智慧型日交易量建模與預測於台灣股票與期貨市場之研究

Abstract

價量關係與預測價格趨勢一直是學界與實務界所關切的議題，關於金融市場的研究大多數都在價格發現中求突破，或是探討交易量與價格之間的關係，鮮少針對交易量單獨做研究。因此本研究旨在探討日交易量建模與預測於台灣股票與期貨市場。本研究主要分為兩階段，第一階段為交易量預測模型實驗，目的是找出三種類神經網路方法與六種相異迴歸方法中預測誤差較低者，並與簡單統計線性迴歸做比較。第二階段為分群建模實驗，目的是探討三種依特徵因子分群後的預測模型之預測準確率是否比第一階段之未分群模型來的高。 實驗結果得出第一階段交易量預測模型實驗之預測誤差較簡單統計線性迴歸低，因此得證本研究之假設一：運用人工智慧與機器學習方法建模的模型比簡單的傳統線性迴歸模型預測準確率高。此外，第二階段實驗結果得出三種分群預測模型之預測準確率皆比未分群模型來的高，因此得證本研究之假設二：使用單一模型無法完全預測多變的金融市場，即日交易量的預測準確率將受限於歷史資料中的多數經驗，故用不同的因子將資料分群並建構各種市況下的多個預測模型。

The relationship between trading price and trading volume had been widely discussed by researchers and financial market practitioners. Traditional financial decision support system mostly focused on the market trend prediction and empirical tests between price and volume. This paper introduced an artificial intelligence system to model and predict the intra-daily trading volume in the Taiwan Stock and Futures Markets. We had implemented three different kinds of artificial neural networks and six regression models as the prediction kernels to substitute for simple linear regression model. In addition to the use of machine learning techniques, the clustering idea had also been applied to further improve the system performance. We found that the proposed trading volume forecasting model may outperform traditional approaches. The major contribution of this paper is to prove that the artificial intelligence and machine learning methods can represent the intra-daily volume changes better than simple linear regression. Moreover, the experimental results also show that the prediction can be further improved by three kinds of clustering model. It means that the adaptive model selection is required in this application to fit the complex and variant financial historical data. In summary, this paper proposed an effective trading volume prediction system based on various intelligent regression methods and clustering model.