{"id":"aa8e59ba-2bc7-4acc-8465-6072f1a0c904","arxiv_id":"1908.05534","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"For a jump-diffusion market with a longevity asset, time-consistent mean-variance equilibria must solve an extended HJB system, and in the no-liability case the optimal strategies are closed-form and numerically robust to jump misspecification.","lead":"This paper derives necessary conditions and explicit optimal trading strategies for an insurer who wants to maximize expected terminal wealth while keeping its variance low in a market with jumps and a longevity bond. The closed-form strategies are shown in simulations to be robust to how jump risk is modeled, which matters for practical hedging.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closed-form equilibrium unverified: Theorem 4.3 is outsourced and the C^{1,2,2,2}/integrability hypotheses for A,a,B,b are never checked; Theorem 4.4 additionally assumes smoothness of g,F in its proof.","rationale":"The reader's conditional verdict is sound. My stress-test confirms the main gap: Theorem 5.1 is presented as a theorem, but its verification is delegated to Theorem 4.3, whose proof is omitted and referred to Björk-Murgoci (2010). In a jump-diffusion setting with a degenerate CIR-type factor, it is not automatic that the Feynman-Kac representations (5.9) and (5.11) yield functions in the required C^{1,2,2,2} class, nor that the integrability assumptions of the verification theorem hold. In particular, σλ(t,λ)=σ√λ in Section 6 violates the global Lipschitz condition in Assumption 2.2 near λ=0, and the candidate b(t,λ) must be shown to be twice differentiable with bounded generator terms on compacts; u_Y(t) grows like 1/Bλ(t,T) as t→T, so the boundedness needed for Dynkin/Feynman-Kac arguments near T must be checked explicitly. I also noticed the necessity proof (Theorem 4.4) uses Dynkin's formula on g^{u*} and F^{u*} although Definition 4.2 and Assumption 4.1 only guarantee the existence of A^{u*}V, not the domain membership of g and F; this does not falsify the theorem but means the regularity hypotheses need to be stated and verified. A single analytical check of the regularity and dominated-convergence hypotheses for (5.9),(5.11) would settle whether the closed-form solution is a genuine equilibrium; until then the paper should remain conditional rather than accepted.","tokens_in":25485,"tokens_out":24323,"duration_ms":249075,"concrete_test":"Take the JCIR yield curve Bλ(t,T) and the candidate b from (5.9). Verify analytically that b(t,λ) is C^{1,2} on [0,T)×(0,∞) by differentiating the Feynman-Kac expectation under the measure P* with dominated convergence, using the moment bounds resulting from Assumption 2.1 and the affine form (6.4); then verify that the B defined by (5.11) satisfies (5.6) and is C^{1,2,2,2} on compacts, including a neighbourhood of t=T where u_Y diverges like 1/Bλ(t,T). If derivatives at λ=0 or t=T fail, state the additional conditions (e.g., λ≥ε or a Feller barrier) under which Theorem 4.3 applies. This single regularity check settles whether the closed-form formulas are genuine equilibria or only candidates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's explicit solution rests on Theorem 4.3 (sufficiency), whose proof is 'omitted' and delegated to Björk-Murgoci (2010). That theorem requires F^{u*} and g^{u*} to be C^{1,2,2,2} and the HJB system to be solved in a strong sense; for the JCIR setting these hypotheses are not checked. In Section 5 the authors write that the Ansatz functions 'are assumed to satisfy the necessary regularity conditions and the limits induced by applying the operators A,L and G are assumed to exist accordingly,' but no proof is given. In the numerical model, σλ(t,λ)=σ√λ is not globally Lipschitz on R+ (violating Assumption 2.2 at λ=0), so the Feynman-Kac representations (5.9) and (5.11) cannot be taken for granted to give C^{1,2,2,2} solutions; dominated convergence and differentiation of b(t,λ) near λ=0 and t→T must be justified. Also, u_Y(t) in (5.13) contains b_λ and b2 and, for the CIR bond, grows like 1/Bλ(t,T) near maturity; the verification theorem needs bounds on the resulting integrals and on A^{u*}B. Separately, Theorem 4.4's proof applies Dynkin's formula to g^{u*} and F^{u*} in Lemmas 4.7-4.10 and Step 3, but Definition 4.2 only defines a regular equilibrium through existence of a feedback equilibrium control, and Assumption 4.1 only postulates existence of A^{u*}V. The smoothness/domain assumptions under which Dynkin is valid are therefore unstated; without them the necessity theorem, the paper's main novelty, is incomplete as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers a time-consistent mean-variance hedging problem of an insurer in a market with stocks, a zero-coupon longevity bond, and a bank account, where asset prices and the mortality force follow jump-diffusions. The authors adopt the Nash subgame-perfect equilibrium concept of Björk and Murgoci (2010), define an extended HJB system, and