Ethnicity and Credit in African Manufacturing

Journal of Development EconomicsŽ .Vol. 61 2000 205–235

www.elsevier.comrlocatereconbase

Short communication

Ethnicity and credit in African manufacturing

Marcel Fafchamps )

CSAE, UniÕersity of Oxford, Manor Road, Oxford OX1 3UL, UK

Received 1 February 1998; accepted 1 March 1999

Abstract

This paper investigates whether market interaction can, by itself, perpetuate the lack ofethnic diversity that is observed in the business communities of many developing countries.Using case study data on manufacturing firms in Kenya and Zimbabwe, we find noevidence that blacks or women are disadvantaged in the attribution of bank credit once wecontrol for firm size and other observable characteristics. In contrast, an ethnic and genderbias is noticeable in the attribution of supplier credit. Although we cannot rule out thepresence of discrimination, the bulk of the evidence indicates that network effects play animportant role in explaining this bias. q 2000 Elsevier Science B.V. All rights reserved.

JEL classification: O16; O17Keywords: Discrimination; Networks; Social capital; Supplier credit; Kenya; Zimbabwe

1. Introduction

In many developing countries, the ethnic makeup of local business communitiesis quite different from that of the population at large. It is not uncommon formembers of a particular ethnic or religious group, or even for residents of foreign

Žorigin, to account for an overwhelming proportion of entrepreneurs e.g., Bigsten.et al., 1998 . The historical record too is replete with examples of a particular

Žethnic group dominating commerce for extended periods of time e.g., Braudel,.1986; Greif, 1993, 1994 . Accounts by historians, anthropologists and sociologists

also demonstrate that the ethnic and religious background of dominant businessŽgroups varies considerably from place to place e.g., Bauer, 1954; Geertz, 1963;

Cohen, 1969; Meillassoux, 1971; Jones, 1972; Amselle, 1977; Geertz et al., 1979;

) Tel.: q44-1865-281446; e-mail: [email protected]

0304-3878r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0304-3878 99 00068-1

( )M. FafchampsrJournal of DeÕelopment Economics 61 2000 205–235206

. 1Staatz, 1979 . It even varies across economic activities within the same country.Lack of business diversity is a priori inefficient: drawing entrepreneurs from asmall talent pool reduces the average quality of local entrepreneurship. It is alsoinequitable as it leads to income disparities between groups that inevitably fuel

Ž .political tension e.g., Marris, 1971; Himbara, 1994 . Animosity toward prosper-ous ethnic or religious groups may, in turn, serve as investment disincentive ifmembers of the group fear subsequently expropriation. Such fears are not baseless,as the historical record demonstrates. 2

Colonial policies 3 and other historical and political factors have often favoredparticular ethnic groups and communities, enabling them to gain a dominantposition in a particular segment of economic activity. These factors, however, donot explain why ethnic concentration persists long after favorable factors andpolicies have been removed. The post-independence experience of many Africancountries, for instance, suggests that, once a group has established a dominantposition in an activity, it can retain its advantage long after initially favorableconditions have been eliminated — and even after having been actively combatedby post-colonial governments. 4 Similar questions have been raised in the U.S.regarding the survival of ethnic disparities between blacks and whites after the

Ž .removal of explicit discrimination e.g., Loury, 1998; Yinger, 1998 . Ethnicconcentration appears to be self-sustaining, at least under certain conditions. Theobjective of this paper is to investigate whether market interaction can, by itself,perpetuate a lack of ethnic diversity in business.

To achieve our purpose, we examine the role that group membership plays inparticipation to market exchange and in access to supplier and bank credit. Usingdata from manufacturing firms in two African countries, we provide preliminaryevidence regarding the respective roles of ethnicity and networks in the establish-ment of business trust. After a brief description of our conceptual framework, weexamine the relationship between ethnicity and credit in Kenyan and Zimbabweanmanufacturing. We uncover no evidence of ethnic or gender bias in the attributionof bank credit but our results show that blacks and women are disadvantaged inthe attribution of supplier credit. Although we cannot rule out the presence ofdiscrimination, the bulk of the evidence indicates that network effects play animportant role in explaining ethnic bias. These results suggest that more emphasis

1 The fish trade in Kenya, for instance, is dominated by the Luos, while textile manufacturing islargely in the hands of Kenyan–Asians.

2 For example, the expulsion of Asians from Uganda in the 1970s or, more recently, the looting ofethnic Chinese shops in Indonesia.

3 Including those of the white-dominated government of the Unilateral Declaration of Independenceera in Zimbabwe.

4 In the late 1960s, the Kenyata government actively favored the transfer of Asians businesses toŽ .ethnic Kenyans, particularly Kikuyus, with very little success e.g., Himbara, 1994 . Similar policies

have been pursued, although less forcefully, by the Mugabe administration in Zimbabwe.

( )M. FafchampsrJournal of DeÕelopment Economics 61 2000 205–235 207

should be put on policies that foster personal connections and information sharingbetween social and ethnic groups.

2. Conceptual framework

The interplay between trust, trade, and ethnicity or religion has long beenrecognized. It has, for instance, been argued that Islam penetrated the East African

Žinterior due to coastal merchants’ preference to deal with Muslims e.g., Shilling-.ton, 1989; Ensminger, 1992 . Sociologists have similarly emphasized that African

Žentrepreneurs prefer to do business with members of their own ethnic group e.g.,.Marris, 1971; Macharia, 1988; Himbara, 1994 . Together with a number of

economists, they have emphasized the role that trust and reputation amongindividuals and communities play in creating an enabling environment for tradeŽe.g., Mitchell, 1969; Granovetter, 1985; Coleman, 1988; Hart, 1988; Milgrom et

.al., 1991; Greif, 1993, 1994; Platteau, 1994 . There is a growing consensus thatsharing the same ethnicity and religion are elements that favor the establishment of

Ž .trust e.g., Gambetta, 1988; Fukuyama, 1995; Cornell and Welch, 1996 .Conceptually, there are several ways by which ethnicity may influence the

Žallocation of credit, e.g., through taste for discrimination e.g., Becker, 1971;. Ž .Akerlof, 1985 , erroneous expectations or ‘prejudice’ e.g., Yinger, 1998 , diffi-

Žculties of communication across cultural boundaries e.g., Cornell and Welch,. Ž1996; Loury, 1998 , statistical discrimination e.g., Arrow, 1972; Coate and Loury,

. Ž1993 , and network effects e.g., Saloner, 1985; Montgomery, 1991; Taylor,

.1997 . There is widespread disagreement as to the relative empirical contributionsthese mechanisms make to ethnic and gender bias in labor and credit markets. 5

Ž .Becker 1971 , for instance, has argued that prejudice and taste for discriminationare costly and should result in lower profits. In a competitive environment, heargues, firms that discriminate on the basis of taste or maintain erroneousexpectations should, in the long run, be competed out by more open-minded,better informed businesses. Becker’s view has not gone unchallenged, however. 6

Unlike prejudice and tastes, statistical discrimination is perfectly compatiblewith the profit-seeking motive and cannot, therefore, be competed out. Wheneverfirms cannot assess clients and suppliers directly, it is rational for them to screen

5 Ž . Ž .See Darity 1998 , Donohue 1998 , and the Spring 1998 issue of the Journal of EconomicPerspectives for recent surveys of the literature in the U.S.

6 Ž .As pointed out by many e.g., Darity and Mason, 1998; Donohue, 1998 , the idea that marketforces should eliminate discrimination and prejudice relies critically on the assumption that markets arecompetitive. In the U.S. South during the Jim Crow era, Donohue emphasizes that employers whohired blacks were ostracized by the white-dominated establishment and feared being targeted by the KuKlux Klan. In such circumstances, the author argues, market forces could not operate and, withoutexternal intervention, discrimination could have perdured indefinitely. Becker’s view has been used tooppose affirmative action in the U.S.


on the basis of whatever observable information they can collect. If groups ofdifferent race or gender differ in unobservable attributes, statistical discriminationwill arise. The role that it plays in explaining actual ethnic bias has, however, beenthe object of much debate. In addition, the presence of statistical discrimination isextremely difficult to prove since it requires the econometrician to have as much if

Žnot more information about applicants than employers themselves e.g., Darity,. 71998 .

Network effects have received somewhat less attention in the discriminationŽliterature, but they have long been studied in labor markets e.g., Granovetter,

.1995 . The basic idea is that information about opportunities for exchange andagents’ types circulates along interpersonal networks. People talk with their friendsand professional acquaintances about jobs, bad payers, and arbitrage opportunities,and they refer job and credit applicants to each other. In such environment,individuals with better networks collect more accurate information, which enablesthem to seek out market opportunities more aggressively and to better screenprospective employees and credit recipients. A rapidly growing literature hasmodeled these processes and has shown that, in a world of imperfect information,

Žthey provide an economic advantage to better connected agents e.g., Kranton,.1996; Taylor, 1997; Fafchamps, 1998 .

To the extent that members of a particular group cultivate close links with eachother, be it for historical or cultural reasons, 8 the group will be seen to performbetter than others in market exchange. If this group recruits its members primarilyalong ethnic or gender lines, ethnic or gender bias will occur although, strictlyspeaking, agents need not have a taste for discrimination and they need not rely onstatistical discrimination. 9 Network effects thus put the emphasis on patterns ofsocialization as an alternative explanation for ethnic or gender bias. 10 Theprimary objective of this paper is to assess how much of the observed ethnic andgender bias in African enterprise credit can be attributed to network effects.

