Pith. sign in

REVIEW 3 major objections 6 minor 2 references

A dedicated neural network for the certainty equivalent lets deep learning solve recursive-utility dynamic programs, including occasionally binding constraints, via first-order and KKT residuals.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 02:51 UTC pith:HOJYQAUK

load-bearing objection Solid methods paper: separate CE network into FOC/KKT residual learning for constrained recursive-utility DP; residual evidence is real but soft, VFI only on one low-dim case. the 3 major comments →

arxiv 2607.09461 v1 pith:HOJYQAUK submitted 2026-07-10 q-fin.CP econ.EMmath.OCstat.ML

Deep Learning for Dynamic Programming with Recursive Utility Using First-order Conditions

Xianhua Peng , Wu Guo , Songyan Wang , Jianfei Zhu This is my paper
classification q-fin.CP econ.EMmath.OCstat.ML
keywords deep learningdynamic programmingrecursive utilitycertainty equivalentfirst-order conditionsKKT residualsEpstein–Zinoccasionally binding constraints
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Dynamic programs with recursive preferences (risk-sensitive, Epstein–Zin, robust control) are hard because a nonlinear certainty-equivalent of next-period value sits inside both the Bellman equation and the optimality conditions, and sample averages do not recover it. This paper claims that learning that state-control certainty equivalent with its own neural network, then training policy and multiplier networks by minimizing model-specific first-order and KKT residuals (while separate networks keep Bellman and certainty-equivalent consistency), yields accurate policies and values without grids, penalty methods, or problem-specific rewrites. On consumption-saving, small-noise robust-control, and DSGE examples, out-of-sample Bellman and optimality residuals are typically 10⁻⁴–10⁻³ over the relevant state region and match value-function-iteration benchmarks where those exist. Expected utility is recovered as the special case of a linear certainty equivalent, so the same architecture covers both settings. A sympathetic reader cares because many modern asset-pricing and macro models use recursive preferences and constraints that classical grid methods cannot scale to.

Core claim

Representing the state-control certainty equivalent by a dedicated neural network, and feeding that approximation into model-specific first-order and KKT residual losses for the policy and multipliers, produces a practical deep-learning solver for discrete-time recursive-utility dynamic programming that handles general equality and inequality constraints, including occasionally binding ones, with small residual diagnostics and close agreement with VFI when available.

What carries the argument

The certainty-equivalent first-order learning (CEFOL) scheme: a certainty-equivalent network C(s,c) that amortizes the nonlinear transformation f⁻¹(E[f(V(s′))]), combined with residual losses built from stationarity (including Euler), Fischer–Burmeister complementarity, and equality constraints, trained with target networks and delayed policy updates.

Load-bearing premise

That small out-of-sample residual diagnostics under nonconvex neural training, plus optional agreement with grid VFI on low-dimensional cases, are enough to treat the learned networks as accurate solutions of the true recursive program.

What would settle it

On any of the reported models, recompute nested Monte Carlo Bellman and Euler/KKT residuals (or a high-resolution VFI benchmark) on an independent dense test set; if residuals remain large or systematically biased away from the training residual classes, or if policies diverge from VFI where VFI is reliable, the claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes CEFOL, a deep learning algorithm for discrete-time infinite-horizon dynamic programs with recursive utility. The central device is a dedicated neural network for the state-control certainty-equivalent value C(s,c), which is then inserted into Bellman targets and into model-specific first-order/KKT residuals used to train policy and multiplier networks. Value, certainty-equivalent, and (when used) multiplier networks are trained on separate residual losses, with stabilizations including target value networks, delayed FOC updates, exploratory control perturbation, and an independent sample-mean product estimator for squared conditional stationarity residuals. The framework is written for general equality/inequality constraints and occasionally binding constraints via Fischer–Burmeister KKT terms. Numerical applications cover risk-sensitive and Epstein–Zin consumption-saving models, a small-noise robust-control problem (with VFI benchmarks), and an Epstein–Zin DSGE model with stochastic volatility. Reported out-of-sample Bellman and FOC/Euler residuals are typically of order 10^{-4}–10^{-3} over relevant state regions, with larger residuals mainly near binding constraints, and close agreement with VFI when available.

