By Andrew Davidson, President, Andrew Davidson & Co., Inc.
Andrew Davidson & Co., Inc. has developed several enhancements to its current MBS analytical capability, which could change the landscape of mortgage analysis.
The three main developments include:
1 Enabling backward induction valuation by using Active Passive Decomposition
2 A new methodology that captures the uncertainty in prepayments and allows the extension of option-adjusted methods to adjust for prepayment uncertainty.
3 Utilizing the enhanced pool data released by the GSEs to bring the tools of loan level analysis to MBS pools.
Currently, Monte Carlo analysis is the primary method used to value the prepayment options imbedded in mortgage-backed securities. However, by addressing the path dependence of mortgages, it is possible to use the much more efficient backward induction methods that other fixed-income securities employ. By applying Active-Passive decomposition, a process that splits the mortgage pool into two path-independent components, fast pay borrowers and slow pay borrowers, we may solve for path dependence issues that arise from prepayment burnout, at least for MBS passthroughs. Once path dependence is removed, it is possible to apply backward induction to mortgage valuation. The backward induction, unlike a Monte Carlo, operates on a grid with deterministic pricing nodes rather than with a large set of random paths.
The use of backward induction opens up additional possibilities for mortgage valuation. First, backward induction is substantially faster than Monte Carlo analysis, allowing for the same degree of computational accuracy in just a fraction the time. As a result, backward induction can be used to value portfolios at the loan or pool level, where Monte Carlo analysis might require evaluating only larger groups of loans or pools. Second, while Monte Carlo methods can only compute the value of a loan or pool at a single point (the starting point), backward induction methods can be used to compute the future value of a loan or pool at any rate level and at any time, present or future. The availability of these values makes a wide range of risk and other analysis possible, often without any additional processing time
One example of an analysis that takes advantage of the enhanced capabilities of backward induction is our new measure, prepayment risk-and-option-adjusted spread (prOAS, pronounced pro-A-S). prOAS looks at the impact of prepayment uncertainty beyond the level explained by interest rates at each pricing node of the valuation grid and adjusts the required return of the mortgage for the impact of that uncertainty. In this way, prOAS truly answers the oft-asked question "How do you adjust OAS for the risk that the prepayment model might be wrong?"
As it turns out, while answering this question, prOAS also explains much of the cross-sectional variability in OAS levels across TBAs with different coupons, IOs and POs. It is widely known that high coupon IOs have very high OAS levels. In fact, many analysts adjust IO OAS levels when performing scenario analysis. With prOAS, that adjustment is endogenous, meaning it is included in the modeling and does not require outside input. prOAS recognizes that higher coupon mortgages and IOs have greater prepayment risk and require additional return, rendering it possible to rapidly compute durations and convexities that account for the market pricing of prepayment uncertainty.
We've further advanced our MBS analysis by incorporating the enhanced data provided by the GSEs to extend many of the features of loan level models to our pool models. Using the enhanced data, we can estimate the relative speed of turnover and refinancing for a given pool. Features like low loan balance tend to decrease both turnover and refinancing, while features like high LTV may lead to slower refinancing, but tend to increase turnover. By segmenting the risk adjustments of these features, we provide a much better indicator of value than a single multiple or score, which attempts to combine the impact of loan features on refinancing as well as the turnover.
Combined, these enhancements open the door to new types of analysis. One such analysis that we have developed is a "pay-up" model. This model enables users to rapidly compute the relative economic value of a specific pool to the TBA. We built this model using powerful .Net technology, which allows the distribution of the results across a firm while pool databases can be centralized. Another analysis is loan level valuation of servicing. With these new tools, we anticipate that investors will have the capability to more fully utilize the data available to them.