[The methods of Cochrane and Orcutt orr Hildreth and Lu to correct linear regressions for first-order autoregression in the disturbances, as usually implemented, underestimate the standard errors of the regression coefficients whenever a lagged dependent variable is included. A convenient transformation is derived from the information matrix to remove this bias. The asymptotic standard error of the estimated serial coefficient is a useful coproduct of the analysis.]

A method for optimization in nonlinear stochastic models is proposed in this paper, and applied to study monetary and fiscal policy in the St. Louis econometric model. Essentially, the method is to simulate the model in a number of stochastic simulations in which the coefficients of the model are treated as random, to use the results of the stochastic simulations to find parsimonious representations of the time form of the policy multipliers by estimating autoregressive moving-average regressions for the effects of policy on the relevant endogenous variables, and then to use these equations in computing optimal policy. The method is applied to the study of monetary and fiscal policy for the St. Louis model over a 60-period horizon with encouraging and sensible results.

The effects of the length and variability of lags on the conduct of stabilization policy are studied using linear difference equation models. It is shown that long lags by themselves do not impair the effectiveness of active stabilization policy and in fact require its more vigorous use. Variability in lags does reduce the effectiveness of stabilization policy and requires its more cautious use; the relative performance of alternative policies depends upon whether the variability of lags is confined to the direct effects of policy or occurs in the dynamics of the whole system. The cautious use of active stabilization policy is shown to be always stabilizing relative to performance under an inactive policy.