Prediction of selected Indian stock using a partitioning–interpolation based ARIMA–GARCH model

Abstract

Accurate long-term prediction of time series data (TSD) is a very useful research challenge in diversified fields. As financial TSD are highly volatile, multi-step prediction of financial TSD is a major research problem in TSD mining. The two challenges encountered are, maintaining high prediction accuracy and preserving the data trend across the forecast horizon. The linear traditional models such as autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroscedastic (GARCH) preserve data trend to some extent, at the cost of prediction accuracy. Non-linear models like ANN maintain prediction accuracy by sacrificing data trend. In this paper, a linear hybrid model, which maintains prediction accuracy while preserving data trend, is proposed. A quantitative reasoning analysis justifying the accuracy of proposed model is also presented. A moving-average (MA) filter based pre-processing, partitioning and interpolation (PI) technique are incorporated by the proposed model. Some existing models and the proposed model are applied on selected NSE India stock market data. Performance results show that for multi-step ahead prediction, the proposed model outperforms the others in terms of both prediction accuracy and preserving data trend.

Keywords

• ARIMA
• GARCH
• Highly volatile data
• Predictive data mining
• Partitioning and interpolation
• Time series forecasting