Description

The island of Puerto Rico is home to many interesting things including beautiful beaches, Old San Juan, Bacardi Rum, and the most egregious accounting in financial services. I am recommending a short in Doral Financial as I believe the company’s true economic earnings power is less than 60% of what bulls expect. Like many specialty finance companies in the late 1990s, Doral has done the financial equivalent of selling its soul to the devil: in return for rapid earnings growth and a soaring stock price, it is massively abusing non-cash gain-on-sale accounting. All of the information in this write-up has been derived from Doral’s public filings, unless otherwise noted.

Doral is the largest originator of mortgages in Puerto Rico. The company originates approximately $8.0 billion of mortgages annually (based on the last quarterly run-rate), approximately 70% of which is in Puerto Rico, and 30% is wholesale U.S. originations. There are a number of reasons why the Puerto Rican mortgage and banking market is actually quite attractive, and Doral is certainly not a bad business. However, Doral is not a 3.9% ROA mortgage bank, as I will detail below.

There is a housing shortage in Puerto Rico, which leads to stability in the mortgage market. This shortage has resulted in rapid home price appreciation which creates a strong purchase market and allows home owners to quickly build equity in their homes. Given that most refinances on the island are driven by homeowners’ desire to take equity out of their home rather than get a lower rate, refinance origination volumes are less rate-sensitive than the conforming mortgage business in the United States. This dynamic is similar to the subprime mortgage market in the U.S. As a result of this market dynamic, I don’t believe that origination volumes are likely to decline materially (although a moderate decline in volume is likely, it is not a part of the short thesis). Doral’s large market share (40%), as well as the retail (rather than wholesale) nature of the mortgage market in Puerto Rico add to the attractiveness of the business. I am not going to harp on the positive aspects of the market, as anyone who is interested in learning more can turn to any sell-side initiating coverage report on the company. The company does a good job of courting sell-side analysts, and while the analysts generally do a terrible job of analyzing the underlying economics, they are quite proficient at taking the market data that Doral gives them, condensing it into paragraph format and putting pretty charts alongside. The banking market in general in Puerto Rico is attractive because the banks on the island don’t pay much in the way of taxes (which is why Puerto Rican residents aren’t so excited about becoming U.S. citizens). Again, I won’t waste space by going into the details on the tax situation, as sell-side reports can explain why this is the case. Doral’s effective tax rate has been approximately 18%.

The gain on sale that Doral books on its mortgages comprises the majority (70% - 80%) of its revenue. The Company’s U.S. mortgage business (30% of total DRL production) is very low margin wholesale production. Doral’s description of its U.S. mortgage business is as follows: [Doral] purchases conforming mortgage loans on a wholesale basis from U.S. financial institutions without the related servicing rights which are generally securitized into FNMA or FHLMC securities and sold into the market. Gain on sale for the conforming U.S. mortgages is probably around 50 basis points. The non-conforming Puerto Rican loans are responsible for the vast majority of the gain on sale revenue (70% in 2003 and 85% in 2004). Non-conforming loans are so named because they can not be sold to Fannie or Freddie. According to DRL, the principal deviations that do not permit non-conforming loans to qualify for such programs are relaxed requirements for income verification or credit history, or a loan amount in excess of those permitted by Fannie and Freddie. In the case of Doral’s non-conforming production the constraint is likely not loan amount, as the average non-conforming loan originated by DRL is about $90K in size (while the ceiling for FNM and FRE is approximately $330K). These non-conforming loans are similar to Alt-A (or more appropriately no-doc subprime) loans in the United States, for those of you that follow mortgage companies. These loans are essentially no documentation, subprime credit, small balance home loans in Puerto Rico. Credit performance appears to have been extremely strong in the past due to the rapid price appreciation in Puerto Rican real estate, though I will note that many people in the market point out that loans to this customer segment are extremely high risk, and Puerto Rican banks will not carry them on balance sheet without a credit guarantee from Doral.

The non-conforming loans are sold in private transactions (with unique structures and accounting – think of it as similar to an MBS securitization) to other Puerto Rican domiciled financial institutions. In this structure Doral basically sells these low balance no-doc subprime conventional (30 year fixed interest rate) loans on a floating rate basis. Doral agrees to receive the fixed coupon on the mortgage and pay to the buyer of the transaction a coupon indexed to 3 month LIBOR. Doral retains any net interest income between the fixed loan coupons they originate and the floating rate loan coupons they sell backed by the fixed coupon loans. This creates an IO (interest only) strip that Doral then values, takes as a gain on sale through the income statement, and books on the balance sheet as an IO asset. Technically this is an Inverse IO (IIO) as the pay portion floats with LIBOR and the receive portion is fixed. The average spread to 3 month LIBOR in these transactions is somewhere around 150 bps (according to the company, buyers of the loans, and a fixed income trader who is familiar with the paper). This paper is attractive to banks and other fixed income buyers in Puerto Rico since it is floating rate, which is rare on the island (floating rate paper helps local banks reduce interest rate risk in their short-dated funding), and it qualifies for tax advantaged treatment. By funding the fixed rate loans with the floating rate paper in the sale, the interest rate risk that the local banks are eager to avoid is transferred back to Doral. It is important to note that the past four years have been a historically profitable time to take this interest rate risk due to the sustained steepness of the yield curve.

