Modelling Fannie Mae and Freddie Mac – Part VII

In Part VI of this series I came to the (non-consensus) conclusion that both Fannie and Freddie were long-term solvent and that the cost to the government of their conservatorship would be zero.  I also suggested that the common stock had value and that the (non-cumulative) preferred shares (currently trading at 4c to 6c on the dollar) would one day receive par. 

There are model sensitivities and economic sensitivities to this conclusion.  In this post I want to (begin to) explore how robust those conclusions starting with the model in Part IV.  I am conducting an “idiot check”.  In the next two posts I will do idiot checks on Parts V and Part II. 

I apologise in advance as these three posts will look a little disjointed compared to Parts I to VI.  In this post I am using all sorts of anecdotal or practical data to test my hypotheses… this is a practical - not a theoretical exercise – and it is as a result messy.

Background

In Part IV I built a model which predicted future credit losses for Freddie (and told you I had done so for Fannie).  We know that there are huge credit losses coming at Fannie and Freddie as delinquency is rising, house prices are awful and default is becoming more commonplace.  My model allows for huge losses.  The main question is whether the losses allowed for are quite huge enough.  The bears (and there are many) seem to assert losses considerably larger than I am projecting.

Well lets start with one of my favourite charts of the traditional mortgage guarantee business.  Again I do this only for Freddie Mac – leaving the reader to do it for Fannie if they want.

In this graph we have $1.907 billion in charge offs during the last quarter and 1.104 billion in the quarter before.  The charge-offs were effected by the foreclosure moratorium which would have meant charge offs were understated in Q1 and overstated in Q2.  Lets call it an average of $1.5 billion per quarter.

At the end of the 2nd quarter there were 25.2 billion in reserves.  I calculated in Part IV that an additional $12.5 billion would need to be provided for over the out-years.  We have thus built 35.7 billion of loss reserves (current reserves and reserves to be taken) into our model.  The key sensitivity question is: is this enough?

Well – if the loss rate does not get any worse then $35.7 billion at $1.5 billion per quarter would last 23.8 quarters or almost 6 years.   I can be fairly confident that if realised loss rates do not rise then model reserves are adequate – as the alternative requires A substantial numbers of people who obtained their mortgage in 2006 to remain current until 2014 and then default.  That is possible if the economy is really sour in 2014 – but it is not an obvious or expected outcome. 

More realistically I am modelling realised losses rising for some time before falling.  Foreclosure stats are rising in aggregate – albeit rising at a slower rate.  If realised losses were to double we could last three years and still be reserve adequate.  That is roughly what the model would imply.  If realised losses triple and remain at that rate then (unfortunately) my 35.7 billion in losses still to come will wind up being an underestimate.  So in a sense – what is required is some comfort that the realised loss rates – particularly in the nasty 2006 and 2007 vintages are unlikely to much more than double from here.

Test 1 – what is happening at on the ground in California

California is the worst state for losses for both Fannie Mae and Freddie Mac.  Some states (particularly Arizona and Nevada) have higher losses as a percentage of loans outstanding.  Some states (particularly Michigan) have higher severity – with many houses recovering less than $2000 on foreclosure.  But California is a massive state with high house prices and high losses.  Arizona and Nevada are simply not large enough to make my estimates wrong.  California is.

Fortunately California has very good data on defaults, notices of trustee sales and recoveries at trustee auction courtesy of Foreclosure Radar.  Paul Kedrosky has pointed out that we have reached the “new normal” in California – a stabilisation of foreclosure processes at high levels. 

 

The point is that the worst state has stabilised.  This tends to indicate that the defaults are not even going to double from here – let alone triple.  On this piece of evidence my default loss estimates are over-stated and Fannie and Freddie will return to full solvency more rapidly than I previously anticipated.  Moreover – adding to the robustness – the proportion of realised losses happening in the “bubble states” (California, Nevada, Arizona, Florida) has been rising by quarter – not falling – so it is reasonable to assume that as-go-the-bubble-states-goes-the-whole-book.

