The formulation of optimal policy in linear rational expectations models is studied using methods analogous to the classical design techniques utilized in linear systems engineering. Specifically, the policy-maker's present-value-like objective function is converted, using the convolution transform, to an equivalent frequency domain, "spectral utility" function. Then the residue calculus and Wiener-Hopf methods are used to maximize spectral utility through the choice of a complex function which represents a sequence of distributed lag coefficients to be applied to current and past values of instrument variables. The solution to this problem is a closed form expression for the decision rule of the dominant player in a particular type of linear-quadratic dynamic game.