the model is gender: G = {F,M}, where F =female and M =male. The theoretical discussion is based on ﬁgure 6.2 below: 200

Using ﬁgure 6.2, I focus on the gender-biased economic interactions between susu collectors and susu clients. As in the case of ﬁgure 6.1, Individuals who are not creditworthy save an amount of zero, and receive a credit amount QL. In the mould of the previous model, the loan amount of QLis provided to clients because she is not creditworthy. Following the discussion of Figure 6.1, creditworthy individuals contribute

∗ savings s and receive credit QH . Let the susu collector be male, and the susu client be female. The belief function of the collector reﬂects a gender-bias against female clients. In the presence of gender- bias (against women), women who are creditworthy contribute relatively more savings for the same credit amount as (men who are creditworthy). Since this assertion reﬂects an out of equilibrium belief, I do not prove the statement, but explain it from the above

~~ﬁgure (6.2). An absence of gender-bias against women would imply that both female and male clients receive QH . However, male collectors are assumed to be biased against 201~~

~~0 ∗ female clients. For this reason, female clients have to contribute s > s to receive QH .~~

∗ ∗ On the other hand, male clients only contribute s < s to receive the same QH . Both female and male clients who are not creditworthy receive QL. A similar explanation holds if female collectors are biased against male clients.

~~6.4.2 Gender Matching and Susu Savings Mobilization and Credit Provision~~

The previous section discussed cross-gender bias: susu collectors may be biased against clients of the different gender. Clients may react to bias by saving more than they would have to if gender-bias was absent. At the same time, bias may occur as a matching problem: susu clients may contribute more savings to collectors of the same gender.

Similarly, collectors may provide more credit to clients on the same gender. To complement the previous discussion on cross-gender bias, I provide a discussion on bias by gender match in this section.

Gender-based matches of collector-clients create susu networks that are potentially important. On the one hand, matching collectors and clients of the same gender may facilitate the susu savings and credit processes. On the other, savings and credit outcomes may be biased toward agents of the same gender, adversely affecting the operation of the susu structure over time. I provide a brief motivation of matched client-collector networks based on a slight variation of Abrevaya and Hamermesh (2011) to focus on gender-based matches of clients and collectors2. This provides a same-gender discussion to complement the cross-gender discussion of the previous section.

~~I provide some illustrations to explain bias by gender match. If female susu clients are relatively more generous in contributing savings (than male clients) when matched~~

2 The derivation is in the Appendix. 202 with female susu collectors, then I assume female clients to be positively biased by gender-match. On the other hand, if female clients are relatively less generous than males when matched with female susu collectors, they are negatively biased by gender- match. Similarly, If male susu collectors are relatively more generous than males when matched with female clients, the bias by gender-match will be positive. If female collectors are relatively less generous than males when matched with female susu clients, the gender-match effect will be negative. Similar explanation holds for male collectors and clients.

~~6.4.2.1 Hypotheses~~

Combining both sections on signaling and gender matching, I now present the following hypotheses. Hypotheses 1 and 2 derive from the signaling model of bias, while hypotheses 3 and 4 follow from the discussion on gender-matched networks.

~~HYPOTHESIS 1: Male clients provide less susu savings contributions than female clients.~~

~~If s∗ QM =the savings of male clients, and s∗ QF =the savings contributions of female clients, then s∗ QM < s∗ QF .~~

~~HYPOTHESIS 2: Male clients receive more credit than female clients.~~

~~If l∗ QM =the credit outcomes of male clients, and l∗ QF =the credit outcomes of female clients, then l∗ QM > l∗ QF .~~

~~HYPOTHESIS 3: Clients in gender-matched networks provide more susu savings contributions than clients not matched by gender.~~

~~If s∗ Qmatch =the savings of gender-matched clients, and s∗ Qunmatched =the savings contributions of clients not matched by gender,~~

~~then s∗ Qmatch ≤ s∗ Qunmatched.~~

~~HYPOTHESIS 4: Clients in gender-matched networks receive more credit than 203 clients not matched by gender.~~

~~If l∗ Qmatch =the credit outcomes of gender-matched clients, and l∗ Qunmatched =the credit outcomes of clients not matched by gender,~~

~~then l∗ Qmatch ≥ l∗ Qunmatched.~~

~~6.5 Empirical Tests: Female Susu Collectors and Gender-~~

~~Matched networks~~

Testing the model’s predictions requires identifying variation in both savings mobilization and credit attributable solely to the genders of susu clients and collectors. The empirical implementation will focus on individual gender impacts of susu collectors and clients, as well as the collective gender impacts of susu collector-susu client networks as noted in the previous section. The average gender impacts of a susu collector on savings contributions and credit outcomes of clients of both genders would be different from the average gender impacts of a susu collector when linked exclusively with clients of the same gender.

