[RST-001]2026
Separating Flow-Driven from Information-Driven Price Impact Across Asset Markets
SSRN Abstract ID 6450561·DOI 10.2139/ssrn.6450561
Abstract. Return covariance is usually treated as a single object, but roughly three-quarters to four-fifths of it turns out to be self-generated. We estimate the market's flow-to-price transfer operator directly, a matrix-valued impulse response mapping signed order flow to price changes across assets and time scales, and use it to split realized covariance into a flow-driven component and an innovation residual that flow cannot explain. The operator is estimated under three hard constraints, checked as optimization gates rather than fixed afterward, cumulative positive semi-definiteness, cross-impact reciprocity, and the exclusion of round-trip arbitrage proxies. A violation produces a recorded failure artifact rather than a silently corrected number. On a calibrated BTC/ETH synthetic baseline, the constrained estimator holds 100 percent primary admissible window survival, 0 percent PSD failure, a mean flow-driven share of 75.5 to 82.5 percent, and out-of-sample R-squared of 76.0 to 82.3 percent. A free study on real Binance Vision trade archives confirms the pipeline executes correctly on public data and produces a coherent contrast between calm and stress periods, though the depth-based falsification test that would confirm the thesis directionally still requires full order-book replay data not available in trade archives alone. We state plainly what the current evidence establishes and what it does not. The calibration and admissibility results support treating this as a market-state upstream of returns, correlations, and factor exposures, requiring a stress-event validation on genuine order-book depth data.
[RST-002]2026
A Common Dealer Constraint Behind Persistent Deviations from the Law of One Price
SSRN Abstract ID 6457180·DOI 10.2139/ssrn.6457180
Abstract. Persistent gaps between prices that should be identical, covered interest parity violations, Treasury cash-futures bases, swap spread dislocations, are usually studied one market at a time. We treat them instead as a single sensor array and invert it. The recovered object is a time-versioned, non-negative vector of implied shadow prices for binding intermediation constraints, solved via a deterministic convex program subject to non-negativity, temporal smoothing, and explicit identifiability diagnostics. Applied to the March 2020 Treasury market stress episode using public CIP, Treasury cash-futures basis, and swap spread data, the recovered three-coordinate constraint state peaks at 152.91 for the balance-sheet coordinate on March 27, the date of the Federal Reserve's emergency repo facility expansion, with a single-day move of +19.11. Quarter-end identification following the difference-in-differences design of Du, Tepper, and Verdelhan (2018) yields p = 0.000777, confirming that the balance-sheet coordinate responds to known regulatory reporting windows. A targeted jackknife diagnostic shows the balance-sheet and secured-funding coordinates are genuinely separated, dropping balance-sheet-dominant sensors shifts the secured-funding coordinate by only 2.6 percent in the three-family build, versus 8.3 percent in the two-family baseline. An out-of-family prediction test, estimating the constraint state from CIP and Treasury basis only, then pricing withheld swap spread observations, yields a mean MAE of 18.68 bp and a latest-date MAE of 3.10 bp, establishing cross-market constraint coherence. The system is built to emit failure artifacts rather than smooth over poor identification. When the loading matrix is ill-conditioned, it reports that the coordinates cannot be separated rather than producing spurious precision. Law-of-one-price deviations across these markets behave as tomographic projections of one latent constraint state rather than separate anomalies, a new measurement primitive upstream of returns, factors, and conventional liquidity proxies.
