期貨基金績效預測之探討 -以混沌現象與類神經網路分析

A Study of Commodity Trading Advisor Performance Prediction -An Analysis of Chaos and Artificial Neural Network

黃俊宏(Jyun-Hong Huang)

指導教授 ： 陳若暉

中原大學/商學院/企業管理研究所/碩士(2011年)

https://doi.org/10.6840/cycu201100290

摘要

全球金融市場的開放，導致投資商品多元的趨勢。期貨基金又稱商品交易顧問基金(commodity trading advisor, CTA)因為具有多空操作靈活以及分散投資組合風險的特性，深受國際投資人的喜愛。國外期貨基金早已行之有年，從1990年至2010年間全球的期貨基金市場淨值達到2,234億美元成長大約20倍。相較於其他投資產品，CTA績效表現卓越，從1980年至2009年的年化報酬率大部分皆為正數，在金融海嘯發生期間也擁有7.64%至14.09%的年化報酬率。隨著投資人對於期貨基金的需求升高，投資人對於期貨基金的價格型態及價格預測也會相對重視。
本研究利用BDS分析、R/S分析和相關維度分析法探討期貨基金價格型態是否具有混沌現象。在價格預測方面，使用適合非線性預測的倒傳遞類神經模型及調適型模糊類神經模型，並加入R/J CRB 指數、MSCI全球指數、Put/call ratio、貨幣供給量(M2)、LIBOR等五項自變數進行模型的預測效果比較，研究樣本分為系統交易型和非系統交易型兩種不同策略型態共20組樣本，研究期間為2005年1月至2010年12月。
研究結果顯示BDS分析的輸出值皆為顯著，且相關維度分析結果也進行收斂，表示期貨基金的報酬型態具有混沌現象，即擁有決定性的價格結構及可預測性。類神經網路預測結果顯示R/J CRB 指數、MSCI全球指數、Put/call ratio、貨幣供給量(M2)、LIBOR適用於期貨基金的價格預測。倒傳遞類神經網路進行期貨基金的預測效果較佳，且非系統交易型期貨基金表現相較於系統交易型期貨基金預測效果更優異。

關鍵字

倒傳遞類神經 ； 混沌現象 ； 期貨基金 ； 調適型模糊類神經

並列摘要

In recent years, the global investment market has many new derivative instruments to satisfy the investor’s diversity demand. The futures fund, also known as “commodity trading advisor, CTA”, having long and short flexible and diversified investment portfolio risk, attracted global investors to invest this product. In the foreign investment markets, futures fund was famous many years. From 1990 to 2010 global market value of the futures fund is 223 billion U.S. dollars to grow about 20 times. Comparing to other investment product performance, futures funds have excellent performance. From 1980 to 2009, the most of annualized return are all positive. It also has 7.64 to 14.09% annualized return during the financial crisis. With increasing demand of futures fund, the value and forecast of futures fund are important for investor.
This paper used Brock, Dechert and Scheinkman (BDS) test, Rescaled Range (R/S) analysis and Correlation dimension analysis to investigate chaos effect of futures fund., This study compared Back-propagation network (BPN) and Adaptive network-based fuzzy inference system (ANFIS) to determine which one is better in the price forecast part. And adding five index such as R/J CRB index, MSCI global index, Put/call ratio, M2, LIBOR in the forecast model examined whether the forecast model is suitable or not. The sample is divided into system and Discretionary strategies and study period is from January 2005 to December 2010.
The research result showed that the outputs of BDS test are all significant, and the outputs of Correlation dimension analysis are convergence, indicating that the price of futures fund have chaos. The price structure is decisive and predictable. Neural network prediction result shows that five index such as R/J CRB index, MSCI global index, Put/call ratio, M2, LIBOR are suitable to forecast the price of futures fund. And Back-propagation Network (BPN) also has good performance in price prediction. This paper compares two kinds of CTAs, and the result shows that the forecast performance of system type is better than Discretionary type.

並列關鍵字

ANFIS ； BPN ； CTA ； Chaos theory

參考文獻

5. 呂奇傑、李天行、高人龍、陳學群(2008)，結合獨立成份分析與類神經網路於財務時間序列預測模式之建構，交大管理學報，28(2)，187-216

9. 洪世章、黎正中、涂敏芳(2004)，以混沌觀點探討產業發展，管理學報，21(3)，391-410。

29. Peng, K. L. and Goo, Y. J. (2004), “Empirical study of chaotic behavior in the Taiwanese stock market”, Pan-Pacific Management Review, Vol. 7, No.1, pp.19-34.

2. 石麗娟(2007)，獨立成份分析法與類神經網路於股價時間序列分析與預測之研究，國立成功大學資訊管理研究所碩士論文。

4. 呂世傑(2004)，渾沌理論於美元匯率預測上之應用：以多度空間向量預測之研究，私立中原大學企業管理學系研究所碩士論文。