Should stochastic or non-stochastic exogenous variables be used in Monte Carlo experiments?

Most Monte Carlo studies of test properties use the same values of the exogenous variables in all replications. We show that a considerable improvement in precision can be obtained by drawing separate samples at each replication.