View article

Market-wide events (eg, financial crises) and regulatory changes empirically have group heterogenous impact on firm outcomes. Inappropriate modelling of the heterogeneity by existing econometric models such as time-fixed effect (assuming a homogenous response to shocks) and industry-year interacted fixed effect (assuming a heterogenous responses to shocks based on industry) is likely to result in biased estimates. This paper investigates the effect of heterogenous responses to common shocks for existing panel studies. We demonstrate theoretically and empirically that ignoring time-varying unobserved heterogeneity that is correlated with regressors in current empirical practices leads to biased estimates and standard errors. To overcome the bias, we propose the use of the “group fixed effect, GFE” class of models, which produce consistent estimates even under the two-way fixed effect and interacted fixed effect data generating processes. We study the finite sample properties of GFE through simulations and demonstrate its economic importance with two empirical applications. We also extend the GFE class of models to accommodate two-stage least squares estimators. Finally, we provide researchers with guidance and user-written functions in statistical packages to overcome the limitations of existing approaches.