{"id":"36cd96ab-5668-4aa6-8485-6032ffb3676c","arxiv_id":"2509.01263","paper_version":3,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A duopoly model with observational learning, random arrivals, and costly search yields mixed-strategy price dispersion whose width shrinks with signal precision and grows with search costs.","lead":"This paper models a duopoly where consumers observe past purchases but not payoffs, receive private signals, and can pay a search cost before buying. It claims that feedback between social learning and search frictions makes firms randomize over prices, generating price dispersion even between identical firms.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reny better-reply security in Appendix A Step 4 is asserted only for atomless opponent mixes; since existence must cover atomic mixed profiles too, the static mixed-strategy equilibrium—the paper's central dispersion result—is unproven.","rationale":"The reader's weakest_assumption identifies the same load-bearing gap: the better-reply security verification in Appendix A Step 4 is asserted, not proved, and it improperly assumes atomless opponent mixes before existence is known. This is not a cosmetic issue: the abstract's headline claim is that a mixed-strategy pricing equilibrium exists and yields price dispersion, and Proposition 5.1 is the only existence result. If the Reny argument fails, the paper has no proof of a mixed equilibrium; Proposition 5.2 is conditional on mixing and does not establish that mixing occurs. The consumer-side results (Lemmas 4.1–4.3, Propositions 4.1–4.2) are cleaner and not the main problem, but the firm-side central claim is the paper's contribution and it rests on an unverified technical condition. The numerical simulations in Appendix H are suggestive but cannot substitute for a rigorous existence proof, especially since the solver uses softmax smoothing and grid approximations that do not verify the exact indifference conditions of Proposition 5.2. Therefore the reader's REJECT verdict is appropriate: the central theorem is unproven as written. A revised version with a complete existence proof—or an alternative construction of the mixed equilibrium—could change this assessment, but the current manuscript does not provide it.","tokens_in":23464,"tokens_out":5214,"duration_ms":69136,"concrete_test":"Test better-reply security directly on the mixed extension: fix an atomic opponent strategy σ_{-i}=r δ_{p1}+(1-r)δ_{p2} for a fine grid of p1,p2,r, and compute whether firm i has a strategy that secures payoff within ε of its best reply at every profile in a neighborhood, including price differences where a reachable posterior equals the cutoff (4.4). If any atomic profile violates ε-security, Appendix A Step 4's atomlessness assumption is essential and the Reny argument as written fails; the existence theorem would need a different proof.","verdict_should_be":"REJECT","load_bearing_attack":"The central claim is Proposition 5.1/Theorem A.1: a (possibly mixed) BNE exists, and Proposition 5.2 turns this into a mixed equilibrium with connected overlapping supports. Existence is delegated to Reny (1999). The verification in Appendix A Step 4 says \"Because K is null and σ_{-i} is atomless, small perturbations of p_i shift the path distribution only on events of arbitrarily small probability... Hence the game is better-reply secure.\" This is not a proof. In the mixed extension, opponents' mixed strategies may have atoms; Reny's better-reply security must hold at every mixed profile, including atomic ones. Atomlessness is a property that has to be derived for equilibrium strategies, not assumed for arbitrary profiles. The proof also does not establish payoff security at profiles where the opponent places mass on a price that, together with p_i, makes a reachable posterior equal the cutoff (4.4) with positive probability; under such a profile the relevant discontinuities are not measure-zero under the opponent's mix. Proposition 5.2 does not fix this: it is conditional on \"any symmetric BNE with mixing\" and never proves mixing occurs for interior κ; the no-atom/gap/overlap arguments are sketches relying on \"standard mixing logic\" and \"continuity of best replies\" that are exactly what is in question. Thus the existence of the mixed-strategy equilibrium that generates price dispersion is not established by the supplied arguments.