Noise-state game explorer

The loop, and the cuts. Every tractable approach beyond the Kyle line severs the loop somewhere: on the privacy of beliefs, on what actions can teach, on the weight of one player, on the flow from observations to beliefs, or on time itself. Cutting it tends to hand back the separation principle, where learning and acting can be treated apart, and with it removes the interaction between information and incentives the model set out to study.

The noise-state. (illustrated) Each player keeps estimates of the primitive shocks instead of belief hierarchies about the endogenous state. Every linear quantity is a deterministic mean plus an impulse-response kernel against those shocks, and taking a player's expectation changes the integrator and nothing else, so the tower of expectations closes. Under perfect information the estimates are the shocks themselves and the kernels are the model's ordinary impulse responses.

The information wedge. (illustrated) The shadow price of changing another player's beliefs. It breaks separation and vanishes when the loop is cut. In the plots it is the part of a player's first-order condition that exists only because the others read its actions in their signals.

The solver. Equilibrium is a deterministic fixed point in the kernels. It iterates best responses, with Anderson acceleration, on piecewise Chebyshev nodes in shock age, or on a triangle of dates and shock times for a finite horizon, until the kernels stop moving.

The games. The tabs follow the chapters: the finite-horizon baseline (1), the stationary game (3), Kyle–Back trading with several informed traders (4), a cycle of local markets (5), and monitored deviations, naive against privy (6). Chapter 2, on delayed signals, has no tab yet; the model editor takes any game the solver supports.