Heterogeneity and School Committee Structure and Performance in Kenya

Document Sample
Heterogeneity and School Committee Structure and Performance in Kenya Powered By Docstoc
					                              You Can’t Save Alone:
       Testing Theories of Rotating Savings and Credit Associations in Kenya

                                                Work in Progress
                                                 January 2003

                                            Mary Kay Gugerty
                                          University of Washington

  Correspondence to: Evans School of Public Affairs, University of Washington, PO Box 353055, Seattle , WA 98177.
Phone: 206,221-4599; email: The author wishes to thank the World Bank Social Capital
Initiative and the MacArthur Network on Inequality for support of this work, as well as ICS Africa for their cooperation
and assistance. I also thank Ted Miguel, Ashok Rai, Michael Kremer, David Laibson, Peter Timmer, Merilee Grindle,
Robert Bates and seminar participants at Harvard University and the NEUDC Conferences at Cornell and Harvard
University for helpful comments. In addition, Sylvie Moulin, Moses Osia, John Ikoluot, Florence Makokha, Carol
Nekesa, George Egisa, Carol Eve Andera, Chris Webo, Elizabeth Beasley, and the staff of ICS-Kenya provided
tremendous research assistance, administrative support, and insight. All errors, however, are mine.


        This paper examines how and why individuals develop and maintain local-level financial savings
organizations known as rotating savings and credit organizations, or roscas. Understanding how
communities can develop financial institutions that overcome market imperfections is important for several
reasons. Theoretic work on income distribution and wealth suggests that disparities in income and wealth in
the presence of credit market imperfections can lower long-run growth. The study of roscas illuminates how
communities can develop institutions that overcome these market imperfections. Saving is essential to asset
accumulation; understanding how and why the poor save is therefore critical to reducing inequality. Unequal
access to resources may also hinder the development of effective mechanisms for local governance, either
because lower levels of trust make strategies more difficult to sustain or because norms of reciprocity and
fairness may diverge. In countries or communities with missing economic markets, weak states or
ineffective local political institutions, it is important to understand how citizens build institutions that
compensate for missing political and economic markets.
        From a policy perspective, understanding how informal financial institutions are organized is
important for several reasons. The growing microfinance movement has adopted many of the principles of
roscas in developing group lending programs, yet there is little empirical evidence on the effectiveness of
these principles in sustaining these programs. Moreover, many microfinance institutions are expanding to
provide a wider range of banking and savings services to the poor; understanding how lower income
individuals save will be critical to designing appropriate and effective banking products for the poor.
        Rotating savings and credit associations (roscas) are among the oldest and most prevalent savings
institutions found in the world and play an important role in savings mobilization in many developing
economies. Roscas are locally organized groups that meet at regular intervals; at each meeting members
contribute funds that are given in turn to one or more of the members. Once every participant has received
funds, the rosca can disband or begin another round. In joining a rosca, an individual agrees to a schedule of
periodic payments in return for which she receives a lump-sum payment at a future date. Roscas often pay
no interest and participants often have little or no control over when they receive the funds. Participants also
bear the risk that other participants may not fulfill their obligations.
        Why would anyone join a rosca instead of saving alone? Economic theories suggest that individuals
join roscas to finance the purchase of a lumpy durable good, as a response to intra-household conflict over
savings, or to provide insurance. This paper evaluates these theories of roscas using a detailed data set on 70
roscas and 1066 rosca members in western Kenya and finds they are not fully adequate to explain the

complex institutional structure of the roscas in this sample. Roscas in this sample are not organized solely to
provide credit for the purchase of durable goods since their inflexible order of rotation often means that
participants do not expect to receive funds sooner than by saving alone. Nor are these roscas primarily
insurance mechanisms, since participants have very little control over when they will receive funds. These
roscas do not provide a solution to intra-household conflict over spending, since husbands usually know
when their wives receive funds and contributions and rewards are shared among household members. There
is also little evidence that rosca participants employ severe community-level social sanctions against
individuals who do not make payments. Instead, roscas participants appear able to distinguish between cases
of malfeasance and cases of genuine need and design internal punishment mechanisms that do not jeopardize
participants‘ social reputations.
        The paper proposes an alternative hypothesis for rosca participation: saving requires self-discipline,
and roscas provide a collective mechanism for individual self-control in the presence of time-inconsistent
preferences and in the absence of alternative commitment technologies. As many rosca participants put it:
―you can‘t save alone.‖ A simple theoretical example identifies the conditions under which roscas can serve
as a commitment device and shows how repeated interaction can sustain roscas even without the use of
community-level social sanctions. The commitment hypothesis is also consistent with the empirical
evidence. First, most rosca participants report that they join roscas in order to commit themselves to saving,
―to get the strength to save.‖ A key feature of roscas is the public nature of deposits and the inability to
withdraw funds once they have been deposited. Second, participants do not always value earlier positions in
the rosca allocation more highly, implying credit is not the reason for joining a rosca. Third, roscas monitor
and enforce an individual‘s payments to herself as well as her payments to the group. Finally, many rosca
participants ―bind their hands‖ through the use of a pre-commitment mechanism in which participants agree
in advance on how they will use their funds and the group monitors the individual to ensure that she honors
her commitment.
        In addition to evaluating economic theories of roscas, the paper also examines theories of roscas that
stress the social connectedness of rosca members and the role of social capital in building and maintaining
roscas. Descriptive evidence from sociological and anthropological studies as well as theoretic work in
institutional economics suggests that the dense network of social relations in which participants are
embedded allows them to undertake credit transactions where formal banks could not. The information and
enforcement advantages held by residents of a close-knit community allow groups to screen out poor risks
through peer selection (Stiglitz 1990) and to monitor other group members‘ level of effort by peer
monitoring (Varian 1990). This informational advantage is backed up by the power of social sanctions:
group members have the ability to exclude defaulters from community networks that provide insurance and
access to local resources. Much of the descriptive evidence on roscas stresses the effectiveness of these

mechanisms and the rarity of any kind of default (Ardener 1964; Ardener and Burman 1995). This paper
finds that default in many roscas in the sample is not rare, and that the social connectedness of participants
has a somewhat different implication: rosca members are able to distinguish between default due to genuine
need and cases of strategic opportunism. Where rosca participants cannot pay for good reason, rosca
members are reluctant to punish their peers by damaging their reputations in the community. Participants are
keenly aware of the value of social relations and social capital and would prefer not to diminish the social
collateral of their peers. The reluctance to use extra-organizational sanctions to ensure perfect payment
discipline means that roscas must design other mechanisms internal to the organization to prevent
opportunistic behavior. Roscas are able to do this in two ways: by repeating the rosca, so that the benefits of
future participation ensure current cooperation, and by instituting a repayment strategy in which creditors of
rosca defaulters do not pay the defaulters in future rounds. The use of such internal mechanisms to sustain
cooperation is consistent with the evidence on the institutional factors underlying successful collective
action. For example, Ostrom (1990) notes that successful and long-standing resource management
organizations do not rely solely on pre-existing social relations and trust, but establish clear organizational
boundaries, use graduated sanctions, and establish low-cost mechanisms for resolving disputes. Roscas
successfully accomplish all these goals.
        The paper proceeds as follows. The next section reviews related literature and the empirical
evidence on roscas. Section 1 describes the general features of roscas and discusses the data set used in the
paper. Section 2 tests the implications of other theories of roscas and shows how key implications do not fit
the data. Section 3 presents a simple model of roscas showing how individuals aware of their self-control
problem construct savings institutions that circumvent their time inconsistent preferences. Empirical
evidence is discussed. Section 4 presents the implications for policy and concludes.

    1. Related Literature

        One of the most surprising features of roscas is their prevalence. Roscas are found worldwide and in
countries with vastly different levels of economic development. Ardener (1964) and Ardener and Burman
(1995) document the prevalence of roscas in Asia, Latin America, the Caribbean, and Africa. Rosca
participation is particularly high in Africa. Estimates suggest that in 1986, 50 percent of the adult population
in the Congo belonged to a rosca, while participation ranged from 50 to 95 percent in many rural areas in
Liberia, Ivory Coast, Togo, and Nigeria (Bouman 1994). In 1992, membership in roscas in Cameroon was
estimated at 80 percent of the adult population (Bouman 1994) and in several villages in Nigeria in 1987,
adult membership was found to be 66 percent of the population. A sample of 115 households in central
Kenya showed that 45 percent were participating in a rosca (Kimuyu 1995).

        Rosca participation is not limited to developing countries. While roscas are often found in
economies where formal credit markets are thin or non-existent, they are also found in more developed
economies among sub-groups who have been traditionally excluded from mainstream financial services, as
well as among individuals who have reasonable access to formal banking institutions. Roscas have been
reported among employees of the IMF (Ardener 1995) and among bank employees in Bolivia (Adams and
Canavesi 1992) and Ghana (Aryeteey and others 1997). In countries such as Taiwan with relatively well
functioning credit markets, as many as 80 percent of adults are estimated to belong to roscas (Besley and
Levenson 1996).
        The use of such savings commitment mechanisms also has a long history in the U.S. Rotating
savings and credit associations formed the basis for early U.S. savings and loan institutions (Grossman
1992). Christmas clubs were also popular in the U.S. in the early 20th century. In a Christmas club,
individuals deposited savings on a regular basis in a low or no-interest account in a bank; these funds could
not be removed until the end of the year. Most clubs were run by small neighborhood banks who knew their
customers and deposits were made in person; the embarrassment of not making a weekly deposit therefore
encouraged savings discipline (Cosgrove 1924).
        While the popularity of Christmas clubs has declined, 10 million Americans still belong to such
savings clubs (Laibson, Repetto, and Tobacman 1998). In addition, there are numerous savings commitment
technologies available to middle and upper income consumers in the U.S., including IRA‘s, defined
contribution plans, and pensions. Over two-thirds of U.S. household assets in 1994 were held as illiquid
assets (Laibson 1997) such as pension funds and life insurance reserves. Many Americans also make use of
the tax system to save by over-withholding on taxes in order to receive a lump sum (but non-interest bearing)
refund at tax time.
        Savings commitment technologies are receiving increasing attention as a mechanism for increasing
savings among the poor since low-income Americans generally have little access to the institutional savings
mechanisms available to their middle- and upper-class counterparts. For example, the savings discipline
inherent in a Christmas club is being harnessed a century later as part of a ―faith-based savings‖ movement
among U.S. churches in low-income urban neighborhoods. Many churches now run savings circles that meet
on a weekly basis. Participants report that the regular weekly meeting with church peers provides the
discipline to save and to avoid trivial expenditure (Tripoli 2000). In addition, programs such as the Earned
Income Tax Credit (EITC) may promote savings and asset accumulation by the poor because the credit is
normally received as a deferred lump sum after taxes are filed (Smeeding, Phillips, and O‘Connor 2000).
Similarly, individual development accounts (IDA‘s) are increasingly being used to provide poorer Americans
with a savings commitment technology similar to an IRA: withdrawals from IDA accounts are limited to
specific purposes, and participants must often commit to a pre-arranged level of savings (Stegman 1999).

1.1    Description of Data and Sample Areas

        The roscas in the sample are located in Busia and Teso districts in western Kenya, near the border
with Uganda. The two districts cover nearly 2,000 square kilometers and have a population of roughly half a
million people. The groups are located primarily in rural areas with a local economy based primarily on
small-scale farming for subsistence and local market trade. There is some limited cash crop production of
cotton, tobacco, and sugarcane. Both districts are relatively poor for Kenya; the average daily agricultural
wage in the area is approximately $0.85.
        The roscas in the sample are all run by local women‘s self-help groups. In addition to running roscas
and agriculture projects, these groups may also provide emergency financial or labor assistance to members.
Many groups also undertake an income–generating project, which might include agriculture, fish farming,
beekeeping, or handicrafts. The goal of most groups is mutual assistance and support, as reflected in some of
the group names: ―In Unity is Strength‖, ―You Reap What you Sow‖, and ―Let‘s Move Ahead.‖
        These mutual assistance groups were originally identified to participate in an agricultural
development project targeted to women and funded by a Dutch NGO working in the area. The NGO
surveyed all women‘s groups in the project area and identified a total of 100 operational groups. Eighty
groups were selected from these 100 groups to participate in an NGO agricultural assistance program during
the period 1997-1998; seventy of these groups run roscas. This sample of roscas therefore represents roughly
70 percent of the groups operating in the project area in 1997-98, with wealthy town groups being somewhat
under-represented. The assistance program provided the groups with agricultural training and implements,
but did not directly affect the operations of any rosca run by the groups.1
        Roscas operate separately from other group activities. Records are kept separately and not all group
members participate in the rosca. This pattern of roscas embedded in mutual assistance groups is quite
common in rural areas in sub-Saharan Africa (Anderson and Baland 2000; Wright 1999; Bortei-Doku and
Aryeetay 1995; Nelson 1995; Shipton 1992; Siebel and Marx 1987) but may not be typical of roscas in other
parts of the world. On average, 83 percent of the self-help group members participate in the rosca.
        The data on the roscas were collected in surveys and interviews conducted in 1998 and 1999. A
trained field officer visited each of the active rosca groups. The group leader, treasurer, and secretary and
group members were jointly interviewed to gather detailed information on the functioning of the rosca. Data
were gathered on rosca structure, size, and meeting frequency, and on the complete cycle of rosca payments

  Gugerty and Kremer (2000) evaluate the impact of NGO agricultural funding on group activities using a prospective,
randomized evaluation and find no effect of funding on any aspect of rosca activity. The rosca sample is evenly
distributed between funded and unfunded groups.

and disbursements for the last completed cycle of the rosca,2 including the timing and amounts of all
payments made (or not made) by rosca participants for the entire round. Limited individual level data
collected from groups is available on all 1,066 rosca members; in addition, more detailed household data are
available on a randomly selected subset of 340 rosca members.

