應用高頻操作於台灣期貨指數之投資策略研究

Abstract

抓住趨勢、順著潮流投資是許多投資者的指導原則，也是許多研究及分析工具想達到的最終也是最重要的目的！但是台灣股市易受歐美等經濟體系之影響，因此投資人常蒙受隔夜風險的損失，再加上當沖保證金減半制度實施、政府連續調降交易稅、電子交易使得手續費逐年下降，使得期貨交易成本大幅降低，因此採用日內投資者與日遽增。但日內投資若是使用一般的技術指標大多是以日為單位所計算出來的，對於日內投資者實在不是非常的適用，因此只能單純以股市的走勢或線圖來判斷，但此種操作既耗時傷神且也無法參考歷史資料的記憶，因此常出現追高殺低的風險；故本研究提出了以秒為單位來計算技術指標，並加入長線保護短線之概念，在極高頻的操作模式下，以期可為投資人帶來一種新的投資策略，並可帶來超額的利潤！

本研究提出以以兩層式類神經網路學習方式，搭配以秒為單位之長短期技術分析，學習台灣加權指數期貨日內趨勢行為，期望可從從歷史股價趨勢中，找出其中之知識規則來預判未來股市走勢。透過多重類神經網路的優勢來增加類神經網路模型的穩定度及準確度，使類神經網路的輸出更具有可靠性，以其建立一個台灣指數期貨極高頻的日內操作的預測模型！

由實驗結果得知，多重類神經網路模型在預測能力上比單一類神經網路模型準確率高約30%，而獲利能力約為單一類神經網路模型的兩倍，由此可見多重類神經網路模型在統整及判斷長、短線的物理力量後，的確可帶來較好的準確度及獲利率。也證實了由本實驗提出的台灣指數期貨極高頻的日內操作模型是有效且可帶來超額的利潤的！

To drive the potential and go with the current are the principles of many investors.This is the ultimate and most important goal of many researches and Analysis Tools However, Taiwan Stock Market is affected by Europe's and America's economic system easily. As a result, investors often undergo the loss of overnight risk. Besides, the intraday trading futures margins reduced by half, the government continuous lower sale tax and electronic transactions lower the fee and reduce prompt sale cost a lot. These lead to increase intraday trade investors. It’s not proper for them to adopt technical specification calculated by “ Day-based Data”. As a result, it’s only can be determined by Taiwan Stock Market Trend or Chart. But it takes too much time and it doesn’t refer to historical data. We will often meet the risk of “ Buying at highest price and selling at lowest price”, so I propose to adopt “ Second-based Data” to calculate technical specification combined with the concept of long-term protection of short-term .In the high-frequency trading strategy may bring new Investment Strategy and extra profit.

This study addresses a multi-neural network model with second-based technical analysis data to study TAIEX intraday trend, and I expect to find the rules of predicting the TAIEX intraday trend by the historical data. By using the advantage of multi- neural network model, I hope to provide a more stabled and reliable model to predict the TAIEX intraday trend by using high-frequency trading strategy!

The result shows that the multi-neural networks is more accurate than single neural network by almost 30%, and get almost twice profit than single neural network. We can confirm after the multi-neural networks integrating the long-term and short-term sub-networks, it shows the better accuracy and profit. And we can verify the model of high-frequency trading of Taiwan index futures market strategy is effective and more profitable!