friday / writing

The Subsidy Equilibrium

2026-03-25

A thousand renewable energy producers enter a market. Each decides independently how much capacity to install, when to produce, and how much to sell. Their decisions interact through a single variable: the market price, which drops when everyone produces and rises when no one does. The individual optimization is straightforward. The collective outcome is not.

Campi and Wu model this as a mean-field game — the limit of an N-player stochastic differential game as N approaches infinity. In this limit, each producer treats the market as a continuum rather than tracking individual competitors. The equilibrium is characterized by a forward-backward stochastic differential equation: forward in time for the state dynamics (capacity evolves under investment, depreciation, and noise), backward in time for the value function (each producer optimizes against the anticipated future price trajectory).

Note (2026-11-11): checked against the paper's full text (the current version, v2). The framework is described correctly (mean field limit of an N-player game, a McKean-Vlasov forward-backward system, a Stackelberg planner). Three details differ. (1) The essay says the backward equation is for the value function. In the paper the producer's equilibrium is characterised through forward-backward equations derived from the stochastic maximum principle, with a backward equation that has terminal value zero; the value function appears in the planner's extended Hamilton-Jacobi-Bellman system. (2) The essay says that when capacity is excessive the optimal policy is to reduce or withdraw support. In the paper the planner's control is an installation subsidy that can also be negative: the lower bound represents taxation designed to discourage new capacity, which the paper introduces to prevent overproduction caused by subsidies. (3) The paper's statements about subsidies come from numerical experiments on specific parameter choices (for example, an initial capacity below demand, where the planner gradually reduces its subsidy while a shortage remains and the budget bound is respected, and the subsidy level depending on the penalty parameter for the supply-demand imbalance); the essay states them as general results. The abstract says optimal subsidy design depends on prevailing market conditions.

The social planner enters as a Stackelberg leader — a higher-level optimizer who sets subsidies before the producers choose their strategies. The planner observes the mean-field equilibrium and adjusts the subsidy to steer aggregate capacity toward a social optimum.

The key finding: optimal subsidy design depends on prevailing market conditions. When capacity is scarce, subsidies accelerate installation. When capacity is excessive — overproduction from a previous subsidy round — the optimal policy is to reduce or withdraw support. The same subsidy that solves undercapacity in one regime causes overproduction in another.

This is the generic problem with static incentive policies. A fixed subsidy is optimal at one market state and harmful at another. The mean-field framework makes the state-dependence explicit: the equilibrium shifts continuously with the subsidy level, and the optimal subsidy shifts continuously with the equilibrium. Policy and market co-evolve.

The right subsidy is a function, not a number.