{"id":"71bb322a-10e1-413d-a9fd-e750c2777ca0","arxiv_id":"2606.22555","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A normalized projection score onto cluster, factor, and sparse covariance geometries forms a 'dependence profile' that is consistent, classifiable, and yields oracle-equivalent variance-estimator selection.","lead":"This paper proposes a geometric framework that summarizes whether economic data are clustered, factor-driven, or sparsely dependent, and uses the resulting 'dependence profile' to choose among robust inference procedures. It proves that this profile-guided choice is asymptotically as good as an oracle that knows the true dependence structure, though the practical example ends in a hybrid, no-clear-choice conclusion.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Factor projection's alternating-projection algorithm has no global convergence guarantee; the estimated profile may not correspond to the theoretical projection, breaking the consistency and oracle-adaptivity proofs.","rationale":"The reader's weakest_assumption identifies exactly this gap between the theory and the computational algorithm for the factor projection. This is the most load-bearing concern because the entire chain from profile estimation to oracle adaptivity depends on the estimated projection being a consistent estimator of the population projection. If the algorithm fails to compute the latter, the asymptotic guarantees are moot. The paper acknowledges the issue but does not resolve it: it calls convergence a 'numerical observation rather than a proven global guarantee.' This is a concrete, fixable gap, but it is material to the central claim. My verdict remains CONDITIONAL, consistent with the reader's assessment.","tokens_in":36546,"tokens_out":3501,"duration_ms":36320,"concrete_test":"On the simulation DGPs (pure factor, cluster–factor hybrids, r=1,2,3), run the alternating-projection algorithm from many random starting points (e.g., 100 restarts) and record the dispersion of bP_F and bω_F across restarts. If the standard deviation of bω_F across restarts is non-negligible relative to the Monte Carlo standard errors in Table 1, the algorithm is not reliably computing the population projection. Additionally, for a small n (e.g., n=10), compare the algorithm’s output against a global optimization solution (e.g., a semidefinite relaxation or exhaustive search on the fixed-rank manifold) to check whether the algorithm attains the global minimum. If it frequently misses it, the consistency and oracle-adaptivity results do not hold for the implemented procedure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central oracle-adaptivity claim (Theorem 7) rests on the estimated profile bω being consistent for ω0 (Theorem 3) and on classification consistency (Theorem 5). Both rely on Lemma 4, which requires bP_d → P_d,0 for each geometry. For the factor class S_F(r), the population projection P_F is the global Frobenius minimizer over a nonconvex set. The implemented algorithm (Online Appendix B.1) is an alternating-projection heuristic that, as the paper itself states, converges to a stationary point that 'need not be the global minimizer P_F(bΓ_n)' except when bΓ_n is sufficiently close to a regular factor point. In finite samples—especially under weak factor signals or with the diagonal initialization—the algorithm can converge to a different local solution. Then bP_F is not P_F(bΓ_n), so the theoretical consistency and Hadamard-differentiability results do not apply to the object actually computed. Consequently, the oracle-equivalence proof (Theorem 7) is disconnected from the implemented estimator. The simulation evidence in Section 8 uses this heuristic, so the reported near-oracle coverage may not reflect the theoretical profile-guided procedure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a geometric framework for learning the dependence structure of an econometric disturbance vector. Candidate structures—cluster, factor, and sparse—are represented as closed subsets (covariance geometries) of the Hilbert space of symmetric matrices, and an estimable dependence operator Γ̂_n is projected onto each class. The normalized squared Frobenius norms of these projections form a 'dependence profile' ω=(ω_C,ω_F,ω_S). The paper claims local identification of ω under a principal-angle separation condition (Thm 1), a first-order indistinguishability theorem when tangent spaces overlap (Thm 2), consistency and asymptotic normality (Thms 3–4), classification consistency with finite-sample bounds (Prop 2, Thms 5–6), and oracle adaptivity of profile-guided inference (Thm 7, Cor 2). The main applied payoff is that one can select a dependence-robust variance estimator in a data-driven way and match an infeasible oracle that knows the dominant geometry in advance. Simulations and an empirical illustration with Fama–French industry portfolios support the framework.","tokens_in":36831,"tokens_out":8136,"duration_ms":84200,"significance":"The geometric formulation is a useful addition to the econometric toolkit: it makes dependence-structure learning an estimation problem rather than a maintained assumption, and the impossibility result (Thm 2) is a clean formalization of ambiguity. The paper also provides explicit finite-sample classification bounds (Prop 2), a transparent projection-residual diagnostic, and a replication package. If the consistency results were shown for the object