在股市操作裡,能夠買低賣高是萬千股票投資人夢寐以求的事,但是股票短線的漲跌已被證明不可預測,聽信各種股市技術分析的效果也有限,而現在學術界所採用的各種統計學方法或機器學習模型,都是在運用歷史數據探討各個影響因子之間的關係且成效良好,但是在預測股價或是股市未來走勢時亦因缺乏未來的輸入數據而困難重重。
以下簡述研究目的、方法、期間及結論:
研究目的:發展短線股價預測操作模型,提高當日沖銷的獲利機會。研究方法:係以使用超調量(Overshoot)的近似值作為某些股票走勢的參考依據,首先假設開盤走高的股票走勢與超調量模型相似,並運用相關性分析作為篩選股票的方法;運用超調量模型與這些經篩選的股票作迴歸分析後計算股價的買點和賣點,研究樣本(實際測試期間):鑒於本模型為預測模型,非一般統計模型,使用的數據均是即時的股票漲跌數據,而評估參數是淨利潤,故沒有抽樣樣本數、信心水準等等的統計參數,實際於 2016 年 3 月底及 4 月初於台股實際操作, 共測試 8 個交易日,實際淨利率為正值,表示本模型具有實際操作的潛力,值得繼續修正測試。
研究結論:運用超調量模型作為股票當日沖銷的工具經過測試每日獲利均為正數,雖獲利甚微,但卻是可以實際上線操作的模型,再經過參數調整設定應有提高潛在獲利
的潛力。

The optimal operating strategy of security trading for numerous investors is ‘’buy low and sell high’’ that is almost impossible to practice in real world. The fluctuations of stock prices
are proven to be stochastic and unpredictable. Most investors use the analysis of different organizations as the references for decision makings. However, the performance of most analysis approaches in security trading markets is also unsatisfactory. In addition to the analysts of security trading markets, most statistical methods and machine learning models in academic fields use historical data to investigate the relation between different factors and valuables i.e. stock prices. Although the performance of these models was outstanding, these models cannot be applied to real time security trading given the lack of the future input data. In order to overcome these shortcomings, this thesis aims to develop a model to increase the profit of short-term transactions in security trading markets. The aim,
methodology, testing samples and conclusion are outlines as follows:
Aim: Develop a short-term stock prices prediction model that enables investors to increase the profit of their day trading.
Methodology: Use an overshoot model which comprises a set of strategies of screening stocks, calculating buying and selling prices.
Testing samples: This model was applied to the real time security trading in Taiwan on 8 days in the end of March and early April, 2016.
Conclusion: Given the net profit made by this model was all non-negative on each day, this model is relatively capable to operate in security trading markets. However, the parameters of the screening and pricing require further adjustment to increase its potential.

中文部分
吳宗正, 2004, 投資技術分析 (第七版), 華泰文化出版社.
李育叡, 2012, 應用倒傳遞類神經網路於台灣 50 指數預測之最佳化模式分析, 國立中興大學碩士論文.
阮榆方 & 陳安斌, 2006, 應用分類元-類神經網路模型於台灣加權指數趨勢預測之研究, 國立交通大學碩士論文.
林惠玲 & 陳正倉, 2002, 應用統計學二版, 雙葉書廊.
林惠玲 & 陳正倉, 2004, 統計學: 方法與應用, 雙葉書廊.
姚伊蔚, 2015, 波動性預測與未來股價預測比較-以臺指選擇權為例, 臺灣大學財務金融學研究所碩士論文.
洪才元, 2009, 結合基因演算類神經網路預測台股指數之模型, 中原大學碩士論文.
洪安裕, 陳安斌, & 林富松, 2004, 以知識規則與類神經網路之整合對台灣加權股價指數收盤走勢預測之行為研究, 國立交通大學博士論文.
張百棧, 王彥文, & 楊雯寧, 2004, 混合式預測模式之發展─ 預測台灣股價指數之波動, Journal of the Chinese Institute of Industrial Engineers, 21(4), 358-368.
陳志龍, 2006, 運用類神經網路與技術指標預測股票型基金漲跌及交易時機之研究-以臺灣 50 指數股票型基金為例, 朝陽科技大學資訊管理系碩士論文.
黃登源, 1998, 應用迴歸分析: 華泰文化出版社.
葉俞佛, 2014, 應用資料探勘技術結合股票分析方法建構投資策略, 中原大學碩士論文.
詹玄維, 2014, 歷史期數在股價預測上的成效比較.
 
英文部分
Bishop, C. M. (2006). Pattern recognition and machine learning: springer.
Malkiel, B. G. (2004). Models of stock market predictability. Journal of Financial Research, 27(4), 449-459.
Ogata, K. (1995). Discrete-time control systems (Vol. 2): Prentice Hall Englewood Cliffs, NJ.