Advances and Applications in Statistics
Volume 19, Issue 2, Pages 113 - 148
(December 2010)
|
|
BENCHMARKING USING WORKING ERROR-MODELS
Zhao-Guo Chen and Ka Ho Wu
|
Abstract: For a socio-economic variable, the process of employing less frequent and more reliable data (e.g., annual reports) to adjust more frequent and less reliable data (with error, e.g., repeated monthly surveys) is called benchmarking. Benchmarking via any statistical method requires the error model and that is usually not provided to analysts. This paper suggests methods of choosing a working model for the error to replace the unknown ?true? model in the benchmarking process, and how to evaluate the impact of the replacement on the benchmarking results. As an example of such a strategy, the paper focuses on the regression method (Cholette and Dagum [8]) and shows that by properly choosing an AR(1) model, the method may work well in most practical situations. An AR(1) error-modeling procedure that uses both monthly data and annual benchmarks is proposed and tested with this benchmarking method. The problem of error-variance estimation after benchmarking is also discussed. |
Keywords and phrases: ARMA model, autocorrelation, benchmarking prediction, difference stationary, regression estimation, signal-extraction. |
|
Number of Downloads: 369 | Number of Views: 1150 |
|