應用金融工程物理學與市場輪廓理論於期貨市場行為之分析

Abstract

『掌握趨勢，擁抱財富』是金融資本市場中投資人努力追逐的目標，但要如何正確的判斷當前趨勢，至今亦沒有明確的方法學能夠充分解釋。眾多的專家學者嘗試利用統計學、投資學、會計學、財務工程等研究方法進行趨勢方向的預測，但往往建立在過多完美無瑕的假設上，以致在實務操作上窒礙難行。此外，伴隨商業活動全球化與金融貿易自由化的發展趨勢，國際資金的移動與相關的投資、套利及避險行為日趨頻繁，各國金融資本市場間的關聯性也日益顯著。金融資本市場資訊的來源日趨多元，因此掌握趨勢的首要工作即是掌握相關知識，盡可能地將資訊落差最小化，進而取得貼近實務市場的有效知識。 股市的交易過程中，依據Steidlmayer [7]所提出之市場輪廓理論，市場中存在新的買賣力、舊的買賣力、大戶買賣力、散戶買賣力、攻擊買賣力及防衛買賣力等諸多買賣力，其中又以攻擊與防衛買賣力的分析最為困難，但卻也是影響形成趨勢或區間運動之關鍵因素，多數成功的金融投資其真正的獲利都是來自於趨勢判斷正確的投資。因此本研究使用台灣期貨資料與美國股市資料，嘗試提出以市場輪廓理論、金融工程物理學與監督式類神經網路為基礎之台灣指數期貨市場走勢預測模型，從中分析市場參與者的投資行為。透過市場輪廓可以清楚描繪該交易區段的市場行為，藉由金融工程物理學的計算可更細膩的瞭解市場的趨勢動能變化，最後經由類神經網路進行各種不同趨勢動能樣本的學習。期望能更準確的掌握市場趨勢的運動方向，進而提升預測模型之準確率與獲利能力。 第一階段實驗結果顯示，加入市場輪廓價格偏離值與美國股市市場輪廓擺動因子等考量，的確可有效的提高準確率與獲利能力。在不同預測區間的比較，發現市場輪廓指標對於長區間的預測能力與獲利能力較佳。第二階段實驗嘗試將市場趨勢動能進行分群學習，試圖找出在具有攻擊防衛行為學習樣本下的市場未來趨勢，實驗結果顯示加入市場趨勢動能分群學習後，能顯著提升模型預測之準確率與獲利能力。

"Being able to identify the market trend and make profit" is the one and only goal that every investor is looking for. However, until today, there is still not a perfect theory that can lead us to making the most accurate prediction on market trend. Many experts tried to predict the market based on researches in statistics, investment, accounting, financial engineering, and others, but most of the predictions were made with ideal assumptions which were not practical in real market. Moreover, along with the commercial globalization and trade liberalization, international capital flows and investments have become more frequent than ever, and therefore, the correlation between financial markets has become increasingly significant. Nowadays the information we get from the markets is diverse, thus, our first priority is to identify which information really matters; filter out the irrelevant as much as possible and find the critical and practical information that can be used. According to Steidlmayer's market profile theory in stock trading, there are new buyers, old buyers, major buyers, individual buyers, offensive and defensive buyers and many other different participants; and among these participants, analysis and prediction on offensive and defensive behaviors are the most difficult ones to make. Despite its difficulty for analysis, offensive and defensive behaviors are often the most critical factors that make the market moves dynamically, and most of the successful investments are made based on making the right analysis and prediction on the trend driven by those factors. Therefore, by using the data of Taiwan Futures Exchange (TAIFEX) and US stock exchanges, this research is to build a prediction model by collaborating financial engineering theories with the Backpropagation Neural Network, and use this prediction model to make analysis on the market trend and the participant's behaviors. First, market profile theory clearly describes the market behavior in a certain period. Then, by using financial engineering physics calculations, we can get a more detailed understanding on how the dynamics has been shifting in the market. Lastly, we let the Backpropagation Neural Network to learn from different market trends. Our expectation is to get more accurate analysis on market trend, and increase the prediction model's accuracy and profitability. In the first phase of the experiment, the result showed that adding factors like market profile price deviation and US markets profile rotation factors was feasible to effectively increase the profitability and the accuracy on prediction; and by comparing the results of the experiments in different interval, we found the market profile theory had better predictability and profitability in the long term prediction. In the second phase of the experiment, we applied cluster analysis into our AI prediction model to identify market trends which were driven by offensive and defensive trading behaviors, and predict such market's future movement based on the analysis. The result showed that adding the mentioned application of cluster analysis into our prediction model could significantly increase the model's prediction accuracy and profitability.