CFPB Research Conference May 8 (6 of 6)


I’ll just turn it over now to Darryl Getter,
who is going to be our moderator for this session.
Hello, everybody. Before I start I was asked to give an announcement. After this session
everyone is invited to a happy hour at the Hamilton on 14th Street by the White House,
so we’re going to hurry up and get started here. Actually, let me modify that slightly, which
is that there is a final sort of brief logistical meeting. Any Scientific Committee members,
ARC members, or anybody that was presenting or discussing a paper, I just want to have
a very brief meeting before we head over to the Hamilton, so come congregate down here
for some final logistics. Thanks. Today’s session, as you’re well aware,
there has been ability to repay requirements that lenders now must be aware of, so that
we can ensure that borrowers have the ability to repay the loans that they get. This has
been applied to mortgages and perhaps payday. However, I think it’s fair to say that academics
are still debating upon what are some of the key contributors to the inability to repay. We have three papers today. We have one that
talks about the people, one that talks about the products, and one that talks about the
prices, so in terms of the people, are they credit-worthy? Did we make loans to people
that weren’t as credit-worthy? Is it maybe that wasn’t as important as the fact that
they had products that were tricky and full of traps, or whatever? And then we have one
that says, well, what about negative shocks? How important was the negative shocks? So without further ado, the first paper that
we’ll hear from today is by Mr. Stefan Hunt from the UK Financial Conduct Authority, and
the paper is entitled “The Impact of Access to Payday Loans on UK Consumer Borrowing Behavior
and Credit Outcomes.” Thank you. I’m going to talk about the impact
of payday loans, and we’re going to focus particularly on consumer credit and debt outcomes,
and then we’ll talk a little bit broader than that as well. This is a joint piece of
work that I did with John Gathergood and Ben Guttman-Kenney, and Ben’s also down in the
audience here today as well. And it’s a great pleasure to be here with my sister organization
in the U.S. For those of you who don’t know, the Financial Conduct Authority is essentially
the UK’s equivalent of the Consumer Financial Protection Bureau. We find it increasingly
useful to talk with the CFPB and we hope that it’s at least somewhat useful in the other
direction. So I’m going to talk today about a piece
of work that we did in the context of having to set a price cap on payday loans. This came
direct from UK’s Parliament. We had about one week’s notice. And they put into primary
legislation a requirement to impose a price cap on high-cost, short-term credit, which
I’m going to refer to as payday loans, because it really is payday loans, for the rest of
this presentation. In order to think about how we were going to set this price cap, we
obviously had to think about, well, what’s going to be the impact on supply. I’m not
actually going to cover that work today—that’s all publicly available—but then we have
to think about is we’re going to think about the effects that we’re going to have on
suppliers if we tighten up the price gap. Obviously we need to think about the effects
on consumers and what are the impacts on consumers. There are two other questions that we focused
on, and those are the focus of the presentation today, are what options are there for consumers
when they no longer have access to payday loans? What do they do? And are these consumers
better or worse off from not having payday loans? At least we want to try to get to that
question as best as we could. As many of you will be aware, on the literature
that exists on payday lending, all the sort of rigorous literature is focused on the U.S.,
and it uses a bunch of different methodologies but most of it focuses on some geographic
variation across the different states, in terms of regulations, and there’s mixed
findings on the effects of payday loans. But we have to think about applying it to the
UK, and unfortunately in the UK there was no historical variation in terms of policy,
so we have to think about how we’re going to identify the impacts of payday loans. It turns out that we decided to do was use
what John Campbell was referring to earlier as the subnatural experiments from the policies
of firms, and this was using regression discontinuity. If we look on the left-hand side graph, at
the bottom, we can look at the credit score for the X axis. We can look at the credit
scores of firms. So it turns out in the UK, for all the major payday lenders, they have
credit scorecards, but those credit scorecards are not known to the borrowers at all. This
is not public information. And when we look at the Y axis, you want to look at what might
be the impact of a particular rule, a particular cutoff for a specific firm, and whether people
got loans anywhere in the whole of the market. This Y axis is not whether you get a loan
from this particular firm. It’s whether you get a loan anywhere in the whole market
at all. Now you’ll notice, of course, the jump doesn’t
go from 0 percent to 100 percent. It’s a fuzzy regression discontinuity design. And
the reason is because, of course, if you’re on the left-hand side, if you’re below the
credit score cutoff here, you might be able to go to other firms and get a loan from other
firms, so also there are some other mechanisms as well. On the right-hand side, albeit you may get
past the credit scoring part of the loan approval process, actually loan approval process is
normally of multi-stages and sometimes there’s stages after the credit scoring process. For
example, there might be four checks. Some people get turned down on the right-hand side.
But overall you can see that there may be a discontinuity here, as shown on the left-hand
side. So what we wanted to do was try and look for
these discontinuities across a whole range of firms, and then try to see, could we see,
when you look on the outcome variables—this is moving to the right-hand side—can we
see any actual discontinuities? Can we see any jumps in outcome measures? That’s basically
what we tried to do with this part of our work, looking at the payday loan price cap,
to look at the impacts of payday loans. With this in mind we luckily have data on
legal powers. John Campbell was talking yesterday a lot about cheese and the desire for data.
It turns out, especially when you have to do something very quickly from Parliament,
you can pretty much ask for the data sets you want, within some bounds. So we used our
legal powers and we went to the top 37 lenders in the industry and said, “Can you give
us the details of all the loans you made in 2012 and 2013, and we want the names and addresses
and loan sizes for every single loan that you made.” We don’t have unique identifiers.
We don’t have Social Security numbers in the UK that we could use. So we get that. That gives us pretty much
the whole market, and then we need to get some discontinuities, so we chose 11 lenders.
Broadly, this was based on size. For those we got all of the applications, whether they’d
been denied or accepted, and the loan level. We also got details of the credit approval
process and the credit scores that happened at the application level, with denied applications
as well. This isn’t relevant for today’s work but we also got fully disaggregated revenues
and costs at the loan level, and that was actually very important for trying to model
the supply side, understanding the effect of a price cap. Because we have the names and addresses for
the individuals and we can get unique identifiers, we can actually match the same individuals
across all the 37 different firms, so we can track them across all the individuals, and
we can then go to, again using legal powers, we then went to one of the major credit reference
agencies and asked for all of the credit bureau files for this whole population as well. So
we matched that in as well. And then, lastly, we commissioned a consumer survey of 2,000
people. Obviously there’s a lot you can’t see in credit bureau files, and we tried to
match that in as well. I’ll give a quick bit of background on the
payday market in the UK. It’s quite concentrated. The top three firms, 72 percent. There’s
around 400 firms. Most of them, of course, are very small. It’s mainly online. Over
80 percent online in the 2012-2013 period, so that’s quite different than the U.S.
The prices, I think, are relatively similar. My understanding, in the U.S. was that you
linearized. When you went to try to go for short-terms loans and you looked at the effects
annually, you just took a linear. If you exponentiate, which is certainly the European rules for
APRs, you get to roughly 5,000 percent APR, which is the typical cost. There were about
1.3 million customers in 2013 and they six loans per customer. The average loan size,
which actually is relatively commensurate to the U.S. is at £260 for 17 days. This is an infographic from our Commence Department.
On the top left you can see that people tend to be a bit younger than the average in the
UK, and they’re a bit less wealthy but actually maybe this was less extreme than I would have
expected. This here, we took all the different loan
approval processes that we could get working, so one firm may have had various loan approval
processes, they changed their processes over time. Sometimes if people come in with different
sets of information they may be different processes. So we took the 17 loan approval
processes that we could get working and we tried to see whether there were discontinuities.
Effectively, we got 9 of the loan approval processes working. That was from four different
firms, covering around 80 percent of the market. And as you can see, these are some examples
of those four different processes, one taken from each of the major firms, and you can
see that there are sharp discontinuities, so actually the size of the discontinuity
varies a little bit by each of the firms. Broadly, the density tests work quite well.
There’s one slight wrinkle on that, which I’m going to skip over today. The density
tests worked pretty well. And this is the first set of outcome variables, when we look
at the number of personal loan applications. If you look at the bottom, you can look at
the four different processes I’ve taken from the nine that worked. You can see before
applying for payday loans we see very smooth, no discontinuity, so essentially that’s
the falsification test. If you look on the right, above, and you look at the top, you
can see that there’s actually discontinuities across all the four different ones, and actually
the scale is slightly different on the Y axes. It turns out in this particular case, the
actual estimated effect is quite similar for all four of these different processes. So in order not to have to show four different
charts for every single outcomes, and what we did in the paper, what we did was we took
the beta coefficients that we just estimated. I just showed you four of them but you can
imagine them for nine different processes. So we actually pooled the beta coefficients
and we showed pooled estimates for most of the coefficients. And as you can see, when
you move down to the chart here, where we have a whole load of outcome variables on
the left-hand side, and we look 0 to 6 months and then 6 to 12 months after applying for
the payday loan, and we’ve got the coefficients and then we’ve also got the means up here,
so you can get some sense of the scale of the effects. And what we find—and I’m
going to go through five different classes of outcome variable, quite quickly, I think—we
can see that actually getting payday loans causes people to apply for more other types
of credit. That isn’t just true immediately after the payday loan but that’s actually
also true in the 6-to-12-month period afterwards, and that’s particularly true for personal
loans, you’ll note as well. So, actually, use of payday loans leads to more applications
for other kinds of credit. Let’s take a look at the actual credit portfolio.
