應用類神經網路於股票技術指標聚類與預測分析之研究

The Study on the Clustering and Prediction Analysis of Stock Indicators Based on the Neural Networks

紀勝財(Sheng-Chai Chi) ； 尤明偉(Ming-Wei Yu) ； 饒瑞倫(Jui-Lun Jao)

《科技管理學刊》 9卷2期 (2004/06) Pp. 53-81

摘要

本研究主要目的是探討有關股票技術指標(stock indicators)起伏對於台灣股價指數(Taiwan stock index)漲跌間的關聯性，以瞭解不同指標變動相互影響下會造成大盤指數走勢如何發展的關係。如此，投資者可以利用技術指標的改變來預測股價指數的變化。
本研究方法，首先應用自組織映射圖類神經網路(Self-Organizing Map neural network, SOM)，取特定期間內之各技術指標值為分析對象，將移動趨勢相似的指標聚類為同一分群。SOM是被公認簡單且經常應用於聚類分析(clustering analysis)的一種工具。而模糊類神經網路(Fuzzy Neural Network, FNN)的運用是透過網路訓練和刪減(network training and cutting)後，所留下來關聯強度較大指標，找出技術指標和股價指數間的模糊關聯規則(fuzzy association rules)。
為了評估自組織映射圖網路分類的效果，本研究利用灰色關聯分析(Grey Relationship Analysis, GRA)來驗證自組織映射圖網路分類的準確度，驗證結果證明自組織映射圖網路分類有相當不錯的準確度。為了進一步改善預測準確度，本研究由每一分群中篩選出各分群影響度高的指標作為整合預測模式的指標，研究結果顯示，整合各分群重要性指標的模式較未整合前各模式預測準確度高。

關鍵字

股票技術指標 ； 自組織映射圖網路 ； 灰關聯分析 ； 模糊類神經網路

並列摘要

The purpose of this research is to study the relationship of changes between the stock indicators and Taiwan stock index in order to understand how the trend of Taiwan stock index change is under the complex influence among the stock indicators. Such that, the investors can be easier to predict the index change in reference to the change of the indicators.
The proposed methodology, first of all, applies the Self-Organizing Map (SOM) neural network to cluster the similar indicators into groups based on their similarity of moving curve within a certain period of time. To investigate the relationship between the stock index and the technical indicators within any of the groups, the Fuzzy Neural Network (FNN) technique is employed to search for the rules about their relationships.
To evaluate the performance of the SOM, the Grey Relationship Analysis (GRA) is used for the verification of how similar of the indicators which was clustered into a group. According to the results, it is clear that the capability of the SOM in clustering is confirmed. To further improve the predication accuracy, this research selects some key indicators from each of the groups as the inputs of the united prediction network and the result completes much better prediction accuracy than all of the previous networks.

並列關鍵字

Stock Indicators ； Self-Organizing Map (SOM) ； Grey Relationship Analysis (GRA) ； Fuzzy Neural Network (FNN)

參考文獻

Azoff, E. M.(1994).Neural Network Time Series Forecasting of Financial Markets.John Wiley & Sons.

Chang, P. T.,Lee, E. S.(1983).The Estimation of Normalized Fuzzy Weights.Computer & Mathematics with Applications.11,229-241.

Jang, G. S.,Lai, F.,Parng, T. M.(1993).Intelligent Stock Trading Decision Support System Using Dual Adaptive-structure Neural Networks.Journal of Information Science and Engineering.9(2),271-297.

Kaufmann, A.,Gupta, M. M.(1991).Introduction to Fuzzy Arithmetic Theory and Application.Van Nostrand Reinhokd.

Kohonen, T.(1988).Self-Organization and Associative Memory.Berlin:Springer-Verlag.