中文摘要：

现有基于参数模型构造的条件异方差检验往往存在模型设定偏误问题。为了避免模型误设对检验结果的影响，并且捕获多种条件异方差现象，本文基于非参数回归构造了不依赖于特定模型形式的条件异方差检验统计量。该统计量可视作条件方差和无条件方差之间差异的加权平均，在原假设成立时渐近服从标准正态分布。数值模拟结果表明本文统计量具有良好的有限样本性质，也说明条件均值模型误设会导致错误地拒绝条件同方差的原假设，凸显了本文引入非参数方法构造条件异方差检验的必要性。实证分析采用本文统计量探讨了国际主要股指收益率的条件异方差现象，得到了与Engle（1982）不同的检验结果，可能意味着股指收益率呈现出非线性动态特征。

英文摘要：

The existing parametric specification based test for conditional heteroskedasticity generally suffer from model misspecification problem. To avoid this problem and capture various form of conditional heteroskedasticity, this paper proposes a model free test for conditional heteroskedasticity. The test statistic could be regarded as the weighted distance between conditional and unconditional variance, and has a convenient asymptotic standard normal distribution under the null hypothesis. Monte Carlo studies not only demonstrate the well behavior of our test in finite samples, but also show that the misspecification of conditional mean model could lead to the false rejection of the conditional homoskedasticity, which highlights the necessity of introducing the nonparametric conditional mean model. In an application to test conditional heteroskedasticity in stock returns, we obtain some results which differ from Engle＇ s （1982） test, indicating the nonlinear dynamics of stock returns.