運用支援向量機與序列探勘建構財務預警模型—以台灣電子業為例

Using Support Vector Machine and Sequential Pattern Mining to Construct Financial Prediction Model- A Case of Electronic Industry in Taiwan

莊曜彰(Yao-Chang Chuang)

指導教授 ： 羅淑娟

國立臺北科技大學/管理學院/工業工程與管理研究所/碩士(2008年)

摘要

財務預警模型在財務上是一個重要的議題。對於投資者和顧問而言，可以利用預警模型提供的資訊做初步的決策，進一步，讓企業能及時預防財務危機。因此，本研究企圖建構一預警模型，以提供使用者初步的資訊。
在過去的預警文獻中，學者利用許多領域的方式建構表現優異的預警模型，學者也指出混合模型在建構預警模型上，有著極大的潛能。雖然許多文獻提供了優異的預警模型，卻鮮少文獻著重在模型分類值的分析。針對以往文獻不足之處，本研究結合二元序列演算法以及支援向量機做為混合模型，以二元序列演算法從支援向量機所產生的模型分類值中擷取隱藏資訊，並利用這些資訊協助支援向量機在後期的預測。
本研究分為兩階段來建構混合模型。第一階段是利用四種不同期間的資料來建構類神經模型，並從這四組模型中選取表現最佳的模型來產生模型分類值。在第二階段中，利用二元序列演算法從這些分類值中擷取出預測樣式，並計算預測樣式在預測上的準確率以及誤判率。
本研究建構的預警模型即使在缺少下一期資料的情形下，仍然可以提供初步的預測資訊，這改善了以往模型需要藉由資料才能提供資訊的缺點。此外，實驗結果指出，藉由合適的預測樣式，能提供較佳的預測結果。

關鍵字

預警模型 ； 混合模型 ； 支援向量機 ； 二元序列演算法 ； 預測樣式

並列摘要

Financial prediction model is an important economic issue. Financial prediction models can provide organization’s stakeholders and investors with preliminary information. Even prediction model could protect firms from bankrupt event in time.
Much literature has indicated the researchers employ various methodologies to construct excellent prediction models. Researchers also conclude that hybrid methodologies have great potential in prediction models. Although many studies have been investigated excellent prediction models, little attention has been paid to analyze binary signals. In addition, prediction models generate these binary signals.
The objective of this study is to use hybrid model to improve on insufficiency of previous prediction models. This study combines Support Vector Machine (SVM) with Binary Sequence Algorithm (BSA) to construct prediction model.
The process of constructing model was divided into two stages. In first stage, we selected the best model to generate binary signals of all firms. In second stage, BSA extracted prediction patterns from binary signals and assisted SVM in predicting next outcome of firms. Moreover, the accurate rates and misclassify rates were employed to evaluate the predictive capability of hybrid model.
This study analyzes the insufficiency of previous prediction models. As users lack for data of next period, previous models can’t provide users with information. However, this study uses patterns to assist SVM in prediction. In addition, the experiment results reported that using appropriate patterns could provide reliable prediction for users.

並列關鍵字

Financial prediction model ； Hybrid model ； Support Vector Machine ； Binary Sequence Algorithm ； Prediction patterns

參考文獻

[20] Law Source Retrieving System of Stock Exchange and Futures Trading (Taiwan R. O. C.), “Operating rules of the Taiwan Stock Exchange Corporation”, http://www.selaw.com.tw/Scripts/newsdetail.asp?no=G0100501.

[1] E. I. Altman, “Financial ratios, discrimnant analysis and the prediction of corporate bankruptcy,” The Journal of Finance, vol. 23, no. 4, 1968, pp. 589-609.

[3] A. Basheer and M. Hajmeer, “Artificial neural networks: fundamentals, computing, design, and application,” Journal of Microbiological Methods, vol. 43, 2000, pp. 3-31.

[4] C. -C. Chang and C.-J. Lin. “LIBSVM: A library for support vector machines, “ Technical Report, Department of Computer Science and Information Engineering, National Taiwan University, 2004.

Available at http://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf