US Mortgage-Backed Securities (MBS) is a $7.3 trillion market composed of 1M pools of Prime Real State loans guaranteed by:

• Fannie Mae Single-Family MBS

• Freddie Mac Single-Family MBS

• Ginnie Mae Single-Family MBS

FNMA_PoolView is a shiny app that displays Fannie Mae, Freddie Mac and Ginnie Mae guaranteed pool and loan characteristics on a pool by pool basis.

The upper region selects the agency as above and the reporting month (for demonstration, a random sample pool is loaded each time). Below are two tabs, one for viewing single pools and one for cohorts or aggregations.

For individual pools, containing the subtabs:

For entering a single pool and displaying it's main characteristics, including it's geographical dispersion.

For displaying all information disclosed by the source plus some calculated fields like prepayment speeds.

ARM-specific information in case the pool is an ARM.

Supplemental information about the pool

Displays loan level information for the pool when available.

Displays aggregations and analytics (prepayment rates) on the cohorts

The publicly available data files monthly provided by the agencies are stored in Amazon Web Services (AWS) S3 file hosting platform. The list of buckets created is:

##                                  Bucket             CreationDate
## 1          fhlmc-mbs-sf-arm-singleclass 2021-01-26T18:23:19.000Z
## 2  fhlmc-mbs-sf-arm-singleclass-datadir 2021-04-20T16:38:38.000Z
## 3              fhlmc-mbs-sf-singleclass 2021-01-09T11:40:48.000Z
## 4      fhlmc-mbs-sf-singleclass-datadir 2021-04-20T16:39:09.000Z
## 5                       fnma-llp-2020q4 2021-04-28T16:11:13.000Z
## 6               fnma-mbs-sf-singleclass 2020-10-27T05:57:35.000Z
## 7       fnma-mbs-sf-singleclass-datadir 2021-04-20T16:37:45.000Z
## 8                             gnma-hmbs 2021-01-09T11:51:03.000Z
## 9                     gnma-hmbs-datadir 2021-04-18T12:35:03.000Z
## 10              gnma-mbs-sf-singleclass 2020-12-25T05:57:33.000Z
## 11      gnma-mbs-sf-singleclass-datadir 2021-04-15T18:07:20.000Z
## 12                              test-tg 2021-02-09T18:27:34.000Z

For example, the first Fannie Mae monthly factor files is:

## Bucket: fnma-mbs-sf-singleclass-datadir 
## 
## $Contents
## Key:            FNM_MF_201910.zip 
## LastModified:   2021-04-20T16:43:00.000Z 
## ETag:           "d7cfe52c7a07971021f5de1f74e6dcfb-4" 
## Size (B):       31689689 
## Owner:          3ed8938a6ec6ccbf8e5544fed9c6be5f74559d6d28ddeda9375f52176205d37b 
## Storage class:  STANDARD

The input files are parsed and stored in binary format in AWS S3:

## Bucket: fnma-mbs-sf-singleclass 
## 
## $Contents
## Key:            FNM_MF_201910.fst 
## LastModified:   2020-10-27T06:05:25.000Z 
## ETag:           "ff8565f0ec8dea21ec118c0bb72433c1-4" 
## Size (B):       32205844 
## Owner:          3ed8938a6ec6ccbf8e5544fed9c6be5f74559d6d28ddeda9375f52176205d37b 
## Storage class:  STANDARD

Package loanroll processes Fannie Mae and Freddie Mac MBS pools.

Examples of its use are:

The MonthlyFactorDataset object is loaded from AWS S3

# devtools::load_all("~/Finance/FNMA/loanroll", reset = TRUE, recompile = FALSE, export_all = FALSE)

Factor_Date <- "2021-04-01"
args.lst <- list(Factor_Date = Factor_Date, bucket_name=fn_mbs_sf_bucket, verbose = FALSE)

MF <- tryCatch( 
  do.call(MonthlyFactorDataset, args.lst)
  , error = function(e) e
)
if(inherits(MF, "error")) {
  stop(conditionMessage(MF))
}

Then monthly aggregations are made. For example, for FNCL:

FNCL <- subset(MF, subset = quote(Prefix=="CL" & Seller_Name != "SCR" & WA_Net_Interest_Rate %in% seq(1,15,0.5) & Security_Factor_Date == Factor_Date))
FNCL_stats <- aggregate(FNCL, by.vars=c('Prefix'))
saveRDS(FNCL_stats, "FNCL_stats.Rds")

FNCL_Coupon_stats <- aggregate(FNCL, by.vars=c('Prefix', 'WA_Net_Interest_Rate'))
saveRDS(FNCL_Coupon_stats, "FNCL_Coupon_stats.Rds")

Package gnmar processes Ginnie Mae pools.

