應用支向量機於公司初次公開發行股價之預測

Application of support vector machine to initial public offerings prediction

葉麗鵑(Li-Chuan Yeh)

國立虎尾科技大學/管理學院/資訊管理研究所/碩士(2006年)

摘要

公司初次公開發行股價之預測(initial public offerings prediction)一直以來都受到學術及業界所關心的議題。對此問題的分析方法也從會計基礎評價模式及統計線性迴歸，一直到人工智慧工具中的類神經網路等方法。本研究將提供新的分析方法，利用支向量機迴歸(support vector regression; SVR)對此問題重新檢視。
支向量機為近年來機器學習領域中最受矚目的方法之一，於許多領域之應用上像是生物資訊、文件的分類、影像辨識，其表現可與類神經網路和決策樹等主流方法相匹敵。本論文中我們將探究應用支向量機於公司初次公開發行股價之預測上，實證將建立三個模型，分別為預測七天後IPO折價幅度、十四天後IPO折價幅度、以及三十天後IPO折價幅度。為了使實驗中的模型更具強健性(robustness)，我們先使用格子點演算法配合5-fold交叉驗證法於訓練資料集上，以求得各種不同參數組合下的誤差，再觀察其訓練誤差及測試誤差的變化作為參數選取機制，來篩選出最佳的參數組合，最後再利用此最佳的參數組合來建構實際的支向量機迴歸預測模型。並且，本研究以敏感度分析來探討不當的參數組合將使模型容易陷於過度適配(over-fitting)或不足適配(under-fitting)的危機中。
最後經由模型績效分析後顯示支向量機迴歸確實優於類神經網路模型的預測能力，綜而言之，支向量機能有效應用於公司初次公開發行股價議題上，可提供管理者或投資者一個參考的方向。

並列摘要

Initial public offerings prediction has always been an issue concerned by academia and Industry. The analysis method is form Accounting-base Valuation and Statistics linear regression to neural network in artificial intelligence trick for this topic. In this study we will offer new analytical method and use SVR to respect this topic.
Support vector machine (SVM) is a popular method of machine learning. Support vector machine has been applied to biotechnology, text categorization and image recognition. Support vector machine has good performance like decision tree and artificial neural network. In this study we use SVR to rediscover initial public offerings prediction. And the experiment will construct three models to forecast the IPO price after the seven days, the fourteen days, and the thirty days respectively. For build stable and reliable prediction model, we use Grid Algorithm and 5-fold cross-validation technique to build the models with train data, and to get the prediction errors of the every different parameter set( ). And then we observe the difference between train error and test error to be parameters selection mechanism and find out the optimal parameter set. Finally, we use the optimal parameter set to construct real support vector regression prediction model. Moreover, this study implements sensitivity analysis technique, the analysis demonstrates that incorrectly selected parameters will lead the model’s results in the risk of over-fitting, or under-fitting.
Finally, the experiment shows that SVR forecast ability significantly better than BP neural network from the model’s Performance analysis. In a word, support vector machine can efficiently apply in initial public offerings prediction, and provide administrator or investors a direction to refer.

參考文獻

15、黃敏菁，支援向量機在財務時間序列預測之應用，輔仁大學金融研究所，民國94年。

18、胡德中、馬黛，由承銷價低估、市場高估與股權結構觀點分析最適釋股策略

19、洪振虔、吳欽杉、陳安琳，非理性投資行為對新上市股票價路績效之影響，

20、馬黛、胡德中，承銷配售機制之決定及其對IPO折價之影響：競價拍賣、詢價圈購與公開申購，財務金融學刊，民國92年四月，第十一卷第一期。

4、陳寬裕，應用支援向量迴歸於國際旅遊需求之預測，長榮大學經營管理研究所博士班，民國93年。