Skip to content

Interview with Erik Hurst

Interview with Erik Hurst (The University of Chicago)

Paper: “The Aggregate Implications of Regional Business Cycles”

The paper’s main takeaways, according to Hurst:

  • For those of us who want to do empirical work using cross region variation to try to learn about the shocks hitting the economy, it is more complicated than we thought when you’re trying to understand what’s driving the aggregate business cycle .
  • Wages are pretty flexible at the local level. When we start building models of aggregate wage stickiness, we have to be confronting the fact that when you look at the difference between Las Vegas and Dallas, unemployment rates and wage rates correlate very strongly. We need to keep that in mind when we’re doing this. Once you do that, wages look a lot more flexible at the local level than they do at the aggregate.

The Federal Reserve Bank of St. Louis hosted its 40th Annual Fall Conference on Oct. 15-16, 2015. David Andolfatto, a vice president and economist with the St. Louis Fed’s Research division, sat down with each of the conference presenters and discussed their work in plain English. The content below is from those interviews. All interviews have been edited for clarity and length, so the content below should not be considered a transcript. 

Erik Hurst

Can you tell me what your paper is about?


There’s a large literature among many applied economists who try to use regional variation to try to learn what was driving the aggregate business cycle during the Great Recession. People say, “Hey, maybe we have variation. Something happened in Las Vegas. It didn’t happen in Dallas. Maybe we could use the variation between Las Vegas and Dallas to tease about mechanisms that were driving the Great Recession.”

We say there are things to learn about the difference between Las Vegas and Dallas. But it’s much harder to draw aggregate implications. What drove the business cycle from just looking at local variation in GDP and such from these local cities?


Can you give me an example of how other researchers have made a mistake in misinterpreting regional variation and ascribing incorrectly the cause of the recent recession?


There’s a large literature about the role of housing and how it affects households—me and you—in terms of our decisions. House prices plummet, and we feel poor, so we stop spending. Those types of stories, people have done by running regressions, just simple statistical correlations between what happened to your house price and what happened to things like local employment in a given area. And they found places where house prices fell the most, like Las Vegas, had the biggest declines in employment relative to places like Dallas.

The problem with those stories is, when you look at what’s going on between Las Vegas and Dallas, policymakers like the Federal Reserve could come along and offset those types of shocks by lowering interest rates, trying to stimulate spending on the household side, which is exactly what the Fed did in 2008, 2009, 2010 and such.

If that’s the case, any shock like that at the aggregate level could be offset by the Federal Reserve. Because the interest rate is the same in Dallas as it is in Las Vegas, all of that gets differenced out.  

Low interest rates were in Dallas. Low interest rates were in Las Vegas. That increased spending in both of those places. So if the housing shock declines spending, the interest rate shock could have increased spending and those could push against each other.

Think about it this way. You have a simple graph with how much house prices declined on the one axis and how much employment fell on the other axis, and each observation in this graph is a city. What the regional variation gives us is the slope of that line. What it doesn’t give us is the intercept of the line, where the line’s going to, how high or how low the line is.

Anything that moves all cities up and down, you have no way to identify this. We’re thinking hard about this variation between Las Vegas and Dallas. You need some structure to do this. You need a little bit more modeling and assumptions. How does that all aggregate up? Then we can talk about policy that affects at the aggregate level as well.


When you say things like “the housing shock,” my understanding is there is a great deal of regional heterogeneity in the housing shock. Las Vegas’ home prices fell a lot. It’s probably not surprising that the construction sector was hit and construction workers got laid off. That seems kind of obvious. And other jurisdictions were not so hit hard. To a lot of people, this just seems like common sense.

But it’s still not clear to me based on this common sense kind of observation: How does an economist potentially go wrong in identifying what ails the economy and what should be done?


During the Great Recession, nominal wages in the U.S. actually rose from 2007 to 2011 despite very weak labor markets, and real wages stayed right on trend. Real wages have been essentially growing at zero rate since at least 2000, and probably before. During the recession, real wages grew at zero percent.

So you look at the data and say, "Wages aren’t adjusting that much during this recession despite the fact that we had a huge decline in employment." You might think if firms don’t want to hire as many workers, the price of workers should fall. So people have told stories now that there’s some friction: That’s just something messing up how the economy works, and the wage-setting process is part of the thing that makes the recession worse.

In a world where wages don’t fall, certain types of shocks could have a bigger effect on employment, and certain types of shocks, like a housing shock, could have bigger effects in the aggregate level if wages don’t adjust.

Now you go look at the local data. Again, this might not be surprising to a lot of people, but when you go to the local data and you look at Las Vegas and the employment is falling, you see wages are falling like crazy. It doesn’t make sense. Anybody who’s working in Las Vegas now probably knows that their wages were growing at a lower rate than it would have been in the prerecession periods.

