Thursday, 11 December 2014

OECD on inequality

Whenever a small country gets mentioned specifically in an international report, that report there gets noticed.

A new OECD working paper claims that income inequality hurts economic growth, and particularly hurt New Zealand growth. Note that this is a working paper rather than OECD position.

Let's walk through their method a bit before discussing.

They use a new OECD panel to estimate effects. Growth, and everything else, is measured at five year intervals from 1970 to 2010.

Rather than use standard fixed-effect OLS modelling, they use a System Generalised Method of Moments approach. I've not used this approach before but here rely on their description: it combines first differenced equations with a set of lagged first-differences of the explanatory variables as instruments.

They find that net inequality (after tax-and-transfer) hurts economic growth, that gross inequality (pre tax-and-transfer) doesn't hurt growth, that changes in human capital (education) do not affect growth one way or another - there's a slightly negative effect of education on growth in the set of specifications, but it's not significant; and, investment doesn't affect growth one way or another.

The set of results is then a little surprising. We usually expect investment to matter a lot for growth - both in physical plant and equipment (investment) and in people (education). They find that neither does anything and that the only thing that matters is inequality.

Further, when they break things down a little, what seems to matter most is the difference between 4th decile income and average income rather than incomes at the top. Incomes in the 9th and 10th decile relative to average income do nothing; differences between the fourth decile and the average matter hugely.

And now we start getting into the plausibility checks. Does this set of results really make sense?

By what mechanism does a sharper gradient between income at the 40th percentile and average income translate into worse growth? Imagine that we took this as policy conclusion: increase the tax on average earners to give money to people slightly poorer than them. Does that seem reasonable? They argue that the effect runs through reduced investments in education in the lower decile cohorts when income inequality is higher, but they found no effect of education on growth. Further, the countries examined, like New Zealand, went through rather a few changes to tertiary education over the period - from free tuition to tuition to student loans. All of these would affect lower-tier access to education, and none are accounted for.

The paper goes on to argue that its results on education must be wrong because everybody knows that education matters, then makes a strong case not for income transfers, but for increased spending on education. And they hang that case on other papers finding a strong effect of education on growth - but that also find that inequality increases growth!

I've a specific concern also about the use of New Zealand in this time series. The potted history of New Zealand inequality and growth. New Zealand growth rates tanked from the late 1970s through the early 1990s as first Muldoonism then necessary restructuring put a pretty high cost on the economy. The Muldoon stuff was nonsense. The economic restructuring set the groundwork for strong growth in the 90s and through the 2000s, but was really really painful. It was painful for laid off workers, and it was painful for a whole pile of firms, both large and small, that had to reinvent themselves for an open and free market. Them ships don't turn on a dime, and so growth tanked.

At the same time as NZ growth rates tanked due to restructuring, incomes at the top jumped - in part due to changes in tax and accounting that brought some of that onto the books where it previously had been hidden, and in part due to that folks with the skills to adjust to the new environment started being compensated for it. That rise in inequality happened almost entirely from 1985 through about 1992, after which it wobbled around but didn't have systematic trends.

If we look then at a long run data series, we get a big increase in measured inequality in New Zealand at the exact same time as economic growth takes a nosedive. Once the growth in inequality stops around 1992: blammo! Growth starts again.

Is it any wonder, then, that a regression approach based on reduced form fixed effect estimation with no dummying out of the reform period or other adjustment for it would find huge effects of inequality on growth in New Zealand? It's stuff like this that's meant that more recent academic work, unlike OECD working papers, has been shifting to use of microdata within countries to try to figure out what's causing what. I don't think the OECD papers get us there.

I note that I have profited from conversations with Matt Nolan on this, though I'm to blame for all errors here.

Update: Here's me and Tim Hazeldine agreeing about the report's merits over at Morning Report.


  1. shanghai, hong kong and singapore top the pisa education rankings, yet have high income inequality

  2. When you say the relevant difference is between 40 percentile and average do you mean between 40 percentile and 50 percentile (median) or between 40 percentile and mean?
    If the latter, is the result an indication that skew in the inequality is important?

  3. It's the average, not the median. You could say that points to skew, but there's like tiny effects in 10th percentile to average, bit bigger in 20th percentile to average, real big in 40th to average, then nothing at all in 90th to average.

  4. I use the B I S index. Beggars in street. I count the number of beggars I encounter about central Wellington. It has increased 350 % over the last 20 quarterly readings. You can use stats to argue anything. Figures can lie, and often do. I always ask, " who is paying for this report?" Who is paying you?

  5. Just a standard paper that backfits data onto a multiparameter polynomial model. Lots of math programs, and I think Matlab, have a module to do that. What is surprising to me--and you can find my analysis here: --is that papers on growth often fail to mention the term "patents" at all, not even in passing. Of 51 Cato Institute sponsored papers on growth, only two mentioned patents in any meaningful way. This is doubly surprising since there is no hard evidence that patents do either good nor ill (see the link above), so why not at least mention that fact? It's small wonder that the general public does not take economists seriously. Sadly, policy makers seem to.

  6. Well if some percentage of education is signalling, then it acts less like investment right?

  7. The average household income in the US is around the 60th percentile. The fourth decile is centered on the 35th percentile. So redistribution from one to the other is to people substantially poorer.