In these loan sale transactions, Doral sells the loans on full-recourse basis meaning DRL retains the credit risk (this is potentially a very important point). The coupon on the low balance non-conforming mortgages are 7.0%-7.5% on average (it has been drifting downward as a result of increasing competition in the last two years and market contacts tell me that most loans are currently being originated below a 7% yield). Therefore Doral’s spread is the 7.25% average coupon, less 3 month LIBOR (currently 2.80%), less a spread of 150 basis points, less a servicing fee and credit loss assumption. Doral strips 25bps off of the loan coupon and values that in its MSR (it should be noted that since these are very small balance loans with a high balance of non-performers, both factors that increase servicing expense as a percent of loan balance serviced, the MSR related to these loans should be valued far lower than conventional conforming MSR) so I will assume 25bps as the servicing fee. The company assumes no credit losses when modeling and valuing its IO cash flows. I will give Doral the benefit of the doubt and use this assumption as well given the historically low losses on Puerto Rican mortgages resulting from a decade of strength in the island’s housing market. The table below walks through the various cash flow assumptions that result in the initial spread on a non-conforming mortgage originated and sold in private transactions by Doral:

Coupon on mortgage: 7.25%
Less: 90 day LIBOR: 2.80%
Less: spread over 90 Day LIBOR: 1.50%
Less: servicing expense 0.25%
Less: credit losses 0.00%
Net Spread to DRL at the outset: 2.70%

Doral books a gain on sale at the time that it enters the non-conforming loan transaction and creates the above calculated spread. The gain on sale should be the present value of the most probably future cash flows that are related to the loan sale transaction. The key assumptions in this calculation are: 1) the average life of the transaction, 2) the net spread received each year of that average life, 3) the rate at which these cash flows are discounted.

So what are these IIOs worth? What SHOULD the gain on sale be? Let’s first determine the key assumptions:

ASSUMPTIONS FOR IO VALUATION

1) the average life of the transaction ------------------> 5 years

This is one area where it is difficult to come to a firm view. There is very little data available on the life of the small balance loan product. We do know that this product comprises 55% of the DRL servicing portfolio and the overall portfolio has had a run-off rate of 20%-28% over the last couple years. This would imply an average life for this product of 4 to 5 years. Since there are prepayment penalties for the first 3-4 years typically, it seems reasonable these loans would survive on the books for that long. So we will assume that the average life of this loan product is 5 years.

2) the net spread received each year of that life --------> 0.90%

Currently, the going in spread on one of these transactions is the 270 bps we calculated above. But, because of the inverse (receive fixed, pay floating) nature of these IOs, that spread will not be enjoyed for the life of the deal. The cash flow assumption used to value this IO must consider what the forward curve indicates LIBOR (ie the funding cost) will be in the future. As there is an upward sloping curve the expectation is that this 270 bps is declining each period going forward for the life of the transaction. To get an accurate assessment of this expected cash flow stream one should look at what forward LIBOR indicates the spread would be for each quarter for the 5 year period being valued (ie what is 3 month LIBOR expected to be for each period for the life of the transaction). A shortcut way accomplish the same result is to use the 5 year swap to approximate the average expected LIBOR for the life of the transaction. The 5 year swap is currently at 4.60%. So incorporating this into the cash flow table from above, we get:

Coupon on mortgage: 7.25%
Less: 5 year swap 4.60%
Less: spread over LIBOR 1.50%
Less: servicing expense 0.25%
Less: credit losses 0.00%
Net Spread to DRL fully swapped: 0.90%

The credit loss used in this analysis assumes that the housing boom in Puerto Rico continues unabated and that losses don’t increase, even as market contacts tell me this loan product is being offered at ever higher LTVs (often greater than 100%) and to ever shakier credits as the mortgage producers stretch for volumes. Any deterioration in the housing market and credit performance would significantly impair this cash flow assumption which assumes a world’s best mortgage credit performance of no losses.

3) the rate at which these cash flows are discounted -----> 10%

Choosing discount rates is always a very subjective exercise. What discount does the uncertainty of the stream of cash flows warrant? What return would the market require in order to exchange cash today for these future cash flows? Similar assets in the United States are discounted anywhere from 12% to 25%.

Plain vanilla conventional conforming MSRs are discounted at the lower end of that spectrum. These MSRs have a more volatile life expectancy, yet the spread is fixed and there is no credit risk. So the market uses 12% for an asset that has a more volatile life, but that doesn’t bear the credit risk or spread risk of the Doral IOs. The value of an appropriately booked MSR can be approximately hedged so the risk of loss is reduced, warranting this low discount rate.

U.S. subprime residuals are discounted at 18%-20%. The life on these assets are less volatile than conventional conforming MSRs, yet there is credit risk and spread risk as these are Inverse IOs much like Doral’s IO asset. This is an asset with a shorter average life than Doral’s IOs. All else equal, that would call for a lower discount rate for the US sub-prime residual as you are relying on a shorter period of cash flows, so there is greater visibility into the performance and a shorter duration. Subprime valuations also assume very high credit losses versus Doral’s 0% credit loss assumption, so the underlying assumptions in a sub-prime residual have a greater margin of safety.

To give Doral the benefit of the doubt, I will use a 10% discount rate, though there are very strong arguments that it should be substantially higher.


INTEREST ONLY STRIP VALUATION -----------------------------> 3.5%-5.5%

IO VALUATION ANALYSIS:

Using the above assumptions, this becomes a simple DCF.

Discount 90bps of cash flow received every year for a 5 year average life at a 10% discount rate. This results in an IO valuation of 3.4%. 3.4% is a phenomenally good gain on sale (loans are sold at par so no capital markets loss to offset the booking of the IO) for a mortgage production business and Doral deserves a great deal of respect for building a market position that allows them to extract these economics from an industry that has become commoditized most other places in the world.

While I do not believe these assumptions are too conservative, lets relax them a bit to give Doral an even greater benefit of the doubt in its mortgage profitability. Lets first extend the average life of the loans to 7 years, the longest Doral states that some of the loans last. If we assume the life of these loans is 7 years we must then also adjust the expected spread based on 7 year funding which costs an incremental 16 bps (ie 7 year swap less 5 year swap: 476bps - 460bps = 16bps). Actually let’s not even charge this incremental funding cost against their spread to give them further benefit of the doubt. Finally, let’s continue to assume no credit losses and discount this cash flow stream at 7 year treasuries (4.35%). Therefore we have a 90bps spread for 7 years at a 4.35% discount rate. This gets us to a 5.3% valuation when discounted. Would you pay this much for this cash flow stream?


DORAL's IO VALUATIONS:

If you look at Doral’s net production margin, you can see a steady march higher in recent years. A few years back, this number was in the 300bps range and in most recent quarters it has risen as high as 800bps-900bps. As I mentioned earlier, a large part of their production (30%) is US wholesale that should have a margin that is at best in the 50bps range. This implies that they are recording gains of over 12.5% on the rest of their production. The remaining production is composed of primarily Puerto Rican small balance non-conforming loans and some Puerto Rican FHA/VA and conforming production. If you look closely at the notes in Doral’s filings, you see that they disclose the IO produced in a given quarter’s non-conforming loan structured sales. They also disclose the non-conforming loan balance sold in these deals each quarter. From this you can calculate the IO valuation as a % of loan balance sold in a given period. For the past 8 quarters, this IO has been booked at 14%-18% of the loan balance underlying the IO.

Non-conforming loan structured sale activity:
Q103 Q203 Q303 Q403 Q104 Q204 Q304 Q404
Loan balance sold 289 457 513 NA 744 656 900 NA
$ IO valuation booked 52 64 76 90 114 111 143 NA
% IO valuation booked 17.9 14.1 14.6 NA 15.3 17.0 15.9 NA

As can be seen, Doral has been valuing its IOs at 15%-17% of the underlying loan balance of in 2004. These are the same IOs that I valued earlier at 3.5%-5.5%. Again, since this is gain on sale accounting, the valuation is based on a set of assumptions. Doral has chosen to use more aggressive assumptions than I felt reasonable in the earlier analysis. Doral’s assumptions are as follows:

Average life: 7 years
Discount rate: 8.25%
Net spread: 270 bps

Doral has taken an aggressive stance on each of the key drivers of the gain on sale valuation. The assumption that leads the greatest variance from our valuation is clearly the net spread. We assumed approximately 90 bps of spread for the life of the deal while Doral is assuming 270bps. What leads to this difference? Doral has decided to conveniently ignore the forward curve. While the market expects LIBOR to rise in the coming years, Doral is assuming that LIBOR does not budge. Doral is valuing a cash flow stream that it is extremely unlikely to receive. My IO valuation of roughly 4% is over 70% lower than the valuation Doral assigns to this asset.

