Table of contents for Live from InterACT 2008
- Live from InterACT – Ian Ayres
- Live from InterACT – New Approaches to Strategies
- Live from InterACT – Insurance in the 21st Century
- Live from InterACT – Building a Decision Engine
- Live from InterACT – Using Risk Applications to Drive Growth
- Live from InterACT – Design for People, Build for Change
- Live from InterACT – Automate, Improve and Connect
- Live from InterACT – The Mortgage Crisis
- Live from InterACT – Scoring Innovation
- Live from InterACT – Optimal Pricing
- Live from InterACT – An Enterprise Decision Engine for Originations
- Live from InterACT – Changing the game
- Live from InterACT – Closing Keynote
Time to listen to some analytic scientists with Larry Rosenberger (analytic Fellow) and Jeffrey Feinstein (Principal Scientist) of Fair Isaac talking about gaining insight by focusing on behavioral drivers. Larry went first and mentioned a presentation be gave at InterACT back in 2006. He had focused then on where ideas come from and talked about “a jarring juxtaposition of ideas” and gave the example of how the FICO score predicts things like likelihood of accidents for drivers, how likely you are to stay fit and other things. The idea that credit behavior predicts these things is odd – jarring – so perhaps there is some underlying factor that impacts both credit and other behavior and it is this that the FICO score measures indirectly. Perhaps character, risk aversion, responsibility or something similar. Could, he asked then, such an underlying behavior be measured directly? Such a factor might be more insightful, more stable over time, more robust under varying conditions etc. Back on April 16th the New York Times ran an article linking what people eat to how they vote – not directly, perhaps, but because what you eat/drink is driven by underlying motivations that do matter. Larry called the underlying factor that drives credit behavior, and these other behaviors, the “X” factor.
This ideas has been developed since then and Jeff went next to explain where it has gone since then – on rethinking the traditional predictive modeling approach by using psychodynamics. Traditionally, analytic models use what you know about someone at decision time to make predictions about that person at a future time. When that time arrives, new data is recorded and used to improve the models etc. This approach derives averages, it is thinking about the aggregate.
Consumer by consumer, however, the behavior of each varies widely even if they have the same score at the start of a period. This happens because it is not just time that passes after a score is calculated but an interaction with a living breathing person. Psychodynamics considers the motivations and issues that drive an individual’s behavior. Three broad areas drive people’s behavior with credit:
- Behavior – what they did in the past
- Psychology – how they think, about their credit for example or about collections
- Situation – other factors
Two conclusions come from this:
- So this “X” factor, this underlying factor, is a function of biography and psychology.
- Behavior is a function of this X factor and the situation.
Traditional models treat cohorts of consumers as similar and update after an event occurs – such as a missed payment. The goal of the psychodynamic approach is to separate consumers in advance of a bad event. For instance, a responsible consumer will behave differently to the same situation to an irresponsible one. The responsible consumer stays a stable risk when the situation changes. An irresponsible one does not.
Jeff used a couple of examples to illustrate this. First was the Credit Capacity Index. Capacity was traditionally measured by FICO scores or income. FICO score scores likelihood of default but is not a measure of handling incremental debt and income is hit and miss. The new Credit Capacity Index is designed to measure the probability of defaulting on this incremental debt.
- High Capacity Consumers more likely to use credit as a tool, thick files, use credit responsibly
- Low Capacity Consumers less aware of credit, thinner files, more likely to use credit irresponsibly
The index is used in combination with risk cohorts to split them into those with high, medium or low capacity for a given risk. It’s a very effective tool for differentiating consumers with similar risk profiles into groups with different behavior.
Further out there is work on using consumer responsibility to work with HELOC consolidators. In particular, can it answer the question of whether HELOC consolidators have an overly high FICO score. The problem is that responsible people use HELOC consolidation to take control of their debt while others use it to expand their data capacity. R&D asked the question if those with scores of 700, say, who consolidate behave differently from those with 700s who don’t consolidate. What they found was that a HELOC consolidator behaved a little better than their score would predict. But this is an overall picture. In fact there is a variety of behavior. Most have balances and utilization decrease and delinquency increases. Some groups see big risk increases, other big risk decreases. This makes it seem like there are two groups:
- Responsible consolidators who use consolidation to recover from their indebtedness, save money should be less risky post-consolidation as they will behave responsibly.
- Irresponsible consolidators use consolidation to do something similar but may also do it because they are buried in debt and don’t know what else to do. Post consolidation they could be expected to succumb to temptation and so run up more debt and become worse risks.
When you look at the credit card balances of consolidators there are two groups – those who hardly add to their balances post consolidation and those who immediately add more debt, typically back up above their pre-consolidation total! The HELOC consolidation risk score, focused on interim behavior and built using these psychodynamic approaches, differentiates these groups very strongly and much better than other scores.
Considering, then, underlying responsibility seems to have value. The research pretty clearly worked for HELOC and bank card behavior. Now the challenge is to generalize this. In addition, psychodynamic modeling is focused on honesty, early adopter mindset and other underlying drivers.
As usual when I hear these guys, I only understand the implications and have no idea how they figure this stuff out. Very cool.