A Heteroskedasticity Test Robust to Conditional Mean Misspecification
提出一种新的异方差检验统计量，无需指定均值回归函数的具体形式，而是用核估计方法非参数地估计回归函数，得到的残差对模型误设具有稳健性，检验统计量渐近服从卡方分布。
This paper proposes a new test statistic to deter the presence of heteroskedasticity. The proposed test does not require a parametric specification of the mean regression function in the first stage regression. The regression function is estimated nonparametrically by the kernel estimation method. The nonparametric residual is estimated and used as a proxy for the random disturbance term. This nonparametric residual is robust to regression function misspecification. Asymptotic normality is established using extensions of classical U-statistic theorems. The test statistic is computed using the nonparametric quantities, but the resulting inference has a standard chi-square distribution. Copyright 1992 by The Econometric Society.