Subscribe
Share
Share
Embed
Market Pulse is a monthly podcast by Equifax, in partnership with Moody’s Analytics. Equifax hosts bring you interviews with industry experts on the latest economic and credit insights that can help drive better business decisions. Whether you’re in financial, mortgage, auto or another service industry, we help make sense of the latest economic conditions that impact you. This podcast series supplements our Market Pulse webinars, which occur on the first Thursday of each month.
.
Welcome to the Market
Pulse podcast from Equifax,
where we break down the latest economic
and credit insights to help you navigate
today's business landscape.
Hello everybody,
and welcome to the special edition
of the Equifax Market Pulse podcast.
I'm Wendy Hannah Olson,
senior Vice President,
Digital Alliances and Strategy Execution,
and I'm delighted to be
hosting today's conversation.
We have a very bright mind
joining us. Joni Baker,
a mathematician and senior researcher
on the behavioral modeling team of
Andrew Davidson and Company. Welcome Joni.
Hi. Thank you for having me.
Absolutely. Thanks for being here.
Today we're going to be diving into a
topic that sits the very heart of the
American Dream, the mortgage process,
and the role of credit reporting.
Since the late 1990s,
the Trier credit report,
which pulls data from all three nationwide
consumer credit reporting agencies,
including Equifax,
briefly known as b
Crass or Credit Bureaus,
has been the gold standard for determining
who gets what loan at what price.
But the ground is shifting.
There has been much heated
debate in the industry.
Should we move away from the Trier
standard in favor of a single or bar me
credit report? A new white paper from
Andrew Davidson Company called The
Impact of Moving Away from the Tri-merge
Standard suggests that this move might
come with hidden
multi-billion dollar cost.
So let's dig in. So Johnny, I'm
going to start with the foundation.
What was the study that you performed?
Okay. Well
basically our study examined the
credit score differences across
the three NCRAs bureaus,
Equifax, TransUnion,
and Experian. And again, I'm really
excited to be here talking about it.
I think this was the first
time that a study like this was
able to be performed on an
unbiased uncensored data dataset.
And so I'm excited to share
the analysis that we did.
So first we examined the raw
underlying score differences between
any two individual NCRA credit
scores for the same consumer.
So in other words,
what does the collection of those
differences look like across all possible
credit score pairs and all consumers?
So we're just looking at
those raw differences.
And then we considered various metrics
that result from choosing or combining
consumers scores in different ways
representing either a dual or a single
credit reporting standard.
And we compared that against the current
tri-merge standard for GSE mortgages,
which is to take the median or the
middle value of all three scores.
And we did this across different
assumptions such as the scores are chosen
randomly, or the outlier scores
chosen or the maximum scores chosen,
and we also subdivided the consumers in
different ways to see if these metrics
varied for different populations.
So I,
I kind of like to put this study into
context and explain why studying the
differences between the bureaus
is important. Absolutely.
And describe a little more how credit
scores are used for mortgage qualification
and pricing. I.
Welcome that. Thanks.
Okay, great. I just think it
might help people sort of,
especially with some of the
things that I suspect, you know,
I might want to talk about later on. Yes.
but so since 2008 the gse,
Fannie and Freddie have used a loan
level pricing adjustment table or LLPA
table to determine the risk-based
fees when a mortgage is
originated.
And the fees vary for
credit scores from six 20,
which was the traditional
cutoff up through 7 79 in 20
point increments.
And the fees are also based on the
loan to value for the mortgage.
And if that loan to
value is greater than 80,
then the required mortgage
insurance or PMI also varies.
The cost of that insurance also
varies based on credit score and LTV.
And meanwhile,
six 20 has been the traditional
hard cutoff for qualifying for A GSE
mortgage.
And while they have very recently moved
away from a hard cutoff for consumers
in that general credit score
range, the credit score of course,
will be a major deciding factor as
to whether they qualify for the loan.
So the question is, what credit
score is used? And as we've said,
it's been the tri-merge
standard requirement,
pull all three credit reports
from all three bureaus,
and then for the consumer's representative
credit score use the median of those
three scores.
So the F-H-A-F-H-F-A has
previously considered relaxing this
requirement to a buy merge.
And under that the lender may
poll just two reports and use the
average of the two scores.
More recently it's been suggested that
the GSEs could even move to a single
report standard,
in which case a consumer's
scores based on just one of three
sets of NCRA data.
So basically to understand the
ramifications of such a move,
it's important to study the score
differences between the bureaus and that's
what we did.
Absolutely.