This is actually the holdings of the products. The evidence is mixed on this slide, less
so with others. We can see, actually, the holdings of personal loans increases. It turns
out that the mortgage result is spurious. You can basically assert that the falsification
test doesn’t work for mortgages, although it does for pretty much everything else. So
we an overall increase in the holdings of products. There’s a bit of nuance here which
I’ll maybe come back to later. If you look at credit balance we tend to see
that credit balances for non-payday products increase after people get payday loans, and
this is particular, too, with personal loans. There’s a small decrease 6 to 12 months
afterwards for credit cards. And then if we move to the next set of outcome variables,
we can actually try and see whether we’re looking at credit events. Now, credit events
is mostly missing payments for products, but it could also be the use of what we call unarranged
overdrafts; in the U.S. what you’d just call overdrafts. We have arranged overdrafts;
you don’t. And we can actually see, when you look at credit events, we start to see
missing payments in other products. So payday loans cause consumers to increase
their unarranged overdraft use, what you would call just overdraft use, and borrowers begin
to miss payments for the non-payday products 6 to 12 months after applying for the payday
loan. We actually looked at their worst account status. This is measured in months. Their
worst account, the account you’re most behind on, in terms of the number of month, you’re
half a month behind more in payments from having taken a payday loan out versus not
having a payday loan out. So we see this increase in missing payments. Then if we move on to the next set of outcomes,
this is delinquency and default. We actually see an increase in non-payday defaults, and
especially if you do non-payday defaults as a percentage of total non-payday balances,
that goes up. And then if we look at the bottom right, this is change in credit score over
a 1-year period around getting the payday loans. We see, in general, this whole set
of people were in a period where their credit scores were getting worse, but those people
who took out payday loans had an extra 20-point drop in their credit scores over this period.
So there’s a series of what appear to be some negative effects, especially when we’re
looking at sub-default delinquency and credit scores. Taking some of those key variables, we can
look at time before and after people take out payday loans. I’m actually going to
start with the bottom right and we’re going to look at the holdings of personal loan products.
As we can see, as people take out payday loans they begin to hold more personal loan products
over time. One might imagine what kind of channels there are, but certainly my immediate
thought is this is people are taking out payday loans, they’re pretty expensive, they’ve
got to pay them back, they’re looking for someone else to go to get loans to pay them
back. If we then move now to the top left, which
is overdraft usage, we can in the actual month that people take out payday loans there’s
a small decrease in overdraft usage. You can see this small decrease. It’s about a 1-percent
drop in overdraft usage, and that’s off a base of about 11 percent, so a very small
decrease in overdraft usage. But by 3 months after taking out payday loans there’s about
a 4-percent increase in overdraft usage, and that’s persistent. It stays up right up
to 12 months after taking out payday loans. Looking at a couple of other variables, this
is the non-payday default bounced as a percentage of non-payday bounces. The default here is
defined as 6 or more months behind in payments, so it’s not surprising there’s a lag on
this one. But as you can see towards the end of the year there’s a very marked increase
in the percentage of bounces who are in default. This worsening credit events means getting
more months behind in payments, and again we can see that this happens after taking
out payday loans. Now, that’s just the credit bureau evidence,
which we then matched into that. We’ve done a consumer survey of 2,000 people. A small
part of the sample had to be used for particular problems that we were looking at, for various
reasons, so we took the remainder of the sample, about 1,600 people, and we put it evenly across
three different buckets—those people who were just below the credit score, those people
who were just above the credit score, and then the third group tried to be relatively
representative of everyone else. We focused most of our analysis on comparing Groups 1
and 2, and I’m just going to give some flavor of the results here today. We want to understand, of course, who people
were, so this was information we couldn’t get from the firms or from the credit bureau,
so we tried to get a whole bunch of extra demographics and financial capability measures
from individuals. We then wanted to understand what substitutes people would use. There’s
a whole load of information measures that people might do. Of course, they might turn
to legal lending. They might borrow from friends and family, so we wanted to measure that.
We also want to think about some form of substitute. Not all forms of substitutes will be captured
by the credit reference agency data. In particular, certain types of overdraft usage, arranged
overdraft usage, isn’t covered in the credit bureau files, so we asked questions on that.
And then we also tried to understand if people are trying to use their savings, reducing
consumptions. And then lastly, of course, we wanted to understand whether people are
better or worse off, so we had a bank of metrics, subjective well-being metrics, a whole bunch
of questions on sort of financial distress. Then, of course, we asked about things like
regret and then some things about reasons for borrowing and how this money is used. It was a telephone survey, and we had a relatively
low response rate, as you can see. It was about a 5 percent response rate. Much to my
surprise, actually, when you look at the population and you compare this population, just a 5
percent sample, actually they looked really, really similar on the observable characteristics,
which I was surprised by, and actually Groups 1 and 2 are pretty similar. Actually, this is not comparing Groups 1 and
2 but comparing the marginally successful to the marginally unsuccessful individuals,
those people that did get loans to didn’t, so it’s not quite a discontinuity comparison.
But what we can see here is we’re comparing what people said they actually did when they
didn’t get payday loans to what people said they would have done if they wouldn’t have
gotten the payday loan. To my surprise, actually, the hypothetical and the real outcomes seem
to be very aligned. So people seemed to be able to predict what they’d do quite well. I’m not going to go more into some of the
fancier stuff we did and just say what immediately stands out is most people just go without.
The second biggest thing that people do is they borrow from friends and family. As I
mentioned, we had questions on whether they would use legal lending and we actually designed
questions with the illegal money lending team in the UK, and we found absolutely no effect
of lack of payday loan usage on usage of illegal lending. I’m going to skip quickly to the conclusions.
What we found in terms of payday loans, payday loans do provide some short-term liquidity.
On loan denial there is almost no short-run substitution to other formal borrowing, though
there was a small increase in unarranged overdraft usage. People borrow more from friends and
family—that’s one thing that happens—but there’s no evidence for substitution to
illegal lending. Most just adjust to loan denial by foregoing or delaying consumptions.
That’s the main thing. Payday loans cause a subsequent increase in
non-payday credit products, in particular what we found was most marked was this increase
in overdraft usage that I showed earlier. Borrowers get further behind in payments on
non-payday products, and the non-payday default debt balances increase in the credit bureau
scores decreases. That said, we didn’t find any evidence of an effect on subjective well-being
or financial distress. And I guess the last is, this is the cap we
came up with. This is obviously building in the supply side model. This about a 40 to
45 percent reduction in the amount that people are charging. So they could charge up to 0.8
percent per day for loans, including rollovers and all aspects of various types of fees.
The most that could ever be paid back was twice the amount that was borrowed, which
I appreciate is still quite high. But it turns out the main thing that we had to consider
in comparison to this was mass exit of all the firms in the market, so this is what we
came up with. Thank you very much. Our next paper is by
Felipe Severino from Dartmouth College, and the paper is entitled “Changes in Buyer
Composition and the Expansion of Credit During the Boom.” Thanks, everyone, and the organizers for inviting
us to present this paper here. The name is “Changes in Buyer Composition and the Expansion
of Credit During the Boom.” It’s co-authored with Manuel Adelino from Duke and Antoinette
Schoar from MIT. Before I start the paper or the presentation,
I want to make sure that or warn you in a way that what I’m going to do today is try
to re-look at some stylized facts with respect to the mortgage crisis, with the aim to challenge
what we know about the mortgage crisis. To do that, we’re going to have to basically
depart from this approach of causal inference and just look at these patterns and see what
we can learn from it and how do we make sense of it. So with that in mind, the main motivation
of this paper is this common view that during the mortgage crisis innovations in the supply
of credit, led to a distortion in the allocation of credit throughout the income distributions,
in particular, allocating more credit to poor households. This distortion in the allocation
of credit led to an increase in house prices, and then when the credit was cut, basically
house prices dropped. The main evidence of this credit supply-income
distortion relies on a negative correlation between income growth and mortgage growth
that was stated in the Mian and Sufi paper in 2009. And in order to make sense of this
type of expense that we’re going to show you today, we’re going to have to try to
challenge those findings. What do we find? The first thing that we find
and the first thing that I want you to remember is that credit expanded throughout the income
distribution, and not just for the poor. So middle-income and high-income households had
a much larger fraction of the total amount of mortgage origination and the amount of
mortgage that was allocated to poor households was small and it stayed small throughout the
whole boom. And that may be super clear here. We’re not saying that there wasn’t an
expansion of credit. We’re just saying that the way that credit was allocated throughout
the income distribution wasn’t what we would call a fundamental story. The second and very important point is that
we find that the majority of credit in default, when we focus on the dollar values of default
during the crisis, is coming from middle-income households, and ZIP codes, and high FICO score
borrowers, even when you look at within subprime areas. And this fact is consistent with a
view that the recent participation in the mortgage market implies a larger fraction
of households close to their maximum debt capacity when house prices dropped. And I’m
going to argue that this has very specific implications in terms of what we think is
the right regulation to prevent things like the mortgage crisis that we experienced. Finally—and this is in order to make sense
of all these findings with respect to previous literature—we’re going to take a focus
on households and not ZIP codes, and the main idea here is that households are the ones
that take out mortgages, not ZIP codes. And why is that important? Because previous results
focused on the total amount of mortgage in a ZIP code, and that measure is a composition
of two things—the average mortgage size and the number of mortgages that people in
that ZIP code are taking. It turns out that when you look at those two things separately,
independent on how you measure income, you find that the individual mortgage size is
always positively correlated with income, and this negative correlation is only coming
from the number of mortgages, so from the extensive mortgage. And even more important,
actually, this negative correlation is coming from the high end of the income distribution,
so it’s basically telling you that high-income areas that
experience a higher increase in income growth,
that the increase in mortgages are the ones that are driving that negative correlation.
