(music)
- Good afternoon.
On behalf of the Center for Global Economy and Business,
Stern School of Business and New York University,
I wanna welcome you all and thank you for coming
to the second David K. Backus Memorial Lecture.
David Backus had very broad interests in economics
that spanned theory, applications, public policy.
It doesn't mean he liked a lot of the research
he read in theory and application public policy,
his interests were deep, they weren't shallow,
they were just very broad.
And Dave had high research standards.
He often,
when people ask him what kind of research do you value,
he often quotes, and Chris will back me up on this,
he'd often quote his friend and co-author
Fin Kydland's recipe for great research,
and it was, step one, ask a really important question.
Step two, acquire the tools it's gonna take
to answer that question.
Step three, take your model seriously.
Step four, respect the data.
And step five, and the most important step is,
be humble about your results.
So it's a great honor
to be able to introduce somebody today
that shared both Dave's broad interests
across lots and lots of areas of economics and finance.
But also who's research has consistently checked
every item on Finn's list.
Dave and I would often sit around,
sometimes over a beer or two,
and talk about who we thought
was doing really great research on economics and finance,
inadvertently, Darrell Duffie's name would come up,
and we would kind of fall into this kinda set piece,
this little comedy routine we would do,
and Dave would say,
you only like Duffy because you're Canadian.
(audience laughing)
and I would say, no Dave that's wrong,
I only like Duffy cause he's Canadian.
(audience laughing)
And we would always chuckle over that.
And it's a great honor and a privilege to introduce today
the lecturer for the David K. Backus Memorial Lecture
from Stanford University, Professor Darrell Duffie.
(audience applauds)
Thanks, thanks everyone,
it's generous of you to come out and honor Davis Backus.
And it's a great opportunity for me to be here
to honor David Backus,
with his wife Marilyn and Jason sitting right here,
thank you.
I think I may not have checked all the boxes on this one.
The last one, the last step five,
be humble about your results,
I think I can do that.
The title is much more provocative
than the results will suggest,
it's extremely preliminary work,
and let me,
we're already on the slide where it says
extremely preliminary results.
So the title, no longer too big to fail,
let me just motivate the title,
and then you can decide how much progress
we've made towards that.
The title refers to the kind of killer moral hazard
that let into the financial crisis, that the largest banks,
the so called globally systemically important banks,
were able to get very cheap credit from markets
because it was widely believed by creditors,
who will all try and convince you of that from the data
that these banks were too big to fail,
that is if they failed,
they would cause a crater on the economy,
and that the government couldn't allow that
so that they would have to be bailed out.
Now Mervyn King is sitting right here,
so I have to be very careful about what I say,
the government or essential bank would do or would not do,
but he can cross swords with me whenever he likes.
And by the way I hope you'll all take the privilege
to interject, say no that's wrong,
or ask questions.
And I'm gonna try to leave enough time at the end,
that we'll have a good opportunity to have a discussion
about the topic.
This is joint work with Professor Antje Berndt
at Australia National University,
and Yichao Zhu, also at Australia National University.
What we're gonna do is we're gonna walk through the data
that we can get from asset pricing
which was David Backus' specialty,
he did asset pricing in a macro-international setting.
And I'm going ta here focus on what asset prices tell us
about the degree to which the creditors of the largest banks
really believed that these banks were too big to fail
or alternatively that the government would allow them
to fail and the creditors to take loses.
And there's a big division point there,
I will try to convince you that after the financial crisis,
the markets assumed that the likelihood of a bailout
was much lower, okay.
So that's the major message.
We'll also get a look how much
that pre-crisis bailout meant
in terms of the subsidy to these large banks
to borrow money and get too big to fail.
Okay so just before doing any analysis,
let's just look at what the markets were saying
about the cost of wholesale credit to the largest banks.
And I'm gonna show you this at the one year maturity
and at the five year maturity.
So one year first is on the left hand side chart,
measured by the difference between
the London Interbank Offered Rate,
which is the cost of unsecured credit
to large banks in London,
and the overnight index swap rate
which is a proxy for the risk-free rate.
So that difference is measure, it's a credit spread,
it's a difference,
it's a measure of the expected losses
on a so called risk-neutral basis
to creditors associated with lending money
to these large banks.
You can see that until the crisis, this credit spread,
if you can see this far, was about 10 or 15 basis points.
So a razor thin charge for cost of credit
because the banks, in my view,
were viewed as too big to fail.
And then after, well of course during the crisis,
the banks were immediately viewed as potentially failable,
but then after the crisis,
those credit spreads did not come back down
to 10 or 15 basis points,
they stayed up kind of in the vicinity of 50 basis points.
And you might have said,
well maybe the banks were lower quality after the crisis,
that would be a real failure of post-crisis regulation,
it would mean the banks are less safe, right?
I'm gonna try to convince you that's not the case
and I think it's pretty much unanimous, almost unanimous,
that the banks are much safer, that is,
there's much lower likelihood
that they will run out of capital.
So we have to explain it a different way.
Before we do, at the five year point,
we're gonna measure credit spreads
using the credit default swap market,
which in, at least an arbitrage-free market,
is the same as the five year par bon spread.
The additional cost for borrowing
associated with expected loses,
we've measured it for both US and European banks,
the five largest US, five largest European,
and you can see again pre-crisis at the five year point,
25ish to 30ish basis points
as the measure of the expected losses to creditors
for lending to the largest banks for five years.
Post crisis, of course it went way up,
and then there was the European debt crisis,
it went up again,
but it's settled out at a level, ups and downs,
but a level far higher.
Again, how could that be if you believed the story
as I do that the banks are a lot safer?
Just in terms of their safety, let's take the,
almost all the data that I have today
is for the large US banks
and I am not sure yet what the answers are gonna be
for the European banks.