And I know right at the beginning you
mentioned a little bit about which data
was used, but I'd like to, for
the audience reiterate that.
So can you explain exactly what
data was used in your study?
Yeah, sure.
We had an anonymized
sample of consumers for a
snapshot from October,
2023 that included for each consumer,
each of the three individual
credit scores based on a, each,
each of those scores based on
a single set of bureau data.
And I should say that for that,
for the credit score model
we used VantageScore four.
Because this model is the
same across all three bureaus,
there aren't any inconsistencies
between the scoring formulas.
So what that means is that
differences between the CRA's scores
generally reflect differences
in the underlying data.
We also had a flag indicating
whether a consumer was
low to moderate income,
meaning that the household income
was below 80% of the median family
income in that metro
area. And when available,
we had another field indicating
race or ethnicity information.
And I should note that this information
is not in a consumer's credit file,
but was attached anonymously
from a third party.
And it's also important again
to note like these weren't just
consumers already holding GSE
backed mortgages or even having a
mortgage at all. In fact,
we had all consumers below 700 that
had credit scores at all three bureaus.
And we had a random 10% sample
for consumers above 700.
So each sampled consumer
above 700 represented 10
consumers. And accounting
for that rescaling.
Of those above 700,
there were almost 250 million
consumers represented in this study
dataset study.
That is a lot of data, two 50
million. Thank you for that.
So can you help us understand how much
difference can scores be between the
three nras since you
had VantageScore four,
which we know leverages the full
breadth and depth of the three
CRAs plus the trended data?
So it's additional alternative
information within the credit reports.
How, how different can the scores
actually be between the three NCRAs?
Yeah, so that was the very
first thing that we looked at.
What are the raw score differences
between any two given scores for a
consumer when those scores are
based on data from a single bureau.
So just thinking like if you were to
choose two reports randomly, you know,
how different are those two scores?
And what we found was that the two
scores were actually identical about
42% of the time. So that
maybe sounds pretty good.
However, 27% of the time, the
two scores by at least 10 points,
14% of the time,
they differed by at least 20 points
and 9% of the time they differed by
at least 30 points.
So that's almost one out of 10 pairs
where you're seeing at least a 30 point
difference and then moving
up from there 4% of the time,
the two scores differed by at least
50 points. That's a lot .
Yeah. So in mathematical terms,
we observed that the distribution of
the two score differences was "heavy
tailed" meaning that
large differences like 50
or 60 or more occurred more
often than you would expect
based on just the average or the
standard deviation of differences.
So yeah, they were often near zero,
but they also had more extreme highs and
lows than you would see in your typical
bell curve.
That's, yeah, that's very
insightful. Thank you.
So what do you believe causes the
difference between the scores, knowing,
as you mentioned before,
VantageScore four, the,
and I'm going to say this, you
can correct me if I'm wrong.
The algorithm used is the same
across all three in crass.
So what could cause the difference?
Well, I mean, credit scores are based on
the data in a consumer's credit report,
including things like number of
accounts, amount of credit available,
utilization, delinquency,
history, and so on.
And the bureaus or nccra have
different relationships with different
data providers such as mortgage
servicers, for example.
And what this means is they might
learn about new accounts or charges or
delinquencies at different times. And
so that can lead to timing differences.
But even aside from timing
differences the CRAs can have
permanent data differences, for
example, due to a competitive market.
They each try to acquire new or unique
data elements that the other ones don't
have.
And there can be differences based on
just the processing that it takes to get
from the trade line data to the inputs
that are required for the credit
score algorithms.
And some models even have slightly
different formulas across the bureaus
which isn't the case
with VantageScore four.
So lastly in some cases a consumer or
a third party may intentionally try
to manipulate or hide certain data
like a bankruptcy or a foreclosure
by falsely disputing it
or claiming identity theft, for example.
And that's called credit washing.
And so while that dispute
is being resolved,
the consumer can qualify for a loan or
get a better rate if the loan is based on
data from just that bureau
where they filed the dispute.
So currently this doesn't
work for a mortgage because
it's hard to like wash all
three bureaus at the same time,
but it can be causes because for
differences between the bureaus.
Wow. Thanks for that insight. I would,
I would imagine lenders would be
very in tune to credit washing
knowing that, you know, later
down the road it could reappear.
So what do you, based off of your
study and the actual data that you've,
you guys have been analyzed,
what are the implications of moving
away from Tri-merge and how could this
affect pricing and loan qualification?
Okay, so this is a, it's
a big question. Yes.
to analyze it for each consumer,
we considered the representative score
differences that would result when moving
away from the Tri-merge standard.