Significance. If the method works as claimed, it is a useful computational contribution for recursive-utility dynamic programming, where nonlinear certainty equivalents enter both the Bellman equation and optimality conditions and are hard to evaluate inside policy updates. The careful FOC/KKT derivation under recursive utility (§2.2), including the structural condition that can eliminate future value gradients, and the modular treatment of vector controls and occasionally binding constraints without penalty reformulations, are genuine strengths relative to expected-utility residual methods and to model-specific deep equilibrium systems. The paper also situates CEFOL relative to Maliar et al. (2021), Friedl et al. (2023), and the companion CEL algorithm, and shows that expected utility is nested. For a computational journal, residual diagnostics plus VFI agreement on a low-dimensional nonlinear case are a recognizable evaluation package; the main open issue is how tightly residual smallness pins down solution accuracy for the harder recursive models.

major comments (3)
  1. [§4 Numerical Results; §3.2–3.3] §3.2–3.3 and §4: The headline accuracy claim (Abstract; §4) rests on out-of-sample Bellman and FOC/Euler residuals of order 10^{-4}–10^{-3}, plus VFI agreement only in the small-noise robust-control model (§4.3). Those residual classes are essentially the same objectives minimized in training (certainty-equivalent, Bellman, stationarity, Fischer–Burmeister). Nested simulation improves the diagnostic estimator relative to the training product estimator, but still measures consistency of a learned fixed point rather than distance to the true value/policy. For the risk-sensitive, Epstein–Zin consumption-saving, and DSGE applications, the paper relies on internal CEFOL/CEFOL-td/CEFOL-vc alignment and residual size. Given the paper’s own acknowledgment of nonconvex training without global optimality guarantees (§1), the accuracy claim needs stronger independent grounding: multi-seed/hyperpara
  2. [§2.2; applications in §4.1–4.4] §2.2, Eqs. (18)–(27): The FOC simplification that eliminates ∂V(s_{t+1})/∂s_{t+1} depends on a structural condition that the transition factors through a scalar y_{t+1}. This is load-bearing for the practical residual forms used later. The applications in §4.1–4.4 implement specialized Euler/KKT residuals, but the manuscript does not systematically verify that (18) holds in each environment or state what is done when it fails (direct automatic differentiation of C and V, alternative envelope forms, etc.). Please state explicitly, for each numerical model, whether the structural condition is used, how F_{t+1,k} is obtained, and whether any residual involves future value gradients.
  3. [§3.1; §4; literature discussion of CEL/Friedl] §3.1 and §4: Three architectures are developed (four-, five-, and three-network), yet the numerical section appears to use only the four-network baseline. Without ablations, the five-network decomposition (Ve, D) and the compact three-network variant remain untested design options rather than demonstrated tools. Either report comparative residuals/stability/cost for the alternative architectures on at least one application, or narrow the main text to the architecture actually used and move the others to an appendix as optional variants. Relatedly, CEFOL is positioned as complementary to CEL (Peng and Guo, 2026) and distinct from Friedl et al. (2023), but §4 contains no head-to-head accuracy or cost comparison on a shared recursive-utility problem; such a comparison would substantially strengthen the methods contribution.
minor comments (6)
  1. [§4 figures and captions] Figures in §4 often evaluate along a one-dimensional slice (cash-on-hand or capital) with other states fixed at steady state. Please state this limitation more prominently when interpreting residual magnitudes, and consider reporting residual quantiles on simulated ergodic samples as a complement.
  2. [§3.4–3.5; §4] Hyperparameters (network widths/depths, learning rates, delay d, τ, residual weights λ_S/λ_FB, exploration ζ_k, N_z, etc.) are numerous but only partially documented in the main text. A compact table of training settings per application would aid reproducibility.
  3. [§4 opening; §4.1.3] In the risk-sensitive examples, absolute Bellman errors are used because V crosses zero; in other examples relative Bellman errors are used. A short note unifying the diagnostic definitions and normalizations (including the Euler residual (106)–(108)) would help cross-model comparison.
  4. [§2–§3] Notation occasionally shifts between C(s,c), C(s,c;θ_C), and script C; keep a single convention for the true certainty equivalent versus the network approximator.
  5. [title page] The manuscript date line reads “July 13, 2026”; confirm versioning/arXiv metadata consistency before journal submission.
  6. [§4 figures] Some figure panels mark horizontal thresholds (e.g., 10^{-4}, 10^{-3}) without stating whether these are formal tolerances or visual guides; clarify in captions.

