(with Johannes Boehm and Etienne Fize)

We conduct a randomized controlled trial to study the consumption response of French households to unanticipated one-time money transfers of 300 Euros. Using prepaid debit cards, we consider three implementation designs: (i) a transfer without restrictions; (ii) a transfer where any unspent value expires after three weeks; (iii) a transfer subject to a 10% negative interest rate every week. We observe participants’ main bank accounts, such that we can compute the impact of the transfer on their overall spending. We establish five facts about MPCs in this setting. First, we find that participants in the baseline treatment group have an average marginal propensity to consume (MPC) of 22 percent over one month. Second, we find that implementation design matters: the one-month MPC is substantially higher for treatment groups where any remaining balance becomes unusable after three weeks (60%) or where remaining balances are subject to the 10% negative interest rate every week (36%). Third, we document that the cumulative consumption responses are concentrated in the first weeks following the transfer and are flat thereafter. Fourth, we find that there is significant MPC heterogeneity by observed household characteristics, including by liquid wealth, current income, proxies for permanent income, gender, and age; the MPC remains high even for agents with liquid wealth exceeding twice their monthly income. Fifth, we estimate the unconditional distribution of MPCs across households and find that a large fraction of households have high MPCs. These facts are difficult to reconcile with the consumption response in standard Heterogeneous Agent New Keynesian models, which is long-lived and driven by a small set of illiquid households at their borrowing constraints. Furthermore, we observe that households in the treatment groups with a short expiry date or a negative interest rate frequently use other means of payment while still having a sufficient balance on the prepaid card to cover their expenses, indicating that participants see money as non-fungible. Our finding that households consume more when presented with an urgent spending need lends support to theories where the salience of treatments affects economic choices. We conclude that implementation design and the targeting of transfers can greatly alter the effectiveness of stimulus policies.