Skip to content

Christopher Carroll, Evening Keynote Address

February 5-7, 2013 | St. Louis Mo.

previous  view previous video in this series | view next video in this series  next

about the event | conference materials

   Julie Stackhouse (4:19)
   Michael Sherraden (7:42)

The St. Louis Fed's Household Financial Stability Research Initiative
   Ray Boshara (9:08)

Session One – Household Balance Sheets: Status and Perspectives
- The Current State of U.S. Household Balance Sheets
   John Sabelhaus, Kevin Moore and Paul Smith (22:09)

- Why Did So Many Economically Vulnerable Families Enter the Crisis With Risky Balance Sheets?
   William Emmons and Bryan Noeth (22:31)

   Session One Moderator: Jim Fuchs (4:16)
   Session One Q&A (25:28)

Session Two – Household Balance Sheets: Homeownership and Retirement Security
- The Effects of Health and Wealth Shocks on Retirement Decisions
   Dalton Conley and Jason Thompson (19:12)

- Homeownership, the Great Recession, and Wealth: Evidence from the Survey of Consumer Finance
   Michal Grinstein-Weiss (7:22)
   Clinton Key (11:47)

   Session Two Discussant: J. Michael Collins (19:50)
   Session Two Moderator: Yvonne Sparks (2:29)
   Session Two Q&A (19:41)

   Introduction: Ray Boshara (4:21)
   Keynote: Michael Barr (56:23)

Session Three – Household Balance Sheets: Education and Social Development
- An Experimental Test of Child Development Accounts on Early Social-Emotional Development
   Jin Huang, Michael Sherraden, Youngmi Kim and Margaret Clancy (18:18)
   Session Three Discussant: Robert Pollak (19:37)
   Session Three Q&A (26:45)

Evening Keynote Address
   Introduction: Christopher Waller (3:06)
now playing  Keynote: Christopher Carroll (1:02:19)

February 7, 2013

Breakfast Keynote Address
   Introduction: James Bullard (3:30)
   Keynote: Governor Jeremy Stein (46:29)
   Keynote Q&A (11:22)

Session Four – Household Balance Sheets: Deleveraging and Economic Growth
- What's Driving Deleveraging? Evidence from the 2007-2009 Survey of Consumer Finances
   Karen Dynan and Wendy Edelberg (22:08)
   Session Four, Part 1 Discussant: John Krainer (20:16)
   Session Four, Part 1 Q&A (14:47)

- Household Balance Sheets, Consumption, and the Economic Slump
   Atif R. Mian, Kamalesh Rao and Amir Sufi (21:57)
   Session Four, Part 2 Discussant: Brian Melzer (18:51)
   Session Four, Part 2 Q&A (15:44)

   Session Four Moderator: Daniel Davis (2:14)

Closing Plenary – Facilitated Panel Discussion: Household Balance Sheets and Economic Growth
   Panelist: David Buchholz (10:24)
   Panelist: Steven Fazzari (11:41)
   Panelist: Deniz Igan (12:13)
   Discussant: Barry Cynamon (20:03)
   Panel Discussion Q&A (27:29)

   Ray Boshara (2:09)


Below is a full transcript of this video presentation. It has not been edited or reviewed for accuracy or readability.

Christopher Carroll: Thank you very much. I'm delighted to be here and would like to thank the organizers for putting together such an interesting conference and inviting me to speak here. I'm going to start with Aristotle, which is I'm sure the oldest reference that anyone will make in this conference. Aristotle in 350 B.C. compiled a list of sort of the most common reasoning errors that people make. And one of those items in his list was what he called the fallacy of division, which is attributing characteristics of the whole, of some whole object, to each of the parts of that object. So if you go on Google today and you search for the fallacy of division it will not take you long before you come up with an example. I have made a strategic substitution, which you will understand in just a second. But an example of the following kind: America is rich. Chris Carroll is American. Therefore Chris Carroll is rich.

Now this is clearly an error. Google does not have this quite right. But Aristotle was way ahead of most macroeconomists, at least most macroeconomists as of 2008. I think perhaps the fallacy of division is in the process of being recognized as a fallacy in the macroeconomics profession now. But the representative agent models that dominated a lot of macroeconomic thinking before the crisis took the form basically that well, it's not important to worry about how household balance sheets are structured and how much debt there is and how much assets because yes, there are some people who have debt; there are other people that have assets and some of that debt is—you know, one person's balance sheet is an asset on somebody else's balance sheet, so that debt is owed to someone and as a result all that really matters is how does it all net out. The distribution of things across asset categories and debt categories, how could it possibly matter? Well it can't in a representative agent kind of a model.

Now it's true that one advantage of representative agent models is that they are simple. Assuming that there's only one person in the economy does make your life easier if you want to analyze that economy. You have only one set of expectations to compute. You have only one possible set of outcomes. But it of course has for a long time been kind of a dissenting point of view, which I hope eventually will become the dominant point of view, that says the gospel of the representative agent is missing some really crucial things.

And so that's what I'm going to try to elaborate on in a fairly kind of hard-core way this evening. But before I do that in my prelude to the overture of the balance sheet model let me remind you of something that you have already been reminded of several times today, which is the argument of the sort of representative agent, dynamic, stochastic, general equilibrium, macroeconomic camp that was a popular argument before 2008, which is that well geez, look at aggregate household net worth in the U.S. I'm plotting it all the way back to the beginning of the data in 1959 and is there any reason to worry about the low saving rate that the American economy is exhibiting in 2007 when well we have the highest wealth-to-income ratio that we have had on record or very nearly.

There was of course the alternative camp of people who emphasized the other graft that you've seen several times today about debt-to-income ratio, which is also at the highest levels ever seen in 2007. And the representative agent camp tended to denigrate the people who are worried about the ever-rising household debt and on occasion treated them as I say here, as people who were sort of roughly on the same par of rationality as the believers of the Mayan apocalypse. But the apocalypse came, not the Mayan one but the one that the debt worry warts had warned about. Now one must acknowledge that saying well the debt-to-income ratio is the highest level it's ever been did lose a little bit of its worry factor after it had been true for 22 years in a row. So it wasn't that easy to know in 2007 that that was really the year in which the apocalypse was about to happen. But nonetheless there were people who were worried about it.

So that was kind of the intro/prelude. Now I want to talk about sort of what have we learned since then and I would characterize, I think there are three really interesting and important data points or sort of characterizations of data that all line up in the same direction here. They sort of all have the characteristic that if you have two entities that you're comparing, Entity I and Entity J, and I had a greater run-up in debt before the crisis, then in the crisis Entity I suffered a worse decline in consumption or any other measure of sort of economic output than J.

And this is true of countries, so the IMF had a very nice study on that subject. It is true of states and localities. Amir Sufi is here somewhere and he and Atif Mian had done some very nice work showing that those states in the U.S. that had big debt run-ups had a much worse experience in the recession. And Karen Dynan and other people but Karen is the one whose work I am most familiar with, has pointed out the same thing is true with households. Now this is like, speaking of apocalypses and planetary alignments, it's never true that data at the country level, at the state level, and at the household level all tell the same story, almost never except now. So when that happens you suspect that there's a real signal there, that it actually means something. It's not actually totally clear what it means, however. It seems to be pretty strong prima facie evidence that balance sheet conditions matter somehow. But part of the point of my talk is to try to tease out whether we know what it means or whether we don't know what it means.

So since I will assume that everyone in the audience, or all of the economists anyway, is familiar with the seminal works of Amir and Karen, I will present only the evidence from the IMF, which is an obscure organization in Washington, D.C. that also did some interesting work on this subject showing a whole bunch of countries, each of which gets a data point: the run-up in the household debt-to-income ratio during the sort of boom years of 2002 to 2006 on the horizontal axis,—and it's pretty impressive actually how big a range there is there and how many countries there are that had big run-ups in the debt-to-income ratio—and the decline in consumption that was experienced by each country in the 2007 to 2010 period. This was from the third chapter of the World Economic Outlook last year. This is of course as you can tell just by looking at it that the ocular regression tells you that it's a highly statistically significant relationship and it looks very much like the famous figure in the Mian and Sufi paper.

But what exactly does it mean? What kinds of elements do we need in a model that might hope to match these interesting data points? Well the first key element that you need to have any hope of grappling with these data is you need a model where there's not perfect foresight, right? If there were perfect foresight we wouldn't have gotten into the mess that we got into. There were an awful lot of people who made an awful lot of decisions that they regretted ex post and would not have made if there was perfect foresight. Of course it's a hard to thing to build sort of stochastic elements, important, big stochastic elements, into models where people are trying to sort of be forward looking at the same time.

The simplest thing that I know of to do that's sort of reasonably tractable that you can work with without too much difficulty is to have a model in which there is labor income risk. So households are worried about losing their job or they're worried about having to take a job with a lower level of income. The model I'll be talking about is one where you can interpret the thing they're worried about is just a decline in their permanent income.

So all we really need here is that even if there is an unemployment insurance system it's imperfect. You're not completely indifferent to whether you lose your job or not. Not a very hard assumption I think. But we're also going to need a framework for thinking about this in which people are different from each other in something other than did they lose their job or not. And the reason for that again if you want to have a framework where you have something useful to say about balance sheets and about the role that they play, well if everybody had the same employment experience or the same employment expectations they would all have basically the same target balance sheet. We wouldn't really have diverse household balance sheets. We wouldn't have people who were in debt and other people that had asset positions sort of ongoing over the long run. So you need to have some kind of heterogeneity across people in order to have a meaningful model of the role the balance sheets might play.

