Group lending with covariate risk

Group-based lending with joint liability has been a major tool microfinance institutions (“MFIs”) have employed to improve lending feasibility. The related theoretical literature typically assumes borrowers face independent risk. This paper examines how covariate risk affects the usefulness of joint liability lending, in the hidden-information setting of Stiglitz and Weiss (1981) and Ghatak (2000). In a benchmark setting where all agents face the same degree of covariate risk, greater correlation renders group lending less effective; this is because the effective rate of joint liability is reduced when borrowers are more likely to fail together. We focus on a setting where the extensive and intensive margins are distinguished: some agents face independent risk while others face correlated risk. We find that an intermediate prevalence of correlated risk can lead to lower outreach than both a low and a high prevalence. Thus, reaching a market with mixed covariate risk profiles, e.g. farmers and micro-entrepreneurs, can be harder than reaching markets with a single profile of either kind. Assuming limited ability of lenders to use information on borrower correlatedness, we find that higher outreach is often achievable by separately servicing correlated and non-correlated borrowers. This can help explain the existence of specialized institutions such as agricultural banks versus standard microenterprise-focused MFIs.