Since the employment initiative was aimed at enfranchising women as susu collectors, I ﬁrst focus on female collectors, before isolating gender-bias as well as (collector- client) gender-match impacts. Tables 6.1A, and 6.1B show summary statistics. Female collectors are only 24% of the surveyed susu collectors, while male counterparts are in the majority (76%). On the other hand, female and male clients are about 60% and 40% of the client sample (respectively). I introduce the empirics in the next sub-section. 204 205 206 207 208

~~6.5.1 Speciﬁcations~~

~~In this sub-section, I estimate the following equation for savings mobilization and credit as a function of the gender of the susu collector and other social and economic factors.~~

~~Preliminary Estimates~~

~~To test whether the gender of a client’s susu collector inﬂuences the client’s savings mobilization and credit outcomes, we run the following OLS regression:~~

~~0 yi = α + β fi + Xi π + εi~~

~~where yi is the economic outcomes (savings contributed or credit received) by client i with susu collector s; f is a dummy variable if the client i has his or her susu collector~~

0 being female; Xi is a vector of the following pre-treatment client i controls: age, gender, married (=1), some secondary schooling, and monthly income. Table 6.2A presents the OLS estimates. Column 1 shows the associations of independent variables with savings contributions while Column 2 shows effects on credit outcomes. While having a female susu collector does not show a significant impact on savings mobilization, susu collection by gender has a large and significant credit effect. The client controls of gender and income significantly affect savings contributions, while being married has a negative impact on receiving credit. Given our discussion on gender bias and the possibility of discrimination in savings and credit outcomes, self-selection may be important into two ways. First, clients may self-select into a savings arrangement with a collector based on the collector’s gender. Secondly, clients may self-select into gender- matched networks based on pre-treatment characteristics. The estimation strategy that follows relies on the above observation that gender impacts may depend on a set of pre-treatment characteristics. 209 210

~~6.5.2 Empirical Strategy: Propensity Score Matching~~

~~The design of this project allows identiﬁcation of gender effects in the presence of dif-~~

ﬁculties ﬁrst presented by Rosenbaum and Rubin (1983). Direct comparisons between treated and non-treated groups may be misrepresentative when the units exposed to the treatment differ systematically from the units unexposed to treatment. The authors show that if groups of subjects have similar propensity scores, they can be expected to have similar values of background data in the aggregate. Propensity scores can summarize all background information, effectively predict the probability that susu clients receive a treatment (instead of a control)– providing unbiased effects of the treatment, or Average Treatment Effects.

Let the differences between treated and non-treated agents be encompassed by observables X, following the literature. A comparison group is accumulated from non- treated groups. The distribution of observables for the comparison group is as similar as possible to that of the treated group. The conditional probability of receiving treatment is estimated using a logit as a function of relevant socioeconomic factors. We use one-to-one matching with replacement and kernel matching using a normal density function3.

~~I ﬁrst use NEAREST NEIGHBOR MATCHING (OR ONE-TO-ONE MATCHING WITH~~

REPLACEMENT) to estimate our Average Treatment Effects based on propensity score matching algorithms. This technique matches treated individuals with comparison individuals based on the similarity of their propensity score. Mapping with replacement means that the matches are usually of a higher quality since an untreated individual can be used more than once in creating a match. Results are based on limiting analyses to the area of common support in each case. Excellent summaries of the advantages of

3See Caliendo and Kopeinig (2008) for a general review, and refer to Dehejia and Wahba (2002) for further discussions on one-to-one matching. Other propensity score methods such as subclassiﬁcation tend to provide similar results to kernel matching (Zhao 2004). Mahalanobis-metric matching are expected to yield similar matching results to my one-to-one approach (Michalopoulos, Bloom and Hill 2004) and are therefore not used in the paper. 211 one-to-one matching with replacement (relative to without replacement) are provided by Caliendo and Kopening (2008) although the differences may not be very signiﬁcant overall (Zhao 2004).