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper embeds BHW-style observational learning in a symmetric duopoly with random arrivals, search costs, and posted prices. Consumers observe past purchases (not payoffs), receive private signals, and may pay a search cost to check the alternative seller. The paper derives a posterior cutoff rule (Lemma 4.1), closed-form absorbing cascade boundaries (Proposition 4.1), and shows wrong cascades occur with positive probability but vanish as signals become precise or search becomes cheap (Proposition 4.2). On the firm side, it claims a mixed-strategy Bayesian Nash equilibrium exists in the static price game (Proposition 5.1), that symmetric equilibria with mixing have connected overlapping supports (Proposition 5.2), and that support width decreases in signal precision and increases in search cost (Proposition 5.3). It also develops comparative statics for absorption and prices, a welfare decomposition with a Pigouvian search subsidy (Propositions 6.3-6.4), and extensions to Calvo price resets (Proposition 7.1) and prominence (Propositions 7.4-7.5).","tokens_in":23801,"tokens_out":14052,"duration_ms":160526,"significance":"The consumer-side analysis is the strongest part: the threshold rule and cascade boundaries are closed-form, and the invariance of absorption probabilities to the arrival rate is a clean observation. If the equilibrium existence and mixing results were rigorously established, the paper would offer a tractable model of price dispersion driven by social-learning feedback rather than exogenous heterogeneity, with sharp comparative statics and reproducible simulation details. However, the central equilibrium existence proof rests on an asserted better-reply security argument that is not supplied, and the mixing characterization is not proved. These gaps block the paper's main claim at present.","major_comments":[{"comment":"The abstract states that the model yields a 'unique symmetric atomless price equilibrium' and that a 'regularized Calvo game' has a stationary Markov equilibrium on a countable reachable state space. Neither claim appears in the main text: Proposition 5.1 only asserts a 'possibly mixed' BNE, and Proposition 5.2 is conditional on an unspecified symmetric BNE with mixing. The abstract should be reconciled with the theorems actually proved.","section":"Abstract / §5"},{"comment":"The proof of better-reply security is asserted only for atomless opponent mixed strategies. Reny's theorem requires better-reply security at every mixed profile, including those with atoms; atomlessness is not an a priori property but a property one would need to derive. Step 3's claim that threshold sets have measure zero under the induced process also depends on the opponent's strategy. Without a rigorous argument covering atomic opponent profiles, existence of the static mixed BNE is not established. This is load-bearing because Proposition 5.1 underpins the abstract's central claim of a mixed-strategy pricing equilibrium.","section":"Appendix A, Step 4 / Prop. 5.1"},{"comment":"The dispersion theorem is conditional on 'any symmetric BNE with mixing,' but the paper never proves that mixing occurs for interior κ. The proof sketch invokes 'standard mixing logic' and 'continuity of best replies' even though best replies are discontinuous. The no-atoms, connected-support, and overlap arguments need a formal proof. As written, the paper does not establish the existence of price dispersion in equilibrium; it only characterizes a hypothetical mixed equilibrium.","section":"Section 5.2, Prop. 5.2"},{"comment":"The text says the planner uses a 'lower buy-threshold' and that the subsidy 'shifts the private cutoff down.' But from Eq. (4.4), a search subsidy lowers the effective κ and therefore raises x*. If the planner values the information generated by search, the planner's cutoff should be higher than the private cutoff, not lower. The formal formula for sP may be correct, but the direction stated in the text and proof sketch is inconsistent with the model. This sign error affects the interpretation of the welfare result and must be corrected.","section":"Section 6.3, Prop. 6.4"}],"minor_comments":[{"comment":"Assumption 7.1 and the paragraph immediately after it both state that each firm receives independent Poisson reset opportunities with hazard α>0; the duplication should be removed.","section":"Section 7.1"},{"comment":"The notation is confusing: Eq. (4.4) defines x* as the posterior cutoff, but §5.1 refers to it as η*. Use distinct symbols or explicitly define the relationship.","section":"Eq. (4.4) / §5.1"},{"comment":"The caption should state clearly that the plotted price distributions come from the approximate grid/MC solver in Appendix H, not from a closed-form equilibrium solution.","section":"Figure 3"},{"comment":"The MIX-SOLVE algorithm is a heuristic best-reply iteration; the paper should not imply that its output verifies Proposition 5.2 or Proposition 5.3.","section":"Appendix