We'll see whether the resolve of the governments
to let the banks creditors take a loss
is actually gonna work in Europe,
but in the US I think I'll convince you
that it is gonna work.
In the US the banks are much more solvent now
then they were.
In this chart is measured by
tangible common equity to assets.
In red, 2007,
in blue, 2015,
a lot more tangible common equity to assets.
For Goldman Sachs, Morgan Stanley, City Bank,
Bank of America, J.P. Morgan and Wells Fargo.
So at least by this measure,
which is only an accounting measure,
the banks are much safer.
If you account for the volatility
as estimated by our model of the assets
and measured instead in how many standard deviations
of changes in asset value
the banks have as a capital buffer,
so this is basically how much capital buffer there is
correcting for volatility.
Pre-crisis is about a point three standard deviation
move of assets away from failure.
Post-crisis is varying but a lot, more than double,
possibly you could argue triple,
and the bozzle program,
not to mention other financial regulations,
have forced the banks to take in a lot more capital.
They should be a lot safer,
yet their credit spreads are a lot worse,
those two facts don't seem to go together,
what's the, how do you apply Occam's Razor?
Well one possible explanation is behavioral.
Perhaps the creditors of these banks
were lulled into a behavioral kind of belief
that these banks just couldn't fail
and they got shocked out of their belief
by the financial crisis and they're still in shock.
That's not the story that I would like to tell,
I'd like to tell you more rational story,
which is that the resolve of governments
to force the creditors to take loses
have taken a grip on the minds of creditors,
at least in the United States,
so that if you're a lending money wholesale
to the largest US banks,
you really believe, post Lehman,
that you may be forced to take a loss.
And in fact, there's a program,
part of regulation that suggests that you will take a loss
through what's called failure resolution.
Let me walk through that.
In the European union it's called a bank resolution
and recover directive, BRRD.
In the US it's title two of the Dodd-Frank Act
as implemented, is says,
that the holding company debt of these banks,
a large portion of it will be bailed in,
meaning those creditors will be told
you no longer have a debt claim,
you now have an equity claim.
We'll get back to that.
The main result is conditional on the bank
approaching insolvency,
creditors believe that they're much more likely
to take a loss,
that tax payers will no longer be on the hook
for as much as they were.
Okay, now, to believe this story you don't have to believe
that the regulators will indeed hit the button,
you only have to believe that the creditors believe
that they won't get bailed out.
You don't have to believe that if the regulators
hit the button for failure resolution
that it's actually going to work.
All you have to believe is that the creditors believe that
if they don't hit the button, they'll take a loss,
and if they do hit the button, they'll still take a loss.
And that's what we're gonna be talking about.
Also, reduced equity subsidies
and reduced subsidy induced leverage.
Now before I get too much further
let me go directly to step five
on the Dave Backus program,
and be extremely humble,
this is all preliminary,
we're getting the emails back and forth
from Canberra, Australia nightly
and the numbers are adjusting
and they may not be the same
by the time this paper gets submitted for publication.
And again we've only covered the largest US banks,
you still have to fold in the European banks
where the story may be different.
We can control for domicile as well.
Here's a preview of the kind of results
that you're gonna see in the next 20 minutes or so.
This is, moving across time,
what would be the five year credit spread
of one of these banks,
where it's capital buffer to be held constant across time,
measured in distance to default,
measured in the number,
that is in the number of standard deviations
by which asset values would have to fall
before these banks would be put to the point of insolvency.
So this bank, this hypothetical bank,
is of constant credit quality,
and you're looking at how it's credit spreads
adjust across time.
There's two reasons for adjustments,
one is market risk premia are changing.
Again that was one of David Backus'
big contributions to financial economics,
how market risk premia are determined
and how they change and what factors change them.
The other big change is post-crisis
has estimated in this model
creditors believed that they would no longer be bailed out
with as greater likelihood.
Now you've noticed I've just backed off a little bit,
I didn't say their no longer to big too fail,
I rather said that the likelihood that they'll get
bailed out has dropped by quite a bit,
and I'll show you the numbers.
What are the rating agencies say?
So let's compare the median in blue
of all publicly rated firms,
US publicly rated firms, according to Moody's.
So the blue is the median rating,
median Moody's rating,
of the senior and secured data publicly traded firms.
And in red the US G-SIBs median credit rating.
Notice that the median US credit rating for the G-SIBs,
the big banks, has come down a lot.
Is it because their capital buffers are reduced?
No, it's because the rating agencies fall,
they credit the governments story that they
won't be bailed out.
They explicitly include in their ratings
what are called sovereign uplifts,
these are extra notches
that they used to give to the big banks
in light of the likelihood that the big banks
would be bailed out if they run out of money.
Those notches are listed for each firm,
and you can look in the research provided with the ratings,
how many notches above lift were given to each bank.
They've all been removed post 2013,
and they were gradually removed
following the financial crisis.
So Lehman was enough of a shark
and then the government said explicitly,
you're not protected anymore.
Okay, a bit of a model,
is gonna be some math coming up shortly
for those of you who need to think about
your grocery list for dinner, or whatever,
you can get ready, that's coming up soon.
But just for a few minutes let me give you the idea
of bailout versus bail-in
versus allowed to go through bankruptcy in pictures,
and then we'll go into the math just briefly.
So this is a bank that's verging on insolvency,
it's assets have fallen to the level
at which the shareholders are willing to allow
the debt to go unpaid.
That's the condition of the bank in this picture.
It has bonds in blue, deposits in green,
its assets have fallen below the level
at which the firm is solvent,
and in fact,
a lot of big banks ran at probably negative solvency
but continue to operate until it just became not worth it
or they were unable to issue enough equity
to keep the banks alive,
and they had to be either bailed out or fail,
Lehman failed.
Here's what a bailout means in our model,
so when you see the numbers coming up later
you'll know what the model assumption is.
