ETHNIC STATE CAPACITY AND CONTRACTENFORCEABILITY∗

RAUL SANCHEZ DE LA SIERRA

State capacity has recently become the workhorse of development scholarship. One rea-son why state capacity matters is that expanding the state legal capacity may increase tradewhere social institutions cannot govern agency relations (Greif, 1993). However, the stateitself may be captured, and thus expanding the state could generate adverse results. Further-more, the state may crowd-out pre-existing informal mechanisms of contract enforcement.Estimating the impact of the state is challenging, because state formation is endogenous,and because in the absence of a functioning state, there is usually no data. I create a de-livery business in the Democratic Republic of the Congo, involving traders and customerswho learned to operate without the state, and randomly introduce state contracts. I findthat state contracts strongly reduce shirking. However, the results uncover an ethnic bias incontract enforcement by the state: only some ethnic groups can draw on the state’s legal ca-pacity, and customers anticipate this bias. Furthermore, ex-ante, I find that state contracts,when they are enforceable, and coethnicity are substitutes to solve commitment problemsthat prevent trade in the presence of agency relations. These results suggest that whilesocial institutions govern some agency relations, they also govern the state administration,therefore distorting the impact of state legal capacity. JEL Codes:.

∗1727 Cambridge street, Harvard University, Room E113, Mailbox 42, Cambridge, MA 02138. Fax: (617) 496-9592. Phone: (917) 488-9151 rsanchezdelasierra@fas.harvard.edu. Jean-Paul Zibika provided invaluable researchassistance and Gauthier Marchais provided an outstanding collaboration. I am grateful to Christopher Blattman,Ritam Chaurey, Pierre-Andre Chiappori, Jonas Hjort, Macartan Humphreys, Supreet Kaur, Christian Mastaki,Suresh Naidu, Bernard Salanie, and Eric Verhoogen, for invaluable contributions. This project was supportedby the Private Enterprise Development in Low-Income Countries exploratory grants (PEDL), Russell Sage SmallGrants in Behavioral Economics, the Center for the Study of Development Strategies at Columbia University.

1

1 Introduction

State capacity has recently become the workhorse of development scholarship(Besley and Persson,

2013). One way in which state capacity can create development is legal capacity. For instance, in a

stateless economy, social institutions can govern agency relations and solve commitment problems

that prevent trade. While social institutions rely on the social structure to monitor and sanction

defectors, the social structure is often fragmented. Hence, social institutions can be inefficient

(Dixit, 2003, Greif, 1993). In this case, creating or expanding the state monopoly of violence and

its legal capacity can increase trade by introducing a commitment device that social institutions

are unable to produce. However, despite its control over violence, the state is often governed by

the interests of administrators (Gonzalez de Lara, Greif, and Jha, 2008, Greif, 2007). Furthermore,

the state system of contract enforcement can undermine the conditions that sustain trade in the

absence of the state, thus crowding-out trade (Benabou and Tirole, 2003, Bowles and Polania-

Reyes, 2012, Kranton, 1996, Lowes, Nunn, Robinson, and Weigel, 2015). Expanding the state may

thus be counter-productive in areas where the state has not previously expanded.

Estimating the impact of the state is challenging. Before states expand, systematic data do

not usually exist. Furthermore, state formation and trading relations are endogenous (Gambetta,

1993, Tilly, 1990). Due to these challenges, the impact of the state remains an empirical question,

unexplored with statistical methods.

In this paper, I create delivery business involving traders and customers in Eastern Democratic

Republic of the Congo (DRC), where the economy has developed in the absence of a functionning

state since its collapse in the nineties, and where social networks and ethnicity govern economic re-

lations (Fund For Peace, 2013, Nest, Grignon, and Kisangani, 2011). The delivery business allows

me to observe the success of economic transactions that would otherwise not occur, and introduce

state contracts using random assignment. A commitment problem is intrinsic to the relationship

between traders and customers, which allows me to compare the impact of the state and of social

institutions on the incentives of the agent, as well as on the willingness of the principal to trust

the agent and engage in trade.

In the first part of the paper, I examine how the introduction of state contracts, which are

previously not used, impacts the behavior of the agents. Traders visit 1,000 customers and of-

fer a household consumption good at a discount (cell phone credit), with the condition that the

2

customer (the agent) promises to pay by cell phone within two days. In collaboration with a

professional lawyer and the Provincial Government, I develop a state contract that exposes a ran-

dom sample of customers to legal action if they fail to pay on time. Randomly introducing state

contracts for customers whose payment data I observe allows me to examine the impact of state

contracts on the behavior of the agent. However, customers can choose to accept or reject the

offer depending on whether they are asked to sign a state contract. This creates a selection bias.

To isolate the incentive effect of state contracts, I implement a design a-la Karlan and Zinman

(2009). While traders require all customers to accept to sign a state contract before they accept

the deal, a random sample of customers who accept the deal gets away without signing a state

contract. Average payment rates are 24% for transactions not involving state contracts, but this

rate is 50% higher for customers who signed the contract.

However, I also find that only the ethnic groups that have captured the administration are

able to enforce contracts. To examine how capture of the state administration by different ethnic

groups impacts contract enforcement, I use random assignment to different ethnic compositions of

the trader customer match. Some traders and customers belong to ethnic groups which have cap-

tured the state administration (Bantu ethnic groups), while the other are excluded from the state

administration (Tutsis). I find that, on average, state contracts increase payment rates by 50%

when used by Bantu traders but have zero effect on payment rates when used by Tutsi traders.

Additional behavioral and survey evidence confirm that customers use the trader’s and customer’s

ethnicity to predict their power to enforcing the state contract, consistent with administrative

capture by Bantus. This behavior provides evidence that administrative power, in particular the

capture of the administration by ethnic groups, governs the enforceability of contracts.

While the results suggest that Bantus are able to enforce state contracts, state contracts may

be too weak to solve commitment problems ex-ante, and may crowd-out informal mechanisms of

contract enforcement. Thus, in the second part of the paper, I examine the effect of state contracts

on trade among different Bantu ethnic groups when agency relations are inherent. To observe the

behavior of the agent, I change the features of the economic activity by implementing a business

extension. I recruit new traders from the population of customers in the first part of the paper,

and deploy them to sell a household consumption good collecting payments before delivery (the

agents). In order to avoid suspicions arising from comparison between the two business activities,

I implement this activity in different areas. This leads me to offer a different consumption good

3

that matches customers’ demands (soaps), and to focus on traders of Bantu ethnic groups for

security reasons. For a randomly selected subset of customers (the principals), the traders offer a

state contract, exposing themselves to legal action if they fail to deliver the good. This allows me

to examine whether principals are more likely to engage in risky trade when state contracts are

made available. If traders are of a different ethnic group than customers and do not offer state

contracts, 40% of customers are willing to accept the trade. In the absence of state contracts,

coethnicity increases the customers’ acceptance rate by 40%. If trader and customer belong to

different ethnic groups, state contracts increase the customers’ acceptance by 40%. This behavior

provides evidence that state contracts and coethnicity are substitutes at generating trade, among

groups that can enforce contracts.

While state contracts and coethnicity increase trade, their effect may reflect that customers

have a preference for the payoff of coethnics or for traders who show contracts. I thus go a step

further and isolate the effect of state contracts and coethnicity on trade through expected behavior

from their effect through preferences. I randomly select customers to two types of sales. In the first

group, the payment is made upon delivery (sales on the spot), while in the second, the payment is

still required in anticipation of delivery within one day (sales on debit). A commitment problem

is only present in the second group of sales. State contracts and coethnicity increase trade when a

commitment problem is present, but have no effect when there is no commitment problem. This

suggests that state contracts and ethnic based social institutions are substitutes at generating

trade because they are substitutes for solving commitment problems inherent in agency relations.

Eastern DRC is a well-suited experimental ground to study the impact of state penetration

into social relations. The Democratic Republic of Congo state is considered a “failed state”, since

its collapse in the nineties.1 To cope with a predatory state and pervasive armed groups, the

economy organized around informal ethnic networks outside the reach of the state’s legal system

(Mathys, 2014). Due to the prevalence of holdup problems that arose in the absence of a function-

ing state, the economy developed around small-scale transactions involving minimal investments

and minimal risk taking (Geenen, 2013, Nest, Grignon, and Kisangani, 2011). However, since its

collapse in the nineties, the Congolese state has recovered state capacity in the East, including a

full-fledged coercive and judiciary system. Thus, expanding the reach of the state as a third-party

contract enforcer can unleash market forces that are currently unexploited .

1Source: Fund For Peace (2013).

4

While a large number of studies documents how individuals and groups solve commitment

problems in transactions taking place outside of a legal framework (Alesina, Baqir, and Easterly,

1999, Greif, 1993, Habyarimana, Humphreys, Posner, and Weinstein, 2007, Hjort, 2013, Miguel

and Gugerty, 2005), this paper is the first to explore the impact of access to the state system of

contract enforcement in a real context aiming to establish causal identification.

This paper complements the literature on contract enforceability by proposing a strategy to

address existing empirical challenges. First, since the choice to use contracts is endogenous, esti-

mating the effect of state contracts in observational studies is affected by endogeneity problems

(Fafchamps, 2000, 2006). I address this challenge by using random assignment. Second, esti-

mating the effect of the social structure is challenging, because social interaction is endogenous

to unobservable characteristics that may predict both the formation of links and the success of

trade (Chandrasekhar, Kinnan, and Larreguy, 2015). To address this challenge, I focus real ethnic

groups, whose divisions have deep historical roots. Drawing on these pre-existing groups, I ran-

domly match traders and customers who have not previously met, but who belong to the different

ethnic groups. This allows me to estimate the impact of ethnic based social institutions on trade,

whether they stem from networks inherited from the past, social norms, or culture, while avoiding

endogeneity issues arising in network formation and matching of partners. Third, since introduc-

ing the requirement to sign contracts can create losses to traders or businesses, researchers usually

draw on laboratory environments to establish causal identification on coethnicity or the impact

of commitment devices (Chandrasekhar, Kinnan, and Larreguy, 2015, Habyarimana, Humphreys,

Posner, and Weinstein, 2007). I complement the existing work by organizing a business with an

inherent agency relation. This allows me to introduce experimentally variations to the features

of the business and to the identity of the traders, and observe the resulting effects in real-world

transactions.

Furthermore, by measuring the impact of administrative capture on contract enforceability,

this paper takes the literature on contract enforceability a step further. In seminal work, a few

scholars have argued that the power of the administrators is a determinant of the behavior of

rulers and states (Gonzalez de Lara, Greif, and Jha, 2008, Greif, 2007). However, administrators

are also required to enforce contracts and thus maintain order necessary for trade. To date, the

administrative foundations of contract enforceability are under-explored. Exploiting the fact that

certain ethnic groups have captured the administration, while other have not, I show that even

5

contract enforceability for politically irrelevant transactions are determined by the power and (eth-

nic) interests of the administration. Although few scholars (Shayo and Zussman, 2011) measure

biases in the judicial system, very little research examines the impact of administrative capture

on economic organization and trade.

This paper’s contribution is also to provide empirical evidence to a theoretically ambiguous

relationship. Existing literature proposes a number of ways in which contracts and social organi-

zation can affect the patterns of trade, and their effect is potentially ambiguous. First, trading

partners who share a social structure may be able to solve commitment problems by exploiting

features of repeated interaction (Greif, 1993, Morjaria and Macchiavello, 2014, Tirole, 1996).2 If re-

peated interaction within groups can sustain cooperation (Alesina and Ferrara, 2004, Dixit, 2003,

Miguel and Gugerty, 2005), third-party contract enforcement can introduce outside options to

trading relationships, thus potentially undermine the conditions that sustain trade within groups

(Kranton, 1996). Since formal contracts potentially solve commitment problems, the overall effect

of contracts is an empirical question. Second, co-ethnics may be able to solve contracting problems

because of parochial altruism (Bernhard, Fischbacher, and Fehr, 2006, Bowles, 2006, Charness and

Rabin, 2002). Chen and Li (2009) draw on social psychology (Tajfel and Turner, 1979) and find

evidence in a laboratory setting that group identities are associated with strong in-group altru-

ism. If social relations are governed by reciprocal social preferences, state penetration can impact

social preferences in unknown ways (Lowes, Nunn, Robinson, and Weigel, 2015). Third, groups

may be able to solve contracting problems by appealing to group norms of behavior, which can be

internalized (Fehr and Gaechter, 1999) or by self-enforcing equilibria (Greif, 1993, Habyarimana,

Humphreys, Posner, and Weinstein, 2007). If the strategic environment in which they operate has

asymmetric information, the introduction of contracts can potentially crowd-out social equilibria

by changing information sets (Benabou and Tirole, 2003, Bowles and Polania-Reyes, 2012, Gneezy

and Rustichini, 2000). In this framework, the impact of contracts is also theoretically ambiguous.

