Some Evidence on the Consistency of Banks’Internal Credit Ratings By Mark Carey Discussion: Xavier Freixas

Summary The paper exploits a carefully designed database, the Loan Loss Database and allows to identify and to predict disagreement in internal credit ratings. But consistency does not mean accuracy

Main results Overall, lenders are neither optimistic nor pessimistic (computed on the required capital so as to avoid an important biais on the riskier borrowers) There is no convergence over time (after one year)

Main results (II) Rating disagreement across lenders are not very predictable from borrower characteristics. Disagreements are less likely for large borrowers and for borrowers that have not drawn down much on their lines of credit. Small disagreements are important for high quality borrowers.

Non significative variables: Surprisingly, the following variables are not significative in explaining internal ratings disagreement –Industry –Location –Time since origination –agency rating

Motivation for the paper An evolutionary view of the market conflicts with prudential regulation. Banks that enter a new segment of business are likely to be less accurate: Learning by lending assumption. Relationship banking and ex post monopoly of information. Incentives to missreport: Aghion Bolton and Fries(2000) and Mitchell (2000)

Reinterpreting the results Regarding overoptimistic lenders, results are reassuring. Still, the lender dummy is highly significative in the regression. This may be due because conditionally on the regression variables some lenders are optimistic and other pessimistic. Regarding the the learning by lending assumption results indicate (unconditional) lack of convergence after one year.

A suggestion on ratings There is more agreement for high risk classes. If confirmed, this suggests that we could refine the risk class partition