Forecasting stock volatility using LSSVR-based GARCH model optimized by Siwpso algorithm????(Open Access)

作者：Geng, Li-Yan; Yu, Fei

来源：Journal of Applied Sciences, 2013, 13(22): 5132-5137.

DOI：10.3923/jas.2013.5132.5137

• China stock markets
• Effective approaches
• Empirical research
• Forecasting accuracy
• Inertia weight
• Least squares support vector regression
• Parametric modeling
• Volatility forecasting

Volatility forecasting plays an important role in derivatives pricing, risk management and securities valuation. As a traditional parametric model, GARCH can't forecast financial volatility well. To improve the forecasting accuracy and the modeling speed of GARCH model, this study proposed a hybrid forecasting model for stock volatility forecasting, in which Least squares support vector regression (LSSVR), combining with stochastic inertia weight PSO (SIWPSO), is proposed to GARCH model. First, LSSVR was used to forecast financial volatility under the GARCH framework. Then, the SIWPSO algorithm was adopted to obtain the optimal hyper-parameters needed in the LSSVR model. An empirical research was performed to illustrate the effectiveness of the proposed method. Empirical results from four high frequency stock indices returns in China stock market indicate that the proposed model provides improvement in volatility forecasting performance. The values of HRMSE, HMAE, LL and LINEX of the proposed model are obviously smaller than those of the other two models: LSSVR-GARCH-CV and GARCH. The searching time for the optimal hyper-parameters of the LSSVR model by SIWPSO is much shorter than that under 10-fold CV method. Therefore, the proposed model is an effective approach for forecasting stock volatility.

• 出版日期2013
• 单位石家庄铁道大学