prove (Theorem 4.4) that a regular equilibrium necessarily solves this system, extending Lindensjö (2016) from a pure-diffusion setting to a jump-diffusion setting including a terminal hedge. For the special case D≡0, they present closed-form expressions for the equilibrium strategies, the value function, and the expected terminal wealth (Theorem 5.1). The final section contains numerical experiments indicating robustness of the expected terminal payoff and its variance with respect to jump-size distributions and to a mis-specification of jump versus diffusion risk, as well as to a maturity mismatch of the longevity asset.","tokens_in":25925,"tokens_out":8632,"duration_ms":80546,"significance":"If the main theorems are correct, the paper provides a useful extension of the equilibrium HJB characterization to jump-diffusion markets and offers a tractable closed-form solution in an insurance setting with basis risk. The numerical findings, if reproducible, are practically relevant. The paper is generally well motivated and the algebraic derivations in Section 5 are careful. However, the sufficiency theorem (Theorem 4.3) is imported without proof, and the regularity and integrability hypotheses needed to apply it to the explicit solution are not verified; the necessity proof also relies on generator applications whose smoothness assumptions are not stated. These gaps affect the main claims as written, but they are potentially fixable within the manuscript's scope.","major_comments":[{"comment":"The necessity theorem is proved without explicit regularity assumptions on V, F^{u*}, and g^{u*}. The proof applies Dynkin's formula and the generator A^{u*} to these functions, but Definition 4.2 only assumes u* is an equilibrium control, and Assumption 4.1 only postulates existence of A^{u*}V. In particular, Lemma 4.7 concludes A^{u*}g^{u*}=0 using Dynkin and dominated convergence, yet no C^{1,2,2,2} or boundedness condition is imposed on g^{u*}; Lemma 4.8 similarly requires A^{u*}F^{u*}. The limit interchanges in Step 4 also need a careful justification, as the liminf over c is replaced by a subsequence limit without stating why the two are equal. As written, the proof of the central necessity result is incomplete.","section":"Section 4, proof of Theorem 4.4 (Lemmas 4.7–4.10, Step 3)"},{"comment":"The closed-form result Theorem 5.1 rests on the sufficiency theorem 4.3, whose hypotheses require F^{u*},g^{u*}∈C^{1,2,2,2} and a strong solution of the extended HJB system. The text only says these functions 'are assumed to satisfy the necessary regularity conditions' without verification. For the numerical model, σ_λ(t,λ)=σ_λ√λ is not globally Lipschitz on R_+, violating Assumption 2.2(ii) at λ=0, so the Feynman-Kac representations (5.9) and (5.11) cannot be taken to yield C^{1,2,2,2} solutions without additional argument. This is load-bearing: without a verification, the formulas in (5.12)–(5.13) are candidate equilibria, not proven ones.","section":"Section 5, after Eq. (5.1); Eqs. (5.9), (5.11); Section 6, JCIR model"},{"comment":"The experiment with T_L=15 and T_L=25 does not match the model setup: Section 2 assumes the insurance horizon and the time to maturity of the longevity bond coincide, and the dynamics of Y in (6.5), the pricing formula (6.4), and the strategy (6.10) are derived for a bond maturing at T. The paper does not specify how the model is modified when T_L≠T, so the claim that maturity differences 'do not add to the variance of the terminal wealth' is not reproducible from the given formulas.","section":"Section 6, Table 4 (Panels E and F)"},{"comment":"Admissibility of the closed-form strategies is not established. In the JCIR example, B_λ(t,T)→0 as t→T, so the coefficients in (6.5) imply σ_L^2+η~_L ≈ O(B_λ^2) while ν_L ≈ O(B_λ); the candidate u_Y in (6.10) then grows like 1/B_λ(t,T) near maturity. The resulting wealth process may fail the condition E[|P_t|^2]<∞ in Definition 2.5. In addition, the Girsanov density Φ in (5.10)/(6.7) is only assumed to be a positive martingale, and the condition C(t,λ,Y,x,x̄)<1 is stated but never verified; without these checks the measure change used for b in (5.9) lacks justification.","section":"Section 5, Eqs. (5.5)–(5.13); Section 6, Eq. (6.10)"}],"minor_comments":[{"comment":"The heading 'Numercal results' contains a typo and should read 'Numerical results'.","section":"Section 6, title"},{"comment":"The phrase 'Definiton 4.2' should be corrected to 'Definition 4.2'.","section":"Theorem 4.4"},{"comment":"The statement 'We take λ_t>0 for all t' appears to conflict with Assumption 2.2, which is formulated on R_+; the CIR volatility σ_λ√λ is not Lipschitz on R_+, so a modified assumption or a localization argument is needed.","section":"Section 2, after Eq. (2.3)"},{"comment":"The Itô expansion leading to Ξ would be easier to follow if the domain of B and b and the integrability conditions on the jump integrals were stated explicitly instead of being left to the reader.","section":"Section 5, Eq. (5.2)"},{"comment":"The