{"id":"927ab8b5-5546-4644-83d7-5023a289a2e6","arxiv_id":"2504.18436","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Refining the set of tradable risk contracts monotonically improves welfare in risk markets and drives equilibrium risk prices to the complete-market price under stated conditions.","lead":"This paper studies markets where people trade financial contracts to share future risks, but only some risks can be traded. It shows that as the contracts are split into finer pieces, total welfare rises step by step and the prices of risk converge to those of the ideal complete market.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unproved welfare-equilibrium equivalence and possible non-existence of a social optimum leave the price-convergence claims unsupported.","rationale":"The reader's weakest assumption already identifies the same load-bearing concern: the equivalence between the social optimization problem (5) and a competitive equilibrium, and the existence of an optimal trade vector. This is indeed the most serious gap because the paper's headline results are about prices of risk in a market, and those prices are only meaningful if the social optimum is decentralizable. The missing proof is not a minor technicality: the optimality conditions of (6) only give subgradients Y_i*, not the common conditional expectation Lambda. Without additional conditions on the risk sets D_i (e.g., invariance under conditional expectations), Lambda need not be a subgradient, so agents would not find the social optimum individually optimal at the stated prices. Independently, the existence of W* is not guaranteed; a concrete specification with one risk-neutral and one CVaR agent makes the social objective unbounded below. This means Theorem 3's 'assumed existence' cannot be taken for granted and must be stated as an explicit hypothesis or derived from coercivity conditions. The secondary concern about Zame's theorem and the uniform topology on L^1 is real but less central, since the countable-state section can likely be repaired by working with exact replication of Arrow-Debreu securities and bounded payoffs. Given all this, the CONDITIONAL verdict remains appropriate: the paper should add existence conditions and prove a decentralization theorem before the price-convergence results can be considered established for actual risk markets.","tokens_in":14629,"tokens_out":14448,"duration_ms":147434,"concrete_test":"For a two-agent market with p=1, Omega=[0,1], F_1 generated by A=[0,1/2], D_1={1}, D_2={Y:0<=Y<=2,E[Y]=1}, and Z_1=Z_2=0, compute the objective of (5) along W_1=-W_2=M(1_A-1_{A^c}). If the infimum is -infinity, the assumed existence of an optimal W* is false. Separately, in a case where W* exists (e.g., Example 7 with CVaR agents), compute the common conditional expectation Lambda=E[Y_i^*|F_1] and check directly whether Lambda lies in argmax_{D_i} E[(Z_i-W_i^*)·] for each i; if not, the social optimum is not a competitive equilibrium at the implied price.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central results, Theorem 4 and Corollary 1, describe the behavior of optimal pricing densities in a risk market. But the paper never proves that the social optimization problem (5) is actually a competitive equilibrium of the risk market, nor that an optimal trade vector W* exists. The proof of Theorem 3 only yields Y_i* in argmax_{D_i} E[(Z_i-W_i*)·] with E[Y_i*|F_m] equal across agents. For agent i to be individually optimal facing the implied price Lambda=E[Y_i*|F_m] on tradable risks, one needs Lambda in argmax_{D_i} E[(Z_i-W_i*)·]; a conditional expectation of a subgradient need not be a subgradient, so decentralization can fail. Moreover, existence can fail outright: take p=1, D_1={1}, D_2={Y:0<=Y<=2, E[Y]=1}, F_1 generated by A with P(A)=1/2, and Z_1=Z_2=0. Along W_1=-W_2=M(1_A-1_{A^c}), the objective in (5) equals -M/3 -> -infinity, so no optimum exists. Without a proof of existence and of the welfare-equilibrium equivalence, the monotone welfare and price convergence claims do not yet describe a competitive risk market.