標題: 以1分單位建構市場輪廓圖於台指期市場極短線行為分析
Market Profile Behavior Analysis based on 1-minute intervals of TPO
作者: 陳俊生
Chen,Chun-Sheng
陳安斌
Chen,An-Pin
管理學院資訊管理學程
關鍵字: 臺灣指數期貨、市場輪廓、倒傳遞類神經網路;TAIEX Futures, Market Profile, Neural Network
公開日期: 2012
摘要: 財政部為吸引更多投資人投入期貨市場,將期貨交易稅率由十萬分之四調整為十萬分之二,預期將活絡股市,提升股市動能,根據以往經驗,只要調降期貨交易稅率,往往交易量即有顯著攀升,效果實為明顯。 市場輪廓理論以圖形化方式,描述價格隨著時間進展的變化過程,所發展出來的分析工具。本研究提出從市場輪廓中,萃取出特徵值,作為倒傳遞類神經網路的輸入值,期望透過期貨市場的物理力量之總合評判,嘗試找出市場的知識規則。 為增進神經網路對於市場輪廓指標的學習效果,本研究提出價格指標及交易量指標之計算方法,分別探討只使用價格指標,以及加入交易量指標,分別建立預測模型,最後透過績效評估,以比較各模型間的獲利能力。 由實驗結果得知,加入交易量的模型,無論在短期間或是長期間預測,均能提升預測績效,尤其是在短時間的績效效果提升最為顯著。最後並加入單純使用隨機交易模型進行比較,發現市場輪廓指標,確實比隨機交易能有更高的準確度與獲利能力,因此驗證了市場理論廓的適用性。
Ministry of Finance in order to attract more investors into the futures market, the futures rate from four hundred thousandths adjusted to two hundred thousandths, is expected to be active stock market, increase the stock market momentum, based on past experience, as long as the futures rate cut, that is often a significant rise in trading volume, the effect is indeed obvious. Market profile to graphically describe the progress of price changes over time, the process developed by the analysis tools. This research proposes contours from the market profile, the extracted feature value, as back-propagation neural network input value, hope that through the physical strength of the futures market aggregate judge, try to find out the market knowledge of the rules. To enhance market profile indicators for neural network learning effects, this study proposes price indices and volume indicators calculation method, using only price indicators were discussed, as well as adding volume indicators, respectively, predictive modeling, and finally through the performance evaluation, in order to comparison between the profitability of each model. From the experimental results, the added volume of the model, both in the short or long period of time during the prediction, can enhance the prediction performance, especially in a short time to enhance the performance of the most significant effect. Finally, add a simple trading model using a random comparison, find the market profile indicators, indeed better than random transaction to have higher accuracy and profitability, so verify the theoretical profile of the market applicability.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079964515
http://hdl.handle.net/11536/72409
Appears in Collections:Thesis