Our paper describes the statistics for
both a single score framework and a
Bi-merge framework.
But I'll give you a few examples for the
single score framework since it's the
one that's been discussed most recently.
And I don't want to throw out
too many numbers, . So,
so note that when choosing a single score
randomly at least a third of the time,
you'll get the median itself. So you'll
actually get the Tri-mergestandard.
But still that random choice in our study,
the random choice differed from the median
by 10 or more points about 14% of the
time. And that difference could be
either higher or lower than the median.
On the other hand, if you consider, well,
what's the furthest single
score for a given consumer?
What if you choose the outlier
score away from the median,
then fully 35% of consumers had
at least one score that differed
from the median by at least 10 points.
Looking at that range like
say from six 40 to 7 79,
where a consumer could potentially move
between those pricing bins that I was
mentioning earlier this
actually increased to 40%.
And a difference of like 10 or more
points might not sound like a lot,
but we found that in that
crucial range from six 40 to
7 79,
a score of 10 or more resulted in
a move between the 20 point pricing
bins either up or down about 83%
of the time. Wow. And then, yeah,
. Yeah. Yeah.
So because it doesn't require a 20
point score difference to move bins,
it just kind of depends on where you are.
Where you are within the bin. Yeah.
Got it.
Where you are within the bin and what
that difference is and whether it's higher
or lower. Yeah. so on the other hand,
a 20 plus point difference will
always result in a move either
up or down to a different pricing
point for consumers in that range.
And we found that more than one
fifth of these consumers or 21%
had at least one score that differed
from the median by at least 20 points.
So basically guaranteeing that that's
going to happen if it's used. Oh wow.
Yeah.
So you are asking how this could
affect pricing and loan qualifications.
So kind of moving on to that I
want to give a sense of what moving
between bins means and sort
of practical dollar amounts.
So let's consider a mortgage
for 350,000 that has an initial
loan to value of 90%.
To qualify for A GST mortgage
alone with an LTV of 90 will
usually require PMI
private mortgage insurance.
And that cost will also depend on the
consumer's representative credit score.
So looking at the difference in
present value for those PMI payments
you know, depending on where you are and,
and combining it with the
difference in the GSCs
LLPA risk fee moving between
20 point pricing bins
can cost between like 3000
and $5,000 in present value
depending on the underlying credit
score. So if you're on the lower end,
maybe you're more, you're
closer to the 5,000. Got it.
For some changes you're
closer to 3000 so that,
you know, the 3000 to $5,000 just
for moving between consecutive bins.
Yes. and yeah, . Yes. So ,
so when it comes to loan qualification
those on the margins could be denied
a mortgage if only the lowest score is
used. Right. And also
those on the margins,
some of them maybe really
can't afford a mortgage,
but if the highest score is used, they,
they might end up qualifying and
then ultimately having to foreclose
and that's worse than if they just
hadn't gotten a loan in the first place.
Wow. That's again, very insightful.
So talking about consumers, I mean we're,
we're down to an example of a
anonymous individual consumer.
What populations had the largest
differences? Because you had mentioned,
you know, you had third party out
different data, so mm-hmm .
Can you explain to us the populations
that you saw the largest differences in
your study?
Sure. So first I'll talk about
differences based on the credit score.
In general, lower
scoring consumers had higher
variations between scores.
So I've already mentioned that for
consumers in the 640 to 779 range
score differences maybe
I didn't mention this,
but they were a bit larger
than for consumers in general.
Once you start reaching the higher end
where the median is between 780 and 850,
there are fewer
differences between scores,
partly because there is that cap at
850, so you just can't get any higher.
Right. so for example, I said earlier,
21% of consumers between 640 and
779 had at least one score
that differed from the median,
either higher or lower by at least
20 points. For those above 780,
it was much lower, just 6%.
And for those consumers it also matters
less because they're above that top
LLPA bucket and as a group they
have the lowest delinquency rate,
so they are lower risk. On the other hand,
when you go down in credit
score for consumers with scores,
say between 600 and 639,
now we're talking about whether or not
you qualify for a GSC loan in the first
place. And more than a
quarter of these consumers,
26% had at least one score that
differed from the median by
20 or more points. In fact,
more than 10% had a score that
differed by at least 40 points.
And and that's compared with
7% of consumers in general.
Which is two bands. That
would be two, two, yeah.
You're guaranteeing two bands
at that point, or you know,
or you're falling way below or above
qualification where maybe you shouldn't.
So so that's showing some of
the credit score variation.
In addition to that, though,
there are differences.
