In this article I look at where the value is coming from in agency mREITs, by decomposing their assets and liabilities. I show that for the most part, these REITs make money by shorting naked options, and the main differentiating factors are first leverage, and then size to some extent. This approach is applicable to Armour Residential REIT (ARR), American Capital Agency Corp. (AGNC), Hatteras Financial Corp. (HTS), and Annaly Capital Management Inc. (NLY) among others.
Note that I am simplifying many things, but that's for clarity sake.
An agency mREIT essentially buys and holds agency residential mortgage-backed securities (MBS), financed through repo, and hedged with a variety of derivative or cash positions, including Treasurys, swaps, swaptions, Treasury options and TBAs. The first step is to understand the assets side, and we will go through the hedging strategies next.
Agency MBS are bonds backed by pools of mortgages, the credit of which is guaranteed by the US Government, either directly (in the case of GNMA MBS) or indirectly (in the case of Fannie Mae or Freddie Mac MBS). Taking credit risk away by no means implies that these assets are risk-free, because they carry substantial interest rate exposure.
The prepayment option
Mortgage borrowers have the option to prepay their mortgage. Now the "option" here is an important keyword, it means that these borrowers hold some kind of right over the mortgage investors. In other words, the REITs, by holding mortgage securities, are short options.
How does that work? We'll focus on fixed-rate mortgages for now, since that's what most REITs buy. Adjustable-rate mortgages are different enough that they would require another long article.
Let's take a large pool of typical 30-year fixed-rate mortgages paying 4%. There will typically be thousands of individual loans in such a security. All the borrowers tend to move, for a variety of reasons, and historically that's been at a rate of about 5% a year (see page 6 of this document on mortgage prepayment modeling for example). So that means that each year, you'd expect about 5% of the thousands of borrowers in the pool to sell their house and repay the loan. The prepayment rate is called CPR (Conditional Prepayment Rate) and most investors who have read REITs reports or presentations in the details have seen such numbers. This prepayment rate is in addition to the normal and scheduled repayment of principal, as part of the borrowers' monthly mortgage payments. All in all, the average duration over which you expect the principal to be repaid will be around 8 years on a typical mortgage pool like this, reflecting the 5% CPR. So these bonds would seem fairly equivalent to 8yr Treasurys: roughly the same maturity, and the same guarantee...
Prepayment directionality and its negative effect
In fact mortgage securities are nothing like Treasuries, because the borrowers behind this pool will likely prepay if interest rates drop, since they could get a new, cheaper mortgage somewhere else. In addition, if rates rise, it becomes less likely that people move because it increases the cost for the new mortgage they'll need to get. When rates drop, CPRs can increase massively, to over 50% sometimes (see page 10 of the prepayment modeling document). And when rates rise, prepayments will decline, as low as 2 or 3% CPR. Note that here it is mortgage rates that drives the borrowers' incentives, but they are strongly correlated to other core interest rates such as Treasury or swap rates. How does this prepayment directionality affect our "Treasury equivalent"?
A quick review of bond math is warranted at this point: A real Treasury bond with an 8-year duration will approximately gain 8% in value when interest rates drop by 1%, and lose 8% when interest rates rise by 1%. That is because when you are getting a fixed rate, 4% in our example, and "interest rates in the market" drop it means that now if you wanted to buy the same type of bond you would only get a 3% coupon. So your bond, the one paying 4%, effectively pays an extra 1% for about 8 years. So it's only fair that it would be worth about 8 points more now that rates have moved. In the reverse situation, if rates rise, it means that now you could go out and buy pretty much the same bond but it would pay 5% instead -- so your bond pays 1% less than the market now, for 8 years, hence its value drops by about 8 points. The sensitivity of bonds to changes in interest rates is called duration -- as you can understand, it's because the more a bond "lasts", the more you will gain or lose in total, relative to what can be obtained in the market.
