Forecasting correlation among equity mutual funds in Taiwan

[[abstract]]投資人常購買各式各樣受積極管理且投資於特定類型資產的基金以規避風險，而O’Neal(1997)的研究發現，多重基金之投資組合所含的風險遠低於持有單一基金之投資組合；此外，Sharpe(1992)亦指出，適當的基金風格分類將有助於投資人有效的風險分散。因此，為達此目的，投資人必須能對於共同基金報酬間的相互關聯作精確預測。投資人在事前正確估計相關性的能力將會決定其投資組合中最佳組成份子的選擇。 本研究以三種不同類型的模型－歷史模型、均數模型及指數模型，估計共同基金未來的相關性，並比較各模型之預測績效。希望能找出最適應用於我國共同基金之相關性預測模型，以期能對投資人於管理共同基金投資組合時有所助益。由於許多實證結果均顯示具有某些共同特徵因子的股票在歷經不同的經濟情況及景氣循環後，仍有相似的報酬；再者，投資風格亦是影響基金報酬的重要因素。故本研究在指數模型中模型之選取方面，除了常見的市場模式以外，亦包含風格指數模型、多重風格指數模型、Fama-French三因子模型及Ferson and Warther(1996)提出之動態模型；此外，本研究另依Buetow, Johnson and Runkle(2000)之觀點，設計符合我國股市交易特性的產業多重風格指數模型，並將其納入前述預測模型中予以一併探討。 本研究實證結果顯示，以電子/非電子及規模或成長/價值及規模差異作為解釋變數之多重風格指數模型及Fama-French三因子模型，不只是最佳預測模型，而且在相同及不同風格類型之基金子樣本均一致，表示其預測績效並不受共同基金風格類別型態影響，具有良好的穩定性；而動態模型亦有不錯的預測績效穩定性，但其在事後有良好產生歷史相關矩陣能力的均數模型在測試期之預測績效卻不佳；此外，在事後績效表現較差的單一指數模型在測試期之預測能力亦不佳。整體而言，多數模型之預測能力較歷史模型為佳。[[abstract]]For diversification considerations, investors often buy various mutual funds that are actively managed with a focus on specific market segments. Recent research by O’Neal(1997) shows that a multi-fund portfolio is far less risky than its single-fund counterpart. Moreover, Sharpe(1962) shows that appropriate style classification will enable investors to diversify effectively. To successfully implement a strategy of diversification investors must obtain accurate estimates of correlation among mutual fund returns. An investor’s ability to produce accurate ex-ante estimates of correlations will determine the optimal choice of securities in their portfolio. This paper forecasts mutual fund correlations using three major model categories — historical, mean, and index models. We evaluate the performance of each model in forecasting correlations among equity mutual funds in Taiwan to provide useful information for investors to manage a portfolio of mutual funds. Many studies have shown that stocks with some common characteristics tend to perform similarly over several economic and market cycles. Furthermore, the investment style is also an important factor that may affect fund returns. Therefore, the index models using by this paper, including the market model, style index model, multi-style index model, Fama-French 3-factor model and dynamic model used by Ferson and Warther(1996). Moreover, based on the findings of Buetow, Johnson and Runkle(2000), we use a industry-based style index model to capture the characteristics of trading behavior in Taiwan’s stock market. Results show that the estimates of future correlations from the industry-based style index model, multi-style index model and Fama-French 3-factor model have the lowest prediction errors. Meanwhile the performance of these models is very consistent in sub-samples stratified by style. The dynamic model also performs well in forecasting future correlations among mutual funds. Furthermore, the single index model does not exhibit fair performance in the either ex ant or ex post. In short, most of the models examined by this paper perform better than the historical model