Get Started
Blog Archives

How Are Ginnie’s New RG Pools Performing?

In February of this year, the Ginnie Mae II program began guaranteeing securities backed by pools of mortgages previously bought out of Ginnie Mae securities because of delinquency. In order to qualify for these new re-performing pools (known as “RG pools”) a loan must meet two (related) conditions: 

  • Borrower has made at least six months of timely payments prior to pool issuance. 
  • Pool issue date is at least 210 days from when the mortgage was last delinquent. 

The novelty of RG pools raises questions about their composition and performance relative to other Ginnie Mae pools. While it remains too early to make many conclusive statements, a preliminary look at the prepayment data indicates speeds somewhere between those of similar vintage Ginnie Mae multi and custom pools, with typical variability from servicer to servicer.  

In this post, we discuss the prepayment behaviors we have observed over the first seven months of RG pool securitization, issuance patterns, and collateral characteristics. 

Prepayments 

Latest September prepayment prints show that RG pools’ speeds generally fell in between those of similar coupon/vintage multi and custom pools.  Below charts shows that 2015/2016 3.5% RG pools prepaid at around 37-38 CPR in September, a couple of CPR slower than similarly aged multi pools and almost 10 CPR faster than custom pools.  


Prepayments for G2 3.5% RG, Custom and Multi Pools by Vintages, September Factor Month Note: Loan level data


Below, we plot S-curves for 49 to 72 wala RG loans against S-curves for similarly aged multi and other custom loans from April to September factor months Speeds for RG loans with 25 to 100 bp of rate incentives have prepaid in mid-30s CPRs (Green line in below figure).  During the same period, similar multi pools have prepaid 5 to 8 CPR faster (blue line) than RG pools while similar custom pools have prepaid around 5 CPR slower (black line) We also overlaid a s-curve for 7 to 18 wala G2 multi pools as a comparison (orange line).


S-curves for RG, Custom and Multi Pools (49 to 72 WALA) April to September Factor Months 
Note: Loan level data, orange line is the s-curve for 7-18 wala G2 multi pools with one-year lookback period 


Not surprisingly, prepayment behavior differs by servicer. Wells-serviced RG pools that are seasoned 49 to 72 months with 25 to 100 bp of rate incentives appear to be prepaying in low 30s CPRs (black line in below figure).  Similar loans from Penny Mac are prepaying 5 to 10 CPR faster, which tends to be the case for non-RG loans as well. 


S-curves for RG loans by servicers, 49 to 72 WALA, April to September Factor MonthsNote: Loan level data 


While the re-performing loans that are being securitized into RG pools are already seasoned loans, prepayments have been increasing as pool seasons.  For example, one-month old RG 3.5% pools have prepaid at 27 CPR while 6- and 7-month 3.5% pools prepaid at 45-50 CPR (black line below). In addition, overall prepayment speeds for same-pool-age 3.0%, 3.5%, and 4.0% have been on top of each other. 


 Prepayments for RG 3.0%, 3.5% and 4.0% Pools by Pool Age, March to September 2021 Note: only showing data points for cohorts with more than 50 loans


Issuance Volume 

Following a brief ramp-up period in February and March, issuance of RG pools has averaged around $2 billion (and roughly 300 pools) per month for the past five months (see Issuance chart below). The outstanding UPB of these pools stands at nearly $11 billion as of the September factor month. 


Note: RiskSpan uses reporting month as a factor month. For this chart, we adjust our factor date by one month to match the collection period.


RG pools already account for a sizable share of Ginnie II custom issuance, as illustrated in the following chart, making up 18% of G2 custom issuance and 3% of all G2 issuance since April.

Note: RiskSpan uses reporting month as a factor month. For this chart, we adjust our factor date by one month to match the collection period. 


RG Pool Characteristics 

Nearly all of RG pool issuance has been in 3.0% to 4.5% coupons, with a plurality at 3.5%. As of the September factor month, almost $4 billion (37%) of the outstanding RG pools are in 3.5% coupons. The 4% coupon accounted for the next-largest share–$2.5 billion (23%)—followed by $2.3 billion in 3.0% (20.9%) and $1.3 billion in 4.5% (11.8%). 


