{"id":"5c827e9c-a19b-479d-964a-a26975ba441d","arxiv_id":"2608.08625","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Clipping a heavy-tailed return at a price limit and retaining a fraction of the hidden overshoot generates same-sign next-day responses and finite same-limit persistence in a minimal stochastic model.","lead":"This paper proposes that when a stock hits its daily price limit, the unobserved excess pressure is partly carried into the next day, creating memory even though daily shocks are independent. It derives and tests statistical predictions: after a limit move, the next-day return tends to continue in the same direction, with a size that grows with the width of the price band.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Empirical section lacks an out-of-sample test: λ_eff is calibrated at C=20% and the response is compared at that same band, while C=2/5/10% violate the wide-band criterion (Appendix E), so no band independently tests the memory mechanism.","rationale":"The reader's CONDITIONAL verdict is appropriate, and my stress-test does not change it. The theory is coherent: the contractive recursion, the regular-variation tail renormalization (6), the single-dominant-shock asymptotics, and the simulations supporting Eqs. (28), (32), and (33) are internally consistent. Credit is due for the independent ν calibration from 13 global equity indices and for the self-flagged limitation in Appendix E. The load-bearing weakness is that the empirical bridge from model to market is not an independent test. Only C=20% satisfies the wide-band persistence admissibility criterion, and that same band supplies λ_eff; the other bands are out of regime, so the qualitative agreement in Fig. 6 is weaker evidence than it appears. The author's own Appendix E acknowledges this, which is a candid limitation rather than a hidden flaw, but it means the data section alone cannot establish the headline mechanism. A finite-C simulation comparison would settle the issue without altering the theoretical results. I therefore leave the verdict unchanged at CONDITIONAL, with the condition interpreted as requiring such a finite-C out-of-sample check.","tokens_in":20480,"tokens_out":14656,"duration_ms":167314,"concrete_test":"Simulate the exact finite-C Markov chain defined by Eq. (3) with Student-t shocks (ν=3, scale s_ε calibrated to NSE daily volatility), for λ=0.942 and a grid λ∈[0,1), at C=2%,5%,10%,20%. Compute finite-C m±, P±±, and P±∓ and compare to Table III and Fig. 6, with bootstrap confidence intervals for the empirical means. If a single λ reproduces both persistence and response across all four bands, the mechanism is supported; if λ must vary per band or with direction, the central empirical claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is empirical as well as theoretical: retained hidden excess creates memory in real price-limited markets. The theoretical derivations and simulations are internally consistent, so the vulnerable step is the empirical validation in Sec. VI. λ_eff is obtained from persistence at C=20% using the wide-band formula (32); the same band is then used in Fig. 6 to compare the theoretical response (28). This is an in-sample calibration, not an independent prediction. For C=2%,5%,10%, Table III reports pooled persistence 0.7885, 0.4478, and 0.2674, all above the maximum 0.1681 allowed by the wide-band asymptote for ν=3 (Appendix E). The author explicitly states these bands cannot be described by the large-C asymptote for any λ. The wide-band lines drawn in Fig. 6 at those C are therefore outside the model's validity domain. Hence no band provides a clean out-of-sample test: 20% is in regime but not independent; 2/5/10% are independent but out of regime. The observed pattern across bands is consistent with the mechanism only in a loose sense; it does not pin down retention. The model could still be correct—finite-C corrections might reproduce the high persistence—but the paper does not provide that simulation, so the empirical section does not independently establish the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a minimal latent-state model in which clipping of daily returns at an exchange-imposed band [−C,C] creates a hidden excess L_t = X_t − R_t, and a fraction λ of that excess is retained in the next day's latent return. Even with independent driving shocks, the retention couples successive days and generates memory. For symmetric shocks with regularly varying tails, the stationary latent return preserves the tail index with an enhanced amplitude (1−λ^ν)^{-1}. In the wide-band limit C≫s_ε, limit closes are argued to admit a single-dominant-shock description; from this the paper derives the next-day mean response, the same-limit persistence probability, and the opposite-limit reversal probability, all as functions of λ and the tail index ν. Simulations with Student-t shocks support the asymptotic formulas. The empirical section uses NSE limit-close data to compare the predicted mean response with data for C=2%,5%,10%,20%. The paper concludes that the data are qualitatively consistent with the predicted same-sign response and its increase across wider bands.","tokens_in":20782,"tokens_out":3974,"duration_ms":46146,"significance":"If the theoretical claims hold, the paper identifies a genuinely new and minimal mechanism for temporal memory: clipping conceals part of a fluctuation, and retaining that concealed part couples otherwise independent time steps. The derivations in Secs. III-V are nontrivial and internally consistent: Eq. (6) for the tail-amplitude renormalization, Eq. (28) for