So, overall, this tells you that there’s not evidence that during the boom there was
a distortion on the allocation of credit. So that’s a preview of the results. Now
I’m going to walk you through the patterns that I told you about, and in order to do
this we basically use mostly publicly available inform data to look at the patterns of mortgage
origination with respect to the income distribution. We’re going to use HMDA data, which we have
already seen many papers using it already, so I don’t think I have to you too much
about it. You basically have the universe of mortgages that were originated in each
year, and you have the balance, so the total amount of mortgages that were taken, and also
you have the self-reported income that people use when they apply for that loan. We’re going to supplement that with IRS
income data at the ZIP code level, house prices at the ZIP code level also from Zillow, demographics
from census, and then because we want to look at ex-post outcomes, such as delinquency.
We’re going to use LPS data, a 5 percent sample of the total population. Today I’m
going to focus on LPS data but we have been basically finding the same patterns using
Black Book data or the agency nonperformance data that is publicly available. So the facts
that I’m going to show you with respect to delinquency are robust through many different
data sources. Before we go into the results, I’m not going
to show you related literature because of the time and also because there’s vast literature
trying to approach and trying to study what was the consequences at the way that credit
was expanded during the crisis, and this study is trying to challenge a lot of those views. The first patterns—what do we see here?
What we see here is for each year we’re going to take the total amount of mortgages
that were originated in that year and then, using the self-reported income, assign that
specific mortgage to an income packet, and we’re going to divide the whole amount of
mortgage origination in five buckets. Then we’re going to basically estimate the fraction
of mortgages that were originated on that bucket with respect to the total amount of
mortgages that you received and that it was originated in that year. And we’re going
to do the same for each year. So as you can see, each column sums up to 100, so this graph
doesn’t reflect the fact that the total amount of mortgages were growing through the
boom. It’s just telling you about the distribution. And what you see here is that basically middle-
and high-income individuals, in this case, are the ones that account for the biggest
fraction of the total mortgage origination, and the location of mortgages on the low income
was small, so 10 percent, and it stayed small during the whole boom. So there could be concerns
about the measure of income that we’re using, so we’re going to repeat the same analysis
but now classifying mortgages based on a ranking, using the IRS income per capita. It turns
out that when you do that, you basically find similar patterns. Low-income guys have a slight
increase in the share, so it went from 10 to 12, but that’s very far from arguing
that there was a change in the way that credit was allocated. The obvious follow-up question is, okay, this
is focusing on first mortgage for purchase, so mortgage origination for purchase, so what
happened with other types of products? What we’re going to do is we’re going to look
at, through LPS, how the allocation of credit was distributed on second liens and cash-out
refinance, which arguably are the ones that could have led to a distortion on the allocation
of credit. Here we basically have the same idea of pictures but now also showing the
magnitude of the amount of debt that each one of the products have. The first one is mortgage for purchase, the
second one is second liens, and the third one is cash-out refis, and I showed for a
specific year, 2006, using, again, LPS data, but the bottom line is that you see that the
way that the allocation of credit was distributed, it was according to income, so there wasn’t
really a distortion there. So, basically, these three pictures are telling you that
when you look at the data this way, you find that there’s no evidence of a disallocation
of credit. And let me be super clear here. We’re not trying to argue that liar loans,
or non-doc loans didn’t exist. That’s not the bottom line of this graph. What we’re
trying to say is that those things, even if they exist, it doesn’t seem to matter in
the way that the credit was allocated with respect to income. Now let’s move to the second point—what
happened ex-post? Here we’re also going to take a different approach to the data.
Instead of focusing on the rates of default we’re going to focus on the amount of dollars
that went into default for mortgages that were originated in 2003, ‘4, ‘5, and ‘6,
and here we’re going to be looking at whether those mortgages were delinquent 3 years onwards,
in the future. And we’re going to basically look at the mortgages that are in delinquency
and multiply it by the amount of that mortgage and calculate the fraction. When you do that, you observe that the inter-range,
so the middle quintiles, are the ones that account for the highest fraction of delinquent
mortgage, and that stays stable again through mortgages that were originated through the
boom. So again, this is consistent with previous literature that I’ve shown, that rates of
default went up but acknowledging that basically a lot of mortgages defaulting on the lower
end have a smaller impact because the amount of those mortgages are small. And if you want
to think about where the consequences of this high credit expansion with respect to the
global economy, we want to argue that this is the right measure because this is the one
that is going to impact financial intermediaries and, at the end, the local economies. So moving forward we’re also going to try
to investigate what happened when you look at FICO scores at the individual level, and
here we also are finding somehow contradicted patterns. Here we’re going to basically
look, what is the fraction of mortgages that are in default, in terms of dollar value,
and then classify them in three buckets. So prime borrowers in white, the light gray is
prime borrowers with a FICO between 660 to 720, and then superprime borrowers with FICO
scores about 720. When you do that, you see that for mortgages that were originated in
2003, superprimes are a big fraction, but as you move through mortgages that were originated
by the end of the boom, the big fraction is coming from what we call prime borrowers,
so high FICO scores. If you put together with the fact that the amt of mortgage that was
originated by the end of the boom was a lot bigger than at the beginning, the impact high
FICO borrowers in terms of default is a lot bigger than the impact of subprime borrowers. So how can we make this consistent with some
very, very convincing patterns that show that the fraction of subprime borrowers in a ZIP
code is very positively correlated with mortgage origination, house prices, and ex-post default?
The answer to that is look at individual data. So what are we going to do? We’re basically
going to classify ZIP codes based on the fraction of subprime borrowers that a ZIP code has,
which is highly correlated with the fraction of subprime borrowers that the ZIP code, and
then look at within those ZIP codes, prime and subprime ZIP codes, who are the individuals
that are defaulting, and what is the contribution of those individuals to default? When you do that, you see that specifically,
in 2006, it’s true that default went up a lot more in subprime ZIP codes, but the
fraction or the contribution of those defaults is coming from high FICO score guys. So again,
by looking at the individual level data, we’re finding that the individuals that are contributing
the most to the ex-post defaults are the high FICO score guys. So that solves the puzzle with respect to
subprime and FICO scores. What happened with credit and income? For this it’s very important
to kind of go back to the specifications that show that negative correlation. What that
specification does is look at the growth of mortgage at the ZIP code level, as a left-hand-side
variable, on the growth on income per capita, as a right-hand-side variable, and counting
fixed effects to control for several account characteristics. And what we’re going to
argue here, in this paper, is that this specification masks two things. One, as I already told you,
on the left-hand side variable, you are basically abstracting from the fact that individuals
are the guys that take on mortgages and therefore you need to take an individual approach to
this analysis, and by doing that you need to decompose the total amt of mortgage in
a ZIP code and the average size and the number of loans, and that’s what I’m going to
focus on in the next slide. But then there’s another important issue
too, is the fact that income per capita in a ZIP code is the income of the residents
but not the income of the guys that are buying houses, and that is historically very different.
But again, we don’t need that and it’s not going to be the focus here. To come back
to the first point, the first column shows the negative correlation that I was telling
you about before. When you look at IRS income growth and total amount of ZIP code growth
you find a negative coefficient. When you break out the left-hand side variable you
see that it’s always positive when you look at the average mortgage size. And again, the
negative correlation is coming from there. Just to conclude, basically, this stylized
facts and this revisiting of the data provides a novel explanation of the observed credit
expansion due to a system-wide increase on leverage, throughout the income distribution
and not just for the poor, and this is consistent with homebuyers and lenders buying into a
housing bubble, and then experiences the consequences as house prices drop. The fact that we see
defaults coming from middle-income areas and high FICO scores is consistent with this view
that the recent market participation led to a higher fraction of households close to the
maximum debt capacity, and then when house prices dropped they basically didn’t have
any financial slack to react, and this fact highlights the need to think about macroprudential
regulation that has to account for the fact that the total amt of leverage in the economy
is going up and that it will not be solved using microfinancial regulation or financial
regulation for institutions. Thanks a lot. I appreciate it.
Thank you very much. Our next speaker is Christopher Palmer from the University of California,
Berkeley, and his paper is titled “Why Did So Many Subprime Borrowers Default During
the Crisis: Loose Credit or Plummeting Prices.” Thank you. Thank you. It’s been a very successful conference
and yet I have to offer one suggestion from my 5-year-old I was dropping off at kindergarten
on my way to the airport to come here. He shouted back at the car as he ran off to the
kindergarten playground, “Good luck at the carnival.” So I’m imagining maybe next
year we could have it say CFPB Research Carnival. It was plenty successful on the enrollment
and registration but think what it could be if it was the carnival. This is another paper in a series of papers
that we’ve seen that are doing forensics on the housing crisis, and I’m focusing
particularly on subprime borrowers. As a motivation, it was really ground zero for a lot of the
housing crisis. Subprime foreclosures accounted for more than half of the foreclosures that
happened during the recession, and the big question is why was there this surge in the
subprime foreclosure rate? Why were so many subprime borrowers entering into foreclosure? There are lots of explanations for this. I
want to focus on two very divergent stories. I’ll show you evidence of these stories
being advocated as somewhat of an either/or explanation for the crisis. One was that there
was a change in the composition of borrowers, and if I think about why is there a surge
in the subprime foreclosure rate it’s because something changed about subprime lending.