Meetings and rosca organization
         Every rosca must make three decisions: first, how to allocate the funds to participants; second, how
to prevent default once a member has received funds, and third, how to balance the size of contribution, the
number of members, and the frequency of contributions. The allocation order for funds should help to ensure
both that the first person to receive funds does not default and that the last person to receive funds is willing
to participate. Funds can be allocated in a number of ways: randomized draws, bidding, or by some other
criteria such as need, seniority, or residential patterns. The size and frequency of the contribution may be
weighed against participants‘ ability to pay and the intended use of the funds. The larger the membership the
bigger the collective pot, but the greater the temptation of default. In addition, each member must wait
longer to receive the funds unless the frequency of meeting is also increased. In the current sample, the
average rosca contribution, number of members, and frequency of meetings appear to be balanced so that
each member receives the rosca money roughly once in a year. In this paper, a ―round‖ refers to one
complete cycle of the rosca, in which all members have received funds. A ―meeting‖ refers to one individual
period in the rosca in which members contribute and distribute funds.
         Table 1 gives characteristics of the roscas in the sample. The average rosca in the sample has been
running for 6.5 years and has completed five rounds. The longest running rosca has been in place for 19
years and 22 of the roscas have been operating for 10 or more years. The average size of the rosca in these
groups is fifteen, though there are roscas with as many as thirty members in the sample. The average
individual contribution is just under two dollars, usually contributed on a monthly basis. Thus, the average
rosca recipient expects to receive a pot of about twenty-five dollars, or about one-quarter of average monthly
household expenditures which were ninety-three dollars in 1994 (Kenya Bureau of Statistics 1996). This is
enough, for example, to pay school fees for a primary student, to buy two bags of maize, or two iron roofing
sheets, or a mattress or a blanket, but it is insufficient to buy a larger good such as a bicycle or to pay fees for
a secondary school student. The average length of a rosca round in the sample is just under one year.

  Of the 80 women‘s groups in the sample, 77 operated roscas. Four groups had not yet completed their first round so
complete records of meetings were not available. Records were unavailable for an additional 3 groups, due to damage
to the records (water, fire, etc), the group secretary moving away and not leaving the records with the group, or loss of
records. In this paper, we generally restrict our attention to those 70 roscas with complete records that have been in
existence for at least one complete cycle.

           Table 1: Descriptive Characteristics of Roscas and Rosca Participants

Variable                                                     Mean      Standard
Characteristics of Participants

Proportion who are women                                       0.91        0.12
Average years of education                                     5.5          2.3
Average age                                                     40          5.8
Proportion who are married                                     0.99        0.04
Proportion of female participants whose husband lives          0.69        0.23
at her home
Proportion who have formal sector (salaried)                   0.12        0.14
Proportion who report no regular source of off-farm            0.56        0.24
Descriptive Characteristics of Roscas

Number of participants                                        15.24        5.99
Average number of rosca rounds completed                       5.21        4.59
Average number of years rosca has been in operation            6.48        4.92
Proportion of roscas contributing items, no cash               0.06        0.24
Proportion of roscas contributing items and cash               0.21        0.41
Average distance in kilometers of roscas from a               15.44       16.05
banking facility
Average distance in kilometers of roscas from paved            9.47       11.79
Average daily agricultural wage in US $                        0.87        0.09
Rosca Structure

Average number of weeks between meetings                       3.16        1.08
Average contribution per meeting, in US$                       1.94        2.86
Average amount hosts receives in US$                           24.9       31.95
Number of Observations                                          70

         Roscas may organize payments in several ways. Members may meet and exchange funds in a face-
to-face interaction, or they may appoint a rosca ―president‖ or other individual who collects and allocates
funds (sometimes called an ―impersonal‖ rosca). In this sample, all rosca interactions are face-to-face and
the full group makes almost all decisions concerning the rosca. All the roscas but one meet at the home of
the recipient of funds, who is called the ―host.‖ Meetings are generally celebratory events, as reflected in
the local name for roscas: ―merry-go-rounds.‖ Food is served in 85 percent of the roscas and singing and
dancing may accompany awarding of funds. The host is responsible for providing the food, though in 20
percent of the rosca participants make a special contribution for food at the time of the rosca meeting.
         In each of the roscas in the sample, the rosca is repeated with essentially the same membership. This
is characteristic of many roscas, particularly in eastern and southern Africa (Rutherford 2000; Wright 1999;
Ardener and Burman 1995; Bouman 1995; Niger-Thomas 1995; Shipton 1992). Membership in the roscas in
this sample is relatively stable. Members may leave the rosca at the end of a round and new members can be
admitted at that point. In the last round of the rosca, 20 percent of groups admitted new members, but no
group admitted more than two new participants. Over the course of the last round of the rosca, 30 percent of
the roscas had at least one member leave the rosca during the round. In total, 64 out of 1066 or six percent of
participants left the rosca during the last round.

         Descriptive characteristics of rosca participants
         The average rosca participant is 40 years old and has 5.5 years of education. Although men may also
belong to roscas, more than 90 percent of rosca members are women. Almost all rosca members are married
and 70 percent of female participants report that their husbands stay on the same compound with them.3
Only 12 percent of rosca participants have salaried jobs and 56 percent of rosca participants report no source
of income other than their farm. Table 2 shows some characteristics of rosca participants and groups.
         The average distance of a rosca to a paved road is 10 kilometers and no groups are located in the
major town center for the area. Rosca participants have little access to formal credit markets. There are only
three towns with banking facilities in the sample area and the average distance to a bank for the roscas in the
sample is 15 kilometers. Less than 20 percent of the groups could identify a moneylender within an hour‘s
walk of the rosca.

         Sorting and screening of participants

  The relatively low rate of spousal residence may be due to two factors: first, the prevalence of polygamous marriages
in which the man may reside primarily with another wife, and second, relatively high rates of male out-migration to
work in urban areas.

        Issues of selection complicate the estimation of causal impacts in the study of groups. Rosca
participants may differ systematically from non-participants for at least two reasons. First, individuals may
self-select into roscas on unobservable as well as observable characteristics. In addition, roscas are likely to
screen members along both observable and unobservable characteristics. While this sample is quite
representative of the roscas run by women‘s groups in this area, it will be useful to understand the degree to
which sorting and screening affect rosca membership. This section discusses the process of formation of
roscas and assesses the degree to which individuals sort into roscas and the extent of the screening process.
        The vast majority of groups, 80 percent, report that their initial membership was based on a
combination of residence patterns and friendships: most group officials reported that the initial rosca
participants were ―friends and neighbors.‖ Rosca participants report that they have many interactions outside
of the rosca meetings. On average, participants report that another rosca participant is likely to ―step foot in
their home‖ outside the rosca meetings fourteen times in a month. It is difficult to tell, however, whether this
interaction is a criterion used in group formation or the result of years of group activity. When the roscas
first formed, three quarters of the roscas reported that almost all the participants knew each other by name
and knew each other‘s homes. Nonetheless, many roscas also report that they investigated the character
and/or financial position of potential members. Half the roscas investigated potential ability to pay by
making an assessment of the individual‘s financial position such as whether she ran a small business or trade
or had some other sources of income with which to make payments. Forty-one groups investigated potential
participants‘ character, usually verifying their trustworthiness with another group member and making a
general assessment of their behavior in the community. Other criteria used included how well individuals
ran their homes, and whether the potential member‘s family would be supportive of their participation. The
roscas that did not take explicit action to verify participants‘ ability to pay report that they already knew this
information. The fact that so many roscas took explicit steps to screen participants suggests that the high
levels of trust and reciprocity necessary to sustain roscas are developed in part through repeated interactions,
a point also emphasized by Ostrom (1999) and Rutherford (2000). In interviews, many group officials
reported that one of the reasons they started a rosca was to develop ―unity‖ among the members, indicating
that they felt trust would be built by the process of running the rosca.
        Very few roscas had any explicit rules about who could join the roscas at their inception. Fifteen
groups required a small registration fee, four required that participants be drawn from nearby villages, and
five required explicitly that members be married. Five groups were comprised primarily of widows. Many
groups had general requirements for participation which included agreeing to follow the group rules and
having good character, which might include things such as ―not a gossiper,‖ ―not a drunkard,‖ or ―someone
willing to unite with others.‖

        Do rosca participants differ systematically from the average resident of the area? Table 2
summarizes the characteristics of female rosca participants in comparison to the characteristics of the
mothers of primary school students in the same geographic zone. These zones are administrative units
comprised of several school districts;4 there are 22 zones in the sample. Data are available on the
characteristics of the mothers of primary school 6-8th grade students in the region. These women, however,
are not a random sample of the area: they are likely to be wealthier and better educated than average, given
that they have children in the upper level of primary schools. Rosca participants are less likely to have a
formal sector job than the zonal sample, but are more likely to have some post-primary education, and these
means are significantly different at the 1 percent level. Wealth is proxied by ownership of an iron roof on the
main house in the compound. Female rosca participants are more likely to have an iron roof than the average
household in the zone, but this may be due in part to the fact that some roscas use their funds to buy iron-
roofing sheets for members. In general, it would appear that rosca participants are probably wealthier than
the average, but not significantly so.

 Primary schools in Kenya do not have clear district boundaries; rather, a school draws pupils from the immediate area
and the typical limit is walking distance to the school.

                            Table 2: Rosca Members in Comparison to Zonal Averages

       Variable                                        Sample mean        Average for School          Average
                                                                          Zones in which the         difference
                                                                          groups are located      between roscas
       Descriptive characteristics of rosca              (standard       (standard deviation)5    and the closest
       participants                                      deviation)                               primary school
                                                             (1)                      (2)                (3)

       Monthly non-agricultural income                    1045.8

       Proportion of homes with iron roof                 0.503***               0.24
                                                          (0.237)               (0.077)

       Proportion of homes with a latrine.                 0.82                  0.84
                                                          (0.137)               (0.079)

       Proportion of women with post-primary              0.24***                0.18
       education                                          (0.42)                (0.056)

       Proportion of individuals with formal              0.10***                0.20
       sector employment -women                           (0.30)                (0.052)

       Size of the largest ethnic group in the            0.83***                 0.76                 -0.05
       rosca as a percentage of total membership          (0.21)                 (0.15)                (0.19)

       Ethno-linguistic fractionalization6                0.23***                 0.36                 -0.09
                                                          (0.24)                 (0.19)                (0.25)

       Proportion of participants who belong to             0.69                  0.64                 -0.04
       the same church                                     (0.22)                (0.11)                (0.24)

       Size of the largest ethnic group when                0.76                  0.72                 -0.11
       group is less that seven kilometers from a          (0.24)                (0.17)                (0.23)
       paved road
       Size of the largest ethnic group when              0.92***                 0.83                  .02
       rosca is located more than seven                   (0.19)                 (0.10)                (0.09)
       kilometers from a paved road

       Means are significantly different at the 90% (*), 95% (**), 99% (***) level.

 Source: Based on a sample of primary school students attending schools in the zones in which the women‘s groups
are located. These figures are the means for the mothers or households of primary school pupils.
    Defined as the probability that two individuals drawn at random come from two different ethnic groups.

         Roscas do not appear to strongly sort along ethnic lines. Column 3 of Table 2 compares ethnic
heterogeneity in roscas to that in the closest primary school for which data are available. Column 3 indicates
that roscas are not systematically more or less ethnically diverse than the population of nearby schools. This
holds whether the measure used is the size of the largest ethnic group in the rosca as a percentage of total
membership or ethnolinguistic fractionalization.7 The means for rosca heterogeneity and school
heterogeneity are not significantly different for any measure.8 Group membership appears to be primarily
influenced by residential location and ethnic residence patterns, largely a product of historical accident rather
than systematic sorting (Miguel and Gugerty, 2002). Roscas often do not want to accept members from too
far away, for fear they will not be able to attend meetings and make payments. On average, 72 percent of
rosca participants in the sample come from the same village.9 No roscas in the sample have explicit (nor,
apparently, implicit) ethnic membership rules, and in some cases, participants note that including close
relatives in a rosca can bring difficulties in group management as domestic disputes spill over into rosca
         The ethnic homogeneity of groups is related to their geographic location, as measured by the group‘s
distance from a paved road, shown in Table 3. The median group is seven kilometers from a paved road. The
size of the largest ethnic group in roscas located less than 7 kilometers from a paved road is 76 percent of
total membership, close to the zonal average of 78 percent. In roscas more than 7 kilometers from a paved
road, the size of the largest ethnic group is 92 percent of total membership. In general, groups that are farther
away from a paved road differ systematically from those that are closer, having more members, a higher
proportion of members relying solely on farm income, and lower average levels of education. Column 3 of
Table 2 confirms that ethnic sorting is slightly more prevalent in areas at a greater distance from a paved
road. Roscas less than seven kilometers from a paved road are more ethnically heterogeneous than their
closest primary school on average, though the means for roscas and schools are not significantly different.
Roscas further from a paved road are more ethnically homogenous than the nearest primary schools, though
again the means are not significantly different from each other.
         Roscas do not appear to strongly sort along other socioeconomic dimensions. While roscas are
likely to have participants who belong to the same church, the average religious diversity of roscas does not

 Defined as the probability that 2 individuals drawn at random come from two different ethnic groups:
1- i(Proportion of ethno-linguistic group I in the population)2
 Pupil ethnic diversity is not a perfect measure of ethnic diversity in the area, as pupils may also sort into schools along
ethnic lines. However, this sorting is somewhat limited by walking distance. See Miguel, 2000 and Gugerty and Miguel,
 A ―village‖ in the Kenyan context has a different connotation from many other places in Africa. A village is as much
an administrative unit as a social, cultural, or economic one. Village boundaries are not clearly demarcated

differ from zonal or school averages. The size of the largest religious group in the rosca is 0.69; this does not
vary with distance to a paved road, nor by ethnic homogeneity in the group. The average size of the largest
religious group in the zone is 0.6.
         In summary, rosca membership and heterogeneity appears to be based in part on relatively
exogenous factors such as residential location. There is little evidence that roscas sort by ethnicity, religion,
or other socioeconomic factors, though participants do screen entrants for character and ability to pay.

