Autoregressive Conditional Parameter Model and Applications

Abstract

This paper proposes a new time-varying parameter regression model, which allows time-varying parameters (e.g., the regression coefficient, intercept, variance, skewness) to be conditioned on past information. The proposed model is referred to as the autoregressive conditional parameter (ACP) regression model. Many time series models such as the rolling linear regression, autoregressive and moving average model, generalized autoregressive conditional heteroskedasticity models, autoregressive conditional duration model, autoregressive conditional skewness and kurtosis models, and Poisson autoregression are covered. The identification, estimation and test of ACP model are examined. Stationary conditions for ACP process are also considered. Finally, the empirical study finds that the effects of US dollar and S&P 500 stock index on global commodity price returns are time-varying significantly.