First, apologies to those dedicated fellow fintech junkies that actually read what I write. I’ve been very delinquent in posting recently but it’s been a busy time personally and professionally. Hopefully I’ll be able to post more regularly going forward….hopefully!
With this said, having just returned from the Money2020 conference I thought I’d share a few thoughts on a topic that I was asked about by quite a few people. It’s not lost on the broad community of investors and entrepreneurs in the lending space that most of the “next-gen” lenders have built and optimized their risk models during a fairly benign economic period. Everyone seems to be worried about how well these models will perform in a downturn and how one can protect against a sudden and massive deterioration in a portfolio’s performance.
The key to answering these questions lies in the meta-question: “How resilient is your borrower base”. And measuring resilience is really about understanding “How many things have to go wrong” at the customer level.
At its core, a lender’s job is relatively straightforward. A loan officer makes loans to “healthy” customers who they believe are willing and able to pay back the loans. But the unfortunate truth is that some borrowers in every portfolio default on their loans. The “why” is pretty clear: Borrower’s circumstances change over time and these changes matter.
Foundationally, a healthy borrower has the following traits:
- A relatively stable source of income that supports one’s obligations/lifestyle
- Enough savings to weather a temporary disruption to one’s income
- Enough savings or free cash flow that can handle the introduction of additional unforeseen expenses
- A willingness to pay one’s debtors when the money is available
- The ability to quickly find a new source of income after a disruption
- The ability and willingness to turn collateral into cash to pay one’s obligations
So the breakdown of a healthy customer can be traced to a fundamental change in their circumstances.
- Temporary reduction of income (job loss, reduced commission, etc)
- Permanent reduction of income (major change in health, retirement, etc)
- Increased cost of living (increased borrowing, new child, etc)
- Unforeseen major expenses (car repair, medical bill, etc)
- Reduction in financial safety net (increased spending, reduction in new savings, etc)
- Reduction in willingness to pay (strategic default, attitudinal change, etc)
Statistically based underwriting models perform better than loan officers because models are able to predict the natural change in circumstances at the customer and portfolio levels. A model doesn’t classify a customer as “good” or “bad” but rather that they have a certain probability of paying back a loan.
But both human and statistical based underwriting models/policies suffer from the same phenomenon – tomorrow isn’t guaranteed to look like today. And while models are able to project an ambient deterioration in a portfolio’s performance, they aren’t fundamentally able to project what will happen in a future they’ve never seen before.
The natural reaction from investors and entrepreneurs who haven’t managed loan portfolios through cycles is to be terrified of what’s to come. Investors want to naturally stop investing in companies that originate loans. Less experienced entrepreneurs don’t know how to build resilient underwriting models and convince investors that all is “OK”.
My advice is simple:
Make sure your models give significant weight to the major drivers of risk (ability to pay, willingness to pay, stability of income, etc). Just exposing a model to hundreds of potential variables isn’t good enough. Credit officers have to make sure their models appropriately weights each and every potential driver. And if a model doesn’t want to use an important variable or driver, a credit officers’ job is to force it into the models or policies.
Why? If an important goal is to create a resilient portfolio, to do this a lender needs to make sure many things can go wrong within the customer base before defaults exceed expectations. For example, it’s critical to avoid lending to customers who are on edge from a capacity standpoint because minor changes in their circumstances will push them from solvency to insolvency. DTI might not show up in the underwriting models because our economy has been great for the past handful of years, but I can definitively say from experience that DTI won’t matter until it’s the only thing that matters. Models don’t understand this. Good credit officers do.
Another example: Make sure customers with less stable professions have enough savings to weather a temporary disruption to their incomes. There’s great data at the Bureau of Labor Statistics regarding unemployment rates by profession. If you’re not studying these statistics and internalizing their impact on the stability of your customer base’s income stream you’re missing out on a great source of information. Reg B (fair lending) has to be considered when designing approve/decline/pricing policies but once again I can definitively say that this data can be used to improve the resilience of a portfolio of loans.
Belts and Suspenders. Just make sure you’re not building a business where a single shock to the system causes issues at the customer level. Trust me — you’ll sleep better as will your investors and customers!