Existence and uniqueness results for a mean-field game of optimal investment

This paper establishes the existence and uniqueness of equilibrium for a stochastic mean-field game of optimal investment involving identical firms interacting through a time-dependent price, covering both finite and infinite time horizons as well as a deterministic counterpart.

The Gist

Imagine a massive, bustling marketplace filled with thousands of identical bakeries. Each bakery wants to make as much money as possible, but they face a tricky problem: if they all bake too much bread, the price of bread crashes. If they bake too little, they miss out on profits.

This paper is a mathematical recipe for figuring out exactly how much each bakery should invest in new ovens and how much bread they should bake to reach a perfect "balance point" (called an equilibrium) where no one wants to change their strategy.

Here is the story of that recipe, broken down into simple parts:

1. The Setting: The Baker's Dilemma

In this world, every bakery has a "production capacity" (how much bread they can bake).

• The Decay: Ovens get old and break down over time (this is the "depreciation").
• The Investment: To keep up, bakeries can buy new ovens (this is the "investment"). But buying ovens is expensive, and the cost goes up sharply if you try to buy too many at once.
• The Market Price: The price of bread isn't fixed. It depends on the total amount of bread all bakeries are baking. If everyone bakes a lot, the price drops. If everyone bakes a little, the price stays high.

2. The "Mean-Field" Magic

Usually, to solve this, you'd have to write down a complex equation for every single bakery, tracking how they react to every other bakery. That's impossible with thousands of players.

Instead, the authors use a clever trick called Mean-Field Games. Imagine that instead of looking at individual bakeries, we look at the "average" bakery.

• Each bakery assumes the price of bread is determined by the average production of the whole crowd.
• The bakery asks: "If I assume everyone else is baking X amount, what is the best amount for me to bake?"
• The "Equilibrium" is found when the bakery's best guess matches reality: the amount the bakery actually bakes turns out to be exactly the average amount everyone else is baking.

3. The Two Scenarios: Short Sprint vs. Long Marathon

The paper solves this puzzle for two different timeframes:

• Finite Horizon (The Sprint): The game ends at a specific date (e.g., 10 years from now). As the deadline approaches, the bakeries stop investing because there's no time to pay back the cost of new ovens. They just let their ovens wear out.
• Infinite Horizon (The Marathon): The game goes on forever. Here, the bakeries settle into a steady rhythm. They find a "Goldilocks" level of production where the price and costs balance out perfectly, and they stay there forever.

4. The Surprising Discovery: Noise Doesn't Matter

In the real world, things are unpredictable. Maybe a storm ruins a shipment of flour, or a machine breaks down unexpectedly. In math, this is called "volatility" or "noise."

The authors found something fascinating: The average behavior of the market doesn't care about the chaos.
Even though individual bakeries might have bad days or good days due to random luck, the average path of the entire market is perfectly smooth and predictable. It's like a crowd of people walking through a foggy park; while individual people might stumble or swerve, the crowd as a whole moves in a straight, predictable line.

5. How They Solved It (The "Secret Sauce")

The math behind this is heavy, but the logic is like solving a puzzle in two steps:

1. Step 1 (The Individual): "If I know the average price is going to be $P$, how much should I invest?" The authors found a clear formula for this.
2. Step 2 (The Crowd): "If everyone follows that formula, what does the average price actually become?"
3. The Loop: They kept adjusting the guess until the "assumed price" and the "actual price" matched perfectly.

They proved that there is only one correct answer. There is no confusion; the market will always find this specific balance point, whether the game is short or long.

6. The Takeaway

This paper gives us a powerful tool to understand how large groups of competitors (like energy companies, tech firms, or farmers) behave when they are all trying to maximize profit.

• For the Short Term: It tells us that as a deadline approaches, investment slows down, and production naturally declines.
• For the Long Term: It tells us that markets naturally settle into a stable, predictable rhythm, regardless of how chaotic the individual players' days might be.