parameter values are described as 'typical in the literature' without a specific source; a brief calibration reference or a sensitivity discussion would improve reproducibility.","section":"Section 6, Table 1"}],"recommendation":"major_revision","confidential_remarks":"The main theoretical novelty—the necessity theorem—is genuinely interesting, but the proof as written has significant regularity gaps, and the explicit solution depends on an unproved sufficiency theorem whose hypotheses are not checked. The numerical section also contains an underspecified change of maturity. I believe the paper is salvageable with added hypotheses and verification arguments, but the current version does not yet establish its central claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know two things about 1908.05534. First, its main theorem—necessity of the extended HJB system for regular equilibria in a jump-diffusion market with a terminal liability—is a real extension of Lindensjö's diffusion-only result. Second, the paper as written does not prove its own sufficiency theorem, and it never checks the regularity hypotheses that the explicit solution needs. So the closed-form equilibrium from Theorem 5.1 is a candidate, not a certified result.\n\nWhat is genuinely new: Theorem 4.4 goes beyond Lindensjö (2016) by covering jumps and a terminal hedge. The proof strategy is transparent, and the algebra in Lemmas 4.7–4.10 is mostly coherent. The explicit formulas in Section 5 for D=0 with a longevity asset appear new and plausible; the separation of variables into A(t)p+B(t,z) and a(t)p+b(t,z) is standard but works cleanly here. The paper also deserves credit for saying out loud, in the conclusion, that a numerical treatment of the terminal hedge is still open.\n\nThe soft spots are where you'd expect them. Theorem 4.3, the sufficiency result, is literally not proved—it says 'the proof can be conducted similarly' to Björk-Murgoci (2010) and is omitted. That is the load-bearing wall for Theorem 5.1. The authors then assume, without checking, that A,a,B,b have the required C^{1,2,2,2} regularity and that all operator limits exist. In the JCIR numerical setup, σλ(t,λ)=σ√λ is not globally Lipschitz on R_+, so the Feynman-Kac representations (5.9), (5.11) need more than a hand-wave. Also, u_Y grows like 1/Bλ near maturity, so bounds on the resulting integrals must be justified. The necessity proof itself applies Dynkin's formula to g and F in Lemmas 4.7–4.8 without explicitly stating the smoothness/domain conditions under which Dynkin is valid; as written, that is a gap in the paper's headline result, not just a technicality. The numerics are scenario experiments: a small table of expectations and variances with no error bars, no standard errors, no description of the simulation discretization. The robustness conclusion is suggestive, not established.\n\nNone of this makes the paper worthless. The central ideas are sound and the necessity result is likely true, but it is not fully proved here. A serious referee would need the sufficiency proof (or a proper citation with verification of hypotheses), explicit regularity checks for the JCIR case, and a tighter numerical section.\n\nBring it to a reading group if the group works on time-inconsistent control or actuarial applications; otherwise it's too specialized. I would not cite the explicit solution until the gaps are patched, but a careful referee should engage with it. Verdict: send to peer review, expect major revision.","headline":"The necessity result is a genuine step beyond Lindensjö, but the paper leaves its sufficiency theorem unproved and never checks the regularity its explicit solution needs; the closed-form equilibrium is a candidate, not a certified theorem.","tokens_in":26392,"tokens_out":3648,"would_cite":false,"duration_ms":33756,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q91","60G57","91G80","93E20","97M30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper characterizes time-consistent mean-variance hedging in a jump-diffusion insurance market: regular equilibria solve an extended HJB system, and explicit optimal strategies exist when no terminal liability is hedged.","keywords":["mean-variance portfolio selection","time-consistency","Nash subgame perfect equilibrium","extended HJB system","jump-diffusion model","longevity risk","basis risk","unit-linked life insurance"],"falsifier":"Simulate the terminal payoff under the closed-form strategies (5.12)-(5.13) in the JCIR setting and test a small deviation of the stock or longevity position over a short interval, computing whether the limit in inequality (3.3) is nonnegative; any deviation that strictly improves the mean-variance objective would disprove the equilibrium claim. Alternatively, check computationally whether $A(t)$, $a(t)$, $B(t,z)$, and $b(t,z)$ satisfy the required $C^{1,2,2,2}$ regularity and integrability conditions under the stated parameters.","tokens_in":25249,"feed_emoji":"📈","tokens_out":7676,"duration_ms":72267,"temperature":0.7,"pith_summary":"This