This paper contributes to this literature by providing an empirical estimate of the effect of con-

tracts on incentives to cheat and to trade, and isolating the mechanisms through which they work.

The remainder of the paper is organized as follows. Section 2 presents the context. Section 3

presents the design, econometric strategy, and results of the main field experiment, designed to

2Greif (1993) describes how the threat of collective punishment and concern with reputation sustained trustand trade among Maghrebi traders in eleventh - century Mediterranean trade. McMillan and Woodruff (1999)document the extent to which former relationships predict current supply of risky trade credit among firms inVietnam.

6

estimate the effect of state contracts on incentives to cheat. Section 4 presents the design, econo-

metric strategy, and results of the extension, designed to estimate the effect of state contracts on

willingness to engage in trade when commitment problems are present. Section 5 concludes.

2 Context

The Eastern Province of Sud Kivu in the Democratic Republic of the Congo is composed of Bashis,

Bahavus, Balegas, Batembos, Bafuliros, Pygmies, Tutsis, and to a lesser extent, Hutus.

The history of Tutsis offers one of the major ethnic divides in Eastern Congo. Tutsis of Eastern

Congo are historically cattle herders who migrated from Rwanda and live mostly in the highlands of

the Kivus.3 They self-identify as belonging to Banyarwandas, Banyabwishas, or Banyamulengues

depending on their location and the migration wave from which they arise.4 Tutsi populations in

Eastern Congo are marginalized, often prosecuted. Except for certain battalions of the Congolese

Army, Tutsis are largely excluded from the state administration and the related networks of

patronage. For instance, the state administration of the Province of Sud Kivu is disproportionately

composed of non-Tutsi civil servants. As a result of their vulnerability, the Tutsis of Sud-Kivu

are quasi-nomadic and alternate between urban Tutsi-dominated neighborhoods, the highlands of

Uvira and Fizi, and Rwanda, depending on the security situation.5

The other ethnic groups belong to the Bantu family and self-identify as distinct from the

Tutsis. The languages of Bashis, Bahavus, and Batembos belong to the Shi-Havu language family

and their lexical similarity is 70%.6 Bashis and Bahavus have been politically active in similar

groups and their division is not salient. Batembos, however, are a minority in Eastern Congo, and

have been a major force in local rebellions known as Mayi-Mayi groups. Batembos live mostly in

remote areas of Sud-Kivu, such as Shabunda and western Kalehe (Kalonge, Ziralo and Buloho),

3Tutsis first recorded migration wave took place well before the colonial period, in the seventeenth century,fleeing taxes in Rwanda. A second migration took place in 1880 (today these are called Banyamulengue - thepeople of the Mulengue mountain). The Belgian colonial administration orchestrated subsequent migrations in the1930’s and 1940’s, sending workers to coffee plantations in DRC. In the second half of the twentieth century, largenumbers of Tutsis fled conflicts in Rwanda to the Congo, culminating in the migration of two million refugees inthe 1994 Rwanda crisis - mostly Hutus. This demographic and land pressure exacerbated the ethnic tensions withTutsis.

4Such groups sometimes gather a mix of Rwandophone populations: Tutsi, Hutu, and Twa groups. WhileBanyarwandas are mostly Hutu, Banyamulengues are mostly Tutsi.

5See Stearns (2011), Ngonzola-Ntalaja (2002), and Newbury (1992) for accounts of current ethnic relations inSud-Kivu and their historical foundations.

6https://www.ethnologue.com/language/shr

7

and have a long history of struggle against domination from other groups, in particular from

Bahavus since the 1940’s (Mathys, 2014, Newbury, 1992). Conflicts over land and power involving

the Batembos are still relevant today. Furthermore, Batembos have been particularly affected by

recent waves of armed conflict, in particular by the FDLR and CNDP. Nevertheless, in contrast

to the Tutsi groups, Batembos’ citizenship and access to the state was never contested.7 Balegas,

mostly present in Mwenga and Shabunda, have good relationships with the Bashis and Bahavus.8

Finally, Bafuliros, mostly located in Uvira, consider themselves close related to the Bashis and

Bahavus.

In the first part of the paper, I examine contract enforceability as a function of access to the

state administration, and thus focus on the Tutsi vs. non-Tutsi as the relevant divide. Half the

traders are Bantus and the other half are Tutsis, while customers include Bantus and Tutsis.

I randomly assign traders to customers within lottery bins defined by the lowest administrative

division (urban avenue) and randomize traders to each administrative divisions. This allows me to

examine the impact of contracts on incentive to cheat, and isolate the incentive effect for traders

who have not captured the state administration, the Tutsis, from the rest of traders.

In the second part of the paper, I focus on traders and customers among groups for whom

enforceability of state contracts is credible, Bantus.9 To compare the effect of salient ethnic social

institutions to the effect of state contracts on agency relations, I examine the interactions between

Batembos and the rest of Bantus, as well as the interactions within Bantus, since Batembos are the

salient ethnic divide with deep historical roots. While Batembos have historical grievances mostly

with Bahavus, Bahavus and other Bantu groups (especially Bashis) are almost indistinguishable

to the eye of the Batembo customer. The analysis in the second part thus focuses on interactions

between Batembos and Bashi/Bahavus, although I am able to isolate a Bahavu effect among

Batembos in particular. Due to the obviously fluid meaning of ethnicity across contexts, this

paper does aim at providing an externally valid estimate of the effect of ethnicity. Instead, given

a particular social structure in which there are salient historical divisions, I examine the impact

of state contracts within and across the groups defined by the contextually relevant social divides.

7See Mathys (2014):“Whilst in the case of the [Ba]tembo cited above this led to the emergence of local conflictsand local contestations of belonging, this did not lead to a contestation of the ’ethnic’ (and thus ’civic’) citizenshipof these populations on the national scene.”

8Balegas, Bashis and Bahavus regularly refer to each other as “brothers”.9For security reasons, it was not possible to include Tutsi in the second part of the paper.

8

3 The aministrative foundations of contract enforceability

In this section, I establish that state contracts decrease the incentives to cheat in trading relations

involving agency problems, but only for individuals whose trading partner belongs to an ethnic

group that has captured the state administration. I first present the field experiment design,

then the empirical strategy, and the results. I find that signing a state-backed contract reduces

the agents temptation to renege upon its promise, but this effect depends on ex-post bargaining

power: groups without links to the administration are unable to reduce shirking by requiring a

state contract.

3.1 The home delivery sector: customers as agents

To examine behavior in the presence of agency relations in a real context, I organized a real de-

livery business, in which traders and customers are residual claimants, and I observe the behavior

of traders and customers. I next describe the delivery business.

Traders sell a basic consumption good, door to door, to customers in 1,000 randomly selected

customers of ethnically diverse semi-urban neighborhoods of Bukavu. Traders deliver cell phone

credit in cards on the spot, at a discounted price below the market price. In exchange, the cus-

tomers’ who accept the deal commit to pay by cellphone within two days through the organization’s

central payment system.10 While traders receive a minimal fixed compensation by the research

project which is independent on sales, traders are residual claimants on all sales and derive the

largest part of their daily income through the sale of the cell phone credit. Customers receive no

payment, except the consumer surplus inherent in the price discount of the offer.

Absent enforceable state contracts or social sanctions, it is in the customer’s best interest to

accept the purchase and renege on payment. To randomly selected customers, traders require

the customers to sign a state contract if they want to proceed with the transaction, in which

they expose themselves to legal sanctions if they were to renege payment. The state contract was

drafted by the a local lawyer, and stamped by the Ministry of the Interior. This design allows me

to examine the impact of state contracts on payment rates.11

10I chose cell phone credit cards after implementing a market study. In the areas in which I implemented thestudy, there was compelling evidence that cell phone credit was in excess demand at a reasonable price.

11The contract reads as follows:“ I, the undersigned... , recognize to have received ... cell phone units of the company ... from ... , for a value

of 500 Congolese Francs per unit. I hereby commit to pay ... in exchange of these cell phone units to ... in the

9

The design tackles a selection problem. Some customers anticipate they may not be able to

pay, and the requirement to sign a contract can lead them to refuse the offer. This can create

a selection bias: the customers who have accepted the offer with a requirement to sign the con-

tract are likely different than those who accepted when the requirement to sign a contract was

absent. Comparing payment rates across the two groups would capture both selection and incen-

tive effects. To isolate the effect of contracts on incentives to pay from their effect on selection

of customers into trade, I implement a two-step randomization, analogous to Karlan and Zinman

(2009): traders offer the deal to all customers with the requirement to sign a contract before

obtaining their approval. Among the customers who have accepted the deal, traders withdraw

the contract requirement in a random sample of customers. I thus compare the payment rates of

customers who signed a state contract and those who did not among customers who accepted the

deal. The timing of the transaction is as follows:

Step 1. As part of the sales protocol, and prior to the customers’ decision to accept the pur-

chase, traders introduce the sale and explain the timing of the transaction. Furthermore, ex-ante,

they show and explain the state contract, and announce that the state contract is required to

proceed with the purchase. The state contract stipulates that the customer exposes himself to

legal sanctions if he fails to pay within the agreed timeline (two days). Let Bt=1 ∈ {0; 1} denote

the customer’s decision. Customers buy, Bt=1 = 1 or reject the offer, Bt=1 = 0, and the trader

records the response of the customer before proceeding to the next step.

Step 2a. Among customers who have accepted the offer, traders withdraw the requirement to

sign a contract to a random sample of customers.12 Customers that accept the offer are thus ran-

domly assigned to two groups: those who sign the contract, F = 1, and those who do not sign the

contract, F = 0. Traders also record the final decision of the customer Bt=2. Since withdrawing

the contract requirement is a positive transfer to the customer, Bt=2 = 0 and Bt=2 = 1 is never

observed.

Step 2b. Among customers who have rejected the offer, traders withdraw the requirement to

interval of TWO days at most. I am ready to bring this contract, if necessary, to a legal representative. I recognizethat in case of no payment, I am exposed to the prosecutions and sanctions that the Congolese law considers forthese cases. Done in... . Date ... . Signature of debtor... Signature of creditor... Signature of witness... .”

12The script reads as follows: “I see I do not have enough contracts. It is therefore not necessary to signthis contract and my protocols stipulates that in such cases we shall proceed with the transaction. [paymentinstructions]”. Forgiving the requirement to sign a contract could induce reciprocity, and bias behavior. To avoidthis, the script specified that traders did not have sufficient contracts to pursue with the signature of the contract.Since the traders had already committed to sell the good, the design does not generate compliance problems -withdrawing the contract requirement is a net positive transfer to the customer.

10

sign a contract to a random sample of customers. Traders then allow the customers who refused

the offer initially, Bt=1 = 0, and for whom the contract requirement was withdrawn to reconsider

their decision. Traders also record the final decision of the customer Bt=2.

Step 3. Immediately after, traders supply the cell phone credit cards to the customers who

have accepted the purchase, and provide instructions on how to implement payment by cell phone.

Traders then implement an exit survey to all customers, regardless of whether they have made the

purchase. In addition, to capture social preferences, all customers are also offered phone credit

card on the spot: in exchange for cash, the trader supplies a fixed amount of phone cards on

the spot. The remaining amount needs to be purchased on credit. Figure 1 provides a graphical

representation of the experiment.13

3.2 Theoretical framework

Let Ei,j ∈ Rn×n characterize the ethnic match between trader i and customer j, both of whom

are drawn from an n-dimensional vector of ethnicities. The variables F ∈ {0; 1}, D ∈ {0; 1}

respectively indicate whether the transaction is formalized by a contract, F = 1, and whether

the sale is on debit, D = 1. Let v(Ei,j, F ) be the private valuation of acquiring the good, and

℘ the price. The term l(Ei,j) ∈ R denotes the expected cost of cheating arising from the legal

system. It is a function of the ethnic characteristics of the trader-customer match, since the

ability to enforce legal sanctions may depend on the ethnicity of the trader and customer. Let

P = {0; 1} indicate whether the customer implements the payment, and the term θ(Ei,j, F ) in-

dicate the informal sanctions incurred by the buyer upon reneging payment, P = 0, if he had

finally agreed to make the payment, Bt=2 = 1. I allow θ(Ei,j, F ) to depend on Ei,j, through some

ethnic sanction technology, and F , if complementarities exist between formal and ethnic con-

tracts. The cost θ(Ei,j, F ) could indicate utility losses stemming from internalized social norms,

such as utility losses arising from guilt, social preferences or reference points. It could also be

the product of any social sanctioning technology specific to the ethnic match. Let Bt=2 denote

whether the customer finally buys the good and 0 otherwise (henceforth B). The buyer’s utility

is: uBP = B [v(Ei,j, F )− ℘P − (1− P ) (θ(Ei,j, F ) + Fl(Ei,j))].

