Strategies

Hyperliquid Weekend Pump Reversal, BTC-Hedged: each Monday 00:00 UTC short a Hyperliquid alt perp whose Fri-close-to-Sun-close log return beat BTC by >=1.5 sigma, long equal-dollar BTCUSD.HYPERLIQUID, fixed 2-day clock (24-name pre-registered HL panel, 1-DAY, market-neutral, flat most of the week)

Outcome: Abandoned

HlWeekendPumpReversalBtcHedged

Outcome Summary

The Research Lead proposed a BTC-hedged short on weekend alt-pump reversals across a 24-name Hyperliquid panel, replicating a promoted Bybit strategy. The pre-study showed +0.76% net per pair-trade, but most of those events came before these Hyperliquid perps were listed on 2023-05-12. On the tradeable sample of 123 positions over 40 event weeks, the strategy was flat to negative: profit factor 0.998, Sharpe 0.03, and pumped alts slightly outperforming BTC instead of reverting. The analyst abandoned it at Backtest Review for negative expectancy after one iteration, so optimization, analyst review and risk review never ran. The verdict applies only to this Hyperliquid version, not to the promoted Bybit one.

Strategy report

Hypothesis

Original Research Lead hypothesis (not seeded). An independent-venue replication of the promoted Bybit 'Thin-Liquidity Weekend Pump Reversal, BTC-Hedged' (198 trades, PF 1.23) on Hyperliquid, which is below its venue quota (14% vs >=20%) and whose only HL promotions so far are long-side down-shock panels; this adds the hedged SHORT side that L256 says is the only short-side shock design that has survived. MECHANISM: on Saturday/Sunday the institutional and market-making liquidity that arbitrages alt-vs-BTC mispricing is thinner, so retail-driven or narrative-driven weekend pumps in individual alts overshoot their BTC-explained move; when weekday liquidity returns (Monday-Tuesday) the idiosyncratic excess partially mean-reverts. The trigger is defined RELATIVE TO BTC and the trade carries a BTC hedge leg, as L256 requires (no raw-return trigger, no unhedged short). PRE-STUDY (frozen rule, measured by the Research Lead on catalog 1-DAY bars, 17 liquid HL names listed before 2021-06, BTC hedge, 2020-08..2026-10): x = ln(alt Mon00/Sat00) - ln(BTC Mon00/Sat00); sigma = std of that name's previous 26 weekend x values (min 20); z = x/sigma. Events with z >= 1.5: 318 pair-trades on 156 distinct Mondays; mean 2-day hedged forward return (short alt + long BTC, each leg equal dollars, measured as a fraction of the alt-leg notional) +0.94% gross, +0.76% net of 2 x 0.09% HL round trips, win rate 61%; by year mean 2.8% (2021), 0.0% (2022), 1.3% (2023), 0.6% (2024), 1.0% (2025), 0.4% (2026 YTD) - positive in 5 of 6 years. z>=1.0 gives 569 events at +0.52% net, z>=2.0 gives 208 at +0.91% net, so the effect is monotone in the threshold rather than a single cliff. The long mirror (z<=-1.5) was weaker (+0.55% net, 93 dates) and is NOT part of this hypothesis. Funding was not in the pre-study: the short alt leg during pumps usually RECEIVES positive funding and the BTC long pays it, so the net funding effect should be small; the backtest charges it via the funding ledger. EXECUTION: the decision is made on the close of the Sunday daily bar (stamped Monday 00:00 UTC); on this 24/7 venue a market order at that close is the next bar's open (the registry's use_equivalent for next_bar_open_fill - substitution named here). Data-coverage check per L237/L259: all 24 legs plus BTC have HL 1-DAY bars in the catalog from 2020-08..2021-01 (the 7 extra candidates are FTM, MATIC, MKR, SNX, COMP, RUNE, ATOM); funding is the only supplementary feed and is cost-only (not a signal), so a funding gap cannot remove events. Pre-registered event count inside the overlap: >=150 distinct event Mondays and >=300 alt-leg trades (measured 156 / 318 on the 17-name subset).

