本文主要是探討台灣市場長期資產報酬率的研究,參考目前美國精算學會(AAA)所建議之動態資產模型,例如:the independent lognormal模式、the ARCH-type模式、the regime switching log normal(RSLN)模式、對數隨機波動模式(SLV)等較為長期之模型,做模型適合度分析,找出適合描述台灣市場的資產模型。在台灣股票市場的實證研究中,使用最大概似估計法(MLE)與有效動差法(EMM)估計長期資產隨機模型之參數,並比較概似函數、AIC(Akaike Information Criterion)或SBC(Schwartz Bayesian Criterion),利用常態機率分布圖檢驗各模型左右尾分配是否合適,最後再根據最低累積給付保證商品的例子來估計動態資產模型之條件尾端期望值的風險。實證結果顯示,GARCH和RSLN模型較能描述台灣市場特性的情況。
This paper mainly investigates long term equity return in the Taiwan stock market. We attempted to find the optimal model among and perform diagnostic analysis of the stochastic asset models, such as the independent lognormal model, the ARCH-type model, the Regime-Switching Lognormal model, and the Stochastic Log-Volatility model, all of which have been proposed by the American Academy of Actuaries. In addition, the most suitable long term equity model was determined. In the empirical study of the Taiwan stock market, the parameter estimations were carried out using maximum likelihood estimate and efficient moment of method. Following, we compared log-likelihood, Akaike Information Criterion and Schwartz Bayesian Criterion and drew Quantile-Quantile plots to test tails, to determine the appropriateness of models. Finally, Conditional Tail Expectation risk measures of stochastic equity models were estimated from simulation of the liability of a guaranteed minimum accumulation benefit contract. The results indicate that the generalized autoregressive conditional heteroskedastic and regime switching log normal models perform much better than other considered models for analyzing the Taiwan financial market.