應用多重類神經網路進行台灣指數期貨跨日走勢行為研究

Abstract

台灣的金融市場在衍生性商品的推出以後，投資人得以從以往單純的股票投資到各式符合投資人需求的商品，像是期貨、選擇權、基金等等，投資人希望藉由這些金融商品能夠達到投資理財的目的。投資理財是一門重要的學問，也是門不太容易的學問，必須敏銳地趨利避險，才能提高獲利的機會。 對所有投資人來說，投資決策中最重要的三大要件為：買什麼、何時上車、何時下車。而隨著行情上下變動，運用定期定額機制化投資，時間一到不論價位就買進，可長期有效的攤平成本；但到底是應該長期持有？還是賺了就跑？該如何設定停損、停利點，正是投資人最難以拿捏的課題。 因此，如何有效的掌握股市波段行為去擬定自身的交易策略，便是本研究的研究目的，期能建立一個穩定的模型與機制，提供投資人一個在進行投資行為上的參考依據，順應近年來人工智慧在金融領域方面的研究盛行，利用人工智慧的方法似乎能改善以往技術分析過於主觀的缺點，因此產生了希望能夠利用人工智慧的方法，來協助投資人在進行金融投資上有個方向，而不再只是憑直覺式的進出場。 由實證結果顯示，可以知道經過多重類神經網路學習的實驗結果表現都較隨機漫步模型與單一子網路的結果來的好，平均的準確率與平均的獲利點數上來看都是透過多重類神經網路這組表現最穩定。說明了長線趨勢行為是能夠保護短線趨勢行為的。

After the derivatives published, people can invest from the previous simple stock investment to people needs to meet all kinds of commodities, such as futures, options, funds. people want to achieve the purpose of investment and financial management by these financial products. For all investors, the most important elements of the investment decisions are: what to buy and when to get on, when to get off. With the market up and down movements, the use of regularly investing a fixed mechanism, when the time comes to buy regardless of price, can be an effective long-term cost-sharing level; but in the end should be long-term hold it? Or make a sell? For investors, how to set stop loss, stop interest point, are the most difficult to draw the line issue. Therefore, the purpose of this study is to act effectively control the stock market to develop their own trading strategy, to build a stable of models and mechanisms to provide investors carrying out investment activities on a frame of reference, comply with the recent the field of artificial intelligence research in the financial prevalence of use of artificial intelligence methods to improve the previous technical analysis seems to be too subjective shortcomings, had attempted to use artificial intelligence methods to assist the investors in making financial investments has a direction rather than just playing into the intuitive style. The results show that, Multiple Neural Network outperforming is better than the random walk model with a single network. On the average of accuracy and average earnings points, Multiple neural network performance is good. Verifies the long-term trend of behavior is to protect the short-term trends in behavior.