[UPDATE: Just to clarify - as I say below, Mulligan claimed that labor supply effects of these programs "a lot of the increase" in non-employment. He doesn't say all. When I talk about being concerns that "he's explaining the Great Recession" with these labor supply effects, I mean I'm concerned that he's using that one gross effect on its own - not that he's alleging that he has completely explained it. If you want to say something about the impact of the safety net on non-employment, you cannot just look at labor supply and then formulate your conclusion. If you want to say something about the impact of the safety net on labor supply, then of course you can just look at labor supply.]
[UPDATE 2: OK, Casey Mulligan just went absolutely nuts on me over email and said some very insulting things - apparently completely misconstruing the post. Let me give some take aways I intend here:
1. I think labor supply effects are important - that's why I wanted to go see him.
2. I think you can't talk about the impact of the safety net without talking about labor demand effects.
3. There are a lot of issues I'm still getting my head around so I want to read more.
4. There are a lot of concerns I had about the seminar that may be in the book so I want to read more.
5. You all should think about the arguments that Casey raises.]
Yesterday I went to see Casey Mulligan speak on his book Redistribution Recession at GMU's workshop on philosophy, politics, and economics. I misjudged how long it takes the shuttle to get from the metro to campus, so I came about ten minutes late into a packed seminar room, but it was worth going and hearing the details of Mulligan's argument (and if I ever attend again I'll definitely be on time!). I'm taking a few slides from his AEI talk, which seems to have been essentially the same as the talk I saw yesterday.
His argument - in my own words - is that unemployment in the Great Recession has been so bad because changes to the safety net have raised marginal tax rates, which has reduced labor supply. The second half of that thesis, stated in that way, is eminently reasonable. Part of my interest in going was that we talked about the marginal tax rates implicit in safety net programs all the time at the Urban Institute, precisely out of this concern for the counter-productive impact on labor supply.
The first half of the sentence, stated that way, is what people are incredulous about - not the idea that the safety net has labor supply effects.
I came out of it intrigued enough by his argument to take a look at his book, despite some very questionable analysis at the New York Times by Mulligan that had put me off him for a while. As you'll see, though, I mainly want to take a look at his book because I either did not get the entire argument in the seminar or I think he made a weak defense of his case in the seminar - not because I think he clearly had the right answer.
I know this is long - the real meat of the problem is in the third section: "Assuming his own conclusions?"
Marginal tax rates
Mulligan began by describing the many changes to safety net programs that occurred during the Great Recession and the impact that these had on marginal tax rates. The idea is that if you have a benefit that's contingent on low income (like SNAP - i.e., food stamps) or not working (unemployment insurance), increasing your labor supply is going to reduce your income because those benefits go away - so it's like you're being hit by a tax by moving from poverty to somewhat-above-poverty. This has been a concern for safety net and labor market policy design for a long time. Mulligan's question is whether this can explain the decline in employment we've been experiencing.
Using an approach I'd have to read the book to get a better sense of (involving eligibility for certain programs and income levels), Mulligan aggregates these marginal tax rates to get an average marginal tax rate for prime age heads of households. It's very hard for me to compare Mulligan's aggregate marginal rate measure over time with other marginal rate measures I've seen calculated over the income distribution, but looking at - say - Gene Steuerle's numbers over the income distribution, it seems like Mulligan's rates that top out at just under 50 percent in 2009 are plausible if you were to take Gene's rates and get a population average.
Mulligan switched back and forth between talking about marginal rates and program generosity, but the measures are essentially the same - take the decline in generosity associated with an increase in income and you've basically got a marginal tax rate. The generosity changes, mapped against hours not at work, are presented below:
Now Mulligan was quick to note that he knows these two series are endogenous. "Don't call the causality police on me!", I believe were his words. I believe he knows that, but I could not for the life of me understand how that knowledge informed the rest of his analysis (which is another reason why now I want to see the book to understand if it ever informed any of his analysis). The endogeneity problem is very serious. Mulligan concludes that a lot of the increase in non-employment is due to safety net generosity, but if in fact safety net generosity is due to increases in non-employment, it's no wonder you would find these results. The actual role of labor supply could be considerably diminished.
He seemed unconcerned with endogeneity. My impression of the reasons for this was that he was taking labor supply elasticities from the literature and using his measure of increases in benefit generosity to estimate the change in labor supply. If these estimates are well identified, then the endogeneity is not a problem because he's not estimating the relationship from the (highly endogenous) relationship depicted in the graph above.
I had a very tough time getting my head around the elasticities he used to simulate labor supply changes (which I'll show you below). He said he was using Frisch elasticities and he was clearly using intensive margin elasticities. The intensive margin is the choice of how many hours to work. The extensive margin is the choice of whether to work or not. What concerns me is that Frisch elasticities (and perhaps others?) are very different depending on whether you estimate it at the micro or macro level. Since we're looking at macro series here, presumably we'd be interested in macro elasticities, but Mulligan and Peter Boettke (the workshop organizer) kept talking about it as a micro story.
The programs have very different impacts on the extensive and intensive margin. Increased unemployment insurance generosity has a big impact on the extensive margin, because if you go back to work you lose all of it all at once. Food stamps, on the other hand, impact the intensive margin because you can get them while you're working, but the level of benefits phase out as your income increases. It would seem tough to combine these.
