整合隱含波動率與技術指標於台灣加權指數的行為分析--應用類神經網路

Abstract

過去許多文獻均對於股票報酬的可預測性多所著墨，並對標的物進行各種投資策略的研究，眾多分析股票方法中，技術分析是相當受到歡迎的其中一種，此外亦有人提出隱含波動率包含人對股價走勢的預期心理。

鑒於上述，本研究運用隱含波動率與標的物之間探討是否有存在的關係，對標的物之價格進行漲跌趨勢預測動作；並利用型態學之技術指標-指數平滑異同移動平均線(MACD)，進行標的物趨勢判定，並利用其一階動量進行分類，定義為下跌趨勢、盤跌趨勢、盤整趨勢以及上漲趨勢，進而針對四種時期進行一、二、五、十日之股價趨勢漲跌預測。本研究定義輸入變數為隱含波動率買賣權差值（由買權主要序列的隱含波動率和賣權主要序列的隱含波動率構成）。經由類神經網路的學習後；研究結果指出，不論是預測準確率或是年獲利率的績效上，系統於十日股價趨勢預測上有顯著的效益；此外研究並發現下跌趨勢亦顯著優異於上漲趨勢的預測績效。

實證結果亦顯示，本系統不論於買賣訊號制定亦或漲跌趨勢判定均顯著優於未分群結果和隨機買賣。本研究亦證實隱含波動率之差值應用在股價漲跌趨勢分析上確有其意義，實能提供投資人觀察選擇權市場與現貨市場間關聯性的預測指標。

In recent years, many literatures have discussed that whether or not stock price is predictable, and researchers use all kinds of investing strategies to optimize the investment in stock market. Among the methods of calculating stock price, Technical Analysis is a popular method in forecasting stock price. Moreover, some research also claimed that implied volatility reflects the investors’ expectation for the trend of the stock price.

To verify the aforementioned claim, this study would probe the fluctuation of Taiwan Stock Market Index and implied volatility and determine whether the relations really exist. And then try to predict the trend of stock price with the relations. Technical indicators of moving average convergence-divergence are taken as input factors to predict next day's stock price trend. The input factor’s value of variance would be classified into four periods. The four periods were defined as falling substantially, falling slightly, backing and filling, and rising. In the four periods, input factors include the differences between implied volatilities calculated from the stock option put and call main series to predict the daily closing price and the closing price of one, two, five, and ten day(s) later. In the present, Neural Network is a novel artificial intelligence methodology which is widely applied to solve complex financial problems. Through the training of neural network, the result had shown that the accuracy of prediction and rate of annual profit were positive in terms of 10-day forecast. At the same time, prediction of falling trend was substantially better than the prediction of rising trend.

The result of this experiment also shows that Neural Network is significantly more effective than random walk model and unclassified model. In this study, the use of the differences between implied volatilities in predicting the trend of stock price is meaningful. The methods used in this study provide investors a tool for analyzing the stock market and the option market.