As bad debt continues to rise, lenders will need the industry’s best approaches to risk management to effectively compete and to grow profit margins. With the upcoming introduction of its third release of Consumer Application Risk Models for originations (ARM 3.0)—developed from a data set six times as large as version 2.0—Fair Isaac is giving lenders greater predictive power, and the ability to apply the models to many more portfolios and to an expanded geography.
“While underwriters have effectively managed risk with our 2.0 models, ARM 3.0 provides even greater predictiveness and flexibility in making originations decisions,” says Fair Isaac Senior Project Manager, Dina Rojas. “With ARM 3.0, lenders can not only further reduce their risk exposure, but they can also expand the range of transactions they assess. That level of power and flexibility is critical in today’s market conditions.”
The development of the ARM 3.0 models, which will be available in Fair Isaac origination software systems (such as LiquidCredit® decision engine) in 2008, resulted in many advancements and benefits than were offered in the 2.0 version (see Figure 1).
Deeper insight, expanded scope
In validation testing for several lenders, the new models have consistently demonstrated a significant boost in detection of bad performing accounts.
“We’ve used Fair Isaac’s application risk models for auto, unsecured, and home equity lending for almost 15 years, with much success,” says William Zysk, vice president, credit services, Pennsylvania State Employees Credit Union.
“Fair Isaac recently completed a validation analysis comparing their newest models against the version we currently use,” says Zysk. “The results are quite compelling—improvements of 11% and 18% on two key measurement criteria. The new ARM 3.0 will improve the quality of our decisions.”
ARM 3.0 models were developed from Fair Isaac’s vast proprietary databases, using multiple modeling and data analysis techniques. Models are available for industry-specific portfolios, such as student lending, direct sub-prime auto, and revolving credit. In addition, whereas version 2.0 had no Canadian-specific models, version 3.0 offers seven models tailored for Canadian markets.
In development and testing, all new ARM 3.0 models demonstrated superior predictive power over ARM 2.0 models, and over the sole use of FICO® scores (see Figure 2). With ARM 3.0, which is based on evaluation of data not available to FICO® scores, lenders have the ability to use various FICO® scores, (such as FICO® Auto Industry Option or FICO® NextGen) as one of the models’ predictive characteristics.
Consumer ARM 3.0 showed a stronger ability to detect bad performing accounts compared to previous generations of ARM and FICO® scores. For example, the sector with the lowest scoring 20% cumulative good accounts identifies 53% of bad accounts, versus approximately 40% of bad accounts with ARM 2.0 and FICO® scores.
Until the 3.0 models become available, lenders currently using the ARM 2.0 models can feel confident in their predictive power. All ARM 3.0 validation testing showed no significant degradation of the 2.0 models. In fact, Fair Isaac’s impetus behind development of ARM 3.0 was not based on the effectiveness of the 2.0 models. Instead, the ARM 3.0 development is a result of the accumulation of data that provides a much greater representation of US lending markets.
Janette Van Meter, a senior vice president and central underwriting manager at Oklahoma-based Stillwater National Bank supports that point. "Since Stillwater National Bank has been using the ARM 2.0, we have seen a significant drop in our past dues and charge offs."
Lenders’ growing support and data participation in model development has enabled Fair Isaac to greatly expand the scope of the new models. ARM 3.0 was based on a random sample of 3 million records assessed, versus 500,000 in version 2.0. The random sample of records was pulled from a total of 7 million records. That’s a reflection of how much ARM 2.0 clients support Fair Isaac’s model development.
Inroads to new business
The new Consumer ARM models provide a cost-effective solution for lenders that are new to using scoring, or for lenders wanting to enter new or niche markets.
ARM 3.0 gives lenders a way to quickly launch new portfolios, or expand their geographic reach. It gives them the ability to bypass the expense and time involved in data gathering and model development.
Steven Vaughan, an industry consultant and former president of several community banks, advocates the use of models, and particularly the Consumer ARM models, for banks seeking growth, or looking to take a competitive advantage.
"Any community bank that has a strategy of either increasing its consumer loans, or trying to achieve the status as the service leader in their market, should consider the use of Fair Isaac’s Application Risk Models to increase the speed, consistency, and efficiency of the consumer lending effort,” says Vaughan. “Fair Isaac is the industry leader in origination analytics. The newer models in ARM 3.0 offer more predictiveness and more loan products than ever before."