7 As much information to interpret a significant sign of race in a wage regression as evidence ofdiscrimination; more information to demonstrate that unobservable attributes vary systematically withrace or gender.

8 For example, persecution.9 In statistical discrimination models, ethnic bias arises when two populations have different hidden

characteristics. In network effects, ethnic bias may arise even when they have the same hiddencharacteristics, provided members of one group can more easily screen members of their own group,and one group has acquired a dominant position, perhaps for historical reasons.

10 The concept of network effects bears some resemblance with another explanation for ethnic biasbased on the existence of a dominant group seeking to protect its supremacy. The difference is thatnetwork effects can arise in a completely decentralized manner, that is, even in the absence of any

Žcollusion among members of the dominant group and without need for metapunishment see, for.instance, the discussion in Donohue, 1998 . Of course, the presence of discrimination or of collusion to

exclude members of other groups can coexist with network effects, and the existence of ethnic-basednetworks can facilitate collusion. The point is that the coexistence of ethnic bias and interpersonalnetworks does not, by itself, imply collusion.


Our analysis focuses not only on bank credit but especially on supplier creditŽwhich has been shown to play an important part in firm finance e.g., Fafchamps,

.1997 . At the core of our argument lies the recognition that normal commercialtransactions require a temporal dissociation between delivery and payment, other-wise the conduct of business would be too unwieldy. Market transactions thusnormally encompass an element of credit. 11 Because recourse to formal collateralis impractical for most business transactions, trade credit gets allocated essentially

Ž .on the basis of trust e.g., Fafchamps, 1996, 1997; Fafchamps et al., 1994 . Tradecredit is thus typically offered on a selective basis; those who do not qualify must

Ž . 12buy cash e.g., Fafchamps, 1996, 1997; Fafchamps et al., 1994, 1995 . Sinceonly very small firms can operate on a cash-only basis, how trust is establishedand with whom dictates not only how trade takes place but also which firms areable to operate in a business-like fashion and which must remain microenterprises.In addition, credit from suppliers is an important source of finance for small and

Žmedium-size firms e.g., Cuevas et al., 1993; Bade and Chifamba, 1994; Fafchamps. 13et al., 1994, 1995 . An examination of the process by which trade and bank

credit are attributed is therefore expected to throw light on the ethnic compositionof business communities.

The difficulty is to disentangle discrimination from network effects, since theycan both generate ethnic bias in the attribution of credit and, to the extent thatnetworks are based on ethnicity, both grant one particular ethnic group aneconomic advantage. Their testable implications differ in two important dimen-sions, however. First, if networks do not matter, possible access to businessinformation through socialization and screening methods should not affect howfirms get and give credit. In contrast, if network effects are present, better accessto information should help predict which firms get and give credit. Second, if biasis the result of statistical discrimination, firms who are discriminated against willthemselves discriminate among their clients. To the extent that all firms face thesame pool of potential clients, the ethnicity of a firm’s owner or manager shouldthen help determine whether the firm gets credit but not whether it gives credit to

11 The duration of trade credit is normally defined as the time elapsed between invoicing andŽpayment. In developed countries, this delay typically ranges between 30 and 60 days e.g., Schwartz,

. Ž1974; Schwartz and Whitcomb, 1979 . Similar delays were found among African manufacturers e.g.,.Cuevas et al., 1993; Bade and Chifamba, 1994; Fafchamps et al., 1994, 1995 .

12 In this paper, we use the phrase ‘cash payment’ to designate payment in currency or certified checkat the time of delivery. Accepting a payment by check is about as risky as granting credit and isconsidered as such here. In the business world, payment for materials and inputs out of petty cash isextremely rare, except for very small infrequent purchases. One of the reasons is that large movementsof cash are unsafe. Among businesses, the word ‘cash’ is often given a meaning different from the oneused here, i.e., to refer to early payment.

13 Although trade credit is formally considered short term financing, it is normally renewed with eachorder so that, in practice, it provides firms with a long term source of working capital.


its clients. 14 In contrast, if ethnicity is correlated with network affiliation, and ifbias is the result of network effects, then insider firms should find it easier notonly to be recognized by suppliers, but also to identify reliable clients. In this case,ethnicity should explain not only whether firms get credit, but also whether theygrant credit to clients. 15 The same holds for gender bias. These testable implica-tions form the basis of our empirical analysis.

3. The data

We use data from two surveys of enterprise finance in African manufacturingconducted under the auspices of the World Bank and coordinated by the RegionalProgram for Enterprise Development of the African Technical Division. The firstsurvey was undertaken in September 1993 in Nairobi, Kenya; the second took

Žplace in August 1994 in Harare, Zimbabwe see Fafchamps et al., 1994, 1995 for.details . Close to 60 firms were interviewed in both countries, two-thirds of which

were randomly selected from two panels of 200 firms previously surveyed byWorld Bank teams. The remaining firms were selected among wholesalers andretailers operating in the same sectors. Both panels are made of manufacturersoperating in textile and garment, food processing, wood products, and metalproducts industries. As a result of sampling design, they overrepresent small and

Žmedium size enterprises and underrepresent microenterprises Bade and Gunning,. 161994 . Since the former display the most variation in trade credit use, the data

are particularly suited for our purpose.The sectoral and ethnic composition of the two samples is given in Table 1. In

both countries owners and managers of medium to large firm predominantlybelong to a minority ethnic group. The ethnicity of the dominant business group,however, varies between the two countries. In Kenya, the bulk of surveyed firmsare in the hands of ethnic Asians; in Zimbabwe, they are mostly managed by

14 In contrast, if discrimination is due to ‘taste’, firms should favor members of their own ethnicityŽ .and one should observe the reverse effect e.g., Darity and Mason, 1998 . By comparing the provision

and attribution of credit, this test is less subject to the criticism that affects standard ‘tests’ ofŽdiscrimination which regress, say, wages or credit received on observable characteristics and race e.g.,

.Darity, 1998; Heckman, 1998 .15 The validity of the test rests on the assumption that credit reliability is not itself a determinant of

the willingness to grant trade credit to clients. If this was the case, being discriminated against wouldbe correlated with customer credit through unreliability. Given that all clients ask for credit, it is hardto believe that firms would fail to realize that granting credit to their clients helps their sales.

16 Ž .Daniels 1994 reports that, in Zimbabwe, enterprises of less than five employees represent around95% of all enterprises of 50 employees or less; the corresponding ratio is 44% in our sample.


Table 1Composition of the sample

Percentage of sample firms No. of All Kenya Zimbabweobservations

Percentage of sample inthe following categories

In manufacturing 75 65 62 68In trade 40 35 38 32In food processing 26 23 17 29In textile and garment 43 38 38 38In wood products 22 19 22 16In metal products 23 20 22 18With an ethnic African 37 32 42 23ownerrmanagerWith an ethnic Asian 29 25 46 5ownerrmanagerWith an ethnic European 38 33 5 61ownerrmanagerWith an ownerrmanager 11 9 7 11from other ethnicityIn Kenya 58 50 100 0In Zimbabwe 57 50 0 100Total 115

whites. 17 In both cases, these communities represent around 1% of the country’spopulation. Such degree of ethnic concentration in medium to large scale enter-prises is not unusual in Africa, although dominant business communities are not

Ž .always of foreign origin e.g., Bigsten et al., 1998 . In contrast, the overwhelmingmajority of microenterprises in Kenya and Zimbabwe are in the hands of ethnic

Ž .Africans e.g., Daniels, 1994 .Data were collected on a variety of issues pertaining to enterprise finance. A

large number of questions were devoted to trade credit. Respondent were quizzed,for instance, on how they screen customers before granting trade credit, how theyrespond to late payment by clients, and how they seek enforcement of contractualobligations. Their responses were later coded into consistent categories. The datasuffer from a number of shortcomings: the coverage and wording of individualquestions vary somewhat between the two surveys; many observations are miss-ing; and certain important pieces of information were not collected. The imperfectcoverage of the data is partially compensated by the detailed qualitative informa-tion gathered during interviews, but more careful data collection is needed to

17 Why is it Asians who dominate in Kenya and whites in Zimbabwe is undoubtedly the result of pastpolicies that favored these two groups. These policies, however, have been discontinued since

Žindependence 1964 for Kenya, 1979 for Zimbabwe after 15 years under Iam Smith’s white govern-.ment .


Table 2Descriptive statistics

Ž . Ž .Group 1 sownerrmanager is an ethnic African. Group 2 sownerrmanager is neither ethnicŽ .African nor member of the ethnic group that dominates business. Group 3 sownerrmanager belongs

Ž .to the ethnic group that dominates business in the country Asians in Kenya, whites in Zimbabwe .