What does a bailout mean?
It means the government injected enough capital
into this bank to recapitalize it
to the point at which the unsecured holding company
that trades at par,
that's our definition, we had to pick a level,
as you know a model is
an abstraction of reality
that's our abstraction that re-capitalization
to par debt values.
So that's how the model will work
if there's a bailout.
There doesn't have to be bailout in this model,
we'll allow some probability called pi of a bailout,
and one minus pi of the alternative,
which is, it's allowed to fail.
Failure means that a lot of the assets are
eliminated by distress,
and now there's even less to pay the bonds.
For modeling purposes, the bank is liquidated,
the bonds take a big loss,
in our model the deposits are either protected by assets
or there is deposit insurance,
so the model includes deposit insurance.
We'll get a little bit more detail on that in a minute.
The probability pi of a bailout
is the number we're looking for,
and I know that there's probably at least one or two
non-finance experts and asset pricing experts in the room,
let me explain our limited ability to identify pi.
This is not the actual probability of a bailout,
it's what's called the risk-neutral probability.
It's what practitioners call
market-implied bailout probability.
So it's the bailout probability
that when you take expected discounted values,
you get market prices.
We can't identify the statistical probability
that a big bank gets bailed out
because we don't have enough data to do that.
Now there is an alternative to bankruptcy
and that's called bail-in,
and that's part of the BRRD
and Dodd-Frank implementation in regulation.
What does that mean?
It means that the creditors are told
you no longer have a bond claim,
you now have an equity claim instead.
So they will probably take a loss if that occurs.
But you're not going to cause all those distress costs
because you're gonna open for business the next morning,
theoretically.
Again, whether you believe this will work
or whether it'll be tried,
is somewhat secondary to what we're able to measure.
All we're gonna really be able to measure
is the likelihood that you'll get to this stage
of not getting bailed out
and we'll have a difficulty distinguishing
between bankruptcy and bail-in,
why?
In both cases the equity prices
would reflect the same outcome, zero.
In both cases the bond prices EXANTE
would reflect roughly the same outcome, a big loss.
So from looking at the market prices of equity and debt,
before we get to this stage,
you would have a very difficult time
identifying the difference between bankruptcy and bail-in.
What you could do, and what we haven't done yet,
is to get different debt instruments
that are differentially affected
by bail-in versus bankruptcy,
or bail-in versus bailout.
And those do exist,
there's AMERAL or TLAC designated debt
that's distinct from other forms of debt
that are market priced,
or you could do banks subsidiary debt
versus holding company debt,
so maybe with enough data I could come back to you
in a future opportunity, and distinguish these,
but I'm not gonna be able to today.
Today I'm only gonna talk about the likelihood
of a bailout.
So that's a limitation of what we're doing.
There is some work on bail-in,
and I've slided it in in the bottom of this slide.
But in the interest of time, I'm gonna pass over that.,
So here is the mathematical model
that's inspired by the work of Hayne Leland
that was done in 1994,
which in turn was inspired by some work
by Fisher, Hinkle and Secknor.
Hayne Leland solved this model
for when a firm defaults explicitly,
and I'm gonna rely on those results.
So this hypothetical firm has,
and this is a big bank,
has assets in place which are
a sarcastic process called the geometric Brownian motion,
which means the change in assets looks like
a risk-free rate minus the payout rate K on assets
times the current assets in place,
in expectation that's the change in assets.
And then there's a volatility term, sigma,
multiplied by the level of assets,
multiplied by the change in a IID process
which for today will be a Brownian motion.
You can substitute with a Brownian motion,
a process that has downward jumps,
and everything that I tell you can be redone.
But today it's gonna be with a Brownian motion.
So that's the fluctuation of assets in place,
it's a very typical risk model for assets in place
for a large firm.
The liability structure includes risk free deposits
of some offering some interest rate big R,
you notice big R,
the interest rate on deposits doesn't need to be
the same as the one as the market risk-free rate.
In case you haven't noticed,
the money you're depositing in the large banks
is probably not paying you a market interest rate,
it's probably paying you a rate that's far below that,
and that's another story for another day.
So we're not requiring that.
And one of the reasons the deposits are risk free
is that they're insured.
For modeling purposes, because it makes the model easier,
we'll take the outstanding amount of debt in face value
to be a constant number, P, or principal of debt,
I see I misspelled principal,
with a coupon rate, c.
And because we wanted that maturity structure
we'll take Leland's trick of having
an exponentially decaying maturity structure
so that the average maturity of the outstanding debt
is one over m, where m is the exponentially decaying rate,
or equivalently, either the minus mt
is the amount of debt outstanding
with maturity at least t.
As bonds mature, the bank is issuing the same face value
of debt in the market at the then market price.
And Leland was extremely adapt at solving these models,
he's already given us an approach to solving this,
we extended to the case of bailout and to adding deposits,
but really it's based on the work that Hayne Leland did
back in the early 1990's.
Okay, so we're not gonna go through all the guts of that.
One calculation that's gonna be coming up shortly however
is the displayed quality in the middle of the screen,
which is the current market value of the bailout subsidy
offered by the government to this particular big bank.
The first factor in this expression
is pi the likelihood of a bailout,
that's obviously gonna be of a constant
proportionality of the market value.
The second factor, which is that parenthetic expression
raised to the power of minus gamma,
is the present value of receiving one dollar
when the bank becomes insolvent.
That's based on the Laplace transform
of the first hitting time of a Brownian motion
on a given barrier.
Here the barrier is called v star,
it's the level of assets at which the equity owners
will give up the bank because they're not willing
to issue any more equity to pay down the debt.
Market value of equity is dropped to zero.
So the first two terms are just the expected present value
per dollar of exposure to the government.
How much is the exposure to the government?
Let's see if this pointer works.
No, okay.