H"},{"comment":"The reference list has a typo: 'V arian' should be 'Varian'.","section":"References"},{"comment":"The existence proof for the Calvo game is a sketch. The claimed Kakutani-Fan-Glicksberg argument requires a precise topology on the space of measurable policies and a proof of upper hemicontinuity of the best-reply correspondence and compactness; these points are not supplied.","section":"Appendix E / Prop. 7.1"}],"recommendation":"major_revision","confidential_remarks":"The paper has a promising framework and the consumer-side results are genuinely useful, but the central existence and dispersion theorems are not rigorously proved, and the abstract overstates what is established. The sign error in Section 6.3 also needs careful checking. I would be willing to review a revised version if the proofs are completed and the claims are aligned with the theorems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper is worth reading for the consumer-side analysis, but the headline result on price dispersion is built on an existence proof that is a sketch with a serious gap. I would send it to referees, but I would not cite it as it stands.\n\nWhat's genuinely new: embedding BHW observational learning in a duopoly with random arrivals, costly search, and posted prices, and deriving closed-form cascade boundaries (Prop 4.1) and absorption comparative statics. The consumer threshold rule (Lemma 4.1) and the wrong-cascade results (Prop 4.2) are clean and I think correct. The intuition for dispersion—learning feedback rotates demand and makes rivals indifferent over a price range—is appealing and not in the cited literature.\n\nThe soft spot is exactly where the reader points. Appendix A Step 4 asserts better-reply security for atomless opponent mixes, but Reny's theorem needs to hold for all mixed profiles, including atomic ones. Atomlessness is an equilibrium property, not something you can assume for arbitrary profiles. And Proposition 5.2 is conditional on 'any symmetric BNE with mixing'; nothing shows mixing occurs for interior parameters. The numerical solver in Appendix H is not a substitute. So the existence of the mixed-strategy equilibrium that generates dispersion is unproven. The abstract's claim of a mixed-strategy pricing equilibrium is stronger than what the formal results support.\n\nMinor issues: Proposition 6.4's Pigouvian subsidy is nearly definitional—the subsidy is set to equal the information externality, so implementation is by construction. That's standard, not a flaw. The Calvo existence argument is also a sketch, but less central. And the arXiv metadata (abstract, author name) doesn't match the full text; looks like a versioning error that should be fixed before any resubmission.\n\nWho is this for? IO and social learning researchers will find the consumer-side machinery useful and the price dispersion mechanism provocative. The paper deserves a serious referee—the gap is technical and possibly fixable—but it needs a real existence proof and a statement of what is actually shown. My recommendation: send to referees, expect major revision.","headline":"A genuinely novel consumer-search-learning model with clean cascade results, but the central mixed-strategy pricing equilibrium is not proven—the better-reply security step is asserted, not shown, and the paper never proves mixing occurs.","tokens_in":24270,"tokens_out":2182,"would_cite":false,"duration_ms":27137,"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":"This paper claims that social learning plus switching costs makes identical duopolists randomize over connected, overlapping price ranges, so price dispersion arises without any cost or preference heterogeneity.","keywords":["social learning","informational cascades","price dispersion","search costs","duopoly","mixed-strategy equilibrium","Calvo pricing","welfare"],"falsifier":"At a parameter setting where a reachable posterior equals the cutoff in equation (4.4) exactly, compute each firm's profit against a candidate atomless rival mixture; if profits jump at that pair or the best-response correspondence fails the security condition used in the existence proof, the mixed-equilibrium existence claim collapses even if the dispersion pattern is later observed.","tokens_in":23360,"feed_emoji":"🛒","tokens_out":9125,"duration_ms":99792,"temperature":0.7,"pith_summary":"The paper studies a duopoly in which consumers arrive one by one, observe past purchase choices but not realized payoffs, receive a private signal about which product is better, and can pay a switching cost to check the rival before buying. It argues that this environment produces informational inertia: a region of beliefs where each consumer