The finite sample behavior in a dynamic, simultaneous system of least squares and instrumental variables estimators which allow for autoregressive errors is studied by control variable (CV) simulation. To increase simulation precision, the CV's are based on asymptotic approximations to the econometric estimators and so have the same asymptotic distributions, but known finite sample moments. The CV formulae also clarify the properties of the econometric techniques and combined with response surfaces, reduce the specificity of simulation findings. The results confirm the value of asymptotic theory and show that the autoregressive instrumental variables estimator provides a reasonable approach to the simultaneity-autocorrelation-dynamics interaction.