Something changed about who the borrowers were and the products they were taking out. The second story is that something changed
with the economic conditions and this affected subprime borrowers, so whether this is negative
equity, whether this is unemployment, et cetera. So both of these are reasonable, and I’ll
show you that there’s evidence that both of them are true. Disentangling them is important
for policy and I’m going to provide a method of being able to quantify their respective
roles. So it helps me to pay attention in a talk
if I have some sense of what’s coming and in what order, and I know where to expect
what I’m most interested in, so I’ll flesh these two stories out a little bit more, and
then I’ll show you the data and what it says about how subprime borrowers and subprime
mortgage products are changing. I’ll briefly describe why a hazard model is the right way
to do this here and what the hazard model specification looks like, and walk through
some coefficients, put some numbers on this. Especially I want to spend some time talking
about how to disentangle the role of prices and lending standards. If you’ve ever gotten
into a discussion about this, as I had with my taxi driver a couple of days ago here in
D.C., it’s hard to argue that prices have an important role to play as distinct from
lending standards, given that we think that prices themselves—and as the following paper,
after mine, will show—prices themselves are, in large part, affected by lending standards
and by credit conditions. So I’ll mention how I go about disentangling the two. And
then I’ll conclude with kind of takeaway points and policy implications. Here is Krugman arguing for my first story,
talking about a graph that I’ll show you in a moment, writing in the New York Times,
saying that “many borrowers are ill-equipped to make judgments about these ‘exotic’
loans,” and he singles out ‘teaser’ rates, prepayment penalties, et cetera, and
maybe this is going to help motivate our financial regulation and help our financial regulation
of sufficient teeth in the future. Here’s Stan Liebowitz offering the counterpoint
in the Wall Street Journal, so it works out well to have New York Times and Wall Street
Journal opposite, saying many policy makers, ordinary people, New York Times columnists
blame the rise of foreclosures squarely on the subprime mortgage lenders who misled borrowers.
What was really behind this? It was negative equity, and the divergence in policy implications
is enormous. So what is this divergence in policy implications?
Let me just mention that briefly. Take story number 1—lending standards fell and what
changed was a decline in borrower creditworthiness and them taking out risky products. Well,
in that case we want to restrict the contract space. We want to be able to say, well, the
following risky products should be outlawed, and we’ve done much of that. It’s very
hard to find people offering subprime-type mortgages with various risky features these
days. So that’s, I would call, a microprudential approach. Let’s change the characteristics
of mortgage products. The second, if it’s a decline in property
values and that’s impeding distressed sales, well, that motivates more macroprudential
policies, more, let’s keep the credit market cooler throughout the entire boom. That might
help prices from falling in the future and it also might help there be an explosion of
people who are at their max debt capacity, as Felipe just talked about. So there’s this important distinction here,
not only for this ex ante, micro versus macro pru, but also for ex-post remediation, how
you think about what needs to happen to modify, for example, distressed mortgages. For stress-testing,
as well, and risk management, whether that’s the government or a private organization,
trying to understand the riskiness of a given portfolio, the results of understanding how
much loads on one and how much loads on two of the riskiness of a given mortgage portfolio
informs all those decisions. So I’m going to take an approach and try
to bring some order to it that lots of people have taken—congressional reports, journalists,
academic papers—of looking and comparing vintages or cohorts of subprime borrowers.
Felipe was just doing this, where he was showing each bar was a vintage or a cohort of borrowers,
and I’m going to do the same thing. The question is, is this indicative of the cause
in defaults? Here’s this graph, my replication of it,
that has motivated a lot of this literature. This is showing the cumulative default probability
of the 2003 cohort. So on the X axis, the number of months since origination, on the
Y axis the total fraction of the 2003 cohort that has defaulted within a given period of
time. So if I go out to, say, 24 months, around 5 percent of that 2003 cohort has defaulted.
This lines shows me the trajectory of the performance of borrowers who took out subprime
mortgages in 2003. If I put 2004 on this, it looks largely similar, but let me point
out that if you go about 5 years out, and you go up, a seemingly higher fraction of
the 2004 cohort has defaulted than will seemingly ever default from the 2003 cohort. So there
seems to be something different about the performance of the 2004 cohort. If I do the
strip tease, go to 2005, 2006, and 2007, largely monotonic decrease in the performance of subprime
borrowers over time. Looking at this graph you can see scope for
both explanations. You could say, well, what happened here is a change in who these borrowers
were. The 2006, 2007 borrowers, they’re just riskier. Banks are becoming more brazen.
People are becoming more and more extrapolative in their house price beliefs, and what’s
happening here is a change in who these borrowers are and what their mortgages look like. On
the other hand you could say, no, it doesn’t have anything to do with who these borrowers
are. It’s a change in what happened to them. So later cohorts are buying at the peak, they’re
underwater almost instantly. So you can see scope for both of these stories and it just
becomes an exercise, really, in what your prior is, and how you read these tea leaves. As I mentioned, it could come from both of
these. It’s not an either/or. I’m going to tell you how much of it is both, and again,
the empirical challenge here is how do you identify the importance of lending standards
in a world where there is a fall in property values, which was potentially caused by that
very change in lending standards? One more graph, just showing you aggregate,
what’s happening differentially to these cohorts. This is the combined loan-to-value
ratio of each of these cohorts, mark to market, so using an updated property valuation of
their property for the denominator in that combined loan-to-value ratio. The first thing
I want to point your attention to is the numbers. Each point at the beginning of a line is the
combined loan-to-value ratio of a cohort at origination, so at kind of the first month
of that year, those first borrowers that are taking that out. So using the assessed valuation
model denominator, you can see it seems like down payments are getting smaller, leverage
is increasing over time. Perhaps this is evidence of these underwriting standards declining. The other pattern that you see is that these
early cohort borrowers, what’s happening to their leverage? It’s decreasing very
rapidly. Why is that happening? It’s not so much that they paying down their mortgages
during the first few years. It’s that they’re experiencing double-digit price appreciation
in much of the country in 2003, 2004. So they have a large equity cushion that insulates
them from the price declines, whereas these late cohort borrowers—2006, 2007—they’re
underwater almost immediately after they take out their mortgage. They’re buying at the
peak. Prices are falling quite quickly. Turning to the data, this is the CoreLogic
Loan Performance data from mortgage-backed securities, and it contains rich origination
characteristics and then also followed the performance of these borrowers over time.
Let me show you what these look like. Here are subprime borrower characteristics. Their
FICO scores are low. Their debt-to-income ratios are high. All of these factors are
indicative of some level of riskiness of this segment of the market, and yet they’re not
changing very much over time. What is changing a lot over time are these
product characteristics, particularly the ones I’m highlighting at the end, so the
share of these mortgages that an interest-only time period or have a balloon component, where
there’s a balloon payment that’s due, or have a second lien attached to them. That’s
changing quite dramatically over time. So that’s what the data says about borrowers
and products changing. Now, just because those individual characteristics such as the FICO
scores are changing doesn’t mean that the people aren’t changing. Assuredly, the people
that take out only-only mortgages are different than people who don’t take out interest-only
mortgages, so borrowers are changing, as measured especially by the types of products that they’re
taking out. I’m going to build a hazard model to analyze
this data. Why is a hazard model the right way to think about this? Well, first of all,
I’m thinking about default. That’s a failure. But think about the structure of my data.
I can see an individual in my data if and only if they have not yet defaulted, and as
soon as they default, I see that they defaulted, and then they’re gone. So I have a sample
selection issue, and if you remember from your math stats courses what a hazard is,
it’s the probability of failure given that I have not yet defaulted. So that works out
very well. It’s exactly the structure of my data. I have a proportional hazards model,
so this lambda naught function is going to be the baseline hazard, which is called up
and down by this linear index, X beta. So how do I fill X beta? It’s Xicgt. I is
an individual loan, so I have those borrower characteristics and those loan characteristics.
Those are static, measured at time of origination. And then C is the cohort that you’re in,
so gamma C, this is a cohort main effect, and this is going to tell me how different,
depending on what I’m controlling for, are these different cohorts, a measure of the
change in prices that they’re experiencing, and then geographic fixed effects. So thinking about what these gamma C’s are,
here’s just a plot where I’m not controlling for anything else, just the baseline hazard
and gamma C. So this is just a one-number measure of how different was the performance
of each of these cohorts relative to 2003. Let me back up for a moment and remind you,
very high level, what we’re trying to do. Where did the rise in subprime defaults come
from? It seems like a lot of them came from these later cohorts. This exercise that I’m
going to walk you through right here is going to decompose where the deterioration in cohort
performance came from. So you can see each cohort is performing much worse than the 2003
cohort. That’s the omitted category. So each of these are positive and they’re large,
all this is doing is running those first few graphs through this hazard model. As I introduce controls it’s going to say
how different would these cohorts have performed if, for example, this next bar is when I add
in geographic fixed effects. So maybe one of the reasons later cohorts perform poorly
is that they are coming disproportionately from risky areas. We can control for geographic
fixed effects and verify that there’s nothing to do with geography that’s causing all
of this. So we can rule that out. We can put in borrower characteristics, and
now what we’re saying is, how different would cohorts have performed if they had all
been in the same place and all have the same borrower observables? And the answer is they
would’ve still been very different. We put in loan characteristics and we start to see
some action. So now what this is saying is, if instead of being as different as they were,
all borrowers took out identical mortgage products in each cohort, there would have
been 30 percent less heterogeneity across cohorts. So 30 percent of the difference across
cohorts, the rise in defaults, would be explained by loan characteristics. So each of these
bars tells me if I match on certain things then how different would cohorts have been?
I can put in borrower and loan characteristics to start to get a feel for this. I’m going to do the same exercise with prices.
I have a Case-Schiller-style repeat price index from CoreLogic, so here’s the punch
line there. Prices are even more important than loan characteristics. They are explaining
over 50 percent of the variation across cohorts here. And so what I’m saying here—and
let me skip to this full model, where you can see that the cohort differences are largely
insignificant, is that if all cohorts had faced the same prices and the same loan and
borrower characteristics, they would’ve been largely the same, which means this model
is explaining where the rise in subprime defaults comes from, and it’s doing this decomposition
exercise for me. So before I put number on that, let me briefly
mention that, again, there’s this problem that if I want to say, well, this fraction
of the variation comes from prices, this fraction of the variation comes from lending standards,
if prices were caused by lending standards it’s hard to disentangle these. I’m not
going to walk through the entire IV strategy in the control function approach—that’s
in the paper—but what I do is I map historical price cycles in each geography onto current
cycles. So, arguably, the things that happened that caused a city’s boom and bust in the
1980s were unrelated to future changes in subprime lending in these areas. And the fascinating thing which was highlighted
a little bit by papers this week has shown that there’s some persistence here. This
is an example just showing that there are lots of cities that have both booms and busts
in both periods, but to just zoom in and pick two cities, here’s Philadelphia and Pittsburgh.