I also use KERNEL MATCHING (WITHA NORMAL DENSITY FUNCTION) to estimate Average Treatment Effects of gender and gender-bias. This methodology yields a comparison group by calculating weighted averages of almost all individuals in the control group. Kernel matching estimates derive from a weighted regression of the counterfactual outcome on an intercept with weights given by the kernel weights (Smith and Todd 2005). I use weights from a normal kernel, and derive our matches from a logit regression used to balance the subgroups. An advantage of kernel matching is a lower variance since relatively more data is used in the estimation process. A demerit is the possibility of adverse matches, although this is reduced substantially by imposing the common support, as motivated by the literature4.

~~6.6 Results~~

The savings and credit Average Treatment Effects are presented in table 6.2B: having a female collector does not have significant savings impacts, but shows strong positive credit impacts. Both one-to-one and kernel matching estimates show that female collectors have positive significant effects on credit outcomes. The positive credit impacts are independent of any significant gender bias in savings impacts. That is, the positive credit effects do not follow from significant savings impacts. It is worth noting that in the results of table 6.2B, the clients involved are not disaggregated by gender. In the next sub-section, I investigate whether client-collector matches (or networks disaggregated by client gender) yield significant economic outcomes relative to unmatched clients. 4Note Heckman Ichimura and Todd ((1997), (1998)). 212 213

~~6.6.1 Susu Collector-Susu Client Gender Networks~~

The presence of gender-based networks may have implications for the mechanisms of susu savings mobilization and credit provision. Gender-bias in the undirected diffusion of funds may be important to economic interactions. Funds transmitted across nodes may or not be reciproal as a result of gender considerations. To explore the relevance of gender bias, we partition our analysis of clients of female susu collectors by gender. I provide some descriptive statistics to show the extent to which gender-matched networks are comparable.

Table 6.3A shows summary statistics of male and female clients. The ﬁrst two columns show male and female clients of female susu collectors respectively. The female and male clients are comparable in terms of the variables education, marital status and income. Clients (by gender) are less comparable in terms of age. I show summary statistics for the susu networks’ outcome variables in table 6.3B. Interestingly, male clients contribute more to female collectors than male ones on average, while female collectors contribute more to male collectors than female ones. On the other hand, female collectors give slightly more credit to female clients relative to male clients.

~~Similarly, male collectors give relatively more credit to clients of their own gender. 214 215 216~~

~~In table 6.4, I show OLS estimates of gender-based susu networks. From column~~

~~1, female clients contribute signiﬁcantly less savings to collectors of the same gender.~~

However, although female clients contribute more to male collectors, the effect is not statistically significant. Similarly, the savings effect of having a male collector is not significant for female collectors although it is negative. Assignment into a network is not random, since the association of monthly income (on savings contributions) is positive. From the second column, female collectors with female clients have large and significant positive credit impacts. Similarly, female collectors with male clients have significant (but slightly lower) credit impacts. 217 218

As per the discussion (in the previous section), the non-random assignment motivates our use of one-to-one nearest neighbor matching in table 6.5. The female collector-male client networks show positive and significant impacts on savings contributions: male clients contribute more to female collectors. On the other hand, the male collector-male client effects on savings contributions are not significant. None of the other matches show significant impacts. The female collector-male client networks show significant positive effects on credit, in a positive relationship with the previous savings association. Although female clients did not give significantly less savings to their male collectors, I find that male collectors granted significantly less credit to them. On the other hand, male collectors did not give significantly more credit to male clients. 219 220

The kernel matched results are similar and stronger in some instances (shown in table 6.6). The female collector-male client networks show significant positive effects on credit, following positive savings impacts as in the previous discussion. The female collectors provide significantly more credit to clients of the same gender, although they receive less savings from them. Again, male collectors give significantly less credit to female clients, although female clients do not significantly lower their savings contributions to male collectors. Male collector-male client networks do not have a significant credit (or savings) impact. I now discuss the results at length. 221 222

~~Discussion of Main Results~~

I find that male clients are significantly more likely to contribute savings to a female collector. On the other hand, female clients contribute significantly less savings to female collectors although they share the same gender. The credit results show that female collectors give more credit to clients of the same gender (inspite of the clients contributing less savings.) Female collectors give more credit to male clients, while male collectors give less credit to female clients. Therefore, male collectors show gender-bias against female clients while female collectors do not show gender-bias against male clients.However, this evidence may only be suggestive since we consider homophily in susu networks under the single dimension of gender. Considering subsets of gender-based networks where homophily is observed in a different dimension could affect savings mobilization and credit outcomes. We provide a discussion in the next section.

