Just one post yesterday, “Correlation, Causality, and Casuistry:”
And I’m back. This real life thing is getting in the way of my blogging, so maybe I should give it up.
The delay has put me behind the curve on two new entries in the Reinhart-Rogoff affair, by Arindrajit Dube and Matt O’Brien. On the other hand, it’s given me time to think about Dube’s post, which I felt I needed to do before posting myself.
Dube tries to take on the issue of whether high debt causes slow growth or vice versa. His approach is clever, comparing the power of debt as a predictor of both past and future growth, but to my taste he skips too many steps in his explanation. So let me try to fill in the gaps.
Imagine one story — the story that R-R are implicitly telling — in which countries differ in their fiscal responsibility, this leads to different levels of debt, and those countries with high debt then suffer from slow growth. In that story, debt should be a pretty good predictor of future growth. You might also expect to see some correlation between debt and past growth, because debt levels change only gradually over time, and a country with high debt now typically had high debt and hence slow growth a few years ago too. But you’d expect the relationship between debt and future growth to be stronger than the relationship between debt and past growth.
Now imagine another story, in which countries aren’t that different in fiscal responsibility, but in which some countries for whatever reason — burst bubbles, declining fertility, structural problems coming from social change or something — have slower growth than others. Very plausibly, slow growth would lead to rising debt ratios, both because of slow growth in revenues and simply because the denominator of the ratio would be smaller. In this case past debt should be strongly related to past growth. You might also expect some relationship between debt and future growth, because growth tends to be “serially correlated” — countries that grew slowly in the past tend to keep growing slowly — but that relationship should be weaker.
So Dube does the exercise, and it looks like this:
Clearly, the data look a lot more like story #2, in which slow growth causes high debt, than story #1, which is what everyone hyping Reinhart-Rogoff claimed.
And the everyone hyping Reinhart-Rogoff very much included Reinhart and Rogoff themselves. Matt O’Brien has the goods. It’s true that their papers never said outright that the relationship was causal, but they weren’t anywhere near that scrupulous in op-eds and other media presentations. And the truth is that the papers may not have stated causation flatly, but it was clearly insinuated. By trying to claim now that they never meant to imply such a thing, R-R are falling down seriously in the menschhood test.
One last thing: even if you take Dube’s forward-looking regression as a causal relationship, which you shouldn’t, notice how weak that relationship is in the relevant range. It looks as if raising debt from 50 to 150 percent of GDP, other things equal, reduces growth by around 0.1 percentage point over the next three years. This is the dreadful consequences that prevents us from doing anything about mass unemployment?