應用類神經網路對極短線之艾略特波浪行為模式研究-以台指期貨日內交易行為例

Abstract

台灣屬於淺碟式經濟體，中長期的走勢往往會受到各消息面的影響而大幅波動。在全球金融海嘯的衝擊下，台灣金融市場所承受的震盪更是激烈；在這種環境下，投資人往往因為隔夜持有股票導致開盤承受很大的系統風險，亦或是持有個股股票而承受非系統風險；因此本研究為了規避這些風險，以台灣加權股價指數期貨規避非系統風險，以日內交易規避隔夜持有之系統風險。 本研究主要探討艾略特波浪理論中物理力量，分別為時間、斜率、角度、位能和長度，與台灣加權股價指數趨勢之間的關係。藉由輸入波浪物理力量的特徵值，倒傳遞類神經網路學習當中的知識規則，並加以運用於台灣加權股價指數的分析，提出可以預測短期指數趨勢的模型。 研究中採用2000年至2008年台灣加權股價指數做為研究資料，先以不同的時間區段截取波浪理論的轉折點，將轉折點所形成的波浪的物理力量特徵值正規化，做為倒傳遞類神經網路的輸入值，預測下一個波浪的走勢。同時設定不同的門檻值與交易策略，計算其準確率與投資績效。研究結果發現，台灣加權股價指數的趨勢確實會受到物理力量的影響。而在日內交易方面，當時間軸以20分鐘的區間來抓取轉折點的投資報酬率最大，因此在進行日內交易的操作時，宜以20分鐘左右為區段來分析趨勢；此外，當門檻值設定為20點時，不論是在交易的勝率或是總獲利都有相當好的表現。

Taiwan stock market belongs to a shallow-plate market .The long –term trend is often affected by news with high volatility. In the impact of the global financial tsunami, Taiwan's financial markets are subject to more intense shock; In this environment, because of holding stocks overnight , investors often bear the systematic risk when the opening of the market or the risk of non-system. This study adapted TAIEX Futures to avoid systematic risks; intraday trading to avoid non-systematic risks. This study focused on the relationship between Elliott Wave theory of physical forces, namely, time, slope, angle, potential energy and length, and the TAIEX trend. Back-propagation neural network can learn the knowledge of the rules by using the physical forces as input to analysis TAIEX and predict the short-term trends in the study. This study used daily TAIEX from 2000 to 2008 as the research data, and picked the turning point of the wave from different time sections. Then normalized the physical force value as input of the Back-propagation neural network to predict the trend of the next wave. At the same time, the different threshold was set with different trading strategies and to calculate their accuracy rate and investment performance. The result shows that the trend of TAIEX is indeed affected by the physical force. In the intraday trading, when the time sections is 20 minutes, the accuracy rate and investment performance was the largest. So when we trade in the intraday, we should use the time section of 20 minutes to forecast and analysis.