Our mortgage bond is trickier than a Treasury, because of the effect of prepayments. If nothing happens, we can expect it to have a duration of about 8 years, so far no issue. But if rates rise, we know that prepayments will slow down, and that will make the bond last longer. Instead of getting principal back over about 8 years now it'll take maybe 13 years, due to the slower than initially anticipated CPRs. But in the higher rate environment, I can get bonds paying 5% instead of the 4% on my mortgage pool, and this 1% difference would go on for 13 years, so the mortgage bond should be down 13 points. If rates drop, then the fast prepayments imply the MBS will be much shorter than anticipated, say about 3 years only. So the extra 1% the bond now pays (4% vs the market's 3%) only lasts for 3 years, and you would expect a price increase of only 3 points. So what do we have? This agency MBS looks like a normal bond at first sight, but it is in fact a little nasty because it always underperforms its Treasury equivalent, because of the short embedded prepayment option. If rates drop, the MBS goes up 3 points, the Treasury 8 points; if rates rise the MBS goes down 13 points, the Treasury 8.
The relative underperformance, in the example above on average 5 points ( (8-3) + (13-8) ) / 2, is called the negative convexity of the MBS. All mortgage securities exhibit a degree of negative convexity. To be fair, the mortgage pool in the example should probably trade 5 points below a Treasury bond paying the same coupon, or alternatively if it traded at the same price it should pay a higher coupon to compensate for this nasty behavior. I'd estimate here that an extra 0.6% would be fair (5% / 8 years = about 0.6%).
Some types of pools have what is called prepayment protection; that is usually in the form of credit impairment, basically anything that prevents borrowers from refinancing. For example, AGNC's positions are listed out in this presentation on page 8, where one can see various particular types of mortgage pools. As a result, when rates drop, their prepays do not accelerate as much, and the value of the MBS rises more than it would have on a more standard pool.
However, a prepayment-protected pool is not necessarily a better investment than a standard one, because it depends on how much that protection costs. If I altered my earlier example and made it prepay protected, with less negative convexity, then it would trade at a lower yield, or in other words it would not pay an extra 0.6%, but maybe only 0.3%. So owning prepayment-protected pools does not necessarily mean better returns, because their prepayment protection costs something.
As explained above negative convexity comes from the prepayment option in mortgages, but one expects the same kind of negative convexity by shorting interest rate options, such as options on Treasury futures for example. Therefore one could view an MBS as a combination of some Treasury or Agency bond, and shorts on a basket of Treasury futures options.
But don't forget duration either
As with any fixed-rate bond, and as we have seen above, MBS have exposure to interest rates. If rates rise their value goes up, and the reverse holds true as well. REITs usually hedge this by shorting Treasury bonds, or by entering swaps. In this context, a swap would pay a floating rate, for example LIBOR, to the REIT, and the REIT would pay out a fixed rate. The advantage of using a swap is that it hedges against the cost of financing for repos: the floating rate that the swap pays to the REIT will typically be close to whatever the REIT pays on its borrowed money, used to buy the mortgages in the first place. The Treasuries duration, or the duration on the fixed leg of the swap, will be set so that the effect of changes in interest rates is locally canceled out with the MBS portfolio. But as we just discussed above that will still leave the portfolio unhedged for negative convexity.
Hedging mortgage risk
The only way to mitigate the effect of negative convexity by hedging is to buy options in one form or another. Some REITs use "swaptions" (options on swaps), other sell TBAs ("to-be-announced", these are futures on MBS, that can be sold forward, and selling something that is short an option is kind of like buying an option). But most REITs do not particularly hedge this risk. This is clearly apparent in the typical table shown in quarterly reports with the impact of interest rates moves on the asset portfolio NAV. For example, in ARR's Q2 quarterly report, these numbers are shown on page 33. Note that reading the footprint is important for these tables in order to be sure that the assumptions include reflecting the effect of interest rate changes on prepayments (which, as we have seen, is very important on MBS).
