Rob Fear , Director Client Solutions Group

Determining Haircut/Margin Requirements for Whole Loan Collateral

Wednesday, April 13th, 2011

The calculation of the appropriate haircut or margin to use when lending against mortgage whole loans that are pledged as collateral is not unlike the process involved when performing a bank “Stress Test” – the type of test that the US Treasury originally required for many of the largest US banks in 2009 that was used to determine capital adequacy.  Using this capital adequacy framework, where assets and liabilities were shocked in order to estimate the amount of capital remaining on the balance sheet under severe economic stress, analysts could determine a bit more clearly which banks may or may not survive an economic catastrophe and which ones required additional capital.  Of course, there was quite a bit of criticism surrounding these bank stress tests at the time, but that was mainly focused on the assumptions and methodologies of the tests themselves.

The same concept and methodologies of capital adequacy embedded in a stress test can be applied to haircut and margin requirements for whole loan lending, but with a clear design and specific approach. A proper stress test for whole loan collateral involves shocking each individual input assumption independently as well as shocking all parameters collectively in order to ascertain the worst or near-worst scenarios affecting the value of the whole loans. The only way to truly estimate these worst-case scenarios for whole loan collateral is by proactively calculating the effect of negative shocks on the value of the loans. 

One of the primary roles of the Federal Home Loan Bank (FHLB) system is to lend money to member banks as part of their advance business in exchange for collateral – which consists mostly of mortgage whole loans. However, doing so comes with significant inherent risks. To mitigate those risks, the FHLB advance business is dependent on applying haircuts, or margins, to the value of the collateral.  Having a robust process in place for setting the correct haircut is integral to the safety and soundness of the FHLBs.  It is well known that the FHLB advance business survived admirably during the recent credit crisis, and in fact was a key provider of liquidity to member banks during this difficult period.  However, their success can be a useful guide for other institutions because now is not the time to be complacent about risk management processes.

The objective of this article is to outline a framework and methodology for accurately quantifying the potential market risk of whole loan collateral, which can be used to set margin requirements. Any owner of a mortgage whole loan portfolio could follow a similar approach for managing market and liquidity risk.

We will first consider a risk management methodology known as Prospective Attribution Analysis.  This is the process of organizing the risk characteristics into pre-defined risk attributes, such as, prepayment risk, credit risk, and liquidity risk, among others.  We organize our scenario testing into these risk attributes and we simulate the price performance of the assets by these risk attributes into forward or prospective scenarios.   These scenarios are performed to quantify the risk of lending against whole loan collateral by determining the widest range of possible loan values through different interest rate scenarios and shocks. We’ll investigate the major sources of risk contained in whole loan portfolios and better understand the sensitivity of that value based on shocking those parameters. Then, we’ll review the Retrospective Attribution Analysis. The primary use of a Retrospective Analysis is to back-test model results to ensure those models used to project the valuation are providing accurate results and, if not, it provides one with the opportunity to re-calibrate those models.  MIAC Analytics™ software suite can be used to perform this analysis, and specifically the module known as MIAC-ALM/VAST™.

What is the worst-case scenario for any institution that is lending against whole loan collateral?  That would be a situation where one of the lender’s clients defaults for one reason or another and the secured lender takes possession of the whole loan collateral.  If this situation were to occur, the lender would have three primary alternatives at its disposal that could be used to recoup its advances.  First, it could sell the loans in the open market as quickly as possible.  This liquidation scenario would most closely resemble a distressed transaction, where the seller is highly motivated to monetize the assets and where the timeline is very short.  In this case, the seller usually has limited choices and would typically end up accepting a price that is inferior to all other options.  The second alternative would be for the lender to gradually sell the portfolio over a period of time.  In this case, the seller has the luxury of time, can break the portfolio of loans up into smaller sub-portfolios to optimize the execution, and can be more selective in the bids it accepts. This would be considered an orderly transaction. Lastly, if selling the loans outright does not appear to be an attractive option and the lender has a strong capital position overall, then the loans could be put on the balance sheet and held to maturity. Under this scenario, the lender would be the beneficiary of all future cash flows for the remaining life of the loans, but it would be liable for any realized losses as well.

Each one of the three alternatives described above would result in a different amount of proceeds ultimately recovered. The usefulness of a Prospective Analysis is that it can help the FHLB estimate the outcome for all three options.

