本研究探討股市報酬不對稱特性，同時建立報酬分配的條件平均數、條件波動率、條件偏態參數及條件峰態參數動態模型，並且分別考量過去訊息對當前條件參數的不對稱效果。以八個國家股市為樣本，研究結果指出除了報酬分配的前兩項動差外，尚需考量高階動差的條件模型，才足以完整捕捉股市報酬實證特性。過去創新對各個條件動差的作用是不相同的，一般而言，過去創新對條件波動率具有明顯的不對稱效果，條件平均報酬的不對稱效果次之。另外，條件偏態及峰態參數明顯呈現時變性，而且在部份樣本中，它受到過去創新的影響而呈現不對稱效果。因此，本文支持報酬分配模型需將高階參數的動態過程予以考慮。另外，我們評估條件模型配適股市報酬的估計效能，以及應用於股市波動率預測的績效優劣。結果發現，考量較完整的條件動差模型並未比只考量條件波動率模型配適良好；但是若將股市報酬條件模型應用於報酬波動率之預測，則本文所提出的同時考量四項參數條件模型比其他未考量高階動差的模型具有良好的股市波動率預測能力。

This paper investigates the asymmetric effects of stock returns by constructing the dynamic models of the conditional mean, volatility, skewness and kurtosis parameters simultaneously. Using the samples from international stock markets, the empirical results suggest that the higher moments of the asset return should be added into the conditional model to capture the empirical features for the stock returns. In general, the impulses are different for the past innovation to the alternative conditional moments; it is more obvious that the asymmetric effects on the conditional volatilities are the strongest. In addition, the conditional skewness and kurtosis series apparently appear time-varying behaviors; they are also influenced by the past innovations in some stock markets. Finally, we evaluate the estimates efficacy of the alternatives conditional higher moment models and measure relative superiority of the competitive models applying on the volatility forecast for stock returns. The results suggest that it is not true for the complete conditional model on the estimates efficacy. However, on the issue of performance evaluations of volatility forecasts, the conditional higher moment model to be introduced in this paper outperforms others that are not included in higher moment settings.