In the past few years, the booking of the IO has been an ever increasing component of pre-tax income:

IO booked in period as % of pretax income:
Q103 Q203 Q303 Q403 Q104 Q204 Q304 Q404
62.1 71.8 73.0 76.6 89.9 79.3 97.7 NA

Clearly, the booking of the IO is the dominant component of this company’s earnings. If it were recorded at the valuations that I believe represent the underlying economics, earnings would be reduced anywhere from 60% - 70%.

Bulls will cite that Doral receives opinions from outside valuation firms and takes the lowest valuation for its gain on sale purposes. It should be noted third party MSR/IO valuations are notorious for their use of management supplied assumptions. Valuation of this type in the industry typically involve the company that hires the valuation firm, submitting the assumptions it wishes the valuation firm to use in the analysis and then the valuation firm returning an analysis yielding the valuation that resulted from the company supplied assumptions (ie the company’s valuation). There is a high degree of flexibility with these valuations and for esoteric assets like Puerto Rican Inverse IOs the bounds are particularly wide.


THE IMPOSSIBLE HEDGE:

Doral will assure investors that the IO is hedged so there is nothing to worry about. The fundamental problem with this argument is that one cannot hedge to lock in an impossible valuation (at least not without spending an amount that brings the value of the IO net of the hedge expense back to the true fair value). Doral would need to lock in spot LIBOR at the current rate for the life assumed in their IO transactions (7 years). The cost of this instrument would be the area under the forward curve and above the current 3 month LIBOR spot fixed for the life of the transaction (plus a premium). One way to think about it is as a gross and net valuation: the gross IO valuation that Doral calculates is actually somewhat accurate under this methodology (if one ignores the aggressive discount rate, credit and life assumptions), but at the time they book the gross gain from the IO they should book an offsetting loss of the hedge (area under the curve plus the 150bps spread) to get back to a net valuation equal to what I calculated earlier in the analysis (3.5%-5.5%). I swapped the funding out 5 years in the initial valuation analysis to capture the expectation that LIBOR would squeeze the IO spread as the life progressed and valued the cash flows net of the hedge.

It is possible to hedge an inverse IO, but the practical options are 1) ‘lock’ in the forward curve (not spot LIBOR) through the life of the asset, or 2) swap the funding for the duration of the asset (as I assumed in the earlier valuation analysis). So the correct valuation methodology would be to either 1) assume the funding costs rise with the forward curve and value this declining/disappearing cash flow stream, or 2) assume a swapped (average LIBOR) funding for the life of the deal and value the stable, but reduced initial cash flow stream over the life. Both of these methods will return approximately the same valuation. Then this valuation could be roughly locked in by either 1) shorting Eurodollar Futures along the curve for the life of the deal (reducing the hedge going out based on the CPR assumptions) or 2) swapping the funding out for the life of the deal. Then during the life of the deal, any interest rate deviation from what the forward curve predicted at the time the deal was initially executed would have offsetting gains and losses in the IO and the hedge (assuming the hedge was put in place at the time of the deal). For example, if the IO was valued based on the forward curve at the time the deal was initiated, and the hedge was put in place at that time, a later shift to higher expected rates reflected in the forward curve would force an impairment to the IO, but also an offsetting gain in the hedge.

Under Doral’s valuation methodology, every time LIBOR rises they will either need to impair their IO valuation, or relax some of their other assumptions (beyond already extremely aggressive levels). And they can not “hedge” against this inevitability. It is not possible to hedge to lock in the valuation that assumes spot LIBOR persists for 7 years without spending the entire area under the curve for the life of the deal, and it is clear from a quick look at their financials that they haven’t expensed what would be required to buy this protection.

It should also be noted that of late Doral has been able to take advantage of their double secret “macro hedge” to shield their earnings from impairments. As far as I know this is some kind of tax deal with the Puerto Rican government that conveniently became available as they started to take IO impairments. Details surrounding this are sparse as, according to the company, they are covered by a non-disclosure agreement with the Puerto Rican government.

(***due to VIC length constraints this write-up is continued in the MESSAGE STRING THAT FOLLOWS***)"

Catalyst

-Continued impairments to the IO
-Investors no longer believe that the reported earnings represent economic earnings
-Potential for earnings restatement
-Potential for off balance sheet assets and liabilities being put back on balance sheet