In practice, the number of passive assets must be limited in some fashion. Our empirical design includes p passive assets, consisting of k benchmarks and m nonbenchmark assets, and the benchmarks are associated with popular asset pricing models. Suppose one admits the possibility that the benchmarks do not price the nonbenchmark assets exactly, that is ®N 6 = ±A, the intercept in (2), is a better measure of skill, in that it is de¯ned with respect to the more inclusive set of passive assets. Of course, that measure might still be nonzero for passive assets omitted from the set of p. The point is simply that.
