A New Framework for Nonlinear Cointegration Analysis and Its Application to Finance

Abstract

With the purpose of analyzing non-stationary time series, this paper innovates the nonlinear cointegration discriminate analysis by introducing support vector machine and innovated feature-weighting model to overcome existing limitations of two methods, that is, the statistical approach and the neural network used by the nonlinear cointegration theory. Then, the application of the innovated method for the investigation of the connection between financial markets is explored. The empirical analysis demonstrates that the support vector machine is effective in analyzing the money demand function and its stability and has advantages in dealing with nonlinear cointegration relations and estimating nonlinear functions over the two methods above.