Analisys of Hidden Cointegration in Financial Time Series

The techniques to analyze time series generally use model identification and estimation procedures based on the stationarity assumption of the Data Generating Process (in short DGPs). This hypothesis is often violated when we study financial phenomena, that show non stationary features. When we consider multivariate time series, the series can present spurious correlation. Engle and Granger faced this problem introducing the idea of linear cointegration. Recently Granger and Yoon have introduced the idea of hidden cointegration to take into account the asymmetric behaviour of a time series with respect to the shocks that hit another one. In this work, we present some applications of hidden cointegration in the analysis of some Italian stocks price time series, for which linear cointegration was not evident. For this reason, we have search a possible hidden cointegration relationship. The series were decomposed in their positive and negative components on which we estimated four hidden cointegration models. Then we performed ADF tests (for models without constant and without deterministic trend) on the residuals of these four models in such a way that we could detect if there exists one model generating stationary residuals. Empirical results show that an hidden cointegration is evident for some couples of components