類神經網路規則萃取-以台股期貨為例

Rule Extraction from Neural Networks – A Study for TAIEX Futures

吳媛媛(Yuan-Yuan Wu)

指導教授 ： 洪智力

中原大學/商學院/資訊管理研究所/碩士(2009年)

https://doi.org/10.6840/CYCU.2009.00844

摘要

本研究使用倒傳遞類神經網路預測台股期貨收盤價格，使用五種由專家統計結果得出的輸入變數：(一)KD指標，(二)前日外資現貨買賣超，(三)前晚Nasdaq漲跌幅，(四)隱含波動率，(五)當日開盤幅度。
用此五種輸入變數訓練完成的模型進行TREPAN規則萃取，解讀類神經網路模型的黑箱(Black Box)，萃取出可被使用者理解的規則，以提供實務交易更多的資訊。最後使用決策樹演算法，比較TREPAN與決策樹演算法兩者的準確率何著為佳。由實驗結果得知，TREPAN的準確率，較決策樹佳。

關鍵字

規則萃取

並列摘要

The study applied NN (Neural Networks ) to forecast the daily close price of TAIEX futures with five inputs chosen by experts:
1. the stochastic indicator.
2. daily trading volume of foreign investors.
3. last trading day’s change of Nasdaq.
4. implied volatility of TAIEX Options.
5. open price of TAIEX futures in the forecasted day.

In order to more deeply interpret the black box of NN, we also applied the algorithm of rule extraction named TREPAN which can provide users further information of the relationship between those inputs. Following this step, we then utilized the decision tree as a comparison. By comparing the results of TREPAN and the decision tree, we would evaluate which one has a better capability of forecasting. The result is that TREPAN is the better.

並列關鍵字

rule extraction ； Trepan

參考文獻

12.袁淑芳(2008)，台灣市場隱含波動率指標之資訊內涵探究，淡江大學財務金融學系博士論文。

16.陳宥佐(2006)，應用類神經網路預測電漿化學氣相沉積膜厚與參數調整建議系統，中原大學機械工程研究所碩士論文。

1.Black F., & Scholes M. (1973), The Pricing of Options and CorporateLiabilities. Journal of Political Economy, Vol.81, pp.637-659.

2.Chu, S. H., & Freund S. (1996), Volatility estimation for stock index options: GARCH approach. The Quarterly Review of Economics and Finance, Vol.36, pp.431-450.

3.Craven, M.W. (1996), Extracting comprehensible models from trained neural networks, Ph.D. Thesis, University of Wisconsin, Madison, WI.