[RST-003]2026
Measuring the Ricci Curvature of Market Consensus
SSRN Abstract ID 6523041·DOI 10.2139/ssrn.6523041
Abstract. Standard risk models treat market belief space as flat, so a given shift in implied volatility or correlation carries the same assumed consequence regardless of where the market currently sits. This paper treats that space as what it mathematically is, a curved Riemannian manifold under the Fisher-Rao metric, and measures the curvature directly. At the market's current belief state θ(t), the Ricci tensor of the Fisher-Rao manifold of market-implied probability distributions is estimated via automatic differentiation and eigendecomposed into per-direction curvatures κα(t), separating belief dimensions that absorb informational shocks (κα > 0) from those that amplify them (κα < 0). No prior work across 17+ venues in information geometry and financial curvature computes this quantity. Two closed-form results are derived and verified against the known curvature of H² and S², the scalar curvature of the positive-definite cone SPD(d), R = −d(d+2)(d−1)/4, giving R₀ = −35.0 at the paper's working dimension d = 5, and the warped-product curvature of the full N(μ,Σ) manifold, R = −d(d+1)²/4. Estimated on real S&P 500 constituent data, 2005-2025, using three parametric families with analytic Fisher metrics, the pipeline passes 62 of 62 unit tests and all five pre-registered falsification tests. Scalar curvature falls materially below its analytic null during each of eight named stress episodes. A rolling Newey-West HAC regression shows curvature leading VIX at 5- and 10-day horizons (R² = 0.109 at h = 5d) with a lead-lag plateau across 0-12 days that differs significantly by regime (Kruskal-Wallis, p < 0.05), though the 1- and 21-day horizons are directionally consistent without reaching statistical significance. A parallel options-implied pipeline, using Breeden-Litzenberger density extraction across a 25-dimensional cross-asset belief state, finds option-implied curvature leading return-implied curvature by roughly three trading days.
[RST-004]2026
The Minimum Price Adjustment Required to Restore Cross-Market No-Arbitrage
SSRN Abstract ID 6524938·DOI 10.2139/ssrn.6524938
Abstract. A no-arbitrage identity is really a claim about consistency between markets, and real markets do not fully satisfy it. We compute the minimum deformation required to map the market-implied world into the nearest jointly admissible world under a structured library of cross-market financial constraints. This deformation field, measured in constraint-dual sigma units, constitutes a new cross-asset state variable that separates structural contradictions (hard admissibility violations carrying nonzero KKT multipliers and non-recoverable with soft slacks) from softer plausibility pressure. The production engine solves a formal quadratic program with KKT-style dual tension extraction, enforcing hard admissibility faces and soft slacks simultaneously, and the active-set anchored solve is the benchmark leader with composite score 1.4842, hard breach 9.6×10⁻⁵, and feasibility gap 2.106. On the live-public evaluation path, 21 persisted runs, official FRED-distributed series spanning listed volatility, rates, credit, FX, and macro, the system clears all five pre-registered falsification tests, holds a stability score of 0.963 under bid-ask perturbation, reproduces exactly under pinned configuration, retrieves the correct historical analog first 62.1 percent of the time and within the top three 85.3 percent of the time across a 95-episode library, and carries 1.972 times the discriminatory signal of a trivial local-volatility proxy. The current live memo identifies a specific fracture, the 1Y equity tail requires 2.9 sigma of deformation against tail-admissibility, with tail-admissibility and cross-horizon transport jointly binding at 0.49 confidence and a lead tension residual of 0.1146 against a hard tolerance of 0.04. We state plainly what the current live-public evidence establishes and what it does not. The deformation field and its constraint decomposition are supported as a genuine structural-inconsistency primitive by the calibration and falsification evidence gathered so far, and broader validation on a licensed institutional data stack is the natural next step to extend it beyond the live-public data plane.
[RST-005]2026
The Rotational Component of Return Covariance and Its Predictive Power for Volatility
SSRN Abstract ID 6597020·DOI 10.2139/ssrn.6597020
Abstract. Close to half of observed cross-asset return covariance comes from a source classical finance has no name for. We split the drift matrix of the standard multivariate Ornstein-Uhlenbeck return model into its symmetric and antisymmetric parts, B = S + Q, the same decomposition used throughout non-equilibrium statistical mechanics, and derive a closed-form Lyapunov equation for the resulting circulatory covariance perturbation, SΔC + ΔCS = −[Q, C]. Across five equity universes spanning 11 to 315 assets and 1963 to 2026, this rotational, non-equilibrium component accounts for 45.5 to 54.9 percent of observed covariance, with permutation z-scores of 19.4 to 117.7 (all p < 0.001). The effect is not a curiosity. The dominant circulatory frequency ω₁, the leading eigenvalue of Q, acts as a stress-peak detector, a one-unit increase predicts a 2.6 to 9.9 percentage-point decline in the VIX over the following one to six months (t = −2.10 to −2.21, p < 0.04, NW-HAC, N = 291 monthly observations), and circulatory beta prices the Russell 1000 cross-section with a significant anti-premium (t = −2.10). No prior work identifies this covariance source or estimates its pricing content.
