Abstract:
|
The Bank of Canada 2015 Retailer Survey on the Cost of Payment Methods (RCPM) faced low response rates and outliers in sample data for two of its strata: chains and large independent businesses. In the presence of outliers, Chambers (1986) and Beaumont, Haziza and Ruiz-Gazen (2013) suggested the use of a bias-corrected estimator based on an appropriate choice of the tuning constant. In this paper, we develop a parametric bootstrap procedure to determine the tuning constant which minimise an estimator of the mean square errors (MSE). Monte Carlo simulations are performed to compare the performance of the robust-estimators for several choice of the tuning constant, including the adaptive method proposed by Beaumont, Haziza and Ruiz-Gazen (2013). The results show that for all the simulation set-up considered in this paper, the estimator based on the conditional bias with the adaptive choice of the tuning constant proposed by Beaumont, Haziza and Ruiz-Gazen (2013) performs better in terms of both bias and efficiency. The methodology is applied to the RCPM survey.
|