分析能源類股股價影響因子與預測漲跌

Explore Stock Price by Applying Fuzzy Clustering Method and Support Vector Machine

股票是台灣投資人熱愛的投資工具之一，根據台灣證券交易所統計，110年12月股票交易市場週轉率為11.16次，高於日本（8.88次）與新加坡（3.17次）等亞洲國家。由此可發現台灣投資人偏向短期投資型態，因此如何在找出影響股價因子並有準確的預測股價趨勢是投資人重要的課題。本研究目的在於找出股價影響因子組合，依據資料特徵值進行資料分群，探討各影響因子與股價變動關係。本研究資料為1323筆能源類別股資料，時間為2014年1月至2015年9月，公司樣本數63家，使用Cart找出影響股價因子組合，將訓練資料使用Fuzzy C-means依據特徵值進行分群，根據不同群聚資料建置不同持向量機預測模型後，將測試資料計算所屬群聚，導入各自所對應支持向量機預測器後並收集預測結果。支持向量機預測結果與實際輸出結果相比，本研究模型預測準確率達71.28%，並根據預測結果進行模擬投資，平均月報酬率為1.77%，年化報酬率為21.24%%，優於其他方法。

According to the statistics of securities market published by Taiwan Stock Exchange Corporation (TWSE), in Dec.2021 the volume turnover rate traded in TWSE was 11.16, which was much more than Hong Kong and Singapore. Investors in Taiwan tend to adopt short term investment strategy. How to discover factors influencing stock price and further create a model to predict stock prices is the important concern of investors. The study collected data from listed companies of energies related industry in TWSE from January 2014 to September 2015 which accounts to 63 companies and total number of data is 1,323. First of all, the classification and regression tree (CART) method was adopted to filter the most effective factors related to stock prices. After using a combination of factors to identify the impact of stock price, Fuzzy C-means clustering was used to cluster training data based on feature values. Then support vector machine (SVM) was applied to build different prediction models. Compare the results from SVM prediction with actual output, the model shows that 283 of 397 test record transactions are accurately predicted. The prediction accuracy rate of this study is 71.28%, the average monthly return rate is 1.77 %, and annual rate of return is 21.24%t the method by this study is much better than other methods.