Full information maximum likelihood estimates of an up-dated and extended version of the Klein-Goldberger Model have been completed. These supersede some earlier estimates that made simplifying approximations and dealt only with linear equations. Simulation and ex post prediction tests are made for comparison among full information maximum likelihood, two stage least squares (using principal components of predetermined variables), and ordinary least squares estimates of the same model. The relative performance of different methods varies depending on whether the simulations are one period (re-setting of lagged dependent variables every period) or dynamic (solution from fixed initial conditions). Also, the test performance varies with the number of principal components used, the best number not necessarily being large relative to the number of predetermined variables.