Intraday Flush-Absorption Panel: Buy a Down Day Whose Range Blew Out but Closed in the Upper 40% of Its Range, Fixed 3-Day Clock, on 13 Bybit USDT Perps (1D, long-only event trade, flat ~95% per name, pure OHLCV)

Hypothesis

Quota check: BYBIT is at 3.2% of experiments against a ≥5% target, while BINANCE USD-M is at 48% against ≤40%. Long-only is over-represented (75%), but L230 says only the long arm of absorption events has survived, so a two-sided version would be unjustified. I first tested a transfer of the promoted 1D/1W macro trend basket to these same 13 Bybit names. The fixed-rule portfolio Sharpe was 0.16, so I dropped it rather than propose a known-weak transfer. The 13 names are disjoint from the over-mined BTC/ETH/SOL/BNB/XRP/ADA/LINK core. Per L237/L238 the pre-registered independent-date count is 104 (≥30), and per L249 a fixed-rule after-cost forward return was measured on the exact names before submission. Risk profile (conservative): each leg is 10% notional; at a ~5% daily std that contributes ~10% annualized vol, far below the 120% limit. Gross is ≤60%, unlevered, so the liquidation buffer is not binding. The expected strategy vol is ~8-15%/yr.

Implementation

Long-only daily event panel on 13 Bybit linear USDT perps (DOT primary + 12 extras). Each leg is evaluated independently at each 00:00 UTC daily close: event when the close is below the prior close, the high-low range / prior close exceeds range_mult x the std of the prior vol_lookback daily returns (excluding the current bar), and the close location (C-L)/(H-L) exceeds close_loc. Flat, eligible legs (listed >= min_listed_bars days, point-in-time via universe_at/listed_since) are bought at market with notional = leg_fraction x account equity, 1x, up to max_concurrent open positions panel-wide, ranked by largest range/sd. Each position is sold at market at the close of the hold_days-th daily bar after entry (from position.ts_opened). No stop, no target, no pyramiding; events on a held leg are ignored. Execution substitution (next_bar_open_fill = use_equivalent): the market order fills at the signal bar's close, which on a 24/7 venue is the next bar's open.

Implements the hypothesis exactly: all three event conditions, sd excludes bar t, loc undefined when H=L, extra-leg bars must be stamped at the primary's decision timestamp and be consecutive days (no stale/future leg data), concurrency cap counts open positions and in-flight orders, ranking by range_t/sd_t. Risk route (b): time exit plus capped sizing (10% per leg, <=60% gross, unlevered); the leverage term in sizing is 1.0. calculate_signal returns the primary's continuous range/sd ratio every bar; panel entries/exits for extra legs are submitted from calculate_signal via _submit_entry_instrument/_submit_exit_position, the primary's through should_enter/should_exit. All 13 1-DAY Bybit bar dirs exist in the catalog. Parameter names and fixed/tunable split follow the optimization plan; no clamps, so no _param_bounds.

Verification Results

Optional: decrement occupied when an exit is submitted this bar, so a slot frees on the exit bar.

A leg exiting on its hold_days-th bar still counts as 'occupied' on that bar. Its exit order is submitted in calculate_signal; for the primary, should_exit runs afterwards. So a freed slot only opens on the next daily bar. The hypothesis does not say whether a slot frees on the exit bar, so this is a conservative reading rather than a mismatch. Same-bar re-entry on the exiting leg is impossible either way.

Acceptable as is. A fully leg-independent panel would need on_extra_bar handling, which is not required here.

min_bars_required() is overridden to max(70, vol_lookback+2) on the PRIMARY (DOT) bar buffer. The whole panel, including extra-leg entries and exits, is processed only from the primary's decision step. So until DOT has 70 bars, and on any day a DOT bar is missing, no leg is evaluated. Extra-leg exits can also be delayed by a day if a DOT bar is missing. DOT has a long Bybit 1D history, so the practical impact is small. The override is deliberate and consistent with the warmup.

The pre-study's effective sample is about 104 independent dates, and 2 of 6 years were negative (2021, 2026 YTD), so the edge may depend on the regime. Funding over the 2-5 day hold is not in the pre-study. In the full backtest, check per-year and date-clustered results and the funding drag. Also check deployment per trade: about 10% of equity per leg means total return and drawdown should scale accordingly (L211).

Backtest Review

Sharpe

0.82

Total return

43.07%

Max drawdown

10.21%

Trades

200

Win rate

58.0%

Profit factor

1.72

Development-period economics warrant optimization: 160 closed trades, 2.11461% mean net trade return, profit factor 1.855 and portfolio return +37.37%. Ledger P&L reconciles to approximately $37,370.

All trades are long and held exactly three days, across 74 entry dates. Entry notionals of approximately $10,000–$14,428 are consistent with capital-relative 10% sizing.

Code implements the stated absorption conditions and excludes the signal bar from trailing volatility. Sharpe 0.867 and annualized volatility 8.98% are plausible.

QA’s regime concern remains relevant: 2025 through September 13 returned -4.53%, after positive 2021–2024 returns. Clustered entries reduce independence relative to the headline trade count.

Funding cost $1,752.28; net profitability survives this drag, although the supplied audit does not independently verify funding coverage.