Philadelphia, blue line, repeat sales index, has a boom and a bust, both around late ‘80s,
early ‘90s, and in the 2000s. Pittsburgh just kind of motors on both periods. So this
is a complier for my instrument, if you will. These are places that if you scaled up what
house price cycle they had in the late ‘80s, early ‘90s, you’d have a decent guess
of the house price cycle that they have in the 2000s, in the sense that there’s a strong
first stage. So control function estimation, I can put
in the residuals, basically, is as houses works or flexible functions of them and verify
that, in fact, this graph looks very similar to the one you saw before. The model does
a great job of explaining why there’s variation across cohorts, why there was a deterioration
in cohort performance, and thus where subprime defaults came from. And the numbers to put
on this are that loan characteristics and the advent of riskier products and the change
in borrowers and mortgage products is causing 30 percent of the rise in defaults. Price
changes, price declines, especially later cohorts buying at the peak and facing strong
price declines right away, is explaining 60 percent of that variation. The model combined
is explaining 95 percent of that story. In Chris’ opening remarks he was talking
about the desire to get some kind of deep parameters that are more useful for policy,
so using this model I can then go back and tinker with the world and say, well, what
if policy had been different? What had been characteristics had been different? Or, what
if observed macroeconomic shocks had been different? How different would the performance
of various cohorts have been? Let me orient you. I’m going to talk about
the 2003 and the 2006 cohorts. Remember, 2003, very rosy cohort, actually defaulted, on average,
4-and-change percent a year, whereas the 2006 cohort defaulted at 12 percent a year. So
what kind of policy things would have to change, or what would have to change about the environment
that these borrowers faced for that gap to be closed? Here’s a table where I walk through some
of those scenarios, and, in particular, let me highlight what happens so clearly if they
faced each other’s prices the gap changes a lot. If I go here to this fourth column,
what if they had both faced flat prices? Well, the 2003 cohort would have surely defaulted
more than they did. The 2006 cohort, they would have happily taken flat prices relative
to what they experienced, and there’s still a gap there, and you can see from the final
column that that gap is closed completely if they had had similar characteristics. So
that’s just telling you, basically, the same thing as the results that I showed you
before. But what I want to highlight here is that even if you thought, from those first
graphs, what we really need to do is return to the good-old-days lending environment of
2003. It’s not so much that those 2003 borrowers, with their lending environment were bulletproof.
They also would have been severely affected by a downturn, which was orthogonal to their
credit standards. So if they had just faced a macroeconomic shock they also would have
had a tremendous increase in default. So this kind of gives us a bound on what we could
hope to achieve in that sense. Thinking about the policy implications here,
I mentioned at the beginning I wanted to draw this contrast between microprudential and
macroprudential and what the right balance here is. Again, if our entire focus is on
microprudential, which, incidentally, is quite justified by my results showing that much
of the increase in subprime defaults came from these risky product types, then certainly
that is going to make a difference, as the simulations show. However, to Felipe’s point,
it could push the burden of financial regulation in the incidence of closing off this credit
market on a particular group of borrowers who were not responsible for the lion’s
share of the defaults. On the other hand, macroprudential policy, provided that macroprudential
policy has enough efficacy to be able to rein in credit in and have a smoother price cycle,
which I think is open for future research, seems to have some scope here as well, given
that 60 percent of this was caused by the macroeconomic fluctuation as embodied by local
price changes. So are price declines relevant only for risky borrowers? No. Everyone is
impacted. You need both of these strategies to be working in concert. So wrapping up, my goal here was to speak
to the debate about the causes of the overall subprime crisis. I offered a new strategy
for disentangling these effects. The numbers you can take home and put under your pillow
are that 30 percent of the lending standards, 30 percent of this decline in cohort performance,
30 percent of the rash increase in subprime defaults came from lending standards, 60 percent
of it came from prices, very little of it came from things that are not captured by
those two factors. Mortgage regulation, just doing this kind of, well, we’re going to
tweak the mortgage contract space, is not going to be enough. It could go too far. There’s
need for this kind of shared sacrifice through macroprudential policy, and there are broad
implications here that hopefully several of you found interesting. Thank you. Our last paper is by Marco Di Maggio from
Columbia University, entitled “Credit-Induced Boom and Bust.”
Okay. So thanks a lot to the organizers for including our paper. This is a joint work
with Amir Kermani, who is here. So let me start with some general motivation.
The idea is that the Great Recession was proceeded by a large expansion of credit and it was
important that it was followed by collapsing house prices, employment, consumption, and
a spike in delinquency. Just to give you some numbers, if you look at, for example, the
flow of funds, mortgage liabilities increased by almost $6 trillion during 2000 to 2006.
If you at employment, employment declined and peaked at 10 percent in October 2009.
Moreover, there is also some important regional differences. In particular, exactly the regions
that experienced the larger increase in the credit availability are also the ones that
experienced the most severe boom and bust in house prices, employment, consumption. Just to give you a sense, the left panel are
mortgage liabilities for households in the U.S., and then you have house prices, employment,
and delinquency rates. As you can see, basically exactly the same countries that experienced
the largest increase in credit during the boom years are also the ones that experienced
the largest boom and bust in house prices and employment, and then for delinquency rates,
delinquency first went down because you were relaxing the credit constraint, and then these
basically spiked during the bust. So what is our research question? The research
question is how much of these fluctuations, how much of these boom-and-bust patterns in
house prices, in general, in real economic activity can be explained by an outward shift
in the credit supply? In particular, much of these can be explained by not a shift in
credit supply towards riskier borrowers. So what is the identification challenge? The
challenge is that if you observe these counties, you’re going to see a correlation between
all of these variables. In particular, counties that experience higher growth are also going
to demand houses, house prices are going to go up, consumption is going to go up, and
they are also going to have higher demand for credit. So in the end you, are only going
to explain, only going to observe, a correlation among all these variables, and credit is not
really doing anything better. So what’s our identification strategy? Our
identification strategy uses an unnatural experiment, like John said before, and uses
two main regulatory changes in the U.S. The first one was that, starting in 1999, many
different states adopted what are called anti-predatory laws. I’m going to be a little bit more
specific on what these really entail, but these anti-predatory laws basically were restricting
the amount and the terms of lending towards riskier borrowers. But the most interesting
thing is that in 2004, the OCC, the Office of the Comptroller of Currency, passed what’s
called the preemption rule. This preemption rule only applied to national banks, so this
is great validation for us, because now you are going to have states with anti-predatory
laws, states without anti-predatory laws, and then you are going to have before and
after these preemption rules, and within states that adopted anti-predatory law you are going
to have more or less exposure to this preemption rule, depending on the fraction or the importance
of national banks within these countries. This is where we are going to use exactly
the passage of this preemption rule as a possible shock to the credit supply towards riskier
borrowers, because remember, these are anti-predatory laws. These are exactly for the high-cost
mortgages in the U.S. So let me tell you, in a nutshell, what are
the main results. First of all, we show first a very important first stage. So if we compare
counties with a high fraction of national banks, in states with the anti-predatory laws,
compared to counties with a lower fraction of national banks, then we see an 18 percent
increase in the credit supply during the boom years, right after the preemption rule, after
the OCC implemented the preemption rule. And this was followed by similar bust in the credit
supply during the 2007-2010 period. Then we used these as an instrument and we
say, okay, how much of the increase in house prices can actually be explained by upward
shift in credit supply? And we see that the 10 percent in the credit supply leads to 3
percent increase in house prices growth, which compounded over the boom years is about 10
to 12 percent. Then we look at employment, especially employment
in non-tradable sector, because our mechanism is going to work through the local demand,
and we show that the 10 percent increase in the credit supply is going to increase employment
by about 2 percent. And the delinquency rates, first they are going to collapse and then
they are going to increase very significantly. In particular, we’re going to look at almost
a 30 percent increase during the bust. Also, another interesting thing that is going
to provide some evidence of really the mechanism behind our effects are heterogeneous effects.
So we are going to look at counties where we expect our effects to be stronger, in particular,
counties with higher fractions of prime borrowers, fraction of more inelastic housing, or less
affordable housing, and those are exactly the counties where all our effects are much
stronger. And then I’m going to try to spend a couple
of minutes just to convince you that this is indeed very robust results to a bunch of
different alternative hypotheses. Now, let me just rephrase our contribution.
Since there has been a lot going on in terms of the research and literature on this debate,
our contribution is the following. If you think all that was going on up to now—for
example, Mian and Sufi type of papers, they had in the back of their mind a positive shock.
This was a shock that might come from a relaxation of the lending standard, inflows from China,
securitization—it doesn’t really matter. And then they were using the elasticity as
a static regional characteristic to measure how this shock propagates in the U.S. Our paper, in contrast, is a completely different
approach. We say we are providing hopefully a credible instrument for an outward shift
in the credit supply. So we have a direct mechanism, a direct measure of these outward
shifts in the credit supply. And then we can control and we can be sure to control for
the regional difference that might or might not be correlated with this credit supply. Now, let me go to the regulatory framework,
and I should skip this slide, but one thing that we take advantage of is the fact that
the different type of lending institutions are regulated by different agencies. Okay.
