本研究的研究樣本為台灣證券交易所2010年1月2日至2015年10月30日共1442個交易日期間皆保存於台灣50名單中的企業共38家的外資、投信、自營商等三大法人和融資融券的交易量每以及月營收等歷史交易資料。
研究中以多元貝式定理演算法分析三大法人和融資融券上半月的交易量以及營收的月增率以及年增率對下半月收盤價趨勢進行分析，將知識庫的訓練範圍分為樣本前70%資料以及滾動式資料範圍，其中滾動式資料範圍又分為兩年及三年，並在最後以模擬交易進行預測結果驗證，模擬交易的過程中針對MMB預測值及獲利率進行門檻值設定，模擬交易結果以產業以及企業資本等級進行探討，分析預測模型在不同的投資需求中可行的調整策略。
研究結果發現，當投資需求為單次高獲利率時以企業資本等級1及等級3的企業較適合，當MMB預測值門檻為0.9且交易門檻為6%時平均獲利率可達到1.63%，預測正確率達到76%，如投資需求為低風險則適合以傳統產業為投資對象，當MMB預測值門檻為0.1且交易門檻為0.6%時預測正確率可達到99%，而平均獲利率則為1.08%。
This research adopted historical share prices, revenue and historical trading vol-ume from three primary institutional investor, margin trading and short selling from january 2, 2010 to October 30, 2015 of all listed companies whicth From component index stocks in Taiwan 50 in the period 2010 to 2015 are stored in the list of 38 com-panies by Taiwan Stock Exchange Corporation.
In the research use multimembership bayesian to analyze revenue and the half month of the trading volume from three major corporate, margin and short selling in order to understand the investment behavior of investors Effect of the stock price.
In the training process of knowledge base, we set three kinds of training range:
(1)The first 70% of the sample.
(2)Two year of the roller data.
(3)Three year of the roller data.
Use these three training range to analyze the stability of investors behavior.In the end, we use simulated trading to verify the predictions.During the simulated trading we set the threshold of the predicted value and the profit rate, and analysis the fore-casting models in different investment demand to establish possible trading strategies.