the linear-in-C response, Eq. (32) for the finite persistence limit, and Eq. (33) for the power-law reversal are all specific, falsifiable predictions, and the simulation comparisons in Figs. 2, 4, and 5 support them. A particular strength is that the wide-band results depend only on λ and ν, not on the full shock distribution. The empirical section is carefully constructed (seven-stage exclusion procedure, corporate-action handling, independent determination of ν), but the validation of the memory mechanism in real markets is incomplete for the reasons detailed in the major comments. The theoretical contribution alone is valuable for a statistical-mechanics audience; the empirical part needs substantial reworking before it can be used as evidence for the central claim.","major_comments":[{"comment":"The empirical validation does not provide an out-of-sample test of the model. The effective retention coefficient λ_eff is estimated from the same-boundary persistence at C=20% using the wide-band formula (32), and the same band is then used in Fig. 6 to compare the theoretical response (28). At C=2%, 5%, and 10%, Table III reports pooled persistence 0.7885, 0.4478, and 0.2674, all above the maximum Q_max(3)≈0.1681 allowed by the wide-band asymptote, as Appendix E itself notes. The theoretical lines drawn in Fig. 6 at those bands are therefore outside the model's validity domain. Consequently no band independently tests the mechanism: C=20% is in regime but not independent, while the narrower bands are independent but out of regime. The empirical section should calibrate on a hold-out subset, predict at a different band, or provide finite-C simulations that cover the observed persistence values.","section":"§VI and Appendix E"},{"comment":"The single-dominant-shock replacement in Eqs. (7)-(9) is the load-bearing approximation for the wide-band predictions (28), (32), and (33). It is justified only asymptotically for C/s_ε→∞, and the paper does not provide finite-C corrections. The empirical persistence values at C=2%,5%,10% lie far above the wide-band ceiling, which shows that finite-C effects are not small in the data. As a result, the apparent agreement of the theoretical lines with the empirical responses in Fig. 6 at intermediate bands is not evidence for the asymptotic theory; it is a comparison between data outside the regime and a formula not intended for that regime. Adding finite-C simulation results, or at least a quantitative estimate of the error of the dominant-shock approximation, is necessary to support the cross-band consistency claim.","section":"§IV-V and Fig. 5"},{"comment":"Even for the one admissible band, C=20%, the direction-resolved persistence is inconsistent with the symmetric model: Table III gives P^{emp}_{++}=0.1325 and P^{emp}_{−−}=0.2045, and the text states that the lower-limit estimate separately remains above the theoretical ceiling. The pooled calibration therefore masks a directional violation of the model's symmetry assumption. Since the same band is then used for the response comparison, the single acceptable data point is not cleanly in the model's support.","section":"Appendix E"}],"minor_comments":[{"comment":"Several axis labels and equations show corrupted symbols (e.g., '10□3' and 'x□(ν+1)'); the final version should fix these rendering issues.","section":"Fig. 2 and equations"},{"comment":"The statement that both parameters are fixed 'without fitting the empirical post-limit-close mean responses' is technically true but potentially misleading; the text should state explicitly that λ_eff is calibrated from persistence at C=20% before the response comparison, and that this is the only band used for calibration.","section":"§VI"},{"comment":"For a paper with a quantitative empirical section, 'available upon request' is weak; depositing the processed daily observations, the band-history reconstruction, and the analysis code would strengthen reproducibility.","section":"Data availability"},{"comment":"The directional asymmetry in persistence is reported but not tested statistically; a confidence interval or test for P^{emp}_{++} vs. P^{emp}_{−−} would make the claimed deviation from the symmetric model more precise.","section":"Table III"}],"recommendation":"major_revision","confidential_remarks":"The theoretical core is sound and publishable in a statistical-mechanics venue, but the empirical section as written does not provide the independent validation claimed at several points in the text. If the editor regards the empirical claim as secondary, the paper could be accepted after the empirical section is reframed as a qualitative illustration and the in-sample calibration caveat is made explicit; if the empirical claim is central, the authors need a genuine out-of-sample design or finite-C simulations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing worth knowing: the theoretical core is solid and genuinely new, and the empirical section is the weak link, not because it is dishonest but because it does not provide the out-of-sample test the abstract implies.\n\nWhat is new: a minimal recursion X_{t+1} = ε_{t+1} + λ(X_t − clip(X_t, −C, C)), where clipping creates a hidden excess that is partially retained. That is a clean mechanism for memory without correlated shocks. The paper shows the stationary tail keeps the noise tail index with enhanced amplitude, and in the wide-band limit a limit close is dominated by a single big shock; the conditional next-day mean response is positive and proportional to C, same-limit persistence approaches 1 − π_0, and opposite-limit reversal is suppressed as C^(−ν). Those results come from a careful application of the single-big-jump principle, with the age distribution of the dominant shock doing the work. Simulations match the predicted scaling. That part is convincing and is a real addition to the literature on clipped stochastic recursions.