A consistent, closed-loop solution for infinite-horizon, linear- quadratic, dynamic Stackelberg games (Q1108211)

We work out Kydland's suggestion to derive a (time-) consistent, closed- loop, solution in discrete-time, infinite-horizon, linear-quadratic, dynamic, 2-player, Stackelberg games by varying undetermined coefficients in players' linear, constant-coefficient, feedback policy rules [see \textit{F. Kydland}, J. Econ. Theory 15, 307-324 (1977; Zbl 0375.90089)]. The problem of consistency in linear-quadratic dynamic Stackelberg games was discussed, e.g., by \textit{M. Simaan} and \textit{J. B. Cruz} [J. Optimization Theory Appl. 11, 613-626 (1973; Zbl 0245.90039)], \textit{F. Kydland} [loc. cit. and Int. Econ. Rev. 16, 321-335 (1975; Zbl 0335.90061)] and \textit{F. Kydland} and \textit{E. C. Prescott} [J. Polit. Econ. 85, 473-491 (1977)]. Using a form of matrix-differential calculus [cf., \textit{J. R. Magnus} and \textit{H. Neudecker}, ``Matrix differential calculus with applications in statistics and econometrics'', Wiley, New York, to appear], we derive nonlinear, algebraic (nonrecursive), Riccati-like solution equations for such games. With the same method, we also derive analogous continuous- time results. For both the discrete- and continuous-time cases, we propose analogous, numerical, solution algorithms, which proceed with recursive versions of the algebraic Riccati-like equations. The algorithms are illustrated with a duopoly model. In the discrete-time case, the solution extends the dynamic-programming (d-p) feedback solution in this game. The present solution is more appealing than the d-p solution because in the present solution the leader is better off, yet, like in the d-p solution, the equilibrium is consistent. The leader is better off in the present solution because the players also optimize with respect to what may be called anticipative elements of control. ``Anticipative control'' is the ability of the leader, midstream in the game, to indirectly influence the evolution of the state vector backwards in time, by taking the follower's reactions into account. It is this effect which causes dynamic Stackelberg equilibria to be inconsistent, although, here inconsistency is avoided by assuming that the coefficient matrices in policy rules are independent of the initial state vector. By virtue of the principle of optimality upon which it is based, d-p cannot take such anticipative control effects into account.