When you compare them to Dallas, whose wages were growing relatively robustly during this period, you could see there’s a variation. This causes a puzzle, and this is the whole part of the paper. The first part of the paper is just setting up this puzzle. I think that’s value added itself.

It sounds crazy, but no one has ever looked at local variation in wages during big recessions to get a sense of how wages adjust. We do that. We put together data to try to measure wages, adjusting for the fact that certain types of people might lose their jobs and we might not be able to observe their wages. You have to make some adjustments for that.

Then you look at the data. Now you have how much employment has changed in each city on one axis and how much wages have changed on the other axis. You see that places with the biggest unemployment increases or the biggest employment declines had the lowest wage increases. So that variation across regions doesn’t look like the variation at the aggregate:

  • At the aggregate, big increases in unemployment, no change in wages
  • At the local level, big increases in unemployment, big relative declines in wages

So that’s the puzzle, and this is why the title of the paper is The Aggregate Implications of Regional Business Cycles. If the variation at the local level is different than the aggregate level, what does that mean when the U.S. economy is just the sum of the local levels?


You’ve identified the puzzle. You develop a theory to interpret the puzzle, or understand it. What’s your theory about what’s driving this puzzle?


Now we try to do like I just said: We’re going to take each one of these parts of the U.S. In our data, it’s actually going to be U.S. states. We’re going to make a model of each one of the U.S. states that aggregates up to the U.S. as a whole. Anything that goes on in these local areas eventually has to sum up to what goes up at the aggregate level. In our model, there are certain types of shocks and certain types of assumptions on what the households do, what the firms do, what the government does, etc.

Let me tell you about these shocks, the things that could drive business cycles in general. These are metaphors for the type of shocks that most of us have talked about during the recession. We have some shock that hits the household sector. That makes them want to spend less, and you can think about that as being like a housing shock. People have told those stories that house prices have fallen. People can’t borrow as much money to fund their consumption.

We have one type of shock that looks like that. We have another shock that hits the firms. Those shocks make firms want to hire less at any given rate, maybe like a productivity shock on the part of firms or a borrowing cost shock for firms. Firms might go to banks and maybe not be able to borrow as much.

The third shock is leisure shock. Some people have said the story that something has made us want to work less than we did in the past. Maybe it’s that our government policy has made us want to work less. Maybe something hit certain sectors like manufacturing, and it takes a while for workers to reallocate to other sectors.

Those are the three types of shock. Every one of these shocks has an aggregate component and a regional component. We define the regional components so that they sum to zero. Anything that moves all regions, we’re going to call an aggregate shock, and anything that moves your region relative to the aggregate we’re going to call a local shock. Every shock has those components.


And is it something you’re going to estimate using the data and using the model?


Exactly. Let me just tell you two other things about the model that I think will be important. The first thing is that regions are connected to each other. They could trade goods with each other. As a result, shocks that hit you might spill over into me because if I produce cars and you’re in Las Vegas and you don’t want to buy as much, it’s going to affect how much I’m going to produce there. So we could trade with each other, and that’s one thing that makes the regions different than the aggregate.

The second thing is that there’s a Fed, and the Fed is in the background. They’re moving around interest rates, and the interest rates are the same in all the places of the U.S. Interest rates are the only shock that has no regional component. So we’re going to have a fourth shock I should have mentioned. There’s a Taylor rule, and sometimes they could do things to move us above and beyond the Taylor rule.

Then the last thing is that the assumption I want to have—and this is what we’re really going to estimate—is how sticky nominal wages are. There’s a local stickiness to local shocks and an aggregate stickiness to aggregate shocks. Because eventually everything’s going to aggregate, the local wage stickiness parameter will then map directly into the aggregate wage stickiness.


You introduce this local nominal wage stickiness because it’s going to be necessary for your model to interpret the data?


It’s twofold. One is that those spending shocks in the household sector don’t really have any effects on recessions unless you have some friction in the model. So we needed some friction.

And the second thing is, when you look at the data, it screams that there is some wage stickiness. Shocks that occur today might not affect your wages until tomorrow. Why? For example, being at the University of Chicago, my wages are set once a year. Things that happen in December don’t show up in my wages for a while.


It’s consistent with what we observe at the local level as well?




So you develop this theoretical framework. You’ve got your interpretation. Now you formally estimate using econometric techniques. You choose parameters so that the theory has a best fit for the data that you have. What are your findings? What is the interpretation that your model offers? And how does it differ from the conventional interpretation?


