Liquidity, trading costs and capacity

Expanded snapshot evidence and strategy diagnostics

Ten predeclared dates test displayed liquidity. Costs are conditional on filling the whole order within ten displayed levels; read them together with coverage and sample counts.

A static book does not establish a fill, market impact, or executable capacity. Order-flow association has not passed the dynamic replay gate.

1 bp equals 0.01%. Costs use the contemporaneous midpoint and include half-spread and displayed-depth consumption; fees, latency, replenishment, queues and own-order impact are excluded. Coverage divides fully covered orders by all valid two-sided books.

Liquidity group comparison

Full findings and reproduction record ↗

56-session strategy diagnostic, January–March 2025

The model uses a 5% daily-amount participation limit, T+1 and price-limit checks. Fractional shares and adjusted-close proxies simplify fills. Cost references include later 2026 observations and are counterfactual assumptions, not an ex-ante 2025 model or stock/time calibration. Original release times, corporate actions and delisting settlement remain unverified. These results do not establish real capacity or net alpha.

Fee sensitivity assumes 1 bp inclusive commission per side and 5 bp seller stamp duty, with zero minimum commission and no extra transfer fee. Static costs replace cash debits; the original 5 bp simulator charge is removed.

Download aggregate evidence (expansion)Download aggregate evidence (strategy)

Early and recent common-sample comparison

The common-security comparison still has different minute coverage. Paired results use only security-minutes with valid books on both dates; neither comparison isolates a market-wide time trend. Common securities also condition on survival.

In the 106 common securities, observed minute coverage is 100% on 2021-01-04 and 96.99% on 2026-04-16. Earlier Shanghai trades have no corresponding order records in this selected sample; Shenzhen order-ID match rates are 99.68%/99.75%. More complete book minutes do not establish more complete event streams.

Minute execution diagnostic: 40 main-board securities

The second half of 2024 and full year 2025 start independently with cash. Their different lengths prevent annual return rankings. This public 40-security main-board sample differs from the earlier 400-security micro-cap / 56-session diagnostic; the results cannot be compared directly.

Full-year 2024 remains unpublished. Reviewed cross-source evidence permits a trial of raw-price accounting without changing adjustment factors. The CNY 1m and CNY 10m fee pilots are incomplete: three held delisted securities (600321, 002621 and 000982) lack supported valuation or disposition proceeds from their delisting dates. Holdings cannot be removed, valued at zero or carried at their last price to fill this gap. Vendor factor internals and historical revisions remain unverified.

Separate assumption-based studies retain delisted shares and change valuation without creating disposal cash. They do not resolve the strict annual evidence gap or replace the 18 scenarios below.

Retrospective validation: the CNY1 price guard improved the assumption-based 2024 account but changed no 2025 fills. Across 24 scenario records (12 simulations and 12 verified ledger reuses), higher returns sometimes coincide with lower completion. These results do not establish live capacity.

Completion aggregates all buy and sell parent orders: filled quantity divided by target quantity, weighted by each order’s target notional at its own arrival reference price. Repeated orders count again; this does not measure the remaining orders at the end of the window.

Execution uses raw minute prices, 100-share lots, T+1 and a 5% cap per minute. Release horizons of 30/120/330 elapsed minutes include lunch. Dated corporate actions are applied; confirmed halts retain nontradable valuation marks.

The minimum-commission scenario assumes CNY 5 once per parent order; actual broker charging has not been verified. Gross scenarios retain execution constraints while omitting fees.

P90 arrival shortfall uses available raw or synthetic arrival references and combines price movement, unfilled opportunity cost and fees. It is an accounting diagnostic, not causal own-order impact. Actual order data are absent, so these simulations establish neither live capacity nor impact as the dominant loss.

What the existing data can support

Both dates reproduce the earliest-minute selections with zero mismatches. On 2026-04-16, the original source is missing 960 minutes across 80 sampled stocks; this is not repaired by rerunning the calculation.