Quota: HYPERLIQUID 14% vs >=20% target, long_short 25% vs long_only cap 55%, and it is a market-neutral event panel (the family with the best survival, event_driven 14%). Mechanism diversity: the only hedged short-side shock design that survived (L256), not yet tested on Hyperliquid, whose HL survivors are all long-side down-shock panels. L250/L249/L231 are respected: this is an event-triggered, flat-most-of-the-week panel, not an indicator-state or continuously rebalanced design, and the frozen-rule after-cost forward return on the exact names was measured before submission (+0.76% net per pair, 156 distinct event dates). L237/L259: all legs have daily bars over the same 2020-08..2026-10 span; pre-registered event count >=150 distinct Mondays. Not BTC-concentrated as a signal: BTC is only the hedge leg.

negative_expectancy: this rejects the Research Lead's Hyperliquid operationalization of the weekend pump reversal; it is not a verdict on the Bybit version that was promoted. On the tradeable development sample (2023-12-11 to the 2025-08-04 cut; 123 closed positions; 83 alt shorts over 40 event weeks), profit factor is 0.998, Sharpe 0.03, total return -0.08% and expectancy -$0.67 per position. The average trade return of 0.126% of notional is below the 0.15% Hyperliquid floor. On price alone the hedged pair lost about $215 before costs: alt shorts -$780, a mean of -0.024% per leg, and the BTC hedge +$565. Funding nearly cancelled across the legs (+$1,940 on the shorts, -$1,807 on BTC). The pre-study's edge came largely from 2020-2023 bars that predate the 2023-05-12 Hyperliquid listing and could not have been traded on this venue. Nothing is concluded about the period after the cut.

Implementation

Weekly event panel on 24 Hyperliquid alt perps (point-in-time universe), with BTCUSD.HYPERLIQUID as both the clock and the dollar hedge. On the daily bar stamped Monday 00:00 UTC (the Sunday close), each alt's Fri-close-to-Sun-close log return in excess of BTC (x) is divided by the sample std of its previous 26 weekend x values (at least 20 required, current week excluded). An alt is eligible when it is listed, has at least 210 prior daily bars and a 30-day median close*volume of at least $1M. Eligible alts with z >= 1.5 are shorted at 10% of equity each (at most 6, largest z first). BTC is bought for the summed alt notional (hedge ratio 1.0, one netted BTC position). Every leg and the hedge close at the close of the bar hold_days (2) days later (Wednesday 00:00 UTC). Each alt short carries an intrabar reduce-only stop-market at entry x 1.25; after a stop-out, that leg's BTC hedge is removed at the next bar close. legs_balanced() is checked every bar and flattens everything if a leg is missing. Execution substitution (next_bar_open_fill = use_equivalent): orders are market orders at the signal bar's close, which on this 24/7 venue is the next bar's open. The continuous signal on every bar is the panel-max current 2-day BTC-excess z; entries fire only on the Monday bar.

Implements the hypothesis rules as written: x is the weekend excess over BTC and sigma comes from the name's own previous weekend x values. Alts with z >= z_entry are shorted against an equal-dollar BTC long, on a fixed 2-bar clock, with a 25% catastrophic stop. Eligibility is point-in-time (is_listed, 210 bars, $1M 30-day median dollar volume), with at most 6 legs. The panel is the 17 pre-study names plus the 7 named extras (FTM, MATIC, MKR, SNX, COMP, RUNE, ATOM), 24 alts in all. All have HL 1-DAY bars in the catalog from 2020-21. FTM, MATIC and MKR are delisted; the base template and point_in_time_universe handle that. Leverage 2.0 is only margin headroom: max gross is 6x10% short + 60% BTC = 120% of equity, which does not fit a 1x margin cap. config.leverage is read in the margin-room check, and leg size stays a fixed 10% of equity (net exposure ~0). The protective stop is a raw reduce-only STOP_MARKET placed on the alt fill, because submit_stop_order() only supports the primary instrument. It triggers intrabar in the simulated venue (intrabar_stop = supported). Per-position state (active flag, entry ts, hedge qty) is kept as plain scalars; rolling statistics are rebuilt from warm-up bars.

Verification Results

No code change needed. The research contract's instrument list should be reconciled to the 24-name panel for the record.

research_contract.instruments lists 20 instruments (BTC + 19 alts), but the config trades 24 alts: it adds FTMUSD, MATICUSD, MKRUSD, COMPUSD and RUNEUSD. The hypothesis description and entry_logic explicitly name this 24-name panel (the 17 pre-study names plus FTM, MATIC, MKR, SNX, COMP, RUNE, ATOM, with delisted FTM/MATIC included while listed), so the code follows the written hypothesis. The contract list is the inconsistent part.

Optionally rebuild the active flags and hedge_qty from open positions on start (paper only).

Per-position state (active_*, entry_ts_*, hedge_qty_*) is plain attributes, and the rolling weekend-x history is rebuilt only from warm-up bars. After a paper restart mid-hold, the hedge and leg bookkeeping depend on that restoration. This does not affect the backtest.

