Selection into and across Credit Contracts: Theory and Field Research
Working Paper No. 03-W23
Christian Ahlin and Robert Townsend
ABSTRACT [article]
Various theories make predictions about the relative advantages of individual loans versus joint liability loans. If we imagine that lenders
facing moral hazard make relative performance comparisons in determining
stringency in repayment, then individual loans should vary positively with
covariance of output across funded projects. Relatively new work also
highlights inequality and heterogeneity in preferences, establishing that
wealth of the agents relative to the bank, and wealth dispersion among
potential joint liability partners, are important factors determining the
likelihood of the joint liability regime. An alternative imperfect
information model also addresses the question of which agents will accept
a group contract and borrow and which will pursue outside options. We
attempt to test these various models using relatively rich data gathered
in field research in Thailand, measuring not only the presence of joint
liability versus individual loans, but also measuring various of the key
variables suggested by these theories. As predicted by one of the
theories, the prevalence of joint liability contracts relative to
individual contracts exhibits a U-shaped relationship with the wealth of
the borrowing pair and increases with the wealth dispersion. (We control
for wealth that can be used as collateral.) Contrary to one theory, we
find no evidence joint liability borrowing becomes less likely as
covariance of output increases. We do find, consistent with our modified
version of the model with adverse selection, that higher
correlation makes joint liability borrowing more likely relative to all
outside options. We also find direct evidence consistent with adverse
selection in the credit market, in that the likelihood of joint-liability
borrowing increases the lower is the probability of project success. We
are able to distinguish this result from an alternative moral hazard
explanation. Strikingly, most of the results disappear if we do not
condition the sample according to the dictates of the models.