A new approach for estimation of long-run relationships in economic analysis using Engle-Granger and artificial intelligence methods

In time series analysis, most estimation of relationships and tests are typically based on linear estimators and most classical co-integration methods and causality tests are based on OLS regresses. However the linear functional specification is not necessarily the most appropriate form. This paper breaks the ordinary rules in econometrics and makes use of time series with artificial intelligence methods, testing for existence of nonlinear relationship. We illustrate the testing exercise using two examples based on OECD health data. In our illustration we confirm that improved nonlinear AEG and VEC, significantly, have a better ability to identify long run co-integration and causal relationships than ordinary linear ones. Ordinary methods and improved-nonlinear methods demonstrate similar results if the variables in a model are approximately linear.