So from the OCC for national banks, LTS, HUD, the National Credit Union, and so on. The interesting thing is that in 1994, the
federal regulators implemented a federal anti-predatory law, and this was trying to limit, for example,
the high-cost mortgages, the interest rate, the prepayment penalties, the points, and
so on, the disclosure rules for these mortgages. The problem was that these only applied to
less than 1 percent of the mortgage market. So starting in 1999, the first state was North
Carolina, implemented these anti-predatory laws, and by 2007, a little bit more than
20 states in the U.S. implemented these anti-predatory laws that actually had a bite in the mortgage
market. Why do they bite? There are a bunch of papers
that actually do studies and they show that these anti-predatory laws have some effects.
For example, after the implementation of these anti-predatory laws there is a reduction in
the full trades, there is a reduction in the usage of complex mortgages, there is a reduction
in prepayment penalties. There is also a paper by Keys and others that uses actually the
passage of these anti-predatory laws also as an instrument for securitization. Why?
Because credit-rating agencies didn’t really want to rate securities so that we’re using
loans originating in states with anti-predatory laws, because they were almost illegal. Then the most important thing that we need
to thank basically the OCC for this is that, in January 2004, they passed the preemption
rule, and what they said was that specifically preempted all regulation pertaining to the
loan-to value ratios, the terms of credit repayment, the aggregate amount of funds that
can be lent, secured by collateral, the access to the credit reports, disclosure and advertising,
the rates of interest on the mortgage loans. So everything about the credit market should
have been regulated by the OCC and not by the state regulators. So we’re going to
use exactly the passage of this preemption rule as a positive shock to the credit supply,
from national banks. Just to give you some anecdotal evidence that
these actually matter, I’m showing here a quote from New Century 10-K, in 2004, and
the first quote basically reads that this would effectively—and it’s referring to
the anti-predatory laws—this will effectively preclude us from continuing to originate loans
that fit within the newly defined thresholds. So they were actually worried about these
laws. And then the other, second important thing is that 2004 is actually the passage
of the OCC preemption rule, and so they also say some of our competitors who are owned
or are subsidiaries of national banks can actually lend to these borrowers, and so these
might actually hurt us, so we are going to be negatively affected by this preemption
rule. The data, this is pretty standard. We put
together HMDA data, because it gives us information on the originators of these different loans,
so we can actually distinguish between a loan made by an OCC-regulated bank versus OTS,
HUD, and so on. And then we also have some information that we take, for example, from
the literature, on the anti-predatory law measures or we really state New York state
prices. We get it from Zillow. Other measures, or county-level measure, we can get it also
from, for example, the New York Consumer Credit Panel. We get there, for example, the FICO
score and the mortgage liabilities. Let’s think about, for one second, the research
design. I basically told you that there are two main sources of variation. One is that
across states, over time, they adopted this anti-predatory law. One first approach might
be, well, let’s start to compare states that adopted these anti-predatory laws versus
states that did not. So, for example, the arrows that are here. The problem is that
the introduction of these laws are going to be endogenous, because these are going to
reflect conditions of the local mortgage market, so we don’t want to do that. A second approach might be, let’s go a little
bit more micro data. Let’s look at within the states and compare counties with the higher
fraction of national banks in the pre-period versus counties with a low fraction of national
banks. The problem here is that we are going to compare different types of lending institutions
and, for example, comparing national banks versus independent mortgage originators, their
growth and their source of funding are going to be very different, just because, for example,
securitization has been much more important for independent mortgage lenders than for
the national banks. So we don’t want to have this approach. So what we do is basically try—so we use
all the variation that is there. So we are going to compare counties with a high fraction
of national banks, counties with a low fraction of national banks, across states and across
time. So the identification assumption is going to be that this difference between the
counties with the high and low fraction of national banks does not differ across states.
I’m going to show you that, for example, the fraction of the counties with the higher
or lower fraction OCC might change, might have different characteristics. The important
thing is that these different characteristics are not going to change across different states. The first thing I want to show you is that
indeed I can use the fraction of lending coming from OCC in the pre-period. What is my pre-period?
We are going to define it as the fraction measured in 2003, so this is the fraction
of all loans made by OCC lenders in 2003. And this is the correlation between the fraction
in 2003 and 2005. The two lines are for states with and without APL laws. Why am I showing
this? Because I want to show you that, first of all, these measures are very stable and
these don’t change across different states. So we are going to use this measure for our
main variation. Let me show you the main result. Basically,
the main result can be seen also just by these summary statistics. We look at summary statistics
for loan amounts of house prices and employment in states with and without, so the states
without anti-predatory laws are the first two columns, the states with anti-predatory
laws are the third and the four columns, and then the fifth column is basically our estimator,
so the difference between these different counties. We first look at the boom years, and as you
can see we show that even just without controlling for anything we show that positive effect
for loan amount, house prices, and employment, while the design is exactly reversed during
the bust years, and there is basically no evidence of pre-trend. So we look at the years
before and we show that there is no differential trends across these different counties, across
these different states. Now let’s go a little bit more specific.
First of all, let me show you the first stage. The first way in which we’re going to show
this is by looking at the loan amount and the main coefficient is going to be the interaction
between APL, which is 1 or 0, depending on if the state has an anti-predatory law, post
after the preemption rule, and here OCC is just an indicator if the loan is coming from
an OCC lender. What we’re showing here is that even controlling for county-agency fixed
effects—so we are absorbing a lot of variation here—or even looking at the counties by
year fixed effect, so any demand level variation, we show that counties with a higher fraction
of national banks, there is a higher growth in lending. Here we show, instead, a similar result, but
instead of looking at only loans made by OCC, let’s look at the whole loan origination
in these different counties. The main variable is always going to be the tribal interaction
between APL, post and the fraction of OCC, controlling always for county and effect,
and we find basically a positive and significant effect during the boom years. And it doesn’t
really matter if we control for population, income, the elasticity of housing supply.
For example, our instrument is completely unrelated to elasticity or housing supply.
It does matter for loan origination but it doesn’t matter for our estimates. And the
same also for the fraction of subprime borrowers in the pre-period. It doesn’t really matter
for us. Here, basically, is a graph that shows you
exactly the same result but over time, so the dynamics of this effect. As you can see,
there is basically no effect during the pre-period, so in 2001, 2002, 2003, but then there is
a boom and a bust. So these also assure us that indeed we are not capturing any difference
in the characteristics of these counties in the pre-period. Let me skip to the economic magnitude, but,
as I say, it’s about at least 11 percent increase after the preemption rule. Let me go to the first main variable, which
is going to be house prices. For house prices, we look at our main introduction, and we see
that basically we’ve observed that in counties with a higher fraction of national banks,
after the passage of this preemption rule, house prices increased more. We can actually
use an IV here, and that coefficient of 0.33 tells you that a 10 percent increase in loan
origination increases house prices by 3.3 percent, in yearly, annual growth. Here is exactly the same graph that I was
showing you before. This is the dynamics. So you see no pre-trend before, and then you
see an increase in house prices right after the preemption, and then a decline over time.
So exactly the boom and bust pattern that I was talking about before. And below, the
first two columns are for the boom years, up to 2006, and then the third and the fourth
column are from 2007 onward. And you see basically the negative effect. Controlling for elasticity,
controlling for the fraction of subprime doesn’t really matter. Employment in non-tradable. So is there any
other effect in the real economy of this expansion of credit? We actually see—and let me go
directly to the IV—we see that a 10 percent increase in loan origination increases employment
in non-tradable sector by 2.2 percent. And the same pattern in terms of boom and bust. In terms of delinquency, what was I saying
before when I showed you the graph at the beginning was that delinquency first goes
down and then spikes, so these table are constructed by looking at the period of 2000 to 2006,
and here you see a negative effect. So exactly in the counties where national banks were
lending like crazy to subprime or riskier borrowers, these were not defaulting, up to
2006. And even by quote a lot, in the sense that this is a decline in delinquency rate
by almost 30 percent. However, if you look then at the bust, then
the third and the fourth columns here show you the delinquency rate spikes, an increase
by almost 60 percent. So those counties where national banks were lending during the boom
years also end up being more fragile because those are the ones that experience the higher
default rates. Now, let me show you some heterogeneous effects,
because I think these are interesting, and also these provide some evidence of our mechanism.
Two that I brought here are about subprime regions and inelastic regions. As you can
see, all the boom and bust, once we interact for the subprime counties—so below 660—we
see that both the boom and bust for loan amounts, house prices, employment, and delinquency,
are much stronger. And the same for inelastic counties. So this tells us the story is probably
exactly what we are trying to push, that these national banks were lending to the low end
of the distribution in terms of household borrows, and these are exactly the counties
where we experienced the stronger boom and bust cycle. Now I want to mention, in the last one second,
the different robustness checks. We do a bunch of different things. We can control for securitization
activity, so we observe how many of these loans were securitized. It doesn’t really
matter for us. We look at state borders, so we can do a regression discontinuity across
the borders. We can look at CRA lending, so all of these are going to show that it is
actually due to household debt and not firm’s investment. If there is any regulatory arbitrage
we found no effect there, and we also show some loan-level evidence that indeed those
characteristics, for example, that Chris was talking about—balloon payments, high costs—those
are exactly the ones that could originate in more by national banks after the preemption. Thank you so much. Okay. Awesome. Our next speaker is Jialan
Wang from the Consumer Financial Protection Bureau. She will be discussing the papers.