For those of you who are not economists I apologize for all the Greek letters here. For those who are you'll be familiar with all of them. I'm working here with a standard model in which you have people who sort of are looking forward and they're thinking about being worried about losing their job and they have some expectations of income growth over time. They believe that there is some probability Mu, which sort of looks like a U, and so it's going to be my unemployment variable. And there is an index Kappa, which is sort of like a K kind of a sounding variable and it's about credit and so you can remember what Kappa is that way. And they have beliefs about all of these things: the future unemployment risk, the future growth rate, the availability of credit in the future. And we, as I said a minute ago, need a model in which there are sort of different kinds of people in order to have a framework where there are people with different balance sheet positions sort of in the long run in equilibrium on average.

The simplest way to talk about there being two kinds of people is to say there are debtors and creditors. Now of course the real world is more complicated than that but I think the sort of key lessons that you get out of more complicated models can be distilled to their essence in a model where you just have these two categories, debtors and creditors. Now you can create a model that has debtors and creditors by assuming that people differ from each other in any of a number of different possible ways. So you could say okay, we've got young people and old people and young people anticipate rapid income growth over their life cycle, and old people are looking ahead to maybe retiring soon. And so that's why people have different targets. When they're young they want to borrow. When they're old they want to save. Or it could be they're all the same age but some of them are just optimists and some of them are pessimists. Some of them think that we're going to experience a new economy indefinitely and others think that we're all going to hell in a handbasket.

Or maybe the difference across people that's most important is some are risk averse and some are risk tolerant. But my point here is that it doesn't really matter very much for the purposes I'm going to be pursuing here. You need to make some assumption that causes people to have different desired positions. Some people don't want to be in debt and other people don't want to be saving.

The simplest thing in the sort of mathematical, analytical terms is to say that the difference across people is in their time preference rates, so some are just more impatient than others. I find that plausible. Some people have a theological objection to the proposition that certain people are more patient than others. For those who have these theological reservations just interpret the model as being about young people versus old people or something like that. But the crucial point is that it does matter, in models where people are different from each other, who it is that has the wealth. It matters because those people behave differently: the ones that are in equilibrium at a high level of wealth and those that are in equilibrium with a low level of wealth. So the model is a fully specified, numerical dynamic optimization framework with all of the usual bells and whistles.

To continue making things as simple as possible I'm going to assume that half of aggregate income goes to the people who are debtors and the other half goes to people who are creditors. And if you want to sort of make the further leap, which isn't necessary but it is plausible, that the creditors are let's say the top 10 percent of income earners in the U.S., earn maybe half of all aggregate income, and the bottom 90 percent earn the other half, then that might help you get an intuition for how things work.

So we're going to assume that all of the debt belongs to the impatient types and we'll call those people the poor people. They are poor in equilibrium in the model, at least in their ratio of wealth to income. And there is a lot of reason to think this is a plausible way to focus the model. Steve Fazzari sent me a while ago a paper with Barry Cynamon that had tried to calculate what portion of the debt run-up that we have seen over the last decade or so in the U.S. is attributable to sort of the bottom 95 percent I think they look at instead of the bottom 90 percent. And they estimate that almost all of the run-up in debt was concentrated in that set of people, so I think this sort of breakdown between the debtors and the creditors is perhaps not as much of a distortion or a simplification as you might think.

In any case the aggregate net-worth ratio in this economy is going to be basically the assets of the rich households minus the debts of the poor ones. The rich households aren't going to have any debt. The poor households aren't going to have any assets. Each of them gets half of aggregate income so the aggregate is going to be the assets minus the debt. So I'm going to set things up to try to calibrate this model roughly to match some aggregate statistics. So in particular I'm going to say well we want to match the aggregate wealth-to-income ratio that we had in 2001 and the aggregate debt-to-income ratio that we had in 2001. And if you're interested in the details this requires a difference in the degree of patience between the patient and the impatient people of about eight percentage points a year. As I say you could motivate it in a number of different ways. There are other parameter values in the model that I won't go into. There's a paper of mine with a co-author that sort of lays out all the gory details of how all that works.

But what we want to do is take this model and say okay, what do we need to do to make that debt-to-income ratio for the economy as a whole do what it did in that figure that I showed you earlier, which is roughly to rise from debt of one year's worth of income to debt of about 1.3 years' worth of income between 2001 and 2007. So we can make the debt-to-income ratio rise in a number of different ways. We can say that there was an expansion in the availability of credit. We can say that there was the Great Moderation and so people became less worried about income risk or we can say people became more optimistic about future growth. So what we're going to do is compare those three different experiments and see what implications they have for us and whether they have the same thing to say about the role of balance sheets or whether they have different things to say about the role of balance sheets.

So in order to make it perfectly clear what the nature of the experiment is, in my analysis here I'm going to only change one thing in order to generate this debt run-up. I'm not going to change the actual evolution of the path of the economy. I'm only going to change expectations. So we're going to have exactly the same actual outcomes for the economy: the same aggregate income, the same interest rates, the same growth rates across all of these experiments. The only thing I'm manipulating is the title of my talk: expectations. And that's going to be sort of a key point is that expectations are really powerful in all of these models. They're more powerful than the sort of intrinsic, inherent dynamics that come from the evolution of the economy in response to expectations.