~~6.6.2 Alternative Explanations: Education and Effort in Gender- Bias, Susu Savings and Credit Provision~~

Within the gender-based economic networks discussed above, other distinguishing features may increase or decrease gender bias. In this section, I discuss different scenarios that may differently inﬂuence the propensity of gender bias to another in savings and credit. I focus on two explanations: the possibility that female collectors and clients may have similar levels of education, which may in turn have implications for savings and credit outcomes by gender. Secondly, I focus on the effort inherent in providing savings by gender, which may have implications for gender-bias in savings mobilization and credit provision.

~~First of all, related network models with microfoundations that involve homophily~~

(the tendency of individuals to associate disproportionately with others with similar 223 traits–such as gender in our case)–also predict that homophily may delay convergence toward consensus as agents average their observations of their neighbors to develop beliefs (e.g. Golub and Jackson 2012). However, these models do not account for whether the incidence of homophily in a different dimension has positive or negative implications on already observed homophilous groups. The bias (male collectors discriminating against female clients) may be reinforced by newer homophily related to educational attainment, or alternatively, lessened by it. For instance, it may be that gender-bias occurs in gender-matched groups, but not in gender-matched groups that have higher education attainment in the sample. We test this in the next section.

~~6.6.3 Survey Evidence and Mechanisms of Education and Economic Effort~~

~~6.6.3.1 Education and Gender Bias~~

The presence of gender-bias in formal-informal ﬁnancial markets is perhaps surprising since the Central Region of Ghana hosts a disproportionate presence of educational institutions, although there still exist signiﬁcant education constraints. In the data, susu clients and clients have attained some level of Junior Secondary schooling and Senior

Secondary schooling (or post-primary education) on average. This suggests that relatively educated female collectors and their gendered-client networks with relatively similar levels of educational attainment may affect gender-biases in outcomes. To explore the relevance of education considerations, I ﬁrst focus on female susu collectors whose education levels exceed the mean. Another sub-analysis restricts the discussion to female and male susu collectors, partitioning their clients by gender, (as done in the previous section) before further partitioning by educational attainment above the mean.

~~In table 6.7, I use one-to-one matching (with replacement) and kernel matching~~

~~(with a normal density function) to isolate the impact of having a female susu collec- 224 tor whose educational attainment exceeds the mean on savings and credit outcomes.~~

Column 1 shows the impacts of highly educated female susu collectors on savings outcomes, while Column 2 shows the effects of educated female collectors on credit outcomes. The savings outcomes are similar to female collectors in general (with no significant impact). On the other hand, the credit results are positive and statistically significant, with larger co-efficients (or higher economic significance). 225 226

We next show susu gender networks’ (which have educational outcomes above the mean levels) average treatment effects on savings mobilization and credit outcomes in table 6.8. Similar to the previous results, I find that female clients contribute less savings to collectors of the same gender while collectors provide more credit. On the other hand, male clients that have a female collector contribute more savings receive more credit. Male clients contribute more savings to female collectors, but receive less credit, although the credit responses are not statistically significant. The economic interactions between male collector-male client networks are never significant. 227 228

Gender-focused savings mobilization and credit provision with educational biases may not reﬂect the actual effort people take in mobilizing savings. In the ﬁnal section, we provide some evidence on how different savings schedules affect savings mobilization and credit outcomes, conditional on the collector and network gender factors discussed so far. If effort is important, it may weaken or strengthen the observed bias caused by gender.

~~6.6.3.2 Economic Effort and Gender Bias~~

In the survey, the susu clients also have different savings schedules that they self-select into, that influence the frequency of meeting their susu collector. Each schedule represents a different degree of financial effort exerted by a susu client. Susu clients may agree to contribute savings on a daily, twice-weekly, weekly, fortnightly, or monthly basis. In the survey data these schedules represent 26%, 24%, 28%, 17% and 0.04% respectively. Companion pieces to this paper showed that the daily schedule required the most effort and commitment, yielding the most savings, and earned the most credit relative to other more relaxed savings schedules. To explore the relevance of economic effort, I perform another sub-analysis, focusing on female susu collectors and networks under different savings schedules. The fortnightly and monthly schedules are dropped from the analysis in the section for less representation across gender-based susu networks. For similar reasons, we limit our discussion to OLS estimates. Given gender- bias, clients may receive more credit from collectors of the opposite gender when the degree of financial effort is relatively high.

