Least Squares Support Vector Regression Based CARRX Model for Stock Index Volatility Forecasting

Abstract

CARRX model is a new volatility model. This paper applies least squares support vector regression to the CARRX model and a LSSVR-based CARRX model is established for predicting the range volatility of Chinese stock index. Out-of-sample forecasting results of using the LSSVR-CARRX model are compared with that of the ANN-CARRX model. Empirical results show that for the RMSE, MAE, MPE, Theil and Mincer-Zarnowitz regression test, the LSSVR-CARRX model outperforms the ANN-CARRX model both in static and dynamic forecasting. Therefore, LSSVR-CARRX model is expected to be important in developing the novel strategies for volatility trading and advanced risk management.