actually computed, the profile would be a practical diagnostic with clear interpretation. However, the current manuscript stops short of proving that the implemented estimator satisfies the theoretical conditions, and the 'oracle adaptivity' result is largely a restatement of classification consistency. These gaps are fixable, but they are central to the paper's main claim.","major_comments":[{"comment":"The factor projection actually computed is not the object for which consistency is proved. Online Appendix B.1 states that the alternating-projection algorithm converges to a stationary point that 'need not be the global minimizer P_F(bΓ_n)' except when bΓ_n is sufficiently close to a regular factor point. But Lemma 4 and Theorem 3 require max_d ||bP_d − P_d,0|| = o_p(1) for each geometry, and Theorems 5 and 7 build on that. If the algorithm returns a different stationary point, bP_F is not P_F(bΓ_n), so the theoretical consistency, asymptotic normality, and oracle-equivalence results do not apply to the computed object. Since Sections 8 and 9 use this heuristic, the reported coverage probabilities and profiles are not covered by the theorems. The authors should either prove convergence to the global minimizer with probability tending to one under the maintained assumptions (e.g., initia","section":"Online Appendix B.1; Lemma 4; Theorem 3"},{"comment":"The paper's own simulation evidence violates the principal-angle identification condition. Table E.2 reports θ(T_off_C, T_off_S) = 0° in every design, so Assumption 1 fails for the cluster–sparse pair. Yet Theorems 1, 3, and 4 are stated under Assumption 1, and Theorem 5 relies on Theorem 3. Section E.6 acknowledges the violation and then proceeds to use those same designs to support the profile-recovery and classification conclusions. This creates a mismatch between the formal sufficient condition and the evidence. The authors should state which results hold under the weaker condition of a positive separation margin Δω alone, or modify the theorems/simulations so that the reported evidence directly validates the claimed conditions.","section":"Table E.2; Assumption 1; Theorems 1, 3, 4"},{"comment":"The 'oracle' in Theorem 7 is defined internally as d* = argmax_d ω_d, the population argmax of the profile itself. On the event {b̂d = d*} the equality bV* = bV_d* holds exactly, and P(b̂d = d*) → 1 by Theorem 5. Thus bV* − bV_d* = o_p(1) is essentially a restatement of classification consistency, not an adaptivity result with respect to an external risk or inferential loss. Moreover, no condition links ω-dominance to which variance estimator is actually valid or efficient for the inferential target. The phrase 'infeasible oracle that knows the dominant covariance geometry in advance' is therefore stronger than what is proved. I recommend redefining the oracle in terms of the inferential loss used in Section 7.4, or softening the claim to 'consistent procedure selection given a fixed dictionary'.","section":"Theorem 7; Section 7.4"}],"minor_comments":[{"comment":"The text says the covariance operator has factor and sparse scores exactly tied at 0.445, but Table 6 reports b̂Δω = 0.001 for the covariance operator. If the scores are tied, the margin should be 0. Please reconcile.","section":"Section 9.5; Table 6"},{"comment":"In the factor-geometry paragraph, the proof states 'at Σ_F ∈ M_r the residual Σ_F − P_F(Σ_F) = D_F'; but if Σ_F ∈ S_F(r), then P_F(Σ_F) = Σ_F and the residual is zero. The sentence appears to refer to projection onto the fixed-rank manifold rather than onto S_F(r). The derivative result is standard, but the argument needs correction.","section":"Online Appendix D.2, Lemma D.7"},{"comment":"The positive-semidefinite shift adds the same constant to all diagonal entries when the minimum eigenvalue is negative, which changes the Frobenius objective. The text says the shift 'leaves the objective unchanged when no shift is needed'; explain why the shift is a valid constrained projection or how it affects the convergence claims.","section":"Online Appendix B.1, Step 3"},{"comment":"There is a typo 'Proposition 1 hlds' in the statement of Proposition 1. Also, the proof of Proposition 1 uses continuity of the projection at Γ0, which requires the regularity conditions from Lemma D.13; this should be stated in the proposition.","section":"Section 4.2"}],"recommendation":"major_revision","confidential_remarks":"The factor-projection gap is the main obstacle: the theory concerns the global Frobenius minimizer, while the implementation is a nonconvex heuristic with no global convergence guarantee. This is explicitly conceded in the Online Appendix and affects the central consistency and oracle-adaptivity claims. The paper is otherwise interesting and likely publishable after the authors either prove convergence to the global minimizer under their assumptions or rework the theory around the computed stationary point, and after they recalibrate the oracle-adaptivity language. I do not see an unrecoverable error, but the present gap is too large for acceptance as is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Name],\n\nYou can skip the oracle-adaptivity theorems if you're short on time. The genuinely new thing is the dependence profile itself—normalized Frobenius projection scores onto cluster, factor, and sparse geometries—and the tangent-space indistinguishability result. That part is clean, well-motivated, and worth knowing.