In table 6.9, I study the impact of having a female susu collector under different savings timetables. Columns 1 and 2 show the impacts of varying degrees of economic efforts of using female susu collectors on savings outcomes. Columns 3 and 4 shows the effects of different female collector savings schedules on credit outcomes. The 229 savings outcomes are positive when the savings schedule is two days every week. On the other hand, the savings results become negative when the savings schedule is relaxed into a weekly schedule. Neither savings nor credit outcomes are significant for the daily schedule. The credit outcomes are very significantly positive for clients with a female collector that save two days every week. For clients with a female collector, saving every week had a negative and significant impact on credit outcomes. 230 231

~~We next show susu gender networks’ schedule effects on savings mobilization and credit outcomes. We ﬁrst focus on female collector-female client networks in table~~

6.10. Within gender-based susu networks, I ﬁnd that female clients do not contribute signiﬁcantly more savings to collectors of the same gender (on a twice-weekly basis), but contribute less on a weekly basis. Female susu collectors provide more credit to female clients who save on a twice-weekly basis. On the other hand, female collectors provide less credit to clients who save on a weekly basis. 232 233

Table 6.11 shows no significant impacts of savings schedules (with cross-gender networks) on savings mobilization, although female collectors lend less to male clients who save on a daily basis. On the other hand, table 6.12 shows a slightly significant impact on female credit from male collectors only when female clients provide savings on a daily schedule. Finally, table 6.13 does not show any significant gender-bias between male collectors and clients irrespective of the schedule.

Of course, gender-based savings mobilization and credit outcomes, even when disaggregated by savings schedlues may not reﬂect gender-based repayments of loans once credit is received. However, given the strength of the results, the evidence that savings schedules are important lends credence to the view that gender-bias is subject to within-network homophily. In addition, the ﬁndings that gender-bias savings and credit may or may not obey a signaling model is consistent with our earlier network- based results. They suggest that gender-bias may both exist independently in savings mobilization and credit outcomes, or may be reciprocated by agents in allocating credit outcomes. 234 235 236 237

~~6.6.4 Discussion of Education and Economic Effort Results~~

I ﬁnd that male collectors discriminate against female clients and show evidence of homophily (people being inclined to associate with others who are similar to them) by gender. Yet, homophily does not explain all of the results, since male clients contribute more savings to female collectors. There are also postive credit impacts of female collectors (with male and female clients) when the analysis is limited to collectors and clients whose education levels exceed the mean. The gender-bias of male collectors’ credit outcomes (against female clients) disappears when education levels exceed the mean. This implies that factors that correlate with homophily (e.g. status or class) are relevant in the incidence of gender-bias. In line with the signaling model, female clients contribute more to male clients when education levels exceed the mean.

For female clients, there are positive and signiﬁcant credit impacts of having a male collector when the female client saves on a daily schedule. The daily schedule represents the highest level of internal commitment (see Chapter 4), and economic effort

(see Chapter 5). Although there is gender-bias against female clients in the general results, the bias is mitigated for females who signal creditworthiness to a relatively high degree. Female collectors give female clients relative more credit conditional on female clients being relatively creditworthy. Female clients on the weekly schedule have a negative credit impact while female clients on a twice-weekly schedule have a positive credit impact. This result implies that economic effort has consequences for gender-bias against women.

~~6.7 Conclusion and Policy Implications~~

The literature on gender in politics and economics tends to focus on agents at the mercy of a principal, although agents may react to perceived gender-bias in economic 238 decision-making. In practical terms gender-bias is often two-sided, although studies that analyze this are rare. I analyze the gender-biased signiﬁcance of female empowerment on economic outcomes in Ghana, focusing on formal-informal susu deposit collection. The theoretical analyses isolates gender bias in credit and savings outcomes as functions of gender-bias across genders and homophily (same-gender matching in networks). The analyses brings together elements from micro theory, political economy and social network analyses. The data show that female collectors generally do not reciprocate the cross-gender biases attributed to male collectors. Female clients generally save more with male clients, a ﬁnding that agrees with a model of signaling gender-bias. If female clients anticipate gender bias in the credit decisions of their male collectors, they save more (than a male client would) for equal credit amounts. On the other hand, gender-bias can be analyzed in a two-sided situation where credit outcomes correlate with savings contributions.

