When SAS chief marketing officer Jim Davis told a roomful of executives in Washington last spring that he "doesn't believe business intelligence is where the future is," it went off like a little bomb in the close-knit circles of financial technology.
Not that Davis seemed to be forsaking SAS' bread-and-butter business intelligence software empire, but because he put his finger on a much broader issue: the blind adherence to enterprise data. The predictive modeling, automated credit decisioning, risk management and business decisions made with confidence behind BI-bred numbers had proved to be a disastrously unreliable guide through the storm of a financial meltdown, and there are ramifications to come for institutions' books.
What is needed, he and others believe, is a new discipline that more closely examines the raw numbers relating to customer and market activity. By adding more variables that ask "what if" instead of just "what happened," bankers and traders are looking for next-generation, deeper-diving metrics that will hopefully erase the blind spots in current data-based decisioning, and prevent future catastrophe. Thomas Davenport, author of "Competing on Analytics," management professor and former Accenture fellow, said he was once told by a former Washington Mutual chief risk officer "if they had had better model management systems, they would have seen they were producing more charge-offs in 2006 than predicted, and they could have changed their practices more rapidly."
It's a science that is still developing, but Davis is positioning SAS toward what he calls business analytics and away from the plain-vanilla query and reporting of business intelligence. SAS isn't alone: IBM just took the wraps off its "Smart Analytics System," on sale in its new Business Analytics and Optimization Services practice, and is supported by a 4,000 consultants, 200 mathematicians and a global network of analytic solution centers. One of the first customers is Sterling Savings Bank in Spokane, Wash.
This business analytics trend is an answer to more than just what's next, but also about getting more out of the data-tools investments previously made for BI-based automated decisioning and predictive modeling. It seems a more acceptable alternative to the radical solutions from those proposed by financial scholar Nassim Taleb, author of "The Black Swan," who feels the answer to bad financial modeling is to simply burn the drawing board. "Numbers don't make you risk-averse. They make you less risk-averse. The answer is to ban these products," he told NPR's Planet Money. "Anything that relies on mathematics for its survival, anything that relies on mathematical models should disappear. Because we know nothing about these probabilities, and the past is no indication of the future."
London tech consultant and mathematician Peter Thomas say that, by Taleb's logic, we should stop decision-making altogether, because all decisions are made using information from the past. Harvard economist Edward Glaeser adds: What else is there?
As much false confidence may come from reams of numbers, there won't be a wholesale abandonment of modeling anytime soon. If anything, banks look like they are doubling down on data. A recent small-scale survey from consultancy Aleri shows 70 percent of banks and brokerages think their firms needs to manage risk in real time-and invest in more data-crunching tools in order to do it.
Bank of America, for example, is looking for someone in Hong Kong to run business analytics in its Merrill Lynch unit, using Dealogic and other research tools to "data mine" banking deals within the region, ensure data integrity with an emphasis on thoroughness of sources and variables, and develop models to "create new views of the business that allow improved study of impacts in the market."
The new parameters on what those models might include are still in development, but top Merill executives lately on the rubber-chicken circuit are dropping clues. New BofA-Merrill Asia president Kim Hong recently said the firm is adding people in investment banking and key research posts across the region in a bid to exploit "inter-regional," cross-border plays for M&A clients, connecting those clients to financing, and trying to steer clients to multiple desks - the old integrated bank idea, at a time when many bank-watchers are advising a back-to-basics approach. Merrill tech and risk exec Jay Morealle told an IDC forum that he expects business intelligence to be about analysis, not reporting; it needs to be fast, intuitive and assist decision-making. Data should flow from a base of power users, not from IT, and publish back into broader systems so information can be widely shared.
On the retail banking side, the $140 billion-asset BB&T Corp. is revamping its ongoing use of SAS risk management software, working closely with SAS partner The Financial Risk Group. BB&T risk manager Mike Stevens is also trying to work more flexibility into modeling. For example, if the "affordability index" of housing prices and personal income is distorted, a correction must be taken account into the lending process, he told an SAS panel. Rather than just predicting a bad loan workout three years in the future with higher asset values, he said, you also build an assumption that asset prices could very well be lower, and the loan-to-value ratio is therefore much higher than expected. "I may then be in a negative equity position and taking a significant loss on this bad loan," he says.
There are questions whether business analytics is just a new mine for fool's gold. Neil Raden of Hired Brains thinks analytics-versus-intelligence is "all fluff." Yet others might just say that underlying assumptions, automated credit scoring and rules engines based on things like FICO scores should now reside in a big smoking hole in the ground, along with the huge marketplace of unpriceable securities built on top of them.
Michael Stefanick, SAS's global head of risk practice, believes the answer is increasing the data going into the models, and increasing the velocity of optimization - the updates of the data points. "Lenders have realized the limitations of FICO and are now basing models on a larger set of risk factors," he says. "This is where you find a more accurate view of how a customer will manage credit, and go beyond a credit snapshot."
There's a paradox pointed out by Eric Lindeen, marketing director for Zoot Enterprises of Bozeman, Mont. He points out regulatory changes and tightening of risk-based capital ratios are going to drive data-handling practices anyway. "Figuring how to do risk-based pricing up-front at a granular level is going to take a lot of system flexibility," he says. "Institutions will need to run models much more frequently and in many more situations." At the same time, he calls for simpler models with a greater degree of flexibility - conceding that complex models in a dynamic market just proved to have "virtually no predictability." The question then becomes, how to do that while making them real-time and increasing the number of risk factors?
Azhar Iqbal, an econometrician at Wells Fargo Securities in Charlotte, still mourns over a model he built in the summer of 2007 while at Wachovia that tracked U.S. light car sales. "For four or five months, we nailed it," he says. After that? "We couldn't identify why our model was totally failing." It was because the credit markets were freezing. "All my other indicators supported one thing, that retail sales should go up. But nobody was lending, so sales went down." He's still tinkering with light vehicles, looking for new variables, for an optimized proxy for credit availability as well as the propensity for its use: perhaps something like the TED (Treasuries/Eurodollars) spread. Iqbal is hunting for variables "outside the cycle," and coming up with interesting findings on the way: he and Wells Fargo senior economist Mark Vitner published a paper showing delinquency rates for consumer loans are driven more rapidly by factors other than unemployment. In other words, delinquency rates for different products, say credit cards and mortgage payments, are driven by different forces.
Wells Fargo Securities now keeps an ongoing record of every one of its operating models. If one misses three consecutive directions, it's taken off-line and evaluated - an example of model management that Davenport said was lacking at Wachovia before its demise. "Econometric theory is being re-evaluated. The lesson is, many theories broke," Davenport says. "And when that happens, you need to modify."