Test 2 – early stage delinquencies

Early stage delinquency (particularly 30-60 day buckets) is the best leading indicator in delinquency data.  90 plus delinquent loans will continue to rise well after the economy or credit cycle has turned because bad loans accumulate – especially when selling foreclosed property is hard.  By contrast loans going through the 30-60 day bucket give you an idea of the flow rate into problem loans.  When the 30-60 day delinquency improves you know the credit problems are ameliorating (even though the 90 day buckets are getting worse).  One of those hopeful leading indicators is that in many categories of loan I look at the 30-60 day buckets are improving – and usually improving more than seasonal factors indicate.  [That bucket is seasonally difficult in February when the Christmas bills come due and easier in summer because there is more overtime or temporary work about…  Incidentally 30-60 day buckets are getting better in some credit card books as well.] 

Fannie and Freddie (unfortunately) do not give early stage delinquency data (I believe because the loan servicers report the data inconsistently).  However the office of thrift supervision has recently conducted a survey.  You will find this chart – and much more data like it on the OTS site.

 

Again this indicates that is unlikely the charge-off will even double.  It is unlikely to more-than-double – and again it seems my loss estimate is too high.

Estimated defaults versus amount in mortgage pool

The model presented in Part IV estimated losses for each year of business.  Here is the table again:

 

Year	Defaults to come by year (billions)	Severity by year (percentage)	End losses (billions)
2000	0.3	10%	0.0
2001	0.6	12%	0.1
2002	0.9	14%	0.1
2003	2.0	15%	0.3
2004	3.4	18%	0.6
2005	9.7	40%	3.9
2006	21.9	60%	13.2
2007	27.5	55%	15.1
2008	14.3	30%	4.3
Losses still to come			37.6

 

Now we estimate that (as of year end 2008) there were 13.2 billion of losses (and 21.9 billion in defaults) left to come in the 2006 year of business.

The Freddie Mac credit supplement gives us other data about the credit in that year of business. 

 

 

There is in the 2006 vintage only 236 billion of unpaid principal balance left – and that number is falling quite fast.  (The rest has refinanced, defaulted already or been repaid.)  We have built into our model a 9.28% default rate from here.  Given that the seriously delinquent loans are only 6.34 percent that seems a little harsh – all seriously delinquent loans need to default and then there needs to be another serious round of new delinquencies.  Given that most delinquencies cure (even in times like this) because people with a default notification often try hard to pay rather than have their house foreclosed on – the required foreclosures being about 1.5 times the current delinquency does appear to be a high estimate. 

I guess – and people will say this – that people could walk-away with loans going from current to default very rapidly.  34 percent of the loans have a current loan to value ratio above 100 percent and almost all of those loans are current.  My only counter – and I know this is a weakness – is that whilst it may be in their interest to walk-away from their mortgage they are not doing so.  It could be that they are unaware that their home loan is underwater (and there is some evidence that Americans know house prices have fallen but delude themselves about their own house).  It could be that they just have good credit , the house is not far underwater and they want to pay.  It could be that they have scattered daddy’s ashes in the backyard and walking away is unthinkable.  More likely – with a 90% LTV loan on a 200 thousand dollar house you might not walk away even though the mortgage is underwater because the mortgage payments are lower than the alternative rent payments.  Whatever – they are not currently defaulting.  That is all I can say on this data.  If people have data that suggests that this will change soon I am interested.  I see no data in aggregate proving that point although there has been some data about the extent to which people are deluded as to the value of their own home and that their level of delusion is economic-conditions correlated.

The shift in housing problems

The housing market crash began with low end housing.  In most markets jumbo mortgages retained fairly good credit until recently – though there is considerable evidence that upmarket housing is experiencing trouble now.  There is also considerable evidence – much of it anecdotal – that lower-end housing prices have stabilised.  [Certainly in most markets it is considerably cheaper to buy low-end housing and make mortgage payments than it is to rent.]  This is generally supportive of my thesis as the critical 2006 and 2007 books at Fannie and Freddie contain no mortgages above $330 thousand.

An observation about second derivatives

As indicated in Part VI we at Bronte purchased Fannie and Freddie preferred stock fairly aggressively at below 2 cents in the dollar.  It was March when we first started doing that – and the world looked like a sour place.  The idea that Fannie and Freddie might actually be solvent seemed unthinkable – but we were busy thinking it. 