Table I maps the parameter space onto the strategy set. There are four possible strategies:

(B = 1, P = 1), (B = 1, P = 0), (B = 0, P = 0), (B = 0, P = 1). I ignore the last strategy,

13Figure A.2 in the online appendix shows the coupon.

11

because it is never observed and because it is strictly dominated. Figure 2 provides a graphical

representation.

The terms αi ∈ {1, 2, 3, 4} denote the mass of agents in cell, defined by observable strategies.

I consider the following partition of the parameter space: cells uniquely identify the strategies

chosen by customers as a function of whether state contracts are used, under the assumption that

state contracts are enforceable. For instance, while α3 and α4 display the same behavior in the

absence of state contracts, α3 values the product enough that he is willing to accept a purchase

that requires him to sign an enforceable state contract. In contrast, α4 would not purchase the

good if he was requested to sign an enforceable state contract that would force him to pay ex-

post.14

Assume that the state contract is sufficient to motivate payment: ℘ < l(Ei,j). In that case, α2

and α4 reject the offer. Thus, if the traders randomly lift the requirement to sign the contract, α3

avoid making a payment. Assuming that formal contracts do not affect the private valuation of

the good or informal sanctions, I can now describe the main testable implications. The functions

v(Ei,j) and θ(Ei,j) then determine the mass of customers in each cell: α1(Ei,j), α2(Ei,j), α3(Ei,j),

α4(Ei,j). Table II presents the selection and incentive effects of state contracts, holding constant

the ethnicity of the match. I next discuss four empirical implications from this setup.

First, contracts have incentive effects (are enforceable) if and only if α3(Ei,j) > 0. Indeed

α3(Ei,j) is the mass of customers who prefer to forgo payment but pay if they were requested to

sign a contract.

Second, Ei,j has a total effect on the incentive effect of state contracts, through v(Ei,j), θ(Ei,j),

and l(Ei,j):∆α3

∆Ei,j= ∆α3

∆v∆v∂Ej

+ ∆α3

∆θ∆θ

∆Ej+ ∆α3

∆l∆l

∆Ej. Since Tutsi traders are excluded from the

state administration but Bantus are not, ex-post, their relative power to enforce a state contract

is weaker than for Bantus, for a given a customer. Thus l(Ei,j)|Ej=Tutsi < l(Ei,j)|Ej=Bantu and

α3(Ei,j)|Ej=Tutsi < α3(Ei,j)|Ej=Bantu if and only if: ∆α3

∆v∆v

∆Ej< ∆α3

∆l∆l

∆Ej+ ∆α3

∆θ∆θ

∆Ej.

Third, if state contracts increase the trader’s ex-post enforcement power, by backward induc-

tion, state contracts should affect the decision of customers to accept the deal, thus potentially

generating selection. Enforceable state contracts are thus a screening device to attract customers

14If the informal cost of reneging payment, θ, is sufficiently high relative to v, I label the customers as “honest.”If the customer’s valuation of the good, v, is higher than the monetary cost, ℘, the customers are “peaches”(lemons otherwise). The terminology captures the notion that trade is socially optimal only when the customerhas a sufficiently high private valuation, which is unobserved. It does not capture whether customers are of highquality, in contrast to the standard adverse selection literature (Akerlof, 1970).

12

with a higher likelihood to pay. To see this, note that pool of customers who accept the offer

when traders do not require a signature on a state contract, α1(Ei,j) +α3(Ei,j) +α4(Ei,j) contains

a larger fraction of customers who will not pay than the pool who accepts the offer when traders

require state contracts, α1(Ei,j) + α3(Ei,j). Furthermore, if state contracts induce a selection of

customers, then conditioning on the sample of customers in which the trader randomly lifts the

contract requirement, payment rates must be higher among those who initially accepted the offer

when the contract was requested α1(Ei,j)+α3(Ei,j), than among those who first reject, but accept

it only when the trader (randomly) lifted the requirement to sign the contract, α4(Ei,j).

Fourth, if state contracts are less enforceable by Tutsi traders, then state contracts are a

weaker screening device when they are used by Tutsi traders. Indeed, customers enter into the

sale depending on θ(Ei,j), v(Ei,j), and l(Ei,j). However, identifying the heterogeneous screening

effect across ethnic groups is challenging. Since formal contracts can affect θ or v, the ethnic

composition of the match has an ambiguous effect on the strengh of the screening device. The

sub-sample of sales on the spot, in which payment is immediate, allows me at least to disentangle

pure social preference channels,∆v(Ei,j ,F )

∆Ej. In sales on the spot, the customer’s decision to buy

depends only on v(Ei,j) and ℘. Therefore, if customers have a preference bias against trading with

Tutsi traders, v(Ei,j)|Ej=Tutsi < v(Ei,j)|Ej=Bantu, the mass of customers in sales on the spot who

accept the sale when the trader is Tutsi must be lower than the mass of those who accept when the

trader is Bantu.15 If customers do not have a preference bias against trading with Tutsi traders,

but if Tutsis are less able to activate legal sanctions, then l|Ej=Tutsi < l|Ej=Bantu (and potentially

θ|Ej=Tutsi < θ|Ej=nonTutsi). If Tutsis and Bantus are able to activate legal sanctions equally, then if

θ is lower for Tutsi traders, contracts will have a stronger screening effect for Tutsi traders – since

in the absence of contracts, there are weaker informal sanctions. If Tutsis and Bantus are able to

activate informal sanctions equally, then if l is lower for Tutsi traders, contracts will have a weaker

screening effects for Tutsi traders – contracts lead to a smaller increase in pool quality, since they

are less likely to be enforced and customers anticipate that. Therefore, in the absence of customers’

preference bias against trading with Tutsis, and in the absence of heterogeneous informal sanction

technologies within Bantus and Tutsis, l is smaller for Tutsi traders if and only if contracts have

a smaller effect on payment when requested by Tutsi traders among Bantu customers.

15This is true even with a homogeneous distribution of preferences within ethnic group. Otherwise, this resultwill depend on the mass of customers whose preference parameter is above the purchasing threshold ℘.

13

3.3 Econometric strategy

Let Bi,j ∈ {0, 1} indicate whether customer i accepts the sale offer from trader j. In sales on

credit, customers make the decision in two steps. Let the dummy B(t = 1)i,j indicate whether

the customer accepts the sale initially (as thet trader announces the state contract is requested),

and the B(t = 2)i,j indicate whether the customer accepts the sale in step 2 (after the trader

implemented the customer level randomization). Let Pi,j ∈ {0, 1} indicate whether the customer

pays. Outcome Pi,j is only observed if the customer accepted the purchase. Let Tj ∈ {0, 1} denote

whether the trader is Tutsi, Fi ∈ {0, 1} whether the trader requires a formal state contract to

customer i, and Ci ∈ {0, 1} whether the sale is credit (payment expected in the future). Each

trader team is composed of a Tutsi and a non-Tutsi trader, who work separately but in the same

administrative unit. I randomly assign teams of two traders to urban avenues. Within each

avenue, I randomly assign traders to customers. Finally, I randomly assign the instruction to lift

the contract requirement at the customer level, using randomization blocks defined by avenue ×

trader. I can thus include avenue ηa and team φe fixed effects to increase precision of the coefficient

estimate, for avenues a = 1, ..., A and teams e = 1, ..., E.

To capture the incentive effects of contracts, I run the following linear probability model:

Pi,j = c0 + c1Fi + c2Tj + c3FiTj + ηa + φe + ei,j (1)

where I condition the sample on Ci,j = 1 and B(t = 1)i,j = 1.16 The parameter c1 captures the

incentive effect of contracts for Bantu traders, while c1+c3 captures the incentive effect of contracts

among Tutsi traders. Contracts are less effective on the agent’s behavior for Tutsi traders than

for Bantu traders if and only if c3 < 0. To measure the customers’ preference bias against Tutsi,

I run the following linear probability model in the sub-sample of sales on the spot:

Bi,j = b0 + b1Tj + ηa + φe + ei,j (2)

where I condition the sample on Ci,j = 0. The parameter b1 captures the mass of customers

who would prefer to purchase if the trader was Tutsi, but not otherwise. There is a preference

bias against Tutsi if and only if b1 < 0. The dummy B(t = 1)i,j ∈ {0, 1} indicates whether the

16Empirically, whenever B(t = 1)i,j = 1, B(t = 2)i,j = 1.

14

customer was initially screened into the pool of buyers willing to purchase the good, despite the

initial request to sign a contract. To capture the screening effects of contracts on customer quality,

I thus run the following linear probability model in the subsample of sales on credit:

Pi,j = d0 + d1B(t = 1)i,j + d2Tj + d3TjB(t = 1)i,j + ηa + φe + ei,j (3)

where I condition the sample on Ci,j = 1 and Fi = 0.17 Since Pi,j is only observed when B(t =

2)i,j = 1, the parameter d1 captures the selection (screening) effect of contracts for Bantu traders

– the difference in payment rates between customers who accepted head on and customers who

accepted only when the state contract was lifted – while d1 + d3 captures the selection effect for

Tutsi traders. Table III presents the testable implications.

3.4 Results: contracts are enforceable, but only by some ethnic groups

I now present the experimental results of the effect of state contracts on incentives for the agent.

I first show the main result that signing state contracts increase incentives to pay when the trader

is Bantu. I then then show that the effect of the contract on incentives to pay depends on the

ethnicity of the trader, consistent with the prior that Tutsi traders have weaker ex-post enforcement

power of state contracts: the impact on customers’ payment rates of signing state contracts on

incentives is smaller if the trader is Tutsi. Additional experimental and survey evidence supports

this interpretation.

3.4.1 Average effect of state contracts on incentives of the agent

On average, 72% of customers are willing to accept the purchase. To examine whether contracts

decrease incentives to cheat, I first condition the analysis the pool of customers who accept the

purchase. Table IV presents the result from econometric specification 1.18 Columns (1)-(4) re-

strict the sample to transactions in which the retailer is Tutsi. Column (1) presents the baseline

specification, without fixed effects. Column (2) includes team fixed effects, column (3) includes

in addition avenue fixed effects, column (4) includes both fixed effects, as well as household-level

17For all three specifications, a conditional logit produces analogous results.18The sample size is smaller than 1,000. This corresponds to the conditioning on the 72% of customers accept

the transaction.

15

controls. As household controls I use the size of the household, as well as a dummy variable

indicating whether the customer purchased the phone credit when offered on the spot.19 Both

variables are aimed to control for unobservable characteristics correlated with purchasing power

in order to increase precision.20 The coefficient on Contract in columns (1) to (4) shows that for

Bantu traders, the requirement to sign a state contract increases the probability that the customer

implements the payment by 38%, and the coefficient statistically significant.

3.4.2 Heterogeneous effects by ethnic composition of the trader and customer match

Table IV also presents the effect of state contracts on payment when retailers are Tutsi, and the

difference in marginal effects of state contracts with transactions in which retailers are not Tutsi.

Columns (1) to (8) show that state contracts have no effect on payment rates when retailers are

Tutsi.21 The controls included in columns (5) to (8) are analogous to columns (1) to (4). Columns

(1) to (4) show that the effect of contracts is different for Tutsi and non Tutsi retailers, and that

the difference is statistically significant. The coefficient on Tutsi is zero, suggesting that among

transactions where no contract is requested, Tutsi traders are equally able to extract payment

from the customers. The coefficient on Contract X Tutsi is negative and statistically significant,

and suggests that the effect of state contracts on payment is 58% weaker for Tutsi retailers.

These results confirm that expanding access to state contracts can be used to discipline in-

centives of individuals who face a commitment problem, even in a weak state. However, signing

a state contracts does not increase payment rates to Tutsi traders. This is consistent with the

prior that the the enforceability of contracts depends on ex-post enforcement power linked to state

capture by members of the traders’ ethnic group.

Despite the results are compelling, there are still competing explanations. Tutsi retailers may

obtain customers of a different type, and these customers may just be less responsive to state

contracts. I next provide additional evidence suggesting that Tutsis have weaker ex-post formal

19The control for purchase on the spot is not endogenous to the treatments, since it precedes the sale on credit.The logic behind the inclusion of this control is that it is akin to a sufficient statistic for purchasing power: itcaptures all variables supposed to affect the willingness to purchase, such as wealth, which are unaffected by thetreatment.

20I am unable to match additional survey measures to the payment and purchase data.21Since the population of Tutsi is very small compared to the rest of ethnic groups, the population average

treatment effect should be computed using weights inversely proportional to the sampling probabilities. Since halfof the selected retailers are Tutsi, Tutsis are thus oversampled. However, the sample average treatment effect allowsme to separately estimate the marginal effect of contracts among non-Tutsi traders, among Tutsi traders, and theirdifference in a transparent way.

