Continuous-Time Heterogeneous Agent Models with Recursive Utility and Preference for Late Resolution

This paper establishes the existence and uniqueness of a constraint viscosity solution to the Hamilton-Jacobi-Bellman equation and investigates the solution properties of continuous-time heterogeneous agent models formulated as mean field games with recursive utility and a preference for late resolution of uncertainty.

The Gist

Imagine a vast, bustling city where millions of people are trying to manage their money. Some days they earn a lot, some days they earn very little, and sometimes they get sick or lose their jobs. Everyone is trying to figure out: How much should I spend today, and how much should I save for tomorrow?

This paper is like a massive, high-tech simulation of that city. It tries to solve a very difficult puzzle: How do we mathematically predict the behavior of millions of different people when they are worried about the future and hate uncertainty?

Here is the breakdown of the paper's ideas using simple analogies:

1. The "Time Traveler" vs. The "Wait-and-See" Person

In most old economic models, people were assumed to be simple. They just wanted to maximize their happiness today plus their happiness tomorrow, regardless of when they found out if they were going to be rich or poor.

This paper introduces a more realistic character: The "Late Resolution" Lover.

• The Analogy: Imagine you have a lottery ticket. You can either find out today if you won (Early Resolution) or wait until next year to find out (Late Resolution).
• The Twist: Most people hate uncertainty. They want to know now so they can stop worrying. But this paper studies a specific type of person who actually prefers to wait. Why? Because they are so good at planning and so risk-averse that they feel better keeping their options open and not locking themselves into a "bad" future scenario too early. They want to keep the suspense alive a bit longer to adjust their plans perfectly.

2. The Two Types of People (The "Heterogeneous" Crowd)

The city isn't made of clones. It has two main types of workers:

• Type A (The Struggler): Earns a low income.
• Type B (The Earner): Earns a high income.
• The Switch: These people randomly switch between being a Struggler and an Earner (like getting a promotion or losing a job).

The paper asks: If everyone is trying to save money, but they are constantly switching between being rich and poor, what does the total pile of money in the city look like?

3. The "Debt Limit" Wall

There is a hard rule in this city: You cannot borrow more than a certain amount. Imagine a wall at the bottom of a hill. If you fall below that line, you are stuck. You can't go deeper into debt.

• The Problem: When people get close to this wall, they get terrified. They stop spending and start hoarding cash just to make sure they don't hit the wall. This is called Precautionary Saving.
• The Math Challenge: The authors had to prove that even with this scary "wall," there is a stable, predictable way for everyone to behave. They proved that a mathematical "solution" exists and is unique (meaning there is only one correct way the city behaves, not a chaotic mess).

4. The "Tug-of-War" (The Interest Rate)

The city has a central bank that sets the interest rate (the reward for saving).

• The Balance: If the interest rate is too low, people don't save enough, and the city runs out of capital. If it's too high, people save too much and stop spending, which hurts the economy.
• The Equilibrium: The paper shows how the interest rate naturally settles at a "Goldilocks" point where the total amount people want to save equals the total amount businesses want to borrow to build factories.

5. The "Blow-Up" Scenario

The authors discovered a dangerous tipping point.

• The Analogy: Imagine the interest rate is the water level in a bathtub. As the water rises, people get more comfortable. But if the water level gets too close to the "drain" (a specific mathematical limit where the interest rate equals the discount rate), the bathtub overflows.
• The Result: If the interest rate gets too high (too close to this limit), people stop saving entirely or save so aggressively that the total wealth in the city becomes infinite (mathematically "blows up"). This means the economy becomes unstable and no steady state exists.

6. The "Digital Twin" (Numerical Examples)

Finally, the authors didn't just do the math on paper; they built a computer model (a "Digital Twin" of the economy).

• They ran simulations with different personalities (some people are very risk-averse, some are less so).
• What they found:
  • When people are very risk-averse (scared of the future), they save more, especially when they are poor. This drives the interest rate down.
  • When people are willing to substitute spending over time (they don't mind spending less today to spend more later), they save less, and interest rates go up.

