應用資料探勘技術結合股票分析方法建構投資策略

Applying Data Mining Techniques in Stock Analysis to Construct Investment Strategies

葉俞佛(Yu-Fo Yeh)

指導教授 ： 李維平

中原大學/商學院/資訊管理研究所/碩士(2014年)

https://doi.org/10.6840/cycu201400505

摘要

股價指數對國家而言是個重要的經濟指標，因上市櫃公司的未來獲利能力與景氣指標會事先反應在股票價格上，而股票投資也是所有投資人最熱衷的金融活動之一，所有投資人皆想找出有效的交易方法，以降低風險並提高報酬率。
　　一般而言，股票分析方法分為基本分析及技術分析兩種，基本分析學派認為應利用財務分析和經濟學上的研究來評估企業價值或預測股票價格的走勢；技術分析學派利用過去金融市場的資訊來預測股票價格的趨勢與決定投資的策略，基本信仰建立在「歷史會不斷重演」，並試圖藉由大量的統計資料來預測行情走勢。兩派皆有其理論依據，亦各有相當多學術研究深入鑽研，但兩者兼備的則較為少有，本研究結合基本分析與技術分析，進行較為廣泛面向的研究。
　　本研究以基本面指標及技術面指標為研究變數，研究期間自2007年至2014年第一季，以臺灣50及中型100之成分股為研究樣本，依循巴菲特選股模式，以年度財務報表資料之基本面指標進行個股的篩選以建構投資組合，再利用倒傳遞類神經網路，針對投資組合的個股依技術面指標及建構股價預測模型，以判斷個股的買進時機。
　　實驗皆採用移動視窗法(moving windows)，巴菲特投資策略選股部分，每次以二年(八季)為單位，以前一年年報之基本面指標進行篩選，於次年四月至再次年三月進行績效測試，共計六次的循環測試。倒傳遞類神經網路預測部分，每次以九季為單位，以前四季進行倒傳遞類類神經網路訓練，第五季進行測試，後四季進行實際預估，總共亦六次的循環測試。僅採用巴菲特選股模式的投資組合，自2008年第二季至2014年第一季，共六年累積報酬率205.56%；加入倒傳遞類神經網路預測模型，累積報酬率可達261.89%。同時期台灣加權報酬指數報酬率28.35%、台灣50報酬率17.65%、中型100報酬率1.46%、高股息報酬率14.93%，2012績效最優基金報酬率68.41%，2013績效最優基金報酬率79.85%，皆落後於本研究，足可見本研究之系統策略對於投資獲利的提昇有相當的幫助。

關鍵字

基本分析 ； 倒傳遞類神經網路 ； 巴菲特投資策略 ； 技術分析 ； 投資組合

並列摘要

The index of the stock price is an important economic indicator to the country because the Listed companies’ future profitability and the profitability indicator will be reflected in the stock price. And the investment of the stock is one of the financial activities of the most enthusiastic of all investors. All of the investors are trying to look for an effective trade method to decrease the risk and raise the return rate.
　　In general, the method of the stock analysis is divided into two kinds, fundamental analysis and technical analysis. Fundamental analysis school think financial analysis and economic research should be used to assess the value of the enterprise or predict the movement of the stock price. The technical analysis schools use the past information in the financial market to predict the trend of the stock price and determine the strategy of the investment. The belief is on the basis of “History will repeat itself.” and a great deal of statistics data is adopted to predict the market trend. The two schools have their own basic theories and a lot of relevant researches, but the combination of both theories is not easy to see. The basic analysis and the technical analysis are combined in this study to proceed with an extensive research.
　　This study used the fundamental indicators and the technical indicators as the variables of the research. The research period ranged from 2007 to the first quarter in 2014. The study samples were the constituent stocks of TWSE Taiwan 50 Index and TSEC Taiwan Mid-Cap 100 Index. Following the Buffett's Investment Strategy Mode, the basic indicator of the annual financial information was used to screen the individual stock and construct the portfolio. And the back-propagation neural network was also used to determine when is the most suitable timing to buy stocks.
　　The experiments all used the method of “moving window”. In Buffett's investment strategy, “two years (eight quarters)” is used as a unit, and the fundamentals of the annual report of the previous year indicators were screened. Regarding “Back-propagation neural network”, nine quarters were used as the unit. The previous four quarters of back-propagation neural network was used to train. The fifth quarter was used to test, and the latter four quarters were to make an authentic prediction. There were six circular tests. Only Buffett's stock selection model was adopted as the investment portfolio. Since the second quarter of 2008 and the first quarter of 2014, the total accumulated return of six years reached 205.56%. When the back-propagation neural network forecasting model was added, the accumulative return rate reached up to 261.89 percent. While the follows were behind this study, TAIEX was 28.35% return; TWSE Taiwan 50 Index was 17.65% return; TSEC Taiwan Mid-Cap 100 Index was 1.46% return; Taiwan Dividend Plus was 14.93% return; the best performance Fund of 2012 was 68.41% return; the best performance Fund of 2013 was 79.85% return. As a result, we can know that the systematic strategy in this study is very helpful in the profitability of the investment.

並列關鍵字

Portfolio ； Buffett's investment strategy ； Back propagation neural network ； Technical analysis ； Fundamental analysis

參考文獻

14. 陳威愷，選股策略方法之研究比較-以台灣上市櫃公司為例，成功大學統計學研究所碩士論文，2012。

11. 周宗南、張輝鑫、黃祥穎，應用證據理論融合不同擇股策略模型以建立最佳化投資組合，財金論文叢刊，第十三期，p.59-73 2010。

12. 洪才元，結合基因演算類神經網路預測台股指數之模型，中原大學資訊管理研究所碩士論文，2008，

23. 董寶蘭，程式交易策略實證研究-以投資ETF0050為例，淡江大學管理科學研究所企業經營在職專班碩士論文，2010。

7. 李仁在，價量關係技術分析之實證研究-以台指五十成分股為例，淡江大學管理科學研究所企業經營在職專班碩士論文，2007。