An Interview on Balance Sheet Recessions with Amir Sufi

I think that Amir Sufi (University of Chicago Booth School of Business) and Atif Mian (University of California, Berkeley) are doing the most interesting and important empirical work on what is going on with this Great Recession. So I was excited to see that they just released two new papers on the subject, “What Explains High Unemployment? The Aggregate Demand Channel” and “Household Balance Sheets, Consumption, and the Economic Slump.” Here’s an editorial — “How Household Debt Contributes to Unemployment” — summarizing their research. They have used a lot of innovative methods and data sets in order to pinpoint the problem of the “household balance sheet” and how debt-to-income and leverage are linked to sluggish growth and employment. These papers have been covered in the blogosphere already (here’s Paul Krugman, Calculated Risk and Kevin Drum all discussing it.) I was able to interview Amir Sufi about this research.

Mike Konczal: To get started, your papers, and much of the similar work on the matter, show how debt is impacting our slow recovery. In order for this to happen in an otherwise functioning economy, your model introduces three frictions.  Can you describe them?

Amir Sufi: From an academic perspective, most macroeconomics is done within a representative agent framework where all types of people are identical. What that means is that leverage can never really matter — because it is one guy basically borrowing from himself.

So the first main ingredient in this paper — and I think Eggertsson and Krugman have said this in the most straightforward way — is that you have got to have some agents in the economy who are borrowers and some who are savers. For leverage to matter in an economic model, you are going to have to have heterogeneity among households in the model. From a practical point of view, when talking about the real world that’s pretty obvious, but for the macroeconomic model you need to add that in.

The second ingredient is some shock that reduces the consumption of the borrowers very sharply. Both the Eggertsson/Krugman paper and another by Veronica Guerrieri and Guido Lorenzoni make the argument that the fundamental shock is to the ability of the borrowers to borrow — they are forced to either default or massively pay back their debt burdens. In the context that I interpret it in the real world, it is the combination of the decline in house prices that took away the home equity channel, along with the collapse in credit card availability because of the financial crisis. Those are the big shocks that matter.

In my view, those two things are noncontroversial. Even when I present them to more right-leaning economists who don’t believe frictions are so important, they’re are willing to accept these assumptions. The third thing is trickier. The standard response of economists who don’t believe in frictions is: fine, the consumption of these borrowers declines massively, but there’s no reason the economy shouldn’t equilibrate itself by the savers making up for the lost consumption. And what mediates that channel generally is the interest rate. When the borrowers reduce their consumption, the interest rate collapses — it’s like a positive shock to savings demand. At that point the savers, seeing the lower interest rate, should start buying cars, redoing their kitchens, and everything we think people do when there’s lower interest rates.

What you need, and this is where the third thing, the zero-lower bound, comes in, is some friction that prevents the savers from making up the shortfall. And that’s where the liquidity trap stuff really comes in. In order to get the savers to consume more, you need the interest rate to get really negative, but it can’t get negative because of the zero-lower bound on nominal interest rates.

And then you get into all kinds of problems, like the Fisher debt-deflation stuff. The normal way you try to get real interest rates negative is through expected inflation, but the only way you can get expected inflation is if you force the current price level down, which is deflation. But the debt burdens are written in nominal terms. If you push the price level down, you get this vicious cycle where the borrowers cut their consumption by even more.

I want to add that the third ingredient — the friction — doesn’t need to be the zero lower bound on nominal interest rates. But that is what has been articulated in the theory work most prominently.

Reading much of the zero lower bounds literature now, it strikes me that it was a conversation among a handful of Princeton and New Keynesian academics when it first started. But looking at it now, it seems very obvious that the zero-lower bound creates a challenge. If right-leaning economists don’t think the zero-lower bound is a friction, what do they think?

I think that right-leaning economists don’t deny that the zero lower bound could be a friction. I think the zero-lower bound does bother them, that they think it is a fundamental friction. I think where they’d disagree with Paul, and to an extent even I disagree with Paul, is that if you look at his model, the optimal policy in those models isn’t necessarily fiscal stimulus, it is writing down the debts of borrowers. That’s the number one policy that fixes the problem.