Summary: Why Does This Matter?

This paper is like a new, more sophisticated instruction manual for the world's central banks and economists.

1. It validates the math: It proves that even with complex human behaviors (like preferring to wait for bad news), the economy doesn't collapse into chaos; it finds a stable path.
2. It explains "Precautionary Savings": It gives a rigorous reason why people hoard cash when they are scared, even if it hurts the economy in the short term.
3. It warns of limits: It shows exactly how high interest rates can go before the system breaks down.

In short, the authors took a messy, uncertain world of millions of different people and showed us that, surprisingly, there is a hidden order to how they save, spend, and worry about the future.

Technical Summary

Here is a detailed technical summary of the paper "Continuous-Time Heterogeneous Agent Models with Recursive Utility and Preference for Late Resolution" by Yves Achdou and Qing Tang.

1. Problem Statement

The paper addresses the mathematical analysis of continuous-time heterogeneous agent models (specifically Aiyagari-Bewley-Huggett type models) where agents possess Epstein-Zin recursive utility. Unlike standard time-additive CRRA utility, Epstein-Zin utility allows for the separation of the coefficient of relative risk aversion ($\gamma$) and the elasticity of intertemporal substitution (EIS, $\psi$).

The specific focus is on the case where agents exhibit a preference for the late resolution of uncertainty. Mathematically, this corresponds to the condition:
$$\gamma \psi < 1$$
Given the paper's assumptions ($\gamma > 1$ and $\psi < 1$), this implies $\gamma \psi < 1$. In this regime, agents prefer to delay the resolution of uncertainty, which introduces significant nonlinearity and analytical challenges compared to the standard CRRA case ($\gamma = 1/\psi$).

The model involves:

• A continuum of agents with idiosyncratic income shocks (modeled as a two-state Poisson process).
• Incomplete markets with a borrowing constraint (debt limit).
• Agents solving a stochastic optimal control problem to maximize lifetime utility.
• A general equilibrium where the interest rate is determined by the aggregate capital supply and demand (Aiyagari model) or fixed (Huggett model).

2. Methodology

The authors employ tools from Mean Field Game (MFG) theory and the theory of viscosity solutions for Hamilton-Jacobi-Bellman (HJB) equations.

A. The HJB System

The agent's optimization problem leads to a weakly coupled system of HJB equations for the value functions $v_j(x)$ (where $j \in \{1, 2\}$ denotes income states):
$$\frac{\rho}{\theta} v_j(x) = \max_{c \ge 0} \{ F(c, v_j) + (rx + y_j - c)Dv_j \} + \lambda_j(v_{\bar{j}}(x) - v_j(x))$$
where $\theta = \frac{1-\psi^{-1}}{1-\gamma} > 1$. The Hamiltonian $H(x, y, v, p)$ is derived explicitly.

B. Constrained Viscosity Solutions

Since the state space is bounded from below by the borrowing limit $\underline{x}$, the authors utilize the concept of constrained viscosity solutions (introduced by Soner).

• Subsolution: Defined on the closed domain $[\underline{x}, \infty)$.
• Supersolution: Defined on the open domain $(\underline{x}, \infty)$.
  This framework is crucial for handling the boundary condition at the borrowing limit where the control (consumption) hits the constraint.

C. Analytical Techniques

1. Comparison Principles: The authors establish a strong comparison principle for sub- and supersolutions to prove uniqueness.
2. Convex Analysis: They utilize properties of semiconcave and semiconvex functions to prove higher regularity (specifically $C^1$ and $W^{2,\infty}_{loc}$) of the value functions.
3. Fokker-Planck-Kolmogorov (FPK) Equation: They analyze the stationary distribution of wealth, which satisfies a linear transport equation coupled with the HJB system via the optimal saving policies.
4. Asymptotic Analysis: They perform asymptotic expansions of the saving policies as wealth $x \to \infty$ and as the interest rate $r \to \rho$ to study the existence of invariant measures.

