Abstract
We assess the impact of the Brazilian government's conditional cash transfer program Bolsa Família on unhealthy consumption by households, proxied by expenses with ultra-processed food, alcohol, and tobacco products. Using machine learning techniques to improve the propensity score estimation, we analyze the intensive and extensive margin effects of participating in the program on the household purchase of unhealthy products. Our results reveal that program participants spend more on food in general, but not necessarily more on unhealthy options. While we find evidence that participants increase their probability of spending more on food away from home, they do not significantly alter their expenditures on packaged food, alcohol, or tobacco products.