»Die reichsten fünf Prozent besitzen mehr Vermögen als angenommen«

One very well established fact about the distribution of wealth, one that is often recurring in the public debate, is its substantial positive skew. That is, only a small percentage of the population hold a considerable fraction of total wealth. For instance, in the US, the one percent richest households hold approximately 30 percent of total household wealth; the five percent richest households hold approximately 60 percent of wealth. Good measurement of these relative shares of wealth holdings are important. For instance, macro-economic models are sometimes calibrated using these shares. In addition, the wealth distribution is a topic of broad public interest.

Up to recently, relatively little was known about the wealth distribution of households for most countries in the euro area. That should not be too surprising. Knowledge of the wealth distribution of households is only possible if it is somehow measured. One way to measure wealth of individual households (and thereafter the distribution of the population of households) is to use extensive household surveys that ask questions of all asset holdings and debts of the household. In the US, the Survey of Consumer finances (SCF), sponsored by the Board of Governors of the Federal Reserve System, is widely conceived to be the best source of information on the US wealth distribution. This household survey has a long standing and much of what we know today about the US wealth distribution comes from this survey. In the euro area a new survey, similar to the SCF, the Household Finance and Consumption Survey (HFCS), from a joint project between the European Central bank, the Eurosystem and a number of national statistical institutes, has been recently released (April 2013). Both the SCF and HFCS are large surveys specifically designed to represent the entire population and to measure wealth at the household level.

This paper is concerned about measuring the share of wealth held by the top one and five percent of households. It starts with the premise that measuring wealth at the top is always difficult. Household surveys are widely believed to suffer from various degrees of non-response and differential non-response. Sampled households don’t participate in surveys for numerous reasons: absence, lack of time, refusal to reveal sensitive information, etc. When these reasons are correlated with wealth itself serious biases can occur in wealth estimation and the shares of wealth can be biased too. Wealthier households are generally thought to have higher non-response rates. If non-responding households are having higher wealth in some systematic way, wealth estimates will be biased downwards, particularly estimates of wealth at the top of the distribution.

One way that survey analysts try to remedy this potential problem is by stratifying the sample and oversampling of the wealthy. This is usually done by drawing the rich housholds from a special sampling frame (e.g. tax records). This way more wealthy households will end up in the sample, and their sample weights can be better adjusted for non-response. This strategy goes a long way in minimizing the downward bias of wealth estimates at the top of the distribution.

However not all wealth surveys oversample the rich and not all oversampling strategies are equally successful in obtaining rich households in the sample. This paper proposes a method to overcomes this bias, which is especially useful when the rich households are not or little oversampled. The sample data of the HFCS and the SCF are combined with Forbes World’s billionaires data. Estimates of wealth at the top can then be obtained by estimating a Pareto distribution for the top wealth holders, as earlier research has shown that the Pareto distribution forms a good approximation of wealth held at the top.

In the paper it is then first shown that combining these sources of data when estimating such a distribution can improve on the estimates of the share of wealth held at the top of the distribution. Thereafter, the paper compares estimates of the upper tail of the wealth distribution using different methods. Direct estimates of tail wealth from the sample as well as estimates from an estimated Pareto distribution are compared. The Pareto distribution is estimated with and without adding Forbes individuals at the extreme tail.

This paper provides a set of estimates on the tail of the wealth distribution in the US and nine euro area countries. The results suggest that differential non-response problems, are particularly high in a number of euro area countries, leading to underestimation of the top wealth shares when using only the surveys to construct tail wealth or using estimates from a Pareto tail without extreme tail observations. When using the extra information provide by the Forbes World’s billionaires list, estimates of tail wealth, and shares of the top one percent and five percent increase. 

Die vollständige Studie finden Sie hier (pdf).