看到有同学讨论xtscc时说到xtscc相对于xtgls能够更好的解决面板数据的异方差和时序自相关，并提到
“事实上，无论是 xtgls, 还是 xtpcse 命令都没有考虑个体效果，他们对截面异质性的处理都是通过 OLS 估计得到的残差进行了，也就是采用OLS估计的残差估得稳健型方差-协方差矩阵。更为重要的是，这两个命令事实上更适合大T小N型面板数据，对于分析上市公司的同志而言，并不适用。”

后来查看了一下stata的手册，其中对xtscc这样描述“xtscc produces Driscoll and Kraay (1998) standard errors for coefficients
    estimated by pooled OLS/WLS or fixed-effects (within) regression. depvar
    is the dependent variable and varlist is an (optional) list of
    explanatory variables.
    The error structure is assumed to be heteroskedastic, autocorrelated up
    to some lag, and possibly correlated between the groups (panels). These
    standard errors are robust to very general forms of cross-sectional
    ("spatial") and temporal dependence when the time dimension becomes
    large. However, because this nonparametric technique of estimating
    standard errors does not place any restrictions on the limiting behavior
    of the number of panels, the size of the cross-sectional dimension in
    finite samples does not constitute a constraint on feasibility - even if
    the number of panels is much larger than T.  Nevertheless, because the
    estimator is based on an asymptotic theory one should be somewhat
    cautious with applying this estimator to panel datasets with a large
    number of groups that have only a short number of observations.
也就是说，由于基于依概率渐进理论，在N远大于T的时候，使用xtscc必须谨慎。
不知我的理解对否，与大家讨论。