應用類神經網路於台灣股市財務指標與報酬率之研究

Abstract

身處於資訊時代，透過網路這個傳播媒介，應該是最接近尤金·法馬（Eugene Fama）所提的效率市場的時代。但是，效率市場這個假說，已經被許多實證研究證明只是個理想狀態，人的行為才是股票是最大的變數，而展望理論完整的詮釋人們對於預期效應的反應。相對而言，這也提供了所有人一個機會，我們是否可以透過資訊系統的幫助，找出一個有效衡量股票價值的模型，藉此獲得超額的利潤? 本研究採用國外知名專家學者所研究的基本面財務指標，來衡量股票的價值及價格，並結合神經網路的學習機制，找出指標與報酬率在一段較期間內的關聯性與可預測性。實驗結果顯示，透過神經網路學習後的模型，可以找出一組報酬率較高的股票，能有效的擊敗台灣五十及中型100這兩個與大盤績效相當為吻合的的ETF。

In the information age of 21st century, the Networking environment now is closed to Eugene Fama’s Efficient Marketing Hypothesis. However, many studies have been found the efficient marketing hypothesis is the ideal environment. One non-predictable factor has influence on the financial marketing is human behavior. In the other words, it provides an opportunity for utilizing the information management system to developing a model to measure the stock values and obtain excessive profit. This thesis used fundamental finance indicators which are proved by Warrant Buffet and other experts, for measuring stock values and prices. This thesis integrated Back Propagation Neural Network’s learning mechanism for discovering the correlation between fundamental finance indicators and returns in the period time. The result was shown that a group stocks selected through the model, the returns beat Taiwan 50 Index and Taiwan Mid-Cap 100 Index.