Page 194 - Contributed Paper Session (CPS) - Volume 6
P. 194

CPS1881 Mike S.C. et al.





                                 Likelihood ratio tests for Lorenz Dominance
                                  Mike S. Chang; Michelle Liou; Philip E. Cheng
                        Institute of Statistical Science Academia Sinica, Taipei, Taiwan, Republic of China

                  Abstract
                  The notion of Lorenz dominance (LD) between two Lorenz curves (LC; Lorenz,
                  1905)  is  useful  for  evaluating  welfare  redistribution  toward  decreasing  the
                  inequality  or  ranking  income  distributions  based  on  expected  utility.  Early
                  studies of estimation and testing for the LD property were mostly discussed
                  based on a grid of finite points on the unit interval. Extending the inference
                  from  a  finite  grid  to  the  entire  unit  interval,  consistent  tests  for  the  LD
                  hypothesis were recently developed upon functions of the empirical Lorenz
                  processes. The asymptotic distributions of the test statistics depend on the
                  unknown distributions and theoretical test p-values were empirically assessed
                  using the bootstrap method. In view of the intersection (once) property of a
                  pair of crossing Lorenz curves discussed in the 1980s, elementary properties
                  of crossing and dominant Lorenz curves will be examined in this study. It is
                  found that distinct patterns of the difference curve between a pair of curves
                  can  be  used  to  characterize  the  LD  and  the  crossing  Lorenz  curves  (CLC)
                  conditions through inequalities of their quantile functions. We will use these
                  patterns to construct likelihood ratio (LR) tests tailored for the LD and the CLC
                  hypotheses, respectively. The proposed LR tests are consistent with respect to
                  the standard test levels and critical regions based on approximate chi-square
                  distributions. A simulation study is conducted using Log-normal, Pareto and
                  Weibull distributions, and reliable performance of the LR tests are obtained.
                  An  exceptional  difficult  case  is  found  under  the  LD  hypothesis  when  two
                  lognormal distributions with close standard deviation parameters are rather
                  close to each other, for which larger sample sizes are useful for making a test
                  decision. An empirical study is carried out for certain pairs of real GDP annual
                  data of 133 countries from “The Penn World Table (Mark 5): An Expanded Set
                  of International Comparisons, 1950-1988”.

                  Keywords
                  Crossing Lorenz curves; Likelihood ratio test; Lorenz dominance

                  1.  Introduction
                      The Lorenz curve provides a graph of overall income proportions shared
                  by  cumulative  proportions  of  people  acquiring  incomes  from  low  to  high
                  levels. The notion of Lorenz dominance (LD) between two curves (one curve is


                                                                     183 | I S I   W S C   2 0 1 9
   189   190   191   192   193   194   195   196   197   198   199