We saw variation between
different minority groups and
variation based on household
income. And for example,
if focusing again on those
20 plus point differences,
18% of all consumers had
at least one such score.
But for a specific minority,
a group that increased
from 18% up to 23% whereas
for white consumers it was only 15%.
So you're looking at a pretty big
difference in like how many consumers have
some difference of at least 20
points from the tri-merge median.
So moving away from the tri-merge
standard would affect different groups in
different ways.
Makes sense. Thank you for that.
So as we're talking about the
single impossible dual bi-merge
debate in the industry,
can you explain to us what score
shopping is and how you analyze that?
Okay,
so score shopping is when the
lender pulls credit reports
from multiple bureaus and picks the
consumer's highest credit score to
get the lowest fees. And the
best PMI rate for the consumer
currently under the tri-merge standard,
it's not possible to do this for a GSE
mortgage because all three reports must
be pulled and the immediate score
must be moved, must be used.
But there's a lot of competition
among lenders to get,
get a consumer's business. So if
the tri-merge standard is relaxed,
lenders may be incentivized to do this.
And even with rules in place to try to
prevent lenders from score shopping,
consumers who shop around between lenders
for the best rate and the lowest fees
would implicitly be score
shopping themselves.
So they would tend to choose a lender
that uses their best possible score.
To analyze this,
we considered metrics around what would
happen if the consumer's highest single
or bi-merge scores used.
How often could the consumer increase
their credit score by a given amount over
the tri-merge median,
and how often would they move
up by at least one pricing bin?
So we found under a
single score framework,
consumers in our study, again,
in that kind of critical range
between 6 47 79 mm-hmm
they could increase their score by
at least 10 points 22% of the time,
and by at least 20 points 11% of the time.
So that's more than one out of 10 of
these consumers that could achieve a 20
point increase, which guarantees,
again, moving up in the pricing grid.
However, as we kind of
mentioned alluded to before,
increases of less than 20 can also result
in a move up in bins just depending on
where you lie within that bin. So in fact,
more than a quarter of consumers in
this range or 26% could get a better
price by score shopping
or comparison shopping.
Meanwhile for consumers on
the margins of approval,
these numbers are even higher .
So you potentially have consumers
appearing to qualify for loans that they,
they cannot afford.
Which we might find out later. Right. If,
if they get the loan in the future.
Right. that would be unfortunate.
Right, right.
not, you know, I'm sure
the consumers would be happy,
but it would be unfortunate
for possibly the market.
How about in your study, you know,
you guys have tried to do some real life.
Do you see a real life dollar
consequence with score shopping?
Were you able to articulate
a sample on that?
Going back to the,
the previous example that
I gave that sample loan,
$350,000 at 90 loan to value,
we mentioned that moving between
adjacent 20 point pricing bins results in
a cost difference between 3000 and
$5,000 in present value.
So now we're talking about that
being a savings for the consumer.
And of course jumping more than one
bin would roughly double that amount.
Right.
so I should have maybe mentioned
that like 4% of the consumers between
six 40 and 7 79 could increase
their score by 40 or more points,
which basically guarantees that
move up by two bins. And so,
you know, on the face
of it, to the consumer,
this represents a pretty
big savings in cost. Yeah.
To the mortgage insurer
and the credit investor,
this represents potentially
underpriced risk.
And if more consumers started to
engage in things like credit washing
to hide derogatory events,
then it could be even worse.
So these stakeholders could end
up resing loan pricing across the
board to offset that
risk that you know, that,
that they're not,
they need to be compensated for the
credit risk that they're taking. Yes.
and so,
so basically rates could go up across
the board and there could be sort of a
feedback thing going on where this
even increases the frequency of score
shopping even more and sort of blocks
in less precise credit pricing.
Yes.
Or I can imagine some
startup company out there,
helping with the
situation, but ,
let's hope not. So
along the debate in the industry,
we've also seen a number thrown
out a few times about, you know,
what about doing a cutoff
for GSE loans at the 700
score range.
So how do you feel about
that cutoff based off
of your study? Would it
eliminate score shopping or.
I don't think so.
You don't think so? Okay.
No, I don't think so. ,
so basically this
hybrid approach allows
the lender to choose a single
report framework if the first
score polled is above 700.
But if the first score is below 700,
then all three reports would be polled.
And the, and you know, all
three reports would be used.
You'd had, you'd have the median,
so you'd be back to the tri-merge
framework for that consumer. Right.
so aside from consumers
between I guess 700 now
and 7 79,
still being able to score shop around for
the lender that uses the highest score
there are consumers with median below 700.