Thank you. Thank you so much. Thanks, everyone, for attending
our first CFPB Research Conference. I hope we’ve really shown you, in the past couple
of days, how much we go above and beyond to really make this a great conference. And,
in particular, I think what distinguishes this conference from every other consumer
finance conference, as you’ve seen, is that for the amusement and entertainment of the
audience we, the CFPB staff, have subjected ourselves to discussing three or four papers
in 10 minutes, and we’ve seen a variety of approaches, of coping mechanisms for how
to do this. We started off with my colleague, Jon Lanning, took the auctioneer speaking
style approach, and I think he has a really promising career ahead of him, if the CFPB
thing doesn’t work out, as a cattle auctioneer or selling pork bellies or something like
that. Dustin took the approach of self-deprecation,
making fun of his ginormous and misshapen head. Today, we’ve seen various approaches—camels,
hockey metaphors—and really beautiful and simplistic approach of just taking more than
10 minutes, which is awesome. I should have thought about that. Sadly, I didn’t before
coming up with these slides. Not to disappoint anyone in the audience, my coping mechanism
for this task is going to be equally futile, so I’m just going to move on ahead. The session today is about credit supply,
and inefficiencies in credit supply have implications, as we’ve seen, for both the macro economy
and consumer protection. Of course, different sets of regulators are tasked with these two
different sets of goals—safety and soundness and consumer protection—and the point I
want to make is that these are distinct but also interrelated goals. In particular, these
two goals are impacted differently by changes in credit supply for secured versus unsecured
credit markets, and we’ve seen papers in this session on both of those types of markets. For the housing market, I think that these
two goals, of consumer welfare and also macroeconomic stability, generally we can think of those
as aligned, in terms of having exacerbated boom and bust cycles in housing prices, it
also has negative impacts for consumers in terms of a lot of foreclosures, lots of externalities,
consumer bankruptcies, et cetera. However, for unsecured credit markets, these
interactions might be a little bit different, in that default might have fewer implications
on both firms and consumers in the unsecured credit market. So one fact that stays with
me for a long time is that credit card issuers actually remained quite profitable, at least
had positive profits despite the depths of the Great Recession. And one potential reason
for this is that they’re able to get lots of fees and interest, and, in particular,
so the Agarwal et al. paper that was mentioned early on today actually showed that the more
subprime, the less credit-worthy borrowers who were more profitable. So this makes it
a little bit distinct from the impacts of default for secured credit markets, like the
mortgage market. And in particular, though, one potential concern
in nonsecured markets that does impact consumers and potentially the macroeconomy is about
consumption, that excess supply of unsecured credit may lead to greater consumption volatility
instead of consumption smoothing, which is one of the goals of having these instruments
be available. I’m going to first tackle the unsecured
market, the first paper, on “The Welfare Effects of Online Payday Loans in the UK”
but Stefan Hunt and colleagues. I just want to take the opportunity to say this is wonderful.
Just absolutely, brilliant work, and I think all the regulators in the room can really
appreciate how much hard work and courage it takes to carry out this type of ambitious
research program in the confines of kind of high stakes, very controversial rulemaking,
with a lot of time pressure. I just absolutely want to salute your efforts. And I think not
only doing the type of research, collecting this data, and collecting primary data on
top of administrative data, I think this really pushes forward the literature about this type
of loan, in that this the frontier in terms of having actual loan records—a lot of the
literature doesn’t have actual payday loan borrowers and is using geographic variation—as
well as most of the markets who were capturing spillover effects across lenders, and combining
both survey and credit bureau attributes. So just to say a few things, I think one thing
that’s really, really interesting is the complementarity between this type of high-cost
loan and other forms of credit, which is something we didn’t know before, and why do we see
this complementarity? One plausible reason is that they have to take out other forms
of credit in order to repay this payday loan, but I don’t think, from your study, we really
know, ultimately, and I think that’s really important question. Another question is, would any type of credit
be bad for this type of consumer that is seeking this type of credit, or is it something about
the particular types of loans that are available in the market today? Is it something about
the cost or the loan structure of these particular types of loans, or would these consumers simply
be harmed by any type of credit? Now moving on to the futile portion of this
presentation, in terms of putting together the three remaining papers about the mortgage
market, which I think really fall together nicely into different sides of a debate about
what really happened in the financial crisis. I’m going to go through just to set up the
debate, a brief history of the financial crisis in 1 minute. In the lead-up, in the early to mid 2000s,
what did we have? Low interest rates. So the famous 1 percent regime after the bust of
the dot-com bubble. So we had low interest rates, and also doubling of global imbalances
during this time, so central banks were investing in Treasury assets, which may have further
depressed interest rates and led to a boom in housing construction, but also a lot of
institutional investors, a lot of global capital was trying to flow into the U.S. mortgage
market. But U.S. household credit worthiness was not really increasing during this period.
In fact, household income was slightly decreasing during the early 2000s, so there wasn’t
a growth in credit worthiness to absorb this large influx of credit supply. The story we tell ourselves, or a common narrative
of what happened during the crisis is how can the U.S. households absorb this large
flow of credit when their credit worthiness and income wasn’t really increasing? The
only way that could have happened is if underwriting standards were deteriorating, and these deteriorating
standards contributed to the boom and bust cycle in housing prices, and also the increase
in default rates that started in 2006 and then filtered out through the entire financial
system, through the magic of CDOs, and led to the Great Recession. But is this narrative actually correct? That’s
what three of our papers today really tried to answer. Marco’s paper I think is generally consistent
with this story. So we see each of these different variables, we see an instrument for credit
expansion that did lead to origination growth and increase, boom and bust cycle in housing
prices, as well as a boom and bust in employment growth and delinquencies. The other two papers
say no. Chris’s paper looks at defaults, doesn’t look at the originations and other
things, but what he finds is that most of the variation in mortgage defaults across
cohorts was actually caused by changes in prices, and very, very little actually changes
in borrower characteristics. The final paper, Felipe’s paper, also is
on the side that it’s actually the price increases that led to what we saw, and, in
fact, the income homebuyers was increasing relative to average residents in high price
growth ZIP codes, and there wasn’t an overall deterioration in lending standards. Just to kind of dig a little bit deeper in
these three papers, what we have is, for the Di Maggio and Kermani paper, I think it’s
a promising new instrument that really pushes forward sort of this literature investigating
the role of credit supply, and I think, as an empirical researcher, one of the sad facts
that I’ve had to confront is that adding more diffs doesn’t actually decrease the
number of assumptions we need for identification, sadly. I have to say I actually completely
believe their results, but the parallel trends assumption is really critical for any type
of different-and-different, or triple-different specification. They do a really nice job—he
showed the graphs where, in the pre-period we really see now effect and it really starts
after 2004, when their instrument is active, but a lot of things were changing on more
than the ordinary magnitude of 10 percent increasing credit supply. So, to be honest,
I actually believe the results but I think maybe a little bit more micro evidence such
as looking at the particular lenders who were OCC-regulated, did we see a discontinuous
change within those individual lenders? But I have to say they did an extremely diligent
job putting together lots of data sets to analyze a lot of alternative hypotheses. I also really loved Chris’s paper. I think
it’s an incredibly clever and well-executed paper, and really, really clearly written.
I think it was shocking to me when I first saw your paper how well this instrument actually
works, and it’s not driven by the things you would think, the subprime expansion and
things like that, that we would maybe normally expect to be driven by these price cycles. But I think one thing I would like a little
bit more discussion on, especially given how well you discussed everything else in the
paper exactly where your identification comes from, is what’s the role of the aggregate
versus cross-section price effect? I think in your talk, as well, you said we use prices,
changes in prices, but in fact there’s an important aggregate component as well as the
cross-sectional component, which is where your instrument really comes in. So he’s
using a cross-section source of variation but I think the loading on that is different
for each time period, in order to capture that boom and bust, which I think sort of
the boom and bust part of price changes is the same across time, and I think the aggregate,
going back to Bernanke and the global savings glut, I think the role of aggregate supply
in potentially impacting prices, is really an important factor, and maybe you could just
explain a little bit more how your approach gets into that. And finally, Felipe’s paper, he was very
clear that it’s not a causal paper. I think they really show a lot of important descriptive
facts that are really important to know, about the relationship between income and loan origination.
And to be honest I have to say we don’t have time, of course, to discuss a lot of
really interesting parts of these papers, and particular data, so I think Felipe and
Marco used HMDA data, Chris uses CoreLogic, and I would really encourage anyone who’s
interested in this to look at Mian and Sufi. They have a response to Felipe’s paper that
really gets into the details of income measurement, which I’m just not going to be able to go
over here. But I think, again, in terms of if we really
want to put it as an either/or—is it prices or is it supply—at least in the version
I read I think Felipe presented a slightly different conclusion this time that didn’t
emphasize saying that price expectations were really the driver of mortgage growth. But
the fact that we still see is that mortgage growth was driven by the extensive margin
in these regions that, ex-post, happen to have high price growth. So there is still
a lot of supply growth and we did subsequently see, ex-post, a lot of price growth. So I
think there could still be a role for supply there in terms of interpretation. I touched on this with all of the papers,
in terms of what I think policy-wise. The elephant in the financial crisis post-mortem
literature is about the role of the aggregate supply and aggregate price dynamics, which
I totally understand in papers why we would want to utilize cross-sectional variation
for identification purposes, but I think for policy we really want to understand a little
bit more the aggregate effect. In summary, this session covered the effects
of credit supply for both unsecured and secured credit markets. And we saw a really nice example
of the negative effects of expansion of unsecured credit, and I think it’s fair to say that
the debate will continue about what exactly was the role of prices versus supply in the
financial crisis in mortgages. Thanks. Okay. I suspect there are New Yorkers in the
audience thinking, boy, those CFPB discussants can really talk fast. How about we start off—we only have 10 minutes
before happy hour—so why don’t we give the panelists an opportunity to respond to
our discussants comments. I’ll go first. My mic is already live. Thanks
for the really savvy comments. I think you touched on a couple of themes that we can
all take away. One is that, at the end of the day, a lot of what we really care about
is aggregate effects, and we end up shining the light on effects that we can identify
with cross-sectional variation because that’s what we can identify. And it’s always nice
to find a way to tie back to this aggregate effects. Personally, I think a lot of paper that are
published, even in top journals that identify off of cross-sectional variation and then
try to do, in the conclusion, a back-of-the-envelope extrapolation to aggregate up to the aggregate
effect, probably get it wrong and we’re probably missing a lot of aggregate effect.
