Estimation and testing of regression disturbances based on modified likelihood functions

✍ Scribed by Mizan R. Laskar; Maxwell L. King

• Publisher: Elsevier Science
• Year: 1998
• Tongue: English
• Weight: 129 KB
• Volume: 71
• Category: Article
• ISSN: 0378-3758
• DOI: 10.1016/s0378-3758(98)00091-3

✦ Synopsis

This paper derives the conditional proÿle restricted likelihood (CPRL) function and likelihood ratio (LR), Lagrange multiplier (LM) and Wald tests of the parameters involved in the covariance matrix of linear regression disturbances based on di erent modiÿed likelihood functions. A Monte Carlo experiment was conducted to compare the small sample properties of maximum likelihood estimators and the new tests in the context of ÿrst-order moving-average (MA(1)) and ÿrst-order autoregressive (AR(1)) regression disturbances. The results show that the use of the modiÿed likelihood functions can reduce estimation bias and variation and yield estimates that are more normal than those from the concentrated likelihood. The small sample sizes of all the tests based on modiÿed likelihood functions are more accurate and their power curves are better centred compared to those based on their classical counterpart.