16

contract enforcement power and that the results cannot be explained by differential selection of

customers of different types among traders of different ethnicities, that would then respond differ-

ently to state contracts. In what follows, I show that the empirical patterns are only consistent

with the interpretation that Tutsis have weaker ex-post enforcement power of state contracts,

L(Ei,j)|Ej=Tutsi < L(Ei,j)|Ej=nonTutsi.

3.4.3 Mechanism: evidence from self-reported beliefs about contract enforceability

I first describe the customers’ self-reported beliefs about the trader’s ability to enforce the state

contract. Upon completing each transaction, traders asked customers what they thought the con-

sequences would be if the customers would renege making the payment, and traders recorded the

answers from customers. To avoid priming the customer’s answers on the trader’s ethnicity, the

question does not mention ethnicity and focuses instead on the customer’s beliefs about the corre-

sponding trader. To avoid priming the customer’s responses to possible outcomes, the question is

open ended, and the trader records the answer by selecting in a list of possible categories available

on his tablet. I generate a vector of dummies, each indicating whether the customer self-reported

that a given outcome was the likely consequence of non-payment in the open ended question.

Table V uses a linear probability model to regress dummies indicating possible consequences,

on dummies indicating the customer and the trader’s ethnicities.22 Columns (1) to (4) report the

results on dummies indicating the answer of the customer to the following question “According

to you, what consequences will there be if you don’t pay?”. Across columns (1) to (4), I include

all customers, since I am unable to observe whether the customer was required to sign a contract

for the exit survey, and thus the likelihood of legal sanctions is an underestimate of the likelihood

when contracts are signed. Columns (1) to (4) use as dependent variables the following dummy

variables: whether the customer answered that legal sanctions would be activated, whether that

the customer answered that he would experience shame, whether the customer answered that he

would lose friends, whether the customer answered that he would suffer physical violence. Col-

umn (5) uses as dependent variable a dummy indicating whether the customer reports that legal

sanctions will be activated if he does not pay, among customers having signed the formal contract.

Traders asked the following question to customers who signed a contract: “Do you think that the

22The ethnicity of the customer could not be matched to payment data, and hence I cannot use it in the mainresults.

17

state contract you just signed can have judicial consequences if you don’t pay?.”23

Customers expect that traders will be able to enforce state contracts, but they expect Tutsi

traders to have weaker power to do so. Column (1) shows that when asked about the consequences

of reneging payment, 23% of customers report without priming that there would be legal sanctions

if they would not pay.24 The result suggests that Bantu customers are 53% less likely to expect

legal consequences if they would sign a contract when asked by a Tutsi trader, and the difference

is statistically significant. Furthermore, consistent with the prior that Bantu traders have higher

ex-post bargaining power, Bantu traders are expected to be 45% more likely to enforce state con-

tracts among Tutsi customers than among non-Tutsi customers, and the difference is statistically

significant at the 10% level. The effect of customers’ ethnicity drops to zero when the trader is

Tutsi (Tutsi Customer + Tutsi Trader X Tutsi Customer). Columns (2) to (3) suggest that Tutsi

traders are less likely to activate shame or loss of friends among Bantu customers who would

renege payment. Furthermore, consistent with the prior that Tutsis are less able to activate the

legal system and are thus more vulnerable, Tutsi customers fear physical violence if they renege

payments, but only if they renege payment on a Bantu trader. This suggests that Bantu traders

are expected to be able to organize violence with impunity against Tutsi customers to sanction

Tutsi defectors, consistent with the view that Bantu traders have captured the administration.

Finally, column (5) indicates that 70% of Bantu customers who sign a contract believe that there

will likely be legal sanctions if they renege payment. This proportion drops by 14% when the

customer signed a contract for a Tutsi trader, and the decrease is statistically significant. Con-

sistent with the interpretation that traders differ in ex-post contract enforcement power of state

contracts, the coefficient on Tutsi Customer suggests that Tutsi customers who purchased from

a Tutsi retailer are 10% more likely to believe that contracts will lead to legal sanctions than non

Tutsi customers who purchased from a non-Tutsi retailer, although the difference is not statisti-

cally significant. Consistent with the view that Tutsi traders have weaker power to enforce state

contracts, this difference drops to zero when the trader is Tutsi, as evidenced by the coefficient on

Tutsi Trader X Tutsi Customer. Figure 3 provides a graphical representation of these results.

Overall, beliefs about ex-post enforcement power by Tutsi traders perfectly predicts the patterns

23I replicated these regressions with controls for customer’s gender, age, and education and the results areidentical.

24The low proportion reflects that column (1) includes customers who did not accept the purchase, but acceptedthe survey, as well as customers who accepted the purchase, but who ultimately did not sign a state contract.Response rates are close to 100%.

18

of customers behavior.

3.4.4 Mechanism: additional behavioral evidence

I begin by showing that customers do not have a preference bias against Tutsi traders, ruling

out preference mechanisms, where v(Ei,j)|Ej=Tutsi 6= (Ei,j)|Ej=Bantu that might lead to bias if v

is correlated with L. If there is a social preference bias against Tutsis, Tutsi traders will be less

successful at generating purchase among sales on the spot since the customers will attach a weaker

social preference weight on the Tutsi trader’s payoff. Table VI presents the results from econo-

metric specification 2. I implement a linear probability model of whether trade occurs, Tradei,j,

on whether the trader was Tutsi, Tutsij, in sales on the spot. Since I randomize the request to

sign state contracts only after the sales on the spot are implemented, the variable Contract is

orthogonal to whether trade occurs.25 I include team and avenue fixed effects in all specifications.

Columns (1)-(3) show the results respectively on: Contract, Tutsi, and the fully saturated model.

As expected, the coefficient on contract in column (1) is negligible. As seen in column (2), cus-

tomers who receive the offer from a Tutsi trader are 4% less likely to accept the deal. There is

thus no evidence in favor of preference-based discrimination against Tutsi traders. I then turn to

the analysis of screening using the sample of sales on credit.

Having established that customers have no preference bias against Tutsi retailers, v(Ei,j)|Ej=Tutsi =

(Ei,j)|Ej=Bantu, the results from econometric specification 1 are simpler to interpret. Suppose for

simplicity that the absence of mean difference in purchase rates for Tutsi and Bantu traders in

sales on the spot reflects that Tutsi traders do not induce to a mean-preserving shift in the dis-

tribution.26 If contracts are less effective for Tutsi traders, it must be that the distribution of

θ when purchasing from a Tutsi trader first order stochastically dominates the distribution of θ

when purchasing from a Bantu trader, or that the expected costs from legal sanctions L are lower

when purchasing from a Tutsi trader. Table IV showed that customers are equally likely to pay to

Tutsi and Bantu traders when customers do not sign state contracts. This suggests that the mass

of θ above the payment threshold is comparable for Tutsis and Bantus. If the mass of θ above the

payment threshold would have been lower for purchases from Tutsi traders, Tutsi traders would

have extracted lower payment rates. A lower value of legal sanctions L from Tutsi traders should

25I include the variable Contract as a balance test only.26True, for instance, when v follows a homogeneous distribution.

19

also lead to a lower screening effect of contract.

I then exploit the two-step randomization in order to identify the selection effects of contracts,

for each ethnic group of traders. I focus on a sample of customers who purchased the good, but

ended up not signing a state contract as a result of the field experiment’s design. Among these

customers, some accepted the purchase already when a contract was initially requested, while

other (the “opportunists”) first rejected the purchase when the contract was initially requested,

but accepted only when the trader lifted the requirement to sign a contract. I can thus estimate

the effect of selection by comparing the behavior of customers who ultimately accepted the trade,

B(t = 2) = 1, but did not initially accept, B(t = 1) = 0, and did not sign a state contract, to the

behavior of customers who ultimately accepted the trade, B(t = 2), and also initially accepted,

B(t = 1) = 1, and did not sign any state contract.

To identify the selection effects of contracts, I begin by focusing on acceptance. I show that the

decision to ultimately accept the trade, B(t = 2), is negatively affected if traders require a state

contract, but only for Bantu traders. I use a linear probability model to regress a dummy indicat-

ing whether the customer ultimately accepts the deal, B(t = 2), on the following dummy variables:

Contract ultimately requested, Contract, whether the trader is Tutsi, Tutsi, and their interaction,

Contract X Tutsi. Table VII presents the results. Columns (1) to (3) report the average effects

of Contract respectively for the entire sample, transactions of Bantu traders, and transactions of

Tutsi traders. Column (4) reports the results from regressing trade on the trader’s ethnicity and

column (5) reports the fully saturated model. Column (1) shows that customers are 6% more

likely to accept to trade at this second step if the trader lifts the contract requirement, and this

difference is statistically significant – this effect is driven by the customers who initially refused,

but then accepted once the trader lifted the requirement to sign the state contract. Columns (2)

and (3) show that this effect is entirely driven by the transactions of Bantu traders, suggesting

contracts have a screening effect for Bantu traders, but not for Tutsi traders, consistent with the

prior that Tutsi traders are less able to enforce state contracts. Column (4) suggests Tutsi traders

are less likely to achieve successful purchase on average, and the difference is marginally significant.

Column (5) confirms that contracts have a screening effect only for Bantu traders. The coefficient

on Contract shows that contracts have a strong screening effect for Bantu traders: customers

are 9% more likely to accept the purchase if the contract requirement is withdrawn. However,

contracts have no effect on trade when the trader is Tutsi. Indeed, Contract + Contract X Tutsi,

20

which captures the effect of contracts among the Tutsi traders, is not significantly distinct from

zero.

I then identify the selection effects of contracts by focusing on the quality of self-selected cus-

tomers. Among customers who ultimately accepted but did not have to sign a contract, I compare

the payment rates of customers who initially accepted the purchase to customers who initially

rejected it. This obtains the screening effect of contracts on customer’s quality.27 Table VIII

presents the results from the estimation of a linear probability model with avenue and team fixed

effects. Columns (1) to (3) show the baseline specification, and columns (4) to (6) add avenue

and team fixed effects. Columns (1) and (4) focus on the entire sample, columns (2) and (5)

restrict the sample to sales by non-Tutsi traders, and columns (3) and (6) restrict the sample to

sales by Tutsi traders. The variable B(t = 1) indicates whether the customer accepted the initial

offer, in which signing the contract was a requirement. The coefficient on B(t = 1) indicates how

much more likely to pay are customers who were initially screened, holding constant their current

contractual conditions. Columns (1) to (3) show that there is no effect of contract requirement on

the quality of the selected customers. When I add avenue and team fixed effects, the coefficient

in column (4) is positive and significant. Columns (5) and (6) suggest that this effect is entirely

driven by Bantu sales: the coefficient in the Bantu sample is almost identical, while the coefficient

on the Tutsi sample is zero.28

Overall, the evidence presented in this section shows that state contracts are enforceable even

in the Democratic Republic of the Congo, but that contract enforceability depends on the ex-post

enforcement power determined at least in part by the ethnic proximity to the ethnic groups who

have captured the state administration – Bantus. However expanding the state penetration into

social relations may not lead to changes in the patterns of trade, unless contract enforceability is

sufficiently credible that agents are willing to take risks in agency relations when they are pro-

tected by state contracts. Furthermore, the impact of state contracts on trade will depend on

whether they substitute pre-existing social institutions that may already govern agency relations.

27Another possibility is that some of the customers who maintained their rejection after the contract waswithdrawn but would have actually valued the transaction without contract are on average better than those whochanged their mind. In that case, it is possible that the average payment rates among the pool of customers whochange their mind is a biased estimate of the quality of customers who were screened out because of the contracts.The bias would be negative if the customers who change their mind are worse than those who do not dare to changetheir mind despite they would value the transaction.

28These results should be interpreted with caution due to the fact that they rely on the behavior of very fewcustomers.

21

In the next part of the paper, I examine the impact of state contracts on the willingness to engage

in risky trade in the presence of commitment problems, and compare the effect of state contracts

to the effect of salient forms of social organization: coethnicity.

4 Do state contracts substitute for social institutions?

In this section, I establish that state contracts increase trade, and that they do so by solving

commitment problems inherent in agency relations of trade. However, state contracts do not out-

perform existing social institutions of ethnic groups that govern agency relations. State contracts

and ethnic-based mechanisms governing agency relations are substitutes.

4.1 The home delivery sector: customers as principals

To observe the willingness to engage in risky trade in the presence of commitment problems, I

now examine the behavior of the principal. One way to observe this behavior is for the traders to

offer a household good at a discount, with the requirement that customers must pay first, in order

for the trader to deliver the good in the future. To avoid creating confusion among customers

in the areas visited by the activity of section 3, I implement this sale in different areas. Traders

visited 1,700 randomly selected customers in semi-urban areas of Sud Kivu and sold soaps instead.

Soaps are particularly attractive items because they are relatively scarce in such areas.29Traders

offer five soaps to each customer (whose market price is 2.5 USD) for the price of two (1 USD).

