Detecting dependencies in smooth regression models
平滑回归模型中的依赖性检测
A class of simple estimators for the correlation of the errors in nonparametric regression models is proposed for the fixed design case. These estimators are based on squared differences of various spans of the data and are consistent so long as the regression function is Lipschitz continuous. The limiting distribution is established and applied to derive asymptotic ‘global’ and ‘local’ tests for different null hypotheses of uncorrelatedness. The finite sample behaviour of…