本文利用包括不同因子的GARCH(1,1)-M模型,有系統地探討了台灣股價指數月、週及日報酬率之線性及非線性變動。在對稱模型中,GARCH現象隨著資料頻率的降低逐漸減弱,但季節性在月報酬上最爲顯著。殘差呈常態分配的假設明顯優於t分配的假設,應與我國股市的漲跌幅限制有關。對台灣股價指數的日報酬而言,包括交易量變數、季節效果及週天效果的不對稱GARCH(1,1)-M模型表現最優。日報酬均數的季節效果及週六效果基本上是穩定而顯著的,變異數的週一效果則會因爲模型納入交易量變數、或考量變異數的不對稱性而消失。又負的風險貼水始終與日報酬均數的不對稱性同時出現。最後,顯著的不對稱性透露,台灣股市有價格追漲的現象,日報酬的波動亦可能因股價下跌、投資人意見紛歧而加劇。
This study explores variations in mean and volatility of Taiwan's stockindex returns using assorted GARCH(1,1)-M models.Conditional heteroskedasticityweakens gradually as sampling intervals increase from daysto months in symmetric models.But seasonality appears most strongly inmonthly returns series.Normal distributions suit error terms better than tdistributions.Concerning the daily returns,an asymmetric model includingvolume and market anomaly variables has the best performance.Seasonaland Saturday effects are significant in the mean equation,whereas there isa trade-off between Monday effect and volume effect,as well as asymmetry,in the variance equation.Negative risk premium always accompanies meanasymmetry.In addition,traders in Taiwan's stock market tend to buy morewhen prices are increasing,and have relatively inconsistent opinions whenprices are decreasing.Daily returns thus become more volatile as prices aredecreasing.