So, V zero is the amount of assets the bank needs to have
in order to reprice the debt to par.
Subtract V star, which is the amount of assets
at the point of insolvency,
that's the size of the government injection of capital.
Now when the government injects all that capital
it gets something in return,
it gets the market value of the equity in the bank,
'cause it's nationalized the bank,
so the government now owns the bank.
And that's happened in a few cases.
We could have partial nationalization,
for modeling purposes we'll just assume
that the government is the sole
new owner of the bank.
Nobody else, by the way,
in the private market would participate
because this is, by debt overhang,
a negative NPV arrangement for the government,
and anybody else that wants to come in at the same price.
The government's free to sell the equity
immediately after bailout,
but that's the loss that it's gonna take,
what it paid, V zero minus V star
minus what it received, the market value of equity,
a negative number, as far as the government is concerned.
Positive number in terms of the size of bailout.
So that's the present value of bailout.
In a little while, like five or 10 minutes,
you'll see how big that number is
as a fraction of the equity in these banks.
And what I don't have on the slide I'll tell you in dollars
because this is being recorded but I'm a little less weary
of saying a number orally than I would have in putting
it on the slides.
For the five top US banks,
the present value of the equities they were receiving
has dropped from pre-crisis to now,
to the post-crisis period,
by somewhere between
two and three hundred billion US dollars
in present value terms.
That's not the annual subsidy,
that's the present value that the model indicates
of subsidy value that has dropped
because the government has said
we're no longer gonna bail you out,
you top five US banks.
And more importantly, creditors believed that story
to at least a certain extent.
That coefficient gamma is not for today.
Okay, here's how were gonna identify that key number,
the probability of bailout.
And by the way, the assumed probability of bailout
will also affect that optimal default boundary, V star.
So they have to be co-determined.
And we'll get to that.
Here's the panel it regression approach
that we're gonna take to identify that number.
We have all the credit default swap rates
on all actively traded firms
in the credit default swap market.
That's around 800 firms,
all the transactions or quotes available,
covering around 1.6 million observations.
That's a bit overkill because you have
multiple observations per firm, per day.
But lots of transactions,
essentially all the available transactions
in the credit default swap market.
And we're gonna try to suss out from the capital structures
of these firms and from the credit default swap rates
that markets are charging for protecting their debt,
we're gonna try to suss out how these bailout probabilities
are being set by creditors for the largest banks.
So we're gonna identify this bailout probability
for each big bank on...
It's marked by time,
but I'm only gonna try to do it once for pre-crisis,
once for post-crisis.
So we're gonna try to identify the bailout probability
for big banks pre-crisis, post-crisis.
There are still some problems to get over,
which we'll get to.
Okay, when we assume a given bailout probability,
as I said, the assumption about,
or the calculation of when the firm
is actually gonna default changes.
The optimal time to default,
or the point in time at which the market value
of equity drops to zero and the firms is insolvent,
depends on the bailout probability.
The distance to default,
which is the number of standard deviations,
depends on your assumption about the bailout probability.
That's gonna be our key explanatory variable,
distance to default for credit pricing.
We're gonna assume that creditors looked to
that solvency buffer
when deciding how much to charge for credit.
And then they'll add on a risk premium,
which could differ for big banks,
and could change over time,
and from that information we'll be able to identify.
Here is the panel regression.
Sorry this is gonna take, I promise, only one bit of math,
it's gonna take another little explanation here.
The left hand side is what we're observing
for the credit default swap rate,
that's CDS in logs for each firm, I, on each day, t,
including all the publicly traded firms.
You noticed that I divided it by the no bailout probability
one minus pi.
Why did I do that?
I want an apples to apples comparison
between non-big banks and big banks.
So if you're a grocery firm
or a consumer cosmetics products firm,
you're not gonna get bailed out,
and we look at your credit default swap rate in the market
and we compare it to that for a big bank
which could get bailed out,
we have to re-normalize by the likelihood
that you don't get bailed out.
Otherwise, the two credit default swap rates
are not commensury for the risk that you face.
Getting bailed out means you don't lose any money,
so it's really, if you're a big bank,
if you're lending money to a big bank,
your only concerned about the likelihood
that you're gonna not lose money.
So that's the variable to be explained,
it's gonna depend on a number of explanatory variables.
First, a constant alpha,
'cause you always put a constant in your regressions.
Second, distance to default,
the r squared when you use distance to default
is on the order of 40 percent-ish.
And as I said,
everything I tell is gonna change as the paper evolves.
So we're explaining about 40 to 45%
before making other changes.
The next factor
is a coefficient gamma times
an indicator of whether you're a big bank or not.
And this is constant across time
and simply reflects our appreciation of the fact
that the creditors to the big banks
could be different creditors
and they could perceive a different kind of systematic risk
for big banks than for other firms,
so we allow them to have an add on for big banks
you wanna correct for that.
The next factor are time fixed effects
because in some prior research we showed
there's a significant variation over time
and risk premia for baring
any kind of pubic firm default risk,
so we allow for time fixed effects,
capture the variation across time in risk premium.
Then the last control is whether or not
this is a big bank after the crisis,
because I've suggested that things
are different after the crisis.
Now there'll be a fraction of you
that will be ready with the question,
why do you need to control for the change in bailout
after the crisis if you've already controlled
for it through the bailout probability,
you don't need to control for it twice?
You're exactly right.
And if we'd gotten the right bailout probabilities
then this coefficient V should be zero.
And that's exactly how we identify
the bailout probabilities.
We search for those bailout probabilities figuratively,
with the property that when you put them in here and here,
I see I've mixed p and pi,
when you put them in here and here
this coefficient turns out to be zero.
So that's how we're gonna identify
these bailout probabilities.
We also include crisis fixed effects,
DSIB, which stands for
domestically systemically important bank effects,
and we can give you the numbers for those as well.