So the first experiment we're going to do is we're going to look at what happens if you have a credit boom. It just gets easier to borrow. And we all know there's a lot of reason to think that's a plausible description of something that happened in this period. But it's also true that this is a period when a lot of economists were writing papers about the Great Moderation and how the aggregate economy had been stabilized and the fed had learned how to run things so that we never had to worry about recessions again. That's a little bit of an exaggeration of what people were saying but at John Sabelhaus, wherever he is, has produced some evidence that household-level income shocks declined during this period. So it wasn't just the macro economy but maybe even the micro economy that looked like it had stabilized a bit. So that's the second kind of plausible thing that you could argue might have something to do with why people felt less need to save during this 2001 to 2007 period.

And then a final thing is maybe people believed that we were entering into a period of faster growth. There's globalization and there's the human genome project and there's a whole host of reasons. There are always reasons why optimists might say that we're going to have faster growth going forward. And so let me re-emphasize again that in the experiment that I'm going to be showing you none of these beliefs is actually true. I'm just going to be showing you the effect of people having these beliefs. So it's all about expectations here.

So unemployment in fact in my simulations will remain constant even if people are believing that the risk has declined. The growth rate will remain constant. Credit availability does not change. And what I do in all three cases though is to say okay, people's beliefs returned in 2008 to what they had been in 2001. So there's this period in which expectations sort of go off kilter and then reality sort of comes crashing back in 2008. It's not implausible to imagine that in 2008 people actually got even more pessimistic than they were in 2001, but I want to make the experiment as simple as I can and so all I do is sort of make reality come crashing back.

All right, so before the credit boom what do I do? I basically just look for the amount of increase in the availability of credit such that the debt-to-income ratio in 2007 matches the data for 2007 and then I have this abrupt reversal. So here's what happens and where we get into the balance sheet mattering. Here's what happens for the two different groups: the more patient group and the less impatient group, the creditors and the debtors. Both groups have exactly the same change in their beliefs about credit availability but of course the debtors are the ones that want to borrow. The creditors are the ones that want to save. They want to lend. And so the creditors don't actually change their behavior very much because they weren't feeling very constrained. They were the ones that didn't want to borrow. The debtors however have a party. Credit conditions have gotten a lot easier and these are not sort of irrational; they're not crazy; they're just a little bit impatient. They're the ones who are anticipating a future that is brighter than the present so they want to borrow and they do.

And so you have a big decline in the saving rate just of the debtors, not of the creditors. All right. So that's what happens if you make it easier to borrow and that's what the boom was about. Here's what happens if in 2002 everybody suddenly decides that we are in this great moderation period. I believe the first of the Great Moderation papers was published in 2001 and so of course everyone in the economy read that article in the Journal of Economic Dynamics and Control or wherever it appeared saying we were in the Great Moderation. And again what you see is that basically the creditors, they weren't that worried about unemployment risk anyway. Maybe they're about to retire or maybe they already retired and so these are people whose life is not going to be ruined by the loss of a bit of employment income.

On the other hand the debtors are scared to death when the unemployment rate goes up, or their beliefs actually about the unemployment rate goes up. And they again hugely increase their saving rate. Where's Daniel Cooper? He produced a little bit of, a smidgeon of evidence from the PSID earlier today that what we've seen in these two figures is actually a good representation of the data. I'll let him wiggle his head back and forth a little bit if he wants to but you did show that it looked like the saving rates had really declined among the debtors much more than among the creditors. So these stories are slightly different in their profile but they both say roughly the same thing.

A story that says something different though is what if everybody got a little more optimistic about growth? Well here actually the people who really react a lot are the creditors. The debtors are sort of short-horizon kinds of people. They're thinking about the next year, the next two or three years, a fairly short timeframe. And so telling them that 50 years from now you are going to be a lot richer than you thought you were, 50 years from now is sort of way beyond the horizon that matters to them, while for the creditors 50 years from now or 25 years from now is a very lively prospect to them. That's their horizon. They are the patient ones. They're thinking in the long term and it turns out that if you tell someone who is thinking about sort of an infinite horizon or a very long horizon that growth is going to be lower over that very long horizon, that has a huge effect on their behavior.

So what we learn from this is that it does actually matter a lot what exactly the nature of the experiment is that caused the aggregate spending rate to drop a lot during the period 2001 to 2007. This is the average of the saving rates of the two groups which is going to be the aggregate saving rate in this model. And as you would be able to tell from the previous graphs either the credit bubble model or the Great Moderation model would say that you get a pretty substantial decline in the aggregate saving rate. But the model that really says the saving rate would have dropped like a rock is the model in which you explain the increase in debt of the debtors by the change in income growth that you need to explain that change in debt of the debtors. That much change in growth that if it affects the debtors by debt goes from 1 to 1.3 it affects the creditors hugely more.