\n\nSoft spots: the empirical validation does not carry the load. λ_eff is fitted from same-limit persistence at C = 20%, then the response at that same band is compared with Eq. (28); that is in-sample calibration. At C = 2, 5, and 10%, the pooled persistence exceeds the maximum allowed by the wide-band formula for any λ, so those bands are outside the model's validity domain. The author says as much in Appendix E. So no band gives a clean out-of-sample test. The observed pattern is qualitatively consistent, but it does not pin down the retention mechanism. I would also note the lack of confidence intervals on the empirical means and the directional asymmetry (lower-limit persistence above the theoretical ceiling), which the paper acknowledges but cannot explain within the symmetric model. The exclusion rules in Appendix C could bias persistence counts, but that is a minor concern given the transparency.\n\nThe math is internally consistent; the wide-band derivation is asymptotic but standard for heavy tails, and the reliance on a cited contraction-recursion theorem is appropriate. The citation pattern is fine.\n\nVerdict: the theory deserves refereeing. The empirical section should be revised—add finite-C simulations, a genuinely out-of-sample band, and error bars—but the core result stands and would be of interest to stat-mech readers and market-microstructure people. I would accept for peer review and would cite it.\n\nRecommendation: send it out.","headline":"The theoretical core—clipping-induced retained excess as a memory mechanism—is new, plausible, and well-supported by simulation; the empirical section does not independently test it.","tokens_in":21286,"tokens_out":2195,"would_cite":true,"duration_ms":23868,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"When a stock closes at its daily price limit, the unobserved part of the move carries over to the next day and creates memory even though the shocks driving prices each day are independent.","keywords":["price limits","hidden excess","temporal memory","heavy-tailed shocks","regular variation","single-big-jump principle","persistence-reversal asymmetry","stochastic recursion"],"falsifier":"Simulate or observe a heavy-tailed price-limited market at two large band widths $C_1<C_2$; the model predicts $m_+(C_2)/m_+(C_1)\\approx C_2/C_1$ and a same-limit persistence probability that tends to a fixed positive constant as $C$ grows. If the mean next-day response flattens with $C$, or if persistence tends to zero, the single-dominant-shock retained-excess mechanism is falsified.","tokens_in":20207,"feed_emoji":"📈","tokens_out":7614,"duration_ms":75902,"temperature":0.7,"pith_summary":"This paper argues that the unobserved overshoot created when a daily price move is clipped at an exchange-imposed limit can itself be the source of apparent memory. The author introduces a minimal recursion in which the next day's latent return is an independent shock plus a retained fraction of yesterday's hidden excess. For heavy-tailed shocks and a wide price band, a close at the limit is typically caused by one dominant shock, whose age and overshoot determine the following day. The paper derives a same-sign next-day drift proportional to the band width, a finite probability of closing at the same limit again, and a power-law-suppressed chance of hitting the opposite limit, and reports qualitative agreement with exchange data.","feed_headline":"Price-limit clipping makes next-day returns predictable","feed_subtitle":"A minimal model links unobserved limit-close excess to same-sign drift and repeat-limit odds, with no correlated shocks.","key_machinery":"The engine is the single-dominant-shock decomposition of a limit-close history. An upper-limit close on day $t$ is ascribed to a shock of age $j$ (with $j=0$ fresh and $j\\ge1$ inherited through retained excess). To survive $j$ rounds of clipping and retention, that shock must exceed $CB_j$, where $B_j=\\sum_{m=0}^{j}\\lambda^{-m}$; the threshold-exceedance ratio $V$ converges to a Pareto($\\nu$) variable. The age distribution weights histories by $B_j^{-\\nu}/Z(\\lambda,\\nu)$, and the retained fraction $A_j=\\sum_{m=0}^{j}\\lambda^{m}$ converts the overshoot into next-day drift and persistence. This machinery turns the intractable clipped recursion into a tractable mixture of single-shock histories.","core_discovery":"The central claim is that clipping alone does not create memory; clipping plus partial retention does. In the model $X_{t+1}=\\epsilon_{t+1}+\\lambda(X_t-\\mathrm{clip}(X_t,-C,C))$, the hidden excess $L_t=X_t-\\mathrm{clip}(X_t,-C,C)$ is the only coupling between days, so with $\\lambda=0$ returns are independent. For symmetric shocks with regularly varying tails, the stationary law of $X_t$ inherits the tail index $\\nu$ but has its tail amplitude multiplied by $(1-\\lambda^\\nu)^{-1}$. In the wide-band limit, conditioning on an upper-limit close yields $m_+(C)/C\\to M(\\lambda,\\nu)>0$ for any $\\lambda>0$, with $M\\sim\\lambda/(\\nu-1)$ as $\\lambda\\to0$; the same-limit persistence probability $P_{++}(C)$ approaches the finite value $1-\\pi_0$; and the opposite-limit reversal probability satisfies $P_{+-}(C)\\sim k C^{-\\nu}\\Psi(\\lambda,\\nu)$. The same statements mirror for lower-limit closes.","pith_inferences":["The persistence-reversal asymmetry