By Frank J. Diekmann
It isn’t so much that credit unions are drowning in big data; it’s that credit unions are drowning in people talking about big data.
The reasons are clear: as one expert has shown, data and analytics are critical to getting pricing right for the future; to knowing not just which loan will go bad, but when, and to better understanding which go-to indicators really shouldn’t be gone to at all.
Every gathering of CUs that I attend includes a database’s worth of discussion about how credit unions have all the data they need/need more data, have more/less data than others, have a leg up on other FIs/are falling behind other FIs, and on and on. I’m sure somewhere there is data on how often data is being talked about.
If there is a theme to be extracted from all of that data talk it is that credit unions recognize they must become better at interpreting the data they already have to better anticipate member/consumer wants and needs. That’s what’s known as “predictive analytics,” the power of which was recently made clear during a standing room only presentation at NACUSO’s annual meeting in Orlando.
At the session CUs learned just what they might be able to do with predictive analytics, where their own strengths and weaknesses exist, why FICO is a misleading indicator, and the data points that should really be getting more attention.
Drawing the big crowd was Joe Breeden, COO and chief scientist with Deep Future Analytics, who said credit unions should not be intimidated by the significantly larger tech budgets at banks, suggesting much of that money is going wasted and that credit unions can build similar analytics “models for much, much less” than what the big banks are spending. But he also cautioned that the big banks are spending in a way that could leapfrog credit unions.
“What are predictive analytics?,” asked Breeden. “It’s seeing the future, which is something we don’t do very well. It works, but it’s very difficult.”
What Predictive Analytics Are Not
Breeden, who was trained as a physicist and who has extensive experience working with major banks in data analytics, said the first value of predictive analytics is understanding the past and the present, and then developing an understanding of “where you’re going in the future.”
“The mortgage crisis in the U.S. is one of those great events where everything goes wrong and nobody knows why, and then you learn the most from the data,” he said. “In the 2009 mortgage crisis we did have models but they were not nearly as sophisticated as what we have today, and we knew we had a problem. Our forecast came true. Our forecast for HSBC was such a big number they spent eight months trying to prove us wrong before they came to grips with it. These things are real and they’re tangible. The question is how do we go forward.”
What predictive analytics are not, stressed Breeden, are FICO scores.
“FICO scores are prescriptive. FICO/Bureau scores measure overall risk on existing loans, not the risk after you give the person a new loan,” said Breeden. “They don’t measure product-specific risk or forward-looking risk. FICO scores are point-in-time management. A FICO score is basically the ability to short. But the act of lending is long-term. FICO scores don’t predict, they summarize. FICO scores are backward looking. I’m not beating up on their product; I want to make sure it’s used right. It doesn’t adapt to the economy.”
Although he was speaking to the issue of predictive analytics, Breeden said they are “just another buzzword.”
“What you should care about is where are people putting their money, where are they placing their bets,” he said. “The big analytics spends today are on regulatory compliance. The Fed, OCC, FDIC and NCUA (for CUs of more than $10 billion) have all put forecasting and stress-testing on center stage.”
What The Fed Wants To Know
The government and regulators, said Breeden, have realized that their own models have not worked and that Basel II has been a failure. “(Basel II) did not predict the mortgage crisis at all. In fact, it even said there was 70% too much capital in mortgages lenders in 2005.”
“What the Fed wants know is can we combine all these scoring concepts and make some loan level models,” continued Breeden. “Can we predict the probability of default? Can we predict loan-by-loan how delinquent you will be next month and the month after that?”
Just a year ago Breeden said he believed that while banks had far deeper financial resources to build analytics models than did credit unions, credit unions still held the advantage because banks were hamstrung with data in legacy systems.
“But that’s no longer true,” said Breeden. “By spending hundreds of millions of dollars they are leapfrogging these legacy systems. What’s happened since last year is the time window got a whole lot shorter. Your competition is doing a moon launch. (The analytics are going) to be product specific, long range, and they’re going to be right.”
Getting the Pricing Right
Where analytics can best be applied is in pricing, especially in loans. Pricing remains an area fraught with misunderstanding and mistakes, according to Breeden.
