I'm usually skeptical of instrumental variable models, but this seems more convincing. I'd have to look closer at it, but I imagine this isn't some tiny independent variation they're trying to hang a result on. I'm guessing Congressional seniority and the number of representatives per person has a big effect on spending decisions (and not much effect on economic performance).
The results pretty much predict what I would expect. This is the abstract:
"We use state and county level variation to examine the impact of the American Recovery and Reinvestment Act on employment. A cross state analysis suggests that one additional job was created by each $170,000 in stimulus spending. Time series analysis at the state level suggests a smaller response with a per job cost of about $400,000. These results imply Keynesian multipliers between 0.5 and 1.0, somewhat lower than those assumed by the administration. However, the overall results mask considerable variation for different types of spending. Grants to states for education do not appear to have created any additional jobs. Support programs for low income households and infrastructure spending are found to be highly expansionary. Estimates excluding education spending suggest fiscal policy multipliers of about 2.0 with per job cost of under $100,000."
Big multipliers on non-grant, non-education stuff, modest multipliers on the whole shebang. A couple thoughts:
1. Krugman notes that this is going to still underestimate multipliers because of spillover. If New Jersey stimulus boosts New York and New York stimulus boosts New Jersey this will cancel that out and miss the effect because the model is identified off the variation between states.
2. So the education multiplier was small, but this also seems like it would be the component of spending where the instrument is weakest. I don't know how the sausage-factory that is the Congress works on these things, but I would guess grants to states are likely to be far more formulaic (and therefore less correlated with Congressional seniority, and therefore biased downward in these IV estimates) than spending on other grants and projects. I'm guessing there's some kind of per-student or per-Medicaid beneficiary or per-dollar-state-budget-shortfall calculations that go into that. So while the low education multiplier isn't especially surprising compared to infrastructure spending, it may still be an underestimate.
3. This goes for support programs for low-income families as well, many of which are run through the states. These programs were already found to have high multipliers. If the disbursements were formulaic rather than based on politics, then the actual multiplier is likely to be even higher.
UPDATE: So for those not familiar with my typical unease with IV models, I should probably say why this is more convincing, otherwise it seems a little self-serving. This instrument does two things that a lot of instruments don't that makes me more confident in it. First - the causal relationship between Congressional representation/seniority and fund disbursements is very real and very clear. There's no subtlety to it - you can take this relationship to the bank. It's not like Angrist and Krueger's birth cohort schtick where you were taking a very, very subtle causal relationship and identifying a model off of it. This relationship is clear and there's no clear alternative channel through which this could be affecting state growth besides through federal spending that even comes close to the effect on spending. The second thing that encourages me about this is the expected direction of the bias. In studies on the returns to education, the endogeneous processes people worry about bias estimates upwards. So when you get a positive result from an IV model on those studies, you're always secretly wondering "is this a real positive effect or not?". In fiscal multiplier analyses, the endogenous process people worry about biases estimates downwards. Unlike returns to education studies, empirical multiplier estimates (not calibration of Keynesian models with various MPC estimates - but actual empirical studies like this) are always conservative estimates. When you get a positive result on one of these, you can be more confident in it (if anything it's an underestimate).
UPDATE 2: Scott Sumner does not agree on this paper. I think he's wrong on at least one point, but I think he makes an additional very interesting criticism of the paper that has some validity. One way to look at this (as Sumner does) is that it is a "micro study" because it is not a "national study". This isn't exactly right (what would constitute a "macro" study in the EU these days? A national study? An EU-wide study?). The difference between macro and micro is an aggregation of economic activity across all markets. It doesn't matter at what level that aggregation is - if you're discussing the behavior of aggregates, you're discussing macro. But it does raise some important questions. A lot of fiscal policy is supposed to operate through the bond market - and that is a national market (and thus it is differenced out in this analysis for the same reason that spillovers across state lines are). I'll try to post more on it this weekend, but if I don't that should give you a flavor of my thoughts on Sumner's critique.
Quote of the Week
3 hours ago