stays with her default seller, so purchases happen but reveal nothing. The main discovery is that when firms post prices once and search costs are interior, the pricing game has a mixed-strategy equilibrium in which two ex-ante identical firms randomize over connected, overlapping price ranges. Dispersion narrows as private signals become more precise and widens as switching and search costs rise, and wrong cascades occur with positive probability but vanish in the precise-signal, cheap-search limit. The paper also shows that random price-reset opportunities compress steady-state dispersion and that a search subsidy can implement the planner's cutoff.","feed_headline":"Herding makes identical sellers randomize prices","feed_subtitle":"Social learning plus switching costs sustains price dispersion—and lasting wrong fads—even in a two-firm market.","key_machinery":"The carrying object is a one-line consumer cutoff, the posterior threshold x*(pA−pB, κ) = 1/2 + (pA−pB−κ)/2, which decides when a consumer buys from the default seller without checking and when she pays κ to switch. Lifted to the public belief, this cutoff generates two closed-form absorbing boundaries for up- and down-cascades. The interval between those boundaries is the informational-inertia region: purchases there are partially informative, while outside it actions herd and beliefs stop moving. The same cutoff enters firms' discounted demand kernel, so the profit discontinuities and the indifference conditions that support mixed pricing are all expressed through it.","core_discovery":"The paper claims that in a symmetric duopoly with fixed posted prices, random consumer arrivals, observational learning, and a search or switching cost, there is a mixed-strategy Bayesian Nash equilibrium in which both firms randomize over prices despite being ex-ante identical. The equilibrium price supports are connected intervals that overlap, the price distributions have no atoms, and expected profit is constant across each support. The width of the support is treated as a summary of price dispersion: it falls with signal precision q and rises with search cost κ. Because actions are observed but payoffs are not, the public belief has absorbing cascade regions; wrong cascades have strictl","pith_inferences":["Beyond the paper's stated results, the support-width comparative statics suggest a structural estimation strategy: cross-market variation in observed price ranges across otherwise similar duopolies could identify signal precision and switching costs.","The model implies a trade-off for platforms: neutral prominence minimizes wrong-cascade risk ex ante, but a platform with even slightly informed priors would tilt prominence toward its favored arm—an extension the paper notes but does not develop.","If the price-reset hazard were a firm choice, a firm whose beliefs are turning against it would want faster repricing; this could generate endogenously asymmetric price flexibility, a question the static Calvo setting leaves open.","Because public reviews act like an increase in effective signal precision, the paper's comparative statics imply that any policy raising the informativeness of observed outcomes should compress price dispersion even without direct price regulation."],"forward_implications":["In markets where quality is opaque and buyers can check the rival at a cost, identical firms should be observed charging different prices; dispersion is a systematic feature, not a sign of cost differences.","The support width gives a measurable proxy for informational frictions: high signal precision should compress price ranges, while high switching or search costs should widen them.","Wrong cascades—markets settling on the inferior product—should occur with positive probability whenever signals are bounded and checking is costly, and that probability should be invariant to how frequently consumers arrive.","A search subsidy calibrated to the expected future welfare gain from a check can align private and social learning incentives; the remaining welfare loss splits cleanly into wrong purchases and excess search.","More frequent opportunities to reset prices discipline the market: faster repricing compresses steady-state price dispersion and, with sufficiently informative signals, lowers the chance of landing on the wrong product."],"supporting_citations":[{"why":"Supplies the canonical observational-learning cascade structure (actions observed, payoffs not) that the paper embeds in duopoly pricing.","marker":"Bikhchandani et al., 1992"},{"why":"Provides the herd-behavior benchmark that motivates the absorbing-cascade regions.","marker":"Banerjee, 1992"},{"why":"Supplies the bounded-signal condition