And so it's not the case that all different stories about how you might cause the balance sheet to get out of kilter are all equivalent to each other. We can distinguish these stories from each other using data on how different people of different types behaved over the course of the crisis. This is just a comparison of what actually happened to the saving rate over the longer period and I have a paper that has a model that says mostly what happened over the long term is in fact an increase in the availability of credit. But that's sort of a side point. I just wanted you to see what the figure looks like.

So the bottom line is, the key point that I want to emphasize, is that sort of in the short to medium run—and that means over frequency time periods like a decade—aggregate dynamics really get jerked around by changes in expectations. There are some dynamics that you can see here when you go from this point to this point that are driven by the sort of internal dynamics of balance sheet adjustment from one period to the next or that they're there or all of these movements after the crisis is over reflect sort of the internal dynamics of balance sheets. But from the standpoint of the Federal Reserve, from the standpoint of someone who is trying to manage macroeconomic dynamics, working out the sort of quarter- to-quarter, year-to-year accumulation/decumulation dynamics of the balance sheets is not going to get you all that far.

On the other hand it's crucial to know what the sort of long-term, stationary differences are in the balance sheets. Okay, so we saw there were huge differences in the reactions of the patient people versus the impatient people. And you need to know what the structure of the balance sheet is sort of on average, roughly across the population in order to get the distributions of who's going to be doing what right. But in order to get the short term, you know, how is the economy going to behave over the next year, right, well you need to know what's going to happen to expectations over the next year.

So to sum up here the models say very strongly that it matters whose expectations change and that the debtors in particular are not surprisingly much more responsive to changes in the availability of credit. The creditors of course—and this maybe is a little bit of a surprise to people who are not intimately familiar with these models—the creditors are actually much more responsive to growth expectations and it seems unlikely that you're going to be able to do a good job of explaining the data that we have by saying well there were big changes in growth expectations. It seems much more consistent with what limited evidence we have to imagine that it's one of these two things: the credit or the unemployment fears.

There are some lessons that the model I think really screams out about how we can improve our understanding of macroeconomic dynamics and our ability to analyze it on a high-frequency basis. For one thing when we have surveys that gather information on balance sheet positions across people, knowing just what the balance sheets' positions are isn't really enough. You want to know what people's expectations are. You need to know what people's expectations are and how those expectations are changing if you want to figure out what the implications of those balance sheets are, at least for those sort of high-frequency kinds of things that the Federal Reserve worries about.

And a related point is that the model has very clear and plausible I think implications about differences in saving rates across groups. Unfortunately we don't have any good data on that. We have really, believe it or not, no really persuasively high quality data on how different groups of people within the U.S. population save. We know what their wealth positions are every three years from the survey of consumer finances but on a higher frequency basis and on leaving out capital gains and losses, we just really don't have good data. The data that Daniel was showing earlier from the PSID is the best that there is and I suspect he would agree with the proposition that it is highly to be regretted that the data that we have is not a lot better than that PSID data. And so I would conclude by giving my personal view that the data that the Federal Reserve needs to understand how the economy evolves in either the short run or the long run do not exist and that the Federal Reserve needs to create that data. We are doing all sorts of exciting monetary policy things like quantitative easing and even more interesting things like public-private investment partnerships. But if we knew how the economy works better we would have a much better ability to judge which of those things are going to work and how they're going to work. And we're not likely to learn the answer from just waiting for another wave of the existing data surveys to be made available. And so if the Federal Reserve Bank of St. Louis has a project on improving our understanding of household balance sheets then I say let's really do it and let's improve the available data instead of just analyzing the bad data that we have now. That's where I will stop.

I'll be happy to take questions. Amir.

Amir Sufi: So obviously we are not very sympathetic to the idea about how heterogeneity matters so let me I guess for the non-economists I guess I'll show how them combative we can be. So let me just kind of quarrel with one of your scenarios, the crediting one. And let me make the argument, just pose as a question: I'm not sure that's really about expectations. So let me put the extreme. Suppose the impatient guys are extremely impatient, so much so that they really almost don't even care about their future self tomorrow. And suppose that there's a belief or expectation of a credit boom, which in my world I'm just interpreting as a looser credit constraint today. So then I eat and I keep eating until credit constraint gets tighter and them I'm forced to basically save. And then that story I think as I've just told it, there's nothing about expectations going on. I'm just a hand-to-mouth guy. The other two on unemployment insurance and growth I completely agree.

