Stationary Regressors in Cointegrating Polynomial Regression

Abstract

This thesis investigates the finite sample performance of the fully modified OLS estimator for cointegrating polynomial regressions (CPR),developed by Wagner and Hong (2016), including stationary regressors to the model. To be precise, this thesis considers regressions including deterministic variables, integrated processes, powers of integrated processes and stationary variables as explanatory variables and stationary errors. The errors are allowed to be serially correlated and the regressors are allowed to be endogenous except for the stationary regressors where both cases, i.e. predetermined or endogenous stationary regressors, are examined in this thesis. The basis for the finite sample performance investigation is a simulation study which shows that the assumption of allowing endogeneity of the stationary regressors can not be made as in this case the FM-OLS estimator seem to be not consistent anymore and statistical inference is no longer feasible for every level of serial correlation and endogeneity.