3. Key Contributions and Results

A. Existence and Uniqueness of the Value Function

• Theorem: Under the assumption $\gamma > 1, \psi < 1, \gamma\psi < 1$, there exists a unique bounded constrained viscosity solution to the HJB system.
• Regularity: The value function $v_j(x)$ is shown to be:
  • Strictly concave.
  • Continuously differentiable ($C^1$) on $(\underline{x}, \infty)$.
  • Locally $W^{2,\infty}$ (twice differentiable almost everywhere with bounded second derivatives).
• Optimal Policy: The optimal consumption and saving policies are derived explicitly from the value function and shown to be continuous.

B. Qualitative Properties of Saving Policies

The paper derives several non-trivial properties of the saving behavior:

1. Productivity Hierarchy: Agents with higher income ($y_2$) always have a higher value function than those with lower income ($y_1$), i.e., $v_2(x) > v_1(x)$.
2. Borrowing Constraint Behavior:
   • Unproductive agents ($y_1$) always save negatively ($s_1(x) < 0$) for $x > \underline{x}$ and $s_1(\underline{x}) = 0$.
   • The derivative of the value function for unproductive agents diverges to $-\infty$ at the borrowing limit ($\lim_{x \to \underline{x}} D^2 v_1(x) = -\infty$), indicating a "kink" in the value function.
3. Productive Agents' Savings:
   • Depending on parameters, productive agents ($y_2$) may save positively near the borrowing limit but eventually switch to dissaving as wealth increases.
   • The paper provides sufficient conditions for $s_2(x)$ to be positive or negative.

C. Existence of Equilibrium (MFG System)

• Aggregate Wealth Continuity: The aggregate capital supply $K(r)$ is shown to depend continuously on the interest rate $r$.
• Blow-up at $r = \rho$: The authors prove that as the interest rate approaches the discount rate ($r \to \rho$), the aggregate wealth $K(r)$ tends to infinity. Consequently, no stationary invariant measure exists when $r = \rho$.
• Equilibrium Existence: Using the intermediate value theorem and the continuity of $K(r)$, they prove the existence of an equilibrium interest rate $r^*$ such that $0 < r^* < \rho$ for the Aiyagari and Huggett models, provided a specific condition on parameters (related to income dispersion and transition rates) is met.

D. Numerical Validation

The paper includes numerical experiments using finite difference and semi-Lagrangian schemes.

• Results confirm that higher risk aversion ($\gamma$) leads to higher precautionary savings near the borrowing constraint.
• Higher EIS ($\psi$) leads to lower equilibrium interest rates (as agents are more willing to consume).
• The numerical schemes remain stable even in parameter regimes slightly outside the strict theoretical assumptions (e.g., $\gamma \psi \ge 1$), suggesting robustness.

4. Significance and Implications

• Mathematical Rigor: This work extends the theory of constrained viscosity solutions to the complex setting of recursive utility with preference for late resolution, a case that was previously difficult to handle due to the non-separability of risk aversion and EIS.
• Economic Insight: It provides a rigorous foundation for studying macroeconomic phenomena (like precautionary savings and wealth inequality) in models where agents' attitudes toward risk and intertemporal substitution are distinct.
• Policy Relevance: The analysis of the "late resolution" preference is crucial for understanding asset pricing and the impact of uncertainty shocks in modern macro-finance, particularly in contexts like climate change economics or disaster risk modeling where such preferences are often assumed.
• Future Directions: The authors note that the case $\gamma \psi > 1$ (preference for early resolution) remains open and requires different analytical tools, as the current viscosity solution theory relies heavily on the specific structure of the $\gamma \psi < 1$ case.

In summary, the paper successfully bridges advanced PDE theory and macroeconomic modeling, establishing a complete existence and uniqueness theory for heterogeneous agent models with Epstein-Zin utility under borrowing constraints.