At the time everything was getting worse at an increasing rate.  The expression is “free-fall” – where you simply accelerate towards some immovable hard object.

The modelling did not feel very robust and the reason it did not feel robust was that all the leading indicators were getting worse.  In the curves in Part IV the 2006 and the 2007 curves were accelerating away from 2000 curve.  They did not look bounded at any multiple of the 2000 curve that I could get comfortable with.  I knew defaults would continue to get worse for a while because there was a big build-up in the delinquency buckets and delinquency is a precursor to default.

What would make me confident (and believe me I was not confident) was a deceleration in the rate things were getting worse.  I was interested in the “second derivative”.  For a while (up to about April) the second derivatives all looked good.  Briefly the second derivative of total delinquency looked bad (as reported in Fannie’s monthly data) and that really rattled me.  I posted that here.  That data-point was – it turns out – an exception to the general trend.  Only recently there have been a few bad data points (for instance the recently reported rise in early-stage delinquencies at Capital One Financial suggesting a “W shaped” recovery).

With anything that looks like a W shaped recovery this model could be wrong.  For instance we could have another big leg down in either property prices or the economy.  The data does not generally support that second leg down but that might change.  The Capital One numbers have given me pause. 

But more generally we are making predictions which are ultimately just guesses.  I hope I have convinced you that they are educated and rational guesses.  But guesses nonetheless.

The main objection I have received so far – is about loans with risk layered terms that Fannie and Freddie have in their traditional book

There have been a few objections (mostly in email) that suggest that I assume away much of the dross in the conventional Fannie and Freddie books.  Here is one email from “bob”, a mortgage market professional.

Well reasoned, but... there is a presumption in the marketplace that Fannie & Freddie's book of traditional business is solid stuff.  One has to really question that.  What about the 100 LTV loans made to borrowers with 570 credit scores and 67 DTI's?  What of the 90 & 95 LTV Interest Only loans made to flippers?  What of the 90 LTV Stated Income loans (many made to flippers)?  What about the LTV's being based on stretched and hyped appraisals?  What about the mortgage company "art departments" which cranked out custom W-2's and paystubs to document loan files with?  What about all of the high LTV loans made to people with 50% DTI's and no money in their checking accounts?  And how do you square it all with a huge and growing percentage of mortgaged homeowners who are underwater- far too many of whom are (or soon will be) unemployed or making substantially less than what they were?  Then to top it off, one can only shake their head when it comes to REO disposition practices.  Bottom line, I think any model must attack the assumption that the GSE's book of traditional business is solid stuff.  Unfortunately, Fannie & Freddie were aggressive hedge funds operated to generate executive bonuses, and under the supervision of a defanged regulator.  I'm afraid that when the tide finally goes out, it will not be a pretty sight.  [Emphasis added…]

I will deal with this in a modelling context – but then also in specifics.  My model – which just assumes that defaults in the 2006 and 2007 vintages follow a curve with a similar shape to the 2000 vintage – makes no assumptions whatsoever about the content of the book.  It just looks at the 2006 book, notes that it is currently defaulting at 2.8 times the rate of the 2000 book, that the difference between the 2006 book and the 2000 book is expanding – and hence the end default may be 4.5 times the 2000 book.  Essentially though I am assuming because things are getting worse they are going to continue to get worse. 

One thing however is generally true about the mortgage market – which is the very bad loans default fast in a crisis – and then the defaults from that pool ease up, whereas good loans default slowly and defaults do not ease up for a long time.  That is consistent with simple models of human behaviour.  If you have a 100LTV loan which you purchased on a property you intended to flip in the Inland Empire then you have probably walked already.  Why?  Because the incentive to keep making the payments on a cash-flow negative property is very low.  You will probably pocket three months rent whilst the bank actually gets around to foreclosing on you – but your motivation to pay is low.

If – by contrast – you are a regular mom-and-pop buyer who purchased a home to live in and put down even 10% (which you saved by dint of hard work) then your incentive to walk-away is low.  You have emotional investment in the property even if your financial investment is wiped out.  The incentive to pay (because you do not want to be forced to move) is high.  Moreover there is a real tendency to self-delusion as to the value of the property and self delusion lowers default rates.

