Piketty

Given my interest in picking holes in other people's numbers you might have thought I would have something to say about the FT's hatchet job on Piketty. At the moment I'm loath to say very much for the simple reason that though the book has sat on my desk for the best part of a month it's unlikely that I am going to have sufficient time to read it with the attention it deserves before the Summer vacation begins. Commenting on things you haven't read is rarely a good idea. In any case a lot of heavy weights have already waded in,  with more knowledge and insight than I'll be able to muster.

I did read  Chris Giles' piece on Saturday and my overall impression was that though he may (or may not)  have uncovered a few data errors - let's face it all data contains errors  but unlike Piketty most people don't  bother to expose themselves to the risk that others will sift through their work to find the odd tasty morsel - he didn't even attempt to make a serious case that these errors - if errors they be - are sufficiently important to torpedo Picketty's argument.

Obviously Piketty will have to respond to Giles' claims and I've no doubt that he will. Science is all about conjecture and refutation, so let's see what happens.

There is one specific claim that Giles  makes  which I do have a hunch about. Let me be clear, it is nothing more than an informed guess and I'm perfectly happy for somebody to come along and demonstrate that I'm talking nonsense.

Giles claims that inexplicably Piketty ignores (or decided not to use) evidence for the UK from the ONS  Wealth and Assets Survey (WAS). As it happens I know a little about this source. When I sat on the National Equality Panel we were given access to preliminary results from the first wave of it. And jolly grateful we were to have them because good micro level data about wealth is like gold dust.

What struck me at the time, and after spending a few minutes yesterday having another look at the survey documentation I see no reason to change my view, is that WAS is probably not a good source for producing sensible estimates of the wealth held by the top 1% of the wealth distribution. The reason is very simple. Error prone household surveys tend to do a poor job of measurement when it comes to the tails of the distribution - either those with very little wealth or those with a lot of wealth. All sorts of errors have a habit of accumulating in the tails and the ratio of signal to noise in these regions is probably quite low. Add to that that WAS only samples about 20000 households and you can readily see that the amount of information about the tails of the distribution is pretty sparse and the confidence intervals must be rather wide. And all this is before we even get into sample selection bias (do the super rich agree to be interviewed?), selective item non-response and all the rest.

No doubt HMRC tax data is also flawed in all sorts of ways, but it is not obvious to me that - in the context of creating a time series - it is obviously worse then  the WAS data.