Referee report 2 Martijn Boermans, January 8, 2011 Dear editor, Hereby I send you my recommendation for the paper “Screening by the Company You Keep” received on November 4, 2010. The author(s) meet your requirements with some “Major Revision”. I stipulate my argumentation below. --The paper provides a new theoretical framework to analyse how joint-liability lending can create information to lenders about borrowers to overcome adverse selection problems in credit markets. Under such contractual mechanism borrowers will try to select their peers based on local information, because the payoffs depend on the number of repayments made by group members. The question is how lenders, especially those of micro finance institutions (MFIs), can utilize this sorting pattern in order to provide more credits. The core result is that joint liability induces ‘positive assortative matching’ where “safe borrowers will end up with safe borrowers as partners, and risky borrowers with risky partners.” (p.602). In this manner the research adds a novel channel to address lending inefficiencies with endogenous group formation. The author(s) set up a simple one-period model of a credit market under adverse selection (see p.604). There is a risk neutral bank and there are safe and risky borrowers who differ in the probability of an investment project’s success (and possibly in the first and second moment of the project’s return). In the case of asymmetric information the bank may charge an interest rate that is higher than the safe borrowers are willing to pay, such that there will be underinvestment a la Akerlof’s lemon market. This is not new. Under joint liability lending, they build on the work of Besley and Coate (1995) where a borrower pays an individual ‘interest rate plus debt repayment’ (r) and is jointly liable/responsible for the group repayment (c). As such if a borrower’s project is successful r will be repaid apart from additional payments of c per group member whose project failed. A crucial assumption is that borrowers know each other’s types, but that this is unknown to the lenders. Following Stiglitz and Weiss (1981) on debt contracts and Townsend (1979) on costly state verification the paper assumes that lending is contingent on outcomes and not on the project’s return. The success or failure of a borrower’s project is verifiable (at no cost). They provide sufficient argumentation for this presumption (p.605; see Section 5, p.623). Another assumption is that there is a limited liability constraint where borrowers have no collateralisable wealth, and, that the participation cannot be satisfied by normal individual loans. This latter point hinges on the fact that a borrower will invest in a presumed socially productive project because of positive opportunity costs of labor µ (p.605). However, in many developing areas with extremely high unemployment, people can be so poor that the reservation payoff µ basically is zero. The author(s) may address how the results are affected if a large share of the potential borrowers are persistently poor and face no participation constraint (µ=0). Following Becker’s assortative matching under the sketched assumptions, the study demonstrates that optimal assortative matching under joint liability lending, even while allowing for transfer payments, always results in a match between borrowers of similar type. The author(s) clarify the intuition of this finding (p.609-610). Another intuition the study may
want to point out for their proof is that risky borrowers, although they strictly prefer a safe borrower, care less about partner’s risk type because they are more willing to cope with the risk that unconditionally on their own project success, which is by assumption already lower than that of save borrowers, the other project fails. As such, a priori to their own project’s probability of success it is not profitable to bribe safe borrowers because they can only do so if their project is successful. Moreover, the result that borrowers choose partners of the same type under joint liability lending completely hinges on the assumption that borrowers have full information about each others’ type. It is suggested that the author(s) provide some empirical evidence for this presumption, because it is not obvious that people in a village are able to distinguish how likely others’ projects are to succeed. Also, what happens if the information about each other’s type is imperfect, but positively correlated with the overall outcome? Would such assumption be sufficient? Moreover, the author(s) implicitly argue that if the group size changes that this will not affect their results. However, as is well-known in mechanism design, increasing n often may bring unforeseen complexity (see e.g. Ahlin, 2009). This subject in relationship to the perfect information set about other borrowers project may be given some attention. Based on this result the author(s) also show that safe borrowers are willing to pay a higher amount of joint-liability c than risky borrowers in order to receive a reduction in the interest rate r. This stems from the fact that safe borrowers have safe partners and they do not have to pay c very often (p.611). However, the study may want to point out that how the sensitivity of the change in c as a compensation for the change in r is affected by the probability of a project’s success (both of safe and risky borrowers). One may expect that this relationship is in fact highly depended on the probability of success and in fact non-linear. As such, the relative difference between safe and risky borrowers in terms of their preferred (r, c) may depend both on the probability level of the project’s success and the relative difference between this success rate between safe and risky borrowers. One may suggest adding such discussion to corollary 1. Next, the author(s) again rely on the separation of risk types (assortative matching) to analyze how joint liability acts as screening device for the lender. It is argued that under certain assumptions optimal separating joint liability contracts for safe and risky borrows exist (p.615). However, it requires that the returns from a project (if successful) are high enough to meet both the individual and joint liability payments (r+c). One may question if such high realized returns are realistic. The same holds under a pooling contract. In addition, the author(s) note that “the limited liability constraint requires that a borrower cannot make any transfers to the lender when her project fails.” (p.613). In the paper it is assumed that c > r. A group of (two) borrowers will therefore pay more if one of them fails and the other succeeds (r+c) compared to when both succeed (r+r). As such, a borrower with a successful project will make side payment to the group member equal to r conditional on the failure of the other, so to gain c-r>0. This point is not addressed in the paper. This is important because by rational expectations (and backward induction) in the final stage of the game the lender will never receive r+c, although it seems that the author(s) do assume that this event actually has a positive probability, and, it may negatively affect the zero profit condition. In the worst case this can result in less lending activity under joint liability, and, a less successful pursuit of micro finance in reaching the poor (without subsidies) than is argued. [p.617] In general, a discussion at the moral level this selection mechanism may be required. In this lending literature, several ethical issues with regard to “peer monitoring” to alleviate moral hazard, and “peer pressure” to ensure better enforcement are already covered. It is suggested
to highlight some implications “peer selection” may have apart from the reduced adverse selection problems - especially under separating equilibria, where each group obtains different (r,c). Could it be that inequality/cleavages will rise within communities with large heterogeneity in risk types, and what impact can this have on human development? Could lenders ‘abuse’ this new screening the company that you keep device in unintended manners? Finally, due to space considerations of the journal it is suggested that he introduction of the paper can be shortened. Especially the second and third paragraphs provide background and motivation that may be known to most readers. Also the fifth and sixth paragraph about screening may be stated more briefly. Section 6 (pp. 624-628) may also be skipped In conclusion, the merits of the paper lie in outlining how joint-liability may foster successful lending in poor countries due to a peer selection effect. This is a useful and original analysis given the enormous outreach of micro finance and explains the success of MFIs in terms of repayment. Overall the paper succeeds in presenting a formal model, provides several different situations and it well-argued. We hope the author(s) address the proposed issues above in a revision of the study.