الفهرس 
Robust Estimation of Integer-Valued Time Series Models
AbstractOnly 14 pages are availabe for public view
 |  | from | 118 |  |  |
| | | from | 118 | | |
Abstract
This work extends a robust estimation method for first order integer-valued
autoregressive models with Poisson innovations to integer-valued
autoregressive moving average models of arbitrary order. It uses a Monte
Carlo simulation to investigate the performance of the extensions relative to
the traditional estimation methods of Yule-Walker, conditional least squares
and conditional maximum likelihood under a variety of design conditions.
Overall, the work concludes that the extensions provide significant
improvement in performance if the data is contaminated with additive outliers.
If the data is contaminated with innovation outliers, conditional least squares
appears to be more suitable for estimation of the autoregressive and moving
average coefficients while the extensions perform better for the estimation of
other parameters. However, the improvement in performance might not be
enough for some applications. In such cases, we suggest that the extensions be
used as part of more intricate estimation procedures.