\n\nThe paper does several things well. The geometric framing is a real shift: dependence structures become objects to learn rather than assumptions to maintain. The off-diagonal tangent-space restriction is clever, and the near-tie classification analysis is honest about ambiguity. The empirical illustration is also useful: it finds a hybrid structure where no single procedure can be recommended, which is a nice demonstration that the framework can be inconclusive without being useless.\n\nNow the soft spots, in rough order of importance.\n\nFirst, the factor projection. The theory assumes the global Frobenius minimizer over the nonconvex class S_F(r), but the implementation is an alternating-projection heuristic. The paper says outright in Appendix B.1 that the stationary point need not be the global minimizer. That is a load-bearing gap. All the asymptotic results for the profile—consistency, normality, and therefore the oracle-adaptivity theorem—hold for the population projection, not for what the computer actually computes. The stress-test note raises exactly this, and it does hold up on reading. This alone moves the paper from 'clean' to 'conditional'.\n\nSecond, the principal-angle diagnostics in Table E.2 show theta(T_off_C, T_off_S)=0 at every design. Assumption 1 therefore fails for the cluster-sparse pair. The paper acknowledges it and argues that the separation margin still does the work. That's plausible, but it means the local-identification theorem cannot be used where sparsity matters most. The theory needs either a weaker condition or a separate argument for that pair.\n\nThird, the oracle adaptivity result is partly built into the oracle's definition. Since d* is argmax of the population profile, Theorem 7 is essentially classification consistency dressed up as equivalence to an infeasible oracle. It's not wrong, but it's less surprising than the prose suggests.\n\nFourth, the simulations. The factor SE initially had a degeneracy (X'Gamma X=0 by construction), and the reported coverage uses a bias-corrected estimator. Correcting is legitimate, but the reader should know that the headline numbers don't come from the raw procedure. And despite stating replication details, no package is shipped.\n\nOverall, this is not a desk reject. The geometric framework is worth a serious referee, but the gap between the theoretical factor projection and the heuristic algorithm is the main thing to fix. I'd send it out with the expectation of major revision.\n\nBest,\n[Name]","headline":"A genuinely new geometric diagnostic for dependence learning, with a load-bearing gap between the factor projection theory and its heuristic implementation — worth refereeing but in need of serious work.","tokens_in":37352,"tokens_out":3064,"would_cite":false,"duration_ms":31776,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Dependence structures in econometric residuals can be learned as a low-dimensional profile, and profile-guided inference is asymptotically equivalent to an oracle that knows the true geometry.","keywords":["dependence structure learning","covariance geometry","dependence profile","cluster-robust inference","factor models","sparse covariance","oracle adaptivity","principal angles"],"falsifier":"Run the alternating-projection factor algorithm on the same simulated factor design from several random initializations and from the diagonal-truncation initialization at a moderate sample size where the eigenvalue gap is small; if the resulting ω̂ values differ substantially, the profile is not well-defined and the oracle-equivalence claim fails. A second check: construct a design where the cluster and sparse geometries have overlapping supports (e.g., small clusters and large sparsity budget), compute the principal angle, and verify that classification consistency still holds; the paper's si","tokens_in":36382,"feed_emoji":"📈","tokens_out":7120,"duration_ms":68451,"temperature":0.7,"pith_summary":"Econometric inference usually assumes a dependence structure—clustering, latent factors, or sparse interactions—before estimation. This paper makes that structure an estimand rather than an assumption: it projects a consistent estimate of the dependence operator onto cluster, factor, and sparse covariance geometries and records the normalized squared Frobenius norms as a dependence profile ω = (ω_C, ω_F, ω_S). The paper proves the profile is locally identifiable when the geometries' off-diagonal tangent spaces are separated by positive principal angles, and that its estimator is consistent and asymptotically normal. When one geometry dominates, the estimated dominant geometry is consistent, and using it to select a variance estimator reproduces the infeasible oracle's first-order behavior. The paper also proves a first-order impossibility: geometries whose tangent spaces overlap cannot be distinguished by any test, formalizing ambiguity as a structural feature.","feed_headline":"Dependence profile matches an oracle that knows the truth","feed_subtitle":"A three-number geometric summary of cluster, factor, and sparse dependence lets data-driven inference track the infeasible oracle.","key_machinery":"The machinery is the geometric projection of an empirical dependence operator Γ̂_n onto three closed covariance classes in the Hilbert space of symmetric matrices under the Frobenius inner product: the cluster class S_C (support constrained by a cluster-support matrix), the factor class S_F(r) (low-rank