                           Table 3: Rosca characteristics and Geographic Isolation

                                                  Groups located <=7            Groups located >7
                                                   kilometers from a            kilometers from a
                                                       paved road                   paved road
                                                         Mean                         Mean
                                                  (standard deviation)         (standard deviation)

  Proportion of members relying solely                    0.50**                        0.63
  on farm income                                          (0.25)                       (0.21)

  Average years of education of female                     5.61                         4.67
  participants                                            (2.44)                       (2.22)

  Size of the largest religious                            0.68                         0.71
  denomination in group                                   (0.22)                       (0.21)

  Size of the largest ethnic group in the                  0.76***                      0.92
  rosca                                                   (0.24)                       (0.19)

  Size of the largest ethnic group in the                  0.72***                      0.83
  school zone                                             (0.17)                       (0.10)

  Different between size of largest                        0.07                        -0.01
  ethnic group in rosca and in zone                       (0.26)                       (0.41)

  Number of members in the rosca                          14.5**                       19.38
                                                          (5.60)                      ( 7.31)

  Number of observations                                    39                           31

geographically. The area is densely settled and families reside on the family farm rather than living in a village and
walking to farms on the outskirts of the village.

Estimation strategy

        The estimation results presented below all contain indicator variables for the four geographic areas in
which the groups are located to control for unobservable variation across divisions. In addition, some
regressions contain zonal socioeconomic indicators: the proportion of primary school children in the zone
living in a house with an iron roof (a proxy for wealth) and the proportion of mothers of primary school age
children in the zone with post-primary education. Where zonal indicators are included, robust standard
errors are clustered on the zone. Where analysis is done at the individual level, robust standard errors are
adjusted for clustering on the group. All probit estimations report the change in the probability for an
infinitesimal change in each independent, continuous variable and the discrete change in the probability for
dummy variables, rather than the estimated coefficient.

2. Theories of Rosca Participation

        The economic literature on roscas has focused largely on three reasons why individuals might join a
rosca: financing the purchase of an indivisible good, intra-household conflict, and insurance. Sociologists
and anthropologists document how roscas are embedded in pre-existing social networks that serve as a form
of collateral. This section reviews the implications of these theories in light of the evidence from this sample
and shows how key implications of the theories do not fit the data.

Theory One: Financing the purchase of durable goods

        Besley, Coate, and Loury (1993; 1995) suggest that individuals join roscas to finance the purchase of
an indivisible durable good. According to these authors, roscas take advantage of the gains from
intertemporal trade between individuals. All individuals except the last improve their welfare by joining a
rosca, as each receives the indivisible durable good sooner than by saving alone. The arrangement is
sustainable if the benefit of default is lower than the (exogenously set) cost of defaulting.
        Besley, Coate, and Loury model roscas as one-shot games and focus on two mechanisms for
allocating funds: bidding for the pot and concurrent randomization. In the latter, the pot is randomly
allocated to one member at each meeting. However, these allocation mechanisms are not employed among
roscas in this sample. Furthermore, the repeated nature of the interaction makes most roscas substantially
different from a one-shot game. All the roscas in the sample are repeated; once the current round has
finished, the rosca begins again with essentially the same membership. Ninety-six percent of roscas
determine the order of allocation of funds before the rosca begins (table 7). When participants know the
order of allocation before the rosca has begun, everyone enters the rosca knowing her position and there is no

uncertainty to resolve. The last person to receive knows she is last, and she will therefore not receive the pot
any sooner than she would by saving alone. If the motivation for roscas were purely saving for an indivisible
good, then the last person could do just as well by saving alone and would not join. In this case the second-
to-last person becomes the last, and no longer wants to join, and so on, causing the rosca to unravel as each
subsequent individual loses her motivation for participation.10
          In the lumpy durables explanation of roscas, individuals will always prefer to receive funds as early
as possible in the rosca because their welfare gains depend on the number of periods in which they enjoy the
benefits of the durable good. But rosca participants in this sample do not always prefer to receive money
sooner rather than later. There are two factors at work here. First, in a repeated rosca with a repeated order
of allocation, each member is doing no better than she would by saving alone. Each individual is essentially
joining an organization in which she saves a fixed amount each month and receives this money back after the
same interval in each round. This may partially explain why participants see no need to pay interest to
individuals with positions late in the rotation, even when they hold those positions in multiple rounds.
(Anderson and Baland 2000; Platteau 1997; Seibel and Marx 1987). As one participant put it, ―the system is
fair, because even the first person will have to wait until the last person has received to receive again.‖
          Second, there are many other factors that influence the desired position in the rosca, beyond
receiving early in the cycle. Many rosca participants care as much about when in the year they receive funds
as whether they receive funds early in the round. Most roscas report that they prefer to start in January and
end in December. In practice, however, the starting month of roscas varies considerably. Not all participants
can receive in a year, and roscas may also be suspended if there are adverse shocks that affect all members.
As roscas are repeated over and over, the starting month becomes increasingly arbitrary. The start date of the
roscas in this sample is distributed relatively evenly throughout the calendar year, though the largest
proportion start in January, as shown in Figure 1. Rosca participants prefer to receive funds at particular
times of the year rather than always preferring to receive earlier in the cycle. Table 4 shows member
preferences for the receipt of funds by month. Figure 2 represents the same information graphically,
grouping the months into the main agricultural seasons. The ―fees‖ period consists of the months January –
March, when most school fees are due and the hungry season begins. April – June represents the hungry
season just before the first harvest. The largest proportion of participants prefers to receive funds in the
period just after the harvest. Participants felt that money received during this period was not obligated to be

  However, individuals might join the rosca and discover after joining that they are last in the order. They may desire to drop out,
but the sanctions or social disapproval that might follow this action force them to remain in the rosca. This explanation seems
unlikely, both because nearly 40% of the roscas are repeated with the same order of allocation and because there is no evidence that
individuals who leave the rosca suffer severe sanctions as discussed in section 3.4. One might argue that an individual joins a rosca
with an inflexible order to finance the purchase of a durable good, and then waits patiently for someone to drop out of the rosca,
hoping she will move up in the order of allocation. In some future round, then, she may receive the good sooner than by saving
alone. She would undoubtedly be better off, however, by joining a rosca that allocates the pot randomly at each meeting.

used for food or fees: ―I can do something with the money, not just use it on the household.‖ Participants
also felt they could use the money on their business or to buy food cheaply, either as stock for the household
or to sell at a profit later. This preference is strongest for younger women, presumably because they have the
greatest desire to purchase household items and the least need to pay school fees or buy food for a large
number of children. Older participants have somewhat higher relative preferences for receiving early in the
year, when school fees are due and food stocks are low. The holiday season (October – December) is the
least favored time to receive the pot; participants report that money gets ―used up‖ in the entertaining of
visitors and preparations for holiday celebrations.
        The valuation of one‘s position in the rosca depends on two additional factors. First, the cost of
―hosting‖ the rosca may vary with season. Hosting a rosca during the hungry season means finding food for
participants at a time when household stocks are low and food prices are high. Second, default rates on
payments may be higher during the hungry season because individuals have more difficulty in obtaining
funds to make payments. Figure 3 shows the average percentage of unpaid obligations over the calendar
year and indicates that default rates are higher at the beginning of the year when school fees are due and the
hungry season is beginning and again at the end of the year during the holiday season. Default rates are
lowest during the harvest season. This may also explain why many participants prefer to receive funds during
the harvest season.

                                                                           Figure 1
                                                                 Month in Which Rosca Began

Number of roscas   10
                               Jan                         Feb    Mar   Apr May Jun         Jul   Aug Sep     Oct   Nov Dec

                                               Figure 2
                         Rosca Participants' Preferences for Timing of Funds
                        Percentage of


                                                                Fees          Hungry        Harvest     Holiday
                                                                (Jan-Mar)    (Apr-Jun)     (Jul-Sep)   (Oct-Dec)
                                                                            Month in Calendar Year

                                                  Figure 3
                                Percentage of Payments Missed and Not Repaid
                                Percentage of Payments

                                                                 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

                          Table 4: Preferred Month to Host a Rosca
Month – primary characteristic      Percentage          Two most important reasons
January – school fees                  16.8        Pay schools fees – 52.1%
                                                   Can use on business – 29%

February – school fees & start         3.7         Fees - 40.7%
of hunger                                          Buy food (famine) - 33.3%

March –hungry season                   2.0         Buy food (famine) - 80%
                                                   Fees - 20%

April- school fees & hunger            9.7         Fees - 40.7
                                                   Buy food (hunger) -33.3%

May – peak of hunger                   7.1         Buy food (hunger) - 78.9%
                                                   Can buy food cheaply - 25%

June – start of harvest                5.0         Money not obligated elsewhere- 21%
                                                   Buy food cheaply (stock) -21%

July – main harvest                    10.7        Can put money into business -43%
                                                   Buy food cheaply (stock) -21%

August – fees, main harvest            25.2        Can put money into business -29%
                                                   Buy food cheaply (stock) - 25%

September – second harvest             6.4         Buy Food cheaply (stock) -29%
                                                   Money not obligated elsewhere - 18%

October – second harvest               2.4         Money not obligated elsewhere - 33%
                                                   Can put money in business -33%

November – second harvest              1.7         Can buy Christmas items-50%
                                                   Money not obligated elsewhere - 25%

December – holidays                    9.4         Can buy Christmas items -50%
                                                   Fees - 25%

Number of observations                 351

        There is also direct evidence from the data that roscas are not formed solely to purchase lumpy
durable goods. Over half of rosca participants use their rosca winnings for more than one purpose, and one-
fifth use their winnings for more than two purposes. For individuals that use their funds for more than one
item, the average proportion of the pot spent on the most expensive item is 62 percent. If the primary
purpose of roscas was to finance the purchase of indivisible goods as quickly as possible, individuals would
organize roscas to receive the pot more frequently. Participants who purchase divisible items or more than
one indivisible item with their rosca money might expect on average to get one of their items sooner by
saving alone than by waiting their turn in a rosca. In addition, the expenditures generally favored by rosca
participants (household items, school fees, and food) are often divisible. School fees may be paid in
installments and food purchased in small quantities (though often at a higher cost). Table 5 shows that rosca
purchases are not confined to indivisible goods. The two largest categories of expenditures are household
items and school fees. School fees are generally due at the beginning of the school year in January. Many
parents are unable to pay fees all at once, but are allowed to pay in installments without penalty (Gugerty and
Kremer forthcoming). It is more difficult to characterize the divisibility of household items. Many
participants use their funds for cups or plates that could be purchased individually, but may be purchased
more cheaply (or more uniformly) as a set. It is clear, however, that rosca funds are used differently than
general income. Food expenditures accounted for 62 percent of overall household expenditures in 1994
(Kenya Bureau of Statistics 1996), while household items accounted for 0.5 percent and schools fees eight
percent of household expenditures.

                       Table 5: Uses of Rosca Winnings

Percentage of total funds received spent on:                Mean
Household items (cups, plates, cooking pot)                   .24
School fees                                                   .24
Food                                                          .14
Small business                                                .08
Livestock                                                     .05
Illness                                                       .04
Clothes                                                       .04
Farm                                                          .03
Paid debts                                                    .02
Ceremony (funeral, memorial service)                          .02
Home construction                                             .02
Blanket                                                       .01
Gave to relative                                              .01
Other purchases                                               .06

        Table 6 examines the factors associated with the number of uses of rosca funds. The only factor that
is systematically associated with the number of uses of funds is the distance of the group from a paved road,
which is positively associated with the number of uses of the funds. Individuals in roscas in more isolated
areas appear to be less likely to buy lumpy goods than those in more centrally located areas. There is no
association between the use of an allocation mechanism that re-randomizes the funds in each round and the
number of uses of rosca funds. This suggests that there is no relationship between the lumpiness of rosca
purchases and the allocation mechanism, as would be suggested in the Besley, Coate and Loury model.
        In summary, the direct and indirect evidence indicates that these roscas do not exist solely to finance
the purchase of a lumpy indivisible good. Many participants could receive a durable good just as quickly by
saving alone. In addition, once the order of allocation is determined, the last person in the order has the
incentive to leave the rosca, since she realizes no advantage through rosca membership. Participants do not
always favor earlier positions in the rosca; many also care about when in the calendar year they receive
funds, the cost of hosting the rosca, and the probability that other participants will make payments.

