Spatial or Common Correlated Effects? A Sequential Pretest and the Risk of the Post-Test Estimator

Macroeconomic panels display two forms of cross-sectional dependence usually treated as rival specifications: unobserved common factors with heterogeneous loadings, which motivate common correlated effects estimators, and spatial spillovers, which motivate spatial autoregressive models. This article specifies and evaluates a sequential pretest: common factors are filtered through cross-sectional averages, the residuals are tested for spatial dependence with Moran’s I, and the estimator is chosen accordingly. A Monte Carlo experiment with 96 cells and 1000 replications each varies the weight matrix, spatial parameter, number of factors and panel dimensions. The annual Moran test on filtered residuals is correctly sized, but a rule requiring rejection in a quarter of the periods is conservative by construction. Detection does not translate into lower risk: with common factors and moderate spatial dependence, a more powerful rule raises the mean squared error of the post-test estimator, even when calibrated by permutation, while a joint spatial–factor estimator keeps a low risk and close to nominal coverage if the weight matrix is correctly specified. Applied to 18 Latin American economies (2000–2021), the pretest selects the factor-based estimator and the mean group joint estimator, unlike the pooled one, finds no significant spatial lag.