An Asymptotically Pivotal Transform of the Residuals Sample Autocorrelations With Application to Model Checking

摘要: We propose an asymptotically distribution-free transform of the sample autocorrelations residuals in general parametric time series models, possibly nonlinear variables. The autocorrelation function is basic model checking tool analysis, but it not useful when its distribution incorrectly approximated because effects parameter estimation and/or higher-order serial dependence have been taken into account. limiting may be difficult to derive, particularly underlying innovations are uncorrelated independent. In contrast, our proposal easily implemented fairly contexts and resulting transformed distributed as independent standard normals uncorrelated, providing intuitive device for presence estimated parameters. also discuss detail alternatives ...