So their tri-merge standard
would be below 700,
but the maximum score is above 700.
And if that maximum score
is the first one polled,
it would become the
representative score for the loan.
So for consumers like these, there would
still be an incentive to score shop.
So we looked at how
often this could happen.
And in our study dataset overall
there were about a hundred million
scored consumers with a
tri-merge median below 700.
And of those 100 million consumers,
about 6% had a maximum score above 700.
And of course these percentages
varied depending on the consumer's
underlying credit score.
So it was more common for
consumers up closer to 700.
So looking at the 660 to 679 range,
about 8% by using the maximum
score could get up above 700.
And if you move up to the 680 to 699
range, now we're looking at about 90,
or sorry, not 90,
29% of those consumers had
a maximum score above 700.
And so they would still
be able to increase their
score through score shopping
under this hybrid approach.
So we've talked about a lot and Joni,
you've shown us that the data between the
bureaus isn't just slightly different.
It can actually lead to significant
different outcomes for both the
consumer and lenders.
If there is one fundamental truth that
lenders and investors should keep in mind
about credit transparency
as the industry evolves
based off of your study, what would it be?
I guess like something for
everyone involved in the mortgage
industry,
to keep in mind is that
when changes are made in the
mortgage industry,
they tend to remain in effect for
many years as the new standard
mm-hmm . So it's
important to get things right you know,
to,
to be cautious and look at all the data
and perform a whole bunch of analysis
and think through the potential
ramifications and the unintended
consequences of reducing the
data set that's used to qualify
and price mortgages. And,
and especially if there's any information
asymmetry that's introduced to
the system
I think we all have the same goal of
ensuring a functioning and fair mortgage
market while increasing affordability
for consumers and getting consumers into
homes and mortgages that they actually
can afford. Mm-Hmm .
However,
those holding the credit risk have an
important role to play and they may
require higher compensation for risk
if new standards result in pricing
uncertainty or adverse selection.
And that potentially means higher
mortgage rates for everyone.
So I guess in short any changes need
to be carefully and thoughtfully
considered.
Yeah.
Well luckily you guys were
able to conduct a study that
I don't believe has ever been done
before. So we greatly appreciate
your ability to do the study.
And let's get an update from our
economists at Moody's Analytics.
There's a lot to get to, but
let's start in the Middle East.
US and Israeli start strikes on Iran
have sent oil markets into a tailspin.
Crude oil prices have surged
since the start of the conflict.
The Strait of Hormuz, which carries
about a fifth of global oil supply,
has essentially closed. Our base
case is that the disruption is real,
but will be relatively short-lived.
We expect prices to moderate,
but stay above $70 into the latter half
of the year before gradually returning
to prior levels.
But there's about a 2010 to
20% chance this gets worse.
And in that scenario, oil hits and stays
above a hundred dollars per barrel.
So for American consumers,
every $10 increase in the price of oil
translates to a roughly 25 cent increase
in the price of gas.
And every sustained $10 increase leads
to an addition of roughly 15 basis points
to inflation in the following year.
So this is a stagflationary pressure
layered on top of what tariffs and tight
immigration policy were already doing.
So on jobs February was rough,
payrolls fell by 92,000,
the unemployment rate ticked up to 4.4%,
and labor force participation
hit its lowest level since late
2021 3-month average job
growth is now just 6,000,
and the labor market has essentially
gone nowhere since the Liberation day
announcements last year. Now that said,
we expect job growth to pick up
but remain anemic in the year ahead
with weakness on both the demand
and the supply side. So on tariffs,
despite the Supreme
Court's recent IEPA ruling,
we believe that the administration has
enough policy tools to keep the effective
tariff rate more or
less. Where it has been,
the composition may shift in terms of
which trading partners bear the burden,
but the macro impact is largely unchanged.
So the bottom line is the US
economy is still advancing,
but the risks are rising swiftly. The
longer the Iran conflict persists,
the steeper the damage
and that clock is ticking.
Thank you immensely for being our
guest on today's Market Pulse podcast.
I really enjoyed my time with you.
And I hope that we run into each
other again in the future, .
Yeah, me too. I enjoyed it too. Thank
you, Wendy. It's stay here, .
Yes. Have a great day. Thank you.
The information and opinions provided
in this podcast are intended as general
guidance only, and are subject
to change without notice.
The views presented during the podcast
are those of the presenter as of the
date.
This podcast was recorded and do not
necessarily reflect official positions of
Equifax.
Investor analysts should direct
inquiries using the Contact us box on the
investor relations section@equifax.com.