A lot of the parameters that were going to be used are kind of standard in literature. Wage stickiness is not standard, so a lot of our paper is on trying to estimate how sticky wages are. As I told you before, when you look at the regional data, you’re going to find some stickiness in wages, but it’s not huge.

When you go to the aggregate data and shocks, particularly the household spending shock, that shock should cause wages in the aggregate level to fall. Some people have come along and said, “If you believe this shock was the cause of the recession, you should see wages fall.” They say wages must be really sticky.

The part of our paper comes along and says, "If they’re so sticky, then we shouldn’t see wages moving at the local level." And reverse it: ""Given the flexibility we see at the local level, it is hard to get this household shock to do a lot of the work for the recession."

Some people believe that the recession was “demand driven,” or driven by the household sector wanting to stop spending, and that’s the sole story for the recession. We reject convincingly, because if so, wages should have plummeted. What we find is that about 40 percent of the decline in employment in the early part of the recession, around 2009, could be attributed to a demand shock above and beyond what the Fed was trying to fight against.

The reason why employment was still low in 2010, 2011, 2012 had very little to do with the demand shock. If it was all demand shock, wages would have adjusted to clear the market. So something happening to the firms in the economy is what we estimated. For whatever reason, firms didn’t want to hire. Maybe their productivity was low. Maybe they had some friction in their lending in the banking sector at the aggregate economy as a whole. I can’t distinguish between those, and we don’t even try to distinguish between those. Those are kind of lumped together. The whole point was: It wasn’t the household side that was causing a drag on the economy in 2010, 2011, 2012.

We stop at 2012, so anything about the recovery in recent years is outside the purview of the paper.


Is this conclusion or this finding of your paper in contrast to, let’s say, Mian and Sufi?1


Mian and Sufi were really talking about 2009. They were trying to explain that a lot of the early part of the recession in 2009 was driven by the household sector. And some inside the Fed used the work of Mian and Sufi to bolster a demand driven type of recession for which the Fed should be working against.

We’re saying that was part of the story, so they weren’t wrong. But once you go through 2010, 2011, 2012, you can’t have their story being the story, because wages would have fallen unless wages were really, really sticky. And we just estimated from the local level that wages aren’t really, really sticky.

From the Fed’s perspective in 2010, 2011, 2012, maybe they’re propping up the banking sector. But maybe it’s to help the firms in the economy more than to help the households in the economy.


Your findings don’t necessarily preclude a balance sheet recession. But the findings are that whatever these forces were, they moved from the household sector to the business sector.


That is a perfect summary.


Your model fits the data nicely at the local level as well, which I like a lot. It offers this intriguing interpretation of the crisis and subsequent developments. Does the model scream for any kind of obvious policy intervention?


There’s two things that I think the policymakers should take away, and I still hear some of it in the rhetoric even within the Federal Reserve right now about the stickiness of wages.

You hear the reverse sides now. The story you’ll hear within this Fed right now is that wages didn’t fall in 2010, 2011, 2012 because they were so sticky. Part of the reason we’re not seeing the wage increase in 2013, 2014, 2015 is because we’re just undoing pent up demand. I just don’t believe that is a true assessment to have.

I’ll even go further, though this is off-topic just slightly. When you look at real wage growth from 2000 to 2007, well before the recession started, it was zero. When you look at wage growth now, it’s zero.

During the recession, if wages were really sticky, maybe that’s why they didn’t fall. But we’re saying wages are pretty flexible. So what was going on is there is something else pushing up wages.

We didn’t talk about this, but the thing in our model that’s helping push up wages that we are estimating as having some effect on aggregate employment is that there seems to be something causing the household sector not to spend less but to work less, whether that is a choice or whether certain sectors have gotten hit and then as a result people have trouble reallocating to other sectors. The data says that is putting upward pressure on prices.

For us and the Fed policymakers, I think what we’re finding is also evidence of the same puzzle of why employment has not bounced back even as of 2015. The unemployment rate has come down, but the employment rate plummeted from 2007 to 2009 and has stayed low for workers who usually work, people 21 to 55.

I think that same shock explained why wages at the aggregate level didn’t fall and why employment is still depressed in the U.S.


This sounds like an identification of the types of disturbances, the nature that they have to take to interpret the data. But in terms of a direct policy implication…


The traditional policies that I think that could respond to that story aren’t the usual ones we think of: the Federal Reserve and Congress and the president doing short-run stimulus. These might be things like job training programs or retooling workers to move them from manufacturing towards IT or some other sectors.


1 Mian, Atif; and Sufi, Amir. "What Explains the 2007-2009 Drop in Employment?", Econometrica, 2014, Vol. 82, Issue 6, pp. 2197-2223.