We investigate how well social collateral does as an alternative to traditional physical collateral. We do so by studying cosigned loans - a borrower´s loan is backed by the personal guarantee of a cosigner. We use a regression discontinuity approach with data from South Indian bidding Roscas. Our main finding is that cosigners do indeed provide social collateral: doubling the number of cosigners halves the probability of arrears for high risk borrowers. We then distinguish between different theories of social collateral. Cosigners may be e¤ective as a monitoring device (a borrower would pay to rid herself of the nuisance of a cosigner) or as an insurance device (a borrower would pay for the benefit of a cosigner). We show that these two interpretations of cosigning have different empirical predictions in the context of a bidding Roscas. We find support for the insurance role of cosigners.