Examples of its use are:

A GinnieMBS object is loaded from AWS S3

options(verbose = FALSE)

readRenviron("~/Finance/GNMA/.Renviron")
# devtools::load_all("~/Finance/GNMA/gnmar", reset = TRUE, recompile = FALSE, export_all = FALSE)
remotes::install_github("canarionyc/gnmar",
#                        dependencies = FALSE,
                        force = FALSE,
quiet = TRUE
)
library(gnmar)

As_of_Date <- as.Date("2021-03-01")

args.lst <- list(
  As_of_Date = As_of_Date
 #                , mf_zip=mf_zip
  #              ,  bucket_name = gnma_mbs_sf_bucket
  , overwrite = TRUE
  , verbose = FALSE)

ginnieMBS <- do.call(GinnieMBS, args.lst)
# show(ginnieMBS)

GNSF <- subset(ginnieMBS, subset=quote(Pool_Indicator=="X" & 
                                         Pool_Type=="SF" &  
                                         Issuer_Number!=9999 & 
                                         Security_Interest_Rate %in% seq(0.5, 11, by=0.5)))
# print(summary(GNSF))
GNSF_stats <- aggregate(GNSF, xvar=NULL,  by.vars=c('Pool_Indicator', 'Pool_Type' )
                        , verbose=FALSE
                          )
saveRDS(GNSF_stats, "GNSF_stats.Rds")

# GNSF_Coupon <- subset(GNSF, subset=quote(Pool_Indicator=="X" & Pool_Type=="SF" &  Issuer_Number!=9999 & Security_Interest_Rate %in% seq(0.5, 11, by=0.5)))
# print(summary(GNSF))
GNSF_stats.by_Coupon <- aggregate(GNSF, xvar=NULL
                        , by.vars=c('Pool_Indicator', 'Pool_Type', 'Security_Interest_Rate' )
                        , verbose=FALSE
                          )
if( "grouping" %in% names(GNSF_stats.by_Coupon)) {
  GNSF_stats.by_Coupon <- GNSF_stats.by_Coupon[ ,-c("grouping")]
}

This shiny app is a tool for exploring the Loan Level Public Dataset provided by Fannie Mae at <https://capitalmarkets.fanniemae.com/tools-applications/data-dynamics> and predicting future performance by a Markov Finite State model.

The application has two panels:

The Origination Year is entered and all the loans originated in that year available are pulled from the database. Next you can select as the time variable either the reporting month or the loan age, and accordingly either a calendar range or a seasoning range (for example, use the 0 to 60 months of age experience of all the loans of the 2020 vintage). Time series of the aggregated value of several relevant characteristics like delinquencies, prepayments and defaults month by month or accumulated can be displayed. For example, the plot below shows the increase of delinquencies at the worst of the 2020 COVID pandemic in March-April 2020.

A second tab not shown displays the (cumulative) transition matrices between Markov states.

In this panel the dollar amount of the initial portfolio of loans is entered and the evolution of the portfolio according to the loan experience is displayed. When the initial state is current and the experience is the full calendar range, the final prepayment, delinquency and default rates agree with the actual values reported by the Agency.

The input files are parsed and stored in binary format in AWS S3:

## Bucket: fnma-llp-2020q4 
## 
## $Contents
## Key:            Data_P_2020.fst 
## LastModified:   2021-06-04T15:47:06.000Z 
## ETag:           "d41d8cd98f00b204e9800998ecf8427e" 
## Size (B):       0 
## Owner:          3ed8938a6ec6ccbf8e5544fed9c6be5f74559d6d28ddeda9375f52176205d37b 
## Storage class:  STANDARD