If a book consists of flippers and 100LTV loans then the defaults will be front loaded.  My model assumes that the defaults are rear-loaded.  The more of the drossy loans described by Bob the more the defaults will be front-loaded and the more my loss estimate is thus an overestimate.  If Bob were right I would be more comfortable with my estimates – not less comfortable.

Unfortunately – despite the protestations of Bob (and others) there are actually relatively few truly risk layered loans in the traditional books of Fannie and Freddie.  For that I need to explain risk layering.  Consider the following three loans:

Loan A is made to a customer with terrible credit (a FICO of 580 reflecting past defaults).  However the customer has clearly reformed and they now hold a stable job which you have verified.  They have saved up $25 thousand and they are buying a very low-end property (say $130 thousand) in a non-bubble state.  This is a low FICO loan or a loan to a subprime borrower – but with otherwise good characteristics.

Loan B is made to someone with pristine credit and a stable income.  You have verified the income is more than adequate to serve the loan.  The family has emotion invested in the house as they have purchased near the school in which they enrolled their children.  However the down-payment is only 3 percent because the equity that they had saved had been blown because the family had recently had medical problems.  This is a high loan to valuation loan - but with otherwise good characteristics.

Loan C is made to a customer with terrible credit (580 FICO reflecting past defaults).  The customer has a stable income which you have verified – but they are purchasing a 250 thousand home with only a 3 percent down-payment.  There is little evidence in their past behaviour that that they are capable of saving money for a rainy day.  This is a risk layered loan in that it has two major risk factors – a subprime borrower and a high loan to valuation ratio.

With these three Loan A and B are probably both good under almost all circumstances.  The first borrower shows little record of past willingness to pay a loan – but in this case they have a real incentive to pay and the ability to pay.  They probably will pay.  The second borrower shows a very good willingness to pay and the ability to pay – but their incentive to pay (being the remaining equity in their home) is low.  Nonetheless they have emotional commitment to the community and foreclosure is a difficult option.  Loan C is terrible.  You have the bad borrower and they lack incentive to pay.

The collapse in underwriting standards that occurred in America was due to risk layering.  Two risk factors is many times more risky than one risk factor.  Fannie publishes a table of how many of their loans have various risk factors. 

 

From this table you can estimate how many of the loans have multiple risk factors.  If you add up the special risk factors you get $1,112 million.  However the actual dollar value of loans with a risk factor is 878 million and those loans are 72 percent of losses thus far. 

The numbers also suggest that at most $233 million have more than one risk factor.  That is less than 9 percent of Fannie’s book has risk layering – but my guess is that those small number of loans will be about 40 percent of losses.

The point is that the proportion of losses from risk factor loans is now declining (consistent with the argument above).  It is loans without risk factors (the heart of the traditional Fannie and Freddie business) which is going to show increasing losses.   

Fannie gives some data on loans with two risk factors – separately breaking out loans with a FICO below 620 (which means the borrowers are truly subprime) and with a LTV above 90% at origination.  There are only 25.4 billion of these – 0.9 percent of the book.  However these loans represent 5.7 percent of all losses – they are more than six times as loss intensive.  Note that the proportion of losses coming from this pool is falling – probably because the loans have a tendency to default fast – you get loss burnout.  The subprime loan pool – 0.3 percent of books but 1.1 percent of losses has had even faster loss burnout.

Anyway – contra to Bob’s email my estimate is more likely to be an underestimate precisely because there is not that much risk layering in Fannie’s book.  If the losses burn out fast then they are not likely to rise far from current loss levels.  If the losses burn out slow (which is what would happen with lower-risk mortgages) then loan losses could rise for a long time as families slowly burn through their financial resources trying to keep current in their mortgage.

Sensitivity as to the income line

I am fairly confident about my estimate of credit losses at Fannie and Freddie.   Those are manageable – and the end loss to taxpayers is very unlikely to be large.  I am much less confident that the income line (which has expanded greatly) can remain so generous.  And for the losses to taxpayers to be zero I need the companies to be able to earn their way out of their current predicament.  I need the current large operating income to continue – or at least not to drop suddenly.

The sustainability of the operating income is the subject of the next post.