Bankers that want to defuse government accusations of biased lending are recruiting statistics experts to challenge the claims.
These unusual hired guns are taking aim at government allegations based on their use of regression analysis. This relatively new weapon in the federal regulators arsenal also ranks among the most powerful for it uses the banks own data to reveal what appear to be previously unnoticed patterns of discrimination.
Trouble is, experts say, this weapon is so touchy that they doubt federal regulators are skilled enough to use it properly. Their advice is both to question the governments methodology and, at the same time, develop internal regression models that could spot in advance whatever Washington is likely to find.
Barnett Banks is doing that as part of its battle against efforts by the Department of Justice to sign on to a fair lending agreement that would impose onerous government intrusion onto the institutions business activities. The Jacksonville, Fla., institution retained Harold Black, a professor of finance at University of Tennessee at Knoxville and an expert on use of computer models to determine lending bias at banking institutions.
In an interview with Mortgage Marketplace, Black said he has set up his own regression model based on what is likely to be the same kind of information that DOJ is using. But the assumptions hes using in his model are different.
The DOJ analysis is inappropriate in its technique, Black contended.
Regression analysis is a method for predicting outcomes based on the relationships, or correlation, between known conditions. The more consistent the relationships, the more predictable the future outcomes.
Regression analysis isnt mainframe rocket science: Lotus 1- 2-3, and, to a lesser extent, Excel can be used to create a multiple regression model that can crunch your banks data in meaningful ways. You can use as many as 75 variables with Lotus, but experts suggest that a valid model may include no more than 15 to 25 once testing and subsequent scrubbing of unpredictable variables are completed.
The best way to stay out of trouble might be to get the Feds regression model and apply it to your data before the examiners arrive. Thats not so easy. Glen Canner, senior economist with the Federal Reserve Board in Washington, says there isnt any one model for regression analysis being applied to banks across the board.
We take into account the underwriting practices and policies of each bank we examine, he said. We use a core set of variables that Im certain any bank would agree was appropriate. And regression analysis is only a part of the process; matches and file audits play an important part as well.