In a principal-agent model with multiple performance measures, we show that
the principal benefits by choosing ex post which variables will be monitor
ed. If it is too costly for one type of agent to mimic all performance meas
ures expected from another type, the principal can hope to catch the agent
on the wrong foot if the agent tries to misrepresent his type. For cases of
small asymmetry of information, the principal can implement the first best
contract. For more serious asymmetries of information, the first best is n
ot implementable. Then the low type may be required to overproduce, which i
s in contrast to the traditional result of second best contracting. We also
obtain a ranking of monitoring instruments according to the frequency of t
heir use. (C) 2001 Elsevier Science BV All rights reserved.