is a general signature: any clipped observable whose hidden part feeds back with a delay should show the same pattern, so the predictions could be tested in order-flow, queue, or neural saturation settings.","The empirical asymmetry between upper and lower limits suggests a natural extension with direction-dependent retention; panic-driven declines would then produce stronger memory than news-driven rallies.","Because the wide-band formulas are parameter-free once $(\\lambda,\\nu)$ are fixed, a clean test is to calibrate $\\lambda$ at the widest band and check whether it predicts the response slope at all narrower bands; the author's own data show this fails at $C=2\\%$, pointing to finite-band corrections."],"forward_implications":["After a limit close, the mean next-day return has the same sign as the close and grows linearly with the band width $C$, for any nonzero retention $\\lambda$.","The probability of closing at the same limit the next day approaches the finite value $1-\\pi_0$ in the wide-band limit, while hitting the opposite limit is suppressed as $C^{-\\nu}$.","Without retention ($\\lambda=0$) there is no memory and the normalized response vanishes; any positive retention, however small, gives a positive response.","The stationary latent return inherits the tail index of the driving shocks but with an amplitude enlarged by $(1-\\lambda^\\nu)^{-1}$.","Exchange data from price-limited stocks show the predicted same-sign next-day response and its increase across wider bands, though the quantitative fit is incomplete."],"supporting_citations":[{"why":"Establishes the single-big-jump principle for heavy-tailed sums, which motivates replacing a limit-close history by one dominant shock.","marker":"[49]"},{"why":"Extends the big-jump principle to correlated increments, justifying the dominant-shock description when earlier shocks survive through retained hidden excess.","marker":"[50]"},{"why":"Gives existence and uniqueness of the stationary distribution for contractive iterated random functions, underpinning the stationary analysis.","marker":"[56]"},{"why":"Provides the regularly-varying stationary-solution asymptotics used to derive the retained tail amplitude in Eq. (6).","marker":"[59]"},{"why":"Supplies the monotone-density theorem used to convert the stationary survival tail into the density tail.","marker":"[57]"},{"why":"Provides the convolution-tail property for subexponential variables used in Appendix A to add the fresh shock and retained excess tails.","marker":"[60]"},{"why":"Supplies the convolution-tail result for regularly varying random variables used in the same amplitude matching.","marker":"[61]"}],"fun_headline_variants":["Hidden excess from price-limit hits seeds return memory","Limit-close excess foretells same-sign next-day drift","Price-limit leftover generates memory and drift","Retained limit excess predicts repeat-limit odds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The wide-band predictions all assume that a limit close comes from one dominant shock, with all other shocks negligible and the latent state on that shock's day equal to the shock itself; that assumption is only asymptotically justified and is not accurate for narrow bands.","fun_headline_variants_meta":{"raw":{"variants":["Hidden excess from price-limit hits seeds return memory","Limit-close excess foretells same-sign next-day drift","Price-limit leftover generates memory and drift","Retained limit excess predicts repeat-limit odds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001392,"raw_usage":{"total_tokens":5650,"prompt_tokens":978,"completion_tokens":4672,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":594,"completion_tokens_details":{"reasoning_tokens":4610}},"tokens_in":594,"tokens_out":4672,"duration_ms":31231,"temperature":1.0,"reasoning_tokens":4610,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:29:53.503592+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate or observe a heavy-tailed price-limited market at two large band widths $C_1<C_2$; the model predicts $m_+(C_2)/m_+(C_1)\\approx C_2/C_1$ and a same-limit persistence probability that tends to a fixed positive constant as $C$ grows. If the mean next-day response flattens with $C$, or if persistence tends to zero, the single-dominant-shock retained-excess mechanism is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the single-big-jump principle for heavy-tailed sums, which motivates replacing a limit-close history by one dominant shock."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extends the big-jump principle to correlated increments, justifying the dominant-shock description when earlier shocks survive through retained hidden excess."},{"cited_title":"Plerou, P","cited_arxiv_id":null,"evidence_quote":"Gives existence and uniqueness of the stationary distribution for contractive iterated random functions, underpinning the stationary analysis."},{"cited_title":"Diaconis and D","cited_arxiv_id":null,"evidence_quote":"Provides the regularly-varying stationary-solution asymptotics used to derive the retained tail amplitude in Eq. (6)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the monotone-density theorem used to convert the stationary survival tail into the density tail."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the convolution-tail property for subexponential variables used in Appendix A to add the fresh shock and retained excess tails."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the convolution-tail result for regularly varying random variables used in the same amplitude matching."}],"review_version":1}