“Pricing is often meet-the-market or moving average,” he said. “In 2005, mortgage lenders were pricing based upon 2003 originations. Its not about the loss forecast, it’s about the pricing. We tend to price on last year’s losses. The ones who did the best in the mortgage crisis were the ones who realized they couldn’t do much with the pricing, but they could step back from the business.
“Pricing often ignores loss timing,” Breeden continued. “New loans are always low risk, but don’t stay that way. One of the illusions in the mortgage crisis is you book all these loans and get the revenue up front, but that’s not when the losses come. You’re building a time bomb. Pricing models rarely consider the future environment or even the current or average environment. Fundamentally what it comes down to is that all crises are pricing failures. You’re in lending, you’re going to lose money, losses are inevitable. The only problems come when you haven’t priced for those losses. The question is how to get a good, quantified perspective on those future losses.”
What’s Your Future Risk Appetite?
Pricing risk is often referred to as risk appetite, and is really about how a portfolio will perform in a downturn when some loans start becoming unprofitable, said Breeden. And that’s the crux of the dilemma, he told the NACUSO meeting: too many lenders price according to their current risk appetite, not what they might be swallowing in the future.
“It’s interesting what people are talking about right now,” he said. “Almost universally people are saying ‘We are growing our portfolio,’ and sometimes they will even say ‘aggressively.’ That often means widening the funnel and going downmarket. The economy, relatively speaking, is in good shape, and if you look at next two to three years, not too bad. But some things are changing. We just had a currency shock. The reason the Fed has not raised interest rates is that the exchange rate got there before them. The U.S./Euro exchange rate now is almost 1/1, and that has the effect of dampening the sale of U.S. products.”
Breeden believes the U.S. economy will see a slowdown in approximately two years. “It’s easy to predict future recessions, because there will always be one,” he joked, before adding, “I think origination right now is probably fine, but I’m going to pay close attention. As soon as I see corporate profits starting to sag, I’m going to be thinking about repricing. That is all a part of risk appetite.”
Breeden said credit unions must recognize how the world has changed and that there are far fewer “fat spots” in what has become a much leaner lending process.
“You need probabilities for everything. If I increase your credit line, what is your increased risk, your probability of defaulting based on that action?,” he observed.
Breeden urged every CU to “collect the data”--even if the data isn’t particularly clean. “We can do a lot with noisy data.”
What CUs Can Identify
With that data, credit unions will be able to identify:
- Balance growth/pay down. “Balance dynamics are sensitive to the product, term and segment. Balances are very dynamic through the life of the loan.”
- Attrition/Pay-off. “How long will the loan be with us? Don’t count the interest income in pricing if the loan pays off early. Different segments behave differently.”
- Loss Timing. “Do the losses come early or late in the product. The loss forecast model is the point of greatest risk. Loss timing is at the heart of everything I do.”
When it comes to credit risk, Breeden said credit unions do have one advantage, and it’s an advantage he’s heard banks acknowledge:
“Credit risk is more than just FICO and LTV,” he said. “Credit unions still have an advantage, they have the relationship with the member.”
He also that in all the mountains of data a credit union may have, it often looks to the wrong data points as most critical.
“When we build models on credit union data, by far the most predictive data is deposit balance,” he said. “The bigger that balance and the log of the balance are what really matter. We can build models that the big banks can’t, because we have relationships the banks don’t have. If I price to the individual, I can do things that they can’t do.”
He reminded that while corporate lending is all about the data, consumer lending remains all about the “psychology” of the borrower and the market.
One More Thing To Keep in Mind
Breeden added one more point he wants credit unions to keep in mind.
“I want to make one point very clear. As much as the big banks are spending, 95% of that spend is what I consider waste. It’s documentation and compliance that no one will ever read. We can build these models for much, much less than they are spending. I want to emphasize we have already done it. The first institution with a loan-level, forward-looking pricing model was a credit union. Pooled repositories across CUs can bridge the data gap and build predictive models first for pricing at a fraction of the cost of regulatory compliance models.”
Frank J. Diekmann is Cooperator in Chief at CUToday.info and can be reached at Frank@CUToday.info