Are There Problems with Piketty's Data? Of Course

By Patrick Brennan

Friday, May 23, 2014

The Financial Times published a stunning claim today: French economist Thomas Piketty made serious mistakes in the research undergirding his book Capital in the Twenty-First Century, and not necessarily just inadvertent ones, but ones that suggest he may have cherrypicked data to uphold his argument.

A refresher on his theory, since I assume the bestselling book is now sitting underneath other unopened books on most people’s desks: Capitalist societies, with rare exceptions, see ever-increasing concentrations of wealth. We’re in a destructive self-reinforcing trend back toward 18th- and 19th-century levels of inequality, he argues using data and economic theory, and we need 80 percent income tax rates and a global tax on wealth to stop it.

Piketty — unlike plenty of economists — made the data underlying his book publicly available so that people far smarter than I can poke at it, and Chris Giles, an economics editor of the FT, is one of the first to do so in depth. Has he gutted Piketty’s empirical argument?

No, but there are real problems here. Piketty’s empirical work is both clearly impressive and ultimately unreliable. And when Giles looks at Piketty’s source data, he doesn’t see much evidence for the idea that wealth inequality has been rising in the rich world over the past several decades.

This is important to Piketty’s thesis. He thinks that the natural tendency of capitalist societies is toward ever-growing wealth inequality, because the returns to capital typically remain higher than economic growth overall. Wealth inequality shrunk in the first half of the 20th century or so, he says, because of high taxation, nationalization movements, and highly destructive wars. His argument relies on both empirical research and theoretical work — what if the evidence that huge inequality is returning simply isn’t there?

Giles takes the biggest issue with the historical British wealth data Piketty cites. Here’s a chart he created, with the wealth-inequality data from Piketty’s sources in red and the trend Piketty calculated to back up his argument in blue (all English charts have fog-colored backgrounds, I’m told):

As Giles explains in this video, the sources of the data that fit Piketty’s line, according to the British government, are supposedly not apt for these kinds of comparisons, while the data that has wealth concentration well below the trend he’s claiming — the two little dashes below the blue lines — are the right numbers. And British wealth data isn’t the only serious issue the FT raises.

In fact, Giles writes, “the combined result of all the problems is to make wealth concentration among the richest in the past 50 years rise artificially.”

This is a big problem — if Giles has found unexplained mistakes or distortions. All academics have to make adjustments to data, especially historical economic data. Piketty has to select from unreliable and incomplete sources of data, interpolate some data further back in the past, and had to decide how to weight it all. The fundamental problem is that this is such a tough task it’s not entirely clear Piketty should mount his grand conclusions (a global wealth tax, an 80 percent income tax) on its empirical result, ever — more on this below.

Giles mounts a convincing case that Piketty has at least weighted Europe wrong. He weighted Sweden, the U.K., and France equally when the latter two are seven times the size of the first — this doesn’t look defensible the way RR’s weighting was. Here is Piketty’s calculation of the Europe inequality trend in blue, and Giles’s in red:

The most dramatic comparison comes together here. Piketty’s key chart shows rich-world wealth inequality rising since 1970:

The FT’s averaging of the data shows it’s not:

Is the FT right? It’s hard to say, as Giles himself admits. “While this post is clear about what is wrong with Piketty’s charts,” he writes, “it is much less certain about the truth.”

What Piketty has been doing for years, and what he set out to do in this book, combines at least two very difficult tasks in economics: measuring inequality and reconstructing historical data.

We shouldn’t trust his choices (or Giles’s criticisms) implicitly, but neither should conservatives dismiss research like this out of hand simply because it’s complicated and sometimes subjective. For one, people who are vehemently skeptical of social science will still prize a result that they find ideologically appealing — remember the highly popular Reinhart/Rogoff paper arguing that high national debt slows economic growth dramatically, which turned out to be (partly) wrong on theoretical and empirical grounds when people looked at the data. Second, understanding the world simply is hazardous but important, and to some extent we have to rely on experts to help us do it.

The Frenchman’s reply doesn’t quibble yet with Giles’s specific criticisms but says he did the very best, basically, with the data he had. He writes:

    As I make clear in the book, in the online appendix, and in the many technical papers I have written on this topic, one needs to make a number of adjustments to the raw data sources so as to make them more homogeneous over time and across countries. . . . I have tried in the context of this book to make the most justified choices and arbitrages about data sources and adjustments. I have no doubt that my historical data series can be improved and will be improved in the future (this is why I put everything online).

Scott Winship, a social scientist and NR contributor who has spent a great deal of time dealing with Piketty’s data on U.S. wealth and income inequality, says he hasn’t seen evidence of monkey business there, so Piketty may have good answers for why he did what he did in his European data, too.

But as Scott points out, there are more fundamental problems with Capital: E.g., Piketty lays his historical analysis and his analysis of recent trends for income inequality in Europe on . . . Britain, France, and Sweden. This is not “Europe” and it’s not clear it’s a representative sample. Extraordinary claims like Piketty’s require extraordinary proof, and as Scott explained in his National Review essay on the topic, Piketty doesn’t really meet that standard. On the other hand, Piketty could be understating the trend: One of his frequent co-authors, Emmanuel Saez, and Gabriel Zucman recently argued that wealth inequality is actually rising much faster than Piketty’s book shows. The growth has just been concentrated in the top .01 percent, they calculate (not using actual wealth data, but working back from capital gains and interest tax returns). Piketty also says that data released since the release of his book reinforces his case, so his calculations were actually conservative.

But wait — a number of French economists who’ve been looking at Piketty’s masterwork for a while wrote a paper arguing that the accumulation of wealth he found can be entirely explained by rises in housing prices, which complicates the explanations Piketty gives for rising wealth inequality.

Where does all of this leave us? Piketty set out to do something much more audacious than prove that income inequality is rising in the United States and in most wealthy countries — that’s relatively easy to prove, even if the increase has been substantially overstated. Rather, he wanted to show that this plays into a loop with increasing wealth that needs to be arrested by huge global interventions. One common objection to Giles’s skepticism tonight has been that increasing wealth inequality is simply an obvious fact of this world — why do we need the data to back it up? Well, Piketty needs the data to back up the arguments he made with it — he needs wealth inequality not just to appear high or to be rising, but to be returning to 19th-century levels as a matter of economic inevitability. The errors he made may not be devastating to the work he’s done to prove this so far, but even without taking them into account, he hasn’t yet justified his dramatic conclusions.