The main idea is to construct an accurate representation of each REIT form the lowest level possible, and in such a way that all the moving parts can be realistically modeled.
The graphic below illustrates how the various inputs feed into the construction of an agency REIT's model. The driving principle is that nothing should be considered a given, and everything should be computed from atomic market quantities (like interest rates, certain mortgage prices, etc). For hybrid REITs the principle is the same but there are more moving parts.
This approach is composed of four parts:
- The rates and derivatives inputs are a very basic building block. They drive the valuation of interest rates derivatives generally used as hedges by mortgage REITs, as well as the value of mortgage securities.
- Then, prepayment behavior is important, because it determines how some MBS might withstand moves in interest rates better than others. For example, credit impaired borrowers who have been through a Government refi program would have less opportunities to refinance than the borrower behind a streamlined loan; as a result in case of a drop in interest rates and refinancing wave, an agency MBS backed by credit impaired borrowers would likely outperform more standard pools. Both interest rates models and prepayment models are used as essential inputs to mortgage valuation models
- MBS price inputs are naturally important. It is not possible to find market prices on every single MBS pool in a systematic manner, so we use our pricing models in order to price all the ones for which we have market prices, and then "calibrate" the models. Once the models perfectly price those products whose price we already know, we can use them in order to price other bonds, for which no price is known.
- Public information from each REIT, in the form of 10Qs, 10Ks or presentations, are an essential part in the mix. Any given REIT holds thousands of MBS pools, and they never disclose which ones exactly. So pricing each and every one of them is of course impossible. We can however infer significant amounts of information from the characteristics they make public. Our optimization process automatically figures out the closest representation of a REIT's portfolio one could make using our more generic MBS bonds.
With these four "atomic" data sources, we model how the value of a mortgage REIT is changing over time, and we can naturally project changes in book values.
Book value projections
Projecting book values is useful in and by itself, as it can sometimes prevent large surprises when the numbers are released. But above all, it serves as an on-going test of the quality of our models for representing mortgage REIT positions. The chart below shows the simple average of projected and actual book values across the mREITs we are tracking. The projections are out of sample (only using information that was available before the numbers were released).
We have generally found that the standard deviation of our errors is slightly under 2%. Our approach is very quantitative, and we try very hard to take out any human judgement or interpretation, so that our models are minimally dependent on opinions. This is why we are comfortable looking at our projection errors in a statistical fashion.
With our realistic and fully functional representation of mortgage REITs, we can also explain how any change to the inputs affects book value at the end, and measure any REIT's risk exposure.
Thanks to the approach and logic explained above, we can stress the various parts and see what happens.
For example, a move in interest rates would affect a REIT at different levels. Among other things:
- The market price of MBS would move
- The value of fixed-income derivatives such as swaps or swaptions would move as well
- The cash flows from derivatives would change (for example as a function of the new level of Libor)
- The cash flows of MBS would change, either directly (for example ARMs resetting to a new index), or through the effect of faster or slower prepayments
- The financing cost for the REIT would change, since repo rates tend to be pegged off of short rates.
All these effects would be summed up inside the REIT, and result in a net book value impact. Generally speaking, increases in interest rates will lead to a decline in MBS values, and an increase in derivatives values. The net effect can therefore be positive or negative, it depends. Therefore it's important to have an accurate representation of the small moving parts and of their interactions.
The following table shows an example of stress scenarios applied to CYS as of the end of Q2 2013. Each scenario capture a simple stress, for example higher or lower rates, or wider option adjusted spreads and higher mortgage rates.
|Stress||Projected change in BV (%)|
|Mtg rates +35 OAS +50||-48|
|Mtg rates m35 OAS m50||0|
|Non agency OAS +100 Mtg rates +35||-22|
|Parallel move down 50||-5|
|Parallel move up 50||-44|
|Vol up 20||-30|
Over a given quarter, a mortgage REIT's management typically makes some trading decisions, for example increasing leverage or reducing overall duration. The strategy followed by management will affect the end book value. We can never know precisely what strategy a REIT's management will follow over a given quarter, so we model a range of possible strategies in order to see the impact of the strategy choice on the quarter end book value. We assume that the strategies are rebalanced monthly. Here are some examples of the strategies we model:
- Basic reinvestment (all income is used to buy more of the assets)
- Flat leverage (assets are purchased or sold in order to maintain a constant leverage ratio)
- Leverage decreased by 1 (a new, lower, target leverage is set and aimed for at each rebalancing)
- Flat duration (the total assets and hedges duration is held flat by adjusting the amount of derivatives held)
Comparable strategies can have very different effects on different REITs, and also depending on the evolution of the underlying assets' values. It is also quite informative to see whether some REITs effectively have more exposure than others to the type of strategy followed.
The following table shows an example of strategies applied to CYS (the same as in the table above) as of the end of Q2 2013.
|Strategy||Projected change in BV (%)|
|Flat leverage and 0 duration||-30|
|Flat leverage and 1 duration||-28|
|Flat leverage and duration||-23|
|Flat leverage and duration down 1||-26|
|Leverage down 1||-19|
Atomic analytics examples
We track and model hundreds of separate mortgage securities and fixed-income derivatives in order to accurately represent the many moving parts inside each mREIT. However, there are a few specific products that capture the core trends in the MBS market, and we show some analytics on these products below.
The option adjusted spread is a good measure of the return afforded by MBS once their interest rate and convexity risks have been hedged.
The charts below shows current OAS levels, as well as the corresponding levels at the beginning of the quarter, and then historical MBS prices, OASs and durations.
TBA prices from Trade Reporting and Compliance Engine (TRACE). All calculations and models from Amerigo Capital Analytics.