John Cochrane makes a good argument against Larry Summers that I've made here before that you simply can't use cross-state studies to talk about a national stimulus package. You could - perhaps - use those studies to provide some insights into national stimulus in some Eurozone country. But even there you have to be careful about what you're looking at exactly.
My concern with Cochrane's discussion is that it conveniently only goes through the things that would bias these estimates upward (thus producing overstimates of the multiplier). There are other reasons to think the estimates could be biased downward, producing underestimates of the multiplier. Cochrane should be up front that both are in play. But the point remains that cross-state studies are going to give you pretty crumby estimates and you can never be quite sure what to make of them.
Mark Thoma, though, also pushes back on Cochranes citation of crowding out and highlights none other than John Taylor pointing out that there's no reason to believe stimulus spending will completely crowd out private investment (it would kind of defeat the point if it did!).
Thank you Daniel for trying to be evenhanded on it. Some others referred to Cochrane's post as a "rant." Why oh why can't we have better Keynesian commentators?
ReplyDeleteFirst - thanks.
DeleteGiven the kind of abuse that's heaped on Krugman, though, I'm not too bothered by simply calling it a "rant". Cochrane compared Summers to medieval doctors, too (perhaps that was what ticked Krugman off??). None of this seems all that bad, particularly relative to what gets flung at Krugman.
What did you think of Krugman's actual post? I think he makes an excellent point, and cites a great passage from Christina Romer (who, over the last several months, I've personally developed quite a crush on). She's hammering the state-level counter-factual point just like Cochrane was. I have a hard time justifying liking Cochrane's post and not also liking Krugman's for the exact same reasons.
Dude, still wrong about state level (and/or county) data. One looks where one has data. The alternatives are the time series of a single experiment or 50 (3000) experiments that leak. We don't have the luxury of a lab, so we have to accept leaky experiments. Your criticism is legit and important to keep in mind, but it's not so powerful that it renders the results invalid (or less valid than the alternative).
ReplyDelete