Optional: on failure, drop only the failed leg's hedge_qty (step 1 already does this when the leg is flat) instead of flattening the whole book.

legs_balanced() is applied to the whole active book. One refused or shrunk alt entry (a leg_failure since the book opened) flattens every leg of that week, not just the orphan and its hedge share. That is stricter than the hypothesis ('close the orphan leg'), but it never carries an unhedged position. In a simulated venue a market order fills in full, so this should be rare.

Pre-study net edge is +0.76% per pair (gross +0.94%) against about 0.18% round-trip cost for the two legs. Per leg-trade that is about 0.38%, so it is fee-robust on paper. The 1:1 dollar hedge leaves residual alt beta above 1, so a market-wide Mon-Wed rally is the main risk. Weekly events cluster by narrative, so the book is correlated. The 2022 flat year and the thin 2026 YTD (+0.4%) deserve a look in the holdout. Funding is charged by the ledger and was not in the pre-study.

Backtest Review

Sharpe: -0.02
Total return: -2.41%
Max drawdown: 15.08%
Trades: 230
Win rate: 51.3%
Profit factor: 0.97

The code does what the hypothesis says. Every entry is on a bar stamped Monday 00:00 UTC, each one shorts alts with a BTC long, and all legs close after 2 days. The ledger has 83 alt shorts and 40 BTC hedge longs over 40 event weeks.

The point-in-time universe works. The listings registry dates every Hyperliquid perp in this panel from the first funding print on 2023-05-12, so the strategy correctly made no trades between 2021-09 and 2023-11. The first trade is on 2023-12-11, once 210 bars of history had built up after listing.

Funding is credited on both legs. The alt shorts received +$1,940 and the BTC long paid -$1,807.

There is no edge before or after costs. Profit factor is 0.998, Sharpe 0.03, total return -0.08% and expectancy -$0.67 per closed position across 123 positions. The average trade return of 0.126% of notional is below the 0.15% Hyperliquid fee floor, and that average is flattered by the funding received on the short legs.

The reversal itself does not appear in prices. Price-only P&L on the alt shorts is -$780, a mean of -0.024% per leg with 41 of 83 legs profitable. The BTC hedge made +$565 on price. The hedged pair therefore lost about -$215 before costs: on average the pumped alts kept slightly outperforming BTC over the next 2 days instead of reverting. The pre-study measured +0.94% gross per pair, but the tradeable sample shows about 0.

Week by week the result is a coin flip: 21 of 40 event weeks were positive and the mean was -$2 per week. By year: 2023 (December only) -$3,844, 2024 +$389, 2025 to the cut +$3,373. The total rests on a few weeks; 2024-03 alone was +8.9%.

The pre-study's 156 event Mondays and 318 pair trades came mostly from 2020-2023 bars, before any of these Hyperliquid perps existed (listed 2023-05-12). Those events could not have been traded on this venue. The tradeable development span from 2023-12 to the cut has only 40 event weeks, far below the pre-registered 150 or more.

Outcome Summary

Check that a pre-study's events fall within the venue's actual listing dates. Most of the +0.94% gross pre-study edge came from 2020-2023 bars that predate the 2023-05-12 Hyperliquid listing, which left only 40 tradeable event weeks against the 150+ pre-registered.

The analyst abandoned it at Backtest Review for negative expectancy, before optimization. Measured on price alone, the hedged pair lost about $215 before costs (alt shorts -$780, BTC hedge +$565): the pumped alts did not revert. Funding on the two legs nearly cancelled out (+$1,940 on the shorts, -$1,807 on BTC).

Each Monday 00:00 UTC, short Hyperliquid alt perps whose Friday-to-Sunday-close return beat BTC by at least 1.5 sigma, hedge with an equal-dollar BTC long, and close after 2 days. The bet was that weekend alt pumps on thin liquidity reverse once weekday liquidity returns. It was a Hyperliquid replication of a Bybit strategy that had been promoted.

On the tradeable development sample (2023-12-11 to the 2025-08-04 cut: 123 closed positions, 83 alt shorts over 40 event weeks), profit factor was 0.998, Sharpe 0.03, total return -0.08% and expectancy -$0.67 per position. The average trade return was 0.126% of notional, below the 0.15% Hyperliquid floor. The full-window backtest report shows 230 trades, total return -2.41%, Sharpe -0.02, profit factor 0.969 and max drawdown 15.1%.

Backtest and paper results are hypothetical. Trading involves risk of loss.