Bankers and regulators are in conflict over how to verify the accuracy of the computer models used to set capital levels. The models must be accurate, they agree.
But they dispute how the industry should verify these programs, which use historical data and market forecasts to predict the performance of a bank's investment portfolio.
Regulators last month proposed requiring banks to compare computer model predictions against actual results. If the two differed more than four times a year, then the bank would have to boost capital levels.
"This ties the accuracy of a bank's model to the capital charge," said Richard Spillenkothen, director of banking supervision and regulation at the Federal Reserve Board. "It provides an incentive for a bank to have accurate models to measure risk."
Bankers, writing in comment letters due by April 8, have said this approach - known as back-testing - is unfair. They demanded a more flexible process.
The back-testing dispute is part of a larger market risk proposal the agencies are pushing. Regulators in July 1995 proposed tying a bank's capital requirements to its exposure to interest rate swings.
They gave banks the choice of using their own models or a program written by the international Basel Committee on Banking Supervision to determine their interest rate risk. The goal is to let banks with safer portfolios hold less capital.
Bankers applauded the decision to let them use internal models, saying this would reduce compliance costs. But the Fed, the Office of the Comptroller of the Currency, and the Federal Deposit Insurance Corp. decided to amend the plan in March to include the back-testing rule.
Market risk models produce a "value at risk" figure, which shows a bank how much money it could lose or gain on a given day. These models are supposed to be 99% accurate, meaning the value at risk will exceed the model's prediction only once per 100 times.
The back-testing proposal requires a bank to verify that the gains or losses it actually incurred on a given day were within the range set by the model.
Banks would have to tally these results four times a year when completing call reports. The proposal allows an institution to be wrong up to four times during the preceding 250 trading days without penalty. After four misses, however, a bank would have to increase capital unless it could prove that an unusual event, such as a sudden spike in interest rates, had caused the model to fail.
Bankers, while praising the concept of back-testing, said the proposal contains three major flaws.
First, industry officials said banks routinely reprogram their models once mistakes are uncovered. So regulators would be punishing a bank for a program it no longer uses, wrote Robert Strand, a senior economist at the American Bankers Association.
David H. Sidwell, controller for Morgan Guaranty Trust Co. of New York, recommended scrapping the punishment approach altogether. Instead, he wrote, agencies should judge the bank on how well it adapted its model to changing market conditions.
Second, bankers said the back-testing approach is so complex that many banks might opt to use the Basel Committee's simpler market risk model. This approach, which generally duplicates models already in use at many banks, would not have to be verified.
Finally, bankers said income from customer fees can distort the amount a bank makes or loses each day. "Inclusion of these sources of gains or losses can negate the value of back-testing because theoretically the gains or losses are not entirely the result of market risk," wrote David J. Vitale, vice chairman of First Chicago NBD Corp.
The alternative, Mr. Vitale wrote, is to use hypothetical gains and losses, which would exclude customer fees.
Morgan's Mr. Sidwell said it could cost a fortune to separate these data from overall gains or losses. "Any attempt to isolate this component would be artificial and pure regulatory burden," Mr. Sidwell wrote.
Mr. Vitale, conceding there is no easy solution to this problem, recommended allowing banks to use either method.
The Fed's Mr. Spillenkothen said the agencies would review the comments during the next several months. A final rule is expected by yearend.
