|
Abstract:
|
Compatibility testing determines if two series, a sub-annual and annual series both subject to sampling error, can be considered suitable for benchmarking. We derive a first test that requires the knowledge of the covariance matrices of the errors. We show that the Wald, Score and Likelihood ratio tests are inapplicable. We suggest an alternate test based on the observed bias in the discrepancies. There, we show that the Wald, Score and Likelihood ratio tests become applicable. We propose two alternative tests statistics independent of the covariance matrices; however, we show that the tests cannot be done unless additional assumptions on the covariance matrices are made. Finally, we discuss future research directions.
|