paper studies an insurance company that invests in stocks, cash, and a longevity bond to optimize the mean and variance of terminal wealth in a market where both asset prices and mortality can jump. Because mean-variance objectives are time-inconsistent, the paper adopts the game-theoretic notion of a subgame-perfect Nash equilibrium among the insurer's future selves. Its main theoretical result is that any regular equilibrium in this jump-diffusion setting must solve a system of partial integro-differential equations called the extended Hamilton-Jacobi-Bellman system, with the equilibrium strategy attaining the supremum in the first equation. For the case with no terminal liability hedge, the paper derives explicit formulas for the optimal stock and longevity-bond positions, the equilibrium value function, and expected terminal wealth. Numerical experiments suggest the resulting payoffs are stable whether jumps are modeled explicitly, how jump sizes are distributed, and whether the longevity asset matures after the insurance horizon.","feed_headline":"Closed-form hedging for insurers facing stock and longevity jumps","feed_subtitle":"A PIDE system characterizes every time-consistent equilibrium; the explicit stock and longevity strategies barely change when jumps are…","key_machinery":"The load-bearing object is the extended HJB system, a coupled system of partial integro-differential equations whose unknowns are the value function $V$, the expected terminal wealth under the equilibrium $g^{u^\\star}$, and an auxiliary function $F^{u^\\star}$; the first equation combines the generator applied to $V$ with a correction term in $g^{u^\\star}$, and the remaining equations fix $F$ and $g$ through their infinitesimal generators. The proof of necessity uses stopping-time perturbations of the equilibrium strategy together with Dynkin's formula, the Borel-Cantelli lemma, and dominated convergence to show that the equilibrium satisfies each row of the system. For the closed-form part, the paper makes the linear-in-wealth Ansatz $V(t,p,z)=A(t)p+B(t,z)$ and $g(t,p,z)=a(t)p+b(t,z)$, reduces the first-order conditions to static optimizations in the portfolio weights, and represents the remaining function $b$ through a Feynman-Kac expectation under a changed measure, with the JCIR model for mortality supplying affine bond prices.","core_discovery":"The paper claims that, in a jump-diffusion financial and mortality market, time-consistent mean-variance hedging can be fully characterized by the extended HJB system (3.5): any regular equilibrium quadruple $(u^\\star, V, F^{u^\\star}, g^{u^\\star})$ satisfies the system and $u^\\star$ is the maximizing control in its first row. This extends a known necessity result from pure-diffusion markets to markets with jumps and to the presence of a terminal liability hedge. When the liability hedge is dropped ($D\\equiv 0$), the paper obtains explicit closed-form optimal strategies: the stock allocation is $u_S^\\star(t) = \\tilde\\mu(\\tilde\\sigma_S+\\tilde\\rho_S\\xi)^{-1} / (\\gamma e^{r(T-t)})$, the longevity-bond allocation is given by a formula involving the gradient of an auxiliary function $b$, and the equilibrium value function and expected terminal wealth split as $V(t,p,z)=A(t)p+B(t,z)$ and $g(t,p,z)=a(t)p+b(t,z)$ with $A(t)=a(t)=e^{r(T-t)}$.","pith_inferences":["Because the necessity proof is built on general infinitesimal generators and stopping-time arguments rather than on the specific mean-variance form, the same equivalence between regular equilibria and the extended HJB system is likely to hold for other time-inconsistent Markovian objectives in jump-diffusion settings.","The closed-form solution's dependence on jump distributions only through the covariation matrix $\\xi$ suggests a testable moment-dependence: strategies may be unchanged for any jump-size law with the same second moments, which could be checked by Monte Carlo with heavier-tailed jump distributions.","The longevity-bond formula shows the optimal hedge is driven by the market price of longevity risk relative to total quadratic variation; a natural extension is to incorporate a terminal liability tied to the insurer's own mortality pool, where the paper's own numerics suggest the effects could be larger."],"forward_implications":["If the necessity theorem is correct, any regular time-consistent mean-variance equilibrium in a jump-diffusion market can be found by solving the extended HJB system; no other equilibrium candidates exist.","The explicit formulas give an insurer a directly implementable hedging strategy in closed form, with the stock position depending on the risk premium, a covariance matrix corrected for jump covariation, risk aversion, and time to maturity.","The numerical results imply that, at least in this model, an insurer who ignores jumps or misattributes them to diffusion only sacrifices little in expected terminal wealth and variance.","The maturity-mismatch result implies that using a