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies incomplete risk markets in which agents evaluate liabilities by coherent risk measures. For a finite scenario set, trades are restricted to scenario bundles; by iteratively refining the partition, the authors argue that social welfare increases monotonically and prices converge to the complete-market outcome (Theorem 1, Proposition 1, Example 1). For countably many states (Section 3), the refinement idea is linked to Zame's asymptotic completeness theorem. For a general probability space (Section 4), the authors replace bundles by a filtration F_m, formulate a social optimization problem (5), characterize its dual via a minimax argument (Theorem 3), and prove upper semicontinuity of the optimal dual solution set at the complete market (Theorem 4), with Corollary 1 stating that incomplete-market risk prices price any zero-sum trade to zero in the limit. The main claimed contribution is a monotone market-completion mechanism with limit price convergence.","tokens_in":14839,"tokens_out":26456,"duration_ms":287512,"significance":"If the missing equivalence and existence results can be supplied, the paper would be a useful contribution to the literature on risk markets: it gives an iterative completion mechanism with monotone welfare improvement, connects the finite-bundle construction to a filtration-based limit, and Theorem 4's martingale argument is largely self-contained and appears correct modulo a local error in the appendix. The numerical examples in Sections 2 and 3 are reproducible and informative. However, because the central object is claimed to be a market equilibrium, the unproved welfare-equilibrium equivalence and the assumed existence of an optimal trade vector are the key risks; the significance of the paper depends on resolving them.","major_comments":[{"comment":"The paper asserts that 'the market clearing problem can be expressed as a social optimization' (Eq. (5)) and, in Section 2, that the equilibrium is equivalent to the social problem (3), but it never states a formal definition of competitive equilibrium for the general probability-space model and never proves the equivalence. Theorem 3 shows that if W* solves (5), then there are Y*_i in D_i with common E[Y*_i|F_m]; it does not by itself show that W* and the price system Λ=E[Y*_i|F_m] make each agent's trade individually optimal. Since the subgradient inequality used in the proof gives E[uY*_i]=E[uΛ] only for u in L_p(F_m), the missing argument is likely available, but it must be supplied. The same proof assumes an optimal W* exists: the text says 'For any elements W* optimal...', and Theorem 4's market-price interpretation of the dual optima is vacuous if (5) has no solution. Boundedness below of (5) follows from Assumption 2, but attainment in the infinite-dimensional space L_p(F_m) does not, and no existence theorem is provided or cited.","section":"§4.2, Eq. (5) and Theorem 3"},{"comment":"The statement that the uniform distance d∞(x,y)=sup_s|x(s)-y(s)| 'defines a complete and metrizable topology on L1(S,P)' is not correct as stated: d∞ is extended-valued on unbounded L1 functions and is not a metric, and the sup-norm topology is not complete in the usual sense on L1. More importantly, the convergence claim for the countable-state case is obtained by invoking Zame's Theorem 2 without checking that the model satisfies Zame's hypotheses (commodity space, preferences, asset span, topology). The displayed security matrix does show exact Arrow-Debreu replication, but asymptotic completeness in Zame's sense requires his theorem to apply; as written, the argument is incomplete.","section":"§3, Definition 2 and Theorem 2"},{"comment":"In the case m_∞<∞, the display (A.3) asserts E[E[Y_i∞|F_m] W_i]=E[E[Y_i∞ W_i|F_m]] for arbitrary W_i in L_p; this equality is generally false (the left-hand side equals E[E[Y_i∞|F_m] E[W_i|F_m]], not E[Y_i∞ W_i]). The proof can be repaired locally: taking W_i=U, W_j=-U with U in F_m in the preceding display gives E[(Y_i∞-Y_j∞)U]=0 for all U in F_m, which directly yields E[Y_i∞|F_m]=E[Y_j∞|F_m]. The authors should correct the argument, since the version in the appendix contains an invalid step.","section":"Appendix A, proof of Theorem 4, Eq. (A.3)"}],"minor_comments":[{"comment":"The statement that L^{p1} ⊆ L^{p2} if and only if p1≥p2 is not correct as an equivalence for arbitrary probability spaces; the forward implication holds on finite measure spaces, but the converse can fail, for instance when all random variables are essentially bounded.","section":"§4.1"},{"comment":"There is a typo in 'In this this section'; it should read 'In this section'.","section":"§2, first paragraph"},{"comment":"Definition 2 says an equilibrium allocation of M_n is 'within ε of' an equilibrium allocation of M_CM, but the metric in which this distance is measured is not specified.","section":"§3, Definition 2"},{"comment":"The prose in §4.3 says the incomplete-market