A0790
Title: Bayesian model averaging for income distributions
Authors: Haruhisa Nishino - Hiroshima University (Japan) [presenting]
Kazuhiko Kakamu - Nagoya City University (Japan)
Abstract: The aim is to estimate the generalized beta distribution of the second kind (GB2) with four parameters for the income distribution. It is known to be helpful in analyzing income distribution. However, estimating the GB2 by the maximum likelihood estimation has some problems, such as difficulty choosing appropriate initial values, which can lead to unstable estimates. An alternative feasible Bayesian method is to estimate it using the Taylored randomized block Metropolis-Hastings (TaRBMH) algorithm. On the other hand, the GB2 distribution encompasses several three-parameter distributions as special cases, such as the Dagum distribution, the Singh-Maddala distribution, the Beta distribution of the second kind (B2), and the generalized Gamma distribution. Therefore, the Bayesian model averaging is also explored using these three-parameter distributions to estimate income distributions. The simulation study shows that comparing the two Bayesian methods indicates that the latter requires less computation time than the former. The two methods are also applied to actual Japanese equivalent income data (individual and group data) from the comprehensive survey of living conditions. The two methods are thus evaluated, and the characteristics and dynamics of income distributions in Japan are investigated.