After the traders presented the offer and collected the payments, traders invited customers to

take an exit survey. The survey contains information on demographics, ethnicity, and measures

of perceptions. While customers are a different population, their behavior reflects their beliefs

on the same population as in the first part of the paper. Indeed, traders were recruited from the

same neighborhoods in which I implemented the activity in the first part of the paper. This design

allows me to have agents drawn from the same population as the customers in the first part of the

paper.

Traders offer the soaps at a discount and promise to deliver the soaps in the near future (two

days after purchase). However, traders require immediate payment. Customers either accept or

29Soaps are of comparable value in relative terms, and even at market prices, were in excess demand in semi-urban areas of Sud Kivu.

22

reject the offer. If they accept, a transaction occurs and the customer may expect the trader to

deliver the soaps within two days. This design of the sale creates a commitment problem (Greif,

1993, Williamson, 1983), that allows me to observe the behavior of the principals, the customers,

caught in agency relations with the agents, the traders – whereby the traders are drawn from

the same population as the customers and traders of experiment one. In the absence of social

or state-based mechanisms that provide traders with incentives to deliver the soaps, customers

would refuse the offer even if they would prefer to purchase the soaps at that price when no risk

is involved. Figure 4 provides a graphical representation of the experiment setup.

I randomize traders to customers. I require the trader to show and offer to sign a state con-

tract to randomly selected customers as part of the sales protocol and before the customer makes

a decision. The state contract exposes the trader to legal sanctions if he does not deliver the

soap.30 This experiment design allows me to identify the marginal effect on trade of contracts

between coethnics and non-coethnics and their interactions in the presence of commitment prob-

lems. Furthermore, it allows me to identify how coethnicity affects trade. However, if customers

have a taste for coethnics or for trade with contracts, contracts and ethnicity may affect in the

private valuation of customers, and thus their willingness to trade.

To disentangle whether contracts and coethncity increase the customers’ beliefs about the be-

havior of the trader from their effect on preferences, I introduce sales in which traders provide the

soaps at the time of payment, in a random sample of customers (sales on the spot, henceforth).

This allows me to identify the effect of contracts and coethnicity for sales with commitment prob-

lems, and for sales without commitment problems. If contracts and coethnicity improve trust on

the trader, then their marginal effect on trade should be significantly larger for debit sales than for

sales on the spot. Figure 4 provides a graphical representation of the experiment setup. Table IX

shows the factorial design of the experiment.

30A local lawyer drafted the contract, and the contract was endorsed and stamped by the Ministry of Interiorof Sud Kivu. Traders carried copies of the original contract. The contract stipulates the following: “I, theundersigned... , recognize to have received... Congolese Francs from... in anticipation of ... soaps of type... ,for a value of 200 Congolese Francs per unit. I hereby commit to deliver... soaps of the type... and of value 200Congolese Francs per unit to... in the interval of TWO days at most. I am ready to bring this contract, if necessary,to a legal representative. I recognize that in case of no delivery, I am exposed to the prosecutions and sanctions thatthe Congolese law considers for these cases. Done in... . Date... . Signature of debtor... Signature of creditor...Signature of witness... .” Figure A.1 shows the state contract with its visible stamp of the Ministry of the Interior.

23

4.2 Theoretical framework

The customer’s utility to depends on the monetary payoff of the trader, through a separable social

preference parameter a-la Charness and Rabin (2002). The social preference weight indicates that

the customer values the trader’s payoff, but it could also indicate dis-utility from violating a social

norm of surplus-sharing with the trader. Legal and social sanctions of traders if they renege on

a promise. The term λ(Ei,j, F ) is the weight that the buyer assigns to the monetary payoff of

the trader. I allow the social preference weight to depend on the trader’s ethnic characteristic

of the match, Ei,j, and whether the sale is formalized, F . I normalize the customer’s utility of

not purchasing to 0. The dummy D indicates whether the sale is on debit. It takes value 1 if

the trader promises to deliver the soaps in the future. Customers discount the future value of

consumption by their subjective probability that the trader will deliver the good η (Ei,j, F,D) and

by their discount factor, β (Ei,j, F,D). The subjective probability of delivery and the discount

factor depend on the social (ethnic) proximity and formalization. For instance, customers may

be more impatient with delivery lags by non-coethnics traders and when the trader signed a

contract; in addition, customers may assign a different probability of delivery to coethnics and

to sales that are formalized with a contract. In sales on the spot, there is no uncertainty of

whether the trader will provide the good, η (Ei,j, F,D = 0) = 1, ∀E,F and since the delivery is

immediate β (Ei,j, F,D = 0) = 1, ∀Ei,j, F . Therefore, in what follows, for both functions, I omit

the argument D. The customer’s utility is:

UB = B(β(Ei,j, F )η (Ei,j, F ) v − ℘+ λ(Ei,j, F )℘)

When sales on the spot, η (Ei,j, F ) = 1. The buyer’s utility is: UB = B(v−℘+ λ(Ei,j, F )(℘−

℘a)). The marginal effect of ethnic proximity for sales on the spot, when a contract is not used

is: ∂UB

∂Ei,j|F=0,D=0 = ∂UB

∂λ(Ei,j ,F )

∂λ(Ei,j ,F )

∂Ei,j|F=0,D=0. When the buyer has ethnically biased social prefer-

ences,∂λ(Ei,j ,F,C)

∂Ei,j|F=0 > 0, coethnicity in sales on the spot increases the buyer valuation of the

transaction, since the trader is making positive profit. The marginal effect of contracts when a sale

is with a non-coethnic and on the spot is: ∂UB

∂F|Ei 6=Ej ,D=0 = (℘−℘a)∂λ(Ei,j ,F )

∂F|Ei 6=Ej

. If the customer

has a preference for the payoff of traders who formalize transactions,∂λ(Ei,j ,F,C)

∂F |Ei 6=Ej> 0, the

use of contracts in sales on the spot increases the buyer’s valuation of the purchase. If contracts

crowd-out ethnic social preferences −∂λ(Ei,j ,F )∂F |Ei=Ej

> −∂λ(Ei,j ,F )∂F |Ei 6=Ej

, the effect of con-

24

tracts on trade is negative. Allowing some transactions to be on the spot allows to separate this

effect of contracts from their effect on expectations of trader’s delivery. I next provide propositions

to guide the empirical analysis.31

Proposition 1. Introducing sale on debit reduces the expected payoff customers derive from the

deal. This effect is a function of both the discount factor and the subjective probability that the

trader will deliver the good as promised when the trader promises to deliver the good in the future.

Proposition 2. If the time preference parameter is independent of the characteristics of the trader,

the effect of coethnicity in sales on the spot is smaller than the effect of coethnicity in sales on

debit, if and only if coethnicity increases the subjective probability of delivery, η(Ei,j, F ).

Proposition 3. If the time preference parameter is independent of whether the trader signs a

state contract, the effect of contracts among sales on the spot is smaller than the effect of contracts

among sales on debit if and only if contracts increase the subjective probability of delivery, η(Ei,j, F )

(η(Ei,j, F = 1) > η(Ei,j, F = 0), ∀Ei,j).

Proposition 4. Contracts affect less the impact of uncertain delivery on trade among coethnics

than they affect the impact of uncertain delivery on trade among non-coethnics, if and only if:

η (Ei 6= Ej, F = 0)− η (Ei 6= Ej, F = 1)− (η (Ei = Ej, F = 0)− η (Ei = Ej, F = 1)) < 0

4.3 Econometric strategy

The outcome is whether customer i visited by trader j buys the soaps, Tradei,j ∈ {0; 1}. I

refer to this event as “trade occurs.” To identify the effect of individual treatments, I compare

trading rates in each cell of Table IX. While the traders offer prices below market prices, I expect

customers to reject the offer for multiple reasons, including liquidity constraints. Since treatments

are randomized, the unobservables affecting rejection are orthogonal to the treatments, so the

sample difference of means is an unbiased estimate of the population effect of contracts.

I estimate the average treatment effect using a linear probability model in all regressions

below.32 Let Ei,j ∈ {0; 1} denote whether the interaction is coethnic, Fi,j ∈ {0; 1} whether formal

31Proofs are in the online appendix32Results using conditional logit are analogous. I use linear probability model for simplicity of interpretation.

25

contracts are used, and Ci,j ∈ {0; 1} whether the sale is made on debit, all of which are randomized

at the dyad level. In addition, let Tvt be a vector of separable village and trader fixed effects:

Tradei,j = a0 +a1Ei,j+a2Fi,j+a3Di,j+a4Ei,jFi,j+a5Ei,jDi,j+a6Fi,jDi,j+a7Ei,jFi,jDi,j+Tvt+ei,j

(4)

Table X presents the testable implications.

4.4 Results: state contracts and coethnicity as substitutes

I first estimate the effect of contracts and coethnicity on trade. I then isolate their effects on trade

that stem from shifts in the principal’s expecations about the agent’s behavior.

4.4.1 Effect of contracts and coethnicity on trade in the presence of agency relations

Figure 5 presents the main result. When transactions do not use state contracts, 48% of non-

coethnics trade, against 62% among coethnics and this difference is statistically significant. If,

however, the trader uses a state contract, there is no statistically distinguishable difference between

the rate of successful trades of coethnics and non-coethnics. The proportion of non-coethnics who

accepts the sale rises from 48% to 69%, while the proportion of coethnic customers who accepts the

deal remains unchanged. A linear probability model suggests that the effect of contracts among

non-coethnics is statistically significantly different from the effect of contracts among coethnics.

4.4.2 Mechanisms: contracts shift principal’s expectations about agents’ behavior

To elicit whether contracts and coethnicity affect the expectations of traders’ delivery, I introduce

sales on the spot. Some customers could have social preferences that are biased in favor of

coethnics ( ∂λ∂E

> 0) or in favor of traders who reveal to have state-backed contracts ( ∂λ∂F

> 0).

Preference mechanisms could lead contracts and coethnicity to increase trade. To isolate the

effects of contracts and coethnicity on expectations about the behavior of the trader, I randomly

assigned whether the sale was made on debit or on the spot.

Table XI reports the econometric results. I use a linear probability model with village fixed

effects to regress whether trade occurs in a fully saturated model. Column (1) presents the main

effect of sale on debit on trade in the whole sample. Column (2) restricts the sample to sales

26

made on the spot and reports the main effect of Contract. Column (3) restricts the sample to

sales made on debit and reports the main effect of Contract. In columns (4) and (5), I restrict

the sample similarly, but focus on the effect of Coethnicity. Column (6) presents the coefficients

in the fully saturated model and Column (7) adds household level controls to the fully saturated

model (age of customer, age of customer squared, and number of children in the household as a

proxy for household wealth).

The coefficient on Sale on Debit suggests that the proportion of customers who accepts the

trade decreases by 21% when delivery is not immediate. Columns (2) and (3) show that contracts

increase trade when delivery is in the future, but not when delivery is on the spot. The coefficient

on Contract is insignificantly different from zero for trades on the spot. However, column (3)

shows that for debit sales, the proportion of customers who accepts the trade increases by 16%

the trader signs a contract Columns (4) and (5) provide similar results for coethnicity.

Column (6) shows the fully specified model. Contract and Coethnicity increase trade, only

when trade is on debit. Indeed, the coefficients on Sale on Debit X Contract and Sale on Debit X

Coethnic are respectively .28 and .29 and are statistically significant. However, the coefficients on

Contract and Coethnic are negative and insignificant, suggesting they do not affect trade when sale

is on the spot. Finally the coefficient on the triple interaction Sale on Debit X Contract X Coethnic

is negative marginally significant, and equal to −.28. This suggests that contracts have a weaker

effect on expectations of delivery among coethnics. Furthermore, inspection of the coefficient

magnitude suggests that contracts do not change expectations of delivery among coethnics: the

magnitude is exactly the inverse of the Sale on Debit X Contract coefficient. Results are unchanged

when I add household level controls in Column (7). This section thus established that contracts

and coethnicity increase trade, because they solve trade related commitment problems.

5 Conclusion

While the impact of the state on economic development has attracted scholars of all social sciences

(Bates, 2011), there is little statistical evidence of economic activity in the absence of the state and

the impact of the state instruments of legal enforcement on economic organization. I implement

a field experiment in East Congo, where the state is relatively absent, and I am able to estimate

the impact of expanding the state.

27

My results suggest that expanding access to the state judicial system by providing contracts

has potentially large welfare gains, especially in populations where social groups are fragmented.

Furthermore, my findings suggest that the design of interventions that ignore the allocation of

material power and the political equilibrium may be misguided (Acemoglu and Robinson, 2013).

State-backed contracts activate threats of judiciary sanctions enforced by the state. However, the

state is embedded in a network of social relations and some groups have captured the state.

HARVARD UNIVERSITY

SUPPLEMENTARY MATERIAL

An online appendix for this article can be found as a separate file.