                                     Table 6: Use of funds and rosca structure

                                             Number of uses        Number of uses       Percent of funds spent
                                               of funds              of funds              on largest item
 Explanatory variables:                           (1)                   (2)                      (3)

 Number of years rosca has been in                 -0.01                  -0.02                   0.006
 operation                                        (0.02)                 (0.02)                  (0.004)

 Number of rosca participants                      0.03*                 0.03*                    -0.006
                                                   0.01                   .02                    (0.004)

 Distance of rosca from tarmac                    0.02***               0.02***                 -0.005***
                                                  (0.004)               (0.007)                  (0.002)

 Average years of education of members             0.02                  0.041                   -0.003
                                                  (0.06)                 (0.06)                  (0.01)

 Proportion of participants with salaried          0.17                  0.001                    0.00
 income                                           (0.69)                 (0.71)                  (0.16)

 Indicator if roscas re-randomize order                                      0.07                 -0.02
                                                                            (0.16)                (0.03)
 Geographic Indicators                                  Y                     Y                     Y
 Zonal socioeconomic controls                           Y                     Y                     Y
 Number of observations                                68                     68                    68
 R-squared                                            0.25                   0.37                  0.37
 MSE                                                 0.578                  0.533                 0.126
 Mean of dependent variable                           1.89                   1.89                  0.77
    Significantly different than zero at 90% (*), 95% (**), 99% (***). Note: OLS estimation with robust standard
    errors All regressions contain indicator variables for geographic divisions.

Theory Two: Roscas and intra-household conflict

        Anderson and Baland (1999) provide a second rationale for the existence of roscas: intra-household
conflict over savings and consumption patterns. Using data from roscas in a low-income neighborhood in
Nairobi, Anderson and Baland note that the participation rates of women in roscas in a low-income area of
Nairobi are higher for women than for men and that participation rates for married women who work are
higher than for unmarried women who work. They explain this gap with a model that shows that married
women are more likely than men to join roscas because of a stronger preference for saving for indivisible
goods. In their model, rosca participation is positively related to a woman‘s bargaining position in the
household, as measured by her contribution to household income. They argue that husbands tend to prefer
status-enhancing current consumption to saving for household items and school fees.
        There are several reasons why intra-household conflict cannot fully explain rosca membership, both
in this sample and in general. If household conflict drove rosca participation and women were unable to
exert control over their own income, it also seems plausible that men would simply not allow their wives to
join roscas. Moreover, it is not clear what would prevent the husband from demanding the money once the
wife has received it. Anderson and Baland argue that once the lump sum is accumulated, the husband prefers
to spend it on indivisible items. If this is the case, then husbands may suffer from time-inconsistency and use
their wife‘s participation as a commitment strategy.
        The structure of the roscas in this sample also indicates that participants are not seeking primarily to
hide money from spouses. The meetings are always held at the home of the person receiving the pot. Fifteen
visitors are likely to be noticed in a small community: the host‘s spouse and all her relatives are therefore
aware that she has received funds. A more conducive structure for hiding funds might be to randomly select
a participant to receive the money at each meeting (which none of the roscas do).
        The empirical evidence also indicates that husbands are often supportive of their wives‘ membership
in roscas (Dzingirai 2000; Ardener 1995) and often contribute for payments (Niger-Thomas 1995). In a sub-
sample of 102 rosca participants taken from the Kenya sample, 41 percent report that their husband gave
them money for their rosca contributions; the total amount given over the course of a round averaged Ksh
750 (U.S. $15). Many women also gave money to their husbands after they had received the pot. Forty-
eight percent of participants gave their husband funds, on average giving them Ksh 716 (US $14). This
indicates that participation in roscas is often a household strategy, rather than a mechanism used to keep
funds away from husbands.11

  This relatively high figure contrasts with the relatively low number given in Table 6. This could be for several
reasons. First these questions were asked about different rounds of the rosca. Second, the difference may reflect some
under-reporting of payments to husbands during the first round of the survey. The second round asked explicitly about

        Married women appear no more likely to participate in roscas than unmarried women or women who
are household heads. Kimuyu (1995) notes that in a sample of 115 households in central Kenya, female-
headed households are more likely to participate in roscas than are households headed by a man. In the
current sample, women whose husbands live on the same compound with them are no more likely to
participate in the rosca. Women participate in the rosca at higher rates than men on average, but women with
a salaried income are no more likely to participate in the rosca than women without a regular source of
income. Table 8 gives the results of a probit analysis for the likelihood of belonging to the rosca, given that
one is already in the self-help group. The strongest positive effect on participation is for individuals who are
members of the largest ethnic group in the group. Among married women, those whose husbands live at
home are no more likely to participate in the rosca, providing evidence against the intra-household conflict
hypothesis. Unlike Anderson and Baland‘s data, however, women who have a salaried income are less likely
to join the rosca, though the data do not allow the testing of their hypothesis that rosca participation first rises
and then falls with the share of female income in the total household budget. This evidence suggests, as
Anderson and Baland also note, that ―essentially, roscas provide a forced savings mechanism for the
household.‖ (1990, p. 4).

                                       Table 7: Order of Allocation of Funds

   Order of allocation decided:                                              Percentage of roscas
                                                                              using this method

   1. One meeting prior to distribution:

            Negotiate                                                              2.86
            Random allocation                                                      1.43
   Subtotal-decide one meeting prior                                                            4.29

   2. Anew each round:

             Random allocation                                                     35.71
            Negotiate order among members                                          22.86
   Subtotal-change order each round                                                            57.57

   3. To be the same each round:

            Randomized order                                                        17.14
            Negotiate order among participants                                      11.43
            Use order of joining, age, or residence                                  8.57
   Subtotal-use same order                                                                     37.14

   Number of observations                                                             70

payments to husbands. Third, if husbands sometimes gave wives money for contributions, then women‘s payments to
their husbands may be repayment of a loan, and reflected under the category of debt repayment.

                                    Table 8: Rosca Participation and Gender

                                           Probability of being a Member of the Rosca
                               All group           Female group      Female group Female group
                               members                members          members        members
Explanatory                       (1)                    (2)              (3)           (4)
Years in group                   0.003                  0.002                0.002               0.001
                                (0.006)                (0.006)              (0.006)             (0.006)

If female                        0.39***

Years of education              0.01***                0.02***              0.02***              0.01**
                                (0.005)                (0.004)              (0.005)             (0.005)

Age                              0.003                  0.003                0.003               0.002
                                (0.002)                (0.002)              (0.002)             (0.002)

Have a latrine in the            0.005                   0.010                0.02               0.012
home                            (0.055)                 (0 .05)              (0.05)             (0.050)

Have an iron sheet               -0.08**               -0.08**               -0.08**            -0.08**
roof                             (0.03)                (0.03)                (0.03)             (0.03)

Have a salaried job             -0.11**                -0.11**               -0.10*             -0.10*
                                (0.052)                (0.06)                (0.06)             (0.06)

Member of the                    0.17***                0.21***              0.21***            0.21***
largest ethnic group             (0.06)                 (0.07)               (0.07)             (0.07)
in the rosca
Married                                                                       0.07
Husband lives on the                                                                             0.02
same compound                                                                                   (0.03)
Chi-squared                      72.78                  32.48                32.28              29.21
Pseudo R-squared                 0.11                    0.04                0.04                0.04
Number of                        1691                   1338                 1329                1248
Significantly different than zero at 90% (*), 95% (**), 99% (***).
Note: Dprobit estimation with robust standard errors. All regressions contain indicator variables for geographic division.

Theory Three: Insurance

         Individuals might form roscas in order to provide each other with insurance, particularly in a setting
such as rural Kenya where formal markets for insurance are virtually non-existent. The ability to bid on the
pot provides an explicit insurance mechanism because participants can access money when they need it.
Klonner (2000) shows how risk-averse participants in a bidding rosca can insure themselves against
idiosyncratic risks. Calomiris and Rajaraman (1998) show that in Indian roscas with concurrent bidding, the
actual amount received by the winner is subject to variation through the bidding process in a manner
consistent with insurance. This section shows that individuals do not join Kenyan roscas for explicit
insurance, but may use the rosca institution to provide limited implicit insurance when they suffer a negative
         The inflexible order of rotation adopted by most roscas in this sample makes it difficult for
participants to access funds when they need them; these roscas do not appear to supply explicit insurance to
participants. No rosca in this sample allocates the pot by bidding and the use of such auction mechanisms in
African roscas appears relatively limited (Anderson and Baland 2000; Platteau 1997; Ardener and Burman
1995; Shipton 1992). Moreover, the order of allocation is determined before the rosca begins in almost all
roscas. But even if roscas do not provide explicit insurance against idiosyncratic shocks, roscas might
nonetheless provide implicit insurance if individuals are able to delay or forgo payments. Udry (1990) shows
how credit contracts in Nigeria provide state contingent insurance: the timing of repayment depends on the
realization of random production shocks for both borrower and lender. Platteau (1997) argues that roscas
which use the needs of participants to determine the order of allocation (but do not compensate late receivers
with interest) are in effect providing insurance in the form of income redistribution from those with later
positions to those with earlier positions.
         There are several ways in which roscas could provide implicit insurance. First, roscas could allow
members some say in when they receive the funds, as occurs in roscas with a ―negotiated‖ order of
allocation. Second, roscas could allow members to switch places after the order has been determined. Third,
roscas could allow for some kind of state-contingent payment, allowing individuals longer to make payments
if they received a negative shock. Fourth, roscas may forgive members a payment if they receive a negative
shock. Finally, rosca participants may make side deals outside of the rosca. This section of the paper
reviews each of these issues in turn.

         Determining the order of allocation by need

         Some insurance may be provided in roscas that negotiate the order of allocation at the start of each
round. These groups may consider any negative shocks an individual has received in the preceding months

in deciding the order of allocation. Table 7 shows that 23 percent of roscas negotiate the order of allocation
in this manner. In a negotiated order, each member gives her preferred date of receiving the funds, gives her
reasons, and the group negotiates a final order. Four of the nineteen groups that negotiate the order of
allocation explicitly mention need as a criterion for allocating position. Two of those four groups decide
who will receive the funds only one period in advance, though this still does not guarantee that individuals
will be always able to access funds when they need them.

                            Table 9: Probability of using a negotiated order of allocation

                                                          Use of a negotiated order of allocation
            Explanatory variables:                             (1)                        (2)

            Number of years roscas has                          0.02*                      0.01
            been operating                                     (0.01)                     (0.01)

            Number in rosca                                   -0.01**                    -0.02**
                                                              (0.006)                    (0.008)

            Proportion of rosca members                        -0.08                      -0.13
            with ironroof                                      (0.22)                     (0.14)

            Average years of education of                     -0.09***                    -0.08**
            female members                                     (0.03)                     (0.02)

            Proportion of participants with a                   0.26                       0.24
            salaried income                                    (0.59)                     (0.42)

            Size of the largest ethnic group                  0.85***
            in the rosca12                                    (0.30)

            Indicator variable if more than                                               0.34***
            50% of group members are                                                      (0.10)

            Indicator variables for                              Y                          Y
            geographic divisions
            Indicator variables for division                     Y                          Y
            socioeconomic characteristics
            Number of observations                               70                         70
            Pseudo R-squared                                    0.32                       0.33
            Chi-squared                                        135.5                      528.9

            Significantly different than zero at 90% (*), 95% (**), 99% (***).
            Note: Probit estimation with robust standard errors clustered at the zonal level. All regressions contain an
            indicator variable for the geographic zone.

     The size of the largest ethnic group in the rosca as a percentage of total.

        The data in this sample suggest that this allocation mechanism is more common in more
homogenous roscas that are likely to have better information on each other. Table 9 shows the results of a
probit analysis that characterizes the factors associated with the use of a negotiated order. The use of a
negotiated order is positively associated with ethnic homogeneity in the group, where ethnic homogeneity is
measured by the size of the largest ethnic group in the rosca. This suggests that members of the same ethnic
group either have better information on each other or have lower transactions costs for negotiating the order
of rotation. An alternative specification finds that in groups where more than 50 percent of participants are
related, the use of a negotiated order is again higher. Negotiated order is also associated with smaller group
size, probably because negotiation among larger groups is difficult and time consuming. The use of a
negotiated order is also negatively associated with the level of education of members of the rosca. The use
of a negotiated order is also significantly but weakly negatively associated with the presence of men in the
rosca (not reported). This suggests that in smaller, more homogenous roscas, participants may strive to
provide some limited insurance to members. In cases where participants have better information they are
better able to provide this form of limited insurance, since they are able to determine whether individuals
have in fact received a bad shock. It seems plausible that smaller, more homogenous groups would be more
likely to operate on a norm of balanced reciprocity.

        Exchanging Positions within the Rosca

        Even when the order of allocation is fixed at the start of the round, 88 percent of roscas will allow a
change in the order of allocation, provided the individual can find another person willing to exchange places
with them. In the last round of the rosca, however, only 23 out of 1066 of rosca members (or two percent)
exchanged places with another participant. This number represents 0.1 percent of all 16,808 rosca
transactions that took place in the sample. If individuals are subject to individual shocks and are using roscas
to deal with them, it seems likely that more switches would be observed. In addition, roscas do not make it
easy for an individual to change her position. In three-fourths of the roscas, a participant who wants to
change her position must secure the agreement of another member willing to exchange places, as well as get
approval from the full rosca or the executive committee of the rosca. Several roscas only allow members to
move later in the order. Exchanging places provides a very limited form of insurance for members who
happen to suffer a negative shock before they have received funds. If this were the explicit aim of the roscas,
participants would do better by allocating the pot one period in advance of the meeting.