Educational attainment and economic effort (savings schedules) are important for interpreting how rigid gender-bias is in the study. For relatively educated collectors and clients, there results are similar, but the gender-bias of male collectors’ credit outcomes

(against female clients) disappears when education levels exceed the mean. This implies that other social factors that correlate with homophily (such as social status) may be relevant in the incidence of gender-bias. Further study should study such cases. The impact of economic effort on gender-bias in credit outcomes imply that gender-bias may be lessened. Female collectors only gave more credit to female clients who were on savings schedules reﬂecting relatively high economic effort (relative to other women on more relaxed schedules). On the other hand, the male collectors have positive credit impacts on female clients when the latter are on savings schedules that reﬂect high economic effort. For the above reasons, signaling by effort may mitigate gender-bias.

Although informal ﬁnance has mainly been biased against women, the issue has persisted in spite of merging formal and informal institutions. The female susu collec- 239 tors were hired in a competition-based system instead of a quota-based intervention.

Such broad-based approaches to female empowerment may be important in similar contexts where the efficiency of empowered women is linked to their legitimacy in certain positions. Other microcredit institutions may benefit from having more female officers in credit roles, given possible homophily effects. Although education initiatives for susu collectors may be helpful, self-selection into matched groups is another area that requires attention. Economic effort is very important in lessening outcomes associated with gender-bias. Other avenues that improve information asymmetries between collectors and clients can be researched and implemented as policy. 240

~~Appendix~~

~~This Appendix contains supplementary information for Chapter Six. The contents are as follows:~~

~~A.1: A model on the Average Treatment Effects of Gender-Matching. The explanation is a more comprehensive discussion of the Abrevaya and Hamermesh (2011) model.~~

~~A: Same-Gender Matched Networks~~

~~Model~~

In this section, I model interactions between susu collectors and clients within gender- based matched networks. Gender bias arises because clients may contribute more savings to collectors of the same gender–who may respond by providing more credit to clients of the same sex. The model will draw from Abrevaya and Hamermesh’s recent work on gender-based favoritism (2012). However, it deviates from that work by focusing on undirected (i.e. reciprocated) gender-bias. This approach is made feasible by using a basic social network-based framework (e.g. Jackson 2008, Golub and Jack- son 2012). I use the model to develop testable implications of gender bias on savings mobilization and credit provision. 241

~~Set-up~~

~~Let a susu collector and her clients represent nodes, N = {1,... ,n}, of a susu network.~~

This susu system is represented by its weighted matrix, which is a symmetric n × n matrix S with entries {0,1}. We allow Si j = S ji = 1, since our discussion is primarily on undirected economic networks as noted earlier.

Every client is linked to a single collector. We let relationships (between a susu collector node and a susu client node) ﬂow in both directions, so that susu clients contribute funds (called savings) to collectors, who respond with funds (called credit).

n We let di (S) = ∑ j Si j represent the number of links of node i, so that average degree is denoted d(S). As a result, the total number of links in the network is D(S) = ∑ j di(S). We focus on gender as a distinguishing feature or “type” that inﬂuences the propensity of a node to connect to another. As such, the network is partitioned into 2 = Nf ⊂

~~N, where Nf refers to female gender nodes. We focus mainly on two generic nodes by gender: female susu clients who contribute savings and female susu collectors who provide credit5.~~

Gender may be important to both the savings contribution and credit allocation events, in two ways. First, gender may be independently important to savings mobilization and credit outcomes. Alternatively, gender may be inﬂuential to both occurrances jointly. We ﬁrst discuss the case where the impact of gender on savings and credit are independently important before dicussing the scenario where gender effects are jointly important.

~~Matched Gender Network Impacts on Savings and Credit Outcomes~~

~~In this section, I develop a model of gender-biased microeconomic outcomes, based on Abrevaya and Hamermesh, henceforth AH (2012), but generalized to a dual-market~~

5Note that since the present analyses is identical for male collectors, I omit it in this section. 242 scenario. This adaptation of the AH (2012) model allows insight into the independent impacts of collectors’ and clients’ genders on savings mobilization and credit outcomes. Given our two-sided emphasis, we only slightly adjust the model in the second part of this section to introduce the credit impacts of gender, embedding our discussion in the basic network framework introduced earlier. We assume for simplicity that the utility of clients depend entirely on credit while the utility of collectors depend on savings.

~~Matched Gender Impacts on Savings Mobilization~~

Susu saving clients are represented by Ci (for i = 1...I) while Susu (deposit) Collectors are notationally D j(where j = 1...J). Gender is represented by f (Ci)and f (D j) for clients and collectors respectively (= 1 for females.)