References

Acemoglu, D., and J. A. Robinson (2013): “Economics versus Politics: Pitfalls of Policy

Advice,” Journal of Economic Perspectives, 27(2), 173–92.

Akerlof, G. A. (1970): “The Market for ’Lemons’: Quality Uncertainty and the Market Mech-

anism,” The Quarterly Journal of Economics, 84(3), 488–500.

Alesina, A., R. Baqir, and W. Easterly (1999): “Public Goods and Ethnic Divisions,” The

Quarterly Journal of Economics, 114(4), 1243–1284.

Alesina, A., and E. L. Ferrara (2004): “Ethnic Diversity and Economic Performance,”

NBER Working Papers 10313, National Bureau of Economic Research, Inc.

Bates, R. (2011): Prosperity and Violence. Cambridge University Press, Cambridge.

Benabou, R., and J. Tirole (2003): “Intrinsic and Extrinsic Motivation,” The Review of

Economic Studies, 70(3), 489–520.

Bernhard, H., U. Fischbacher, and E. Fehr (2006): “Parochial altruism in humans,”

Nature, 442(7105), 912–915.

28

Besley, T., and T. Persson (2013): “Taxation and Development,” STICERD - Economic

Organisation and Public Policy Discussion Papers Series 41, Suntory and Toyota International

Centres for Economics and Related Disciplines, LSE.

Bowles, S. (2006): “Group competition, reproductive leveling, and the evolution of human

altruism,” science, 314(5805), 1569–1572.

Bowles, S., and S. Polania-Reyes (2012): “Economic Incentives and Social Preferences:

Substitutes or Complements?,” Journal of Economic Literature, 50(2), 368–425.

Chandrasekhar, A., C. Kinnan, and H. Larreguy (2015): “Social Networks As Contract

Enforcement: Evidence From A Lab Experiment In The Field,” .

Charness, G., and M. Rabin (2002): “Understanding Social Preferences with Simple Tests,”

The Quarterly Journal of Economics, 117(3), 817–869.

Chen, Y., and S. X. Li (2009): “Group Identity and Social Preferences,” American Economic

Review, 99(1), 431–57.

Dixit, A. (2003): “Trade Expansion and Contract Enforcement,” Journal of Political Economy,

111(6), 1293–1317.

Fafchamps, M. (2000): “Ethnicity and Credit in African Manufacturing,” Journal of Develop-

ment Economics, 61(1), 205–235.

(2006): Spontaneous Markets, Networks and Social Capital: Lessons from Africa. Oxford,

address =.

Fehr, E., and S. Gaechter (1999): “Cooperation and Punishment in Public Goods Experi-

ments,” Discussion paper.

Fund For Peace (2013): “Failed states Index,” Policy report, Fund For Peace.

Gambetta, D. (1993): Sicilan Mafia: the Business of Private Protection. Harvard University

Press.

Geenen, S. (2013): “Who Seeks, Finds: How Artisanal Miners and Traders Benefit from Gold in

the Eastern Democratic Republic of Congo,” European Journal of Development Research, 25,

197 – 212.

29

Gneezy, U., and A. Rustichini (2000): “A Fine is a Price,” Journal of Legal Studies, 29(1),

1285–1320.

Gonzalez de Lara, Y., A. Greif, and S. Jha (2008): “The Administrative Foundations of

Self-Enforcing Constitutions,” American Economic Review, 98(2), 105–09.

Greif, A. (1993): “Contract Enforceability and Economic Institutions in Early Trade: the

Maghribi Traders’ Coalition,” American Economic Review, 83(3), 525–48.

(2007): “The Impact of Administrative Power on Political and Economic Development:

Toward Political Economy of Implementation,” in Institutions and economic performance, pp.

17–63. Harvard Univ. Press.

Habyarimana, J., M. Humphreys, D. N. Posner, and J. M. Weinstein (2007): “Why

Does Ethnic Diversity Undermine Public Goods Provision?,” American Political Science Review,

null, 709–725.

Hjort, J. (2013): “Ethnic Divisions and Production in Firms,” CESifo Working Paper Series,

(4449), 1243–1284.

Karlan, D., and J. Zinman (2009): “Observing Unobservables: Identifying Information Asym-

metries With a Consumer Credit Field Experiment,” Econometrica, 77(6), 1993–2008.

Kranton, R. E. (1996): “Reciprocal Exchange: A Self-Sustaining System,” American Economic

Review, 86(4), 830–51.

Lowes, S., N. Nunn, J. Robinson, and J. Weigel (2015): “The Evolution of Culture and

Institutions: Evidence from the Kuba Kingdom,” .

Mathys, G. (2014): “People on the move: Drontiers, borders, mobility and history in the Lake

Kivu region 19th-20th century,” Doctoral dissertation.

McMillan, J., and C. Woodruff (1999): “Interfirm Relationships And Informal Credit In

Vietnam,” The Quarterly Journal of Economics, 114(4), 1285–1320.

Miguel, E., and M. K. Gugerty (2005): “Ethnic diversity, social sanctions, and public goods

in Kenya,” Journal of Public Economics, 89(11-12), 2325–2368.

30

Morjaria, A., and R. Macchiavello (2014): “Competition and Relational Contracts: Evi-

dence from Rwanda Coffee Mills,” Job market paper.

Nest, M. W., F. Grignon, and E. F. Kisangani (2011): The Democratic Republic of Congo:

Economic Dimensions of War and Peace, International Peace Academy occasional paper series.

Lynne Rienner Publishers, London.

Newbury, D. (1992): Kings and Clans: Ijwi Island and the Lake Kivu Rift, 17801840. University

of Wisconsin Press; 1St Edition editio.

Ngonzola-Ntalaja, G. (2002): The Congo From Leopold to Kabila: A people’s history. Zed

Books, London and New York.

Sanchez de la Sierra, R. (2014): “Are Monopolies of Violence States?,” Unpublished.

Shayo, M., and A. Zussman (2011): “Judicial Ingroup Bias in the Shadow of Terrorism,” The

Quarterly Journal of Economics, 126(3), 1447–1484.

Stearns, J. (2011): Dancing in the Glory of Monsters. Public Affairs, New York, NY.

Tajfel, H., and J. C. Turner (1979): An integrative theory of intergroup conflict. W. G.

Austin and S. Worchel. Monterey, CA: Brooks/Cole, New York.

Tilly, C. (1990): Coercion, Capital and European States. Blackwell, Cambridge, MA.

Tirole, J. (1996): “A Theory of Collective Reputations (with Applications to the Persistence of

Corruption and to Firm Quality),” Review of Economic Studies, 63(1), 1–22.

Williamson, O. E. (1983): “Credible Commitments: Using Hostages to Support Exchange,”

American Economic Review, 73(4), 519–40.

31

Tables

Table I: Characterization of the four types of customers: customers as agents

Strategies B = 1, P = 1 B = 0, P = 0 B = 1, P = 0

Preference ordering 1, 1 � 1, 0 0, 0 � 1, 0 1, 0 � 1, 11, 1 � 0, 0 0, 0 � 1, 1 1, 0 � 0, 0

1, 0 � 1, 1 � 0, 0 1, 0 � 0, 0 � 1, 1

Conditions θ + Fl > ℘ θ + Fl > v ℘ > θ + Fl ℘ > θ + Flv > ℘ ℘ > v v > θ + Fl v > θ + Fl

v > ℘ v < ℘

Mass α1 α2 α3 α4

Label Honest peaches Honest lemons Moral Hazards Dishonest lemons

Notes: This table characterizes the strategies in the parameter space in the absence of legal contracts. The firstline presents the strategies. The second line presents the preference orderings associated to the correspondingstrategies. Line three presents the parameter relationships implied by the observable strategies. The terms in linefour, αi ∈ 1, 2, 3, 4, are the mass of agents in each of the cells defined by observable strategies. For instance, whileα3 and α4 display the same behavior when no state contract is involved, α3 values the good enough that he wouldbe willing to accept a deal even when he is required to sign an enforceable contract, while α4 would not. I labeleach mass according to two dimensions. First, if the cost of not paying, θ, is sufficiently high, compared to theindividual’s private valuation v of consuming the good, I label them honest. This reflects that if they buy, theywould never renege payment. Second, if the individual’s private valuation of consuming the good, v, is higher thanthe monetary cost, I label them peach. If the individual’s private valuation of consuming the good, v, is lowerthan the monetary cost, ℘, I label them lemon. The peach/lemon terminology captures that trade is only sociallyoptimal when the customer has a sufficiently high private valuation, which is unobserved.

32

Table II: Screening and Incentives: customers as agents

Step 1: Selection F=1

B = 0 B = 1

Who buys? α2 + α4 α1 + α3

Step 2: Incentives F=0 F=1

Who pays? (P = 1) α1 α1 + α3

Notes: This table presents the selection and incentives effects when customers are agents. To disentangle contractincentive effects from their effect on selection, prior to customer’s decision, traders request customers to sign acontract guaranteeing that they will pay within 2 days by cellphone. Individuals in the groups α2 and α4 rejectthe offer, since they don’t value the good enough so as to pay for it given the expectation of contract enforceability.Once customers have self-selected, at the time of the transaction, the trader announces in a randomly selectedsubset of customers that he cannot offer a contract because he does not have enough contracts, F = 0, while theremaining are still required to sign the contract to proceed with the transaction. The last line shows the mass ofcustomers who pay. Among the customers in which the contract requirement was withdrawn at step 2 (F = 0), α3

prefers to avoid payment, while α1 still prefer to pay, despite they ended up not signing a contract.

33

Table III: Testable Implications: cutomers as agents

Hypothesis Testable implication

Preferences-based ethnic bias b1 < 0

Contracts have incentive effects when trader is non-Tutsi c1 > 0

Contracts have incentive effects when trader is Tutsi c1 + c3 > 0

Contracts have screening effects when trader is non-Tutsi d1 > 0

Contracts have screening effects when trader is Tutsi d1 + d3 > 0

Contracts have stronger incentive effects when trader is non-Tutsi d1 + d3 > 0

Smaller screening effect for Tutsi traders d3 < 0

Notes: This table presents the testable implications when customers are principals. Sales are either on credit oron the spot. In sales on credit, the trader first provides the good, and asks the customer to pay in the future bycellphone. In sales on the spot, the trader provides the good on the spot immediately upon receiving payment.The left column describes the hypothesis. The right column describes the testable implication in the frameworkof the econometric specification. In particular, it indicates the implied sign of the parameter in the correspondingspecification. Let Bi,j ∈ {0, 1} indicate whether the customer accepts to buy. For sales on credit, B(t = 1)i,jindicates whether the customer accepted the sale initially (when signing the contract was requested) and B(t = 2)i,jindicates whether the customer accepted the sale after the randomization was implemented and they were askedto reconsider their choice. Let Pi,j ∈ {0, 1} indicate whether the customer pays for the transaction. This is onlyobserved if the customer accepted the purchase. Let Rj ∈ {0, 1} denote whether the trader is Tutsi, Fi,j ∈ {0, 1}whether formal contracts are used, and Di,j ∈ {0, 1} whether the sale is made on credit. In addition, let Ta ∈ {0, 1}be avenue fixed effects and Te ∈ {0, 1} denote team fixed effects. Trader teams of two are randomly assignedto avenues. Within each avenue, traders are randomly assigned to customers. Finally the contract treatment israndomly assigned within avenue for each trader. The linear probability model specifications are as follows. Tocapture the incentive effects: Pi,j = c0 + c1Fi,j + c2Ri,j + c3Fi,jRi,j +Aa +Te +ei,j where I condition the sample onDi,j = 1 and B(t = 1)i,j = 1. To capture ethnic preferences: Bi,j = b0+b1Ri,j +Aa+Te+ei,j where I condition thesample on Di,j = 0. To capture the screening effects of contracts: Pi,j = d0 + d1B(t = 1)i,j + d2Ri,j + d3Ri,jB(t =1)i,j +Aa + Te + ei,j where I condition the sample on Di,j = 1 and Fi,j = 0.

