Estimation Methods for the Upper Tail of the Wealth Distribution

This paper investigates methods to estimate the upper tail of the wealth distribution. I compare data types and estimation methods using data from the Netherlands for the period 1993–2018, exploiting the unique availability of multiple types of data for this context. In addition to comparing the existing methods of OLS regression, Maximum Likelihood, and Generalized Pareto interpolation, I develop a new method to combine data from several sources. This method, called Robust Pareto Regression, combines local estimates of wealth concentration from individual data sources, and uses fixed effects methods to correct for the heterogeneity across data sources and years. Several conclusions emerge: (i) No data source on its own accurately captures the top tail, meaning that all sources need to be adjusted or combined to estimate top wealth. (ii) Combining surveys with rich lists is highly sensitive to the quality of the underlying data sources; generalized Pareto interpolation partly addresses this concern, but straight Pareto regression and Maximum Likelihood methods do not. (iii) Robust Pareto Regression is preferable to existing methods, since it more adequately adjusts for data heterogeneity, and easily shows trends over time.