The hypothesized impacts of gender on savings and credit outcomes are simpliﬁed and presented below. We begin with the gender impact on savings. The utility of collector D j when matched with customer Ci is:

~~U(Ci,D j) = µi + ψ j + α f (Ci) + β f (D j) + λ f (Ci) f (D j) + εi j. (6.1)~~

We consider µi to be an idiosyncratic value of the savings contribution, while ψ j is the deposit collector’s valuation of that amount. The effect εi j refers to an unobservable. We say that the collector is satisﬁed with the savings contribution amount if U (Ci,D j)is greater than zero. This event occurs when the probability of the Right- Hand Side of (1) is greater than zero.

Since µi and ψ j are assumed to be random, a composite error term can be con- structed from (1) whereby Ei j =µi + ψ j+εi j. Let the cumulative distribution function of −Ei jbe G(.). This allows the isolation of a crude average treatment effect of gender simpliﬁed below. 243

~~AVERAGE TREATMENT EFFECT OF GENDER MATCHED NETWORKSON SAV-~~

~~INGS~~

~~• The effect of female deposit collector on male client saving: G(α) − G(0)~~

~~• The effect of female deposit collector on female client saving: G(α + β + λ) −~~

~~G(β)~~

~~• Then, the treatment (difference-in-difference) effect is:~~

~~[G(α + β + λ) − G(β)]-[G(α) − G(0)].~~

If female susu clients are relatively more generous than males when matched with female susu collectors, then this impact is greater than zero. If female clients are relatively less generous than males when matched with female susu collectors, this effect will be negative.

~~Matched Gender Impacts on Credit Outcomes~~

We now consider the effect of gender effect on credit. We assume that the prior gender inﬂuence in the savings event does not inﬂuence the present event. The utility of client Ci when matched with collector D j is:

~~U(D j,Ci) = υ j + νi + γ f (D j) + δ f (Ci) + τ f (D j) f (Ci) + ε ji. (6.2)~~

~~Consistent with our recent discussion we construct a compositve error term so that e ji = υ j + νi + ε jiand let the cumulative distribution function of −e jibe H.~~

~~AVERAGE TREATMENT EFFECT OF GENDER MATCHED NETWORKSON SAV-~~

~~INGS 244~~

~~• The effect of female client on male collector credit: H (γ) − H (0)~~

~~• The effect of female client on female collector credit: H (γ + δ + τ) − H (δ)~~

~~• The corresponding average treatment effect is:~~

~~[H (γ + δ +tτ) − H (δ)] − [H (γ) − H (0)]~~

If female susu collectors are relatively more generous than males when matched with female clients, this impact will be positive. If female collectors are relatively less generous than males when matched with female susu clients, this effect will be negative. 245

~~Chapter 7~~

~~Conclusions and Future~~

~~Research~~

Social scientists have presented much evidence on a negative relationship between formal and informal ﬁnancial institutions, with adverse implications for state capacity in developing countries. Previous scholarship has also provided the important concept of inclusive economic institutions—mitigating factors that rely on political foundations. Systems favoring pluralism are mechanisms by which institutions may improve state capacity and extend formal systems to the disenfranchised. However, this valu- able research has paid less attention to the political and behavioral norms governing informal economic institutions, in spite of their ramiﬁcations on state capacity.

This dissertation has provided several general mechanisms through which the political and behavioral characteristics of inclusive economic institutions are significant for formalizing informal arrangements. I have proposed and tested hypotheses about how improved information flows, internal commitments, economic effort, and gender bias are all consequences of merging formal and informal finance. The work yields an initial step toward what should be an important goal for the political economics of 246 development: what are the causes and consequences of merging formal and informal institutions? In this chapter I revisit what I have learned, discussing some shortcomings and implications of the study.

~~A main contribution of the study is to focus on Ghanaian susu collection as an inclusive economic institution that contributes to state capacity through information~~

flows to the state. Susu collection in Ghana is an significant economic institution that has been underrepresented in empirical work, although formal-informal susu institutions are undergoing significant institutional change. The study has used a variety of theoretical frameworks to probe the validity of the main hypotheses it presents. The formal models show that state capacity can be built out of a generally informal financial sector; moreover, behavioral and signaling mechanisms are important for analyzing this finding. While the predictions of formal economic models (supported by a case study) are put to the test of observational field data, the political phenomenon of gender discrimination provides some boundaries to the policy implications of formalizing informal finance. At the same time, female empowerment in the formal-informal susu institution mitigates these qualifications.