Attribution Analysis

Let’s define what a Prospective Analysis and Retrospective Analysis entail. First, it is important to define a time period within which one performs these analyses. Let’s assume that the lender typically values its whole loan collateral on a monthly frequency, so in this case it would make sense to use a monthly period for the analysis, using two measurement dates. The first measurement date would be considered ‘Month 0’, and the second would be ‘Month 1’ for comparison. For the Prospective Analysis, one would begin with the base pricing results as of Period 0 and then measure the price sensitivity at that point of time. Using the model, one would shock all of the key individual baseline assumptions in isolation and then also in combination.

What is the proper magnitude of each shock contained in the Prospective Analysis? In order to apply realistic and meaningful results, it is critical to measure the historical volatility of each input assumption over a reasonable time period.  The magnitude of the shocks should be based on an analysis of the historical ranges and should also exceed those observed ranges. The resulting analysis would represent the price sensitivity of the whole loan portfolio based on possible events that would affect the valuation assumptions, and the data generated from such an analysis would be extremely valuable for setting appropriate haircuts.

Now, let’s turn to the Retrospective Analysis. At the next measurement date, the investor would re-price the portfolio using all updated loan data and current valuation assumptions, which would represent another mark-to-market, but as of Month 1 in this case.  The next step is to quantify the impact on the change in price due to each assumption change in isolation, from Month 0 to Month 1. For example, if we know that interest rates (or a benchmark rate) rose by 10 bps over that period, then we would revert back to the portfolio at Month 0 and only increase rates by 10 bps, but leave all other assumptions the same. This would quantify the effect of the change in price only attributable to the increase in rates. After this same process is completed for all key assumptions that changed over the period, one would then be able verify that the actual change in value attributed to each assumption change is consistent with what was projected in the Prospective Analysis. The results of this analysis would be a wealth of information that would allow re-calibration of models to improve model predictability.

Clearly, the validity of prospective shocks is highly dependent on how one constructs the framework of the analysis. Ideally, one would identify all of the key variables that have a material impact on the value of the whole loan collateral. We have identified six key input assumptions that I will discuss later on, but for each of these parameters one would need to assess their historical trends and then design shocks based on the standard deviation of each parameter over a reasonable, historical time frame. With common statistical measures, such as standard deviation, one could apply shocks for each parameter that may be based on units of standard or ‘sigma’ (σ). The shocks could represent a 1σ, 2σ, or 3σ change from the baseline assumption, for example.

The other fundamental consideration in constructing the framework is the specific behavior models selected, since the foundation of the entire analysis is based on the accuracy of the projections and therefore it is critical to choose the right ones. There are many different sources for collateral behavior projections and practically any of the commercially available products can be used in the MIAC Analytics software suite. It is important to understand, however, that each collateral behavior model will result in different terminal/horizon pricing and cash flow projections. The selection of any collateral behavior models should be based on a thorough understanding of the impact and viability of the results of each chosen model.

Only after the above steps are completed would one be ready to perform an accurate attribution analysis. In contrast, considering the historical price sensitivity of the pledged whole loan collateral alone is inadequate for the purpose of this analysis, since the collateral attributes change over time as new loans are continuously added to replace delinquent or paid-off loans. In addition, because of the diverse composition of the loan pools, their prices will be difficult to accurately compare at the portfolio level in a meaningful way. Quantifying the direct price impact of the actual whole loan collateral pool due to changes in the particular assumptions themselves is critical to building a useful analysis.

Identifying Risk Factors – Prospective Analysis

Liquidity

Liquidity essentially represents the ease with which one can sell loans in the marketplace. The primary assumption used to reflect relative liquidity when pricing loans is the discount rate or yield applied when computing the present value of the cash flows. A lack of liquidity for a certain product would be represented by a higher yield, whereas more liquid products would be assigned a lower yield. Some mortgage products are generally more liquid than others and the mortgage loan market as a whole does experience various levels of liquidity over time, so assessing the impact to the price due to changes in required yield is important when quantifying the risk within your pledged collateral.

Liquidity generally deteriorates when a whole loan pool underperforms, becomes ‘scratch & dent’ for various reasons, is a distressed sale situation, or when investor appetite subsides for a particular product.  Also, if the loan pool must be sold very quickly on an aggressive time schedule, then this could possibly impact the buy-side yield requirements.  Generally speaking, more uncertainty about the outcome of the cash flows would decrease liquidity overall. Due to this concept, oddly enough, today we see that non-performing loans (NPLs) have some of the best liquidity in the marketplace and are trading more frequently than any other type of pool. This is because NPLs are usually far enough along in the foreclosure process and therefore closer to liquidation and that cash flow happens to be more predictable than that of a current loan.

