Please ensure Javascript is enabled for purposes of website accessibility Forecasting MBS Prepayments Using Daily Prepayment Data from the GSEs
top of page
  • LinkedIn
  • Facebook
Writer's pictureSharad Chaudhary

Forecasting MBS Prepayments Using Daily Prepayment Data from the GSEs

Introduction

Payoff data on MBS cohorts that is accumulated for each day of the servicer reporting cycle associated with a factor date provide a valuable and increasingly accurate estimate of expected prepayments for that month. The GSEs have been publicly providing this level of transparency starting March 2023 by publishing a daily prepayment report (DPR for short) that provides data on full voluntary payoffs leading into the upcoming monthly factor date (4th Business Day/BD4) for a selection of cohorts. Full voluntary prepayments correspond to the borrower repaying the loan in full prior to the loan’s maturity because they either sold the mortgaged property, refinanced the loan, or just decided to retire their debt. Partial prepayments (curtailments) and buyouts (involuntary prepayments) are currently not tracked in the DPR report.

 

DPR is published on a weekly basis each Wednesday (at 4:30pm ET)[1] and spans each day in the previous week (defined here to be Saturday-Friday). The most recent report is cumulative for the current month and replaces the prior week’s report until the final report for the factor month, which is of course inclusive of all voluntary payoff activity (as described above) for that month. DPR aligns with the monthly servicer reporting cycle that begins with the calendar day following the second business day of the month (BD2 + CD1) and continues through the second business day of the following month (BD2).

 

With respect to cohort coverage, the report includes all cohorts with an aggregated outstanding unpaid principal balance of $500mm or more as of the most recently disclosed factor information.  Cohorts are grouped by security type, coupon, and issuance year.[2] Security types include: UMBS 30-years, UMBS 20-years, UMBS 15-years, Jumbos, and ARMs.

 

The following securities are excluded from the report: (a) Securities issued in the current reporting month (b) Multifamily securities; (c) ARM securities issued prior to 2006; (d) Level 2 securities; and (e) Pseudopools.[3]

 

Patterns in DPR Data

To analyze patterns in DPR data by factor month we can rely on an extensive historical data set released by Freddie Mac that contains daily prepayment information for all applicable cohorts for factor months from 2019-02 to 2023-03, plus the weekly reports issued by both GSEs beginning with the 2023-04 factor month.

 

Some noticeable trends that can be extracted from this data:

 

  • Day-of-week effect: Payoffs tend to be concentrated towards the middle of the week, Tuesday-Thursday. Interestingly, both GSEs report non-zero payoffs on Saturday, and both show no payoffs on Sunday.

  • Day-of-the-month effect: Payoffs tend to be somewhat backloaded towards the end of the reporting cycle.

  • Holiday effect: Only Freddie Mac shows non-zero payoffs on Federal holidays; these payoffs are generally below-trend after controlling for day-of-week and day-of-month effects.

  • GSE effect: Apart from the difference in payoff behavior on holidays, the GSEs display individual variation in the relative strength of their respective day-of-the-week and day-of-the-month effects.

 

A substantial portion of these witnessed patterns can most likely be attributed to the GSE processes that underpin how servicers are required to process and report loan-level transactions over the accounting cycle tied to each factor month. This month-long accounting cycle is very closely tied to but distinct from the DPR reporting cycle because of lags between payment collection, reporting, and GSE processing. In what follows, we will discuss Freddie Mac’s accounting cycle, Fannie Mae’s is broadly similar.

 

Freddie Mac’s accounting cycle is a one-month period which begins on the first of each month and ends on the last day of the month. In each such cycle, servicers need to report “activity” for every mortgage they service to Freddie Mac. Activities come in two flavors: P&I payments and exceptions. P&I payments correspond to the regularly scheduled monthly payments made according to the terms and conditions of the security instrument executed by the borrower. Exception activity is any deviation from the regular schedule of P&I payments that changes the status of the mortgage such as a full payoff, reinstatement, a 3rd-party foreclosure sale etc.

 

There are two reporting options for loan-level transactions: Daily Reporting and All-in-Reporting. Daily reporting occurs when loan-level transactions are reported every day as borrower payment activity for different mortgages is received through the cycle. For example, Jill’s mortgage payment received on July 1 is reported on the same day.  An individual loan is reported only once in the cycle unless additional payment activity is received. Continuing the example, if Jill pays off her loan on July 12th, then the servicer could again use Daily Reporting to submit a revision by summarizing both the payment and the payoff into one transaction. In general, daily reporting is encouraged but not required.

 

All servicers need to submit at least one loan-level transaction for every loan in their portfolio, even if the borrower has not made a payment prior to 2:00AM EST, by the day following the P&I Determination Date.[4]  The "P&I Determination Date" is the 15th calendar day of the month.[5] This bulk reporting where a loan-level transaction is reported for every loan in the portfolio on or by the P&I Determination Date is referred to as All-in-Reporting. Revisions are accepted daily from the first business day of the current month to the first business of the following one.[6]

 

To submit daily loan-level activity to Freddie Mac, servicers can use Freddie Mac’s Loan Level Reporting (LLR) tool, their own proprietary system, or use a service bureau to submit reports on their behalf. LLR has a very specific calendar of processing times:

 

  • Payoffs are processed 5 times a day starting 4:00am and ending 11:00pm on Monday through Saturday, including Federal holidays.

