scientific article; zbMATH DE number 4005394
The aim of this paper is to discuss and illustrate the use of marginal likelihood for testing for serial correlation in regression models. We consider a 'classical' model with first order autocorrelated errors. In that context a widely applied procedure for inference about the autoregressive parameter is to replace the regression coefficients and the error scale parameter in the likelihood by their maximum likelihood (ML) estimate and examine the resulting profile as a function of the parameter of interest. This procedure leads to a biased ML estimate of the AR(1) parameter in small samples and to a poor power of the likelihood ratio (LR) test. For purpose of inference, we propose a LR statistic constructed from marginal likelihood. This method (that is often indicated as modified or restricted likelihood) allows for the effect of nuisance parameters by using only that part of information in the data which can not be affected by ignorance of the regression. We discuss the procedure in details, giving the general expression of the marginal likelihood for the regression model with AR(1) error structure. Also, we show that for a given value of the AR parameter the marginal deviance differs from the profile deviance by a correcting factor, which depends on the regressors and is asymptotically negligible. This factor contributes to reduce the small sample bias in the estimate of the error model parameter and leads to improvements in the sampling properties of the LR statistic. The comparison between the 'profile' and 'marginal' LR test has been performed by simulation. This has focussed on a model with a constant term and a trend-like regressor. The test statistic that we propose is very close to a \(\chi^ 2_ 1\) and shows excellent power with respect to the upper envelope power function, computed for the specific model. In a further study these results have been confirmed for different choice of regressors.
- Improved inference for first-order autocorrelation using likelihood analysis
- Marginal likelihood and unit roots
- Testing for Serial Correlation Against an ARMA(1, 1,) Process
- scientific article; zbMATH DE number 3921780
- A small-sample correction for testing for \(g\)th-order serial correlation with artificial regressions
- Marginal likelihood for parallel series
- Estimation and testing of regression disturbances based on modified likelihood functions
- Improved inference for first-order autocorrelation using likelihood analysis
- scientific article; zbMATH DE number 3921780 (Why is no real title available?)
- The likelihood function and a test for serial correlation in a disequilibrium market model
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3756363)