What’s Wrong with Survey-based Top Wealth Shares? Evidence from Housing Wealth of French Households

This paper investigates the causes of biases in survey-based top wealth shares by comparing the 2017 French wealth survey with a new benchmark database on housing wealth of French households. It contributes to preliminary work toward a major overhaul of the French wealth survey currently conducted by the French National Statistical Institute.

Compared to this benchmark, only 10% to 20% of the top 10% of the housing wealth distribution is missing in the survey (both in population share and in wealth share), but this proportion increases to 40%-50% for the top 1%. I link all sampled households to the benchmark database, respondents and non-respondents alike, and use this linked data to measure the discrepancy induced by each step of the survey process, based on an innovative decomposition approach. I conclude that the downward bias in survey-based top wealth shares comes in equal parts from two causes. First, households belonging to the top 1% are strongly underrepresented in the survey mostly because they are somewhat more difficult to contact and much more reluctant to participate than the rest of the population. The weight adjustment procedure does not fully compensate this underrepresentation and calibration unexpectedly distorts the wealth distribution. Second, wealth underreporting is more intense among wealthy households.

I then compare reported housing assets with the assets actually owned by respondents and show that households tend to report assets they have full legal and daily economic control upon, and that asset underreporting increases sharply with the number of housing units owned by households and among high net wealth households. I compare market values reported by respondents for their primary residence with both statistical estimates and prices observed in real estate transaction data and conclude that households at the bottom of the housing wealth distribution tend to overestimate the value of their home, whereas households at the top tend to underestimate it.

I finally compare reported housing wealth with both the benchmark database and wealth tax returns and show that wealth underreporting among the very wealthiest households (with an estimated housing wealth above 5 meuro) amounts to approximately 40%. Based on these findings I suggest potential improvements to wealth survey methodology.