Volatility Forecasting of Stock-Market Index- An Intelligent Approach with Neuro-Fuzzy Computing and Swarm Intelligence

Abstract

In this paper, we report the study results using the theory of neuro-fuzzy system (NFS) and swarm intelligence to the problem of volatility forecasting for stock index. A new multi-group particle swarm optimization (MGPSO) is presented in the paper. For fast learning, the MGPSO is combined with the famous recursive least squares estimator (RLSE) to form the MGPSO-RLSE hybrid learning method to adapt the NFS. The proposed approach is applied to the Taiwan stock exchange capitalization weighted stock index (TAIEX) and the National Association of Securities Dealers Automated Quotation (NASDAQ) index, respectively. For performance comparison, we design four cases of NFS predictors using different methods, respectively. For fuzzy If-Then rules, the first two cases are by experience-based design while the last two are automatically generated by the subtractive clustering (SC) method. The performances by the four NFS predictors are summarized. According to the experimental results, all the NFS predictors have excellent performance for volatility forecast. Among them, the cluster-based NFS predictor with the MGPSO-RLSE hybrid learning method shows the best performance in forecasting accuracy.