longevity bond with time to maturity longer than the insurance horizon does not materially increase terminal-wealth variance, easing the practical illiquidity constraint.","Because the solution is a Nash subgame perfect equilibrium, the derived strategy is one the insurer will not want to abandon later, resolving the precommitment problem of classical mean-variance optimization."],"supporting_citations":[{"why":"Supplies the extended HJB system and the verification theorem that solving it is sufficient for an equilibrium; the paper's Theorem 4.3 is delegated to it.","marker":"Björk and Murgoci (2010)"},{"why":"Proves the necessity of the extended HJB system in diffusion-only markets, the result extended here to jump-diffusions with a terminal hedge.","marker":"Lindensjö (2016)"},{"why":"Introduces the recursive Nash-equilibrium formulation for dynamic mean-variance asset allocation whose Ansatz structure the explicit solution follows.","marker":"Basak and Chabakauri (2010)"},{"why":"Provides the Dynkin formula and generator calculus for jump diffusions used throughout the necessity proof.","marker":"Øksendal and Sulem (2005)"},{"why":"Gives the precommitment mean-variance solution that the time-consistent approach of this paper contrasts with and extends.","marker":"Zhou and Li (2000)"},{"why":"Motivates affine mortality models with jumps as an adequate description of mortality risk and supplies the CIR/JCIR calibration discussion.","marker":"Luciano and Vigna (2008)"},{"why":"Provides the affine bond-pricing formulas used to price the zero-coupon longevity bond under the JCIR model.","marker":"Brigo and Mercurio (2006)"},{"why":"Supplies the Lévy-measure decomposition and no-arbitrage modelling framework for jump processes used in the market setup.","marker":"Cont and Tankov (2012)"},{"why":"Gives the Feynman-Kac representation used to write the auxiliary function b as an expectation.","marker":"Kromer et al. (2015)"}],"fun_headline_variants":["Explicit hedging strategies for unit-linked life insurance under jumps","Time-consistent hedging for insurers with jump and mortality risk","Closed-form strategies for insurer mean-variance hedging with jumps","Robust hedging payoffs for unit-linked life insurance under jumps"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument relies on the previously established verification theorem being applicable to the explicit solution: the paper does not prove that the closed-form functions are smooth and integrable enough, so unless those regularity conditions are checked the formulas are candidates rather than proven equilibria.","fun_headline_variants_meta":{"raw":{"variants":["Explicit hedging strategies for unit-linked life insurance under jumps","Time-consistent hedging for insurers with jump and mortality risk","Closed-form strategies for insurer mean-variance hedging with jumps","Robust hedging payoffs for unit-linked life insurance under jumps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000568,"raw_usage":{"total_tokens":2693,"prompt_tokens":951,"completion_tokens":1742,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":1674}},"tokens_in":567,"tokens_out":1742,"duration_ms":11397,"temperature":1.0,"reasoning_tokens":1674,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:11:41.807489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the terminal payoff under the closed-form strategies (5.12)-(5.13) in the JCIR setting and test a small deviation of the stock or longevity position over a short interval, computing whether the limit in inequality (3.3) is nonnegative; any deviation that strictly improves the mean-variance objective would disprove the equilibrium claim. Alternatively, check computationally whether $A(t)$, $a(t)$, $B(t,z)$, and $b(t,z)$ satisfy the required $C^{1,2,2,2}$ regularity and integrability conditions under the stated parameters.","supporting_citations":[{"cited_title":"and Chabakauri, G","cited_arxiv_id":null,"evidence_quote":"Introduces the recursive Nash-equilibrium formulation for dynamic mean-variance asset allocation whose Ansatz structure the explicit solution follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the precommitment mean-variance solution that the time-consistent approach of this paper contrasts with and extends."},{"cited_title":"and Vigna, E","cited_arxiv_id":null,"evidence_quote":"Motivates affine mortality models with jumps as an adequate description of mortality risk and supplies the CIR/JCIR calibration discussion."},{"cited_title":"and Mercurio, F","cited_arxiv_id":null,"evidence_quote":"Provides the affine bond-pricing formulas used to price the zero-coupon longevity bond under the JCIR model."},{"cited_title":"and Tankov, P","cited_arxiv_id":null,"evidence_quote":"Supplies the Lévy-measure decomposition and no-arbitrage modelling framework for jump processes used in the market setup."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Feynman-Kac representation used to write the auxiliary function b as an expectation."}],"review_version":1}