Full metadata record
DC FieldValueLanguage
dc.description.abstract試圖在股市或是衍生性金融商品中投資獲利者不可不重視大盤趨勢的變化,然而影響指數的因素極為廣泛,對於如此大量且複雜的資訊,實非一般投資人可以輕易掌握的。近年來,拜資訊科技快速發展之賜,人工智慧應用於財務金融領域變得可行,它的優點是可以只利用少量且不足的資訊,針對非結構化的問題於複雜的空間中求得有效解。 本研究使用:1.非現貨交易時段之指數資訊;2.股市與期貨的原始交易資料變數;3.國際股市的漲跌變化;4.技術分析指標等四大類輸入資料變數,利用類神經網路工具建構台灣加權指數當日收盤漲跌預測模型與隔n日收盤漲跌預測模型,然後與同一段時間之內採取多元迴歸模式進行優劣比較,進一步選取最適之預測模型,並依據預測結果建立台指期貨當沖買賣交易策略。 在資料樣本方面,訓練期資料期間為2002年5月1日至2004年7月31日共計492個交易日;測試期資料期間為2004年8月1日至2004年10月31日共計52個交易日。實證結果發現類神經網路模型預測當日收盤漲跌的MSE為0.0999,RMSE為0.3161,預測方向準確率為73.0769%;而多元迴歸模型的MSE為0.1685,RMSE為0.4105,預測方向準確率為65.3846%。依據預測結果所建立之台股期貨交易策略,扣除交易成本而計算實證測試期的報酬,在52個日交易天數中,有高達287點的指數期貨報酬的和262.8205%的年報酬率。 本研究成果為:1.使用四大類輸入變數,較能完整掌握影響台股指數之因素;2.縮短模型預測天數,可用於短線交易。並將此類神經網路模型的預測績效與傳統多元迴歸模式相比較,實證結果顯示類神經網路模型不論是在預測收盤漲跌點數或是漲跌方向上,其預測能力皆優於多元迴歸模型。zh_TW
dc.description.abstractFor those who try to profit from the stock or the derivative market, the fluctuation of the benchmark can never be left behind. The factors that could affect such movements may not be easy to probe into. Thanks to the contemporary technology. Nowadays, with the applications of artificial intelligence in the financial field, such movements in the both stock and derivative market are converted into insight from foresight. This study makes use of the followings: 1. the index information from the non-trading period; 2. the raw data from the both stock and future markets; 3: the fluctuation in the international stock market; 4: the technical analysis. These four changing factors will be exported into the neural network with the purposes of predicting the daily closing price and the closing price in n days. These will be compared to the multiple regression model. Non-dominated model will be picked among these three and suitable trading strategies will be developed accordingly. As far as the data sample is concerned, the data of training period is from 2002/5/1 to 2004/7/31, which sums up to 492 trading days; while the data of testing period is from 2004/8/1 to 2004/10/31, which sums up to 52 trading days. The results show that the MSE of neural network is 0.0999 and the RMSE of neural network is 0.3161. The accuracy is 73.0769%. While the MSE of multiple regression is 0.1685 and the RMSE of multiple regression is 0.4105. The accuracy is 65.3846%.The TAIEX futures transaction strategy is constructed according to the forecasting results. After subtracting the transaction fee in 52 trading days, the profits are as high as 287 points at TAIEX futures and the annual return is 262.8205%. This study reveals a couple findings: One is that when incorporating four changing factors stated above, one could possibly get hold of momentum from TAIEX; second is that by shortening the forecasting periods, intraday transactions will be enabled. Lastly, according to the experiment results, the neural network performs better in every presumed spectrum such as up-or downward movement in the stock market or the closing points in TAIEX.en_US
dc.subjectArtifical Neural Networksen_US
dc.subjectMultiple Regressionen_US
dc.titleA Behavioral Study of Combining Knowledge Rules and Neural Networks to Forecast the Intraday Trend of TAIEXen_US
Appears in Collections:Thesis

Files in This Item:

  1. 150401.pdf
  2. 150402.pdf