No. of valid All Black Other Domin.Ž . Ž . Ž .observations firms 1 2 3

Firm characteristicsMedian number of employees 111 48 4 65 73Median year that the firm was created 113 1975 1975 1970 1973Percentage of firms that are formally 113 65% 41% 88% 73%registeredrincorporatedPercentage of firms that are a subsidiary 113 14% 14% 6% 17%Percentage of firms headed by a woman 114 12% 30% 13% 2%Percentage of firms reporting severe cash-flow 103 35% 45% 38% 29%problems in recent past

Bank financePercentage of firms with overdraft 113 69% 40% 69% 85%Percentage of firms who ever got a loan 92 43% 27% 54% 50%from a financial institution

Credit from suppliersMedian share of credit

Ž .purchases in total purchases % 111 75 17 85 90Ž .Median duration of supplier credit days 110 30 10 30 30

Ž .Median cash discount on supplier credit % 57 3.75 2.5 3.75 3.25Ž .Median implicit monthly interest rate % 53 2.5 5 3.75 2.5

Percentage of firms 113 85% 62% 94% 97%that ever purchase on creditPercentage of firms who ever get 92 68% 35% 60% 84%offered credit from first purchase

Credit to clientsMedian share of credit sales 81 50 0 33 75

Ž .in total sales %Percentage of firms 114 79% 57% 88% 90%that ever sell on creditPercentage of firms who ever offer 54 63% 42% 50% 74%

acredit from first purchase

SocializationPercentage of firms that do not 109 46% 46% 7% 10%

bsocialize with suppliers at allPercentage of firms that socialize 109 36% 11% 64% 44%

bwith suppliers outside businessPercentage of firms that do not socialize 56 29% 38% 22% 26%

abwith clients at allPercentage of firms that socialize with 56 21% 8% 44% 21%

abclients outside business


Ž .Table 2 continued

No. of valid All Black Other Domin.Ž . Ž . Ž .observations firms 1 2 3

SocializationPercentage of firms that do not 102 41% 72% 20% 32%

bsocialize with bank staff at allPercentage of firms that socialize 102 15% 3% 33% 16%

bwith bank staff outside business

( )Screening of customers multiple answeres allowedPercentage of firms 98 51% 30% 71% 56%that ask clients to fill formsPercentage of firms that rely 98 56% 48% 64% 58%on trial period to screen clientsPercentage of firms that rely on 98 73% 59% 93% 75%reputation to screen clientsPercentage of firms that rely on direct 98 37% 59% 50% 23%investigation to screen clients

a Data were collected in Zimbabwe only.bOther firms socialize during business hours.

confirm our results. The findings presented here should thus be considered aspreliminary and illustrative only. In spite of these shortcomings, they provide auseful initial assessment, given the near total absence of data in this area.

Descriptive statistics are given in Table 2 for all firms and by ethnic group.Only 12% of sample firms are headed by women. There appears to be a strongcorrelation between ethnicity and firm size. Similar — if not stronger —

Žcorrelations have been found in other studies e.g., Bade and Gunning, 1994;.Daniels, 1994; Himbara, 1994 . Ethnicity is also correlated with the amount of

trade credit the firm receives and gives and, to a lesser extent, with availability ofbank finance. Black firms are less likely to socialize with their clients andsuppliers than other firms, an indication that black entrepreneurs largely remainoutside the main business network. Probably for this reason, mainstream firms aremore likely than black firms to screen customers using some sort of informationsharing, and less likely to investigate clients directly. Black firms are also lesslikely to be ‘formal’: fewer of them are registered, and they are less likely torequest clients to fill forms.

4. Multivariate analysis

Table 2 indicates that ethnicity and credit are related, but a number of otherfactors — firm size, sector of activity, susceptibility to liquidity problems — mayaccount for this relationship. To test for ethnic bias, we must rely on a multivariateanalysis. To this effect, we begin by regressing the surveyed firms’ proportion ofpurchases and sales made on credit on a series of firm characteristics thought to


influence the supply and demand of trade credit, plus dummy variables measuringethnicity and network effects. We also use limited information available onwhether firms ever purchase from or sell on credit to first-time commercialpartners, and on credit terms and implicit interest in credit purchases. If statisticaldiscrimination and network effects are absent, the latter variable should benon-significant. 18 The accuracy of the test for discrimination rests critically on the

Ž .absence of omitted variable bias e.g., Heckman, 1998; Yinger, 1998 . If suppliersobserve certain characteristics of prospective clients that are not recorded in thedata but are correlated with ethnicity, we may erroneously conclude that ethnicitymatters. 19 To minimize this bias we include as regressors most of firms’ easilyobservable characteristics, such as sector of activity, registered status, and sub-sidiary status, plus characteristics for which the econometrician probably has betterdata than creditors, namely, the actual number of employees of the firm and itscash-flow history. Still, the possibility of omitted variable bias should be kept inmind before a significant coefficient on ethnicity or gender is taken as evidence of

Ž .discrimination, as the work of Neal and Johnson 1996 reminds us.The estimated regression for credit purchases is:

P sg qg X qg D qg N S qe 1Ž .i 0 1 i 2 i 3 i i

The dependent variable P is the proportion of purchases made on credit; it cani

take any value between and including 0 and 1. The vector X contains firmi

characteristics such as country, sector, legal status, and size. Vector D stands fori

ethnicity and gender, while vector N S captures available information on networkiŽ .effects with suppliers. The error term e is assumed to be i.i.d. Eq. 1 is estimatedi

as a two-sided censored Tobit on the pooled data and on each country sampleŽseparately. Firm size is measured as the log of the number of employees plus

.one . Given that firms with better access to trade credit may grow faster, wecontrol for simultaneity bias by replacing firm size W by its predicted values fromi

the following equation:

W sl ql X ql D ql N S ql A ql A2 qe 2Ž .i 0 1 i 2 i 3 i 4 i 5 i i

Ž .Eq. 2 is run on each country sample separately. The age and age squared of thefirm — A and A2 — serve as identifying restrictions; 20 they are jointlyi i

18 Ž .As Heckman 1998 pointed out, such a test can only show whether discrimination has no effect onŽ .market outcomes. Discrimination by certain but not all individuals is quite compatible with the

absence of market discrimination if individuals who are discriminated against can always choose todeal with non-discriminating firms.

19 It would have been useful to obtain the information about surveyed firms that is publicized by thecredit reference bureau operating in Zimbabwe. Similarly, it would useful to get access to theinformation banks have on surveyed firms. Unfortunately, this information was not collected.

20 Although the age of the firm may have a small effect on access to credit for very young firm, theŽeffect is likely to be much smaller than the effect of firm age on size e.g., Hoogeveen and Tekere,

.1994; Risseeuw, 1994 . When regressing access to trade credit on uninstrumented firm size and firmŽ .age not shown , age is not significant.


significant and explain a large portion of the variation in firm size. EstimationŽ .results for Eq. 2 are given in Table 3.

Given that Zimbabwe enjoys the presence of a formal credit reference bureauwhile Kenya does not, we expect the country dummy to be positive, reflecting therelative ease with which Zimbabwean firms screen trade credit applicants. Foodproducts are typically perishable and turnover is fast, so that less trade credit may

Žbe offered in that sector see Nadiri, 1969; Schwartz, 1974; Schwartz and.Whitcomb, 1979; Ferris, 1981; Emery, 1984 . Larger firms are expected to

purchase a larger share of their inputs on credit since it would be inconvenient forthem to operate on a cash basis. Firms that are registered and incorporated, andsubsidiaries of large holding corporations are more likely to elicit suppliers’confidence than informal partnerships. A dummy variable is also included thattakes the value of one if the respondent experienced severe cash-flow problems inthe past. Presumably, incompetent firms are more likely to run into problems, notonly because they are incompetent but also because they receive less credit, which

Ž .makes them more vulnerable to shocks e.g., Fafchamps et al., 1994, 1995 .Having faced serious cash-flow problems in the past is thus a signal of incompe-tence; if information on such occurrences circulates, one should observe a negativerelationship between past problems and current credit.

The regression of P on firm characteristics is shown in the first two columnsi

of Table 4a–c for the pooled sample, the Kenya sample, and the Zimbabwesample, respectively. Firms in the food and wood sectors receive significantly lesscredit than others. Large firms, registered firms, and subsidiaries receive moresupplier credit, firms with cash-flow problems less. We then add ethnicity and

Table 3Prediction equation for firm size

Ž .Dependent variables log number of employeesq1 .

Kenya Zimbabwe

Coefficient t-ratio Coefficient t-ratio

Intercept y4381.7 y2.27 y1468.5 y1.02Manufacturing dummy 0.96 2.95 0.48 0.96Food sector dummy 0.06 0.12 0.14 0.27Wood sector dummy y0.40 y0.91 0.09 0.14Metal sector dummy y0.20 y0.47 y1.50 y2.55Year firm was created 4.50 2.28 1.53 1.03Year squared y11.55 y2.30 y3.98 y1.05Registered business dummy 0.95 2.69 1.02 1.92OrM is a woman y0.96 y1.73 y1.55 y1.99OrM is black y1.07 y1.73 y0.63 y0.96OrM belongs to dominant business group y0.56 y1.06 0.07 0.12Number of observations 55 55

2R 0.64 0.47

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

216

Table 4Tobit regressions on the share of credit purchasesDependent variable is the share of credit purchases in total purchases from suppliers.The reported estimates are two-limit censored Tobit. Standard errors are not corrected for the presence of predicted variables in the regression.

Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

( )a Pooled dataIntercept 1.15 1.20 0.76 4.16 0.17 1.36 0.67 3.42Zimbabwe dummy 0.13 1.11 0.36 2.99 0.04 0.41 0.25 2.04Manufacturing dummy y0.08 y0.81 y0.05 y0.53 0.00 0.01 y0.01 y0.06Food sector dummy y0.56 y4.69 y0.52 y4.78 y0.48 y4.13 y0.51 y4.57Wood sector dummy y0.36 y3.07 y0.41 y3.70 y0.27 y2.26 y0.37 y3.11Metal sector dummy y0.11 y0.93 y0.33 y2.66 y0.08 y0.70 y0.29 y2.25

Ž .Log no. workers , instrumented 0.16 3.32 y0.03 0.44 0.14 3.02 0.01 0.09Registered business dummy 0.20 1.68 0.39 3.01 0.14 1.28 0.30 2.35Subsidiary dummy 0.39 2.36 0.41 2.70 0.44 2.80 0.44 2.94Past cash-flow problems dummy y0.23 y2.52 y0.15 y1.81 y0.21 y2.50 y0.16 y2.03OrM is a woman y0.36 y2.26 y0.30 y1.92OrM is black y0.35 y3.43 y0.29 y2.63OrM does not socialize with suppliers y0.18 y1.60 y0.08 y0.75OrM socializes with suppliers outside business 0.16 1.85 0.07 0.84Selection term 0.36 0.32 0.33 0.31Number of observations 97 97 93 93Log-likelihood y51.74 y43.77 y43.66 y38.84

2Pseudo R 0.41 0.50 0.48 0.54

( )b Kenya onlyIntercept 0.17 1.06 0.85 3.65 0.17 1.22 0.59 2.45Manufacturing dummy y0.20 y1.15 y0.12 y0.76 0.10 0.63 0.07 0.46Food sector dummy y0.95 y4.59 y1.00 y5.42 y0.80 y4.67 y0.93 y5.56Wood sector dummy y0.51 y2.72 y0.63 y3.73 y0.43 y2.69 y0.57 y3.58Metal sector dummy y0.36 y2.32 y0.63 y3.87 y0.25 y1.98 y0.50 y3.24

Ž .Log no. workers , instrumented 0.22 2.67 y0.01 y0.06 0.17 2.51 0.07 0.84

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

217

Registered business dummy 0.23 1.38 0.50 2.96 0.06 0.43 0.28 1.79Past cash-flow problems dummy y0.15 y0.98 y0.07 y0.52 y0.25 y2.08 y0.19 y1.64OrM is a woman y0.60 y2.91 y0.47 y2.69OrM is black y0.27 y1.71 y0.08 y0.51

UOrM does not socialize with suppliers 0.23 y1.82 y0.19 y1.58UOrM socializes with suppliers outside business 0.33 2.56 0.21 1.61

Selection term 0.38 0.32 0.29 0.26Number of observations 50 50 47 47Log-likelihood y28.10 y21.65 y16.83 y13.10

2Pseudo R 0.40 0.54 0.62 0.71

( )c Zimbabwe onlyIntercept y0.07 y0.27 0.47 0.97 y0.08 y0.29 0.67 1.39Manufacturing dummy 0.02 0.19 0.01 0.12 y0.03 y0.19 y0.05 y0.38Food sector dummy y0.19 y1.53 y0.15 y1.35 y0.17 y1.31 y0.15 y1.26Wood sector dummy y0.17 y1.28 y0.17 y1.40 y0.15 y1.08 y0.19 y1.45Metal sector dummy 0.32 1.84 0.12 0.52 0.33 1.89 0.03 0.13

Ž .Log no. workers , instrumented 0.20 3.45 0.09 0.81 0.21 3.19 0.06 0.53Registered business dummy y0.05 y0.34 0.05 0.26 y0.08 y0.54 0.07 0.35Subsidiary dummy 0.34 2.40 0.35 2.78 0.29 2.01 0.30 2.34Past cash-flow problems dummy y0.15 y1.52 y0.12 y1.32 y0.14 y1.27 y0.11 y1.11OrM is a woman y0.06 y0.23 y0.13 y0.53OrM is black y0.29 y2.32 y0.38 y2.92OrM does not socialize with suppliers 0.21 1.10 0.34 1.88OrM socializes with suppliers outside business 0.03 0.25 y0.01 y0.14Selection term 0.28 0.26 0.28 0.24Number of observations 47 47 46 46Log-likelihood y14.51 y11.89 y13.55 y9.44Pseudo R2 0.51 0.60 0.53 0.67

U Ž .Coefficients jointly significant: F-test 2,36 that both socialization variables have null coefficient in last regression is 3.04 with p-value of 0.06.

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

218

Table 5Probit regressions on supplier credit at first purchaseDependent variable is one if supplier credit is usually offered from first purchase, zero otherwise.Standard errors are not corrected for the presence of predicted variables in the regression.


( )a Pooled dataIntercept y0.48 y0.95 1.02 1.32 y0.96 y1.71 0.58 0.66Zimbabwe dummy y0.77 y1.89 y0.40 y0.90 y0.79 y1.82 y0.47 y1.02Manufacturing dummy y0.39 y1.07 y0.39 y1.03 y0.43 y1.06 y0.50 y1.20Food sector dummy 0.39 0.79 0.26 0.49 0.64 1.17 0.39 0.69Wood sector dummy y0.75 y1.73 y1.08 y2.30 y0.46 y0.95 y0.96 y1.77Metal sector dummy y0.06 y0.15 y0.57 y1.19 0.11 0.25 y0.49 y0.92

Ž .Log no. workers , instrumented 0.45 2.98 0.16 0.84 0.45 2.75 0.18 0.92Past cash-flow problems dummy y0.22 y0.62 y0.14 y0.39 y0.07 y0.18 y0.01 y0.02OrM is a woman y0.98 y1.49 y0.99 y1.47OrM is black y1.12 y2.47 y0.94 y1.79OrM does not socialize with suppliers 0.19 0.36 0.53 0.91OrM socializes with suppliers outside business 0.92 2.38 0.67 1.62Number of observations 82 82 79 79Log-likelihood y43.34 y38.60 y38.36 y35.38

2Pseudo R 0.17 0.26 0.23 0.29

( )b Kenya onlyIntercept y0.28 y0.41 1.43 1.28 y0.61 y0.79 1.29 0.91Manufacturing dummy y1.35 y1.60 y1.04 y1.11 y1.02 y0.99 y0.97 y0.90Food sector dummy 0.48 0.62 0.00 0.00 0.27 0.31 y0.29 y0.30Wood sector dummy y0.06 y0.11 y0.50 y0.79 0.14 0.24 y0.48 y0.64

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

219

Metal sector dummy 0.58 1.42 0.05 0.11 0.72 1.58 0.21 0.37Ž .Log no. workers , instrumented y0.43 y0.56 y0.11 y0.13 y0.89 y1.03 y0.31 y0.30

Past cash-flow problems dummy 0.23 0.39 0.36 0.58 y0.08 y0.12 0.11 0.15OrM is a woman y0.39 y0.38 y0.35 y0.34OrM is black y1.39 y1.77 y1.35 y1.46OrM does not socialize with suppliers y0.36 y0.44 y0.44 y0.52OrM socializes with suppliers outside business 0.54 0.93 y0.12 y0.16Number of observations 38 38 36 36Log-likelihood y22.59 y20.23 y19.87 y18.38

2Pseudo R 0.07 0.17 0.13 0.20

( )c Zimbabwe onlyIntercept y4.03 y2.54 y1.26 y0.58 y5.38 y2.43 y3.10 y1.11Manufacturing dummy 0.47 0.82 0.46 0.77 0.13 0.19 0.04 0.05Food sector dummy 1.06 1.36 1.32 1.45 1.66 1.68 1.49 1.58Wood sector dummy 0.06 0.09 y0.22 y0.33 0.71 0.82 0.34 0.38Metal sector dummy 0.99 1.34 0.35 0.40 1.21 1.40 0.80 0.74

Ž .Log no. workers , instrumented 0.81 3.02 0.34 0.92 0.90 2.57 0.54 1.23Past cash-flow problems dummy y0.28 y0.55 y0.31 y0.57 0.18 0.29 0.05 0.08

a aOrM is a womanOrM is black y1.20 y1.51 y0.62 y0.65OrM does not socialize with suppliers 0.93 0.93 1.21 1.03OrM socializes with suppliers outside business 1.79 2.29 1.63 1.89Number of observations 44 40 43 39Log-likelihood y18.30 y16.27 y14.73 y13.41

2Pseudo R 0.34 0.28 0.46 0.40

aGender predicts outcome perfectly; consequently, gender variable and four observations were dropped from the regression.

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

220

Table 6Tobit regressions on payment terms made by suppliersDependent variable is the log of the number of daysq1 that separate delivery and payment; it is equal to 0 for firms that do not receive supplier credit.Standard errors are not corrected for the presence of predicted variables in the regression.