Innovation within formal-informal financial hybrids is relatively rare in the various social science literatures, so this aspect is another contribution of the study. Although we often associate formal interventions with improved efficiency, the creation of twice- weekly, weekly, fortnightly and monthly susu savings schedules generally show that the creation of new savings schedules may be an endeavor best left to informal financial institutions. Nonetheless, the relative success of the MASLOC policy initiative and collaborations between rural banks and susu collectors provide much support for formalizing informal finance while suggesting new directions for development.

There are several possible additions to this work. One is obviously to broader the range of the informal sector under consideration. In this study, the topic selection was driven partly by the phenomenon under study in Ghana, but also partly by the avail- 247 able data which lent themselves readily to studying conditions under which formalizing informal finance may be significant. Particularly because susu collection is understudied, isolating financial behavior from other economic outcomes was an important goal.

Ghana is a country where the susu institution is dominant, a fact that motivated the general approach to analyzing data in Chapters Four, Five and Six. Still, future work could broaden the subset of the informal sector under consideration. While I apply signaling to studying creditworthiness, the concept might be relevant for informal labor arrangements. Other ﬁeld studies may investigate the extent to which other avenues of merging formal and informal economies may be important, such as health. Developing arguments on formalizing traditional medicine may be important in African countries where formal health services are relatively rare.

Another possible direction is to extend the theoretical approaches of improved information ﬂows (to the state), which were motivated by the Ghanaian case, to a greater range of cases. In particular, developing arguments for (or against) relatively non- intuitive government investments in development may be a fruitful area for research.

While one may think that engaging informality may risk weakening state capacity, the arguments provided in the Ghanaian case suggests that this may not be the result. Al- though institutions similar to susu exists elsewhere in Africa, Latin America and Asia, it is important to note that the level of collaboration with a robust rural and community banking system may differentiate the Ghanaian case from other situations.

Finally, there is the important matter of studying the impacts of formal credit on otherwise informally financed entrepreneurship. In an important sense, this study has taken the existence of fruitful investments as a given; it has defined and analyzed creditworthiness, and yielded results on economic effort and gender bias. Still, a more sustained inquiry into the future impacts of formal credit over time may further encourage or discourage the policy of formalizing informal finance.

~~However, even if the longer-term impacts of credit are not certain in this discus- 248 sion, the impacts of identifying creditworthiness on state capacity are relatively clearer.~~

Since the apparatus to tax small-scale entrepreneurship already exists in Ghana, the important missing link to improving state capacity from a policy perspective is gaining credible information on creditworthiness. The study is important since it yields spe- ciﬁc metrics on internal commitments, and uses economic effort to provide a ﬁrst step towards identifying the productive poor. The factors motivating internal commitments to save and creditworthiness (constrained by gender discrimination and improved by gender empowerment) may then be desirable because have implications on repayments at the micro level and state capacity at the micro level.

If more policies promote inclusiveness, we might expect state capacity to become more entrenched, especially if inclusive institutions have better information on the disenfranchised. However, cases with underlying participatory factors relatively similar to Ghana’s inclusive institutions may have relatively encouraging micro-level and state capacity outcomes. This study is not, then, a discussion that applies to every single situation where an informal economy dominates formal systems. It is instead a series of arguments that seek to understand a particular set of economic, political and social consequences that may follow from formalizing informal institutions that are mainly

~~ﬁnancial in character. 249~~

~~Annex: Matching Protocol~~

~~7.1 Matching Protocol for Propensity Score Estimations~~

~~1. Specify and estimate a logit model to obtain the marginal probabilities.~~

~~2. Standard propensity score matching step: (single and multiple treatments; re-~~

~~stricting to region of common support).~~

~~3. Compute Average Treatment Effects (ATE), given the Conditional Independence~~

~~Assumption (CIA). The CIA states that, given a pre-treatment set of covariates~~

~~X, potential outcomes are independent of treatment assignment (i.e., I assume~~

~~unconfoundedness given a propensity score (Rosenbaum and Rubin 1983)). For~~

~~the main empirical analysis of Chapters Four, Five and Six, I use propensity score~~

~~matching (see the main text for detailed explanations of one-to-one matching~~

~~with replacement and kernel matching using a normal density function). 250~~

7.1.1 Logit Regressions: Determinants of Susu Savings Schedules (Savings) 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300

~~7.1.2 Logit Regressions: Determinants of Susu Savings Schedules (Credit) 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318~~

~~7.1.3 Logit Regressions: Determinants of Susu Savings Schedules (Gender) 319 320 321 322 323 324 325 326 327 328 Bibliography 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345~~