An Approach Based on Multi-feature Wavelet and Elm Algorithm for Forecasting Outlier Occurrence in Chinese Stock Market

The prediction of outliers plays an important role in stock arbitrage and risk avoiding. While most of researches focused on detecting outliers and removing them to forecast time series data, few focused on forecasting the occurrence of outliers. The main goal of this work is to forecast outlier occurrence in Chinese stock market. Firstly, we detect abnormal points of two market indexes and six individual stocks based on multi-feature wavelet method. Compared with single feature wavelet method, multi-feature wavelet method was more reliable, since it captures more information of the market and avoids “masking effect”. Moreover, according to the detected results, we construct an outlier forecasting model for the two market indexes and six individual stocks based on multi-feature extreme learning machine (ELM) algorithm. Finally, we find out the optimal number of hidden nodes to forecast when the mean of forecasting accuracy was the highest which approximately equals 98.255% and the variance of forecasting accuracy was only 0.000193.