{"id":"e28b46f5-81a7-40b9-8af5-b227f473f1d9","arxiv_id":"2603.07782","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":7,"one_line_summary":"Existence of equilibrium interest rates is established for continuous-time heterogeneous-agent models with Epstein–Zin utility under late-resolution preferences (γψ<1).","lead":"The paper proves existence of stationary equilibria for continuous-time Aiyagari–Huggett incomplete-market models when agents have Epstein–Zin recursive utility and prefer late resolution of uncertainty. It supplies the viscosity-solution and Fokker–Planck analysis needed to treat these preferences inside the mean-field-game framework used in modern macroeconomics.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The reader correctly isolates γψ<1 as the weakest (and most load-bearing) modelling assumption: every sign that drives the viscosity comparison, the construction of barriers, and the monotonicity of aggregate capital with respect to r is controlled by θ>1. Because the paper never claims results outside that regime, and because the intermediate-value argument that yields r* is complete once the continuous dependence of K(r) and its blow-up at ρ are established, there is no internal inconsistency that would force a change of verdict. The suggested concrete test merely checks whether one intermediate qualitative property (s1<0) survives without the sign restriction; a negative answer would leave the theorems intact, while a positive answer would only enlarge the paper’s domain. Hence the reader’s CONDITIONAL verdict, already conditioned on the deferred numerical analysis and the early-resolution case, remains the appropriate assessment.","tokens_in":39363,"tokens_out":515,"duration_ms":5002,"concrete_test":"Independently re-derive the key inequality (3.21) that forces s1(x)<0 (Prop. 3.17) from the first-order condition (2.7) and the comparison v1≥û1, without invoking θ>1; if the inequality still holds under the opposite sign θ<1, the late-resolution restriction is not essential for that qualitative step and the paper’s scope claim can be sharpened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central existence claim (Theorems 5.5–5.6) is internally consistent under the paper’s explicit standing assumptions. The comparison principle (Prop. 3.3), the sign of Hv and Hvv ((2.10)–(2.12)), the sub-/supersolutions (Prop. 3.5), the saving-policy signs (Props. 3.17, 3.20–3.22), the blow-up of K(r) as r→ρ (Cor. 5.4), and the intermediate-value argument that produces r*∈(0,ρ) all rely on θ>1 (i.e., γψ<1). That restriction is stated at the outset, used uniformly, and acknowledged as the regime of late resolution; it is therefore not a hidden gap. The two-state income process and the deferred numerical analysis are genuine limitations of scope, but they do not undermine the existence theorems as proved.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper extends continuous-time Aiyagari–Bewley–Huggett heterogeneous-agent models to Epstein–Zin recursive utility in the late-resolution regime γψ<1 (equivalently θ>1), cast as a stationary mean-field game. It establishes a constrained viscosity comparison principle and existence/uniqueness for the weakly coupled HJB system with state constraints (Props. 3.3, 3.7), C¹ regularity, strict concavity and local W^{2,∞} regularity of the value functions, and detailed sign properties of optimal saving policies (Props. 3.17–3.22). It then analyzes the associated FPK system, proves blow-up of aggregate capital as r→ρ (Cor. 5.4), and obtains existence of a stationary equilibrium interest rate r*∈(0,ρ) for the Aiyagari and Huggett models under the explicit sufficient condition (5.7) (Theorems 5.5–5.6). Numerical illustrations are reported in Section 6.","tokens_in":39581,"tokens_out":1366,"duration_ms":40134,"significance":"The work fills a genuine gap between the continuous-time MFG analysis of CRRA heterogeneous-agent models (Achdou–Han–Lasry–Lions–Moll and follow-ups) and the recursive-utility literature used in asset pricing and climate economics. The constrained-viscosity theory, the careful handling of the v-dependence in the Hamiltonian, and the saving-policy sign lemmas are technically substantial and carefully adapted; the comparison principle and the intermediate-value existence argument under (5.7) are clean. Explicit sub-/supersolutions, continuous dependence