Two raw snapshot dates are checked against the saved first quote of every sampled minute. Saved event references and fixed-day five-minute paths are reported separately. This is a bounded audit, not a complete warehouse certification.

Original-file selection and missing records

Each raw security/date/clock is rescanned; comparison retains invalid first books. The daily positive-trading-amount universe is a retrospective cross-check, not a historical eligibility master or a supplier completeness certificate.

How to interpret a 0.4% flat trading cost

0.4% equals 40 bp. If it is the total all-in cost, do not add the same fees or slippage again. It is higher than the 100,000-yuan static median in this selected sample, but cannot guarantee coverage of large orders, waiting or missing fills. A 0.4% round-trip cost corresponds to 20 bp per side only under a symmetric split.

At 250 sessions with buying and selling each 60% of NAV daily, linear expenses are 120% or 60%. This sums expenditure at constant capital; it is not a realized NAV loss or a compounded return. These flat assumptions are stress scenarios, not calibrated cost estimates.

These checks cannot establish real slippage. Vendor event definitions and cross-stream exchange order are still needed for full book reconstruction; actual order and fee records are needed for calibration. Missing securities require historical status and supplier coverage records.

From static books to execution calibration

Event replay: quality gate remains closed

Early Shenzhen reference lookup is audited, but event types, cancellation meanings and cross-stream exchange order remain unverified. A full book has not been reconstructed or reconciled to snapshots.

Real-order entry is ready; actual calibration is pending

Canonical orders, fills and past available quotes feed signed decision/arrival costs, unfilled opportunity cost and actual fees. Training parents must finish before the holdout starts. The 80-parent synthetic check verifies the pipeline, not real slippage. No actual order files have been supplied to this study.

Market-adjusted terminal responses and chronological prediction errors are implemented in the research entry. They describe association; identifying causal own impact needs additional controls and reviewed execution-comparison evidence.

Turnover definitions and cost sensitivity

Illustrative assumptions: 250 sessions, constant NAV, no leverage, 1 bp inclusive commission on each side and 5 bp sell-side tax. Confirm the fee date and broker invoice before use; do not add included levies again. 1 bp = 0.01%.

For balanced fully invested rebalances without flows: one-way turnover = half of the sum of absolute weight changes. A 60% sum means about 30% buying and 30% selling.

Annual linear cost expenditure = days × [buy turnover × (buy fees + buy slippage) + sell turnover × (sell fees + sell slippage)]. Rates use decimals.

This table is arithmetic sensitivity, not observed annual return or NAV loss; trading frequency and funding are not simulated. Slippage must use a named benchmark. Midpoint-based book walking already includes half-spread and depth cost: do not add spread or the same impact twice. Unfilled orders and opportunity costs require a separate ledger.

What Level-2 can establish

Ten-level snapshots measure quoted spread, depth and static buy/sell benchmarks. Own-order impact remains uncalibrated. Report P50/P90/P95 and depth coverage. Insufficient depth means incomplete coverage, never extrapolated fills.

Orders and trades: validate event order, units, cancellation links and zero-price semantics before estimating signed flow or price response. Historical replay cannot identify the causal impact of a fictional order.

Actual order and fill reports: calibrate implementation shortfall against the decision price, including partial fills, cancellations, fees and unfilled quantity. No such calibration is published here yet.

Next experiments and publication gates

The original simulator’s 5 bp charge has been removed before applying the new fee assumptions. Next, reconcile dated fee rates and minimum commissions against actual broker statements.

Ten dates and trading-amount, market-value and volatility groups are published. Next, verify earliest source records per minute, historically eligible securities that are entirely absent, and predeclared normal and stressed sessions.

The earlier eighteen fee scenarios remain available. A separate 40-security diagnostic now reports eighteen funded minute scenarios across 2024 H2 and 2025, including fixed splitting and unfilled orders. Real orders are needed for execution calibration before estimating live capacity.

Publish only after semantic checks, sample manifests, benchmark definitions, exclusions and time-based out-of-sample validation are reviewable. Raw vendor data and order records stay outside this repository.