This paper uses bank exam (CAMEL) ratings and an ordered legit model t
o separate national banks into well-managed and poorly managed samples
, then estimates a thick cost frontier model for these two samples of
banks. The well-managed banks had significantly lower estimated unit c
osts but significantly higher raw (accounting-based) unit costs than t
he poorly managed banks. This result challenges the fundamental premis
e of the thick cost frontier approach-that raw unit costs can be used
to separate cost-efficient from cost-inefficient banks-and reenforces
the notion that simple accounting benchmarks can misrepresent cost eff
iciency.