In short, the authors have written the "instruction manual" for how a chaotic crowd of competitors can find a calm, stable, and unique way to coexist and thrive.

Technical Summary

Here is a detailed technical summary of the paper "Existence and Uniqueness Results for a Mean-Field Game of Optimal Investment" by Calvia, Federico, Ferrari, and Gozzi.

1. Problem Formulation

The paper investigates a stochastic Mean-Field Game (MFG) modeling optimal investment decisions by a continuum of identical firms competing in a Cournot-like market.

• State Dynamics: The production capacity $X_t$ of a representative firm evolves according to a geometric Brownian motion with depreciation and investment control:
  $$dX_t = -\delta X_t dt + \sigma X_t dB_t + u_t dt$$
  where $\delta > 0$ is the depreciation rate, $\sigma \geq 0$ is the volatility, $B_t$ is a Brownian motion, and $u_t \geq 0$ is the irreversible investment rate (control).
• Objective Function: The representative firm aims to maximize the expected discounted net profit over a time horizon $T \in (0, +\infty]$:
  $$J_T(\xi, u) = \mathbb{E}\left[ \int_0^T e^{-\rho t} \left( X_t q_t^{-\beta} - \frac{1}{2}u_t^2 \right) dt \right]$$
  Here, $\rho$ is the discount rate, $\beta > 0$ is a demand elasticity parameter, and $q_t$ represents the average production capacity of the entire population of firms.
• Mean-Field Interaction: The interaction is modeled through the market price. The price is determined by an isoelastic inverse demand function where the price depends on the aggregate production. Specifically, the instantaneous profit term $X_t q_t^{-\beta}$ arises from a limit of an $N$-player game where the price depends on the average capacity $\frac{1}{N}\sum X^i_t$.
• Equilibrium Definition: A mean-field equilibrium is a pair $(\hat{u}, \hat{q})$ such that:
  1. $\hat{u}$ is the optimal control for the representative firm given the path $\hat{q}$.
  2. $\hat{q}_t = \mathbb{E}[X_t^{\hat{u}}]$ (consistency condition), meaning the assumed average production matches the actual expected production under the optimal strategy.

The paper analyzes both finite time horizons ($T < +\infty$) and infinite time horizons ($T = +\infty$), and also considers the deterministic counterpart ($\sigma = 0$).

2. Methodology

The authors employ a fixed-point approach rather than the standard PDE-based methods (Hamilton-Jacobi-Bellman and Fokker-Planck equations) or the Master Equation approach. The methodology proceeds in three distinct steps:

1. Optimal Control for Fixed Path: For a given deterministic trajectory of average production $q = (q_t)$, the authors solve the representative firm's optimal investment problem.

   • Due to the linear-quadratic structure of the state and cost (once $q$ is fixed), the problem admits an explicit, deterministic optimal control $\hat{u}(q)$.
   • The optimal control is found to be $\hat{u}_t = z_t$, where $z_t$ is a deterministic function derived from the future path of $q$.

2. Consistency Condition: The authors compute the expected state trajectory $\mathbb{E}[X_t^{\hat{u}}]$ induced by the optimal control.

   • Using the explicit solution of the SDE and Fubini's theorem, they derive an expression linking the expected state to the control.
   • Imposing the consistency condition $\hat{q}_t = \mathbb{E}[X_t^{\hat{u}}]$ transforms the problem into finding a fixed point.

3. Integral Equation Characterization: The fixed point condition reduces to solving a nonlinear integral equation (or an equivalent integro-differential equation) for the equilibrium average production $\hat{q}$.

   • Finite Horizon: The equation involves a backward-in-time integral term, which is non-standard.
   • Infinite Horizon: The equation is analyzed on $[0, +\infty)$ with specific decay conditions.