I think everything is probably above board here. Mulligan finds large effects, which is what you'd get if you use macro labor supply elasticities that incorporate both intensive and extensive margins. It was a little confusing that it was referred to as a "micro story" in the seminar (maybe they just take "micro" to mean anything that's behavioral?), but I know he was definitely looking at macro labor supply. Anyway - the elasticities were one area where I'd be interested in taking a look at the book.
Assuming his own conclusions?
All the material above is just my general sense of the issues and some things I want to learn a little more about. The really glaring problem with Mulligan's presentation that I asked him about but I feel like he kind of dodged is that he just ignored the demand side of the programs he looked at. After explaining the aggregate marginal tax rates, he showed us a simple aggregate labor supply and demand graph he put together using actual data and estimates of the labor supply response to changes in marginal tax rates (see below). The origin is the 2007 Q4 equilibrium. The 2009 Q4 demand schedule is to the right of that because population grows and that sort of thing. It's presumably not as far to the right as it might have been because he did take wealth effects into account. This may be an overly generous assumption on the demand side, I don't know. Let's give him that one, though, because even bigger problems come up later.
Mulligan's primary conclusions come from the simulations of labor supply at 2009, Q4 with the actual safety net (the red line) and labor supply at 2009, Q4 if there had been the same less generous safety net that there was in 2007 (the pink line). It makes sense that the pink line is to the right of the red line, because we do expect the marginal tax rates to have a negative effect on labor supply (so removing them has a positive effect). What's incomprehensible to me is how he draws any conclusions from this, because he does not simulate what 2009 Q4 labor demand would look like if benefit generosity was reduced to 2007, Q4 levels!
To put it another way, Casey Mulligan sat in the seminar room and essentially told all of us "if you look at the supply effects but don't show any demand effects it looks like supply matters a lot and these programs I'm looking at explain a large share of the downturn". And I was the only one to ask him about it!
I tried to keep the comment simple because there was a lot of trashing of Keynesians in the seminar room. Bryan Caplan talked about how he knew a lot of Keynesians at Princeton and that they are not Keynesians because they care about the data - they are Keynesians because of the feelings of introspection they have about workers. Peter Boettke called it "magical thinking". These comments got lots of chuckles of approval, so I didn't want to come out brazenly in a seminar room that clearly didn't think much of the point (after already having snuck in late!), so I just asked why he didn't simulate a shift in labor demand the way he simulated the shift in labor supply. Mulligan's response was that he did do it in the book along with some other sensitivity tests. What happens in those analyses? No word on that.
What concerns me a lot is that he is going around presenting half a supply and demand model as his headline result, when the labor demand impact of these policies is substantial. Remember, nobody is taxed today to pay the unemployment insurance benefits today. People are taxed, of course, but most of the benefits come out of a trust fund. Food stamps, Medicaid, and tax credits are a little more complicated. They don't come out of a trust fund but they added to the deficit which is financed with money that was not going into investment otherwise.
Multiplier estimates are good to look at to think about the net effect of combined supply and demand shifts because they take both into account, and these estimates suggest Mulligan is making a big mistake by not presenting a full supply and demand story. Zandi estimated that the multiplier for food stamps is 1.73, suggesting that (for that program at least), the labor demand effect that Mulligan ignores is bigger than the labor supply effect that he estimates.
This also makes me wonder what's in the book, but if he is presenting this as his primary result and didn't give me details when I asked, I'm concerned.
My prior on the economics and the policy of the Great Recession is that we had:
1. A big negative demand shock in response to the financial crisis.So in the section above I talk about how Mulligan covers #3 and ignores #4 (or apparently leaves it in his book but doesn't think it's important enough to present). The other problem is that he completely ignores #1 and #2, and tries to assert that he's explaining the Great Recession!
2. A big positive monetary/fiscal/TARP response. You can say it wasn't enough so it wasn't "positive". I'm just saying that it was a big positive relative to doing nothing. My baseline is no policy change, rather than "the policy change that would have fixed everything", which makes it a positive.
3. A small negative effect from the labor supply effects of the safety net.
4. A medium positive effect from the labor demand effects of the safety net.
While I may get an answer to what he thinks about #4 in the book, I doubt I'll get much of anything on #1 and #2.
Notice that endogeneity bias and identification issues come back in the picture here. Even if the observed fall in hours is accounted for by the labor supply effects of the safety net (I don't think it is for the reasons I stated in the previous section), you can't conclude that and say demand side stories don't matter! You have to know whether fiscal and monetary policies improved the economy relative to the counterfactual of what would have happened otherwise.
That's a very hard thing to do, of course. You are going to need a lot more data - more countries and more time. Mulligan said he's interested in this, but hasn't had time to do it. That's fine. What really concerns me is that he's willing to draw conclusions like this before getting the other data.
When you see people like Barro, Romer, Ramey, or Vernon arguing over fiscal multipliers they look at all this data in an effort to generate a counterfactual.
To put it bluntly - based on the seminar presentation - Mulligan seems to think it's sufficient to (1.) note a labor supply effect while not mentioning the labor demand effect, (2.) use labor supply elasticity estimates to estimate the magnitude of that effect, and (3.) determine your counterfactual on the basis of that result and some accounting for population growth and that sort of thing.
I may be misunderstanding something (again... why I think it might be worth looking at the book), but I'm not personally left convince by it.