Christopher Carroll: You're right analytically and I think you're wrong as a description of the evidence that we have. If you look at the predictive power, for example, of the Michigan survey's unemployment expectations index for either the aggregate saving rate or for consumption growth, you'll find that that variable has more predictive power than almost anything else. And in your scenario so there's a famous paper by Greg Mankiw, or a couple of famous papers, in which he advocates the saver-spender's model of understanding where the aggregate capital stock comes from and aggregate consumption and that sort of thing, which is sort of roughly along the lines of what, Amir, we're saying. There's a set of people who just they are hand to mouth. Every dollar they get they spend that same day. And then there's a set of people who are sort of the patient, infinite horizon types of people. You cannot make a model like that match the fact that unemployment expectations matter a lot for high-frequency macroeconomic dynamics. You need to have people who are not completely in the "I'm going to spend every dime that I get" mode in order for unemployment expectations to play the quite substantial role that—you know, I have this paper that I mentioned earlier, that says quite a substantial part of this big increase in the saving rate in the Great Recession was due to the huge increase in unemployment expectations. And that model is estimated sort of overall of the history up to before the Great Recession and finds that unemployment expectations systematically have played a big role in movements in the saving rate. So I agree. If there were no evidence that a precautionary motive matters then the Mankiw saver-spender model would be more attractive than this one because it's simpler. But I think that just misses a huge portion of what the business cycle tends to actually be about.

There's also a fairly substantial microeconomic literature that finds that people who face higher unemployment risk or higher uncertainty tend to hold larger amounts of assets.

Other questions. Yes, Deniz.

Deniz Igan: I'm going to echo what Amir said a little bit. It's like the credit boom started to have—and I also have some difficulty seeing this as an expectation story. But I'm going to pose a slightly different one and that will say instead of saver/spenders, patient/impatient, I'm going to say how about haves and have-nots? So this is also a period where we saw a lot of increase in inequality. Maybe those were the guys they just wanted to stay where they were. They wanted to remain as middle class and the way to do it at that point, without an increase in their income, was to just to borrow and so that they can afford their homes; they can afford their houses and their cars and whatnot. So this would both explain and I'm wondering how it fits with your story.

Christopher Carroll: I'm sympathetic to the possibility that that may be an important explanation as well. From the standpoint of sort of my main theme tonight though, which is that in order to really have a hope of understanding the movements in the saving rate, the movements in aggregate demand, you got to see what the balance sheets look like at the microeconomic level and how they are evolving over time. And that would be true in the case of your scenario as well. I am sympathetic to the proposition that sort of middle class people in America were behaving in a way that might have been a bit difficult to rationalize in a model where everyone sort of recognizes reality and recognizes income stagnation for what it is in real time. But I want to have a framework that makes my key point about how important it is to measure balance sheets, that basically any economist worth their salt is going to have to acknowledge has some power to it.

There are people in our profession who tend not to be very sympathetic to the people that sort of had foolish expectations for many years in a row but those people can't really quibble with what I have done. If I had done what you wanted me to do they would be able to roll their eyes and run out of the room and call me a maniac. But they have more trouble doing that in a model where everyone is rational and optimizing and forward-looking and all the I's are dotted and all the T's are crossed. And it still matters a lot where the wealth is, not just how much there is in the aggregate.

So, Jim.

Jim: Okay, so expectations are important. I can buy that. So what is going on in the experiment where during the period where they're making expectational errors is there any learning going on about the errors that they're making? Are they converging to some, I don't know what, alternative steady state and then this all gets shattered? And critically did they know that their belief is going to come to an end at some point?

Christopher Carroll: No. So they don't know that they have made a mistake until 2008 when they all of a sudden realized that. I did it this way because sort of there's a host of plausible, sort of more realistic scenarios that would involve learning or catching up or that sort of thing. But again it would be hard to argue that it's clear that this model is better than that one and you would also really want if you did that to have a model where in some sense there was a correspondence between beliefs and what was actually happening. So if people were believing that income growth was faster year after year and it wasn't, then that generates a tension and either you want them to learn that their expectations are implausible or you want the income growth to actually be growing faster. But then you get sort of these complications that I think in the end the truth would be the same, that expectations are hugely important. But it would be a lot harder to see that point in the model where you have the learning going on and you have the growth changing from period to period and that sort of thing.

Jim: So I'll go over again a version of that story where there is a true productivity shot where you think and you perceive that to be permanent but it's actually only temporary, so you only gradually learn over time that this thing isn’t permanent. It would fit with kind of the tech boom perception that productivity increased that occurred during the tech boom was going to continue into the infinite future, which a lot of people were saying at the time. And then this gradually turns out not to be true so all the borrowing that I did during this period is regretted ex post.

Christopher Carroll: Yeah. Right. I'm sympathetic to that but in the end I think the part of the message that I wanted to get across is there's too much discussion in the popular press of—and among economics bloggers as well—that sort of takes the form of that deleveraging cycle we are halfway through; we're three-quarters of the way through; we're one-quarter of the way through—as though there was—well the analogy that I use is there's a lake near my house and it's got a dam. And when there's a rain storm there's like a bit more water in the lake and then it all drains down and it's like the metaphor that people have in their mind is there's some absolute unchanging target debt level for example that we're always going to return to. And in fact you sort of search the blogs and you'll find some of these people saying oh well, the target debt level is the average ratio of the past 30 years. And then someone else will say no, it's the—the point is that the thing that really moves the macro economy in models like this isn't sort of the gradual drainage of the excess water in the lake. It's the movement of the level of the dam. And that was the point that I really wanted to sort of emphasize.

