Stock Index Forecasting Research Based On ARIMA-LSSVM Hybrid Model

The study on time series is very important in many applications. It can provide valuable reference for decision makers to improve the accuracy of forecasting. The research of stock index forecasting can help investors making investment strategy and getting stable reward. Autoregressive Integrated Moving Average(ARIMA) model has been one of widely used traditional linear models.However, ARIMA model can’t capture the nonlinear pattern in time series.Support Vector Machine(SVM) is a new machine learning method developed from statistical learning theory and based on VC dimension theory and structural risk minimization. SVM has successfully resolved many nonlinear regression estimation problems. Least Squares Support Vector Machine(LSSVM) has improved the standard SVM. Standard SVM solves the quadratic programming problem, while LSSVM solves the linear programming through transformations.LSSVM reduces the complexity of computation and improves computing speed.The integration of different models may improve the accuracy of forecasting effectively, especially when there is a big difference between the individual models.SSE 180 index is one of China’s main stock indexes. Stock index time series were influenced by many factors, among which the relationship between linear component and nonlinear component is complex. ARIMA model will be applied to SSE 180 index. Then a novel hybridization of ARIMA and LSSVM based on Khashei’s idea is proposed and it considers the characteristics of SSE 180. The forecasting accuracy of ARIMA-LSSVM hybrid model will be compared with individual models, traditional model, BP neural network and ARIMA-BP hybrid model. The forecasting accuracy is measured by RMSE and MAPE indicators.The RMSE indicator measures the absolute error between the actual value and forecasting value, and the MAPE indicator measures the relative error between them.The empirical results indicate that the forecasting performance of ARIMA-LSSVM hybrid model outperforms other models in this paper. In addition, not all hybrid models are superior to the individual models in forecasting. It indicates that only individual models combined properly will the hybrid models have a better performance in forecasting.