So that really is the golden goose in empirical work, is to be able to identify some aggregate
effects. I think a lot of us highlighted some of these. It would be really nice to know
about the effect and incidence of the global savings glut or the effect and incidence of
macroprudential policy. A lot of policy operates at this national level and it’s hard to
identify its effects. So those are all things that are definitely worth thinking about,
and I’m glad to have had them highlighted. Thank you. Jut highlighting one thing of where we could
next with some of this payday work, and actually touching slightly upon what Darryl mentioned
earlier as well, which is the ability to repay what we call affordable lending, I think there’s
one potential to use this kind of analysis, to look at heterogeneous effects. That’s
one thing we really didn’t focus on here was how different types of individuals could
be differentially affected. And I think we’ve got enough power here to try and use quite
general methods to try and look at some maybe machine earning and these kinds of things,
to try and look at exactly which kind of variable are related to differential effects, and then
we can actually try and think of what would constitute good, responsible lending roles.
And rather than just as we’ve done historically, essentially give it over to firms to kind
of think about, what should be a responsible lending role, we could actually try to use
this sort of analysis to kind of further ourselves. First of all, thanks a lot for the great discussion.
Let me say that, for example, this is just our first step to this question, and this
was the macro paper, in terms of understanding what are the aggregate implications of a shift
in the credit supply that comes from the regulation. Then we have a follow-up paper that actually
looks at much more micro, long-level evidence and points out exactly that some of the evidence
that Chris was showing, that during the boom years one thing that changed was the type
of contracts. And we show that these types of deregulation affected exactly the types
of contacts the national banks were offering to these riskier borrowers. So here in what I showed you was only the
aggregate amount and the loan amounts for these households, but then when you go to
the loan level you actually can see how many of these features, exactly penalties, interest
only, negative amortization, balloon payment, and if you look at loans originated by national
banks after the preemption, they would start originating at 10 to 15 percent more over
loans with exactly these type of characteristics. And then you can trace out, for example, default
rates for these households, matching on observables, and these are exactly the ones that they’re
much more likely to default ex-post. So thanks again. This is, I think, just the
first step for much broader research. Thanks a lot for the discussion. It was great.
I guess it was a great challenge to try to put together in 10 minutes this discussion
of four papers, so I appreciate it. I just want to say two things to make it extremely
clear, kind of how robust our results are. Basically, everything that I told you today,
it doesn’t rely on any measure of income. You can throw literally whatever measure of
income you want, and with the patterns that I showed you they will stand up. The other
thing is the extensive margin is not coming from low-income areas. It’s coming from
actually the high-income ones. So it’s basically, if you want, almost like the opposite interpretation
that the original paper had. But having said that, I just want to leave
you with a thought, in a way. I think that we’re trying to push this paper and this
agenda just because we think that, as researchers, we get super comfortable with the narrative
that we understood, and it was basically highly exposed with respect to what happened during
the crisis. Basically, just by looking at the data in a slightly different way you get
a very different view of it, and that puts a lot of pressure in trying to understand
precisely what happened, because the implications, going forward, to try to address these type
of things in the future, are very different. And that’s what this basically attempts
to challenge various findings is trying to do. So thanks a lot again for the opportunity
share that with you. We’ll go with a question over here, Chris
Carroll. Thanks. I wanted first to complement Jialan
on an amazing job, but to point out that she had the advantage of being the last discussant
and so she was able to combine all of the strategies of the prior discussants—the
incredible swift speaking style, and the modest self-deprecation, and the taking a little
bit of extra time. But that’s probably the optimal strategy
to learn a little bit from everyone that came before. I loved this session. I loved all the sessions.
The one thing that I think might be very valuable for all of these kinds of papers to add, to
the extent that there are any data that can speak to the question, would be data information
or models on house price expectations, because, as we saw in the paper earlier on expectations
in house price dynamics, in some sense you must have some story of expectations to explain
the consequences of the supply of credit on booms and crashes and bubbles, and until we
have that channel integrated into our story, I don’t think you can answer the question
without having a coherent story about expectations and how they are affected by the supply of
credit, both at a moment of time and dynamically. Jialan touched on that at the end, and appropriately
didn’t say much more about it because there isn’t much more to say. But I think that’s
a very important venue going forward. It’s unfortunate we don’t have better data on
expectations. I think this is all very interesting data
and I liked all the instruments. On the payday lending, I was worried about that, thinking
of it as a causal thing in terms of all the other loans. I mean, presumably there’s
some shock that happened in people’s lives that made them need to borrow, and they would
try to borrow on all margins, and they sometimes could do payday loans and other things, and
not always exactly at the same time. The opportunity to do something else could’ve come up a
little later. So I don’t think you want to assume that payday loan caused that. On this other thing, contrary to what Jialan
was saying, I don’t think these things were debated at all but just painting different
sides of the same picture, a picture where you have credit expansion to kind of all the
income groups fairly equally, where the sense in which you have worse quality is mostly
about a higher loan-to-value ratio, not even a ton on the FICO scores, though may be a
little bit on that. But primarily the same people in terms of income and FICO scores
but higher loan-to-value ratio. That certainly, even mechanically, that gets you more demand
plus making it attractive to some people. That does have an effect on house prices that
gives you a multiplier. But on the way down you have some very specific macroeconomic
things, in particular, the way they were packaged to affect the banks, that then are a huge
part of the story later on. I love having all this in one session, to
paint that combined story, but I don’t see any conflict at all between these different
pieces of the puzzle. To me, there’s at least one totally coherent story that fits
all these things, and since I believe the identification in each case, I’d want to
look for a story where all of these thing were true, rather than to talk about it as
a debate. I’ll take that first. In terms of the payday
loans, what we’re comparing is people who were denied the loans to people who were accepted
for the loans. There’s some sense in which it’s not a completely pure control group.
It’s not like the people on the left-hand side did nothing, and then you have a treatment.
But I think we can ascribe all the differences that we see after the fact to that. Again,
we’re controlling for everything else right at the margin, and you’ve got sort of local
polynomial regressions on both sides. I mean, that is the difference. I mean, whether it’s
purely from the payday loan, where there’s an effect on being denied directly, maybe
there are some effects, but I don think what we know is we can’t say it’s causally
related to comparing being accepted versus being denied for a payday loan. I’ll add we also see each other as consistent
and circumscribing a lot of truth into one great hole, if you will. To Chris’s earlier
point about house price expectations, data is a big challenge in this, as has been mentioned
over and over again. I have a section of the paper that I don’t get to, even when I have
an hour and a half, let alone 20 minutes, where I do look and say, well, what is it
about house price declines that seems to be so effective? And I can put in a measure fairly
well form CoreLogic of your equity position, and a very flexible function of how far under
water you are, and even over and above that, recent house price momentum has a big effect. So I can’t say that there’s like a missing
first stage there that maybe I have to appeal to Charlie’s paper for, that that is because
of expectations. It’s also just because of when houses prices are falling no one’s
showing up to your open house, whether you have positive or negative equity, so you have
a very liquid asset. But there seems to be scope for expectations playing a role here. Hi. Good afternoon. My name is Asuntha Chiang-Smith,
and I actually refer to the State of Maryland Housing Department, so I’m very much on
the ground on what we’re seeing locally and nationwide. One area maybe for research
that we would love to find out about is, we’re living with the recession and the foreclosure
crisis, and I continuously get asked, “When can we see the end of this foreclosure crisis,
in a sense, to reach levels before the Great Recession, or are those levels never going
to come back and are we looking at new picture?” I think it’s a fantastic question and one
that we all think about, and yet we are not certified as experts by the virtue of the
particular papers that we’ve written today to opine on it. Credit does seem to be pretty
tight right now, and there’s certainly a worry, as I’ve talked with policy-makers,
what’s the right balance here between having accessible credit, which both fuels equity
and price recovery, but also could potentially fuel a future boom and bust in the future?
So just looking at the distribution of credit scores of people who are able to get mortgages
right now, it’s much, much higher in terms of the mean and the distribution than it used
to be. So credit seems to be tight. That seems to be inhibiting a recovery. On the other
hand, we’ve seen this rash of defaults that happens in a price cycle when you loosen credit.
So there are these tradeoffs and no one seems to have the right balance in mind. I’m not
sure that we’re at that balance there now. One interesting thing I’m sure that you
see on the ground in Maryland is that there’s tremendous heterogeneity across space in the
recovery. Arizona, for example, recovered pretty quickly. Nevada has not recovered almost
at all. And we see different things all over the place, and one thing that is very interesting
for policy is the fact that foreclosure timelines are so different. There are some studies that
are starting to come out that say—and I have some grad students working on this—that
when you have a long, long foreclosure timeline it can inhibit this kind of recovery, and
there’s almost something about ripping it off like a Band-Aid, where you can get the
market to start to heal. So there’s lots of variation in these policies and I don’t
think anyone’s really hit one out of the park yet in answering those questions. Okay. I will turn this over to Chris to give
his closing remarks, or Ron. Ron and I are up here just to say thank you
to everyone that made this conference the tremendous success that we think that it has
been. I particularly would like to thank the CFPB staff, all of whom really stepped up
to the plate and did an amazing job in organizing this event. Let me turn it over to Ron. So I get thank the honor of everybody else
for having submitted their papers, for having participated today, presenting their papers,
discussions, questions from the floor, and with that, to officially close our first CFPB
Research Conference, so thank you all very, very much. All the best.

Leave a Reply

Your email address will not be published. Required fields are marked *