The robustness, reliabiligy and power of heteroskedasticity tests
研究了线性回归模型中多种异方差检验方法，发现渐近临界值在有限样本下不可靠，而自助法可解决此问题，并推荐了修正的Glejser和Koenker检验。
Several tests for heteroskedasticity in linear regression models are examined. Asymptoticrobustness to heterokurticity, nonnormality and skewness is discussed. The finite sample eliability of asymptotically valid tests is investigated using Monte Carlo experiments. It is found that asymptotic critical values cannot, in general. be relied upon to give good agreement between nominal and actual finite sample significance levels. The use of the bootstrap overcomes this problem for general approaches that lead to asymptotically pivotal test statistics. Power comparisons are made for bootstrap tests and modified Glejser and Koenker tests are recommended.