In many organizations, agents operate on a fixed budget and allocate funds across sub-periods of the fiscal year. Commonly, unspent budget must be returned to the principal at year-end to prevent policy drifts. However, this savings constraint may induce inefficient spending patterns, such as expenditure surges before the budget expires. This study characterizes optimal budget rollover policy in a model which captures this trade-off. Applying it to Swiss federal consulting spending, I show that substantial welfare gains are possible by allowing agencies to retain one third of unspent funds. Most gains at the optimum stem from a larger share of the budget being returned to the principal, while using the remainder more efficiently to provide a similar amount of public good as under a use-it-or-lose-it rule. The model’s predictions are validated empirically using causal evidence from a staggered reform that liberalized budget rollover rules.

The latest version of the draft can be downloaded here.  

We leverage a novel spatial IV approach to develop a reduced-form estimator that maps local trade shocks into aggregate outcomes, accounting for inter-regional spillovers. For the China shock in the U.S., we find strong evidence for employment spillovers at the local level, which appear to propagate through input-output linkages rather than labor mobility. They shift the shock’s employment ramifications away from the Pacific and North Atlantic towards the South Atlantic region. For aggregate employment, our model rationalizes the 30% difference between Autor et. al (2013) and the structural follow-up literature but implies that it is insignificant from a statistical standpoint.

Draft available upon request.

The threat of spillover bias looms over the findings of many causal empirical studies in the social sciences. This article analyzes how and when simulating contagion effects can help to establish robustness of direct and total average treatment effects against such violation of the Stable Unit Treatment Value Assumption. In particular for quasi-experimental settings, direct estimation of contagion effects may be infeasible, such that sensitivity analysis becomes the second best option. I establish conditions under which the implied distortion to the observed effect can be interpreted as an upper bound on the distortion from contagion in the presence of nonlinearities. Finally, I apply the method to check the spillover-robustness of a paper's findings, which estimates the causal effect of deworming treatments in middle school on socio-economic outcomes later in the participants' lives in Kenya.

Draft available upon request.