Gauti and Paul’s model in particular has a tightened borrowing constraint on borrowers that pushes down their consumption, which in turn leads to zero lower bound problems. The quickest and most effective way in their model would be some type of transfer from the savers to the borrowers to offset this dramatic decline in consumption. Principal forgiveness is exactly such a transfer. Fiscal stimulus is a form of this transfer where we borrow from future generations to make up for the shortfall in demand. But it strikes me as much less direct and potentially more distortive than principal forgiveness.

I come from a finance micro background, so if I were to criticize the zero-lower bound literature, which I use, it is that fiscal stimulus doesn’t fall so naturally out of it. Paul goes to lengths to argue against the argument “how can more debt solve a debt problem?” and explains it is because the borrowers are constrained, and there’s some truth to that. But the fundamental problem in these models, what generates the zero-lower bound problem, is a sharp reduction in consumption by borrowers. Why not attack that problem head on? If you look at Rogoff’s opinion against Krugman’s, I think this is the main difference. They agree on the zero-lower bound nature of the problem, but have different tactics on how to fight it. I tend to agree with the view that directly targeting the household debt problem seems to make more sense than fiscal stimulus.

Looking at these models, the real world implication is that a sharp drop in housing prices and a subsequent increase in debt-to-leverage should cause a decrease in consumption. But it isn’t necessarily clear why this must be the case. Most of the papers don’t develop this, often taking a debt limit as exogenous, though one could imagine people going about their spending decisions in much the same way before or after a housing crash. I was wondering if you have an answer for this.

I think there’s a few ways to think about it that we outline in our consumption paper. Why does the shock lead to such a strong reduction in consumption? One, and this is based on previous research, a lot of the consumption by the indebted households during the housing boom was being financed through home equity withdrawal. So just mechanically, consumption can’t stay at the same path because they no longer have their homes to borrow against to finance consumption.

The second thing is that the deleveraging effect is real. The Survey of Consumer Finances shows that up until the 90th percentile of the distribution, as of 2007, made up around 65 percent of people’s net worth. If you see a massive decline in the value of your home, it is kind of mechanical that if you are thinking about savings and retirement you’ll think, “I was planning on having enough equity in my home when I retire that I could just borrow against it for the rest of my life. Now I don’t, so I have to adjust my consumption path immediately.”

The third thing comes from the credit supply channel. These guys can no longer refinance into lower rates, therefore their income in a relative sense goes down because they can’t get these lower interest payments. Hence you’d see their relative consumption against those that can refinance decline. Also, the act of delinquency itself reduces consumption — your credit score is shot, foreclosures have an effect on durable consumption. Regardless of what you think of the theory, the empirical evidence in our stuff is undeniable: highly indebted households see very sharp relative declines in spending.

A lot of the underwater mortgages in the country are in a handful of states. How well does your research deal with local conditions? Many of these places would have had a major housing construction boom too.

We take on the local difference, and this is where our housing supply elasticity instrument becomes so critical. You are correct: if you unconditionally look at the high debt-to-income places, a lot of it is highly correlated with places that had construction booms and a lot of migration. To get rid of the construction and migration effects, we try to use exogenous variation in debt-to-income ratios that is driven by housing supply elasticity. This is a technique called “two-stage least squares.” We regress the debt-income ratio on how hard it is to build in an area. The idea is that this instrument allows us to disentangle the effect of debt levels from construction and migration. The results after doing this are very strong. Any place that had a high debt-income ratio, whether or not it had a construction boom, is suffering massively now.

A response to your argument is that the savings rate hasn’t gone up that much. It’s up, but lower than historical averages and has stayed pretty much in the 5 percent region. If your theory relies on people saving more, is this a problem?

First, it is in some sense about the derivative. It might not be high historically, but it is high when compared to 2002 through 2007. Every single quarter since 2008, it’s been higher than during the expansion preceding this recession.

But the more fundamental issue is that the savings rate is a very misleading number because it is endogenous. It’s easiest to see this argument by looking at (1 – the savings rate), which is how much consumption you are doing out of your current income. It is true that consumption is not that much lower relative to total income than it has been historically. But of course that is a silly way of looking at it. The point is that consumption is way down, period. We don’t care whether it low relative to current income.

