Paper: arXiv 2305.09783

Authors: Benjamin Fan, Edward Qiao, Anran Jiao, Zhouzhou Gu, Wenhao Li, Lu Lu

Abstract

We develop a methodology that utilizes deep learning to simultaneously solve and estimate canonical continuous-time general equilibrium models in financial economics. We illustrate our method in two examples: (1) industrial dynamics of firms and (2) macroeconomic models with financial frictions. Through these applications, we illustrate the advantages of our method: generality, simultaneous solution and estimation, leveraging the state-of-art machine-learning techniques, and handling large state space. The method is versatile and can be applied to a vast variety of problems.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 3.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper is mathematically dense, featuring continuous-time general equilibrium models, PDEs (HJB, Kolmogorov), and detailed deep learning theory (PINNs, loss functions, gradients), but lacks any empirical backtesting, datasets, or implementation code, focusing solely on theoretical methodology.

Research Flowchart

  flowchart TD
  Start["Research Goal: Solve & Estimate Continuous-Time Macro-Finance Models"] --> Method["Methodology: Deep Learning Framework"]
  Method --> Data["Inputs: Model Specifications & Economic Data"]
  Data --> Comp["Computation: Simultaneous Solution & Estimation via Neural Networks"]
  Comp --> Ex1["Application 1: Industrial Dynamics of Firms"]
  Comp --> Ex2["Application 2: Macroeconomic Models with Financial Frictions"]
  Ex1 --> Outcomes["Key Outcomes: Generality, Large State Space Handling, Simultaneous Solving/Estimation"]
  Ex2 --> Outcomes

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a methodology that uses deep learning, specifically physics‑informed neural networks (PINNs), to simultaneously solve and estimate canonical continuous‑time general equilibrium macro‑finance models.

Source sentence
We develop a methodology that utilizes deep learning to simultaneously solve and estimate canonical continuous-time general equilibrium models in financial economics.

Assumes 3

  • Entry incurs a one‑time cost ce and each entrant starts with the same initial equity e0. model
    Source sentence
    We assume that entry incurs a one-time cost ce, and each entrant has the same initial equity of e0.
  • The stationary bank distribution g(e,z) exists and is normalized. model
    Source sentence
    which means the following assumption holds. Assumption 1.
  • The Kolmogorov forward operator satisfies regularity conditions: σ²(x) is uniformly elliptic and coefficients are smooth with linear growth. model
    Source sentence
    1. σ2(x) is uniform elliptic, i.e.,⃗ yT [σ2(x)]⃗y ≥α|⃗y|2, ∀⃗y ∈Rn. 2.

Data

Synthetic data only.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Training loss decreases steadily and the endogenous variable rl converges to its true value within about 20,000 iterations. performance
    Source sentence
    The training loss decreases steadily (Fig. 2A), and the endogenous variable rl quickly converges to its true value within around 20000 iterations (Fig. 2B).
  • The L2 relative error for the value function v(e,z) is 0.54%. statistical
    Source sentence
    In fact, the L2 relative error for v(e, z) is 0.54%.
  • The L2 relative error for the cumulative distribution of g(e,z) is 4.32%. statistical
    Source sentence
    Here, we compute the L2 relative error of gc, which is defined as the CDF of g, i.e., gc(e, z) = R z z R e 0 g(e′, z′)de′dz′, to be 4.32%.
  • The final predicted values of rl and m after training have very high accuracy (0.01% and 2.12% error respectively). theoretical
    Source sentence
    The final predicted values of rl and m after training are listed in Table 3, and both have very good accuracy (0.01% and 2.12%, respeticely).

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 13 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (2 by independent authors)

Code

https://github.com/lu-group/pinn-macro-finance — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.