The MBS market has long recognized the valuable prepayment characteristics associated with certain loan and borrower attributes such as loan size, Alternative-A documentation, and New York geography. In 1997 the practice of harvesting these loans for separate pooling took on new dimensions as the universe of loans with small balances was demonstrated to have better convexity characteristics than typical TBA pools. Today loan filtering takes place at nearly every stage of the pooling and delivery process as originators, dealers and the agencies identify, segment and extract a premium for (or place in portfolio) loans with the best prepayment characteristics. Indeed, as a first step in the pooling process, originators and dealers now routinely screen raw loan files for a range of characteristics that command a premium to TBA. Even at the pool level the market has become much more discriminating with respect to WAC and seasoning ("low WAC" pools will be addressed at the end of this report).
In the context of the traditional TBA market, this adverse prepayment selection increases the probability that a TBA investor (who by definition is at the end of the "food chain") will be delivered a pool that is more negatively convex than the aggregate universe of pools with similar age and weighted average coupon (WAC). This fact has been highlighted by the current refinancing event where large-scale filtering has inflated true TBA prepayments relative to the universe. This presents TBA investors with two fundamental problems when assessing the prepayment risk of TBApools. First, actual prepayments for the true TBA sector are masked by agency prepayment reports that reflect the entire universe of MBS in a given coupon/vintage bucket. Second, since prepayment models are estimated on aggregate data rather than true TBA data, there is the potential for significant modeling error when using these models to forecast prepayments for TBAs. We will address both of these concerns in this report.
While the concept of adverse loan selection is well understood by the mortgage market, our ability to measure its effect on the TBA universe is limited by our access to loan level information. For example, while we can segment pools by average loan size, issuer, and geography, other common filtering criteria such as LTV and CRA eligibility are not available.
Nevertheless, we can still create a reasonable proxy for the true TBA universe by extracting pools conditional on the following three criteria: loan size, Alternative-A issuers (Greenpoint, IndyMac and DLJ) and New York geography. We note that this analysis probably understates the full effects of adverse selection since our criteria are a subset of the full characteristics used to screen loans.
By applying these criteria we filter out approximately 10% of the total TBA universe across the coupon spectrum as shown in the table below.
We can then compare recent 1-month speeds of the universe to the true TBA coupon speeds that are net of LLB, New York, and Alternative-A pools. Note that the percentage of MBS excluded from the true TBA sector as well as speed differentials increase as we move up the coupon stack. This can be attributed to the heavy concentration of Alternative-A collateral in coupons above 8.0%. Based on this analysis the true TBA sector is currently prepaying anywhere from 1 to 8 CPR faster than the reported universe.
How do we value and monitor the TBA sector?
If we calibrate our prepayment model to the empirical we can quantify the effects of recent adverse loan selection on the TBA market. We reiterate that traditional agency prepayment models do not capture the full consequences of adverse selection since these models are estimated on aggregate prepayment data.
The greater negative convexity of the true TBA market lowers the implied TBA prices (new production 6.0% to 8.5% coupons) by between 0:01 and 0:10. In the benchmark Trust IO sector (new production 7.0% to 8.0% Trusts) implied prices decline between 0:18 and 0:23+.
Finally, to address concerns about monitoring and valuing the effects of adverse prepayment selection in the TBA market, Bear Stearns will provide two new tools to its MBS customers: Beginning with next month's agency prepayment report, along with our standard reports, we will provide Bear Stearns MBS customers with a true TBA report that will be representative of the "cheapest to deliver" pools in the TBA market. The true TBA pool sample will be drawn from pools backing recent IO Trusts.
Beginning April 16, we will make available a true TBA prepayment model that is consistent with actual prepayment observations in the current refinancing wave.
Why are pay-ups not justified on low WAC pools?
One obvious omission from our filtering analysis is "low WAC" pools. Over the past year, investors have begun to assign increased value to agency pools that have very low gross WAC relative to the pool's net coupon. By convention, a pool qualifies for "low WAC" status if its gross WAC is no more than 0.375% higher than its net coupon. The market pay-up for these low WAC pools has been in the range of 2 to 4 ticks since last year, depending on coupon. Is there any value in this most recently defined subset of the TBA universe? After examining the data, we find that the low WAC story is another example of pool adverse selection that works to the detriment of the investor both on a practical and a theoretical basis.
Investors will benefit from low gross WAC pools only if other pool characteristics do not change. However, as seen in the table to the right, low WAC pools have consistently (and in many cases significantly) higher average loan sizes than pools with higher gross WAC. The reason? We believe smart originators are more likely to sell large balance, prepayment sensitive loans into high net coupon (low WAC) pools since they retain less servicing from these loans in this transaction. Conversely, servicers are more likely to concentrate low balance, less prepay sensitive loans in pools where they retain lots of servicing (high WAC pools).
While the lower gross WAC can delay the time when a pool reaches the refinancing trigger point in a declining rate environment (such as today's), low WAC pool speeds actually exceed higher WAC pool speeds once they are refinanceable. For example, low WAC FHLMC 8.0s (2030 WAM) paid 62.1 CPR in March, versus 58.4 CPR for the remainder of that cohort. Much of this speed difference can be attributed to the larger average balance of low WAC pools. Moreover, we note that the difference in average loan size is generally larger in the lower coupons, and particularly large in FNMA pools ranging from $11,000 to $39,000.
We believe that a correct valuation of low WAC agency pools must include the wide divergence in loan size, since the impact of loan size on prepayments has been conclusively demonstrated in all sectors of the mortgage market. Accordingly, when we calibrate our agency prepayment model to take into account loan size differences, we can use it to value low WAC pools (using empirical WAC and average loan size from existing FHLMC and FNMA pools).
Using market TBA pricing and the adjusted model, we find that the market is clearly demanding more of a payup than can be jus-tified on a theoretical basis for 7.5% and 8.0% pools with low WAC. Indeed, the bias toward higher average loan size completely overwhelms the potential advantage of lower WAC in the FNMA 7.5% coupon, such that the theoretical value of these pools should actually be less than regular TBA. Low WAC 8.0% pools from both agencies have a lower theoretical value than regular TBA. In the 7.0% coupon, FNMA low WAC pools are also clearly overvalued (again, the impact of much larger average loan size, as seen in Table 5), while FHLMC low WAC pools have a market value that is roughly 50% of theoretical - the only area where there is even marginal theoretical value in the sector. We believe that these results do not support the current pay-ups in the low WAC sector, regardless of coupon and agency.