of policies on r, and the second-order asymptotic expansion of savings as x→∞ when r=ρ (Appendix B) are reusable tools. The restriction to late resolution and two-state income is stated up front and is not hidden. If the existence theory can be extended beyond the knife-edge condition (5.7), the paper would become a standard reference for recursive-utility incomplete-market MFGs.","major_comments":[{"comment":"Theorems 5.5–5.6 rest on the sufficient condition (5.7), which forces s_j≡0 at the borrowing limit for small r (via Prop. 3.20) and hence K(r)=x. Under the standing assumption θ>1 one has ρ/(θλ₂)<ρ/λ₂. For the calibration of Section 6 (ρ=0.05, λ₂=0.4, y₂/y₁=5) one obtains ρ/λ₂=0.125 while (y₂/y₁)^{1/ψ}−1 is already of order 1–50 for the reported ψ values, so (5.7) fails in every numerical test—including Tests 2 and 4 that lie inside the theoretical regime γψ<1. Yet equilibria are still found. The paper should either (i) supply an alternative existence argument covering the complementary (and economically standard) regime in which s₂(x)>0 for all small r, or (ii) explicitly quantify the parameter region where (5.7) holds and reconcile the numerical findings with the theorem hypotheses. As written, the main existence theorems apply only to a set of parameters of limited economic interest.","section":null},{"comment":"Section 6 reports four equilibrium interest rates and six figures, including two tests with γψ>1 that lie outside the paper’s standing assumption (2.1), but neither the discretization (finite differences vs. semi-Lagrangian), grid parameters, nor the fixed-point procedure for r* is described. Section 7 defers numerical analysis to future work. For the reported equilibria to be reproducible and for the claim that “algorithms continue to perform well” outside the theory to be assessable, a minimal description of the schemes and solver tolerances is needed in the present manuscript (or the out-of-regime tests should be removed).","section":null}],"minor_comments":[{"comment":"Proposition 3.7 states existence for 0<r≤ρ, while several later statements (e.g. Prop. 3.13, Section 4) allow 0≤r<ρ. Clarify the r=0 case, especially construction of sub-/supersolutions when r=0.","section":null},{"comment":"Table 1 introduces the auxiliary parameter b and the scaled discount ζ=ρ/θ; ζ is used in the proofs (e.g. Prop. 3.5) but the HJB is written with ρ/θ. A single consistent notation would help.","section":null},{"comment":"In the proof of Prop. 3.17 the comparison (1−γ)ṽ₂/(1−γ)ǔ₁ leads to (b/r)^{(1−γψ)/(1−γ)}≤1; the sign of the exponent relies on γψ<1 and should be flagged explicitly for the reader.","section":null},{"comment":"Figures 1–6 are referenced but the manuscript text does not specify axis units, the percentile truncation used for the Dirac mass, or which curves correspond to which test beyond the caption. Adding a short legend note would improve readability.","section":null},{"comment":"The abstract and introduction emphasize “preference for late resolution”; a one-sentence pointer to the early-resolution case (γψ>1) and why the viscosity comparison fails there would orient readers familiar with the Epstein–Zin literature.","section":null},{"comment":"Typos/notation: “ex ante identical butex postheterogeneous” (missing spaces, p.1); “thecomparisonprinciplethengives” (p.10); occasional missing spaces after commas in displayed equations.","section":null}],"recommendation":"major_revision","confidential_remarks":"Mathematically the HJB analysis is solid and suitable for a math.OC venue. The main weakness is that the headline existence theorems are proved under a sufficient condition that is essentially never met by the calibrations the paper itself (and the broader macro literature) uses; the numerics already suggest existence is true more generally. I would not reject on that ground, but I would insist on either a broader existence argument or a frank, quantitative discussion of the gap before acceptance. The deferred numerical analysis while still publishing equilibrium figures is a secondary but real reproducibility concern."