Risk Model Management – Challenges in a Shifting Economy

Navigator Edition: March 2015
By: Dan Kreis and Abby Lee

One of the key areas the Consumer Financial Protection Board (CFPB) has been tasked with is the scrutiny of Risk Model Management for all consumer lenders. As defined in the OCC Bulletin 2011-12, the basic requirements focus on:

  • Scorecard design challenge
  • Scorecard use and management
  • Scorecard validation
  • Governance, policies and controls

An essential part of this monitoring is to identify and respond to shifts in various factors including the market and consumer behavior. A perfect example of this can be seen in the data released last year by VantageScore Solutions showing the changes in predictive contribution of key scoring concepts:

At first glance, one might assume consumer behavior has been erratic, since predictive contribution has shifted greatly. Recent Credit, for example, as a predictive factor has slid from a dominate 30% influence to a marginal 5% contribution in the overall predictive value. One might argue a decline in employment may drive this type of shift. An alternate theory – and perhaps more likely – is that the variation in Recent Credit was not driven by consumer behavior, but by lending behavior.

Figure 1: Information Predictive Contribution
Fig-1_-Information-Predictive-ContributionSource: Credit Scores, The gateway to credit and how the scoring landscape is changing, Barrett Burns, VantageScore, July 17, 2014.

The unsound lending practices leading into the 2007-8 crisis (e.g., no down payments, no income validation, no-doc mortgages) expanded the population with Recent Credit and exacerbated the negative impact of new borrowers from this period. Conversely, in 2010-11, it was difficult to obtain a loan and the population in Recent Credit contracted and shifted towards a more affluent population.

As the predictive power associated with Recent Credit declined, other characteristics (Payment History and Depth of Credit) exhibited gains. While these shifts are less dramatic, they further highlight the volatility in recent consumer data and the challenges in developing robust predictive models.

So what does this mean for 2015?

Given the increase in lending activity and even pressure from the CFPB to expand lending (see December 2014 Consumer Credit Reports: a study of medical and non-medical collections), we anticipate that Recent Credit’s marginal contribution to predictive power will increase by year end.

Models may therefore be understating the negative impact of new account opening and lenders should take action. Merely satisfying the regulatory check list will not inform lenders on what current underwriting practices should be revised. Some potential actions may include:

  • Segment accounts with multiple new account opening
    –   Establish a higher approval threshold
    –   Lower credit line assignments
    –   Reduce loan terms
  • Employ Constrained Optimization in model development
    –   Place greater emphasis on more stable characteristics

In our estimation, shifts in predictive value will become more acute with emerging technologies and alternative sources of data. It will be critical to employ a well-informed intuitive approach to model development that goes beyond the statistics.

For more information, please contact Dan Kreis, Director of Portfolio Management,; or Abby Lee, Senior Analyst, specializing in Portfolio Management and Bankcard Issuing,

To read the rest of this article, please subscribe to

The Navigator Newsletter