運用分類元學習技術進行隔夜價差行為模式分析－以台灣加權指數為例

Abstract

本研究嘗試導入具有動態學習能力之人工智慧方法XCS分類元系統，應用於股票市場之隔夜價差行為的分析，並以台灣加權指數為例。 因公開市場的交易時間限制所形成之投資風險，稱為隔夜風險。此風險為由於在今日收盤至隔日開盤的期間，無法進行買賣交易之限制，而形成隔夜效應所產生的風險。而這種斷層的現象與股市投資的走勢預測及投資效益具高度相關。是以在過去的相關證券投資，由於無法掌握此行為現象，常被迫中斷獲利或採取避險的動作，如此常會使投資效益無法達到最大。然而，如能適時的掌握隔夜價差之行為模式，即可以更精確的提前採取平倉或避險的動作。如此，可使證券投資的相關決策更趨完備，並獲得更高的投資效益。 在本研究中以全球股票市場之間相關資訊為預測模型的第一類輸入因子，及以股市交易資料的長、短期自我相關現象為預測模型的第二類輸入因子，透過XCS分類元系統的關連與歸納之特性，並以股市歷史交易資料來進行模型之學習而產生相關行為規則，最後再以此行為規則進行實證測試。在實證研究中，本研究所提出的XCS預測模型系統成功地掌握台灣加權指數之隔夜效應行為模式，並且在測試階段所獲得之預測準確率及累積之投資效益都有顯著成果。

This study attempted to applied the artificial intelligence approach, extend learning classifier system (XCS) with dynamical learning ability, on overnight price-difference behavior for stock market analysis.

The restriction on the trading time in stock market brings the investment risks and the overnight risk which is effected investors who make any bargain during the close time of this today to the opening of the following trading day. By the way, it was mightily interrelated with the trend prediction of the stock market and the investment profit. Investors often could not achieve the better profits due to they did not know this phenomenon and then must take hedging or to stop the profits while uncertainly situations rise. However, investors will obtain more portfolios by this overnight price-difference, and then they will take the better investment to offset or to hedge on that right time.

Thus, this study built up a model to predict the trend of the stock market. The input factors of this model have been comprised not only the relative information between two global stock markets but also the autocorrelations of long term and short term of stock trading data. By the generalization and association abilities of XCS, it is applied in this study to produce some rules of the behavior patterns by trained from historical trading data. After that, to evaluate the performance of the model is using these rules as investment strategies, to the empirical experiments. Meanwhile, it’s accumulated profits with four kinds of strategies. The result has demonstrated that the overnight effect indeed effected the Taiwan economic market, and the proposed model successfully has been simulated the phenomenon with remarkable outcome.