Circularity Check

1 steps flagged

Residual training objectives reappear as out-of-sample diagnostics (standard for residual methods); VFI supplies limited external grounding and companion CEL is complementary, not load-bearing.

specific steps
  1. other [Abstract; §3.2–3.3 (losses); §4 (diagnostics)]
    "By using first-order and KKT residuals to learn the policy... out-of-sample Bellman diagnostics and model-specific optimality residuals, including Euler or first-order residuals where applicable, are generally of order 1.0e-4 to 1.0e-3"

    Training explicitly minimizes the certainty-equivalent, Bellman and FOC/KKT residual losses; the headline accuracy claim is that the same residual classes are small out-of-sample. For residual methods this is expected rather than a forced identity, but the diagnostic is not fully independent of the training objective (except where VFI is available).

full rationale

The paper derives model-specific FOC/KKT residuals from the recursive Bellman equation (§2.2), then trains networks by minimizing certainty-equivalent, Bellman, stationarity and Fischer–Burmeister losses (§3.2–3.3). Evaluation reports the same residual classes on independent test states plus nested simulation (§4), together with internal consistency (CEFOL vs CEFOL-td vs CEFOL-vc) and VFI agreement only for the small-noise robust-control model (§4.3). This is the ordinary residual-method loop, not a self-definitional identity or a fitted parameter renamed as prediction: out-of-sample states and a different Monte-Carlo estimator keep the diagnostics from being literally identical to the training loss. The companion CEL citation (Peng & Guo 2026) is explicitly complementary (Bellman maximization versus FOC residuals) and is not used to force uniqueness or forbid alternatives. No uniqueness theorem is imported from the authors, no ansatz is smuggled via self-citation, and the FOC derivation itself is self-contained. Score 2 reflects only the mild, expected self-reference of residual diagnostics; the central algorithmic claim retains independent content via the separate certainty-equivalent network and the VFI benchmark where available.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 2 invented entities

The central claim is computational: that CEFOL’s residual-minimizing multi-network scheme yields accurate recursive-utility solutions. It rests on standard recursive-utility and KKT theory, differentiability and interchange assumptions, an optional structural transition condition for simplified FOCs, and many free training/architecture choices. The main invented object is the CEFOL architecture (especially the dedicated certainty-equivalent network and residual product estimator under recursive utility), not a new physical entity.