[RST-006]2026
An Upper Bound on Achievable Sharpe Ratios from Market Irreversibility
SSRN Abstract ID 6606258·DOI 10.2139/ssrn.6606258
Abstract. Every standard risk metric in finance, variance, covariance, Value-at-Risk, beta, gives an identical answer whether fed a return series forwards or backwards in time. That symmetry is a mathematical property of the metrics themselves, not an empirical claim about markets, and it means the entire apparatus of modern risk management is blind to which way time is actually running. We make that arrow visible directly. Treating the market as a multivariate Ornstein-Uhlenbeck process, we extract the probability current field generated by its asymmetric drift, compute the entropy production rate this current implies, and split that rate into a housekeeping component, the cost of maintaining the market's current correlation structure, and an excess component, the cost of that structure actually changing. A Thermodynamic Uncertainty Relation bound then converts this single physical quantity into a hard ceiling on the Sharpe ratio any linear strategy can achieve at a given horizon. Across five equity universes, the S&P 500, GICS sector ETFs, macro futures, financials, and technology, the entropy production rate exceeds a symmetric null by up to 398 standard deviations and is essentially uncorrelated with realized volatility, Spearman ρ = 0.131, confirming it measures something classical risk metrics cannot see at all. At the sector and macro level the housekeeping-excess split correctly separates the 2008 financial crisis, dominated by friction, from COVID-19, dominated by a genuine regime shift, with a relative difference of 0.65 at the macro level. That same split fails once individual stocks replace sector aggregates, since a correlated sell-off inflates both components at once, a result we report as a real boundary of the method rather than hide. The Sharpe ceiling itself is not vacuous, it lands at 1.19 per asset for the S&P 500 at a one-month horizon, squarely inside the range systematic strategies actually realize.
[RST-007]2026
The Capital Required to Distinguish Market Outcomes Under Financing Constraints
SSRN Abstract ID 6639600·DOI 10.2139/ssrn.6639600
Abstract. Two markets with the same tradable menu and the same classical measures of completeness can require very different capital to act on the difference between them. We introduce the capital separation function, the minimum cost of a portfolio that distinguishes two future market states inside real financing, margin, and concentration limits. It reduces exactly to classical Arrow-Debreu completeness as frictions vanish, a result proved as a theorem. On a 30-snapshot synthetic stress-regime study, calibrated to the historical volatility and term-structure paths of the March 2020 and April 2025 windows, six matched pairs share an identical instrument menu and identical classical rank diagnostics. The clearest pair, March 5 versus March 6, 2020, has identical payoff-span, Jacobian, and Greek-space rank, yet requires more than twenty times the capital to separate. All seven pre-registered kill rules clear. The result is a market that looks complete by classical diagnostics yet remains too expensive to trade, a gap risk frameworks cannot see.
[RST-008]2026
Forced Rebalancing in Leveraged ETFs Predicted from Public Disclosure Rules
SSRN Abstract ID 6715359·DOI 10.2139/ssrn.6715359
Abstract. A rule that forces a fund to trade on a specific day can be read directly off its own prospectus, without needing to estimate anything statistically, and that is the entire premise here. For leveraged and inverse exchange-traded funds, the daily reset itself defines a closed-form rule, ΔE = β · A · (β − 1) · r, for exactly how much notional a fund is mechanically compelled to move given only its stated leverage factor, its disclosed asset base, and the day's realized return. We apply this rule to an append-only, SHA-256-hashed registry of 18 certified ProShares products, deriving sign and materiality boundary surfaces at four pre-registered thresholds, automated crossing detection, five integrity gates, and a one-page per-cohort memo that reports ACCEPT, REROUTE, KILL, or UNIDENTIFIED. Replaying two real stress windows, the August 14, 2015 and February 2, 2018 information cutoffs, against ProShares Trust II's actual SEC EDGAR filings, the predicted direction of quarter-end VIX-futures position changes across SVXY, UVXY, VIXM, and VIXY matches the realized direction in all 8 of 8 ticker-episode pairs. At daily resolution the result is honestly weaker. The rule ties the strongest adversarial baseline built from the same closed-form structure in both replay windows, and the August 2015 ablation suite fires a pre-registered hard-kill artifact rather than being suppressed. The pre-registered threshold, 20 or more crossing-aligned daily events beating that baseline, is not yet crossed, and the paper states plainly that the next valid evidence is daily official action data, not a relaxed proxy. Every run, artifact, and failure state is persisted under a parent-hashed manifest chain, and 43 tests reproduce both replays byte-for-byte.
