應用類神經網路於台股權值股籌碼面的知識發現

Abstract

目前股票分析實證研究上，大多以技術面、基本面等資訊對動態的股票市場做預測與投資，但這些屬於落後之資訊，不但常常出現過度或過慢反應市場趨勢，導致預測準確率降低與過多的投資損失，因此本研究本著「量先價而行」之理論，改以籌碼分析買賣雙方交易行為對未來股價趨勢影響之探討，期望找出更具參考性之投資決策依據。由外資、投信及自營商所組成的三大法人具有主導台灣金融交易市場的地位，三大法人所操作的資金在股票市場進出的動向經常受到投資人的高度矚目。機構投資人可被認為是主動交易、交易資訊充足且具有影響力的交易者，交易時擁有足夠的資訊，並對股價有一定影響性。三大法人通常所投資的標的多屬於大型股。而一般小型股多被認為易受人為炒作影響，因此本研究將進行台灣權值股籌碼面的研究。 本研究主要以籌碼面指標配合類神經網路針對未來股價趨勢做預測，學習股價短線趨勢行為，嘗試從股價趨勢中，找出知識規則，建構一個有效的股價預測模型，達到輔助投資者擬訂投資策略。 由實驗結果得知，類神經網路模型在預測能力以及獲利能力上，表現較隨機漫步模型優異，準確率提升。由此可知神經網路經由總和評判，統整股票籌碼物理力量後的效果確實會優於隨機漫步模型，也證實了該模型能提供投資人投資操作上的建議與幫助。

Recent empirical studies on stocks depend mostly on technical and fundamental analysis, which are used for investment and forecast. However, these are laggard information could not reflect the behavior of stock price immediately and investors could not do forecast well by it. Therefore, this study proposes a volume is the lead indicator of price methodology to analyze the trend of Taiwan Weighted Stock. The research expects to provide a more accurate model for stock price forecasting. By foreign investors, investment trust and dealers consisting of three major institutional, which with the leading market position in Taiwan, The operation of three major institutional funds in the stock market trends often get great deal of attention by investors. Institutional investors can be considered active trading, transaction information is sufficient and influential traders, who have enough information on transactions, and have a certain influence of stock. The three major institutional investors are usually the subject of large stocks, however the small stocks generally are considered more vulnerable to manipulate by human, and therefore this research study will be focus on the traded volume of Taiwan Weighted Stock. This study forecasts stock price by using trade volume and neural network. The network is applied to learn a rule from stock price change within a few days and to build a reliable prediction model for investors. Construct an efficient neural network model on stock price prediction. To assist investors develop investment strategies. The experiments demonstrate neural network outperforms random walk model in forecasting accuracy, providing better advice on investment operation.