以考慮交易量之市場輪廓預測台指期市場之行為

Abstract

本研究嘗試利用市場邏輯理論所提出的市場輪廓原理，並加入交易量變化的考量，作為倒傳遞類神經網路的輸入變數，以期能建構一優於單純隨機交易之模型，並比較加入交易量變化的考量，是否能有效提升投資績效，進而從中發掘市場邏輯規則與知識。 為使倒傳遞類神經網路能學習市場趨勢變化的關鍵，本研究透過市場輪廓價格偏離值、擺盪因子及價值區間變化指標的計算，再加入交易量的考量，試圖比較以市場輪廓為出發點所建立的指標，是否能較隨機交易具有較佳的預測能力，並比較包含交易量變化考量的市場輪廓模型是否較未加入交易量變化考量的市場輪廓模型，更能準確預測股價的漲跌。 實驗結果顯示，不論有無加入交易量變化的考量，以市場輪廓為出發點所建立的模型均較隨機交易之績效為佳，且以預測區間為5日的績效為最佳，顯見市場輪廓為一有效的投資評估工具。另外，實驗結果亦顯示，加入交易量變化的考量後，並未顯著提升系統預測的準確率，原因可能在於本研究僅以交易量變化情形作為指標，而未將相對應的價格變化關係納入考量所致，爰建議後續研究者可建構一綜合價量關係的指標，以有效提升投資績效。

This research applies market profile, which is mentioned by the theory of market logic, and considers the change in trade volume to establish parameters to back propagation neural network to construct a better model than random trading behavior. This research also checks whether the model has a better investment performance with considering the change in trade volume and eventually extrapolates market logic and knowledge. To let back propagation neural network learn the key to market trend changes, this study uses the indicators of price biases in market profile, rotation factor, the change in value area and the change in trade volume to see if the model constructed with the indicators derived from market profile has better predicting ability than random trading behavior does. This study also tests if the market profile model with considering the change in trade volume predicts the stock price more accurately than the model without considering the change in trade volume does. Experimental results show that the model based on market profile has a better performance than random trading does no matter whether the change in trade volume is considered, especially the 5-day prediction. Therefore, market profile is an effective instrument for making investing decisions. Moreover, experimental results also show that the model that includes the parameter of the change in trade volume does not significantly predict the market trend better. The reason may be that the indicator of the change in trade volume does not include the relating price variation. As a result, the subsequent researchers can establish an indicator which contains both price and volume information and experiment its effectiveness on investment performance.