Paper: arXiv 2512.14967
Authors: Felipe J. P. Antunes, Yuri F. Saporito, Sebastian Jaimungal
Abstract
We present a novel numerical method for solving McKean-Vlasov forward-backward stochastic differential equations (MV-FBSDEs) with common noise, combining Picard iterations, elicitability and deep learning. The key innovation involves elicitability to derive a path-wise loss function, enabling efficient training of neural networks to approximate both the backward process and the conditional expectations arising from common noise - without requiring computationally expensive nested Monte Carlo simulations. The mean-field interaction term is parameterized via a recurrent neural network trained to minimize an elicitable score, while the backward process is approximated through a feedforward network representing the decoupling field. We validate the algorithm on a systemic risk inter-bank borrowing and lending model, where analytical solutions exist, demonstrating accurate recovery of the true solution. We further extend the model to quantile-mediated interactions, showcasing the flexibility of the elicitability framework beyond conditional means or moments. Finally, we apply the method to a non-stationary Aiyagari–Bewley–Huggett economic growth model with endogenous interest rates, illustrating its applicability to complex mean-field games without closed-form solutions.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical tools like McKean-Vlasov FBSDEs, elicitability theory, and pathwise loss functions, indicating high math complexity. It demonstrates empirical rigor with neural network implementations, validation on models with analytical solutions, and application to complex economic models, showing backtest-ready methodology.
Research Flowchart
flowchart TD
A["Research Goal: Develop & Validate Deep Learning Method for MV-FBSDEs with Common Noise"] --> B["Methodology: Picard Iterations + Elicitability + Deep Learning"]
B --> C{"Data/Input: MV-FBSDE Formulations<br>Systemic Risk Model / Aiyagari-Bewley-Huggett Model"}
C --> D["Computational Process: Train NN to approximate Backward Process & Conditional Expectations<br>Loss: Path-wise Elicitable Score"]
D --> E{"Outcome: Accurate Solution Recovery<br>Validated on Analytical Systemic Risk Model"}
E --> F["Key Finding: Flexible Framework<br>Extends to Quantile Interactions & Non-Stationary Economic Models"]
F --> G["Conclusion: Efficient Method for Complex Mean-Field Games"]
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 novel numerical method that combines Picard iterations, elicitability, and deep learning to solve McKean‑Vlasov forward‑backward SDEs with common noise, using neural networks to approximate both the backward process and the conditional expectations without nested Monte Carlo.
Source sentence
We present a novel numerical method for solving McKean-Vlasov forward-backward stochastic differential equations (MV-FBSDEs) with common noise, combining Picard iterations, elicitability and deep learning. The key innovation involves elicitability to derive a path-wise loss function, enabling efficient training of neural networks to approximate both the backward process and the conditional expectations arising from
Assumes 1
- Lipschitz continuity and linear growth of the coefficients (µ, f, σ, σ0, G) with constants L, C, L0, ensuring existence and uniqueness of the MV‑FBSDE solution. model
Source sentence
Assumption 1 (Lipschitz Continuity). To guarantee existence and uniqueness of solution for (1)–(2), we assume there exist L, C, L0 > 0 such that the following properties hold: (i) Linear growth: ∥µ(t, 0, 0, 0, 0, s)∥+ ∥f(t, 0, 0, 0, 0, s)∥+ ∥σ(t, 0)∥+ ∥σ0(t, 0)∥≤C(1 + ∥s∥), for all t ∈[0, T] and s ∈Rp; (ii) Lipschitz continuous coefficients: ∥µ(t, x, y, z, z0, s) −µ(t, x′, y′, z′, z′0, s)∥+ ∥f(t, x, y, z, z0, s)
Data
- Systemic risk inter‑bank borrowing and lending model (CFS15) synthetic public
Source sentence
The first experiment solves the systemic risk banking model introduced in [CFS15]. This model can be solved analytically, hence it is useful as a benchmark against our numerical methodology.
- Quantile‑mediated interaction variant of the systemic risk model synthetic public
Source sentence
Building on this foundation, our second experiment modifies the systemic risk model by replacing interaction through the mean of the agents for an interaction through a quantile of the population. To test our algorithm in more challanging example, we apply the algorithm to a modified version of system (19) where the interaction term St is the α-quantile, with α = 60%, see Equation (13).
- Non‑stationary Aiyagari‑Bewley‑Huggett economic growth model synthetic public
Source sentence
Finally, we apply the method to a non-stationary Aiyagari--Bewley--Huggett economic growth model with endogenous interest rates, illustrating its applicability to complex mean-field games without closed-form solutions.
Costs
Not applicable
Finds 3
- The algorithm accurately recovers the analytical solution of the systemic risk model, matching true solution processes. theoretical
Source sentence
We validate the algorithm on a systemic risk inter-bank borrowing and lending model, where analytical solutions exist, demonstrating accurate recovery of the true solution. 4 CONCLUSION Our numerical experiments demonstrate the effectiveness of the proposed method across problems of varying complexity.
- The elicitability framework successfully handles quantile‑mediated interactions, demonstrating flexibility beyond conditional means. theoretical
Source sentence
We further extend the model to quantile-mediated interactions, showcasing the flexibility of the elicitability framework beyond conditional means or moments. The extension to quantile-mediated interactions showcases the flexibility of the elicitability framework in handling statistics beyond conditional means.
- The method is applicable to a complex non‑stationary economic growth model without closed‑form solutions, illustrating its broader applicability. theoretical
Source sentence
Finally, we apply the method to a non-stationary Aiyagari--Bewley--Huggett economic growth model with endogenous interest rates, illustrating its applicability to complex mean-field games without closed-form solutions. Finally, the economic growth model illustrates the method’s applicability to more complex, economically meaningful problems where analytical solutions are unavailable.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 9 related papers). Coverage is partial, so this is neither replication nor refutation.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.