So, John.

John: So I'm curious about another expectation which is not in the model, one that's come up quite a bit today which is expectations about house prices. What sort of role would they play in a model like this? When you think about describing balance sheets over the last 10 years you focus on how that debt is concentrated in the bottom 95 percent of the population. That's because 90 percent of the debt growth was in mortgages and obviously the top 5 percent can only own so much housing. So I'm not sure that's the delineating factor but how would you talk through something like that? Do we have suddenly this increase in expectations that this asset is now going to be worth a lot more? We want a lot more of it and we put a lot more of it in place and we suddenly learn it's not worth what we thought it was.

Christopher Carroll: That's a good question and as I was working through this I was anticipating that it would be a question that someone would ask. In the end I think the crucial point again goes back to people took out those loans because they had expectations and because they believed the credit was available and it was available. So if you had not had the changes in the availability of credit and you had not had changes in expectations then a higher level of house prices temporarily without any increase in the availability of credit or more importantly here, there actually isn't any change in the availability of credit. It's only people's belief about the availability of credit that changes here. And so that's the point I would come back to is that really it's beliefs about the future that are—we tend perhaps when we get lost in the weeds of the details of the balance sheet to forget that actually changing beliefs will instantly have a bigger affect than the accumulation/decumulation.

John: So I could argue that it actually might breathe a little bit of life back into the third possibility that it was this general belief about greater economic prospects going forward. Also the second, which says it's lowering that varies because you would need this belief that we're all going to be better off for me to be the person who would take on this higher debt to buy this higher asset. I would have to believe that I'm going to be able to afford those payments so I would need to believe that my income is going to continue to grow.

Christopher Carroll: But actually again I think a challenge for that point of view, and I maybe should have said this to Jim earlier, is it says that it should have been the rich people that changed their saving behavior a lot. As long as we all shared the same expectations of growth the sort of creditors, the model says, are going to be a lot more sensitive. We don't know, okay? So this is the other point I was making. We really want to know whose saving rate is it that changed. And maybe it really is actually that the rich people went on a spending spree.

John: [Inaudible 00:50:55] fewer assets. It's not that—you're only looking at the debt side.

Christopher Carroll: That's a different side.

John: You’re not looking at the positive statement side. They didn't accumulate the capital. We borrowed everything from other cities.

Christopher Carroll: Maybe that's true. That's why we need to measure the data at the microeconomic level. Maybe my perception that most of the decline in the saving rate, and maybe Barry and Steve's calculations that most of the decline in the saving rate was from the bottom 95 percent, maybe that's wrong because our data are so lousy. And so the Federal Reserve needs to produce better data for us so that we can prove that Jim is right and I'm wrong.


Female: Perhaps sort of a non, a less economist-focused question for some of the policy folks in the room. And I wondered how, wondering in this presentation, which seems like sort of focused on rational actors who are all making these really good predictions about the future, squares with what I've heard the rest of the day and then research about sort of behavioral economics and how people are making suboptimal decisions and especially one of the first papers about how economically vulnerable families are not touching risk in their balance sheet. Is that because the first paper had that data or is this a disagreement or what am I missing?

Christopher Carroll: I would say the world is more complex than you're suggesting or than I'm suggesting. What you need in order for models of this kind to have some sort of verisimilitude is you need essentially people to have some degree of worry about the future and that that degree of worry changes over time. They may not get it right. They may be less worried than they should be on average. They might be more worried than they should be on average. Some of them might be not worried enough and others of them might be too worried. But in order to have sort of macroeconomic effects, you need the level of worry to change over time. That's really all you need. And saying that the level of worry changes over time, I mean I think any behavioral economist would be perfectly willing to accept that that's possible, in fact that that's plausible. And in fact the evidence about the relationship between unemployment expectations and the saving rate suggest that it's not only possible and plausible but possibly even true.

So I don't have any brief to carry for the sort of perfect rationality. Everybody is able to solve these dynamic stochastic optimization models. Models in economics are metaphors and I'm saying people sort of, they can understand that there's an uncertainty out there. The degree of worry changes. And that's not such a hard thing even for a behavioral economist to believe.

