Note: Underwriting simply refers to the process of reviewing an application for credit. In our case, it’s how we review loan applications
There’s no such thing as a “magic” underwriting algorithm–one that, using data alone, accurately predicts the likelihood of an applicant paying back a loan. In fact, one could argue that so-called “data driven risk models” do more harm than good (think about the financial collapse and all the statistical geniuses who were behind Collateralized Debt Obligations and other weapons of financial mass destruction).
Still, there’s no shortage of people trying to convince investors and the public that they have just such an algorithm. An especially poignant example is a company called Zest Cash, the founder of which boldly proclaims that “all data is credit data…” Wow, that sounds fantastic–data mining, algorithms…powerful stuff. Except that when you go to their website, you see that the Annual Percentage Rate on their loans is 390%. Let me make this extremely clear: if you are charging 390%, your algorithm is worthless. You could close your eyes and make loan decisions based on the roll of a dice and still make money at that rate. It’s laughable.
Here’s why I think relying on purely data driven algorithms is stupid for applicants with poor credit. FICO, which has an algorithm that gives a score of 350 – 850 (a “prime” score is above 640), is widely used for the vast majority of credit decisions. And the higher the FICO score, the more predictive it is; put another way, if someone has a 750, chances are you don’t need a whole lot of other information to approve an application. Where FICO fails, especially for low-income people, is when the score goes down. You see, someone may have a low score due to irresponsibility–i.e., they walk away from a debt–or, in many cases, it may simply be due to the fact that they are living on the margin. For these applicants, we look for examples of a willingness to work with creditors; unfortunately, creditors rarely work with them, for which reason they fall behind on debts, and some of those debts go to collection agencies…Then their score goes down, their borrowing costs go up, and the cycle begins.
In my opinion, it is well-nigh impossible to use quantitative data alone to determine whether someone who has fallen behind on debts in the past will honor a debt in the future. For one thing, there are so many factors involved: the economy, life events, the attitude of the creditor. But perhaps most importantly, it is extremely hard to account for how people react to a strong relationship with a creditor, one that is built on mutual respect and flexibility. After all, how else can we explain why we are seeing 93% repayment rates on our loans despite the fact that the average FICO score of our borrowers is 592? We are outperforming an algorithm that likely has trillions of data points informing it!
The simplest answer is that we have gotten really good at looking at the quantitative stuff, but the most important element of our success is the relationships we build. Our clients know that we are in this together, that we are rooting for them, that we are not in it make money (but we still need to be repaid), that we will work with them when they fall behind.
We’re Getting Things Right!
Anyway, two recent stories have helped to validate our unique approach to underwriting. First, several years ago we realized that taking into consideration medical debts when reviewing an application simply doesn’t make sense: 60% of bankruptcies are due to medical debt, and it isn’t the fault of our clients that our health care system has failed them. Well guess what? According to the New York Times, the latest version of the FICO score will “no longer weigh medical debts–which account for about half of all unpaid collections on consumers’ credit reports–as heavily as it did in previous iterations.” It feels pretty damn good to know that “the creator of one of the most widely used and influential credit scores” in the country has decided to do something we’ve been doing for years!
And another example, also from the New York Times in an article titled Improving Default Rates on F.H.A Loans, shows that we are on the right track. The article talks about the fact that “default rates for loans backed by the Federal Housing Administration have consistently been higher than those on loans guaranteed by the Department of Veterans Affairs.” A study from the Urban Institute suggests that a key difference “involves underwriting. In addition to measuring [an applicant’s] debt-to-income ratio, the V.A. uses a “residual income” test. Borrowers must show a certain level of income after mortgage payment, taxes, utilities and job-related expenses…are subtracted from gross monthly income.”
I imagine you have already guessed that we’ve been doing that for three years! How much of a difference has this made? For loans closed in 2007, “36 perfect of [FHA] loans have gone at least 90 days delinquent. By contrast, the default for VA loans from that year was 16 percent.” That’s a HUGE difference.
So while there is no such thing as a magic algorithm, good old fashion smart underwriting–looking at the right data points while getting to know the applicant–has always worked and will always work. It’s only when lenders try to speed things up so as to maximize profits that they run into trouble. The old adage that the best banking should be boring holds true; boring, but absolutely critical to economic growth, social justice and reductions in poverty and inequality.
I would be remiss if I didn’t acknowledge the work of Joseph Holberg, George Steele, José Fonseca, Peter Vail and myself on the creation and refinement of our algorithm!