A COMPARATIVE-STUDY FOR THE ESTIMATORS OF THE GENERALIZED PARETO DISTRIBUTION

Various estimation methods for the three-parameter case of Generalized Pareto distribution are given, including an alternative method based on least square (LS). Modified formulae for computing estimators are provided. The performances of these methods are compared by Monte Carlo simulation. It is found that the LS method has generally a lower root mean square error (RMSE) than that obtained using other methods. The LS method also performs best in terms of BIAS when the shape parameter is greater than zero (c > 0), while the probability weighted moments (PWM) method performs best when c < 0.