prices Y_i on F_m 'converge to the complete-market price Y0 as m→∞', which is stronger than Theorem 4. Theorem 4 only gives upper semicontinuity of the solution set, and if the complete-market solution set is not a singleton, arbitrary selections need not converge to a unique limit. Corollary 1's quantitative statement is accurate, but the surrounding prose should be softened unless uniqueness is proved.","section":"§4.3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is in scope for math.OC and the central idea is plausible, but I would not accept until the authors prove or explicitly replace the welfare-equilibrium equivalence in Section 4.2 and establish existence of an optimal trade vector; the Section 3 use of Zame should either be made rigorous or de-emphasized. The self-citation of [3] is appropriate given the direct dependence of the paper on that work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper proposes a monotone market-completion scheme for risk markets with coherent risk measures: refine the bundle partition (finite case) or the filtration (general case), solve the social optimization problem, and welfare increases monotonically while the dual pricing densities converge to the complete-market price. The genuinely new result is Theorem 4 / Corollary 1: upper semicontinuity of the optimal pricing set in the filtration parameter, giving convergence of prices on zero-sum trades. The finite-dimensional Theorem 1 is mostly a relaxation consequence of Ralph–Smeers, but it is correctly stated and the examples make the behaviour concrete. The proof of Theorem 4 is largely sound: the minimax argument behind Theorem 3 is standard, and the closedness proof for the feasible multifunction is handled carefully. I would cite Corollary 1 if I worked on risk-market design.\n\nThe soft spots are in the infinite-dimensional framing. Section 3 contains a factual error: the uniform topology is not complete on L^1, and Zame's theorem is invoked without checking its hypotheses. This section is the least rigorous and should be rewritten or explicitly labelled heuristic. Section 4 asserts, rather than proves, that the social optimization problem coincides with competitive equilibrium and that an optimal trade vector W* exists. Those are real gaps. Note that the specific counterexample in the stress-test note does not hold up: for the two-agent data given there, the objective is |t|, not unbounded below. The conditional-expectation concern about decentralization is also weaker than it looks: since agents can trade only Fm-measurable risks, the first-order condition only requires the market price to be the conditional expectation of a subgradient, not a subgradient itself. Still, the paper should prove existence and decentralization or restrict to cases where they are automatic, e.g., CVaR.\n\nThe priority claim in the introduction—\"first method\" for monotone completion—is overstated; the monotone improvement is immediate from constraint relaxation once the equilibrium-equivalence is accepted. The citation pattern is otherwise fine.\n\nWho this is for: researchers on risk trading, incomplete markets, and energy or insurance market design. The paper deserves a serious referee. I would send it out, with instructions to ask for a corrected Section 3 and a proper existence/decentralization argument for Section 4.","headline":"Worth a careful referee: the finite refinement argument is clean, and the dual-price convergence theorem is a real new result, but Sections 3–4 rely on unverified functional-analytic hypotheses and an unproved equilibrium-equivalence claim.","tokens_in":15400,"tokens_out":17231,"would_cite":true,"duration_ms":192682,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B30","91B50"],"pacs":[],"model":"deepseek-v4-flash","headline":"Repeatedly refining the bundles of scenarios in an incomplete risk market drives total welfare up monotonically and pushes equilibrium risk prices to their complete-market limits.","keywords":["risk markets","incomplete markets","coherent risk measures","market completion","welfare enhancement","risk pricing","asymptotic completeness","filtration refinement"],"falsifier":"Choose two agents with identical risk sets (for instance, both using CVaR with the same $\\beta$) and identical endowments on the dyadic filtration of $[0,1]$, so that the dual problem (7) has many optimal pricing densities at every $m$; then compute, for a fixed zero-sum trade $W$ with $W_1=-W_2$, the values $\\sum_i E[W_i Y_i^m]$ along different choices