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a careful technical extension of Achdou–Han–Lasry–Lions–Moll to Epstein–Zin recursive utility, restricted to the late-resolution regime γψ<1. The new pieces that matter are the comparison principle for the recursive HJB system with state constraints, the sign properties of the saving policies (especially s1<0 everywhere and the conditions that force s2(x)>0 or s2≡0), the second-order asymptotics of savings as x→∞ when r=ρ, and the resulting blow-up of aggregate capital that lets them run an intermediate-value argument for equilibrium r*∈(0,ρ). Theorems 5.5–5.6 are the payoff: under an explicit and checkable condition on the parameters, both the Aiyagari and Huggett systems admit a stationary equilibrium.\n\nThe viscosity analysis is done properly. They construct explicit sub- and supersolutions, get uniqueness via a strong comparison principle that adapts the usual doubling-variables argument to the recursive Hamiltonian, prove C1 regularity and local W2,∞, and extract the qualitative features of consumption and saving that feed the FPK analysis. The continuous dependence of K(r) and the non-existence of an invariant measure at r=ρ are clean. The two-state income process is a genuine limitation of scope, but it is the same setting used in the CRRA predecessor, so the comparison is fair. The numerical section is only illustrative and the schemes themselves are deferred; that is a soft spot, but it does not touch the existence theorems.\n\nThe standing restriction γψ<1 is not a hidden flaw: it is stated up front, used uniformly for the signs of Hv and Hvv, and acknowledged as the late-resolution case. Early resolution will need different tools; they say so. Citation pattern is appropriate (they build directly on their own earlier continuous-time HA work and the standard recursive-utility references).\n\nThis is for people who already work with continuous-time heterogeneous-agent models or viscosity methods for MFGs with state constraints. It supplies the existence theory and the qualitative saving properties that quantitative papers in macro and climate finance have been using without a full proof. I would send it to referees; the core claims are solid under the stated assumptions and the paper is self-contained enough to be checked. Worth engaging if you care about the mathematical foundations of these models.","headline":"Solid, careful extension of the continuous-time HA MFG framework to Epstein–Zin under late resolution; existence theorems hold under the stated assumptions and the math is clean.","tokens_in":40229,"tokens_out":578,"would_cite":true,"duration_ms":8211,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q91","49L25","91B51","91A16"],"pacs":[],"model":"grok-4.5","headline":"Continuous-time Aiyagari–Huggett models with Epstein–Zin recursive utility admit a stationary equilibrium when agents prefer late resolution of uncertainty.","keywords":["mean field games","Epstein-Zin recursive utility","constrained viscosity solutions","Aiyagari model","Huggett model","Hamilton-Jacobi-Bellman","Fokker-Planck-Kolmogorov","incomplete markets"],"falsifier":"Numerically solve the same HJB–FPK system for a parameter set that violates the explicit condition ρ/(θ λ_{2})>(y_{2}/y_{1})^{1/ψ}−1 and check whether a market-clearing interest rate still appears in (0,ρ); absence of such an r* would falsify the necessity of the condition for existence.","tokens_in":40209,"feed_emoji":"⚖️","tokens_out":1061,"duration_ms":9192,"temperature":0.7,"pith_summary":"The paper extends continuous-time incomplete-market models with idiosyncratic income risk and borrowing limits to Epstein–Zin recursive utility in the regime where agents prefer late resolution of uncertainty. Agents’ optimal consumption-saving problems become a pair of state-constrained Hamilton–Jacobi–Bellman equations whose solutions feed into Fokker–Planck–Kolmogorov equations for the stationary wealth distribution; market clearing then determines the equilibrium interest rate. Under an explicit parameter restriction that keeps the interest rate below the discount rate, the authors prove existence of a stationary equilibrium and establish qualitative properties of optimal savings, including that unproductive agents always decumulate and that aggregate capital blows up as the interest rate approaches the discount rate. A sympathetic reader cares because the separation of risk aversion from intertemporal substitution is central to modern macro-finance, yet rigorous continuous-time theory with state constraints had previously been available only for time-separable utility.","feed_headline":"Late-resolution recursive utility