Sectoral fixed effects among public firms,
and we have some other controls like
we can control for investment grade,
non-investment grade, things like that.
(audience member speaks faintly)
Yes pi is assumed to be zero,
and there was an exception,
we know that GM was bailed out during the crisis,
sorry, that's an example where our method
doesn't take account of the fact that the government
might bailout other firms.
Doesn't happen as often in the US as in other countries,
so far we're only looking at US data,
maybe a better model would allow for that.
Okay.
We're only gonna allow pi for big banks
as Tamar just suggested
and we're only gonna allow two levels,
'cause we want a good identification.
So one for pre-crisis, one for post-crisis.
Because we have so many control firms
we can do this also bank by bank.
Today, again because I'm very hesitate,
and I'm not to step five in the Backus program yet,
I'm not gonna show you the numbers for the other banks,
but I'll tell you orally,
the effect is much bigger for the two large US banks
that pre-crisis were investment banks.
Why should that be?
Well because they had much greater leverage
and they were also, until we found out otherwise,
protected from being too big to fail,
and our numbers bear out,
that there's a much bigger drop in bailout probability
for those two firms than for the three, five in our setting,
money center big banks.
But I'm not gonna tell you the numbers yet,
'cause I'm not confident enough to tell you the numbers
other than it's quite a lot different for those two firms.
For now we'll just force this number to be the same
for all the largest US banks.
As I mentioned,
I've already controlled for big bank versus non-big bank,
both through one fixed affect that applies at all time,
and also through this bailout probability,
those are the only two ways,
if you suggest other reasons
that big banks should be different,
than you can take your shoe with this later,
but it's been absorbed by those two things.
So again we're gonna, I just said this,
we're gonna search for those
two pre and post-crisis bailout probabilities,
such that, when we go back and re-run that regression
we get a zero coefficient on the post-crisis big bank effect
other than through the bailout probability.
So that's how we get identification.
Now, okay, so let me back up a little bit
before I get to the least salutary aspect of this research,
which is the average bailout probability
is kind of the same idea
as the effective average recovery of assets.
In other words,
not getting bailed out, it's a loss,
having a low recovery of assets, it's a loss,
it's impossible for us to identify
the difference between the expected loss
associated with the stress on average,
and the expected loss associated
with no bailout on average.
All we can do is difference across the crisis,
pre-crisis versus post-crisis.
So between these two probabilities,
pre-crisis bailout and post-crisis bailout,
we can only identify a one dimensional restriction
on those two numbers.
What we're gonna do is we're gonna fix
any number you like for the post-crisis bailout probability
and then we'll tell you
the pre-crisis implied bailout probability.
Again, with apologies, here's an example,
if you decide subjectively
that the post-crisis bailout probability
for these large firms is 20%, not an unreasonable number,
but it's anybody's guess 'cause we don't have
any examples post-crisis.
But if you were to decide 20%,
all the data that I've just told you about
would imply a 65% likelihood of a pre-crisis bailout,
so substantially larger.
And again, with the caveat,
that the number is gonna change as we improve the model.
But a large number relative to
the post-crisis bailout probability.
If you were to say, no,
I'd buy whole heartedly into the story that the government
will not put a penny into these banks,
and the post-crisis bailout probability is zero,
then the implied pre-crisis bailout probability is 55%,
so there's a bit of a nonlinearity in here.
It's gone down by a lot either way,
but the difference, the change,
depends on what you're willing to assume
about one of those two bailout probabilities
relative to the other.
Soon we're getting to the point
where you can grill me with questions,
so get your questions ready,
we've got about five more minutes of me
and then the rest for you.
Okay this is the same chart that I showed you before,
but now I'm confessing that I fixed the assumed post-crisis
bailout probability at 20%
and everything else is implied by the data.
And that's what you'd get for a hypothetical big bank
whose solvency is always the same,
credit quality is always the same,
the market is assigning different credit spreads
for five year debt,
at the holding company level,
this is all based on holding company debt, unsecured.
Much lower pre-crisis than post-crisis on average,
mostly because of the reduction in bailout probability.
That's what the numbers are telling us.
Now, there's also an implied effect on leverage.
If you subsidize the debt of these firms,
what would you do?
You don't even need to think about moral hazard,
all you have to do is say,
oh the market is letting me buy assets
with really cheap funding,
so I'm gonna buy a lot a assets,
and that's exactly what the largest US banks
were doing pre-crisis.
Zoom, zoom, zoom, big increases in assets.
Post-crisis, not so much, for a number of reasons.
Capital requirements also played a significant role,
but funding costs, as I've emphasized in some other work,
are critical to the capital structures of these firms.
If you raise their debt funding costs,
relative to risk-free rates,
they won't borrow as much money.
So pre-crisis,
the too big to fail moral hazard
was really an incentive to buy a lot of assets
funded with debt.
That incentive has been reduced, substantially.
These are just, these are not model numbers,
these are just straight out of the 10 k's of these firms.
And this is J.P. Morgan, Bank of America, Citi,
Well's Fargo, Goldman Sachs, Morgan Stanley,
Lehman Brothers and Bear Sterns added up.
What about the market value of equity?
First, just what do the accounting
and market value's tell us?
And then we'll look at what the model tells us.
The accounting numbers tell us that pre-crisis
these firms were trading
at market to book multiples above two.
Both the investments banks, or so called dealers,
Goldman Sachs, Morgan Stanley, Lehman,
Bear Sterns and Merrill Lynch,
and the money center banks.
Post-crisis, the market to book ratio has dropped a lot.
Why is that?
Well they're less leverage,
that lowers the option value of default, that's some of it,
their franchise values are possibly reduced somewhat,
a lot of it, we will claim,
is due to the fact that you've lost your subsidy,
you can no longer get subsidized debt.
That subsidized debt is worth a lot.