        State contingent repayments and forgiveness

        Roscas may provide state contingent insurance if participants are allowed to defer payments
temporarily or forgo payments if they suffer a negative shock. There is some evidence that this takes place,
but the time period of forgiveness is probably too limited to provide a high level of insurance. The rules of
most roscas stipulate that missed payments must be repaid on or before the next rosca meeting. In most
roscas there is no penalty or sanction for missing a payment, provided that advance notice has been given to
the group and payment is made by the next meeting.
        Before discussing the potential of default to act as state-contingent insurance, however, it will be
helpful to clarify the forms of default that occur in these roscas. The most serious form of default in a rosca
is when a member receives her share of the funds and then leaves the rosca without making any further
payments. This type of default is rare. In the sample, only 2 of the 1066 individuals in the roscas appeared
to engage in this kind of strategic default. Both were subsequently removed from the group.
        A member may also miss one or more payments before she receives funds. In this case the group
has two options: it can refuse to give that individual funds until she has made all payments, potentially
moving her to the end of the rosca if necessary. Alternatively, many roscas have adopted an internal
mechanism for dealing with cases of default, which they call ―tit for tat.‖ In this mechanism, if member A
does not pay at member B‘s house, then member B is under no obligation to pay when member A‘s turn
arrives. This strategy economizes on the transaction costs of disciplining and sanctioning members and also
avoids the unpleasant task of sanctioning a member who may be having difficulties at home.
        In most roscas, participants are expected to make payments to themselves when they host the rosca.
These payments are recorded in the register and individuals who do not pay themselves are recorded as being
in default. There is little a rosca can do, apart from using pressure, to force an individual to pay herself and
the tit-for-tat strategy is obviously not an option.
        Finally, a rosca participant may also miss a limited number of payments after receiving funds. This
is different from strategic default, when a member receives funds and makes no subsequent payments. In
the case of limited default, the participant makes some, but not all, payments after she has received funds.
Once a participant has received funds, the group has less leverage over her, since the group can no longer
threaten to withhold payments when her turn comes. Given that the roscas are repeated, however,
participants may agree to exercise tit-for-tat in the next round of the rosca. Their only other option is open
pressure, which roscas normally exert either by calling the person in front of the group to explain her
behavior, or by sending a delegation from the rosca to visit the person‘s home.

           Table 10 presents the payment rates for the roscas.13 At the time of the interview, rosca officials
were asked about their most recent completed round (a round being one complete cycle of the rosca in which
everyone has received funds). In some cases the round had only recently finished; default rates might appear
higher in these roscas if individuals who missed payments in the later rounds repaid after the interview.
Table 10 therefore presents missed payments both for the entire round (column 1) and excluding the last
three meetings (column 2). Row 1 indicates that thirteen percent of participants do not make their payments
at any given meeting. Row 5 shows that by the end of the round, 95 percent of members had made all
payments. Late payment is relatively widespread; on average over half of all participants are late with at
least one payment during the round (row 4). Only 9 roscas report perfect payment records. The individuals
who repay their missed obligations, however, are in default for only one month on average. Thus some state-
contingent repayment appears to be taking place, though the insurance element in these repayments cannot
be very large, given that most ―loans‖ last only one month.

     Repayment data for one rosca is incomplete.

                                        Table 10: Missed Payments

                                                            All Meetings   Excluding Last
                                                                            3 Meetings
                                                               Mean            Mean
                                                             (standard       (standard
                                                             deviation)      deviation)
A. Missed payments per meeting
1. Proportion of members at each meeting who miss a             0.13
payment                                                        (0.12)

2. Proportion of participants who never repay                   0.04             0.04
                                                               (0.07)           (0.06)
3. Average number of months members are in default              0.95             0.81
                                                               (1.49)           (1.35)
B. Missed payments per round

4. Average proportion of members that miss at least one         0.51
payment during round                                           (0.34)

5. Proportion of members who had made all payments by           0.95            0.96
the end of round                                               (0.07)          (0.06)

6. For participants who repay, average number of months a       0.99           1.07
payment is in default (N=53)                                   (0.58)          (0.70)

7. Average number of months a person who misses a               5.62            5.79
payment is in default during the round, including those        (7.25)          (6.43)
who never repay (N=60, 58)

C. Default rates of hosts per meeting

8. Proportion of hosts missing at least one payment after       0.07           0.02
receiving the pot, per meeting                                 (0.13)          0.06

9. Of hosts that miss any payment, proportion missing at        0.02           0.006
least 3 payments                                               (0.07)          0.02

10. Of hosts that miss any payment, proportion of hosts         0.04            0.01
who do not repay at least one payment after receiving the      (0.07)          (0.03)

Number of observations                                          69              69

        It is somewhat more difficult to ascertain whether missed payments are forgiven, but the data do
suggest some limited forgiveness of unpaid obligations. In the discussion that follows, cases of default
against oneself and cases of tit-for-tat are excluded. Out of the 1066 rosca participants in the sample, there
were 126 individuals who had outstanding payments at the end of the round. Of these 126 individuals, 40
left the rosca, all but three before they received funds. The average number of payments missed by these
individuals was 7.5. Most roscas generally exercise discretion as to whether or not they return funds to
individuals who leave the round early. In general, if participants have not yet hosted and need to leave for a
good reason, their payments may be refunded. If a participant leaves the rosca after receiving funds, she
must either refund the money received or complete the remaining payments, depending on when in the cycle
she leaves.
        The remaining 86 individuals who defaulted on obligations remained in the rosca for the rest of the
round and missed 2.7 payments on average. All but seven of those individuals received funds in that round;
the average number of payments missed by those who did not receive funds was 12. The remaining 79
individuals received rosca funds. Of the individuals receiving funds, 21 missed their first payment before
hosting and missed an average of 1.7 payments. These individuals were allowed to host, however,
suggesting that their unpaid obligations were either forgiven, or deferred. An additional 58 individuals
missed their first payments after receiving funds and missed an average of two payments. It is not clear what
action, if any, was taken against these two sets of defaulters. It is possible that groups use the tit for tat
mechanism, in which case the creditor member will not pay the defaulting member in the next round of the
rosca. Otherwise, the members may not be invited to join the next round of the rosca, or the missed payments
may be forgiven.
        The data just reviewed suggest that while roscas do provide participants with some state-contingent
payment options, the amount of implicit insurance provided by these options is small. In the last round of the
rosca, if all missed obligations were forgiven, 37 participants would have had their payments refunded (data
is not available on refunds), and 79 would have had their payments forgiven. This is a total of 150 forgiven
payments, which represent 0.8 percent of all payments made during the round, affecting 14 percent of all

        Side payments

        Roscas might also provide insurance if participants are able to make side deals with one another
outside of the rosca. An individual with an early position might lend or grant funds to another member with
a later position or individuals might agree to exchange places outside of the rosca. We have seen already that
the ability to switch positions within the rosca is relatively limited and there is no evidence that any members

exchanged places by exchanging funds outside of the rosca. Most individuals report that they are hesitant to
lend large amounts of money to other members outside of the rosca because they do not feel they will be able
to enforce repayment. There is some evidence, however, that smaller loans are sometimes made between
members. In a sub-sample of 102 rosca members, 20 percent reported that they lent some funds to another
member after hosting the rosca, far less than gave funds to spouses or relatives. The average amount loaned
was also quite low, roughly $3.00 or just a little more than the equivalent of the average rosca contribution.
Many of these loans appear to be given for payment of rosca dues. In 48 percent of roscas, the group reports
that at least one participant gave a loan to another participant to help her pay her obligations; in roscas where
this occurred, on average nearly four members gave such a loan. These loans, however, are not private, since
the entire group was aware they had taken place and they were recorded in the group records. Again, this
appears to be a limited insurance mechanism. Out of the 16,808 rosca transactions that took place among all
the roscas in the sample, participants were loaned funds to make payments in 73 cases, or 0.4 percent of
         The roscas in this sample are clearly not formed to provide explicit insurance. For most of the roscas,
the ability to provide implicit insurance is also quite limited. The order of allocation is determined at the
beginning of the round in almost all of the roscas. The 23 percent of roscas that use a negotiated order may
provide some insurance, to the extent that they allow the order to be determined by need. Exchanging places
and state-contingent repayments appear to provide relatively limited insurance functions, accounting for less
than 1 percent of all payments. Note that many of these implicit insurance mechanisms do not require a rosca
structure to implement, and do not require this particular rosca structure. They cannot account for the
patterns of rosca structure observed in this sample.

Another view: Roscas and community social sanctions

         The sociology and anthropology literature has long stressed the important role of social relations in
sustaining the economic benefit of roscas. If participants are involved in on-going relationships outside of
the rosca, they may sanction non-payers by excluding them from social networks that provide access to
resources or insurance. Moreover, the loss of reputation that may be suffered in such circumstances will
lower the individual‘s ability to participate in other economic and social organizations. Defaulters may
therefore resort to dire actions in order make payments, perhaps borrowing from moneylenders or selling
crucial assets. Consider the following observation: ―a member may go to great lengths, such as stealing or
selling a daughter into prostitution, in order to fulfill his obligations to his association; failure to meet
obligations can even lead to suicide‖ (Ardener, 1964 in Ardener and Burman, 1995). Because of these dire
sanctions, default in roscas is thought to be quite infrequent. Many descriptive accounts of roscas stress the
rarity of default (Burman and Lambete 1995; Srinivasan 1995). One researcher noted that among rosca

participants interviewed, ―The question ‗What if someone does not pay?‘ gets little response, as non-payment
is practically unheard of…‖ (Srinivasan 1995, p. 203). In contrast, many of the roscas in this sample have
rules regarding default, and as the previous section demonstrates, this is not an extremely rare event. Default
rates are still relatively low, given the risky environment, but default –when defined as missing at least one
rosca payment is clearly not ―unthinkable‖ for many participants.
        It is important to understand how roscas employ sanctions to prevent default, since such sanctions
also provide one of the theoretic mechanisms underlying group lending schemes. In group lending, all
group members are held jointly liable for the repayment of individual loans; if an individual‘s loan is not
repaid, then group members may not qualify for additional loans. If peers are able to sanction non-payers to
enforce repayment, then group lending programs can reach participants that formal lending institutions may
not (Ghatak and Guinnane 1999; Besley and Coate 1995; Besley, Coate, and Loury 1993). The specific way
in which social sanctions are used in informal financial organizations, however, has not been thoroughly
investigated and most formal models assume an exogenous sanction cost sufficient to ensure repayment
without examining the specific sanction technology. Research has begun to suggest that social
connectedness may not always improve repayment rates in group lending programs (Morduch 1999; Wydick
1999; Handa and Kirton 1999). This paper suggests that social networks among rosca members are useful not
just as venues for punishment, but also as venues of information that allow participants to provide limited
insurance such as that discussed in the previous section.
        When asked whether someone who has failed to make payments in a round will develop a negative
reputation among community members, rosca participants are at pains to point out that payment issues are
kept secret and the community or administration are involved only in very serious cases of malfeasance:

        There is no way it [missing payments] can affect your life outside the group, as it is a group
        This is a group affair, which is unknown to other people.
        No pressure is exerted outside the group. People are willing to lend her [a person who has
        missed payments] money and she is comfortable in the community socially.

        Rosca participants are often aware of when other participants are suffering difficulties and are
reluctant to impose additional harm on them. As one rosca member put it, ―It is difficult for us to be harsh
with a member if we can see she is having problems at home.‖ This presumes, of course, that rosca
participants are able to distinguish cases of need from cases of strategic default. The close relations among
most rosca members makes this possible.
        What actions are taken by rosca members in the case of an unpaid obligation at the end of the round?
The evidence suggests that participants find it difficult to severely sanction other participants. This is

reflected in the comments of rosca officials about what actions they might take if a member does not pay for
several meetings in a row:

        There is no action we can take [if a member doesn‘t pay], as we don’t want to lose members,
        so we just request the affected member to pay all the dues she owes.
        There is no action we can take, except the member is encouraged to cope with the other
        group activities and contributions.
        The only action we could take is to except the member herself may request to stop the merry-
        go-round [rosca] and continue with other group activities.
        The affected member is summoned to the meeting to express herself; then she is allowed to
        be dormant in the rosca.
        If one misses to contribute three times or so, she will be asked to rest for some time.
        You address the issue and ask them to pay, but if you push too hard, then the person feels
        like leaving or feels bad because the group is harassing them.

        Three quarters of roscas state that they would allow members an extended time to make payments
under certain circumstances. Those circumstances were almost always listed as illness, school fees, and
funerals. In the last round of the rosca, 47 percent of roscas reported that the rosca ended with at least one
outstanding unmet payment. There is only one case in which a group decided to take the individual‘s name
to the local government administration to try to recover funds. Otherwise, the incident was either dealt with
by calling the individual to the group to explain (5 cases), granting an extension (7 cases), using the tit for tat
mechanism (5 cases), or pardoning the person in case of extreme personal difficulties (3 cases). Thus,
while the social exclusion that might follow an outright strategic default appears to be sufficient to mitigate
this kind of extreme opportunistic behavior, there is no evidence that participants are forced to go to extreme
measures to make regular payments. This may be in part because individuals have good information on each
other and can distinguish between true economic hardship and opportunistic behavior. Again, the prevalence
of missed payments should be seen in perspective. While over half the groups had at least one outstanding
payment at the end of the round, all but five percent of payments were eventually made. Out of the 16,808
payments made in the last round of the rosca, only 792 payments were never made. Of those, 181 were
payments in which the group agreed to use tit for tat, and 78 were individuals who failed to pay themselves
during their turn.
        While there is little evidence that groups take strong explicit action against participants who do not
make payments, an individual might be sanctioned implicitly through exclusion from relationships with
participants outside the rosca. The evidence from this sample indicates that this rarely occurs. In a sub-
sample of 102 rosca participants, only 12 percent reported that they would not help a rosca member outside
of the rosca if she had missed payments in the previous round. Typical reactions to this scenario were:

        I would help on a friendly basis with a Christian spirit not really considering whatever
        happened in the group.
        I would help as a friend or neighbor because l might need his/her help one day.
        I would help because need befalls anybody. Today it is me, tomorrow it is somebody else.
        I would help on a friendly basis not considering whatever happened in the merry-go-round

Some members point out that their willingness to help might depend on the legitimacy of the default.
        While roscas may provide participants with consumption-smoothing possibilities through state-
contingent payments, the use of these payments is relatively limited and does not provide an explanation for
why individuals joined roscas in the first place. The evidence from this sample suggests that social relations
clearly underpin these roscas, but not because they provide the opportunity for community-based social
sanctions. Instead, social connectedness implies that participants have the information necessary to
distinguish opportunistic from non-opportunistic behavior. These roscas view themselves as mutual
assistance groups and the ties that hold people together also prevent them from causing harm to each other in
times of difficulty. The reluctance to use social sanctions to ensure strict payment discipline means that
roscas must design other mechanisms internal to the organization to prevent opportunistic behavior and false
claims of need. The main mechanism roscas use is tit for tat, which was used in 26 percent of missed
payments in the last round of the rosca. The next section will argue that repetition of the rosca provides
another mechanism for sustaining payments without the use of severe punishment regimes.

