Being part of Fair Isaac, I have never really doubted thecritical role predictive analytics could play in business. I see real-life examples day in and day outbut when I read this New-York Times article, I found one that was painfully telling…
Let me point your attention to a few points that Charles Duhigg makes in this investigation. I will not comment though on the politics involved.
Fannie Mae has been playing a crucial role for the lending industry as we all know. Their early success relied on their ability to predict which borrowers would be able to repay, assessing the premium required to compensate for the risk they took. With an effective model, you can pinpoint good risk versus bad risk and therefore make safe decisions that ensure the business will prosper. This is pretty much the essence of predictive analytics: once you know the probability to repay for people with a given set of characteristics, you can extrapolate and estimate how much reserves you need to build in order to beat the odds that you will not get you money back for each population segment. This is how premiums are calculated.
Unfortunately with new types of mortgage product and thel ack of associated data, Fannie has not been able to produce a robust predictive model. Not being able to tell how borrowers would behave in the long run, there was no way to estimate how much risk they were exposing themselves to. When you combine that with the absence of a CRO (Chief Risk Officer),which is like driving with your eyes closed, no wonder we got where we are.
This is fairly atypical black and white example but it drives the point: with a robust model, you thrive / without a model, you’ll hit a wall… eventually. I am not saying that predictive models are the only way to make good decisions but they definitelyhelp manage the uncertainty.
One might wonder how they could have built a robust model in absence of data. This is a fair question. Well, how about better integrating a feedback loop into the system to detect the early signs of the meltdown? This is what we call Decision Improvement. Another potentially complementary approach might have been to take a less aggressive stand: knowing that we don’t know for sure, premiums could have been calculated with more room for caution. This is where business rules and predictive models work hand in hand.
Again, this is not taking into consideration other factst hat influenced the situation such as the mandates from Capitol Hill or the laxity of credit agents.
