Ignace De Vos and his co-author Ovidijus Stauskas show that while the pooled CCE (CCEP) estimator is popular for eliminating unobserved components—even when variables exhibit different orders of integration—it suffers from a disruptive asymptotic bias in typical macroeconomic panel settings. Their study establishes that the cross‑section bootstrap successfully replicates the distribution of CCE estimators, enabling straightforward bias correction and valid confidence intervals under very general factor structures. This broadens the method’s applicability in empirical macroeconomic analysis, where unobserved stationary and non‑stationary components often coexist.
The authors further demonstrate that their method performs strongly in simulation experiments, outperforming competing bias‑correction techniques and alternative estimators. They also apply the approach to a gravity trade model using the Serlenga and Shin (2007) dataset, illustrating its practical value in real‑world empirical work. Their findings underscore the importance of robust bootstrapping tools in modern econometrics, especially when dealing with complex interactive effects and pervasive unobserved heterogeneity.
The paper is titled Cross-Section Bootstrap for CCE Regressions with General Unknown Factors and is available online.