still yields Aiyagari equilibria","feed_subtitle":"State-constrained HJB–FPK systems admit stationary interest rates when agents prefer late uncertainty resolution","key_machinery":"The constrained viscosity solution of the weakly coupled Hamilton–Jacobi–Bellman system with Epstein–Zin aggregator; comparison, regularity and continuous dependence of this solution on the interest rate supply the continuous map from r to aggregate capital needed for a fixed-point argument that yields equilibrium.","core_discovery":"Under the standing assumptions γ>1, 0<ψ<1 and γψ<1, together with the explicit inequality ρ/(θ λ_{2})>(y_{2}/y_{1})^{1/ψ}−1, the Aiyagari (respectively Huggett) mean-field-game system admits at least one stationary equilibrium interest rate r*∈(0,ρ). The value functions are unique constrained viscosity solutions of the coupled HJB system, are strictly concave and C^{1}, and generate optimal saving policies whose qualitative properties (negative savings for low-income agents, possible positive savings only near the borrowing limit for high-income agents) close the Fokker–Planck equation and produce continuous aggregate capital that diverges as r\toρ.","pith_inferences":["The comparison-principle machinery developed here is likely to adapt, after suitable changes of variables, to the early-resolution regime γψ>1 that the authors flag for future work.","Because the two-state Poisson income process is used only to obtain explicit asymptotics of savings, the existence proof should extend to more general continuous-state Markov income processes once those asymptotics are replaced by Lyapunov-type arguments.","The continuous dependence of saving policies on r supplies a natural path to local uniqueness or comparative-statics results that the paper leaves open."],"forward_implications":["Stationary incomplete-market equilibria continue to exist once risk aversion is separated from the elasticity of intertemporal substitution, provided agents prefer late resolution.","Aggregate capital supply remains continuous in the interest rate and diverges as r approaches the discount rate, so the usual fixed-point argument for equilibrium still closes.","Unproductive agents always run down wealth; productive agents may accumulate only near the borrowing limit and eventually decumulate, producing a Dirac mass only at the debt floor.","The same viscosity framework yields monotone numerical schemes that remain valid for the recursive-utility HJB system."],"fun_headline_variants":["Aiyagari equilibria persist under late-resolution recursive utility","MFG systems admit stationary rates with late uncertainty preference","Heterogeneous agents preferring late resolution yield Aiyagari r*","Continuous-time Epstein-Zin models still close Aiyagari equilibria","State-constrained HJB-FPK yield equilibria for late-resolving agents"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The entire theory requires that agents prefer late resolution of uncertainty (γψ<1), because that sign makes the Hamiltonian decreasing and concave in the value variable and permits the comparison principle used throughout.","fun_headline_variants_meta":{"raw":{"variants":["Aiyagari equilibria persist under late-resolution recursive utility","MFG systems admit stationary rates with late uncertainty preference","Heterogeneous agents preferring late resolution yield Aiyagari r*","Continuous-time Epstein-Zin models still close Aiyagari equilibria","State-constrained HJB-FPK yield equilibria for late-resolving agents"]},"model":"grok-4.5","effort":"low","cost_usd":0.004284,"raw_usage":{"total_tokens":1208,"prompt_tokens":693,"num_sources_used":0,"completion_tokens":75,"cost_in_usd_ticks":42840000,"prompt_tokens_details":{"text_tokens":693,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":440,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":693,"tokens_out":75,"duration_ms":3829,"temperature":1.0,"reasoning_tokens":440,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-15T13:03:03.436118+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Numerically solve the same HJB–FPK system for a parameter set that violates the explicit condition ρ/(θ λ_{2})>(y_{2}/y_{1})^{1/ψ}−1 and check whether a market-clearing interest rate still appears in (0,ρ); absence of such an r* would falsify the necessity of the condition for existence.","supporting_citations":[],"review_version":1}