of the optima. If there is any selection of optimal $(Y_1^m,Y_2^m)$ whose weak* limit is not an optimal solution of (8), or for which $\\sum_i E[W_i Y_i^m]$ fails to go to zero, the asserted upper semicontinuity of the optimal-solution multifunction is false.","tokens_in":14368,"feed_emoji":"📈","tokens_out":11417,"duration_ms":106350,"temperature":0.7,"pith_summary":"The paper studies risk markets in which agents trade bundles of possible future outcomes rather than every single contingency, so the market is incomplete. It proposes a completion mechanism—repeatedly splitting each bundle into finer pieces, which in general probability spaces amounts to enlarging the available information—and claims two guarantees: total welfare rises monotonically with each refinement, and as the refinement approaches the complete market, equilibrium prices of risk converge to the complete-market prices. The finite-scenario version follows from a simple relaxation of constraints, the countable-state version uses asymptotic completeness of the market sequence, and the general version is a continuity result for the optimal pricing densities. A sympathetic reading takes the paper's contribution to be a principled, welfare-improving path from an incomplete to a complete risk market, not just a statement that completeness is desirable.","feed_headline":"Finer risk bundles lift welfare, then prices converge","feed_subtitle":"As markets grow complete, welfare never falls and prices of zero-sum trades converge to zero.","key_machinery":"The load-bearing object is the pair of optimization problems (5) and (7): the social welfare problem restricts trades to the information available at level $m$ and requires $\\sum_i W_i=0$, while the dual pricing problem maximizes $\\sum_i E[Z_i Y_i]$ over risk sets $D_i$ subject to the equality of the conditional expectations $E[Y_i|\\mathcal{F}_m]$. That equality is the market-consensus condition that agents agree on the price of every tradable risk. The proof chain uses a minimax theorem to identify the dual of the welfare problem, weak* compactness of the risk sets to obtain optimal densities, and a perturbation-sensitivity result plus martingale convergence to show the optimal-solution map is upper semicontinuous at the complete market.","core_discovery":"The paper's central claim is that an incomplete risk market can be completed by a sequence of refinements—splitting scenario bundles, or enlarging the filtration $\\mathcal{F}_m$—and that this sequence has two guaranteed properties: total welfare is nondecreasing at every step, and in the limit the prices of risk approach those of the complete market. In the general probability-space setting the argument is carried by a dual characterization: the equilibrium pricing densities $(Y_1,\\dots,Y_n)$ are the optima of the problem in which each agent's density lies in his risk set $D_i$ and the conditional expectations $E[Y_i|\\mathcal{F}_m]$ coincide across agents. Theorem 4 asserts that the set of such optima is upper semicontinuous as $m\\to\\infty$, and Corollary 1 states the concrete consequence that for any zero-sum trade $(W_1,\\dots,W_n)$ in $L^p$, $\\sum_i E[W_i Y_i^m]\\to 0$, the price the trade would fetch in a complete market.","pith_inferences":["The monotone welfare property gives a practical design rule for real markets: new instruments can be introduced one refinement at a time, and the dual prices of (7) identify which bundle carries the largest welfare gain, so a sequence of refinements can be chosen greedily.","The convergence statement suggests a testable measure of market completeness: record the price of a fixed zero-sum trade as tradable instruments are added; the rate at which this price approaches zero can serve as an index of incomplete-market distortion.","The zero-priced-bundle result implies that refinements must be coordinated: refining only bundles that nobody prices can stall welfare gains, an observation relevant to financial innovation in illiquid or segmented markets.","The paper's proof strategy depends on exact equilibrium/social-welfare equivalence, so the conclusions should be expected to need modification for markets with transaction costs, bid-ask spreads, or non-convex risk preferences."],"forward_implications":["In the finite-scenario model, any refinement of the scenario bundles admits an equilibrium whose total welfare is at least as high as before (Theorem 1).","In the countable-state model with the cumulative-bundle scheme, every single-scenario security can be replicated