Nonlinear earnings persistence

This study employs panel smooth transition regression (PSTR) models with different lagged variables of earnings components as regressor to evaluate earnings persistence effects. The models can resolve collinearity problems between predictors, reflect firms' volatile or irregular earnings streams that are likely derived from long-run investments, and provide more useful information for improving forecasting performance. Most importantly, they can describe differential earnings persistence effects between different regimes that have not been verified by previous studies. Our empirical results support these arguments.