free parameters (6)
  • Neural network architectures and widths/depths for V, c, m, C (and optional Ve, D)
    Capacity and inductive bias of each block are chosen by the authors; accuracy claims depend on these design choices.
  • Learning rates α_V, α_c, α_m, α_C (and α_e, α_D) and optimizer settings
    Nonconvex training outcomes depend on step sizes and optimizer hyperparameters not uniquely determined by the model.
  • FOC update delay d and target soft-update rate τ
    Stabilization knobs that change training dynamics and final residual levels.
  • Residual weights λ_S, λ_FB, λ_EQ and component normalizers v_S, v_FB, v_EQ
    Relative scaling of stationarity vs complementarity vs equality losses is hand-tuned and affects the learned policy/multipliers.
  • Exploration scale ζ_k and path-simulation design (N, N_z, mini-batching)
    State-control coverage and Monte Carlo noise in CE and FOC losses are controlled by simulation design choices.
  • Model preference/technology calibrations (e.g., σ, β, γ, ρ, ψ, DSGE parameters)
    Numerical success is demonstrated on selected calibrations; they are not free parameters of the algorithm but define the test suite the claim rests on.
axioms (6)
  • domain assumption Recursive utility admits a Bellman representation V(s)=max_c w(s,c,C(s,c)) with C=f^{-1}(E[f(V(s'))|s,c]).
    Foundation of the entire residual system (§2.1); standard in recursive-utility theory but model-dependent for existence/uniqueness.
  • domain assumption Relevant functions are differentiable and derivatives may pass through conditional expectations when forming FOCs.
    Stated in §2.2 as needed to derive stationarity integrands F_{t+1,k}.
  • domain assumption Optional structural condition: next-state map factors through a scalar y(s,c) so future value gradients can be eliminated via envelope and next-period stationarity.
    §2.2 equations (18)–(27); used to simplify implementable Euler/KKT residuals without ∂V/∂s.
  • standard math KKT conditions characterize the constrained optimum (stationarity, primal feasibility, dual feasibility, complementarity).
    Used to build policy/multiplier losses including Fischer–Burmeister terms (§2.2, §3.3).
  • standard math Conditionally independent residual product estimates the squared conditional mean residual (all-in-one / independent sample-mean estimator).
    §3.3–3.4.3, adapting Maliar et al. (2021); unbiasedness of the product for (E[R])^2 under conditional independence.
  • ad hoc to paper Neural-network residual minimization yields a practically accurate solution despite nonconvex training.
    Acknowledged in the introduction as a general limitation of gradient-based deep learning; the paper relies on residual diagnostics rather than a proof.
invented entities (2)
  • CEFOL multi-network architecture with dedicated state-control certainty-equivalent network no independent evidence
    purpose: Amortize nonlinear CE evaluation and insert C(s,c) into Bellman targets and FOC/KKT residuals for recursive utility.
    Core algorithmic object of the paper; not an external physical entity, but a new computational construct relative to cited solvers.
  • Five-network CE decomposition into conditional expectation Ve and nonlinear difference D no independent evidence
    purpose: Separate linear expectation from Jensen-type recursive-utility gap for interpretability/training.
    Optional architectural invention in §3.1.2; no external evidence beyond the paper’s own training narrative.

pith-pipeline@v1.1.0-grok45 · 44831 in / 3939 out tokens · 43824 ms · 2026-07-13T02:51:27.185706+00:00 · methodology

0 comments
read the original abstract

This paper proposes the certainty-equivalent first-order learning (CEFOL) algorithm, a deep learning algorithm for solving discrete-time dynamic programming problems with recursive utility. Dynamic programming with recursive utility is challenging because nonlinear certainty equivalent appears in the Bellman equation and the first-order optimality conditions but is difficult to evaluate. By introducing a separate neural network to represent the certainty equivalent, CEFOL enables the exploitation of the Bellman and model-specific first-order optimality conditions. In addition to certainty equivalent, CEFOL also uses neural networks to learn the value functions, policy functions, and Lagrange multipliers by using model-specific first-order conditions to construct residuals for minimization. By using first-order and KKT residuals to learn the policy, CEFOL directly accommodates general equality and inequality constraints on the controls, including occasionally binding constraints, without requiring penalty functions or problem-specific reformulations. We apply the algorithm to risk-sensitive and Epstein--Zin consumption-saving problems, a small-noise robust-control problem, and a DSGE model with recursive preferences and stochastic volatility. Across these applications, out-of-sample Bellman diagnostics and model-specific optimality residuals, including Euler or first-order residuals where applicable, are generally of order 1.0e-4 to 1.0e-3 over the relevant state regions, with larger values mainly near binding constraints, and the learned value and policy functions closely match VFI benchmarks when available. The CEFOL algorithm also works for dynamic programming problems with expected utility, as expected utility is a special case of recursive utility.

Figures

Figures reproduced from arXiv: 2607.09461 by Jianfei Zhu, Songyan Wang, Wu Guo, Xianhua Peng.