by finitely many traded instruments, so the incomplete market sequence is asymptotically complete and equilibria converge to the complete-market equilibrium.","In the general probability-space model, the price of any zero-sum trade in the incomplete market converges to zero as the filtration grows to the full $\\sigma$-field (Corollary 1), so risk prices approach complete-market prices.","Agents in an incomplete market agree on the price of every tradable risk, since $E[Y_i|\\mathcal{F}_m]$ is equal across agents, while they may still disagree on non-tradable risks.","Refining a bundle whose price is zero, while leaving other bundles unchanged, does not improve welfare (Proposition 1)."],"supporting_citations":[{"why":"Defines coherent risk measures, the objects agents use to evaluate and trade risks.","marker":"[1]"},{"why":"Establishes the complete-market welfare/equilibrium equivalence that this paper extends to incomplete markets.","marker":"[3]"},{"why":"Supplies the asymptotic-completeness theorem used to prove countable-state market sequence convergence.","marker":"[13]"},{"why":"Provides the minimax theorem used to pass from the social welfare problem to the dual pricing problem (7).","marker":"[16]"},{"why":"Gives the parameterized optimization result from which Theorem 4's upper semicontinuity is derived.","marker":"[17]"},{"why":"Provides the martingale convergence theorem used to show the feasible set of pricing densities is closed at the limit.","marker":"[18]"}],"fun_headline_variants":["Market completion lifts welfare, prices converge","Welfare never falls as risk markets complete","Refining risk sets improves welfare, sets prices","Zero-sum trade prices vanish as markets complete","Incomplete markets completed: welfare monotone up"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the social welfare problem has an optimal trade vector and that its solution and the prices that come out of it are exactly the competitive equilibrium of the risk market; if that equivalence or the existence of an optimal trade fails, the monotone welfare and price-convergence results may not describe what actually happens in the market.","fun_headline_variants_meta":{"raw":{"variants":["Market completion lifts welfare, prices converge","Welfare never falls as risk markets complete","Refining risk sets improves welfare, sets prices","Zero-sum trade prices vanish as markets complete","Incomplete markets completed: welfare monotone up"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000178,"raw_usage":{"total_tokens":1248,"prompt_tokens":846,"completion_tokens":402,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":462,"completion_tokens_details":{"reasoning_tokens":335}},"tokens_in":462,"tokens_out":402,"duration_ms":4627,"temperature":1.0,"reasoning_tokens":335,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:17:21.312469+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Choose two agents with identical risk sets (for instance, both using CVaR with the same $\\beta$) and identical endowments on the dyadic filtration of $[0,1]$, so that the dual problem (7) has many optimal pricing densities at every $m$; then compute, for a fixed zero-sum trade $W$ with $W_1=-W_2$, the values $\\sum_i E[W_i Y_i^m]$ along different choices of the optima. If there is any selection of optimal $(Y_1^m,Y_2^m)$ whose weak* limit is not an optimal solution of (8), or for which $\\sum_i E[W_i Y_i^m]$ fails to go to zero, the asserted upper semicontinuity of the optimal-solution multifunction is false.","supporting_citations":[{"cited_title":"Ralph, Y","cited_arxiv_id":null,"evidence_quote":"Establishes the complete-market welfare/equilibrium equivalence that this paper extends to incomplete markets."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the asymptotic-completeness theorem used to prove countable-state market sequence convergence."},{"cited_title":"Fan, Minimax theorems, Proceedings of the National Academy of Sciences 39 (1) (1953) 42–47","cited_arxiv_id":null,"evidence_quote":"Provides the minimax theorem used to pass from the social welfare problem to the dual pricing problem (7)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the parameterized optimization result from which Theorem 4's upper semicontinuity is derived."},{"cited_title":"Durrett, Probability: theory and examples, Vol","cited_arxiv_id":null,"evidence_quote":"Provides the martingale convergence theorem used to show the feasible set of pricing densities is closed at the limit."}],"review_version":1}