and the pathological-outcome logic used for wrong-cascade results.","marker":"Smith and Sørensen, 2000"},{"why":"Gives the classic search-cost price-setting benchmark that emerges as the high-κ limit.","marker":"Diamond, 1971"},{"why":"Sets the baseline model of price dispersion from heterogeneity that this paper contrasts with learning-driven dispersion.","marker":"Varian, 1980"},{"why":"Provides an equilibrium price-dispersion benchmark from sampling technologies, used as a comparison for the dispersion mechanism.","marker":"Burdett and Judd, 1983"},{"why":"Supplies the sequential-consumer-search framework that the paper's search-and-switching technology extends.","marker":"Stahl, 1989"},{"why":"Analyzes pricing in dynamic social learning and establishes signal boundedness as a key determinant of asymptotic learning.","marker":"Arieli et al., 2022"},{"why":"Connects random search with observational learning, informing how observed choices shape search incentives.","marker":"Garcia and Shelegia, 2018"},{"why":"Provides the random price-reset device used in the paper's staggered-pricing extension.","marker":"Calvo, 1983"}],"fun_headline_variants":["Switching costs make sellers randomize prices","Price dispersion from herding and switching costs","Fads persist when identical firms randomize prices","Switching costs seed price randomness in duopoly"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that each firm's profit jumps only on knife-edge price pairs that a rival's mixed strategy never hits; the paper asserts this rather than proving it, and the asserted atomlessness of the rival's mixture is part of what existence must establish.","fun_headline_variants_meta":{"raw":{"variants":["Switching costs make sellers randomize prices","Price dispersion from herding and switching costs","Fads persist when identical firms randomize prices","Switching costs seed price randomness in duopoly"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000973,"raw_usage":{"total_tokens":3958,"prompt_tokens":714,"completion_tokens":3244,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":458,"completion_tokens_details":{"reasoning_tokens":3196}},"tokens_in":458,"tokens_out":3244,"duration_ms":24440,"temperature":1.0,"reasoning_tokens":3196,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:42:39.123764+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"At a parameter setting where a reachable posterior equals the cutoff in equation (4.4) exactly, compute each firm's profit against a candidate atomless rival mixture; if profits jump at that pair or the best-response correspondence fails the security condition used in the existence proof, the mixed-equilibrium existence claim collapses even if the dispersion pattern is later observed.","supporting_citations":[{"cited_title":"A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades,","cited_arxiv_id":null,"evidence_quote":"Supplies the canonical observational-learning cascade structure (actions observed, payoffs not) that the paper embeds in duopoly pricing."},{"cited_title":"A Simple Model of Herd Behavior,","cited_arxiv_id":null,"evidence_quote":"Provides the herd-behavior benchmark that motivates the absorbing-cascade regions."},{"cited_title":"Pathological Outcomes of Observational Learning,","cited_arxiv_id":null,"evidence_quote":"Supplies the bounded-signal condition and the pathological-outcome logic used for wrong-cascade results."},{"cited_title":"A Model of Price Adjustment,","cited_arxiv_id":null,"evidence_quote":"Gives the classic search-cost price-setting benchmark that emerges as the high-κ limit."},{"cited_title":"Equilibrium Price Dispersion,","cited_arxiv_id":null,"evidence_quote":"Provides an equilibrium price-dispersion benchmark from sampling technologies, used as a comparison for the dispersion mechanism."},{"cited_title":"Oligopolistic Pricing with Sequential Consumer Search,","cited_arxiv_id":null,"evidence_quote":"Supplies the sequential-consumer-search framework that the paper's search-and-switching technology extends."},{"cited_title":"The Implications of Pricing on Social Learning,","cited_arxiv_id":null,"evidence_quote":"Analyzes pricing in dynamic social learning and establishes signal boundedness as a key determinant of asymptotic learning."},{"cited_title":"Consumer Search with Observational Learning,","cited_arxiv_id":null,"evidence_quote":"Connects random search with observational learning, informing how observed choices shape search incentives."},{"cited_title":"Staggered Prices in a Utility–Maximizing Framework,","cited_arxiv_id":null,"evidence_quote":"Provides the random price-reset device used in the paper's staggered-pricing extension."}],"review_version":1}