( )a Pooled dataIntercept 0.57 1.29 1.46 2.20 0.83 1.79 1.53 2.11Zimbabwe dummy 0.09 0.24 0.38 0.86 y0.12 y0.32 0.13 0.28Manufacturing dummy y0.12 y0.36 y0.04 y0.11 0.12 0.35 0.13 0.39Food sector dummy y1.04 y2.68 y0.99 y2.57 y1.02 y2.61 y1.05 y2.70Wood sector dummy 0.09 0.24 0.00 0.00 0.03 0.08 y0.12 y0.29Metal sector dummy 0.59 1.45 0.25 0.54 0.58 1.51 0.30 0.64

Ž .Log no. workers , instrumented 0.61 3.54 0.36 1.51 0.61 3.48 0.43 1.82Registered business dummy 0.26 0.63 0.44 0.94 0.13 0.31 0.27 0.57Subsidiary dummy y0.29 y0.66 y0.11 y0.23 y0.13 y0.30 y0.03 y0.06Past cash-flow problems dummy y0.52 y1.76 y0.38 y1.26 y0.58 y2.01 y0.49 y1.67OrM is a woman y0.38 y0.62 y0.32 y0.52OrM is black y0.69 y1.81 y0.53 y1.28OrM does not socialize with suppliers y0.78 y1.99 y0.62 y1.52OrM socializes with suppliers outside business 0.02 0.06 y0.12 y0.37Selection term 1.30 1.28 1.24 1.23Number of observations 97 97 93 93Log-likelihood y154.2 y152.4 y143.7 y142.8

2Pseudo R 0.13 0.14 0.15 0.15

( )b Kenya onlyIntercept 0.30 0.39 1.10 1.01 1.14 1.45 1.53 1.16Manufacturing dummy y0.58 y0.70 y0.46 y0.55 y0.02 y0.03 y0.13 y0.15Food sector dummy y1.92 y2.36 y1.90 y2.30 y1.70 y2.09 y1.83 y2.13Wood sector dummy 0.26 0.35 0.13 0.18 0.31 0.41 0.19 0.24Metal sector dummy 0.30 0.44 0.05 0.07 0.27 0.41 0.05 0.06

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

221

Ž .Log no. workers , instrumented 0.72 1.87 0.47 1.04 0.57 1.48 0.49 1.10Registered business dummy 0.77 1.07 0.96 1.24 0.73 1.00 0.90 1.12Past cash-flow problems dummy y0.39 y0.64 y0.22 y0.34 y0.57 y0.99 y0.48 y0.75OrM is a woman y0.44 y0.42 y0.50 y0.47OrM is black y0.61 y0.83 y0.14 y0.16OrM does not socialize with suppliers y1.68 y2.63 y1.64 y2.38OrM socializes with suppliers outside business y0.42 y0.65 y0.59 y0.77Selection term 1.68 1.66 1.54 1.54Number of observations 49 49 46 46Log-likelihood y81.29 y80.80 y72.21 y72.08

2Pseudo R 0.12 0.12 0.17 0.17

( )c Zimbabwe onlyIntercept 0.89 1.35 2.26 1.84 0.74 1.16 2.42 2.07Manufacturing dummy y0.28 y0.92 y0.18 y0.61 y0.49 y1.58 y0.42 y1.38Food sector dummy y0.34 y1.09 y0.32 y1.04 y0.45 y1.47 y0.43 y1.47Wood sector dummy y0.06 y0.19 y0.16 y0.46 y0.12 y0.36 y0.27 y0.81Metal sector dummy 1.25 3.00 0.68 1.16 1.23 3.10 0.54 0.98

Ž .Log no. workers , instrumented 0.69 4.53 0.38 1.31 0.76 4.95 0.41 1.49Registered business dummy y0.84 y2.07 y0.45 y0.82 y0.95 y2.42 y0.53 y1.03Subsidiary dummy y0.27 y0.87 y0.17 y0.54 y0.55 y1.75 y0.45 y1.46Past cash-flow problems dummy y0.28 y1.09 y0.25 y1.00 y0.31 y1.22 y0.28 y1.16OrM is a woman y0.55 y0.81 y0.58 y0.91OrM is black y0.51 y1.50 y0.72 y2.15OrM does not socialize with suppliers 0.96 2.27 1.15 2.77OrM socializes with suppliers outside business 0.19 0.72 0.11 0.42Selection term 0.78 0.76 0.74 0.70Number of observations 48 48 47 47Log-likelihood y57.24 y56.05 y53.66 y51.40

2Pseudo R 0.67 0.68 0.69 0.70


gender dummies. Results are shown in the third and fourth columns of Table 4a–c.The coefficient on ethnicity is significantly negative in all three regressions;gender is significant in the pooled sample and in Kenya. Other things being equal,the share of credit purchases in total purchases is 27–35 percentage points lowerfor black firms, and 36–60 percentage points lower for female-headed firms.Although these results do not, per se, constitute evidence of discrimination, theynevertheless suggest that ethnicity and gender are obstacles to supplier creditirrespective of firm size. In fact, the effects of gender and ethnicity are so strongthat the coefficient of firms size becomes non-significant in all regressions. Thesignificant coefficient of firm size in column 1 thus appears entirely due its

Ž .correlation with ethnicity and gender see Table 2 . The country dummy alsobecomes significant: as anticipated, firms give out more trade credit in Zimbabwethan Kenya.

What remains unclear is why ethnicity and gender matter. To investigate thisissue, we replace ethnicity and gender with network effects. Respondents wereasked to describe their relationship with their suppliers and the extent to whichthey socialize during and outside business. From these responses, two dummyvariables were created. The first one identifies firms that deal with suppliers in anentirely anonymous fashion; the second takes the value of one when the respon-

Ždent socializes with suppliers outside business e.g., through sporting events,.community gatherings, and religious celebrations . Respondents who socialize

during business hours constitute the omitted category. Results show that en-trepreneurs who socialize with suppliers receive significantly more trade credit inthe pooled sample and in Kenya; in Zimbabwe, the effect is not significant,however. Network effects thus appear stronger in Kenya than Zimbabwe, a resultin line with the absence in Kenya of a credit reference bureau that circulates creditworthiness information widely. When gender, ethnicity, and network variables arecombined, results are more mixed, probably because of multicollinearity in the

Ž . Ž .data last two columns of Table 4a–c : network effects are jointly significant inKenya and Zimbabwe, but gender and ethnicity factors remain present. From this,we conclude that gender and ethnicity influence access to supplier credit in waysthat are at least partly accounted for by network effects.

To further investigate the determinants of access to trade credit, we examinewhether respondents ever receive instant credit from first-time suppliers. To do so,

Ž .we run a probit regression similar to Eq. 1 on whether or not the respondent isusually offered credit by first-time suppliers. Results are shown in Table 5a–c.Although less significant, they confirm that ethnicity and socialization play a rolein accessing supplier credit.

Next, we examine whether payment terms vary with ethnicity. Results, reportedin Table 6a–c, suggest that size, sector of activity, and past cash-flow problemsinfluence the duration of supplier credit. Gender is not significant, but being blacktranslates into shorter credit terms in the pooled data and in the Zimbabwesubsample. Network effects are significant and have the expected sign in the


pooled sample and in Kenya, but they have the wrong sign in Zimbabwe. There,firms that deal anonymously with suppliers receive longer credit terms. Onepossible interpretation is that the presence of a credit reference bureau makes itpossible for firms to deal at arms length.

We conduct a similar analysis on the implicit monthly interest rate charged bysuppliers. Sectoral dummies control for various factors including the effect of themarket power of suppliers on the likelihood of price discrimination. 21 To controlfor selectivity bias — interest payments being observed only for firms receivingsupplier credit — we run a two-step Heckman procedure. The regressors that

Ž . 22appear in Eq. 1 are used for the selection equation. Due to the possiblepresence of omitted variable bias, the results must be interpreted with caution butthey suggest that female-headed firms pay a higher implicit interest rate. Othervariables are not significant. 23 We find, therefore, little evidence of discrimina-tion in credit terms. This concludes our analysis of supplier credit received by

Ž .surveyed firms Table 7 .Next, we examine the determinants of bank credit in the form of both overdraft

facilities and bank loans. We have information on whether firms have an overdraftfacility — by far the most common form of bank finance — and whether theyhave ever received a bank loan. Probit regression results are summarized in Table8, using the same regressors as for supplier credit. Results show that, in contrast tosupplier credit, ethnicity and gender have essentially no effect on the use of bankloans and overdrafts once firm size is controlled for. Network effects are signifi-cant, however: not socializing with bank staff has a strong negative effect onaccess to bank finance. Socialization with bank staff outside business also anegative effect on access to bank credit. What matter most for bank credit thusseems to be personal interaction of a business-like character. These findingsprovide little evidence of ethnic or community network bias in the attribution ofbank credit in the two surveyed countries.

21 Unfortunately, we do not have data on the actual market power of suppliers in various sectors ofactivity.

22 Since we do not dispose of truly convincing instruments — a common problem in sample selectioncorrection models — identifying restrictions must be imposed in a somewhat ad hoc manner. Wesimply omit from the interest rate equation the variables that we think are less directly related tointerest charges and focus on variables of interest such as size, ethnicity, and network effects.

23 Discussions with respondents indicate that firms explicitly mention cash discounts only to clientswho may take advantage of them. They usually refrain from mentioning cash discounts to those whoare very unlikely to pay early. Respondents argue that doing otherwise would weaken their position in

Ž .price negotiations with the client see Fafchamps, 1997; Fafchamps et al., 1994, 1995 for details . Iftrue, this attitude generates another form of sample selection bias that could explain why ethnicity andsocialization are not significant.


Table 7Heckman regression on implicit interest rate for supplier creditStandard errors are not corrected for the presence of predicted variables in the regression.Dependent variable is the log of the implicit monthly interest rateq1, computed as cash discount=30divided by the delay between delivery and payment; one-step estimator using ills ratio correction.