Figure 1
Figure 1. Figure 1: Value function and learned recursive expansions for the risk-sensitive consumption [PITH_FULL_IMAGE:figures/full_fig_p042_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Consumption policy for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p043_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Bellman error for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p043_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Euler residual for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p044_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Value function and learned recursive expansions for the risk-sensitive consumption [PITH_FULL_IMAGE:figures/full_fig_p044_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Consumption policy for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p045_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Bellman error for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p045_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Euler residual for the risk-sensitive consumption-saving model with [PITH_FULL_IMAGE:figures/full_fig_p046_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Value function comparison for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p050_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Policy function for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p050_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Bellman error for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p051_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FOC residual for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p051_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Value function comparison for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p052_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Policy function for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p052_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Bellman error for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p053_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FOC residual for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p053_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Value function comparison for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p054_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Policy function for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p054_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Bellman error for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p055_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: FOC residual for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p055_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Value function comparison for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p056_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Policy function for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p056_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Bellman error for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p057_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: FOC residual for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p057_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Value function comparison for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p058_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Policy function for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p058_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Bellman error for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p059_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: FOC residual for the Epstein–Zin consumption-saving model under [PITH_FULL_IMAGE:figures/full_fig_p059_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Value function and one-step expansions for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p065_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Relative difference between the CEFOL value network and the VFI benchmark [PITH_FULL_IMAGE:figures/full_fig_p065_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Consumption policy for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p066_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Relative difference between the CEFOL consumption policy and the VFI bench [PITH_FULL_IMAGE:figures/full_fig_p066_32.png] view at source ↗
Figure 33
Figure 33. Figure 33: Relative Bellman error for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p067_33.png] view at source ↗
Figure 34
Figure 34. Figure 34: Euler residual for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p067_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: Value function and one-step expansions for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p068_35.png] view at source ↗
Figure 36
Figure 36. Figure 36: Relative difference between the CEFOL value network and the VFI benchmark [PITH_FULL_IMAGE:figures/full_fig_p068_36.png] view at source ↗
Figure 37
Figure 37. Figure 37: Consumption policy for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p069_37.png] view at source ↗
Figure 38
Figure 38. Figure 38: Relative difference between the CEFOL consumption policy and the VFI bench [PITH_FULL_IMAGE:figures/full_fig_p069_38.png] view at source ↗
Figure 39
Figure 39. Figure 39: Relative Bellman error for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p070_39.png] view at source ↗
Figure 40
Figure 40. Figure 40: Euler residual for the general small-noise model with [PITH_FULL_IMAGE:figures/full_fig_p070_40.png] view at source ↗
Figure 41
Figure 41. Figure 41: Value function comparison for the DSGE model. Capital [PITH_FULL_IMAGE:figures/full_fig_p077_41.png] view at source ↗
Figure 42
Figure 42. Figure 42: Policy functions for the DSGE model. The figure reports the consumption ratio [PITH_FULL_IMAGE:figures/full_fig_p078_42.png] view at source ↗
Figure 43
Figure 43. Figure 43: Bellman error for the DSGE model. Capital [PITH_FULL_IMAGE:figures/full_fig_p078_43.png] view at source ↗
Figure 44
Figure 44. Figure 44: FOC residuals of labor and consumption for the DSGE model. Capital [PITH_FULL_IMAGE:figures/full_fig_p079_44.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

2 extracted references · 2 linked inside Pith

  1. [1]

    M., Xu, Z

    Aboussalah, A. M., Xu, Z. and Lee, C.-G. (2022). What is the value of the cross-sectional approach to deep reinforcement learning?,Quantitative Finance22(6): 1091–1111. Andreasen, M. M. (2012). On the effects of rare disasters and uncertainty shocks for risk premia in non-linear dsge models,Review of Economic Dynamics15(3): 295–316. Azinovic, M., Gaegauf,...

    Pith/arXiv arXiv 2022

  2. [2]

    and Wang, T

    Dumas, B., Uppal, R. and Wang, T. (1998). Efficient intertemporal allocations with recursive utility. E, W., Han, J. and Jentzen, A. (2017). Deep learning-based numerical methods for high- dimensional parabolic partial differential equations and backward stochastic differential equations,Communications in Mathematics and Statistics5(4): 349–380. Epstein, ...

    Pith/arXiv arXiv 1998