Defaults

No discussion about the risks of mortgage loans would be complete without measuring the Probability of Default (POD) and how it impacts the value of a whole loan pool. POD is a function of borrower credit grade (FICO), occupancy status of the property, loan documentation type, the amount of equity the borrower has in the home, and macroeconomic factors such as unemployment, among others. Default probability can increase if the borrower’s credit situation deteriorates, but as the lender lending against this collateral, it would be difficult to know unless it received an updated credit score beyond the one obtained at origination. Enriched data such as updated FICO scores and property values provided by a third-party source could materially improve your predictive modeling capabilities surrounding defaults.

While it would be nearly impossible to shock each unique driver of default probability in a Prospective Analysis, it is quite possible to shock the projected default rate itself, thereby emulating the effect of any of the underlying causes on default behavior.  While a higher default rate usually results in a lower value, it could have a positive impact on the price for loans already priced/purchased at a discount that have very low loss severities.  In this case, the default and liquidation is more like an early repayment and would therefore return nearly all the principal at an earlier date than originally anticipated.

Principal Loss

The principal lost by the investor in the event of a mortgage loan default and ultimate liquidation of the property is commonly known as loss severity. The current, combined mark-to-market loan-to-value ratio (MTM CLTV) and also the loan size itself are primary factors that drive actual loss severities and also loss severity estimates when pricing a pool of these loans. The MTM CLTV is a measure that reflects the underlying value of the property itself relative to the balance of the loan. Severities rise dramatically when the property value declines, especially when the value of the property becomes less than the loan amount.  In the absence of a formal property appraisal, the updated MTM CLTV could be estimated using a home price index (HPI), a broker price opinion (BPO), or an automated valuation model (AVM), all of which could be supplied by a third party and used in the pricing model.

For the purpose of quantifying the risk of owning mortgage whole loans, it is critical to shock the loss severity estimates, since there are several reasons that severities could be higher than originally anticipated.  The actual sale price of the underlying property in the event of liquidation could be less than modeled due to a multitude of reasons including local economic factors, such as mass layoffs by a local employer. In addition, severities could also increase in the near future due to a flood of supply of foreclosed homes on the market.  Since the investor will never really know the extent of the loss until the property is sold, it is therefore critical to prospectively estimate the value of the collateral using progressively higher or lower levels of severity.

Principal Recovery Timing

The impact on the value of a pool of loans due to foreclosure and liquidation timelines could simply be defined as the time value of money based on an estimate of when the property sale will actually occur.  After a default occurs, the expected date when the proceeds are ultimately received by the investor from a property disposition will directly affect the value of that pool.  But if the timeline changes for any reason, then the present value of the cash flows will change and the value of the loan pool could be higher or lower than expected.  For instance, timelines could be extended due to foreclosure moratoriums, servicer capacity constraints, aggressive state Attorneys General, legal backlogs or delays in court proceedings, or an over-supply of homes on the market, which could all reduce the value of that cash flow.

While the baseline pricing should include a geographic designation of foreclosure and liquidation timelines to reflect the differences between judicial and non-judicial states, the Prospective Analysis itself should attempt to quantify the impact of shortening or possibly lengthening these base timelines in order to gauge their impact on the value of the whole loan collateral.

Prepayments

The assumption that has the most significant influence on the value of any mortgage asset is the prepayment speed.  While prepayments could affect different loans in different ways, the fundamental concept is that slower prepayments extend the life of the asset and reduce the value of the whole loan collateral, whereas faster prepayment speeds result in a quicker return of the principal.  That in itself is a good thing; however, in a decreasing rate environment when prepayments tend to increase, one ends up with re-investment risk, because it would then be more difficult to replace that loan with another asset that provides the same yield. Regardless, faster prepayments are bad for premium coupons and good for discounts.  Prepayments also change the composition of the overall pool of loans, as the higher note rate loans would typically prepay first, leaving the lower note rate loans in the pool and lowering the coupon.  Also, since prepayments directly reduce the principal balance of the pool, a slower prepayment speed may increase overall cumulative losses because there would be more loans in the pool for a longer period that could potentially default over the life of the asset.

A Prospective Analysis would need to include several scenarios with faster and slower prepayment speeds in order to properly quantify the potential change in value of the whole loan collateral and assist in setting the haircut.