The pre-study and development backtest differ in period and portfolio constraints, so their event counts are not directly comparable. Forward robustness remains untested; no conclusion is drawn about the held-back period.

Analysis

Final parameters improve full-history Sharpe from 0.819 to 1.056, return from 43.07% to 71.16%, and trade count from 200 to 269. Profit factor is 1.829 and mean net trade return is 2.114%, comfortably above the supplied 0.22% economics floor.

Pooled chronological OOS Sharpe is 1.126, corrected HAC t=2.252 and DSR=0.9878, clearing the registered significance hurdle. Five windows each searched one tunable parameter with 30 trials plus neighbourhood evaluations.

The final close_loc=0.551603 is classified stable. All contract-fixed parameters remain unchanged.

Measured worst realized trade loss is 1.924% of entry equity; largest position is 10.3687%. The capped-sizing/time-exit route reports no violations. Drawdown of 9.9486% remains below the declared 20% limit.

Low confidence: not statistically proven across both protected scopes. The 409-day, 55-trade holdout has Sharpe -0.0285 and HAC t=-0.0298. It is inconclusive, not statistically adverse. Consistency z=-0.792 does not establish deterioration, but provides no affirmative forward confirmation.

Two of five outer windows are negative, including the latest. Inner PBO ranges approximately 0.32–0.73, warning of selection fragility.

Sensitivity deteriorates above close_loc=0.60; the final value lies below that cliff. These report-only findings remain robustness warnings.

The forced-selling explanation is not independently verified; signal IC t=1.09 is weak. Survivorship remains in the surviving-contract panel.

Measured full-history results exceed research predictions, and pooled OOS Sharpe exceeds the predicted 0.7. These are different scopes from the preliminary event study, which omitted funding; predictions are context, not gates.

Backtest-level deflated Sharpe remains unknown. Final ledger risk measurements rely on supplied risk-control evidence; the available local trade detail contains the baseline.

Risk Review

The signal-exit route holds: three-day exits, equity-scaled 10% leg allocations and six concurrent positions. Measured maximum position is 10.0211% of equity, worst realized trade loss 1.5378%, and worst adverse excursion 3.1201%. A stop-loss is not required. Realized volatility of 10.3573% and peak position volatility of 15.778% fit the conservative profile. Leverage is 1x; no liquidation or risk-control violations are reported. The null liquidation-buffer measurement is not an independently measured pass.

Evidence supports labelled low-confidence paper observation only. The visible backtest ends on 2025-09-13; protected holdout measurements and the final full-history ledger are unavailable for independent inspection. The 409-day holdout is inconclusive, and final parameters are not statistically accepted. Pooled chronological OOS HAC t=2.2516 does not establish holdout success. Backtest-level deflated Sharpe remains unevaluated.

Development drawdown is 9.3365% against the registered 20% ceiling, with recovery lasting up to 248 days. August 2024 stress drawdown reached 6.7785%. The latest chronological test window returned -1.5682% with Sharpe -0.1526; two of five walk-forward windows were negative. Stable parameter choices do not eliminate regime dependence or surviving-contract bias.

Book correlation and deployable capacity are unmeasured. Existing absorption, capitulation and reversal panels can share altcoin exposure. No members are recorded in breach and no execution items are open, but collecting fidelity states, divergent replays and parity warnings leave execution confidence incomplete.

All leg exits depend on primary DOT daily decisions; missing bars can extend the holding period. Funding is folded into returns, but supplied funding audits mark coverage_verified=false, leaving completeness unverified.

Reviewed parameters are reasonable and sizing scales with equity. Development produced 214 closed trades, 61.21% win rate, profit factor 2.2778 and mean net notional return 2.6621%; clustered entry dates reduce independence. The in-sample Sharpe advisory is non-blocking and needs no waiver. No soft floors are in play or hard failures reported.

Admit only to paper with the persistent low-confidence label. This review does not authorize real-money trading or allocation increases.

Keep reviewed 1x leverage, 0.1 leg_fraction, six-position cap and three-day hold. Before shared-account allocation, enforce the 40% correlated-exposure account limit across overlapping strategies; this panel can target approximately 60% gross exposure.

Before increasing allocation or seeking live release, establish strategy-specific decision and fill parity, quantify book correlation and synchronized stress exposure, and validate liquidity and costs at the proposed notional.

Monitor delayed daily bars and overdue exits. Strategy, portfolio and factory circuit breakers apply automatically.

Outcome Summary

Positive pooled OOS evidence should be weighed alongside an inconclusive holdout, parameter sensitivity and selection fragility when evaluating this absorption rule.

Backtest review recommended optimization, and the analyst issued a promote verdict with no hard failures after the pooled OOS significance and economics criteria passed. The strategy reached paper_stage with low confidence because the protected holdout provided no affirmative forward confirmation.

Buy unusually wide-range down days that close in the upper part of their range across 13 Bybit USDT perpetuals, seeking a rebound after absorbed selling with a fixed three-day exit.

The baseline returned 43.07% with Sharpe 0.819 across 200 trades; the analyst reported final-parameter results of 71.16% return, Sharpe 1.056 and 269 trades, with 9.95% maximum drawdown. Pooled chronological OOS Sharpe was 1.126 with HAC t=2.252, while the 409-day, 55-trade holdout was inconclusive with Sharpe -0.0285.