中文摘要：

现有随机波动（SV）模型依赖于参数条件分布形式假设,无法充分描述金融资产收益的偏态厚尾等典型特点,而非参数分布能够更全面地刻画这些特性。本文将SV模型和非参数分布相结合,构建一类半参数SV模型;同时在贝叶斯框架内,发展有效MCMC抽样解决模型的参数估计难问题,并利用对数预测尾部得分（LPTS）法分析模型的极端风险预测能力;最后以我国美元/人民币汇率市场为例,对半参数SV模型在收益特性刻画以及极端风险预测方面的实际效果进行了检验。

英文摘要：

The existing SV model replies on the assumption of parametric conditional distribution, and can not describe the stylised facts such as skewness and fat-tallness of financial asset returns, but nonparametric distribution is able to more fully describe these facts. This paper aims to combine SV model and nonparamteric distribution to build one type semiparametric SV model, and develops an effective MCMC algorithm to solve the problem of parameter estimation; then proposes to use the LPTS to study model forecasting ability; lastly, we use USD/RMB exchange rate data to check the practical ability of semiparametric SV model in modeling the characteristics of financial returns and forecasting the extreme risk.