Coefficient t-statistic Coefficient t-statistic

Intercept 1.59 3.04 1.65 3.18Zimbabwe dummy 0.76 2.12 0.80 2.08

Ž .Log no. workers , instrumented y0.14 y1.07 y0.16 y1.13Past cash-flow problems dummy 0.04 0.18 0.05 0.18OrM is a woman 0.81 1.84 0.81 1.81OrM is black y0.23 y0.67 y0.28 y0.82OrM does not socialize with suppliers 0.10 0.27OrM socializes with suppliers outside business y0.02 y0.09

Selection equationIntercept y0.32 y0.14 y0.32 y0.17Zimbabwe dummy y0.04 y0.03 y0.04 y0.03Manufacturing dummy y2.88 y1.48 y2.88 y1.58Food sector dummy y3.42 y2.30 y3.42 y2.43Wood sector dummy y1.72 y1.23 y1.72 y1.36Metal sector dummy y0.82 y0.41 y0.82 y0.61

Ž .Log no. workers , instrumented 1.47 1.80 1.47 1.86Registered business dummy 0.91 0.86 0.91 0.91Past cash-flow problems dummy y0.21 y0.19 y0.21 y0.19OrM is a woman y1.90 y1.24 y1.90 y1.31OrM is black 0.45 0.33 0.45 0.40OrM does not socialize with suppliers y1.03 y1.06 y1.03 y1.24OrM socializes with suppliers outside business y0.60 y0.37 y0.60 y0.37Correlation between errors y0.36 y0.42Number of observations 59 59Log-likelihood y56.04 y56.59

We now turn to the credit that respondent firms give to their clients. Theestimated regression is:

ˆ ˆS sv qv X qv D qv P qv O qv C qe 3Ž .i 0 1 i 2 i 3 i 4 i 5 i i

Independent variables X and D are as before, except that the legal status of thei i

firm is dropped from X ; there is no reason for it to influence firms’ willingness toi

give credit. We expect Zimbabwean firms to sell more on credit because screeningis facilitated by the presence of a credit reference bureau. Manufacturing firms inthe sample seldom retail to final consumers; they are therefore expected to sellmore on credit than trading firms. Firms in the food sector should offer less creditgiven that their output is perishable and turnover is fast. Large firms are antici-pated to sell in larger quantities and thus to resort more to credit sales.

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

225

Table 8Probit regressions on overdraft facility and bank loan: pooled dataStandard errors are not corrected for the presence of predicted variables in the regression.

Ž . Ž .Dependent variable is one if the firm has a bank overdraft facility first regression or one if the firm has ever received a bank loan second regression .

Overdraft Bank loan


Intercept y2.01 y2.41 y0.97 y0.74 y0.91 y1.03 0.76 0.62Zimbabwe dummy y0.70 y1.27 y1.37 y1.77 y1.16 y2.04 y0.99 y1.59Manufacturing dummy y0.52 y1.23 0.06 0.11 0.32 0.83 0.75 1.63Food sector dummy 0.14 0.30 0.29 0.48 0.55 1.21 0.62 1.26Wood sector dummy 0.07 0.16 y0.12 y0.19 0.35 0.74 0.12 0.24Metal sector dummy 1.65 2.70 1.82 2.30 y0.45 y0.71 y1.16 y1.45

Ž .Log no. workers , instrumented 0.99 3.14 1.27 2.71 0.19 0.64 y0.14 y0.38Registered business dummy y0.63 y1.16 y1.49 y1.99 0.85 1.45 0.88 1.32Subsidiary dummy y1.60 y2.78 y1.85 y2.67 y0.86 y1.71 y0.67 y1.31Past cash-flow problems dummy y0.41 y1.12 y0.67 y1.42 y0.05 y0.15 y0.09 y0.23OrM is a woman 0.49 0.72 0.60 0.62 0.41 0.56 y0.17 y0.20OrM is black 0.03 0.07 0.37 0.69 y0.10 y0.21 0.06 0.11OrM does not socialize with y1.88 y2.98 y1.00 y1.95bank staffOrM socializes with y2.39 y3.09 y0.34 y0.63bank staff outside businessNumber of observations 100 91 79 73Log-likelihood y42.65 y28.06 y42.22 y37.63

2Pseudo R 0.31 0.47 0.22 0.26

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

226

Table 9Tobit regressions on the share of credit salesDependent variable is the share of credit sales in total sales to clients.The reported estimates are Two-limit censored Tobit. Standard errors are not corrected for the presence of predicted variables in the regression.


( )a Pooled dataIntercept y0.02 y0.09 y0.76 y3.60 y0.43 y1.43 y0.36 y1.28Zimbabwe dummy 0.18 1.36 0.15 1.31 0.19 1.53 0.11 1.10Manufacturing dummy 0.27 2.74 0.49 4.79 0.45 4.40 0.33 4.04Food sector dummy y0.54 y4.41 y0.30 y1.93 y0.41 y2.43 y0.46 y3.21Wood sector dummy y0.14 y1.13 y0.14 y1.00 y0.23 y1.49 y0.15 y1.21Metal sector dummy y0.18 y1.29 y0.20 y1.50 y0.22 y1.64 y0.17 y1.49

Ž .Log no. workers , instrumented 0.12 2.23 0.00 0.05 0.04 0.58 0.03 0.55OrM is a woman y0.10 y0.50 0.01 0.05 0.06 0.37OrM is black y0.30 y2.65 y0.24 y1.74 y0.11 y1.01Share of credit purchases, instrumented 0.91 3.29 0.56 1.64 0.26 0.92Prob. overdraft, instrumented 0.41 2.12 0.24 1.14 0.36 2.08Past cash-flow problems dummy y0.07 y0.62 y0.10 y0.93 y0.13 y1.54Client screening using forms y0.05 y0.45Client screening using reputation 0.39 3.25Client screening using trial period 0.02 0.31Client screening using personal investigation y0.19 y2.35Selection term 0.37 0.32 0.31 0.23Number of observations 79 68 68 59Log-likelihood y41.58 y27.88 y26.32 y6.83

2Pseudo R 0.41 0.53 0.56 0.84

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

227

( )b Zimbabwe onlyIntercept y0.23 y0.69 y0.79 y2.67 y0.53 y1.16 0.15 0.44Manufacturing dummy 0.36 3.80 0.56 4.90 0.50 4.23 0.23 3.01Food sector dummy y0.51 y4.50 y0.52 y3.97 y0.55 y4.17 y0.43 y4.50Wood sector dummy y0.10 y0.81 y0.12 y0.80 y0.16 y1.06 y0.17 y1.69Metal sector dummy y0.04 y0.32 y0.18 y1.08 y0.12 y0.69 0.00 y0.01

Ž .Log no. workers , instrumented 0.17 2.87 0.10 1.29 0.16 1.76 0.11 1.90OrM is a woman 0.12 0.59 0.13 0.64 0.01 0.05OrM is black y0.21 y1.86 y0.20 y1.25 y0.20 y2.07Share of credit purchases, instrumented 0.64 1.76 0.28 0.65 0.07 0.23Prob. overdraft, instrumented 0.29 1.11 0.14 0.49 y0.03 y0.17Past cash-flow problems dummy y0.07 y0.64 y0.12 y1.07 y0.10 y1.45OrM does not socialize with clients y0.18 y2.29Client screening using forms y0.16 y1.33Client screening using reputation 0.36 2.74Client screening using trial period y0.11 y1.85Client screening using personal investigation y0.40 y4.82Selection term 0.29 0.28 0.27 0.15Number of observations 51 45 45 40Log-likelihood y14.16 y11.77 y10.59 12.47

2Pseudo R 0.58 0.63 0.67 1.62


As discussed at the beginning of this section, ethnicity should have no effect oncredit sales if ethnic bias is solely the result of statistical discrimination. If, incontrast, it is due to network effects, then it should have a significant negativeeffect on the share of sales firms make on credit. Finally, the effect of ethnicityshould be positive if discrimination is due to taste. As emphasized before, thesetests are vulnerable to omitted variable bias. In particular, lack of working capitalamong black firms could generate a spurious correlation between ethnicity and Si

if not properly controlled for. Three additional variables are therefore included inˆŽ .Eq. 3 to control for access to working capital: P , the share of purchases oni

ˆcredit, O , whether the firm has a bank overdraft facility or not, and C , whetheri i

the firm experienced a serious cash-flow problem in the past. Firms that facedcash-flow problems in the past or which have difficult access to working capitalfinance are expected to offer less credit to their clients. Because of potential

ˆ ˆ 24 ˆendogeneity bias, P and O are instrumented. For P , the instrumentingi i iˆregression is given in the last column of Table 4a–c. For O , it is given in Table 8.i

In both cases, the identifying restrictions are the socialization variables, theregistered business dummy, and the subsidiary dummy; the former are creditcategory specific, the latter should have no direct effect on credit sales.

Ž . 25Coefficient estimates for Eq. 3 are reported in Table 9a and b. The first setof regressions use firm characteristics X , plus gender and ethnicity D . Asi i

expected, manufacturing firms and large firms sell more on credit, firms in thefood sector less. The Zimbabwe dummy has the correct sign but is non-significant.Gender appears to have no effect that is not already captured by other variables.Being black, however, has a significant and independent effect on credit sales. Totest whether access to working capital influences credit sales, we reestimate Eq.

ˆ ˆŽ .3 with P , O , and C but without gender and ethnicity. Results indicate that, asi i iŽexpected, access to working capital — particularly supplier credit overdraft

.finance is significant only in the pooled sample — influence customer credit:buying an additional 10 percentage points of one’s inputs on credit translates intoan additional 9.1 percentage points of credit sales. To control whether ethnicity isin fact capturing difference in access to working capital, we then reestimate Eq.Ž .3 with ethnicity and working capital variables. Pooled sample results suggest thatboth factors are at work: the coefficient on ethnicity is significantly negative,while the coefficient on supplier credit remains positive and is very close to beingsignificant. Signs are correct but coefficients are not significant in the Zimbabwe

24 C is a past event and can thus be regarded as predetermined. Our identifying restrictions rest oni

the continued assumption that incompetence is not directly correlated with the willingness to grantcredit to clients. This implies that, where patterns of trade credit dictated by statistical discrimination,indicators of competence affecting access to credit should not affect the granting of credit exceptthrough their effect on the working capital of the firm.

