An improved method for stock market forecasting combining high-order time-variant fuzzy logical relationship groups and particle swam optimization

Abstract

Fuzzy forecasting approaches are mainly based on the modeling of fuzzy logical relationships of the historical data. In this paper, an improved model for forecasting stock market indices which combines the High-order Time-Variant Fuzzy Logical Relationship Groups (HV-FLRGs) and Particle Swarm Optimization (PSO) is presented. Firstly, HV-FLRGs are more effective to capture fuzzy relations on time series data than the conventional time-invariant fuzzy logical relationship groups. Secondly, PSO is employed to optimize the length of intervals by searching the space of the universe of discourse. To verify the effectiveness of the proposed model, the historical data of Taiwan Futures Exchange (TAIFEX) are examined. The simulation result shows that the proposed model outperforms the previous forecasting models based on the highorder fuzzy time series. These results are very promising for the future work on the development of fuzzy time series and PSO algorithm in real-world forecasting applications.