Theory and methodology for dynamic panel data: tested by simulations based on financial data

A new method is introduced for panel-data models. Asymptotic robustness is used for a multivariate model with latent variables for a family of estimators. It is shown numerically that in comparison to standard methods we obtain: 1) better predictions in out-of-sample occasions; 2) smaller asymptotic standard errors (a.s.e.s); 3) more accurate a.s.e.s; 4) very small bias. Our methodology handles dynamic models with lag-independent variables, individual and time effects, time heteroscedasticity, non-normality, non-stationarity, fixed variables, non-linear and variant-over-time coefficients, and unbalanced data, by using restrictions on the parameters and the multi-sample technique (m.s.t.). Also, a novel formula for the duplication matrix is provided and a solution for a matrix equation is given.