Once you think about it this way, the problem with the savings rate becomes a lot more obvious. The right benchmark for judging whether consumption is low or savings is high is not relative to current income, it is relative to the consumption you had before the recession. When you say, “People don’t seem to be consuming that much less as a fraction of their total income,” I say, “Who cares?” The point is that their income is way lower precisely because of the recession. And their income is lower because everybody is consuming less. This is why the savings rate is kind of a silly number when talking about a deleveraging recession.

The right way to look at it is to say how much has consumption fallen since 2006. It has gone down tremendously! And that’s prima facie evidence that consumers are deleveraging. People are earning less, because they are consuming less, which is the essence of the deleveraging argument.

Finally, you also have to take into account that interest rates are basically zero. If interest rates are zero, then people are really saving a huge amount of money, because they are saving 5 percent at a zero interest rate. You have to adjust for interest rates to determine whether or not savings rates are historically high.

Another response to this model is that the debt-to-income ratios don’t actually matter that much. What is really driving this is a wealth effect. People feel poorer from losing housing value, and thus they spend less. James Surowiecki just had a piece arguing against these balance sheet recession models in The New Yorker, “The Deleveraging Myth.” Dean Baker from CEPR makes this argument as well.

Well obviously I disagree 100 percent with that for both theoretical and empirical reasons. The theoretical reason is that housing should not be thought of in a pure wealth sense. We all have to consume housing going forward. And the value of my house going down is also the same value of the price of housing going down. The easiest way to imagine this is to picture a young couple that currently rents and will buy a house in the future. If housing prices decline, it is good for them because they can then more easily buy a house in the future. Clearly, this is not a negative wealth effect for the young couple.

But as far as I understand it there are studies that find a wealth effect in housing.

This is a semantic point on what you call it. I’m saying as an economist that if you call something a wealth effect, then it has nothing to do with borrowing constraints and debt levels, and that effect in theory should be zero. To the degree that we observe that when people’s house prices go up they consume more, that’s not a wealth effect — that’s a borrowing constraint being alleviated, and people borrowing against collateral that they couldn’t before. Which is a very different thing, and it matters empirically. My own research on this topic shows definitively that people consume aggressively out of housing wealth because of borrowing constraints, not a simple wealth effect.

Here’s why I fundamentally disagree with the “wealth effect” argument. Suppose you have an economy that looks like the U.S. before the recession, where you have an extremely skewed net wealth distribution. The wealth effect argument is that the response of the economy to house price declines would have been the same if you flattened that out versus if you had the polarization that we have now. And I disagree with that fundamentally, and that’s what the research shows. The net wealth distribution matters. People who have very high debt-to-income ratios cut their spending very dramatically, and there is no way a pure wealth effect can explain the magnitude of the cut.

How necessary is debt forgiveness?

I’ll say this: We are about four years into this mess, and we still don’t have any sense what the elasticity of consumption would be with respect to principal forgiveness. The reason we don’t have that estimate is that there’s been no principal forgiveness government programs. Of all that has been allocated, there’s been virtually nothing allocated to principal reduction to see if it works.

I’m not willing to come out and say principal forgiveness will solve all of our problems. But at the very least, shouldn’t we have some basic idea of how responsive spending of highly indebted households would be to principal forgiveness? We’ve tried a ridiculous number of things in terms of government policy during this downturn: fiscal stimulus, homebuyer tax rebates, cash for clunkers, etc. Can’t we at least give principal forgiveness a chance, even if it is on a very small scale?

Any concluding remarks?

The distribution of net wealth matters a lot. Let’s suppose there’s $100 of wealth in the economy and there’s a hundred people. If everybody had $1 of wealth, and then there’s a massive drop in house prices, my argument is that this recession wouldn’t have been nearly as severe. It’s because the five guys at the top have all of the $100 and are just lending to the other 95, that’s why the recession is so severe when house prices collapse. Paul said this a few times on his blog, and he’s usually very clear, but I don’t think he’s been clear enough on explaining this. These models on why deleveraging matters are all about the net wealth distribution. We shouldn’t be surprised that this recession and the Great Depression were preceded by very large increases in wealth inequality. This is well documented during the 1920s and the 2000s. This is why I get a bit annoyed at the guys who are saying it’s just a pure wealth effect, because it’s something bigger than that.