透過您的圖書館登入
IP:3.238.142.134
  • 學位論文

整合基因演算法與類神經網路於股價走勢之預測

Combination of neural network and GAs for forecasting the Taiwan stock market

若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在股價預測研究上分為三種理論學派:1. 基本分析學派、2. 隨機漫步理論學派、3. 技術分析學派。基本分析是透過總體經濟分析、產業分析、公司基本面分析三個步驟,定位該公司股票的實際合理價值;隨機漫步理論認為,股價複雜之程度近於隨機變化,極難預測,因此衍生出買入持有策略;而技術分析則只需針對供需的變動分析即可。在相關研究裡,以往較常使用的是時間序列、複迴歸分析等數量統計模型,但也因為運用數量統計模型,需做諸多的先前假設與限制,且也不符合屬於非線性型態之股票市場,故本研究以類神經網路中的倒傳遞網路為主,以技術指標作為類神經網路的輸入變數,並運用基因演算法決定網路架構與輸入變數組合,進行未來股價走勢之預測。研究資料係取自於台灣證券交易市場部份歷史資料,研究樣本共五十七支個股,資料範圍於1991年至2006年。本研究將預測模型分為四種,1. 探討在日、週、月不同週期之下,何種準確率為佳;2. 將個股以產業別做區分,探討在此模型下,何種類股其準確率為佳;3. 將個股以資本額大小做區分,探討在此模型下,何種準確率為佳;4. 縮短訓練資料,探討屬於時間序列之股價歷史資訊,捨去過於久遠的資訊,對其準確率的影響。研究結果顯示,預測週期以週、月之準確率較為理想;產業別以電子類股、紡織類股、塑膠類股為佳;在資本額方面,股本較大者,因不易受到市場炒作的影響,故網路學習較理想,準確率也會較佳;捨去過於久遠的資料,大部份都可以保持相同,甚至更好的準確率。本研究並結合了股票擇時與股票選擇,以台積電、正新個股進行模擬交易,計算總報酬率。模擬結果顯示,選股前後之總報酬皆超越買入持有策略。

關鍵字

無資料

並列摘要


In the research of stock price filed, it divided into three parts: fundamental analysis, random-walk theory and technical analysis. The fundamental analysis theory was focused on the reasonable value by three steps with macro-economical, industry analysis and corporation basis. In the random-walk theory, some researchers think that the stock price complexity was approximate to the condition of randomly. Therefore it needs to make strategy by buy and hold. The technique analysis is focused on the fluctuation of supply-demand. in relational research, they often use time series, multi-regression etc. but it needs to assume and some constricts. And it was not like the real stock market. In this thesis, we forecast the volatility of the stock with Back Propagation Neural Network, BPN, and try to decide the network structure and input variable set by genetic algorithm, GAs. There are 57 samples received form the Taiwan stock exchange corporation during 1991 to 2006. We divide the model into four categories. 1. To explore the accuracy during decade of daily, weekly and monthly; 2. To explain the model accuracy by industry; 3. To explain the model accuracy by capitalization; 4. To explore the history data by time series analysis and to abandon the far history data to reduce the training data size. The result show the forecasting decade accuracy are good with weekly and monthly data and in the industry with electric , textile and plastics industries. The accuracy is better with the large capitalization, because the large capitalization don’t effect with the market easily. In this thesis, the research integrated the stock time analysis. In addition, we simulate the examples with TSMC and CST Corporation to trade and calculate the total returns. The result of simulation shows the value of total returns exceeds the strategy of buy and hold.

並列關鍵字

無資料

參考文獻


2. 朱正修,2003,台灣股市與國際股市連動性之研究,國立成功大學統計學研究所,碩士論文。
9. 洪俊瑋,2003,以建諸於移動平均法之類神經模糊系統預測股價指數之變動,靜宜大學企業管理研究所,碩士論文。
27. Balvers, R. J., Cosimano, T. F., and McDonald, B., 1990, “Predicting stock returns in an efficient market”, Journal of Finance, vol.55, pp.1109–1128.
28. Bessembiner, H., and Chan, C., 1995, “The profitability of technical trading rules in the Asian stock markets”, Pacific –Basin Finance Journal, vol.3, pp.257-284.
29. Breen, W., Glosten, L. R., and Jagannathan, R., 1990, “Predictable variations in stock index returns”, Journal of Finance, vol44, pp.1177–1189.

延伸閱讀