34

Table IV: Effect of trader’s ethnicity on incentive effects of contracts: customers as agents

(1) (2) (3) (4) (5) (6) (7) (8)VARIABLES Pay Pay Pay Pay Pay Pay Pay Pay

Contract 0.09** 0.08* 0.09** 0.09** -0.06 -0.07 -0.07 -0.06(0.04) (0.04) (0.04) (0.04) (0.05) (0.05) (0.05) (0.05)

Tutsi -0.01 0.00 0.01 0.01(0.04) (0.04) (0.04) (0.04)

Contract X Tutsi -0.14** -0.16** -0.16** -0.15**(0.06) (0.06) (0.06) (0.06)

Constant 0.24*** 0.41*** 0.44* 0.48** 0.24*** 0.44*** 0.44 0.52*(0.03) (0.07) (0.22) (0.23) (0.03) (0.09) (0.27) (0.29)

Observations 668 668 668 667 295 295 295 294R-squared 0.18 0.17 0.22 0.22 0.18 0.13 0.19 0.20Team FE NO YES YES YES NO YES YES YESAvenue FE NO NO YES YES NO NO YES YESHousehold controls NO NO NO YES NO NO NO YES

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: Table IV presents the effect of state contracts when traders are Tutsi, and the difference in marginal effects with transactions in which retailers arenot Tutsi. Columns (1)-(4) restrict the sample to transactions in which the retailer is Tutsi, and columns (5)-(8) present the sample average treatment effectsincluding all transactions. Since the population of Tutsi is very small compared to the rest of ethnic groups, the population average treatment effect shouldbe computed using weights inversely proportional to the sampling probabilities. Since half of the selected retailers are Tutsi, Tutsis are thus over-sampled.However, the sample average treatment effect allows me to separately estimate the marginal effect of contracts among non-Tutsi traders, among Tutsi traders,and their difference. Column (1) presents the baseline specification, column (2) adds team fixed effects, column (3) adds avenue fixed effects in addition tothe team fixed effects, and column (4) adds household controls. Similarly, column (5) presents the baseline specification with the fully saturated interactionswith contract and Tutsi dummy, column (6) adds team fixed effects, column (7) adds avenue fixed effects in addition to the team fixed effects, and column(8) adds household controls.

35

Table V: Belief about contract enforceability: customers as agents(1) (2) (3) (4) (5)

Legal Shame Loss of Physical Legalsanctions friends violence sanctions

VARIABLES yes/no

Tutsi Trader -0.127*** -0.0837*** 0.0872*** 0.00513 -0.108***(0.0256) (0.0231) (0.0197) (0.0125) (0.0383)

Tutsi Customer 0.109* 0.0150 -0.0192 0.0983*** 0.0744(0.0615) (0.0553) (0.0472) (0.0301) (0.0901)

Tutsi Trader X Tutsi Customer -0.0985 0.0117 0.0700 -0.123*** -0.0737(0.0749) (0.0674) (0.0575) (0.0366) (0.110)

Constant 0.224*** 0.165*** 0.0449*** 0.0299*** 0.668***(0.0170) (0.0153) (0.0131) (0.00834) (0.0245)

Observations 971 971 971 971 764R-squared 0.035 0.015 0.032 0.013 0.014

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results from a linear probability model to regress dummies indicating possibleconsequences, on dummies indicating the customer and the trader’s ethnicities. Columns (1) to (4) report theresults on dummies indicating the answer of the household to the following question “According to you, whatconsequences will there be if you don’t pay?” Across columns (1) to (4), I include all customers, since I am unableto observe whether the household was required to sign a contract for the exit survey, and thus the likelihood of legalsanctions is an underestimate of the likelihood when contracts are signed. Columns (1) to (4) use as dependentvariables the following dummy variables: whether the customer answered that legal sanctions would be activated,whether that the customer answered that he would experience shame, whether the customer answered that hewould lose friends, whether the customer answered that he would suffer physical violence. Column (5) uses asdependent variable a dummy indicating whether the customer reports that legal sanctions will be activated if hedoes not pay, among customers having signed the formal contract. Traders asked the following question: “Do youthink that the state contract I have shown you can have judicial consequences if you don’t pay?” and positiveswere coded as 1.

36

Table VI: Preference against Tutsis, sales on the spot: customers as agents

(1) (2) (3)VARIABLES Trade Trade Trade

Contract -0.0328 -0.0514(0.0271) (0.0369)

Tutsi -0.0426 -0.0621(0.0272) (0.0379)

Contract X Tutsi 0.0391(0.0541)

Constant 0.0962 0.101 0.127(0.152) (0.152) (0.153)

Observations 1,009 1,009 1,009R-squared 0.201 0.202 0.203

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: All regressions consider only sales on the spot, in which the trader provided the good on the spot inexchange of immediate payment. I regress a dummy variable indicating whether the trade occurred (Trade) on thedummy Contract, Tutsi, and their interaction. Contract indicates whether the household ultimately was requestedto sign the contract in which he commits to pay by cell phone. Randomization was implemented as randomlywithdrawing the requirement to sign the contract in some households after they had accepted the deal (the scriptspecifies that the trader does not have enough contracts to allow him to request a contract here). Tutsi indicateswhether the trader is Tutsi, and Contract X Tutsi is their interaction. In column (1) I regress Trade on Contractonly. Column (2) I regress Trade on Tutsi only. In column (3) I report the fully saturated regression model. Allregressions are a linear probability model with avenue and team fixed effects. Mixed ethnic two-person trader teamswere randomly allocated to avenues and households within avenues were randomly allocated to each trader. Thecontract randomization was implemented within trader. Since traders’ ethnicity is fixed within trader, I do notinclude traders’ fixed effects. The variable Tutsi does not decrease Trade when sales are on the spot, at conventionallevels of statistical significance. This suggests absence of preference-based ethnic bias against Tutsis.

37

Table VII: Selection effect of contracts on purchase, by traders’ ethnicity: customers as agents

(1) (2) (3) (4) (5)VARIABLES Trade Trade Trade Trade Trade

Contract -0.0568** -0.0958*** -0.0159 -0.0848**(0.0271) (0.0355) (0.0425) (0.0365)

Tutsi -0.0463* -0.0770**(0.0275) (0.0383)

Contract X Tutsi 0.0622(0.0543)

Constant 1.185*** 1.214*** 1.286*** 1.181*** 1.224***(0.156) (0.196) (0.203) (0.156) (0.157)

Observations 960 525 435 960 960R-squared 0.192 0.241 0.212 0.191 0.196Traders ALL Non TUTSI TUTSI ALL ALL

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents a linear probability model to regress a dummy indicating whether the household ulti-mately accepts the deal, B(t = 2), on the following dummy variables: Contract ultimately requested, Contract,whether the trader is Tutsi, Tutsi, and their interaction, Contract X Tutsi. Columns (1) to (3) report the averageeffects of Contract respectively for the entire sample, transactions of Bantu traders, and transactions of Tutsitraders. Column (4) reports the results from regressing trade on the trader’s ethnicity and column (5) reportsthe fully saturated model. Column (1) shows that customers are 6% more likely to accept to trade at this secondstep if the trader lifts the contract requirement, and this difference is statistically significant – this effect is drivenby the customers who initially refused, but then accepted once the trader lifted the requirement to sign the statecontract. Columns (2) and (3) show that this effect is entirely driven by the transactions of Bantu traders, sug-gesting contracts have a screening effect for Bantu traders, but not for Tutsi traders, consistent with the prior thatTutsi traders are less able to enforce state contracts. Column (4) suggests Tutsi traders are less likely to achievesuccessful purchase on average, and the difference is marginally significant. Column (5) confirms that contractshave a screening effect only for Bantu traders. The coefficient on Contract shows that contracts have a strongscreening effect for Bantu traders: customers are 9% more likely to accept the purchase if the contract requirementis withdrawn. However, contracts have no effect on trade when the trader is Tutsi. Indeed, the linear combinationContract + Contract X Tutsi, which captures the effect of contracts among the Tutsi traders, is not significantlydistinct from zero.

38

Table VIII: Selection effect of contracts on customer quality, by traders’ ethnicity: customers asagents

(1) (2) (3) (4) (5) (6)VARIABLES Pay Pay Pay Pay Pay Pay

B(t=1) 0.246 0.244 0.248 0.346* 0.386* 0(0.193) (0.216) (0.435) (0.181) (0.213) (0.409)

Constant 0 0 0 -0.0621 -0.0353 0.429(0.192) (0.214) (0.433) (0.207) (0.219) (0.419)

Observations 355 201 154 355 201 154R-squared 0.005 0.006 0.002 0.268 0.377 0.375Team FE YES YES YES YES YES YESAvenue FE YES YES YES YES YES YESTraders ALL NON TUTSI TUTSI ALL NON TUTSI TUTSI

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results from the estimation of a linear probability model with avenue and teamfixed effects. Columns (1) to (3) show the baseline specification, and columns (4) to (6) add avenue and team fixedeffects. Columns (1) and (4) focus on the entire sample, columns (2) and (5) restrict the sample to sales by non-Tutsi traders, and columns (3) and (6) restrict the sample to sales by Tutsi traders. The variable B(t = 1) indicateswhether the household accepted the initial offer, in which signing the contract was a requirement. The coefficienton B(t = 1) indicates how much more likely to pay are households who were initially screened, holding constanttheir current contractual conditions. Columns (1) to (3) show that there is no effect of contract requirement onthe quality of the selected customers. When I add avenue and team fixed effects, the coefficient in column (4) ispositive and significant. Columns (5) and (6) suggest that this effect is entirely driven by sales by Bantus: thecoefficient in the Bantu sample is almost identical, while the coefficient on the Tutsi sample is zero

39

Table IX: Experiment design: customers as principals

Coethnic trader non-Coethnic traderContract I II

No contract III IVSales on the Spot

Coethnic trader non-Coethnic traderContract V VI

No contract VII VIIISales on Debit

Notes: This table presents the factorial design when customers are principals. There are eight treatment cells,according to whether the trader was coethnic, whether the trader was instructed to offer a state-backed contract toguarantee the sale, and also whether the trade was implemented on the spot or on debit. In sales on the spot, thetrader provides the good on the spot immediately upon receiving payment. In sales on debit, the trader promisesto deliver the good in two days, in exchange of immediate payment by the household. In cells I to IV are thehouseholds in which trade is implemented on the spot. In Cell I, the trader is coethnic and offers a contract. In cellII, the trader is non-coethnic and offers a contract. In cell III, the trader is coethnic and does not offer a contract.In cell IV, the trader is non-coethnic and does not offer a contract. In cells V to VIII are the households in whichtrade is implemented on debit. In Cell V, the trader is coethnic and offers a contract. In cell VI, the trader isnon-coethnic and offers a contract. In cell VII, the trader is coethnic and does not offer a contract. In cell VIII,the trader is non-coethnic and does not offer a contract.

40

Table X: Testable Implications: customers as principals

Hypothesis Testable

implica-

tion

Households prefer to trade with traders of their ethnic group a1 > 0

Households prefer to trade with traders that have revealed to have access to

state-backed contracts (when there is no risk involved)

a2 < 0

Contract dis-taste is larger among coethnics a4 < 0

Households value more immediate delivery than future delivery a3 < 0

Households trust more traders of their ethnic group a5 < 0

Households trust more traders who use contracts to back their delivery promise a6 < 0

Contracts improve trust less among coethnics a7 < 0

Contracts reduce/increase trust among coethnics a6 + a7 < 0

Notes: This table presents the testable implications when customers are principals. The left column describes the

hypothesis. The right column describes the testable implication in the framework of the econometric specification.

Let Ei,j ∈ {0; 1} denote whether the interaction is coethnic, Fi,j ∈ {0; 1} whether formal contracts are used,

and Ci,j ∈ {0; 1} whether the sale is made on debit. In addition, let Tv be village fixed effects. The regression

specification is: Tradei,j = a0+a1Ei,j+a2Fi,j+a3Di,j+a4Ei,jFi,j+a5Ei,jDi,j+a6Fi,jDi,j+a7Ei,jFi,jDi,j+Tv+ei,j .

Sales are either on debit or on the spot. In sales on debit, the trader promises to deliver the good in two days, in

exchange of immediate payment by the household. In sales on the spot, the trader provides the good on the spot

immediately upon receiving payment.

41

Table XI: Effect of contracts and coethnicity on trade: customers as principals(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES Trade Trade Trade Trade Trade Trade Trade Trade

Sale on debit -0.161*** -0.443*** -0.442*** -0.441***(0.0313) (0.131) (0.135) (0.130)

Contract 0.0591 0.0953*** -0.00975 -0.0283 -0.0229(0.0430) (0.0320) (0.146) (0.151) (0.145)

Coethnic -0.0897 0.106** -0.0913 -0.0840 -0.119(0.0879) (0.0536) (0.118) (0.120) (0.117)

Contract X Coethnic 0.0775 0.0891 0.0937(0.154) (0.158) (0.153)

Sale on debit X Contract 0.284* 0.301* 0.283*(0.172) (0.178) (0.171)

Sale on debit X Coethnic 0.293** 0.283** 0.293**(0.136) (0.140) (0.136)

Sale on debit X Contract X Coethnic -0.281 -0.263 -0.281(0.182) (0.188) (0.181)

Constant 0.752*** 0.675*** 0.569*** 0.786*** 0.524*** 0.808*** 0.518** 0.971***(0.0243) (0.0304) (0.0226) (0.0826) (0.0495) (0.113) (0.214) (0.218)

Observations 1,308 450 858 450 858 1,308 1,179 1,308R-squared 0.069 0.084 0.131 0.082 0.126 0.083 0.095 0.098Trader FE NO NO NO NO NO NO NO YESHousehold controls NO NO NO NO NO NO YES YESSample ALL SPOT CREDIT SPOT CREDIT ALL ALL ALL

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the main results when customers are principals. I implement a linear probability model of the dummy Trade (indicating whetherthe trade was successful) on the following dummies: Sale on Debit, Contract, Coethnic. Sales are either on debit or on the spot. In sales on debit, the traderpromises to deliver the good in two days, in exchange of immediate payment by the household. In sales on the spot, the trader provides the good on thespot immediately upon receiving payment. Sale on Debit is a dummy takes value 1 if the sale was on debit and 0 if the delivery was implemented on thespot.Column (1) presents the average effect of Sale on Debit in the whole sample. Columns (2) and (3) present respectively the average effect of Contract,in the sample of sales on the spot and the sample of sale on debit. Columns (4) and (5) present respectively the average effect of Coethnic, in the sampleof sales on the spot and the sample of sale on debit. Column (6) presents the fully saturated model. Column (7) includes household level controls (age, agesquared, and number of children) and Column (8) adds trader fixed effects. All regressions include village fixed effects.

