A MONTE CARLO STUDY OF THE REGRESSION MODEL WITH AUTOCORRELATED DISTURBANCES
RAND CORP SANTA MONICA CA
Pagination or Media Count:
The paper gives a description of the relative performance of estimators based on the results of a Monte Carlo experiment, under the assumption that disturbances are generated by a first-order autoregressive process. To generate artifical data for the experiment, eight structures were specified. For each structure, 300 samples were drawn and estimates of unknown parameters were calculated for each sample by five different methods, namely, maximum likelihood, Theil-Nager, approximate Bayes, Durbin, and least squares estimators. The task was first to examine the performance of the various estimators and second, to check the behavior of several commonly used tests of independence regression analysis. Characteristics of the various structures were chosen to represent a variety of circumstances that might be reasonably encountered in practical work.
- Statistics and Probability