So the next slide will show you our estimates
of the model implied equity subsidies relative to
the total market value of equities of these firms.
And here are the numbers, they're noisy,
but pre-crisis, the model implied subsidies
were on the order of 75% of the market value
of equity of these firms.
You might be, at least I was,
a little surprised at the sizes of these numbers,
they're pretty big.
Post-crisis, they're still there,
'cause we're again,
we're assuming a point two probability of a bailout,
that turns out to be still substantial,
but much reduced.
And I mentioned for the five largest US banks
the order of magnitude of the reduction
in the dollar value of these subsidies
from pre-crisis to post-crisis on average,
a substantial reduction.
Okay, last slide.
We're not the first to address this problem,
I wanna mention a few of the papers
that have addressed this issue
of how much the bailout subsidy has dropped
or how much has bail-in had an effect.
So let me just mention a few of the papers,
there's a 2016 paper by your colleague Viral Acharya
and Anginer and Warburton,
but they only looked at the small time window,
120 days around the Dodd-Frank Act.
And focusing on that time window,
there was an early paper,
didn't give them the advantage
of having this long time series
and they went at it a different way,
and they found no significant effect
of the Dodd-Frank Act as an event
on the implied reduction in subsidies
or increase in credit defaults,
they were looking at credit defaults swap rates as well.
Sorry, bond yield spreads, the same idea.
There's a very interesting paper
by Neuberg, Glasserman, Kay and Rajan,
which is mostly a descriptive paper
that looks at credit default swap rates
in Europe, not in the US.
Because in Europe there was a change in the
triggering mechanism for a credit default swap in 2014.
The new contract said,
as a buyer of protection in the CDS market,
you are now protected for bail-in.
The old contract, which was a 2003, ISDER standards,
said you're not protected for a bail-in.
Well this is kind of like what
every financial economist loves.
You got difference between the two credit default swap rates
tells you the market premium for bearing specifically
the risk for a bail-in.
And interestingly they showed some credibility
from 2014 to 2016,
pardon me two thousand and, yes, 14 to 16,
some credibility for this bail-in assertion
that the government would bail-in.
But when the Italian bank Monte Dei Paschi of Sienna
was not bailed in
or was actually given government capital and bailed out,
the market confidence implied
by the credit default swap market of bail-in reversed,
that is, the credibility of bail-in dropped,
so that event had a big effect on,
at least according to this study,
a big effect on what creditors believed.
There's also a very recent paper
by Atkeson, d'Avernas, Eisfeldt and Weill
which goes at what we're doing from the viewpoint
of estimating the market value of the equity subsidy
of the too big to fail,
by building a composite, representative US bank,
not just big bank, all US banks,
mushing together all the accounting data for this firm
and then using a Gordon dividend-discounting model
and a simple Markoff chain analysis of this firm,
an estimating the fraction of equity value
that's associated with bail-out subsidy that's disappeared.
And that fraction is 23%.
It's a smaller number than I was showing you a moment ago,
which was more like 50%
reduction in market value associated with subsidies.
But remember this is a composite US bank,
8,000 different banks,
so you kind of expected it to be a lower number,
completely different approach to this problem.
Doesn't use credit default swap data and market pricing,
uses accounting data.
So that's the end of what I prepared to tell you
and now I'm gonna shield myself from your interrogation.
So please, anybody that has a question,
Laura you can kick it off.
- Alright, so I'd like to push back
on what you called the behavioral alternative explanation.
I now argue it could be perfectly rational
to be complacent pre-crisis and still shocked post-crisis.
Why?
Because agents estimate distributions,
nobody knows what the true distribution
of returns for a bank is,
and pre-crisis we'd never seen a bank
on the verge of default.
So it would be perfectly rational,
I'd be wrong, but a good econometrician can be wrong, right?
So it could be perfectly rational to estimate
a very low probability of a bankruptcy
that you've never seen and then post-crisis,
having seen big banks on the verge of collapse,
estimate a much higher probability
and once I've seen it, it's in my data set,
so I continue to estimate big crisis levels,
so this behavioral, or econometrics explanation,
may be true for a lot of firms,
and in fact we see in equity that the skew index,
a measure of the left tail in equities went up
in the crisis and never came back down,
and it could be particularly true for banks
and explain your facts if we simply believe
that maybe banks load more heavily on this kind of
tail risk than do other firms.
So in light of that, might you be overestimating?
I don't doubt that some of what you say is going on,
but maybe some of that effect could be that fact
that we learned something in the crisis
and that's affecting our perception of this.
- Yep, no that's quite possible.
I remember how shocked I was when Lehman was not bailed out.
I remember that weekend, quite well,
I'm sure it's seared into the memories
of many people in this room.
Wow, they really didn't bail it out.
Maybe they're not going to do that anymore.
That's what we're trying to say here.
That's what this is about,
it's not necessarily about bail-in, or legislation,
it's about a change in the belief by creditors
that this bank would not be given government capital.
Now, you could say well maybe they never,
they updated, so they did exactly what you just said.
They had a rational prior,
which was based on no bank ever having been bailed out,
pardon me, no bank ever having failed,
having no bank ever not having been bailed out,
then suddenly, one of them was allowed to fail,
that conditional probability
that a bank would get bailed out just dropped.
That's what we're trying to say.
- I guess I was claiming there were two things we learned.
One is, you might not get bailed out, as you said,
but the other is we also didn't know they would,
a big bank would ever get that close to being,
to needing a bailout,
because we hadn't seen that risk before.
- Ahhh, okay.
We can use this, how our model identifies this,
equity prices and the volatility of equity prices
to infer from the data
what that likelihood of reaching the point of insolvency is.
That's one of our model identifying approaches.
- [Audience Member] Put the skew in this.