3. An Alternative Hypothesis: Roscas and Commitment

        This paper has argued that intrahousehold conflict and finance for durable purchases cannot fully
explain the structure of roscas in this sample. While roscas may provide some insurance to members, this
insurance is quite limited in scope and cannot explain the specific structure of these institutions. This section
presents an alternative hypothesis to explain rosca participation: roscas are a commitment mechanism
designed by individuals who want to save but lack access to good commitment technologies. Individuals
must therefore construct collective institutions that commit individuals to savings.
        This section proceeds as follows. First the evidence on time inconsistency and commitment
technologies is reviewed. The paper then presents a simple theoretical example of how individuals with
time-inconsistent preferences may enhance their welfare through participation in a rosca. The section
concludes by presenting empirical evidence on the use of commitment mechanisms in roscas.

Time inconsistency and self-control

        How widespread is the problem of self-control? The psychological evidence on preference
formation suggests that self-control problems are widespread. Under the standard assumptions about
economic behavior, individuals have preferences that are constant over time. These preferences are expressed
by an exponential individual utility function in which individuals discount the future at a constant rate. These
preferences imply that when individuals plan a certain behavior in the future, such as saving, when the future
arrives their preferences will not have changed. They will continue to prefer saving over current
consumption implying that their marginal rate of transformation across time periods is constant. Time
inconsistent individuals, however, have preferences that change as the date of decision-making nears. As a
somewhat trivial example, I would prefer a double ice cream cone in ten days to a single cone in nine. But I
most certainly prefer a single ice-cream today to waiting for a double cone tomorrow. Examples of time
inconsistent behavior are quite common. It is a problem as old as Ulysses and the Sirens (Elster 1979), and
as new as the use of illiquid forms of savings such as CD‘s or individual retirement accounts (Laibson 1997).
        Individuals with a time inconsistency problem must find ways of committing themselves into the
future. The falling discount rate of these individuals implies a conflict between today‘s preferences and the
preferences they will hold in the future. The psychology literature presents several mechanisms, both
intrapersonal and extrapersonal, for encouraging commitment. (Ainslie 1986). Intrapersonal mechanisms
include behavior such as removing the temptation from sight or making private rules with oneself governing
behavior. Extrapersonal mechanisms may include physical or social constraints on future choices, such as
opening a 401(k) account. Economists are not well equipped to explain intrapersonal processes,14 and these
processes may be observationally equivalent to other strategies. The empirical emphasis has therefore been
on identifying and modeling visible precommitment mechanisms, or what O‘Donahue and Rabin (1999) call
―smoking guns.‖
        Just how widespread are these commitment devices empirically? Many people join support groups
to lose weight, quit smoking, or as in the present case, to save. As noted above, there is strong evidence that
consumers, even in a market economy with low transactions costs like the U.S., are willing to pay to have
their options limited. Laibson, Repetto, and Tobacman (1998) note that financial advice to consumers often
involves removing money from bank accounts ―before you can spend it.‖ Ten million Americans still belong
to Christmas Clubs, and ―savings circles‖ run by churches are an increasingly popular mechanism for lower-
income individuals to commit themselves to savings.
        The use of commitment technologies is not limited to developed economies. Shipton (1992)
documents the use of ―lock boxes‖ in the Gambia in which individuals hire carpenters to construct a box with

no opening, except a small slit to push money through. Small amounts of money cannot be removed, except
by smashing the box. Rutherford and others (1999) cite several commitment devices that villagers in East
Africa use to stick to savings plans, including buying a lock box and throwing away the key and the use of
―money guards‖ in which individuals entrust their savings to someone else so that they cannot spend it.
Deposit collectors are another way of tying one‘s hands: in many countries individuals actually pay someone
to collect and hold their savings for them (Rutherford 2000; Wright 1999, Robinson 1994). In our fieldwork,
many individuals reported that as soon as they had accumulated savings, they invested in grain, small
livestock, or in a small business, thereby rendering the savings somewhat illiquid. One woman planted sweet
potatoes in her field in September, timing the planting so that they could be harvested just in time (but not
before!) school fees were due in January. The main limitation to all these technologies is that it is difficult to
prevent oneself from reneging on promises. Lock boxes can be smashed and business inventories easily
liquidated. The strongest commitment technologies involve surrendering control over one‘s savings to
someone else. For example, a woman in our sample reported that she used a lock box for saving money in
her house, but gave the key to someone else. When she needed money, she had to go and fetch the person
with the key, explain to her the situation, and convince her to come back with her and open the box. In this
way, she explained, she could keep herself from using the money for petty expenditures and she could ensure
that when she opened the box, she took out only the amount of money she needed. This is similar to the
strategy underlying a rosca. A collective mechanism is required because a time-inconsistent individual
cannot make deals with herself. Future selves (with different preferences) will always be able to renegotiate
this contract. The usefulness of a rosca is that funds paid in cannot be retrieved, and control over the funds is
essentially given over to the rosca.

An example of a simple rosca with time inconsistent individuals

           To show how roscas might be a collective response to individual self-control problems, this section
presents a simple example of a rosca formed by individuals with time-inconsistent preferences. A common
approach in the literature to capture time inconsistent behavior is to model the choice between present and
future decisions as a bargaining game among sequential selves in which current selves control current
behavior and future selves control future behavior (O‘Donahue and Rabin 1999; Laibson 1997; Loewenstein
and Prelec 1992; Thaler and Shefrin 1981; Phelps and Pollack 1968). In order to constrain future behavior,
therefore, a person must be sophisticated, that is, they must be aware of their self-control problem
(O‘Donahue and Rabin, 1999). A sophisticated but time inconsistent individual who desires to save then
faces the task of constructing a mechanism to constrain her current self from consuming instead of saving.

     Although O‘Donahue and Rabin (1999) show that if individuals are aware of their self-control problem, they can

This mechanism could be individual, as in the case of an illiquid asset, or joint, as in the case of joining a
         The standard exponential model of individual behavior can be characterized by a simple
maximization problem:
                 1   U t (u t , u t 1 ,...uT )     uT
                                                        t

         Utility is the sum of utility in all periods up to T, discounted by the factor .  lies between 0 and 1
and represents the time consistent exponential discount factor. A common approach to modeling hyperbolic
preferences is to use what Laibson (1997) calls ―quasi-hyperbolic‖ utility. The utility maximization problem
with quasi-hyperbolic utility is given by:

                   [2] U t   t u t  
                                                 u
                                                    t 1

         In this case,  represents standard exponential utility, while  represents time inconsistent
discounting. When  <1, a person gives more weight to period  when it arrives than she gave it in any
period prior to period . In this case, the short-term discount rate is higher than the long-term discount rate.
Following Laibson (1997) and O‘Donahue and Rabin (1999) I assume  =1 to abstract away from time
consistent discounting, leaving only , the time inconsistent parameter, so that utility is simplified to:
                  [3] U t  u t        
                                          t 1

         The basic results follow with <1.
         To see why roscas can help overcome time inconsistency, imagine a simple game in which two
agents, A and B, wish to save in order to provide for consumption at a future date. The game lasts four
periods, 1 to 4. In periods 2 and 3 (for example, the ―harvest periods‖) each individual receives one unit of
surplus income. In each period, the individual must decide whether to save or consume this surplus.
Consuming the surplus gives each individual one unit of utility. If she saves both units of surplus, then in
period 4 she may trade these in for a durable good, or enjoy higher consumption in the non-harvest month of
period 4. The individuals have no access to credit markets.
         If A and B have standard exponential utility as given in [1], assuming that  =1 and  =1, then the
benefit of consuming their two periods of surplus (with  =1) is equal to two, the sum of the two periods of

create commitment mechanisms for themselves as long as they can correctly anticipate their future lack of self-control.

surplus. If they save, they can buy a good in the fourth period that gives a one-period benefit of c. A and B
will save in periods two and three as long as c >2. We assume in future discussion that c always greater than
two, implying that individuals will always want to save.
        Under time-inconsistent preferences, however, the decision looks quite different once period two
arrives. Assume A and B have quasi-hyperbolic utility as in equation [3], again with  =1. But now  <1, so
that both individuals have a preference for current over future consumption. Viewed from period one, the
benefit of consuming the surplus in periods two and three for both A and B is given by:  [1 + 1]= 2. If
each individual saves his two periods of surplus, the future benefit of saving for the period one self is  c.
Thus A and B will save if  c > 2 or c >2, just as in the time consistent case. When period two arrives,
however, the picture has changed. Self two is now in control. The benefit of consuming the surplus in period
three is discounted by , so the current valuation of benefits is 1+ . The benefit of saving is  c. A and B
will want to save if c > (1+)/. This expression is greater than 2 for all  <1. The smaller the discount
parameter  (i.e. the greater the time inconsistency), the greater the benefit needed to induce savings, with
the result that benefits that would induce savings in the time consistent case, will be insufficient to induce
savings in a hyperbolic consumer. A and B are unable to improve their welfare because they are unable to
credibly commit themselves to savings for all c <(1+)/. Thus there is a range of values of c greater than
two for which the hyperbolic consumer will not save.
        If A and B are aware of their self-control problem, they may want to find some mechanism for
overcoming their time inconsistency. Suppose they decide to form a rosca. The rosca works as follows. In
period one, A and B both wish to save in periods two and three as long as c is greater than 2. Assuming this
holds, they agree to meet in period two and to flip a coin. The loser of the coin toss will give her savings to
the winner, who will do the same in period three. Viewed from date 1, the utility from joining a rosca for
both players (assuming =1) is  c. The benefit of consuming in periods two and three is also as in case two:
2. Both A and B will be better off forming a rosca as long as c >2, just as before.
        Suppose A wins the coin toss in period two and is therefore to receive funds first. B now views her
own benefits as follows: The benefit from joining the rosca and receiving the good in the third period is 0 +
c. If B does not join the rosca, she receives 1 + . B will therefore join as long as:
                                  1 
          c  1     or    c

        Again, the lower is  (i.e. the higher the amount of time inconsistency), the higher the c required to
induce B to join. If A joins the rosca she gets c in period three so her benefits viewed from period two are 
c. If she does not join she receives 1 + . Her condition for entry is therefore the same as B‘s. If, however,
A enters and defaults, she receives 0 in the first period, but in the second period receives the benefits from

her good plus the one unit of surplus she does not give B, discounted by the time-inconsistent parameter, :
0   (c  1)   c   .
        Because c +  > c, A will join the rosca and then default. Seeing this, B will never join and a
rosca is impossible. While both would be better off if they could commit to saving through rosca, neither is
able to credibly commit to the rosca. This is true no matter what allocation method is used.
        The potential for roscas to provide savings discipline might therefore seem limited, but the folk
theorem provides a mechanism for sustaining roscas: cooperation is possible in repeated games so long as
players care sufficiently about the future. If A and B are sufficiently long-lived, they can create an
endogenous mechanism to sustain the rosca, by offering the opportunity to repeat the rosca once a round is
complete. While no rosca member expects to live forever, or, indeed, for her rosca to last forever, the infinite
horizon framework captures the situation where players are analyzing a long-term situation without assigning
a specific status to the end of the world (Osborne and Rubenstein, 1994).
        In n-person repeated prisoner‘s dilemmas, cooperation can be maintained by a ―trigger strategy‖ in
which a player‘s reaction function tells him to cooperate in this round if and only if all players cooperated on
the previous round, otherwise defect forever (Fudenberg and Maskin 1986). In this example, the trigger
strategy would be played if a player receives funds and then makes no further payments. This happened with
two individuals in this rosca, who were immediately excluded from future participation in the rosca. If A
defaults, for example, the extra benefit from defaulting is equal to the extra period of consumption received,
. The cost of default is exclusion from participation in future rounds of the rosca, with utility loss  c in
each round. Assuming no discounting within rounds, but a discount parameter of  across rounds, the rosca
can be sustained as long as
               or when  > 1 – c.

