This paper assesses the performance of classical strategies to control for (pre-)trends in difference-in-differences designs.
We focus on three main approaches: controlling for or matching on pre-trends, extrapolating linear trends, and controlling for group-by-time fixed effects. Through Monte Carlo simulations using real labor market data, we examine incidental trends that may emerge due to correlations between treatment and unobserved characteristics.
Drawing on these simulations and supporting analytical results, we provide intuitive insights into the performance of these methods and further formalize the conditions that justify their application.