We use a human-subjects experiment to test the effects of a simple mechanism designed to increase cooperation and efficiency in the trust game. In the equilibrium of the standard trust game, the investor does not invest, foreseeing that the allocator would have kept all of the returns from investment. Our mechanism adds a preplay escrow stage, in which the allocator places an amount (possibly zero) into escrow, to be forfeited if he keeps the proceeds of investment for himself. In the experiment, we vary the amounts that can be put into escrow. Our baseline treatment has no escrow. In a second treatment, only low escrow choices are possible, so the equilibrium is unaffected. In our third treatment, there is an escrow amount high enough that, in equilibrium, investment and sharing of the proceeds will occur. Two additional treatments mirror our second and third, except that in these, the escrow amount is randomly chosen and imposed on the allocator. We find that the high escrow amount, when chosen, does lead to the predicted efficient result. Contrary to the equilibrium prediction, we also find substantial investment in both the baseline and “low-escrow” treatments, leading to markedly higher efficiency than predicted (albeit well below that when the high amount is chosen). Over time, however, cooperation and efficiency after low or zero escrow amounts decline. We find only weak evidence for “crowding-out”, which predicts that given a low or zero amount placed into escrow in non-baseline treatments, investment and efficiency would actually be lower than in the baseline. We also find that initially, investment is more likely after allocators place the maximum possible amount into escrow – as if this action by allocators is being (mistakenly) read by investors as a signal that allocators plan to share. All of these results are seen when escrow choices are imposed as well as when they are voluntary.