        As c increases, the rosca is sustained for more impatient individuals. The trigger strategy is not the
only strategy that might sustain roscas in the context of infinite play. Nonetheless, it is clear that the
cooperation necessary to sustain a rosca is possible within a range of punishment strategies and this helps to
explain why individuals are able to sustain roscas even in the absence of strong community-based social

Empirical Evidence

        Self-reported reasons for joining a rosca

        One of the most frequent reasons rosca participants give for joining a rosca is self-commitment.
Consider the following responses to the question of why individuals joined a rosca:

        ―You can’t save alone—it is easy to misuse money.‖
        ―Saving money at home can make you extravagant in using it.‖
        ―Sitting with other members helps you to save.‖
        ―It [a rosca] makes you look harder for money to save.‖
        ―It is difficult to keep money at home as demands are high.‖

        Table 11 presents the responses of 308 rosca members as to why they joined a rosca. The question
was first phrased in an open-ended manner: ―What is the most important reason you joined this rosca?‖ 37
percent of respondents reported that it was difficult to save at home because money got used up in small
household needs. 22 percent reported that it was difficult to save alone, that they ―got the strength to save‖
by sitting with others. Only 10 percent of individuals reported that they joined a rosca as a response to
household conflict, fear of theft, or demands by kin. The question was subsequently re-phrased to ask the
respondent to choose the most important reason from among seven choices. Table 11 shows the seven
choices given and the responses. The most frequently chosen response was that roscas gave individuals the
―strength‖ to save. Again, only eight percent of respondents gave fear of theft or demands of husbands or
families as reasons for joining a rosca. Individuals may have been uncomfortable talking about household
circumstances to enumerators, but the overwhelming number of individuals reported difficulties in self-
control rather than family or household control issues. This is also supported by many field studies of
savings in developing countries, which indicate that many individuals lack good savings commitment
technologies (Rutherford and others 1999; Wright 1999; Robinson 1992; Shipton 1992).

                       Table 11: Self-reported Reasons for Joining a Rosca

Here are 7 reasons why you might belong to a         Number of       Percentage
rosca, which is most important to you               respondents

Get strength to save from sitting with                 105              35.5
others/can‘t save alone
Group wanted to make sure that each member              74              25.0
had a certain item
Can‘t save at home, money gets used on other            51              17.2
To visit each other‘s houses and see how the            39              13.2
person is living
Safer place to save/ fear of theft                      11              3.7
Can‘t save at home, my family will use                  6               2.03
Can‘t save at home, my husband will use                 6               2.03
Other                                                   5               1.7

Number of observations                                 303

What is the most important reason why you            Number of       Percentage
joined a rosca? (open-ended)                        respondents

Can‘t save at home, money gets used on other           109              36.9
Get strength to save from sitting with                  65              21.9
others/can‘t save alone
Group wanted to make sure that each member              50              16.9
had a certain item
Can‘t save at home, my family will use                  18              6.1
Can get a lump sum to buy a big item                    17              5.76
Safer place to save/ fear of theft                      8               2.7
Can‘t save at home, my husband will use                 3               1.02
To visit each other‘s houses and see how the            3               1.02
person is living
Other                                                   22             12.53

Number of observations                                 303

        Evidence from rosca structure

        This sub-section explores the empirical implications of the hypothesis that roscas are formed for
their commitment technologies. A perfect test of the hypothesis would be able to demonstrate that
individuals have hyperbolic preferences. In practice, this is virtually impossible. While experiments have
indicated that individuals preferences are roughly hyperbolic (Ainslie, 1992), the economic implications of
these preferences may often be indistinguishable from an exponential economy (Laibson, 1996). Hyperbolic
consumers are distinguished from exponential ones by their extreme short-run impatience, combined with a
willingness to get ―locked in‖ to long term savings strategies. Evidence suggests that delays of as little as
one week are sufficient to deter hyperbolic consumers from diverting from their planned consumption and
savings strategy (Laibson, Repetto, and Tobacman 1998).
        The commitment hypothesis is consistent with several empirical regularities already discussed. First,
roscas are often repeated with the same order of allocation, suggesting that individuals do not mind getting
locked in to savings, even when they cannot reclaim their savings for long periods. In roscas that randomize
the order of allocation each round, it is possible for an individual to be allocated the first position in one
round, and the last position in the next. In a rosca of 15 individuals, that person must then wait 30 periods
before receiving funds. Moreover, individuals who participate in a rosca with a repeated order of allocation
or who have unlucky draws in random roscas do not realize any intertemporal gains by receiving funds
sooner than they would by saving alone. This suggests that individuals value the commitment mechanism
that roscas provide rather that its value in speeding the accumulation of funds. Second, roscas share many
features with defined contribution plans in more developed economies. Once a participant has entered the
program, she has committed to making set payments at set intervals. Weekly or monthly meetings with
one‘s peers help to ensure payment discipline in a setting where devices like automatic deductions are not
available. In addition, funds once deposited are not easily refundable. They are given immediately to
another participant and spent. While individuals who leave the group may be able to reclaim their savings,
they lose the opportunity to benefit in future rounds.
        Several additional rosca characteristics support the commitment hypothesis. All roscas monitor
participants‘ payments through the public paying in of deposits at each meeting. Surprisingly, roscas also
monitor participants‘ payments to themselves. Meetings are held at the home of the recipient of funds, who
must also contribute to the pot in front of her peers on that day. Her name is duly recorded in the register and
her funds returned to her along with the contributions of the other members. If she does not contribute,
however, she is considered in default. Her name is recorded as a defaulter and the group follows up with her
until she pays this debt to herself at a future meeting. This is an extreme form of self-commitment.

        One of the strongest indications that the motivation for roscas is savings is the widespread use of a
precommitment mechanism by rosca groups. Fifty-nine percent of the roscas in the sample use such a
spending agreement. In these agreements, rosca participants commit to the use of some or all of their funds
before they receive them. Table 12 shows the use of these agreements among roscas. Of the roscas that
agree on the use of funds, 15 percent agree at the start of the round that all members will receive the same
item. In 17 percent of roscas, the member tells the group what she wants and the group brings the item or
items to the rosca meeting. In 34 percent of roscas the members either buy the item for the host and bring it
to her house or accompany her to purchase the item. In groups that do not purchase the item for the member,
the group also monitors the member to make sure she has used the funds as agreed, either by visiting the
person‘s home or by asking her to make a verbal report about the use of funds in the rosca.

                                       Table 12: Spending Agreements

       How do members precommit?                    Number of roscas        Percentage of all

       Group agrees at the start that each                   6                   14.63
       member should receive an item
       Member selects item, group brings                     7                   17.07

       Member buys, group follows up with                    14                  34.15
       home visit or report
       Go with the member to buy                             14                  34.15

       Total                                                 41                  58.57
       No agreement                                          29

        If the sole purpose of joining a rosca is the expectation of receiving funds earlier than by saving
alone, it is difficult to see why this form of inflexibility would be necessary. Participants are explicitly tying
their hands by not only committing to savings, but by committing to spending as well. Participants report
that they do this to avoid spending sprees once they have received funds. Table 13 shows the characteristics
associated with the use of a spending precommitment in roscas. The use of such an agreement is negatively
associated with distance from a paved road, suggesting that more rural groups are unlikely to use this
strategy. This is plausible since individuals living in more developed areas probably have more temptations
for spending sprees and therefore a higher need to bind their hands. The use of a spending precommitment is
positively associated with the proportion of participants with an iron roof and the proportion of members
with a salaried income. The use of a spending precommitment is negatively associated with average levels
of education, but positively associated with ethnic homogeneity in the rosca, suggesting that it may be easier
to monitor individuals from one‘s own ethnic group. Age dispersion in the group, measured as the

probability that two members are drawn from different age groups, is negatively associated with the use of a
spending commitment, indicating again that homogeneity is positively associated with commitment
                             Table 13: Probability of Rosca using a Spending Agreement

                                                              Probability of Using a Spending Agreement
       Explanatory Variables:                                     (1)             (2)            (3)

       Total number of rounds rosca has                          -0.01               -0.01           -.01
       completed                                                 (0.02)              (0.02)         (0.02)

       Number of rosca participants                              -0.01               -0.02         -0.001
                                                                 (0.01)              (0.01)        (0.009)

       Distance of rosca from paved road, in                     -0.02**             -0.02**        -0.01
       kilometer                                                 (0.01)              (0.01)        (0.009)

       Average years of education of members                     -0.07**             -0.07**       -0.08**
                                                                 (0.04)              (0.04)        (0.03)

       Proportion rosca participants with ironroof                0.50*              0.64**          0.54*
                                                                 (0.26)              (0.26)         (0.30)

       Proportion of rosca participants with latrine             -0.20               -0.33          -0.45
                                                                 (0.77)              (0.75)         (0.73)

       Proportion of members with salaried                       2.12***             2.21***       2.09***
       income                                                    (0.66)              (0.67)        (0.51)

       Age dispersion in group                                   -1.09**             -1.05***      -1.01**
                                                                 (0.45)               (0.39)       (0.47)

       Size of the largest ethnic group in the                                       1.40***       1.54***
       rosca15                                                                       (0.52)        (0.51)

       Number of uses of rosca funds                                                               -0.21**
       Indicator variables for geographic regions                    Y                  Y             Y
       Indicator variables for zonal socio-                          Y                  Y             Y
       economic characteristics
       Number of observations                                      70                  70             68
       chi-squared                                               152.93              320.63         94.98
       Pseudo R-squared                                           0.27                0.31           0.26
       Significantly different than zero at 90% (*), 95% (**), 99% (***).
       Note: Dprobit estimation with robust standard errors. All regressions contain an indicator variable for the
       geographic zone.

     Size of the largest ethnic group as a percentage of total rosca participants.

4. Conclusions

        Roscas are among the most prevalent savings institutions in the world. In spite of the prevalence and
importance of roscas, however, there has been little systematic micro-level evidence collected on these
organizations. This paper has addressed this gap by providing a detailed dataset that examines in detail the
structure and functioning of these organizations and finds that existing theories of roscas cannot fully explain
their complex institutional structure. This paper has argued that a key feature of roscas is their ability to
harness collective action to provide individual commitment to savings. One of the most remarkable features
of these institutions is the way in which they combine a commitment to savings with the flexibility to adapt
to local circumstances. These groups operate in areas that are extremely institution-poor, both politically and
economically. Markets and institutions providing credit, insurance, and saving are extremely limited in
western Kenya. The empirical literature has documented the thinness of many markets in rural Africa, as
well as the absence of more complicated inter-linked contracts for credit and insurance (Aryeteey 1996;
Collier 1983). Roscas may be responding to these failures by evolving into fairly complicated institutions
with manageable transaction costs that solve multiple market failures.
        The data show that roscas are not organized primarily to provide credit for the purchase of durable
goods. If that were the case, many rosca participants could do better by saving alone, since they do not
expect to receive the good sooner by joining the rosca group. In addition, most participants use their funds
for multiple purchases. Roscas also do not provide a good solution to intra-household conflict over
spending, since the identity of the recipient of funds is not secret and contributions and rewards are shared
among household members. But roscas undoubtedly provide participants with the opportunity for safe
storage of funds and by keeping funds safe from one‘s own splurges, roscas may also help to keep funds safe
from the demands of others.
        Rosca flexibility is perhaps most apparent in the organizational response to payment issues. Roscas
appear to provide some limited insurance to members in the form of state-contingent payments and limited
debt forgiveness. The key to combining limited flexibility with commitment in these roscas is good
information. Where rosca participants are able to observe negative shocks, they are able to provide some
insurance (in the form of consumption smoothing) against those shocks. Thus this paper suggests that
successful collective action in these organizations depends on the social connectedness of participants. The
social connectedness of roscas means that participants have the information necessary to distinguish
opportunistic from non-opportunistic behavior. This allows roscas to provide some state-contingent
insurance while maintaining the ability to recognize and punish opportunistic deviations and enforce savings
commitment. More homogenous, closely connected roscas are more likely to provide insurance
opportunities through the use of a negotiated order of allocation.

          The paper also shows that collective action in roscas plays a role that has been largely ignored in the
literature: roscas provide a collective mechanism for individual self-control in the presence of time-
inconsistent preferences and in the absence of alternative commitment technologies. As rosca participants
would say, ―you can‘t save alone.‖ Roscas enforce personal savings through close monitoring of
contributions, even to oneself. Many roscas rosca participants also ―bind their hands‖ through the use of a
precommitment mechanism in which they agree in advance on how they will use their funds. The rosca
group monitors the individual to ensure she honors her commitment.
          This work has a number of implications for theory and policy. The paper suggests that self-control
problems are widespread and that savings commitment mechanisms are desired and appreciated by the poor.
Providing the poor with the opportunity to commit to savings could have high payoffs in terms of savings
mobilization and asset accumulation in many economies. The research also provides evidence that social
sanctions may not always be effective in promoting repayment in microfinance schemes when participants
have good information and are reluctant to punish their peers. Instead, this research suggests that dynamic
incentives and regular, peer-monitored payment regimes may be more effective in creating sustainable
finance for the poor. Finally, the paper also indicates that local-level institutions are quite creative in their
response to local conditions. This implies that organizations seeking to transplant successful models of
informal finance to new settings would do well to consider the impact of local conditions on organizational

Adams, Dale W. and Delbert Fitchett. 1992. Informal Finance in Low-Income Countries. Boulder:
       Westview Press.

Adams Dale W. and Marie L. Canavesi. 1992. ―Rotating Savings and Credit Associations in Bolivia.‖ in
      Adams and Fitchett, Informal Finance in Low-Income Countries.

Ainslie, George. 1992. Picoeconomics. Cambridge: Cambridge University Press.

Anderson, Siwan and Jean-Marie Baland. 1999. ―The Economics of Roscas and Intra-household Resource
       Allocation,‖ mimeo.