  • Certain days of the month skip one or more (but not all) of the five processing runs.

  • There are no processing runs on Sundays.

 

This setup explains why payoffs are processed on Federal holidays and Saturdays by Freddie Mac even though they are not business days; presumably there is some spillover of payoffs that were not processed on the previous day to the LLR tool.

 

Inter-GSE variation in the strength of the various effects can also likely be attributed to their differing use of remittance types. Fannie Mae largely uses Scheduled/Scheduled remittance with some use of Actual/Actual (for Cash Window loans) and Scheduled/Actual. On the other hand, Freddie Mac offers only Scheduled/Actual remittance.[7] 

 

Forecasting VPR

The framework for forecasting the month-end VPR for a cohort using DPR data can be outlined as follows:

 

  • Estimate day-of-week, day-of-month, and holiday effects for each GSE from historical data

  • For each DPR report associated to a factor month, use the accumulated payoff data on a cohort and the effects estimated in the previous step to extract a normalized trend as of the report date.

  • Apply the trend to the remaining fraction of the reporting period after applying day-of-week, day-of-month, and holiday effects to generate a “de-normalized” forecast for each day.

  • Stitch together the realized and projected payoff data to get the final month-end VPR forecast for the cohort.

 

Forecasting Factor Month Prepayments

To forecast pool- (or cohort-) level prepayments for any factor month we need an estimate of voluntary (both full and partial) and involuntary prepayments (i.e., buyouts). Thus, the VPR prediction we can extract from the DPR report needs to be supplemented with a forecast of curtailment and buyout rates for the cohort in question. These forecasts are taken to be the curtailment[8] and buyout rates realized in the previous factor month.

 

Let’s work through a concrete example for amalgamating these pieces of information to construct a forecast for Freddie Mac UMBS 30-year 6.5s 2023 in the October 2024 factor month, as of the October 2nd DPR report for that month:[9] · 


  • Projected prepayment rate = 24.7% CPR

    • Projected full VPR = 23.0% CPR. The cumulative VPR based on 20 reporting days (the cycle-length for October 2024 is 24 days) is 18.2% CPR. A naïve extrapolation to the rest of cycle gives us a forecast of 22.7%; extracting a normalized trend and applying it to the remainder of the reporting cycle yields 23.0% CPR.

    • Projected Curtailment = 1.3% CPR (estimated September 2024 pool-level curtailment rate for the cohort) 

    • Projected buyout rate = 0.50% CPR (realized September 2024 buyout rate for the cohort)

  • Actual prepayment rate reported on October 4th, 2024 (4BD) = 25.2% CPR

    • Actual full VPR = 23.8% CPR

    • Actual Curtailment (loan-level) = 1.1% CPR

    • Actual CBR = 0.3%


Table 1. Forecasting October 2024 Factors for FHL UMBS 30-year 6.5s 2023 (DPR Report Date: October 2nd, 2024)


ACTUAL

PRED

DIFF

CPR

25.2

24.7

0.5

Full VPR

23.8

23.0

0.8

Curtailment

1.1

1.3

-0.2

CBR

0.3

0.5

-0.2


Notes

[1] Or the next business day if Wednesday is not a business day.

[2] The two GSEs have different conventions for calculating issuance/vintage year: Freddie Mac defines the issuance year as the current factor date minus the weighted-average loan age (WALA), while Fannie Mae uses at-issue data: issuance date minus WALA-at-issuance.

[3] Pseudopools are groups of loans with similar payment characteristics backing excess servicing securities.

[4] If a payment is not received by the P&I Determination Date, then $0 principal and forecasted scheduled monthly interest is reported since Freddie Mac requires Scheduled/Actual remittance.

[5] Or the next business day, if the 15th calendar day falls on a weekend or a holiday.

[6] This results in an overlapping accounting cycle since we process loan-level transactions for two different accounting cycles on the first business day of every month.

[7] Once a borrower’s monthly mortgage payments are submitted (or not) to the servicer, the different remittance types specify the timing and amount of these payments that are due to the GSEs. At a big-picture level, Actual/Actual requires a servicer to submit the actual P&I received; Scheduled/Scheduled requires remitting scheduled P&I regardless of whether a payment is collected from the borrower; and Scheduled/Actual requires remitting scheduled interest (regardless of actual collection) and actual principal. The treatment of float and prepayments varies across the different remittance types.  

[8] The GSEs do not report curtailment rates at the pool-level; we could use loan-level data to estimate the previous month’s curtailment rate, but adopt a different approach. Since we know the cohort prepayment and buyout rate once factors are released, and the month-end VPR from the DPR report, we can directly derive the curtailment rate for the cohort.

[9] Note that the numbers below don’t add up precisely because of the non-linearity of the SMM to CPR conversion.


182 views0 comments

Recent Posts

See All
bottom of page