- So at least as far as getting to the point of insolvency,
if you go with geometric Brownian model,
or if you wanna change it to a jump model or whatever,
you do have the ability to estimate how likely it is
that you will approach insolvency,
day by day, from the data.
So in theory,
you can calculate the probability of insolvency now,
and the model is supposed to deal with that.
Okay.
Mervin go ahead.
- Related to that question, but slight different.
I think you equate the phrase bailouts
with a government injection of capital.
- [Darrell Duffie] Correct.
- Now in your stylized models,
that flows naturally from the set up,
you've got long term debt, deposits,
which are protected by deposit insurance.
In the actual crisis, there were of course,
an awful lot of very short term finance provided wholesale
which were not protected by deposit insurance.
One of the aspects of government intervention
was to replace that finance by lending,
and the phrase bailout, in practice,
I think referred as much to that
as to government injections and capital.
Are your results capturing the value
of both of these types of intervention,
and that the stylized model only refers to one?
But the actual numbers you get--
- [Darrel Duffie] Okay.
- [Audience Member] Can you comment on that?
- Yeah absolutely.
So what Doctor King has just suggested
is that there are liquidity issues
as well as solvency issues facing this bank.
Some, although not myself,
would use the work bailout to describe let's say,
a central bank providing liquidity against collateral,
and assisting these banks from avoiding failure
due to a liquidity crisis.
So that's not what we're talking about here.
We're talking about the fiscal authority
actually injecting new capital into the bank.
How we're a model have to be interpreted
to separate the two effects.
If you take it as its face value
would have to be the central bank,
or whatever the provider of liquidity is,
is completely unfettered in preventing
a liquidity-based failure
and will and did, let's say case by case,
step in as needed.
And the only question was, where's the capital coming from?
And that would be the fiscal bailout
that we're talking about,
that's how we would have to interpret the model.
Now whether the central bank would in fact step in
and provide liquidity or under what circumstances,
and there's a gray area,
was just heavily discussed by Biernacki and Polson
over the last few weeks,
they've been all over this issue.
Was it liquidity?
Was it solvency?
Did they have the necessary tools
to use central bank liquidity as opposed to a bailout
or a let it fail?
There's a gray area there.
It's a great question.
Yes Tamar.
- So I was trying to think about
whether you can think about another problem,
(speaking faintly)
and on one side of the bias,
I think that's what Laura was saying,
which is in the data, two things occur at the same time,
some learning and some about the true risk,
and some learning about the right bailout condition--
- [Darrell Duffie] Right.
- So it seems to me that the,
(speaking faintly),
is whether the equity volatility,
(speaking faintly)
is that enough to capture that?
I guess what I'm thinking about what Laura was saying
was in a jump based model we also need to have that data.
(speaking faintly)
Can we get a sense of learning the truth,
versus the bailout on one hand,
and on the other side for the post-crisis,
(speaking faintly)
it's also about the recovery of assets.
- [Darrell Duffie] Yeah.
- But now, some of it we've done post-crisis
also improve what would happen in case of a bankruptcy.
So if these are successful,
then that means that
if there was to be a rebound today,
hopefully the recall rate would be higher
- [Darrell Duffie] Higher, right.
- So that buyers estimate--
- [Darrell Duffie] That makes my results conservative.
- Exactly, so that's what I'm saying.
So Laura's point back then,
(speaking faintly)
- Yep, so, okay.
So the second part of your question that we just addressed
which is, if recovery's, if the liquidity program, LCR
and all of the other TLAC and so on,
make recoveries higher then presumably the...
We're understating the reduction in bailout probabilities.
What about, and going back to Laura's question,
what about the robustness of the model
to our ability to estimate the likelihood that
you would reach the insolvency?
So we haven't, again early days,
we haven't tested the model
against other models of asset risk.
If you take the model literally
and you put in jumps,
you would get another estimate that would be again
subjected to the criticism that we might be confusing
the learning about the likelihood of hitting insolvency
from market pricing data
with the change in bailout probability.
And what we could do is do a range of different model types
to try to convince you that most of the effect
is coming from a change of bailout probability.
Again,
our ability to get the average bailout probability
across both pre and post-crisis is very limited.
So if you increase asset risk,
that increases the likelihood of insolvency
both pre and post-crisis,
and we're again unable to distinguish that effect
versus the average bailout probability.
All we can tell you is,
if you assumed the technology of the bank
hasn't changed a lot,
then the difference between
the pre and post-crisis bailout probability is large number.
Marty go ahead.
- [Audience Member] Taking account the change
in the definition of,
(speaking faintly)
because that's clearly,
(speaking faintly)
So I think that means that the numbers you have
post 2009 will be lower,
you'd expect it to be lower than the ones before.
I mean obviously you could have some market
that needs restructuring,
that's one part of the whole scenario.
- That's a good question, so let me address that.
As Marty has suggested, the contractual definition for CDS
of when you get paid has changed based on whether or not,
whatever firm it is, is getting restructured.
So that has been a change over, migration over time,
in the contractual language.
So we can account for that with these time fixed effects
on average across all firms.
And now the remaining criticism might be,
but for the big banks,
the likelihood of getting restructured
without going through an insolvency procedure has changed,
my guess is no.
Because for any of these big banks,
restructuring causes cross-acceleration,
it causes immediate termination of the swap book
and the repo book,
so any kind of materially adverse restructuring
that would trigger a credit default swap
would instantly trigger the failure
of one of these firms.
Other than the bail-in, other than the bail-in--
- In context of your model,
you had distinction between defaulting from other causes
rather than restructuring.
I mean this is something that's really hard to figure out.
- Yeah, so that's,
the changes again,
changes in contractual definitions across time
are accommodated in our model through time fixed effects
which absorb also changes in risk premia.
We can't separate, we don't try to separate all of that,
it's all absorbed into the time fixed effects.
But the restructuring point would make a big difference
in our results if you could claim
that there was a big change in the likelihood
that a large bank would get restructured without failing.
