This is from Wojciech Kopczuk in his recent NBER paper:
The methods that rely on direct measurement of wealth — that is, those based on the Survey of Consumer Finance and on the estate tax — show at best a small increase in the share of wealth held by the top 1 percent, while the capitalization methods show a steep increase.
These methods start diverging in their estimates in the 1980s, and the paper has a very useful discussion of their strengths and weaknesses. This is a notable paragraph:
The most striking feature of the estimates for 2000s is a huge run-up of fixed income-generating wealth in the capitalization series. In fact, this run-up accounts for virtually all of the increase in the share of the top 0.1% between 2000 and 2012 and most of the increase since 2003. The underlying change in taxable capital income (reported by Saez and Zucman, 2014, in their Figure 3) is nowhere as dramatic. The fixed income actually falls in relative terms, as would be expected when yields fall. Instead, the (almost) tripling of the fixed income component on Figure 3 (from 3.3% of total wealth in 2000 to 9.5% in 2012) is driven by an increase in the underlying capitalization factor from 24 to 96.6. This is precisely what the method is intended to do: as yields have declined, the capitalization method should weight the remaining income much more heavily. This increase – if real – would correspond to enormous re-balancing of the underlying portfolios of the wealthy throughout the 2000s. An alternative possibility is simply that the capitalization factors are difficult to estimate during periods of very low rates of return resulting in a systematic bias.
Overall Kopczuk does not favor the capitalization method and thus there seems to be a very real possibility that U.S. wealth inequality has gone up by only a modest amount.
For the pointer I thank Allison Schraeger.