I get tired of correcting Keynesians who don’t understand how to estimate the multiplier. But as long as they keep saying things like this, I’ll have to keep doing so:
In general, cross-sectional comparisons are proving to be a very good way to test some propositions in macroeconomics. I’d cite, for example, the Nakamura-Steinsson paper (pdf) that uses fluctuation defense spending — which has very unequal impacts across states — to estimate the multiplier on fiscal policy (it’s about 1.5).
Actually cross-section studies are a lousy way to test for the multiplier, as they suffer from the fallacy of composition. I’ve pointed this out many times, but Paul Krugman obviously doesn’t read my blog, otherwise he wouldn’t keep making these elementary errors. Of course he’s told us that he doesn’t like to read conservative blogs because they contain nothing of value.
Then there is this, from the very same post:
Calculated Risk sends us to two papers by Amir Sufi and Atif Mian using county-level data to investigate the causes of the recession. Their work strongly supports the balance-sheet view: a fall in demand from highly indebted households is the big story, and you can confirm that by showing that the big declines in nontradable employment — that is, employment in industries that sell locally — is in those countries where debt levels were high.
And let me guess, counties with economies dominated by autos and steel usually suffer bigger job losses in recessions than counties dominated by hospitals and colleges. Does that tell us anything about what causes recessions? Doesn’t EC101 also teach that correlation doesn’t prove causation?
Let’s suppose recessions were caused by tight money, not balance sheet problems. And let’s suppose that tight money reduces nominal income. And let’s suppose that most debts are nominal, not indexed to inflation. In that case wouldn’t you expect tight money to lead to bigger spending declines in highly indebted areas, even if debt played no role in causing the recession?
Fallacy of composition. . .
Correlation doesn’t prove causation. . .
Krugman’s textbook must have something to say on those topics.