Nothing here hinges on—as I said in response to Jim, these are people whose expectations are systematically every period deviating from what actually happens. So I'm not imposing a sort of perfect foresight kind of assumption here. I'm just saying if expectations change that has huge effects compared to the sort of trickle effect of the gradual reconfiguration of balance sheets that happens on a quarter-by-quarter, month-by-month basis. And what I'm really reacting to is a lot of the sort of I think silly commentary even from Goldman Sachs and Deutche Bank and places like that about, you know, we're 72.9 percent through the deleveraging cycle, which is just silly. We will be at the point when we're no longer deleveraging when people stop expecting doom, and the change in the expectations of doom is the thing that's going to make the difference and not have people manage to pay down their debt by one more percentage point this quarter and suddenly we've reached magical tipping point. That's the sort of misguided view that I'm trying to criticize.


Wendy Edelberg: So two thoughts. So one, it seems like you're finding that the saving rate can't be explained by the increase in wealth seems really critical to all of this because if you think that the wealth effect was big enough then you really don't need expectations to change to explain what was happening to savings. And I think as a corollary to that I can imagine a world where all of these people are borrowing against their wealth, increasing their debt to income. They have some expectation of some—they have some distribution shocks in mind, you know, of expected outcomes. And this really low-probability, terrible event happened to them. Yes, ex post they sure wish they had saved more but that's true whenever you get a low-probability, bad event that happens.

Christopher Carroll: So let me respond to the wealth shock one first because that's another one that I expected someone would ask. And what I have to say about that is that again there's—you can't understand—and here I'm going to sound sympathetic to Jim—you can't understand why there was a big run-up in the stock market and all these various different asset prices without believing there was some change in expectations. And if once you have granted that then I've sort of won my point, that it's important to measure expectations at the same time that we measure the balance sheets. We've got some surveys that look at expectations. We've got other surveys that look at balance sheets but what this tells you is that if you want to have a serious model that can help you analyze the balance sheet thing then the crucial dynamics come from expectations. So we need both expectations and balance sheets for the same people at the same time.

On the hundred-year flood point yeah, it's true that if the right way to think about the events of the last few years is as asteroid hits the earth and there was nothing that was kind of mistaken and ex-ante in the way anyone was thinking about this.

Wendy Edelberg: Wait, wait, wait. There's nothing—I want to forgive the households.

Christopher Carroll: Oh yeah. Well right, okay.

Wendy Edelberg: I'm not letting off the hook lots and lots of other factors.

Christopher: Okay, well, but the house, that's right. But again if I were to add into the model an actual change in actual credit availability—which I don't have; I just have the beliefs about credit availability change—then, you know, I'd get the same kinds of results. My purpose here was just to show how super important expectations are. Even the absence of a change in objective-reality expectations are hugely important. And if the objective reality changes and expectations change at the same time well that's even doubly important.

Yeah, John.

John: I think there are two related implications that are worth exploring here. One is that the simplest model—and it's nice to do the these sorts of things—but if we think about how credit is intermediated and it's intermediated through leveraged institutions. And so you could add a third actor there and you just need a small amount of losses, financial systems impaired, and then everything shuts down and then you can have a smaller change in expectations ramping through. And that doesn't invalidate everything you said. It's just another wrinkle. The other thing I thought—

Christopher Carroll: All right. But let me just briefly push back that on that in the sense that I suspect that—so Ben Bernanke famously testified in Congress in the spring of 2007 that the interesting events in subprime market were the consequences, I believe his exact phrase is, were likely to be contained. On the basis of what I think is perfectly sensible analysis he did some calculations or the Fed stats did some calculations of well, it's just not that big. It's 300 billion dollars and even if it all goes bad it's a 200 trillion dollar world economy and it's a 50 trillion dollar U.S. economy and 300 billion dollars is pocket change. If there hadn't been any effect on expectations, if the subprime thing hadn't metastasized into sort of a deep fear about the whole stability of the financial system, then I think it would have been, Ben Bernanke would have been right. But it was actually—

John: Yeah, I think we're actually in agreement here. I think it's more just a question—

Christopher Carroll: Yeah, yeah. So it was the fact—but the point is that the expectations channel is sort of the key transmission thing here.

John: The other aspect of this is housing is a thinly traded asset so and this was one point where some people made mistakes. I won't say who. It's a full equilibrium. Five percent of the households [unintelligible 01:01:27]. That's where the house price is determined. You see the house prices listed in the paper. We think we're worth that. But if a small share of the population all of a sudden can get credit that bids up the price of the housing but that credit is not sustainable. We all think our values are higher. We take the wealth numbers for granted in the thinly traded asset which it is, and then it comes crashing down, that's just another way that your expectation stuff could be amplified. And that thinness is important because if we think back on all the major cycles, it's usually two things that are in common. One is damage to the financial system and two—it's going to sound like [unintelligible 01:02:14]—but real estate seems to be the root of all evil here. It’s really bad. Seriously. We have a thinly traded asset subject to these misperceptions. Once again it doesn't negate what you say. It's just a different wrinkle on it that I think might be worth exploring.

Male: Why don't we stop. This sounds like good beer conversation. So we'll continue on. We have an early start in the morning. I want to thank our speaker. Thank you very much.