[RST-009]2026
Why Monthly Disclosure Cannot Identify the Distribution of Unrealized Gains
SSRN Abstract ID 6781319·DOI 10.2139/ssrn.6781319
Abstract. Fund disclosure rules assume that knowing how much of a portfolio changed each month is enough to tell who is sitting on gains and who is underwater. We test that assumption directly, on the one market where the truth is fully observable, Bitcoin. Using the complete public transaction record, we reconstruct the actual distribution of unrealized gains and losses among short-term holders, then degrade our own information to match the monthly reporting rhythm the SEC requires of registered funds. In a calm accumulation regime, that monthly rhythm is enough. In a post-bubble regime, it is not. The identified range of possible outcomes nearly doubles, from 38 percent of the price range to 67 percent, precisely because volatility is high, which is exactly when investors most need to know who is exposed. Monthly disclosure works when markets are calm and fails when the truth matters most. A sweep across faster and slower reporting cadences shows the boundary lies between 45 and 60 days, not at one month, for the regime where it counts. We report this honestly as the primary finding, not a footnote to a simpler result the monthly test alone would have hidden.
[RST-010]2026
Overlapping Institutional Trading Calendars and the Direction of Overnight Returns
SSRN Abstract ID 6818821·DOI 10.2139/ssrn.6818821
Abstract. Most research on the closing print studies one institutional calendar at a time, index reconstitution, options expiration, futures rolls, leveraged ETF rebalancing. We ask what happens when several of these clocks land on the same asset in the same closing window, and whether that overlap itself predicts returns. We measure it directly from public institutional calendars and public price and volume data, decomposing scheduled trading mass exactly into a self-mass term and a cross-clock overlap term. Across a fifteen-ETF daily panel from 2016 to 2024, 32,130 ETF-days, the overlap term is genuinely non-spurious, sitting at the 0th percentile of a 1,000-iteration calendar-randomization null, with a power analysis confirming the design can detect an effect as small as 2.9 bp per standard deviation. The headline result runs against what we pre-registered. Higher coordination predicts overnight return continuation, not the transitory reversal we expected, an overlap coefficient of −5.03 bp per standard deviation, t = −2.61. The effect is specific to modern market structure, absent across the full 1993 to 2024 history at t = −0.83, present from 2010 onward at t = −2.46, and strongest in the most recent years, tracking the growth of leveraged ETFs and short-dated options. A second, sign-consistent but underpowered pattern reemerges at the most extreme coordination events, quarter-end days, t = 1.94 in individual stocks. We pre-registered and rejected four directional hypotheses across five separate phases, recording each rejection as its own artifact. What the evidence establishes is that scheduled coordination is a real, measurable primitive whose dominant signature is continuation rather than transient pressure in liquid instruments. What it does not yet establish is that isolating the reversal channel with confidence requires closing-auction imbalance data this study does not have.
[RST-011]2026
The Backlog of Required but Unpublished Disclosures and the 2024 Schedule 13D Reform
SSRN Abstract ID 6866499·DOI 10.2139/ssrn.6866499
Abstract. Every mandated disclosure rule is a forward map from a fact that is already true to a date on which the law forces it into the open. We build a conserved inventory of facts already true but not yet public across Form 4, 13F, and Schedule 13D and 13G, using only free EDGAR and Yahoo Finance data. The timing of each fact is computed exactly from public rules, its dollar magnitude is bounded rather than guessed, and the whole ledger is hashed for byte-identical reproduction. The conservation identity holds at all 428 evaluation dates, and a deliberately corrupted input correctly breaks it. On 1,536 real Form 4 facts, the bounded interval contains the true reported value 1,535 times. We then use the February 2024 Schedule 13D deadline change as a natural experiment. Compliance within the new window jumps from 39 to 80 percent while the average filing lag barely moves, an honest and mixed result. A pre-registered test of whether the aggregate predicts abnormal returns fails against a calendar null, a result we report in full, while a narrower cross-statute signal beyond single-statute holdings data survives false-discovery correction.