Technical Tools:

• Finite Horizon: Existence is proven using Schauder's Fixed Point Theorem applied to a compact set of functions, relying on a priori estimates. Uniqueness is established via a contradiction argument exploiting the monotonicity of the integral operator.
• Infinite Horizon: Existence is proven using the Fréchet-Kolmogorov compactness theorem in weighted $L^p$ spaces, constructing the solution as a limit of finite-horizon solutions. Uniqueness is shown by analyzing the associated second-order ODE and proving that solutions must be monotone and converge to a unique steady state.

3. Key Contributions

1. Non-Linear MFG Analysis: Unlike many existing MFG results that focus on linear-quadratic (LQ) games, this paper addresses a problem where the profit functional is non-linear in the mean-field term ($q^{-\beta}$). This prevents the use of standard LQ-MFG techniques and requires ad-hoc analysis.
2. Explicit Characterization: The authors provide a complete characterization of the equilibrium as the unique solution to a specific nonlinear integral equation. They derive explicit formulas for the optimal control and the value function in terms of the equilibrium path.
3. Deterministic Equivalence: A significant finding is that the equilibrium average production and optimal control are deterministic and independent of the volatility parameter $\sigma$. This allows the authors to rigorously prove that the stochastic MFG equilibrium coincides with the deterministic MFG equilibrium (where $\sigma=0$).
4. Infinite Horizon Stability: For the infinite horizon case, the paper proves that the equilibrium average production converges monotonically to a unique stationary level $y_\infty$, regardless of the initial condition.
5. Numerical Validation: The paper includes numerical experiments comparing finite and infinite horizons, demonstrating the convergence of finite-horizon solutions to the steady state and visualizing the impact of volatility on individual firm trajectories versus the aggregate mean.

4. Main Results

• Existence and Uniqueness: The paper establishes the existence and uniqueness of a mean-field equilibrium for both finite and infinite time horizons.
  • Finite Horizon: The equilibrium is the unique solution to the integral equation (4.1) (or the ODE system 4.4/4.5).
  • Infinite Horizon: Under the condition $\beta < 1 + \rho/\delta$, a unique equilibrium exists. The average production $\hat{q}_t$ is strictly monotone and converges to the stationary solution $y_\infty = [\delta(\rho+\delta)]^{-1/(\beta+1)}$.
• Independence from Volatility: The equilibrium aggregate path $\hat{q}$ and the optimal control $\hat{u}$ do not depend on the diffusion coefficient $\sigma$. The stochasticity only affects the variance of individual firm capacities, not the aggregate deterministic equilibrium path.
• Convergence: As the finite time horizon $T \to +\infty$, the sequence of finite-horizon equilibria converges to the unique infinite-horizon equilibrium in the weighted $L^1$ space.
• Deterministic Counterpart: The results hold identically for the deterministic model ($\sigma=0$), confirming that the mean-field interaction drives the system to a deterministic equilibrium even in the presence of noise.

5. Significance

• Economic Insight: The model provides a rigorous framework for understanding investment dynamics in competitive markets with a large number of firms. The result that the aggregate equilibrium is deterministic despite stochastic firm-level shocks suggests that market forces (mean-field interaction) effectively average out individual risks, leading to stable macroeconomic predictions.
• Methodological Advancement: The paper demonstrates that for specific classes of MFGs with scalar interactions and non-linear profit functions, one can bypass the complex PDE systems (HJB/Fokker-Planck) and solve the problem directly via integral equations and fixed-point arguments. This offers a more tractable alternative for problems where PDE approaches are analytically intractable.
• Policy and Strategy: The explicit characterization of the steady state $y_\infty$ allows for sensitivity analysis. The authors show that higher discount rates or depreciation rates reduce the long-term production capacity, while demand elasticity ($\beta$) plays a crucial role in strategic capacity limitation.

In summary, this paper provides a robust mathematical foundation for optimal investment games in large populations, proving that unique, stable, and deterministic equilibria exist even in stochastic environments with non-linear market interactions.