25 The number of valid observations was too small to compute meaningful estimates for Kenya alone.


only regression, possibly because of smaller sample size and multicollinearityproblems.

Ž .We also have information about socialization with clients Zimbabwe only andabout the screening practices followed by respondent firms. 26 This information isused to construct one network variable N C and four dummy variables T . The firsti i

of these dummy variables takes the value of one if the respondent requires clientsto fill forms before granting trade credit. The second is one if the respondent usessome sort of information sharing mechanism to assess the reliability of prospectiveclients: e.g., credit reference bureau, references from members of the community.The third is one if the firm observes a client’s purchase and payment behaviorover a period of time before granting trade credit. The fourth take the value of oneif the respondent investigates clients directly, for instance, by visiting their homeor place of work. The estimating equation becomes:

ˆ ˆ CS sv qv X qv D qv P qv O qv C qv N qv T qe 4Ž .i 0 1 i 2 i 3 i 4 i 5 i 6 i 7 i i

Variable N C only exists for Zimbabwe. If information sharing truly helps screeni

reliable clients, firms that use reputation when screening should sell more oncredit. In contrast, firms that must spend time and effort directly investigatingclients instead of relying on their reputation should offer less credit to theircustomers. Formal procedures, by themselves, should make no difference unlessthey are used to directly assess the customer or to seek information from others.

Results shown in the last columns of Table 9a and b indicate that screeningmethods are a major determinant of credit sales. Individual coefficient estimatesare largely consistent with expectations: firms that rely on information sharing toscreen prospective credit recipients sell an additional 36%–39% of their output oncredit relative to others; firms that must investigate clients directly sell 19%–40%less of their output on credit. The latter results is consistent with the impressiongathered from interviews that firms rely on direct screening only when they do nothave access to information sharing. Lack of socialization with clients and thereliance on trial periods are associated with lower credit sales in the Zimbabweregression. Forms per se have no effect on credit sales, unless they are used incombination with an information sharing network. All these results conform with

26 Because screening procedures are chosen by respondents, they are potentially subject to endogene-ity bias. We do not, however, dispose of suitable instruments that would predict the choice of screeningmethod. Regressing screening variables on X , D and N C indicates that very little of the variation ini i i

screening methods can be explained by observed firm characteristics. Qualitative information gatheredin the field leads us to suspect that screening methods are largely dictated by the options available toindividual respondents. Firms that are part of an information sharing network rely on it to screenclients; those that are not must rely on trial period or personal inspection. The use of a particularscreening method thus serves as a precious, even if potentially biased, indicator of its availability to therespondent.

()

M.F

afchampsr

JournalofD

eÕelopm

entEconom

ics61

2000205

–235

230

Table 10Probit regressions on trade credit at first sale: Zimbabwe onlyDependent variable is one if respondent usually offers trade credit to first time clients, zero otherwise.The reported estimates are standard probit. Standard errors are not corrected for the presence of predicted variables in the regression.


Intercept y0.67 y0.45 y1.84 y1.41 y4.00 y1.50 y3.26 y0.48Manufacturing dummy 0.12 0.27 0.74 1.27 0.94 1.40 6.60 1.61Food sector dummy 0.15 0.29 1.20 1.75 1.30 1.78 10.52 1.75Wood sector dummy 0.03 0.05 1.34 1.53 1.71 1.75 y0.41 y0.21Metal sector dummy 0.30 0.47 y0.64 y0.79 y0.59 y0.68 y7.57 y1.47

Ž .Log no. workers , instrumented 0.19 0.73 y0.41 y1.05 y0.34 y0.79 y2.62 y1.49aOrM is a woman 0.63 0.75 1.41 1.28

OrM is black y0.60 y1.19 0.48 0.48 y2.92 y0.96Share of credit purchases, instrumented 4.54 1.92 5.42 1.80 41.95 1.66Prob. overdraft, instrumented y0.46 y0.33 0.31 0.19 7.44 1.08

aPast cash-flow problems dummy 0.06 0.13 0.09 0.17OrM does not 17.38 1.74socialize with clientsClient screening using forms y25.54 y1.67Client screening using trial period y10.50 y1.96Client screening using y4.69 y2.04personal investigationNumber of observations 52 45 45 43Log-likelihood y32.05 y23.75 y22.83 y7.83

2Pseudo R 0.06 0.17 0.20 0.71

a Variables had to be dropped due to insufficient variation in the data.


expectations and speak strongly of the importance of information sharing as ascreening mechanism.

Ž .Finally, we run Eq. 4 on a dummy that takes the value one if respondents everŽ .offer credit to first time customers Table 10 . Data is available only for Zim-

babwe and certain variables are dropped due to multicollinearity. Results confirmthe paramount role that screening practices play in credit sales. They also confirmthat firms that receive supplier credit are more likely to offer credit to their clients.

5. Conclusion

We have examined how African manufacturers gain access to supplier and bankcredit, and how they grant credit to their customers. A proper understanding ofthese processes helps better assess not only barriers to enterprise development inAfrica, but also how markets emerge. It is well-known that statistical discrimina-tion and network effects can exclude certain groups of firms from credit marketsand, more generally, from normal commercial practices. Using data from Kenyaand Zimbabwe, we provide preliminary evidence that network effects are presentand deserve serious attention.

The two surveyed countries display high levels of ethnic and gender concentra-tion in manufacturing. Black entrepreneurs and female-headed firms appear tohave a harder time getting supplier credit, but ethnicity and gender plays nosignificant role in access to bank overdraft and formal loans. Variables measuringsocialization and information sharing — what we called network effects — play adeterminant role in access to trade and bank credit, and have an overwhelmingeffect on the granting of trade credit to clients. Although we cannot rule out thepresence of discrimination, our results largely support the idea that network effects

Žplay an important role in explaining patterns of market interaction e.g., Loury,.1998 . Based on discussions with respondents, our interpretation is that black and

female entrepreneurs are penalized by their lack of connections with the businessestablishment, and by the difficulties they face distinguishing themselves from themass of small, inexperienced microenterprises headed by blacks or women. Thesefactors lead to the partial or complete exclusion of many black and female firmsfrom trade credit practices.

Lack of access to supplier credit is likely to hinder firm growth and to preventthem from joining the mainstream. Moreover, to the extent that delaying paymentto suppliers is a major avenue through which firms absorb cash-flow variations,firms that are denied supplier credit are probably more fragile and are expected to

Ž .fail more frequently e.g., Daniels, 1994; Fafchamps et al., 1994, 1995 . Exclusionmay thus become a self-perpetuating process. The presence of these negativefeedbacks could explain why small groups are able to dominate particular indus-tries or activities.


Excluded firms are forced to resort to alternative and, generally, less efficientways of contracting with each other. Together, they create what could be labeled a‘flea market economy’, that is, a secondary economy in which instantaneoustransactions predominate and market institutions remain underdeveloped. Firmsthat operate in that market may invent ingenious, alternative ways of raisingworking capital — like susu collectors and rotating savings and credit associa-

Ž . Ž .tions ROSCAs e.g., Aryeetey and Steel, 1993; Besley et al., 1993 . But theyremain cut from mainstream institutions and cannot gain the experience requiredto compete in the global market place. Because large segments of the en-trepreneurial population are prevented from reaching their potential, growth anddevelopment remain stunted.

The close ties that existing firms have weaved among themselves are them-selves an efficient response to asymmetric information and contract enforcement

Žproblems. Networks constitute a valuable form of social capital e.g., Fafchamps.and Minten, 1998a,b . Attempts by governments to alter the ethnic makeup of

business through forceful removal of non-indigenous groups and other strong-handapproaches can result in a massive loss of network capital and result in asignificant deterioration in the level of market sophistication. The conceptualapproach proposed here suggests another way out of the quandary. Non-indige-nous groups in Kenya and Zimbabwe appear to owe at least part of their success totheir ability to identify each other. One way to assist indigenous business couldthus be to ensure that credit reference information circulates widely, so as tominimize the role of old boys networks. 27 More research is needed on theseimportant policy issues.

Acknowledgements

I benefited from discussions with and comments from Chris Udry, Avner Greif,Masa Aoki, Jonathan Conning, Pradeep Srivastava, Tyler Biggs, John Pender, PaulCollier, Jan Willem Gunning, two anonymous referees, and participants to variousseminars and conferences and to the symposium on Markets in Africa held inStanford on March 29–31, 1996.

27 The experience of Zimbabwe suggests that such a move is alone insufficient to break the existingbarriers that indigenous firms face. It may have to be combined with another approach adopted, forinstance, by the Kenya Industrial Estates project. This approach consists in setting a small creditprogram that monitors repayment closely and keeps track of the credit history of its members. Theinformation can then be disseminated and help reliable small business graduate into a larger firm pooland gain wider access to credit.


References

Akerlof, G.A., 1985. Discriminatory, status-based wages among tradition-oriented, stochasticallyŽ .trading coconut producers. Journal of Political Economy 93 2 , 265–276.

Amselle, J., 1977. Les Negociants de la Savanne. Editions Anthropos, Paris.´Ž .Arrow, K.J., 1972. Models of job discrimination. In: Pascal, A.H. Ed. , Racial Discrimination in

Economic Life. Heath, Lexington, MA.Aryeetey, E., Steel, W.F., 1993. Individual savings collectors in Ghana: future financial intemediators?