The true spread of agency MBS
Typically, average yields on the assets are provided in REITs quarterly or annual reports. But as you can understand from the above discussion, they don't really mean anything, since in order to determine how much value the REIT really brings you would need to take out the cost of hedging duration and convexity.
For example, see the table on page 26 of ARR's Q2 2012 report. The asset yields are stated, along with funding costs (for repos), but there is nothing about how much it truly costs to properly hedge all the relevant interest rate related risks.
The big question this all boils down to is, what does a specific REIT bring that one would not achieve simply through a leveraged purchase of Treasuries and a leveraged sale of interest rate options (say Treasury futures options for the sake of argument)? If REITs only add a fraction of a percent of yield on top of this simple strategy then it's just as best to short a bunch of options and pray that short rates do not rise too much and the long end of the curve does not move too much (just as you ought to do if you own agency REIT stocks!).
Using complex analytic models, once can price various agency MBS, and determine the so-called "option-adjusted spread" or OAS. This is the spread brought by a mortgage bond over the swap curve, factoring in the cost of negative convexity. Essentially this technique looks at how much buying options (caps, floors, swaptions etc.) to hedge the mortgage's risks would cost, and subtracts that from the MBS's yield. See this (technical) presentation for an introduction. This OAS being calculated relative to the swap curve, it factors in the cost of hedging duration, as well as financing cost, so it really captures the net value-added of owning agency MBS, if all the interest-rate related risks were properly hedged.
Some example OAS are shown on page 2 of this Credit Suisse document. It is not clear as of when these numbers were calculated but they are typical of specified pools OASs, as you can see ranging from -10 to +35bps, that is -0.1% to 0.35%. The OAS of the current mortgage coupon (the most liquid mortgage securities) over time can be found here.
Across mortgage products (excluding inverses and mortgage derivatives) OAS is in the range of -0.1% to 0.1% for "on-the-run" MBS, that is basic paper with no prepay protection or anything particular, and up to 1% for off-the-run bonds, "story" bonds that are less liquid.
This is why it makes sense to look at REITs holding the less liquid / prepay protected bonds. Not because they prepay slower (since that is normally priced-in), but because their relative value, all risks factored in, is substantially higher.
In addition, it's important to realize that there are not large amounts of illiquid paper. Well if there was, it'd be pretty liquid and there wouldn't be any particular relative value advantage in buying it. So the large players who have to invest tens of billions are at a significant disadvantage. This is why I am very surprised to see some views expressed that it is a good thing that a REIT would grow significantly. Based on this analysis, I believe that is not true. If a REIT is so large that it has to buy generic paper, then it is likely to be constrained to purchase the more liquid paper offering flat or even negative option-adjusted spreads.
A REIT is not a static portfolio, and is actively managed. A good portfolio management team can add value through a more astute hedging strategy, and by generating trading profits. These are however not the first order contributors to the bottom line, as relative value variations in the agency MBS market is measured in basis points, hundredths of a percent. Of course, differences in management fees can be material and will also drive the bottom line by the same order of magnitude.
At the end of the day, I estimate that the "true" spread on the assets that a typical agency REIT (invested in fixed-rate MBS) can generate is in the order of 1% to 1.2%, allowing for some value added due to active trading. There is not a very large difference between most reasonably sized REITs, because at most the relative value difference between a very rich bond and a very cheap bond is less than 1%. The rest is pure bets on either the direction of rates, or their realized volatility.
Hence what truly distinguishes REITs such as ARR, AGNC, HTS or NLY to pick a few, is size (where the bigger ones typically will have more limited opportunities), and leverage. Per dollar of stock purchased, I find that ARR gets you about $10 in agency MBS at one end of the spectrum, and NLY about $6.8 at the other end of the spectrum. The "net" return on equity I would find on ARR is therefore highest, about 10%, versus NLY's around 7%, as a straight consequence of the differences in leverage.