Interest Rate Environment

The final parameter that can have a significant impact on the value of pledged collateral is the interest rate environment itself, and one way to assess this risk is to perform a parallel shift of the entire yield curve in specific increments.  To do this in a Prospective Analysis, one would move all benchmark rates in the Libor/Swap and Treasury yield curves, as well as all mortgage rates and any other input rate that affects the value of the loan in the model. The result would quantify the combined impact of prepayment speeds, ARM resets, and the adjusted discount rates (yields) for each parallel shock.  A simple interest rate shock does not capture all components of risk – only interest rate risk, of course, but similar interest rate processes could also be used to enhance this type of analysis such as Value at Risk (VaR).  In addition, the investor might want to perform a non-parallel yield curve shift that includes flattening and steepening scenarios.  Lastly, if the pricing model is using a monte-carlo process, then performing interest rate volatility shocks would also be beneficial. A Key Volatility Analysis (KVA) would involve shocking discrete points on the volatility surface to calculate the sensitivity to those changes. Performing interest rate sensitivity should be a regular process for the lenders to estimate the value of the pledged collateral through various interest rate scenarios.

Once the Prospective Analysis for all six of the risk factors outlined above has been completed, one final step should be performed. All of the individual shocks should be combined in such a way that all positive factors, and then all negative factors, are adjusted in unison, which would provide an exhaustive and fairly definitive range of the absolute best and absolute worst cases for the value of the collateral, assuming shocks have been appropriately defined at the beginning of the process. (Figure 1 illustrates the combined best and worst-case results of this sensitivity analysis for a hypothetical portfolio.)  Finally, at this point, one would be fully prepared to summarize all of the results of the Prospective Analysis and leverage this information to refine the haircuts that should be applied to the value of the pledged collateral.

Retrospective Attribution

Let’s now change gears and move to the Retrospective Analysis. Recall that the Prospective Analysis is performed at Month 0 and is intended to project the value of the pledged collateral as market conditions change. On the other hand, the Retrospective Analysis is performed at Month 1 and looks back at what actually transpired over that time period for the purpose of back-testing the ability of the pricing model to predict portfolio value changes accurately. Both of these attribution analyses need to be performed on a regular basis in order to be effective risk management tools.

In the interest of simplicity, let us examine how prepayment speeds alone would be addressed in the Retrospective Analysis. Assume the value of a portfolio of loans in Month 0 is 79.87 points with a conditional prepayment rate (CPR) of 34.74%. Now, when the portfolio is re-priced in Month 1, the portfolio price decreases to 79.69 points and the CPR is now 33.88%.  This means that over this single time period the CPR dropped by 2.47% based on all of the changes in market conditions, portfolio attributes, and assumptions. The next step would be to go back to the portfolio in Month 0 and decrease the CPR by precisely 2.47% and calculate the change in price, which in this example ends up being 41 basis points.  That means if one didn’t change anything else, but only decreased the CPR by the actual amount that it changed over that period, then the value of the portfolio would fall by 41 bps. Now this result can be compared to the Prospective Analysis to see how accurate it was. We know from our Prospective Analysis that if we slowed speeds by 10% the price would fall by 160 points.  Through linear interpolation, we can calculate the prospective price change for a 2.47% decrease in CPR, which equals 39.5 basis points.  Now we can conclude that the difference between the projected price change and the actual price change, isolated to the change in prepayment speeds, was only 1.5 basis points.

If the difference were materially larger, then the proper course of action would be to understand why the prepayment model was not predictive and then possibly modify the parameters in the prepayment model itself in order to improve the predictability. If one repeats this process for each and every risk factor and also all risk factors in combination, the precision of the prospective risk metrics that are generated and used to set suitable haircuts will dramatically improve.

In conclusion, it should be recognized that adopting a Stress Test approach for setting margin requirements has its strengths compared to simply using historical price sensitivity.  Emphasis should be placed not on observed price changes by simply analyzing a historical time series, but instead on determining in a justifiable manner the shocks that might occur and then applying these shocks to individual risk parameters to calculate realistic outcomes based on potentially catastrophic events. Prospective Analysis should become an integral part of a haircut methodology. In addition, one should use the Retrospective Analysis as an opportunity to refine the behavior models and improve the accuracy of the Prospective Analysis.  By performing a regular Retrospective Analysis, one can continually back-test the predictive behavior of the prospective shocks.  Ultimately, if one is responsible for ensuring the lenders do not suffer a loss, then adopting a more sophisticated approach for defining that risk and setting margin requirements can only help one reach that goal in the long run.

 

 [BACK TO MAIN PAGE]

Add a Comment

We welcome your thoughtful comments. Please comply with our Community Rules. All comments will show your real name. There maybe a time lag before your comment appears.

You must be logged in to Add a Comment. Click here to Login or Sign Up.