A Monte Carlo Study of AR (1) Estimators under Several Performance Criteria.
Abstract
The small sample performance of several AR(1) estimators is investigated through the use of Monte Carlo comparison studies. The performance of these estimators is compared with respect to the criteria of bias, mean squared error, mean absolute error, and mean squared prediction error. Statistical performance groupings at various fixed parameter values from (0,1) are determined based on pairwise multiple comparisons of estimator performance results. Two types of two-step adaptive estimators are developed. One type relies on the use of only standard estimators, while the other type includes ad hoc modifications to standard estimators. The efficacy of performance of these estimators is validated through the use of additional Monte Carlo runs based on three different conditions of parameter selection for data generation. The sensitivity of these estimators to their use with larger sample sizes is also investigated. Based on the various simulation results, recommendations regarding estimator selection for use in applied estimation are given. The applicability of the adaptive estimators is discussed and an example illustrating their application in forecasting an economic series is given. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1981
- Accession Number
- ADA097533
Entities
People
- H. S. Hill
- R. F. Ling
Organizations
- Clemson University