as we wanted. Therefore, @ is indeed a bias-adjusted estimator up to order O(T~'). It is very easy to see that @ has the same asymptotic normal distribution as the original estimator a, as it usually happens with bias-corrected estimators. The simple argument involves the fact that the original estimator is Op(T~!/2) while its bias is O(T7~!). 0: CML; 0: SD; 6: modified CLS; @: corrected SD. 6: CML; 0: SD; 6: modified CLS; @: corrected SD. Table 4. Integrated relative absolute bias 6: CML; 0: SD; 6: modified CLS; @: corrected SD. Table 5. Mean squared errors (A = 1.0) 0: CML; 0: SD; 8: modified CLS; 6: corrected SD. Table 6. Mean squared errors (A = 3.0) 6: CML; 0: SD; 6: modified CLS; 6: corrected SD. Table 10. Descriptive statistics Table 7. Mean squared errors (A = 5.0) 6: CML; 6: SD; 6: modified CLS; @: corrected SD. 0: CML; 6: SD; 6: modified CLS; 6: corrected SD. Table 11. Estimates of the parameters (standard errors in parentheses), RMSE, MAD and MAE for counts of burns claims Figure 1. Time series plot, autocorrelation and partial autocorrelation functions of the series of burns claims from 1985 to 1994.