plus diagonal, PSD), and the sparse class S_S (at most k_n nonzero off-diagonal entries). For each class, the projection P_d(Γ̂_n) minimizes Frobenius distance; the similarity scores S_d = ||P_d(Γ̂_n)||_F^2 are normalized to give ω. Identification hinges on the off-diagonal tangent spaces T^off_d and their principal angles: positive angles yield local identification (Theorem 1), zero angles y","core_discovery":"The paper's central claim is that the dependence architecture underlying econometric residuals can be summarized by a low-dimensional profile ω = (ω_C, ω_F, ω_S) of normalized squared Frobenius projections of an empirical dependence operator onto cluster, factor, and sparse covariance geometries. Under a principal-angle separation condition on the off-diagonal parts of the tangent spaces, the profile is locally identified and its estimator is consistent and asymptotically normal. When a dominant geometry exists with positive separation margin, the estimated dominant geometry is consistent and the profile-guided variance estimator is asymptotically equivalent to an infeasible oracle that know","pith_inferences":["The paper's own empirical illustration shows the dependence profile differs between covariance and correlation operators, which implies that scale choices for the dependence operator carry information; a natural extension would be to develop an operator-selection or sensitivity procedure around this scale dependence.","Because Theorem 2 identifies an impossibility region, follow-up work could characterize the rate at which geometries become distinguishable as the overlap angle grows, giving a sample-size-adjusted margin for classification.","The algorithmic gap between the population factor projection and the alternating-projection fixed point suggests a testable robustness prescription: report profile estimates under multiple factor initializations, and flag cases where the profile is initialization-dependent.","The framework's oracle-adaptivity result implies that in large samples, profile-guided inference should dominate any fixed robust procedure in terms of coverage accuracy regardless of the true geometry, a claim that could be checked in a multi-design Monte Carlo study."],"forward_implications":["Researchers can choose between cluster-robust, factor-adjusted, and sparse-dependence procedures based on the data rather than a maintained assumption, with first-order behavior matching an oracle (Theorem 7).","Near-ties in the profile are a signal of structural ambiguity; the recommended practice is to report inference from multiple procedures rather than force a single classification.","Projection-residual diagnostics provide an absolute check: a large minimum residual indicates the covariance dictionary is misspecified, and profile scores should be interpreted with caution.","The framework is dictionary-general: any closed covariance class satisfying projection regularity (spatial, network, long-run) can be plugged in, extending the same identification and adaptivity logic.","The profile-weighted variance estimator V̂_avg = Σ_d ω̂_d V̂_d is a natural model-averaging extension, though its efficiency theory is left for future work."],"fun_headline_variants":["Geometric profile learns dependence structure from data","Data-driven dependence selection matches infeasible oracle","Three-number summary guides inference like an oracle","Dependence profile: learnable, adaptive, oracle-approximating","Profile-guided inference rivals an oracle that knows truth"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The asymptotic results rely on the population projections P_d, especially the factor projection onto the nonconvex class S_F(r), being locally unique and Hadamard differentiable at Γ_0; the paper's own alternating-projection algorithm converges to a stationary point that need not be the global minimizer, so if the algorithm lands on a different stationary point, the estimated profile need not be consistent and oracle adaptivity would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Geometric profile learns dependence structure from data","Data-driven dependence selection matches infeasible oracle","Three-number summary guides inference like an oracle","Dependence profile: learnable, adaptive, oracle-approximating","Profile-guided inference rivals an oracle that knows truth"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001156,"raw_usage":{"total_tokens":4608,"prompt_tokens":709,"completion_tokens":3899,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":453,"completion_tokens_details":{"reasoning_tokens":3825}},"tokens_in":453,"tokens_out":3899,"duration_ms":26208,"temperature":1.0,"reasoning_tokens":3825,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T10:32:01.938896+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the alternating-projection factor algorithm on the same simulated factor design from several random initializations and from the diagonal-truncation initialization at a moderate sample size where the eigenvalue gap is small; if the resulting ω̂ values differ substantially, the profile is not well-defined and the oracle-equivalence claim fails. A second check: construct a design where the cluster and sparse geometries have overlapping supports (e.g., small clusters and large sparsity budget), compute the principal angle, and verify that classification consistency still holds; the paper's si","supporting_citations":[],"review_version":2}