Ardener, Shirley and Sandra Burman, eds. 1995. ―Money-go-Rounds: The Importance of rotating savings
       and credit associations for women.‖ Washington, D.C. BERG.

Ardener, Shirley. 1964. ―Comparative Study of Rotating Credit Associations.‖ Reprinted in Ardener and
       Burman, 1995.

Aryeetey, Ernest. 1996. ―Rural Finance in Africa: Institutional Development and Access for the Poor.‖ in
       Bruno, Michael and Boris Pleskovic, editors. World Bank Annual Conference on Development
       Economics. Washington, DC: The World Bank.

Aryeetey, Ernest and Fritz Gockel. 1991. ―Mobilizing Domestic Resources for Capital Formation in Ghana:
       The Role of Informal Financial Sectors.‖ African Economic Research Consortium, Nairobi:
       Initiatives Publishers.

Banerjee, Abhijit V., Timothy Besley, and Timothy Guinanne. 1994. ―They Neighbor‘s Keeper: The Design
       of a Credit Cooperative with a Theory and a Test.‖ Quarterly Journal of Economics¸ 109(2): 491-

Berenbach, Shari and Diego Guzman. 1994. ―The Solidarity Group Experience Worldwide.‖ in Otero and
       Rhyne, eds. The New World of Microenterprise Finance. West Hartford: Kumarian Press.

Besley, Timothy, Stephen Coate, and Glen Loury. 1994. ―Rotating Savings and Credit Markets, and
        Efficiency.‖ Review of Economic Studies. 61(1): 701-719.

______________. ―The Economics of Rotating Savings and Credit Associations.‖ American Economic
       Review. 83 (4): 792-810.

Besley, Timothy and Stephen Coate. 1995. ―Group Lending, Repayment Incentives, and Social Collateral.‖
        Journal of Development Economics. 46: .

Bortei-Doku, Ellen and Ernest Aryeetey. 1995. ―Mobilizing Cash for Business: Women in Rotating Susu
        Clubs in Ghana. In Ardener, Shirley and Sandra Burman, eds, 1995. Money-go-Rounds: The
        Importance of rotating savings and credit associations for women. Washington, D.C. BERG.

Bouman, F.J. 1995. ―Rotating Savings and Credit Organizations: A Development Perspective.‖ World
      Development. 23(3): 37-384.

Burman, Sandra and Nozipho Lembete. 1995. ―Building New Realities: African Women and Roscas in
      Urban South Africa.‖ in Ardener, Shirley and Sandra Burman, eds, 1995. Money-go-Rounds: The
      Importance of rotating savings and credit associations for women. Washington, D.C. BERG.

Callier, Phillippe. 1990. ―Informal Finance: The Rotating Savings and Credit Association—An
         Interpretation.‖ Kyklos. 43(2) 273-276.

Calomiris, Charles and Indira Rajaraman. 1998. ―The Role of ROSCAs: Lumpy Durables or Event
       Insurance?‖ Journal of Development Economics. 56(1): 207-16.

Carillo, Juan and Mathias Dewatripont. 1999. ―Promises, promises…‖ mimeo, ECARE, Universite Libre de

Collier, Paul. 1983. ―Malfunctioning of African Factor Markets: Theory and a Kenyan Example.‖ Bulletin
         of Economics and Statistics. (2).

Cosgrave, Lloyd, M. 1924. ―Christmas Clubs.‖ Quarterly Journal of Economics. 41(4): 732-739.

Deaton, Angus. 1989. ―Savings in Developing Countries: Theory and Review.‖ Proceedings of the World
       Bank, Annual Conference on Development Economics.‘ Washington, DC.

Dzingirai, V. 2000. ―Saving to death: A study of group based and other saving arrangements in rural Chivi
       District, Zimbabwe.‖ FAO, Social Development Dimensions Report.

Elster, Jon. 2000. Ulysses Unbound. Cambridge: Cambridge University Press.

Elster, Jon. ed. 1986. The Multiple Self. Cambridge: Cambridge University Press.

Fudenberg, Drew and Eric Maskin, 1986. ―The Folk Theorem in Repeated Games with Discounting and
       Incomplete Information.‖ Econometrica. 54, pp. 533-54.

Geertz, Clifford. 1962. ―The Rotating Credit Association: A Middle Rung in Development.‖ Economic
        Development and Cultural Change. 10(3).

Gugerty, Mary Kay and Michael Kremer. forthcoming. ―The Impact of Development Assistance on Social
       Capital: Evidence from Kenya.‖ Iin Grooteart, Christiann and Thierry van Bastelaer, eds, The Role of
       Social Capital in Development: An Empirical Assessment. Cambridge University Press.

Gugerty, Mary Kay and Michael Kremer. 2000. ―Outside Funding of Community Organizations: Benefiting
       or Displacing the Poor?‖ National Bureau of Economic Research Working Paper 7896.

Grossman, Richard S. 1992. ―Deposit Insurance, Regulation, Moral Hazard in the Thrift Industry: Evidence
      from the 1930‘s.‖ American Economic Review. 82(4): 800-821.

Handa, Sudhansu, and Claremont Kirton. 1999. ―The Economics of Rotating Savings and Credit
       Associations: Evidence from the Jamaican ‗Partner.‘‖ Journal of Development Economics. 60(1):

Hollis, Aidan and Arthur Sweetman. 1998. "Microcredit: What Can We Learn From the Past?" World
        Development. 26(10):1875-1891.

Kenya Bureau of Statistics, 1996. Welfare Monitoring Survey, 1994. Nairobi: Government Printers Office.

Khwaja, Asim Ijaz. 2000. ―Can Good Projects Succeed in Bad Communities? Collective Action in the
       Himalayas.‖ Mimeo, Harvard University.

Kimuyu, Peter Kiko. 1999. ―Rotating Savings and Credit Organizations in Rural East Africa,‖ World
      Development. 27(7): 1299-1308.

Klonner, Stefan. 2000. ―Rotating Savings and Credit Organizations as Insurance.‖ mimeo, University of

Kovsted, Jens and Peter Lyk-Jensen. 1999. ―Rotating Savings and Credit Organizations: The Choice
       Between Random and Bidding Allocation of Funds.‖ Journal of Development Economics. 60(1):

Laibson, David, Andrea Repetto, and Jeremy Tobacman. 1998. ―Self-Control and Saving for Retirement.‖
       Brookings Papers on Economic Activity, 1: 91-196.

Laibson, David. 1997. ―Golden Eggs and Hyperbolic Discounting.‖ The Quarterly Journal of Economics.
       May 1997.

Laibson, David. 1996. ―Hyperbolic Discount Functions, Undersaving, and Savings Policy.‖ NBER Working
       Paper 5635, June 1996.

Levenson, Alec and Timothy Besley. 1996. ―The Anatomy of an Informal Financial Market: Rosca
       Participation in Taiwan.‖ Journal of Development Economics, Volume 51.

Lowenstein George and Richard H. Thaler. 1989. ―Anomalies: Intertemporal Choice.‖ Journal of Economic
      Perspectives. 3(4):181-193.

Lowenstein, George and Drazen Prelec. 1992. ―Anomalies in Intertemporal Choice: Evidence and an
      Interpretation.‖ Quarterly Journal of Economics. 107(2): 573-597.

Lowenstein George and Jon Elster, 1992. Choice over Time. New York: Russell Sage Foundation.

Morduch, Jonathan. 1999. ―The Microenterprise Promise.‖ Journal of Economic Literature. 37(4): 1569-

Mugwanga, E.H.A. 1999. ―Use and Impact of Savings Services for Poor People in Kenya.‖ Background
     paper for Rutherford, Stuart et al, ―Savings and the Poor,‖ 1999.

Mutesasira, Leonard. 1999. ―Savings and Needs: An Infinite Variety.‖ Mimeo, MicroSave Africa, Kampala,

Nelson, Nici. 1995. ―The Kiambu Group: A Successful Women‘s Rosca in MathareValley, Nairobi, 1971 to
        1990. in Shirley and Sandra Burman, eds, 1995. Money-go-Rounds: The Importance of rotating
        savings and credit associations for women. Washington, D.C. BERG.

Niger-Thomas, Margaret. 1995. ―Women‘s Access to and Control of Credit in Cameroon: The Mamfe
       Case.‖ in Ardener, Shirley and Sandra Burman, eds, 1995. Money-go-Rounds: The Importance of
       rotating savings and credit associations for women. Washington, D.C. BERG.

O‘Donahue, Ted and Matthew Rabin. 1999. ―Incentives for Procrastinators.‖ The Quarterly Journal of
      Economics. 114(3): 769-817.

O‘Donahue, Ted and Matthew Rabin. 1999. ―Doing it Now or Doing it Later.‖ The American Economic
      Review. 89(1): 103-121.

Osborn, Martin and Ariel Rubinstein. 1994. A Course in Game Theory. Cambridge: MIT Press.

Ostrom, Elinor. 1990. Governing the Commons. New York: Cambridge University Press.

Otero, Maria and Elisabeth Rhyne, eds. 1994. The New World of Microenterprise Finance. West Hartford,
        CT: Kumarian Press.

Phelps Edmund S. and Robert A.Pollack. 1968. ―On Second-best National Saving and Game-equilibrium
        Growth.‖ Review of Economic Studies. 35(2): 185-199.

Platteau, Jean-Phillippe. 2000. Institutions, Social Norms, and Economic Development. Harwood Academic

Platteau, Jean-Phillippe. 1997. ―Mutual Insurance as an Elusive Concept in Traditional Rural Communities.‖
        Journal of Development Studies. 33(6): 764-796.

Robinson, Marguerite. 1992. ―Savings Mobilization and Microenterprise Finance: The Indonesia
       Experience.‖ in Otero, Maria and Elisabeth Rhyne, eds, The New World of Microenterprise Finance.
       West Hartford, CT: Kumarian Press.

Rutherford, Stuart. 2000. ―The Poor and their Money: An essay about financial services for poor people.‖
       Mimeo, Institute for International Development Policy and Management, University of Manchester.

Rutherford, Stuart, and others. 1999. ―Savings and the Poor: the Methods, Use and Impact of Savings by the
       Poor of East Africa.‖ Report prepared for MicroSave-Africa, May 1999.

Shapiro, Matthew D. and Joel Slemrod. 1995. ―Consumer Response to the Timing of Income: Evidence
        from a Change in Tax Withholding. American Economic Review. 85(1): 274-283.

Shipton, Parker. 1992. ―The Rope and the Box: Group Savings in the Gambia.‖ In Adams and Fitchett,
        Informal Finance in Low-Income Countries.

Siebel, Hans Dieter and Michael T. Marx.1987. Dual Financial Markets in Africa Case Studies of Linkages
        between Informal and Formal Financial Institutions. Fort Lauderdale: Breitenbach Publishers.

Smeeding, Timothy, Karen Ross Phillips, and Michael O‘Connor. 2000. ―The EITC: Expectation,
       Knowledge, Use, and Economic and Social Mobility. Mimeo, Center for Policy Research, Syracuse

Stegmann, Michael. 1999. Savings and the Poor. Washington, DC: The Brookings Institution.

Stiglitz, Joseph. 1990. ―Peer Monitoring and Credit Markets.‖ World Bank Economic Review. 4(3): 351-66.

Thaler, Richard.1992. The Winner’s Curse: Paradoxes and Anomalies of Economic Life. Princeton:
        Princeton University Press.

Thaler, Richard H. 1990. ―Anomalies: Saving, Fungibility and Mental Accounts.‖ Journal of Economic
        Perspectives. 4(1): 193-205.

Thomas, Duncan. 1990. ―Intra-household resource allocation: an inferential approach.‖ Journal of Human
      Resources. 25(4):635-64.

Townsend, Robert. 1995. ―Consumption Insurance: An Evaluation of Risk-Bearing Systems in Low-Income
      Economies.‖ Journal of Economic Perspectives. Volume 9(3): 83-102.

Tripoli, Steve. 2000. ―Financial Help from the Church?‖ NPR Broadcast, Morning Edition, September 22,

Udry, Chris. 1990. ―Credit Markets in Northern Nigeria: Credit as Insurance in a Rural Economy.‖ The
       World Bank Economic Review. 4(3): 251-69.

Van der Brink, Rogier and Jean-Paul Chavas. 1997. ―The Microeconomics of an Indigenous African
       Institution: The Rotating Savings and Credit Association.‖ Economic Development and Cultural
       Change. 45(11): 745-72.

Varian, Hal 1990. ―Monitoring Agents with Other Agents.‖ Journal of Institutional and Theoretical
        Economics. 146: 153-74.

Von Pischke, J.D., Dale Adams, and Gordon Donald. 1992. Rural Financial Markets in Developing
       Countries: Their Use and Abuse. Baltimore: Johns Hopkins University Press.

Von Pischke, J.D. 1992. ―Roscas: State-of-the-Art Financial Intermediation‖ In Adams and Fitchett,
       Informal Finance in Low-Income Countries. Boulder: Westview Press.

Woolcock, Michael.1997. ―Induced Development and the Dilemmas of Social Sustainability: Lessons form
      a Comparative Study of Group-Based Financial Institutions.‖ Paper presented to the 9th International
      Conference on Socio-economics, Montreal, Canada.

Wright, Graham. 1999. ―A Critical Review of Savings Services in Africa and Elsewhere.‖ Mimeo, Micro-
        save Africa, Kampala, Uganda. (

Wydick, W. Bruce. 1999. ―Can Social Cohesion be Harnessed to Repair Market Failures?
      Evidence from Group Lending in Guatemala,‖ Economic Journal. 109(457): 463-7.


Shared By:
zhangyun zhangyun http://