Fannie Mae is one of the largest financial institutions in the world with $4.2 trillion in mortgage volume. One-quarter of all single-family homes in the US was purchased or refinanced via Fannie Mae and the company is one the largest financers of the multi-family market.
With that said, these mortgages are not made directly by Fannie Mae – these are bought from lenders, packaged up into securities, and sold off to investors. All of this boils down to smart risk management at massive scale. “Risk management isn’t just part of our business,” says Kimberly Johnson, EVP and COO at Fannie Mae, “it is our business.”
This risk management includes everything from establishing creditworthiness of buyers and the value of properties to understanding broader macroeconomic factors. Further, it is not just a matter of establishing a loan and moving on – all of these mortgages are continually assessed, meaning many millions of loans to contextualize. This means handling a lot of data, including confidential personal information, which falls under regulatory purview.
With all of this data, complexity, and regulation, one might think Fannie Mae would be among the last to move operations to the cloud. However, Johnson says after several proofs of concept, Fannie Mae is seeing everything from massive-scale risk calculations to user-facing services as a suitable fit on AWS with higher performance and some cost reductions.
“Several years ago, we built our serverless high performance computing workload with AWS Lambda to do Monte Carlo simulations on 20 million mortgages,” Johnson said at AWS re:Invent. “More recently, we did a proof of concept running AWS RDS on Graviton-2 and I have to tell you, the early results look great. We’re seeing performance improvements of 54 percent and cost improvements of 11 percent.”
Over time, Fannie Mae has added to the stable of AWS-native technologies to build out new services. It is lucky that had existing footing because there were never more pressing demands on the mortgage industry than early 2020 when the pandemic sent unemployment soaring and left mortgage companies without historical precedent to navigate by.
“Unemployment swelled by 25 million jobs, we had no historical reference point to predict how borrowers or mortgages might perform under those conditions. We need more information, to discover new data sources, develop solutions for troubled homeowners,” Johnson recalls. “It would have taken us months or even a year normally, but we had to move faster. We were already using AWS Kinesis as our streaming data platform. That let us ingest new data in real time. S3 took storage constraints out of the equation and SageMaker gave us analytics and the insights we needed.”
The end result of this quick spin-up on AWS were unique forbearance on 1.4 million single family loans. Since that time, 1.1 million of those have exited that stage and those homeowners are back on their feet.
While Johnson did not disclose Fannie Mae’s pre-AWS systems strategy or how AWS now interacts with any on-prem datacenters, the takeaway is that AWS is far more than just a provider of hardware. The mortgage giant uses countless AWS-native tools, from Elastic Map Reduce (EMR), SageMaker, RDS and more as software platforms and talks mightily about the advantages of S3, RedShift, and other data management and storage platforms inside AWS.
For instance, one of Fannie’s Mae’s goals is to make access to quality housing more affordable but getting to that point requires a rethink of the traditional metrics for creditworthiness. For some underserved segments, what credit scores measure is not always in line with what they have had access to. This meant Fannie Mae had to look for non-traditional proof points of on-time payments. One of the best metrics is on-time rent payments – something credit scores don’t factor in. The challenge is, these can be paid in multiple ways (ACH, check, cash, split payments, etc).
“We used our decision system, desktop underwriter, to see if a loan meets our requirements. That system can now look at a sea of cashflow data and identify timely rent payments. This might sound simple but it’s quite complex because of all the ways rent is paid,” Johnson explains.
“So we used a new data source [bank statements] and AWS RedShift and S3 for storage and EMR to turn unstructured data into structured data. To make these sources ready for production, we used machine learning to develop algorithms that identify payments in statements with SageMaker to address all the combinations. And in September, we were finally able to use rental payments as part of our underwriting system.”
“Fannie Mae is working to solve the biggest challenges in housing. We are accelerating the digital transformation of our business to make it safer, simpler, and less expensive for lenders to originate a mortgage. Our ability to innovate directly improves our ability to meet our housing mission by making it more equitable and affordable to buy or rent a home,” Johnson adds.