RG Pool Outstanding Amount by Coupon — September Factor Month 


 The following table compares the characteristics of RG pools issued since February with those of G2 single-family custom and multi pools issued during the same period.  The table highlights some interesting differences: 

  • Issuance of RG pools seems to be concentrated in higher coupons (3% to 4%) compared to issuances for G2 custom pools (concentrated on 2.5% and 3.0%) and G2 multi-lender pools (concentrated on 2.0% and 2.5%). 
  • Loan sizes in RG pools tend to fall between those of G2 customs and smaller than G2 multis.  For example, WAOLS for 3.5% RG pools is around 245k and is around 50k smaller than multi pools and 30k larger than other custom pools. 
  • RG pools consist almost exclusively of FHA loans while G2 multis have a much higher share of VA loans.  Almost 98% of 3.5% RG loans are FHA loans. 

 G2 RG vs. G2 Custom and G2 Multi (pools issued since February), Stat as of September Factor Month 

Wells Fargo and Penny Mac are far and away the leaders in RG issuance, accounting collectively for 62% of outstanding RG pools.  


RG Pools by Servicer, September Factor Month 


 How to Run RG Pools in Edge Perspective 

Subscribers to Edge Perspective can run these comparisons (and countless others) themselves using the “GN RG” pool type filter. The “Custom/Multi-lender” filter can likewise be applied to separate those pools in G2SF. 


Contact Us

Contact us if you are interested in seeing variations on this theme. Using Edge, we can examine any loan characteristic and generate an S-curve, aging curve, or time series.


Machine Learning Models: Benefits and Challenges

Having good Prepayment and Credit Models is critical in the analysis of Residential Mortgage-Backed Securities. Prepays and Defaults are the two biggest risk factors that traders, portfolio managers and originators have to deal with. Traditionally, regression-based Behavioral Models have been used to accurately predict human behavior. Since prepayments and defaults are not just complex human decisions but also competing risks, accurately modeling them has been challenging. With the exponential growth in computing power (GPUs, parallel processing), storage (Cloud), “Big Data” (tremendous amount of detailed historical data) and connectivity (high speed internet), Artificial Intelligence (AI) has gained significant importance over the last few years. Machine Learning (ML) is a subset of AI and Deep Learning (DL) is a further subset of ML. The diagram below illustrates this relationship:

Due to the technological advancements mentioned above, ML based prepayment and credit models are now a reality. They can achieve better predictive power than traditional models and can deal effectively with high-dimensionality (more input variables) and non-linear relationships. The major drawback which has kept them from being universally adopted is their “black box” nature which leads to validation and interpretation issues. Let’s do a quick comparison between traditional and ML models:

behavioral models versus machine learning models

Within ML Models are two ways to train them:

  • Supervised Learning  (used for ML Prepay and Credit Models)
    • Regression based
    • Classification based
  • Unsupervised Learning
    • Clustering
    • Association

Let’s compare the major differences between Supervised and Unsupervised Learning:

Supervised learning versus unsupervised learning

The large amounts of loan level time series data available for RMBS (agency and non-agency) lends itself well for the construction of ML models and early adopters have reported higher accuracy. Besides the obvious objections mentioned above (black box, lack of control, interpretation) ML models are also susceptible to overfitting (like all other models). Overfitting is when a model does very well on the training data but less well on unseen data (validation set). The model ends up “memorizing” the noise and outliers in the input data and is not able to generalize accurately. The non-parametric and non-linear nature of ML Models accentuates this problem. Several techniques have been developed to address this potential problem: reducing the complexity of decision trees, expanding the training dataset, adding weak learners, dropouts, regularization, reducing the training time, cross validation etc.. The interpretation problem is a bit more challenging since users demand both, predictive accuracy and some form of interpretability. Several interpretation methods are used currently, like PDP (Partial dependence plot), ALE (accumulated local effects), PFI (permutation feature importance) and ICE (individual conditional expectation) but each has its shortcomings. Some of the challenges with the interpretability methods are:

  • Isolating Cause and Effect – This is not often possible with supervised ML models since they only exploit associations and do not explicitly model cause/effect relationships.
  • Mistaking Correlation for Dependence – Independent variables have a correlation coefficient of zero but a zero correlation coefficient may not imply independence. The correlation coefficient only tracks linear correlations and the non-linear nature of the models makes this difficult.
  • Feature interaction and dependence – An incorrect conclusion can be drawn about the features influence on the target when there are interactions and dependencies between them.

While ML based prepay and credit models offer better predictive accuracy and automatically capture feature interactions and non-linear effects, they are still a few years away from gaining widespread acceptance. A good use for such models, at this stage, would be to use them in conjunction with traditional models. They would be a good benchmark to test traditional models with.


Note: Some of the information on this post was obtained from publicly available sources on the internet. The author wishes to thank  Lei Zhao and Du Tang of the modeling group for proofreading this post.


Get Started