Stock Index Forecasting Model and Its Application Based on Compound Knowledge Mining

In this thesis, we build a compound model, REP-Tree + BP + GA (hereinafter referred to as REPBG Model), to carry out data mining and text mining on quantitative and qualitative factors with impact on the movement of stock index, then we forecast the Shanghai Composite Index for 10 trading days from May 27 to June 10, 2011. First, we use BP+GA model to carry out data mining on quantitative indicators; then, through correlation analysis, we extract text factors with impact on index movement and put them into various categories; next, we introduce the forecasting errors from BP + GA model and the text factors abstracted from the above step into REP Tree for IF-THEN rule analysis, and the result is a group of adjustment rates which consist of compound knowledge data of both qualitative and quantitative factors. We use this group of rates to readjust the forecasting value and accuracy is increased considerably. The above method is used in positive analysis on movement of Shanghai Composite Index. The experimental result indicates: Text mining on top of data mining can increase forecasting accuracy considerably, and this indicates the significance of compound knowledge mining.