Dynamic space–time panel data models: An eigendecomposition-based bias-corrected least squares procedure

Jin et al. (2020) proposed an efficient, distribution-free least squares estimation method that utilizes the eigendecomposition of a weight matrix in a dynamic space–time pooled panel data model. Their three-step approach is very powerful compared to the well-known instrumental variable techniques. Unfortunately, for short panels, their method can lead to biased estimates of the autoregressive time dependence parameter and the spatio-temporal diffusion parameter, even when using their bias-corrected estimator. We propose a bias correction method inspired from Bun and Carree (2005, 2006) of the Jin et al. (2020) procedure. We also extend their eigendecomposition-based least squares procedure to the random effects model, the fixed effects model, the Mundlak-type and Chamberlain-type correlated random effects models, the Hausman–Taylor model and the common correlated effects model. Extensive Monte Carlo experiments show the good finite sample properties of the proposed estimators. An application on the link between pollution and economic activities, using a dynamic space–time STIRPAT model with common correlated effects on a panel of 81 countries over 1991–2015, shows the relevance of this approach. It underlines the importance of human activities in the pollution growth while reforestation is one of the most important levers to reduce the CO${}_2$ emissions per capita.