建構趨勢切割法與支撐向量迴歸於股票買賣時機之預測

A Trend Based Segmentation Method and the Support Vector Regression for Stock Trading Forecasting

潘伊芳(Yi-Fang Pan)

指導教授 ： 張百棧

元智大學/資訊學院/資訊管理學系/碩士(2012年)

https://doi.org/10.6838/YZU.2012.00088

摘要

股票市場是一個現代最普遍額外獲利的投資管道，但因為股市環境變化迅速，使投資人往往無法即時且準確的判斷買賣時機，而導致資金虧損，因此如何幫助投資者掌握買賣點且獲得穩定甚至更多的利潤顯得格外重要。過去許多研究中，多數使用線段切割法進行股價的買賣最適時機辨識，但對於預測模型的建立只基於漲跌趨勢的訓練。
本研究嘗試建立一個股市交易決策系統，其主要是利用趨勢切割法（Trend-Based Segmentation Method, TBSM）與支撐向量迴歸（Support Vector Regression, SVR）的結合，達成股票轉折點（買賣時機）之預測。本研究所提出之趨勢切割法有別於傳統線段切割法之判斷方式，考慮股價之特性包含上升、下降及持平趨勢，使得整體預測模型更加穩定及精準。支援向量迴歸用於預測模型的建立具有高穩定性及正確性，結合趨勢切割法的決策依據，所訓練之預測模型將可改善股票買賣時機的偏差問題，期望能達成高穩定且高投資利潤之股市交易決策系統。
根據實驗結果的投資獲利率顯示，本研究模擬數個美國股票的自動交易決策，本研究所提出之模型確實優於其他預測模型，其獲利狀況相當穩定。在發展技術上，本研究提出之趨勢切割法確實可提升預測準確性。在實務應用上，本研究所建立之股市交易決策系統可協助投資人做投資考量指標，以選擇買賣股票的適當時機，有效降低投資風險並提高報酬。

關鍵字

趨勢切割法 ； 支撐向量迴歸 ； 股票買賣時機預測

並列摘要

Stock market is a common investment way which it can obtain excess returns. The variation of stock environment is very quickly which investors cannot aright determine buy/sell timing. The important issue is how to help investors decide buy/sell timing and obtain steady investment returns. In past references, most of researchers used Piecewise Linear Representation (PLR) to analyze buy/sell points because they are only considered up-trend and down-trend.
In this paper, we will develop a stock trading decision system to recommend investors buy/sell timing which is using Trend-Based Segmentation Method (TBSM) and Support Vector Regression (SVRs). We propose a segmentation approach is called TBSM; it can analyze several kinds of trends based on price trend and different to traditional PLR approach. We consider three characteristics of price including up-trend, down-trend and hold-trend which can improve stability and accuracy for forecasting model. The forecasting models of SVRs have high performance in several forecasting problems. In this paper, we integrate TBSM and SVR to make trading decisions which can reduce investment risk and improve trade timing bias.
In experimental results, our proposed system has high excess returns on several U.S stocks with an automatic trading decision. However, the TBSM-SVR model outperforms other stock trading systems which are high profit rate and stability. In practical application, this stock trading decision system can helps investor easy to investment and capture high profit ratio. Moreover, it can easy show buy/sell suitable timing, reduce investment risk and increase profit.

並列關鍵字

Trend-Based Segmentation Method ； Support Vector Regression ； Buy/Sell timing forecasting

參考文獻

3. 張凱婷，「應用支撐向量迴歸及模糊規則於股價買賣點之預測」，元智大學，碩士論文，2011。

37. 賴志銘，「叢集式類神經網路在股價轉折點預測之應用」，元智大學，碩士論文，2008。

15. 李惠妍，「類神經網路與迴歸模式在台股指數期貨預測之研究」，國立成功大學，碩士論文，2003。

22. 李龍麟，「應用類神經網路於製程品質參數之即時監控-以印刷電路板之DES製程為例」，元智大學，碩士論文，2002。

21. 楊雯寧，「台灣股價指數預測模型之探討」，元智大學，碩士論文，2002。