Francisco Costa

Assistant Professor of Economics at Fundação Getulio Vargas (FGV/EPGE) 
On leave at UCSD Department of Economics.

Research interests: Environmental/Energy Economics, Development Economics, Empirical Trade.

francisco.costa [at]
(+55) (21) 3799-5871
Praia de Botafogo, 190/1100, 
Rio de Janeiro, RJ, 22250-900, Brazil.
This paper studies how one may overestimate the social cost of a long-run corrective policy by neglecting the possibility of hysteresis, i.e. that the policy in earlier periods may have lasting impacts in later periods. In a price-theoretic framework, we show that one statistic is key to evaluate such a bias: the long-term impact of a similar but temporary policy that was known to be temporary. We then provide evidence of the importance of hysteresis, and estimate such a statistic, for a policy-relevant behavior: residential electricity use in a developing country context. We study the 10-year impact of a 9-month long policy in Brazil, which aimed at large temporary reductions in residential electricity use. We exploit the fact that customers of some distribution utilities were not subject to the policy through a difference-in-difference strategy. Using utility-level administrative data, we find that the temporary policy led to a long-run and stable reduction in average electricity use of 11%, or about half of the short-run impact. Using individual monthly billing data for one distribution utility, we find that 69% of customers were still consuming less electricity four years after the policy ended. Household-level microdata suggest that the main mechanism of hysteresis is a persistent change in consumption habits. Incorporating our estimates into our framework illustrates that, neglecting the possibility of hysteresis, one could dramatically overestimate the social cost of long-run corrective policies.
We employ a density discontinuity design to evaluate the deterrence effect of more severe punishments around the legal age of criminal responsibility in Brazil. Motivated by the criminology literature, we propose a novel proxy based on the inherent risk underlying criminal activities. Using violent death rates as a proxy for individual violent criminal involvement, we find no discernible deterrence effects. We additionally study arrest data from the country's third most populous state, Rio de Janeiro, and discuss advantages of our proxy in light of potential underreporting biases from using criminal records.