The World Bank, Washington, DC, mimeograph.Bade, J., Chifamba, R., 1994. Transaction costs and institutional environment, the manufacturing sector

in Zimbabwe: dynamics and constraints. Free University of AmsterdamrUniversity of Zimbabwe,RPED Country Study Series, The World Bank, Amsterdam.

Bade, J., Gunning, J.W., 1994. The survey, the manufacturing sector in Zimbabwe: dynamics andconstraints. Free University of AmsterdamrUniversity of Zimbabwe, RPED Country Study Series,The World Bank, Amsterdam.

Bauer, P.T., 1954. West African Trade: A Study of Competition, Oligopoly and Monopoly in aChanging Economy. Cambridge Univ. Press, Cambridge.

Becker, G.S., 1971. The Economics of Discrimination. The University of Chicago Press, Chicago,1957.

Besley, T., Coate, S., Loury, G., 1993. The economics of rotating savings and credit associations.Ž .American Economic Review 83 4 , 792–810.

Bigsten, A., Collier, P., Dercon, S., Fafchamps, M., Gauthier, B., Gunning, J.W., Isaksson, A., Oduro,A., Oostendorp, R., Patillo, C., Soderbom, M., Teal, F., Zeufack, A., 1998. Contract flexibility andconflict resolution: evidence from African manufacturing. Department of Economics, StanfordUniversity, Stanford, mimeograph.

Braudel, F., 1986. Civilization and Capitalism. Harper & Row, New York.Coate, S., Loury, G.C., 1993. Will affirmative action policies eliminate negative stereotypes? American

Ž .Economic Review 83 5 , 1220–1240.Cohen, A., 1969. Custom and Politics in Urban Africa: A Study of Hausa Migrants in Yoruba Towns.

University of California Press, Berkeley.Coleman, J.S., 1988. Social capital in the creation of human capital. American Journal of Sociology 94,

S95–S120.Cornell, B., Welch, I., 1996. Culture, information and screening discrimination. Journal of Political

Ž .Economy 104 3 , 542–571.Cuevas, C., Hanson, R., Fafchamps, M., Moll, P., Srivastava, P., 1993. Case Studies of Enterprise

Finance in Ghana. RPED, The World Bank, Washington, DC, draft.Daniels, L., 1994. Changes in the small-scale enterprise sector from 1991 to 1993: results from a

second nationwide survey in Zimbabwe. Gemini Technical Report No. 71, Gemini, Bethesda, MD.Darity, W.A., 1998. Intergroup disparity: economic theory and social science evidence. Southern

Ž .Economic Journal 64 4 , 806–826.Darity, W.A., Mason, P.L., 1998. Evidence on discrimination in employment: codes of color, codes of

Ž .gender. Journal of Economic Perspectives 12 2 , 63–90.Donohue, J.J., 1998. Discrimination in Employment. The New Palgrave Dictionary of Economics and

the Law, Vol. I. MacMillan, London, pp. 615-623.Emery, G.W., 1984. A pure financial explanation for trade credit. Journal of Financial and Quantitative

Ž .Analysis 19 3 , 271–285.Ensminger, J., 1992. Making a Market: The Institutional Transformation of an African Society.

Cambridge Univ. Press, New York.Ž .Fafchamps, M., 1996. The enforcement of commercial contracts in Ghana. World Development 24 3 ,

427–448.


Ž .Fafchamps, M., 1997. Trade credit in Zimbabwean manufacturing. World Development 25 3 ,795–815.

Fafchamps, M., 1998. Market Emergence, Trust and Reputation. Stanford University, Stanford,mimeograph.

Fafchamps, M., Minten, B., 1998. Relationships and traders in Madagascar. Journal of DevelopmentStudies, forthcoming.

Fafchamps, M., Minten, B., 1998. Returns to social network capital among traders. Department ofEconomics, Stanford University, Stanford, mimeograph.

Fafchamps, M., Biggs, T., Conning, J., Srivastava, P., 1994. Enterprise Finance in Kenya. RegionalProgram on Enterprise Development, Africa Region, The World Bank, Washington, DC, draft.

Fafchamps, M., Pender, J., Robinson, E., 1995. Enterprise Finance in Zimbabwe. Regional Program forEnterprise Development, Africa Division, The World Bank, Washington, DC.

Ž .Ferris, S.J., 1981. A transaction theory of trade credit use. Journal of Political Economy 96 2 ,243–270.

Fukuyama, F., 1995. Trust: The Social Virtues and the Creation of Prosperity. The Free PressPaperbacks, New York.

Gambetta, D., 1988. Trust: Making and Breaking Cooperative Relations. Basil Blackwell, New York.Geertz, C., 1963. Peddlers and Princes: Social Change and Economic Modernization in Two Indone-

sian Towns. University of Chicago Press, Chicago.Geertz, C., Geertz, H., Rosen, L., 1979. Meaning and Order in Moroccan Society. Cambridge Univ.

Press, Cambridge.Granovetter, M., 1985. Economic action and social structure: the problem of embeddedness. American

Ž .Journal of Sociology 91 3 , 481–510.Granovetter, M.S., 1995. Getting a Job: A Study of Contacts and Careers, 2nd edn. University of

Chicago Press, Chicago.Greif, A., 1993. Contract enforceability and economic institutions in early trade: the Maghribi traders’

Ž .coalition. American Economic Review 83 3 , 525–548.Greif, A., 1994. Cultural beliefs and the organization of society: a historical and theoretical reflection

Ž .on collectivist and individualist societies. Journal of Political Economy 102 5 , 912–950.Hart, K., 1988. Kinship, contract, and trust: the economic organization of migrants in an African City

Ž .slum. In: Gambetta, D. Ed. , Trust: Making and Breaking Cooperative Relations. Basil Blackwell,New York.

Ž .Heckman, J.J., 1998. Detecting discrimination. Journal of Economic Perspectives 12 2 , 101–116.Himbara, D., 1994. The failed Africanisation of commerce and industry in Kenya. World Development

Ž .22 3 , 469–482.Hoogeveen, H., Tekere, M., 1994. Entrepreneurship: who is a successful entrepreneur? Free University

of AmsterdamrUniversity of Zimbabwe, RPED Country Study Series, The World Bank, Amster-dam.

Jones, W.O., 1972. Marketing Staple Food Crops in Tropical Africa. Cornell University Press, Ithaca.Kranton, R.E., 1996. Reciprocal exchange: a self-sustaining system. American Economic Review 86

Ž .4 , 830–851.Loury, G.C., 1998. Discrimination in the post-civil rights era: beyond market interactions. Journal of

Ž .Economic Perspectives 12 2 , 117–126.Macharia, K., 1988. Social networks: ethnicity and the informal sector in Nairobi. Institute for

Development Studies, University of Nairobi, Nairobi, Working Paper No. 463.Marris, P., 1971. African businessmen in a dual economy. Journal of Industrial Economics 19,

231–245.Meillassoux, C., 1971. The Development of Indigenous Trade and Markets in West Africa. Oxford

Univ. Press, Oxford.Milgrom, P.R., North, D.C., Weingast, B., 1991. The role of institutions in the revival of trade: the law

Ž .merchant, private judges, and the champagne fairs. Economics and Politics 2 19 , 1–23.


Mitchell, J.C., 1969. Social Networks in Urban Situations: Analyses of Personal Relationships inCentral African Towns. Manchester Univ. Press, Manchester.

Montgomery, J.D., 1991. Social networks and labor-market outcomes: toward an economic analysis.Ž .American Economic Review 81 5 , 1408–1418.

Nadiri, M.I., 1969. The determinants of trade credit in the U.S. total manufacturing sector. Economet-Ž .rica 37 3 , 408–423.

Neal, D.A., Johnson, W.R., 1996. The role of premarket factors in black-white wage differences.Ž .Journal of Political Economy 104 5 , 869–895.

Platteau, J., 1994. Behind the market stage where real societies exist: Part II. The role of moral norms.Ž .Journal of Development Studies 30 4 , 753–815.

Risseeuw, P., 1994. Firm growth in Zimbabwe 1981–1993, the manufacturing sector in Zimbabwe:dynamics and constraints. Free University of AmsterdamrUniversity of Zimbabwe, RPED CountryStudy Series, The World Bank, Amsterdam.

Ž .Saloner, G., 1985. Old boy networks as screening mechanisms. Journal of Labor Economics 3 3 ,255–267.

Schwartz, R.A., 1974. An economic model of trade credit. Journal of Financial and QuantitativeŽ .Analysis 9 3 , 643–657.

Ž .Schwartz, R., Whitcomb, D., 1979. In: Bicksler, J. Ed. , The Trade Credit Decision, Handbook ofFinancial Economics. North-Holland, New York.

Shillington, K., 1989. History of Africa. St. Martin’s Press, New York.Staatz, J.M., 1979. The Economics of Cattle and Meat Marketing in the Ivory Coast. Livestock

Production and Marketing in the Entente States of West Africa, University of Michigan.Taylor, C.R., 1997. The old-boy network and the young-gun effect. Department of Economics, Texas

A&M University, College Station, TX, mimeograph.Yinger, J., 1998. Evidence on discrimination in consumer markets. Journal of Economic Perspectives

Ž .12 2 , 23–40.

Documents

Ethnicity and Credit in African Manufacturing