Title: On a modification of Efron bootstrap method for heavy-tailed distributions
Authors: Hannah Opayinka - University of Ibadan (Nigeria) [presenting]
Abstract: The nature of the upper tail of a heavy-tailed distribution is the major reason for the poor performance of classical bootstrap methods, which results in large standard error. Efron's bootstrap method is modified to address the challenges faced when dealing with heavy-tailed distributions. The methodology involves stratifying the observations into homogenous subgroups and using proportional allocation method in selecting bootstrap samples. Real and simulated data sets drawn from Lognormal and Singh-Maddala distributions respectively were used. The two distributions have finite variance. The performance of the Modified Efron Bootstrap (MEB) is compared to that of Efron's bootstrap using SE (standard error) and RMSE (root mean squared error). The findings show that SEs and RMSEs for all sample sizes considered were consistently smaller in MEB than Efron bootstrap. Hence, MEB outperformed Efron's bootstrap when applied to heavy-tailed distributions for the two cases considered.