And I would say for these banks,
it's pretty unlikely that they could go through
a restructuring without failing.
- I'm curious Darrell about your thoughts
about doing this cross country.
So in particular you talked about Europe
and the prior to the common deposit insurance scheme,
the fiscal capacity of different countries,
arguably differed a lot in terms of their ability
to bailout their banking system.
So presumably if you did this country by country,
and you could extract very different probabilities
of these bailouts,
and maybe also after European deposit insurance scheme,
those probabilities will change again.
- That's a great question.
And it's gonna be probably a dilemma
once we pull in the European data
because we have Switzerland which has two G-SIBS,
we have the United Kingdom which,
although it was part of BRRD
or still is for another few months,
might have more resolve than some European countries.
So I could name one or two European countries
that have very large banks which I would predict
would bail them out rather than let them fail,
despite the resolve of the SSM
and all of the other machinery
that Europe is developed around this.
Let me also mention China,
we're not gonna get data on China,
but of the largest banks in the world,
the four largest, the first largest bank,
the second largest bank, the third largest bank,
and the fourth largest bank in the world,
are all China banks.
And they have the backing of their government
almost without question.
And in part because they really are too big to fail.
You couldn't possibly, if you were a government,
and you allowed it to get to the point
where they were insolvent,
you couldn't possibly stand aside.
And in any case,
the China's government would not stand aside.
So you're absolutely right.
It's gonna be country by country.
And I'm really interested to see
what the data for Europe say,
but we're not there yet.
There's a question way up there.
- Hi Darrell, I was curious,
the subsidy, are you talking about the subsidy from size,
or is this a subsidy from systemics in general?
- Okay it's, if you take the model literally,
it's what the creditors believe is the present value
added to their debt claims associated with
government injection of capital when the firm becomes
on the boundary of insolvency.
That's all the model can tell us.
And we're only applying the fixed effect
for the nine largest US G-SIBs in this study so far.
We can do it bank by bank,
we can do all nine banks together as I did today.
We also have done it for D-SIBs,
not on the slides, but let me tell you,
there is also a substantial reduction in bailout probability
for the domestically systemically important banks,
but it's about half as big.
Roughly.
I think I may have exhausted both the amount of time
and the questions.
Mervin you have one more.
- [Audience Member] Well first can I thank you
from admirably clear expositions.
- [Darrell Duffie] Thank you.
(audience applauds)
- [Audience Member] Very Duffie-esque if I may say so.
(audience laughing)
Quickly, I'm wanna get you to speculate about the future,
because you've been comparing
before the crisis and afterwards.
And do you think, this is not to do with results,
it's you introspecting on this,
and how you interpret it.
Could you imagine that as we go forward
over the next 20 years,
that the probabilities would change in such a way,
correlated to Laura's question,
but complacency might step in again,
and that the value of the subsidy,
despite the change in machinery,
could build up again.
Would you believe that somehow
this is a discrete change that's permanent?
It's a question of your judgment and speculation.
- I'm probably not the best judge
of human behavior in the room,
but I think this has a lot to do with human behavior
and how long memories persist,
as I said, most of us have a searing memory
of the financial crisis,
but the median financial worker these days,
wasn't in the financial work place
at the time of the crisis,
and government legislatures come and go,
and the temptation to relax on capital requirements
and relax on the resolve to make these firms fallible
can weaken over time.
So if you had to guess on a trend,
I would guess the trend is towards gradual increase
in creditors assumption of bailout as time moves on.
I'd be interested in hearing your comment actually.
What do you think?
- I think we should hear from Stanford.
(audience laughing)
- So Darrell, thank you very much.
Kim has asked me to give you a couple of symbolic gifts.
They're not symbolic of our appreciation
for your great talk,
I'm startin' to realize they're symbolic of
deep out of the money options that are in your paper.
Like a guy who lives in Palo Alto
we're giving you an umbrella.
(audience laughing)
- We have the rainy season.
- And for a guy who's an expert
on financial markets and regulation,
we're giving you a book written by some of our colleagues
here at Stern on how to regulate Wall Street.
So again, another out of the money option just in case.
- I read all your stuff.
- So again, please join me in thanking Darrell
for a great talk.
(audience applauds)
And I'd like to invite you all to head down
to the first floor of this building
on the west side, Gardner Commons,
it's got a number?
- [Audience Member] M1 100.
- M1 100, and for refreshments.
- [Audience Member] Downstairs.
- Yes downstairs to the first floor, that way.
Okay, thank you very much.
♪♪
For more infomation >> Opel Corsa 1.4 S&S 90pk 5d Color Edition - Duration: 0:53. 
For more infomation >> Verdeliss defiende a Miriam Saavedra en 'GH VIP 6': "Son injustos los comportamientos contra ella" - Duration: 3:27. 
For more infomation >> Opel Corsa 1.4 90pk 5d Cosmo | Climate Control | - Duration: 1:09. 
For more infomation >> Flake: I wouldn't be troubled if FBI didn't talk to Ford - Duration: 2:04. 
For more infomation >> Grande Fratello vip, Lory Del Santo il ricordo straziante del figlio: 'Non usciva mai… - Duration: 2:36. 
For more infomation >> Uomini e Donne, puntata di oggi 3 ottobre: 'Sei un maschilista' | Wind Zuiden - Duration: 2:23. 
For more infomation >> WWE Want BULLET CLUB! Neville To New Japan?! | WrestleTalk News Oct. 2018 - Duration: 4:57. 
For more infomation >> "Same Day Lay of the Year 2018" - Full Infield to Bedroom - Duration: 1:28:55. 
For more infomation >> KAT-TUN - CAST Live IN YOKOHAMA Fanreport (14-15 Sept 2018) - Duration: 18:25. 
No comments:
Post a Comment