42

Figures

Figure 1: Experiment design: customers as agents

�����

����

�����

����

�����

����

�����

����

����

���

��

����

���

��

����

���

��

����

���

��

HHHH

HHHHH

����

���

���

��

HHHH

HHHHHH

HHHHHH

HHHHH

Household chooses Randomization Household chooses

v, −℘a

v − θ, ℘− ℘a

v − θ − L, ℘− ℘a

v − ℘, −℘a

0, 0

Sale on creditContract requested

Hou

seho

ldac

cept

s

Household

rejects

Contract req

uested

Contract withdrawn

Household Pays

Household Defects

Household Pays

Household Defectsu�������������

@@@@@@@@@@@@@

Notes: This graph illustrates the structure of sales when customers are agents. Traders present the offer to thecustomer. The trader informs the customer that he will make the good immediately available to the customer, ifthe customer promises to pays by cell phone in the near future (two days). However, all customers are informedthat in order for the sale to be possible, the customer needs to sign a state-backed contract in which he commitsto pays within two days. This is the sale credit. The customer then accepts or rejects. Once the decision has beenrecorded, a random sample of customers is selected in which the requirement to sign the contract is withdrawn. Inthese customers, the trader announces “I do not have a sufficient number of contracts today that would allow meto have you sign a contract. Since we have decided to make the deal, I must go ahead without having you sign acontract”. In the remaining customers, the requirement to sign the contract is maintained. The trader then leavesand the customer can pays or defects. The figure includes the payoffs that the trader and the customer wouldobtain, with linear and separable utility.

43

Figure 2: Customers’ best responses in the parameter space: customers as agents

��������������������

u11 > u10

u11 < u10

u11 < u0 u11 > u0

u10 > u0

u10 < u0

B = 0

B = 0

B = 1;P = 0

B = 1;P = 0

B = 1;P = 1

B = 1;P = 1

α2

α4

α3

α1

v

θ

Notes: This figure maps the the parameters to the best responses of the household. Thick lines delineate areaswhere the observed strategies are different. Dotted lines delineate areas where the preference ordering changes, butleads to no change in observed behavior whether contracts are required or not. αi ∈ 1, 2, 3, 4 are the mass of agentsin each of the cells. For instance, while α3 and α4 display the same behavior in the absence of contracts, α3 valuesthe good enough that he would be willing to accept a purchase requiring enforceable contract while α4 not.

44

Figure 3: Beliefs about contract enforceability: customers as agents

Notes: This figure presents the beliefs result, dis-aggregated by households’ and traders’ ethnicity. Traders areimplemented on credit. In sales on credit, traders provide the good, and ask the household to pay within two daysby cellphone. At the end of the transaction, the trader asks the household what sanctions he thinks he incurs if hedoes not pay as promised. In this figure, I present the proportion of households who believe that there will be legalsanctions if they do not pay. I separate matches by whether the household is of the majority ethnic groups andwhether the household is of the Tutsi minority. For each type of household, I include the proportion who believethat there will be legal consequences when the trader that visited the household is not Tutsi, and when the traderthat visited the household is Tutsi. I label each of the four interactions according to the ethnicity of the trader,followed the ethnicity of the household.

45

Figure 4: Experiment design: customers as principals

������������

������������

������������

������������

������������

������������

������������

����

��������

���������

���������

���������

���������

HHHHH

HHHHHHH

HHHHH

HHHHHHH

HHHH

HHHHH

Randomization Household chooses Trader chooses

0, 0

−℘, ℘− θ − FL

v − ℘, ℘

0, 0

v − ℘, ℘

Household

Sale

onDeb

it

Saleon

theSpot

Household acce

pts

Household refuses

Delivery

Defection

Household acce

pts

Household refuses

u�������������

@@@@@@@@@@@@@

Experiments I, II

Experiment II

Notes: This graph illustrates the structure when customers are principals. I randomly allocate customers to eitherSale on Credit or Sale on the Spot. In sale on credit, the trader makes the offer and requests payment immediately,in exchange for the promise of delivering the good to the customer in the near future (in a few days). In sale on thespot, the trader makes the offer and requests payment immediately, but makes the good available to the customerimmediately upon payment. Once the offer has been made, the customer can choose to accept it, in which casetrade occurs, or reject it. In sales on credit, if the customer rejects, the sale ends, and if the customer accepts, thenthe customer makes the payment to the trader. Later, the trader may deliver the good, or may defect. In sales onthe spot, if the customer rejects, the sale ends, and if the customer accepts, then the customer makes the paymentto the trader and the trader immediately provides the good. I have included the payoffs that the trader and thecustomer would obtain, assuming that the utility function is linear and separable.

46

Figure 5: Effect of contracts and coethnicity on trade: customers as principals

Notes: This figure presents the main result when customers are principals. Traders implement sales on credit. Insales on credit, the trader promises to deliver the good in two days, in exchange of immediate payment by thehousehold. The vertical axis indicates the share of attempted sales that were successful. There are four columns.The first two columns from the left indicate the share of successful sales among households that are non-coethnicsof the traders. Among these households, the first column reports the share for households in which traders didnot show a contract, and the second, the share for households in which the traders showed a contract. Columns 3and 4 have the same interpretation, albeit for sales in which traders and households are coethnics. Red intervalsindicate the 95% confidence interval.

47

A Mathematical Appendix

Proof of Proposition 1 To see this, note that the effect of debit on the buyer’s utility from purchase

is:

UB (E,F,D = 1)− UB (E,F,D = 0) = −v(1− β(E,F )η(E,F ))

The effect of contracts on the buyer’s utility from purchase when delivery is in the future is:

UB (E,F = 1, D = 1)−UB (E,F = 0, D = 1) = v(η(E,F = 1)−η(E,F = 0))+(℘−℘a)(λ(E,F =

1)− λ(E,F = 0)).

Finally, the effect of coethnicity on the buyer’s utility from purchase when delivery is in the future

is: UB (E = 1, F,D = 1)−UB (E = 1, F,D = 1) = v(η(E = 1, F )−η(E = 0, F ))+(℘−℘a)(λ(E =

1, F )− λ(E = 0, F ))

Proof of Proposition 2 Assuming the discount factor is independent of the ethnic proximity

of the trader, E, and on formalization, F , the difference in the two marginal effects of contracts

implies η(E = 1, F = 0) > η(E = 0, F = 0) If trust is higher among coethnics, i.e., if η(E =

1, F ) > η(E = 0, F ), then it must be that, whenever v > 0:

∆D (E,F ) = |∆UB∆D|E=0,F=0 −

∆UB∆D|E=1,F=0

= v (β(E = 1, F )η(E = 1, F )− β(E = 0, F )η(E = 0, F ))

> 0

Proof of Proposition 3

∆ (E) = ∆D (E,F = 1)−∆D (E,F = 0) = vβ(η(E,F = 0)− η(E,F = 1))

< 0

Proof of Proposition 4 ∆ (E = 0) > ∆ (E = 1) if η(E = 1, F = 0) > η(E = 0, F = 0) = 1 and

η(E = 1, F = 1) = η(E = 0, F = 1),

48

A Online Appendix

A.1 Additional tables and figures

Figure A.1: State contract

Notes: This figure shows the state contract when customers are principals.

49

Figure A.2: Instructions for payment by cellphone: customers as agents

Notes: This figure shows the payment instructions given to the customer, when customers are the agents.

A.2 Incentive compatibility of traders

While they also receive a fixed wage, traders are residual claimants on sales, which is a central

element of external validity.33 The revenues from sales are a non-negligible part of the income of

traders. In experiment I, if all trades are successful during the selling day, each trader generates

9 USD per day of sales. In experiment II, the equivalent number was 12 USD. The figures are

reduced by 25% less than the total revenues from sales due to project reimbursements, which I

describe below. Since traders are residual claimants, their incentives pose a threat to the quality

of the research. I next describe the strategies I use to minimize this threat.

If traders accept payments and do not deliver the goods, their profits increase. To avoid traders

reneging on their delivery promises, I design a cell phone monitoring system. Traders provide

customers with a project cell phone number and instructions for how to register a complaint. In

addition, I require traders to collect the customers’ cell phone numbers during the exit survey.

I inform the traders that the supervisor will contact a random sample of respondents to check

whether the sales were implemented as planned.34 I inform traders that their salaries would be

withdrawn if they fail to deliver the soaps, which we verify with the villages in which they operate.

Finally, traders collect the GPS coordinates of every customer in both the urban and rural areas,

33The fixed component was set after discussions with traders and surveyors, so that their profit, adjusted foruncertainty, was equal to the market wage they would otherwise obtain as surveyors, or as school teachers.

34I recorded no instance of fraud or cheating by traders, among all customers in which the Supervisor imple-mented the verification.

50

Figure A.3: Disaggregation of the main effect by ethnic sub-group: customers as principals

Notes: This figure presents the main result when customers are principals, dis-aggregated by customers’ and traders’ethnicity. Sales are implemented on credit. In sales on credit, the trader promises to deliver the good in two days,in exchange of immediate payment by the customer. The vertical axis indicates the share of attempted sales thatwere successful. There are two groups of columns. The first group columns on the left (dark columns) indicate theshare of successful sales among customers visited by a Bashi trader. Columns are grouped in two for each ethnicgroup of the customer: the first column reports the share for customers in which traders did not show a contract,and the second, the share for customers in which the traders showed a contract. The second group columns onthe left (light columns) indicate the share of successful sales among customers visited by a Bahavu trader and itsinterpretation is identical. Red intervals indicate the 95% confidence interval.

51

which decreases their incentives to shirk, especially for Experiment I in rural areas. 35 Despite

strict monitoring, the design of the experiment could affect the incentives of traders.

First, traders may be tempted to accept payments below the price set by the research project,

hence extracting strictly positive surplus from customers that would otherwise have refused the

purchase, which is a standard problem of industrial organization. To avoid this, I require traders

to pay a fixed amount that is lower than the sales price for each pack of 5 soaps they sell. The

supervisor verifies the stock of soaps and traders pay in proportion to the missing soaps. This

strategy reduces the set of prices below the recommended price at which the traders would make

positive profit. I recorded no sales at lower prices than recommended.

Second, traders may be tempted to sell above the price set by the project to extract additional

surplus. I allow traders to sell above the price set by the project if customers agree to pay

the higher price. To reduce the risk that traders would reallocate soaps to customers offering

higher prices, I give traders enough soaps for all households that they had to visit. Also, traders

could be tempted to violate the random allocation of households and select richer households to

extract higher surplus. However, discovering the wealth distribution in the village is difficult.36

Furthermore, I inform traders that researchers use statistical techniques such as randomization to

verify implementation violations.

A.3 Sampling of customers

Traders sample randomly selected households within each village in rural areas and within an urban

neighborhood in Bukavu. In the first day in the village (or in each urban neighborhood), traders

establish a village census with assistance from village (or neighborhood) authorities. Traders based

the random selection of households and their treatment on a list of randomly selected numbers

that were previously created using a statistical package.37 Experiment I includes 2,684 households

composed of the following ethnic groups: Balegas (n=1,208), Bashi-Bahavus (n=993), Batembos

(n=188), Tutsis (n=161), Other groups (n=134) and their main economic activities are agriculture

35More importantly, traders work in a long-term basis for various research projects for the authors and fear theirloss in reputation if caught cheating.

36See Sanchez de la Sierra (2014) for a description of how armed groups who are settled in the village struggleto know its wealth distribution.

37For each village size, I generated a sequence of random numbers lower than the total number of households.Traders then selected the households whose numbers in the census they drew coincides with the randomly selectednumbers.

52

and mining. All target customers are randomly selected males within the selected households.

53