Saturday, May 24, 2014

Some thoughts on the Piketty data thing

This is just transcribed from what I said on facebook, but I thought some readers might be interested in it here that aren't friends with me on facebook:

I'm a little concerned at where this data debate is going, and it makes me worry that people got the wrong impression of how these things ought to go from the Reinhart and Rogoff template. The R&R problems were pretty explicit excel errors. They did not demolish the case (actually Dube's econometric work did more on that front), but if I understand it right (I have not read the CJE article from the critics) the unambiguous excel errors did make the 90 percent threshold go away. And then, the important thing is, they published it and it was clear where the errors were.

The Piketty situation is quite different. True, there are a few excel errors too. But the meat of the criticism is NOT of that nature. As a result it's very hard to even verify that Giles is right and Piketty is wrong. People who are jumping to conclusions here don't seem to understand how work with raw data works. Raw data come with a lot of problems, and it's not just the 19th century data sources. There are blindspots and limitations to many data sources from before the 90s, and I assume the wealth data is no different. Moreover, wealth combines many disparately and poorly measured components that have to be brought together. Is Giles's estimate including all of those pieces or is it missing one of them because (unlike Piketty) he hasn't been neck-deep in wealth data for the last fifteen years. I don't know. And nobody commenting on this knows either so DON'T let them convince you that they do. But that's really the heart of the matter (unless something turns out to be wrong due only to the excel errors - the miscopied lines and the averaging - if something is very wrong just from that then we have a problem pretty comparable to R&R).

A lot of times adjustments and imputations are made that are not easy to discern unless you dig into it. And it's really nuanced stuff usually. In a paper we just resubmitted to the Review of Industrial Organization some of the variables a referee wanted us to add - the yield spread premium - had to be imputed. It was only available for certain home loans in the dataset, so we had to generate our own data that don't exist anywhere else in the world by modeling YSP based on the data that did have it and using that model to predict values for the cases that didn't. It wasn't perfect of course. But the more important point is that I think we just said we made an imputation for some cases and left it at that (we have all the program files so if anyone cares they can come to us - we used standard approaches that other people have used before). If you just took our raw data and naively ran the numbers you would get something different. But if you just took our raw data and ran the numbers you would be wrong. Even if there might be improvements on what we did you would still be much more wrong to use the raw data. This is a minor case - I'm sure wealth calculations across countries and time must have much bigger examples of legitimate but not immediately transparent adjustments and imputations.

We are definitely not merely in Reinhart and Rogoff territory with this.

To get to the bottom of this, we simply need more information and review. And someone probably needs to publish it like the R&R critics do. It's an important revelation. I don't think I'll get around to the book until I'm past my comp and into the summer, and at that point I fear there's going to be another set of 600 pages worth of reviews to read too.

And for people who think this might be ideological, it's worth saying something about my priors before Reinhart and Rogoff and Piketty came into the public discussion:
1. I did believe wealth inequality was rising along with income inequality
2. I did not believe r > g
3. I did not believe there should be an international wealth tax
4. Whatever flaws there were in capitalism I did not believe it would doom capitalism
5. I did think that high levels of government debt risked financial crisis
6. I did think the level of public debt that posed a danger depended on the institutions and strength of the economy, and therefore I
7. Didn't think much of a single 90 percent threshold
8. I did worry about the problems posed by entitlement programs for long-run debt projections

For better or worse these haven't moved all that much. I consider most of the Piketty-related points to be open questions to reconsider now that the book is out. So although it's a mixed bag I am at least as much a Reinhart/Rogoff partisan as I am a Piketty partisan.

3 comments:

  1. This is off-topic, Daniel, but...why don't you tag posts anymore?

    ReplyDelete
  2. Bob Murphy has documented / identified a host of other problems with Piketty's work http://mises.ca/posts/blog/are-you-kidding-me-now-allegations-that-piketty-fudged-his-wealth-data/

    ReplyDelete
    Replies
    1. Yes I've been following his posts. This one of his is just repeating Giles's material and what I said above still applies here.

      Delete

All anonymous comments will be deleted. Consistent pseudonyms are fine.