SolHyperliquidDailyTSMomentumLongShort

Hyperliquid SOL Daily Time-Series Momentum, LONG/SHORT (Single-Instrument HYPERLIQUID Perp, Pure OHLCV, Flip-Via-Flat, Fixed-Fractional Sizing, SOL-Validated TS-Momentum Mechanism on the One Confirmed Clean-Sizing Venue — Fills the Dominant Long-Only Over-Representation AND the Under-Weight HYPERLIQUID Venue, Pre-Registered LS-Sizing Probe, Tail-Safe, Low-Parameter)

A LONG-SHORT, SINGLE-INSTRUMENT, DAILY time-series (absolute) momentum strategy on SOLUSD.HYPERLIQUID. This simultaneously fills the TWO largest quota gaps: (1) DIRECTION — long_only is 85.8% vs the <=55% target (30.8pp over, the single biggest miss); the factory needs long_short, and HYPERLIQUID is the ONLY venue where long_short sizes correctly (BINANCE USD-M perp shorts/flips are engine-corrupted by the proven over-exposure defect; spot-CASH MTM, multi-instrument aggregation, and options settlement are all also defective; census signals liq/OI/taker/account-ratio are data-walled, so signals are restricted to OHLCV + funding). (2) VENUE — HYPERLIQUID is 6.5% vs the >=20% target (under-weight), so adding here is doubly correct. MECHANISM is asset-specifically VALIDATED: the time-series-momentum long-short family reached paper_stage on the SOL sibling (SolTimeSeriesMomentumLongShortSingleAsset1D), but has NEVER been run on the clean HL venue. This honors the recent ASSET-SELECTIVITY lesson (HL daily breakout worked on BTC PF 1.76 but FAILED on ETH PF 0.70 -> validate per instrument): SOL is a historically strong, high-amplitude trender, and the TS-momentum mechanism is precisely the one that promoted on SOL elsewhere. It also diversifies HL beyond BTC (HL currently: BTC Donchian optimizing + BTC TS-mom LS pending + SOL Donchian long/flat pending) by adding the SOL two-sided momentum book. Construction is deliberately simple and bug-avoiding: pure OHLCV, ONE fixed-fractional size per entry (NO vol-target re-scaling, NO leverage stacking), leverage 1.0, and flips routed THROUGH FLAT (on a sign change, exit fully to flat on one bar, then open the opposite side next bar) rather than direct reversal — so the engine never nets opposing fills, the cleanest long-short construction. PRE-REGISTERED ACCEPTANCE / SIZING PROBE (doubles as the HL long-short sizing test, since clean-HL evidence so far is all long/flat): avg_position_pct ~<=100% and no physically impossible single-day equity return (SOL at 1x is ~+/-25-30%). If violated, the over-exposure defect reaches the HL short/flip path -> abandon and escalate as the HL-LS reproducer. If clean, this + the HL BTC TS-mom LS establish a correctly-sized two-sided trend sleeve on HL. Low parameter count (one lookback, one trend filter, one ATR stop).

Implements the hypothesis as a two-sided trend sleeve on the only venue confirmed to size long/short cleanly (HYPERLIQUID), simultaneously addressing the long_only over-representation and the HYPERLIQUID under-weight quota gaps. Mechanism is the asset-validated TS-momentum family that promoted on the SOL sibling, honoring the asset-selectivity lesson by keeping it on SOL (a strong high-amplitude trender). Construction is deliberately defect-avoiding: (1) single-instrument, pure OHLCV (no funding subscription, no extra legs) so multi-instrument aggregation and funding-exclusion defects cannot apply; (2) flip-via-flat is achieved structurally — should_exit closes to flat on a reversal and the base engine only evaluates an entry when position is None, so the opposite side opens on a later bar and opposing fills are never netted; (3) ONE fixed-fractional size (notional = 0.95 * equity, qty = notional/price) with no vol-target rescaling or leverage stacking, and leverage left at 1.0 so the engine margin model and in-strategy notional agree. The vol-normalized momentum score keeps the signal continuous and conviction-bearing and makes a single entry_threshold behave consistently across volatility regimes. Daily bars match the long-term momentum timeframe and keep HL fee drag (~0.09% RT) negligible given few trades/year. Pre-registered sizing probe: avg_position_pct should be <=~100% and single-day equity returns bounded near SOL's daily move at 1x (~+/-25-30%); a violation would indicate the over-exposure defect reaches the HL short/flip path and should be abandoned/escalated as the HL-LS reproducer.

Failed deflated Sharpe: DSR=0.0, expected-max=7.51 over 225 trials — the optimized Sharpe (1.48) is far below the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO=0.6151>0.5, deflated_sharpe=0.0, sharpe_ci_low -0.615 straddles 0). Every decisive generalization gate fails in concert even though SENSITIVITY PASSED (0 cliffs): the walk-forward out-of-sample Sharpes are [0.0, 0.0, -1.50] (avg -0.499, NEGATIVE) against in-sample [8.14, 4.40, 4.26] with is_overfitted=TRUE — the edge does not merely vanish OOS, it loses money — and the holdout FAILED with a NEGATIVE holdout_sharpe -0.68 (ratio 0, passed=false). Sensitivity passing only confirms in-sample smoothness; it does not measure generalization, and the generalization gates (OOS, holdout, DSR) all fail. The attractive-looking headline is a single fat-tail trade: the initial +109% return rests on the 2024-01-21 +262.9% parabolic-SOL trade (skew 12.87, kurtosis 183.36, tail_ratio 0.0132 — upside is ~1% of downside by magnitude, profit_factor only 1.157), and the risk profile is severe (max_drawdown 61%, 764-day drawdown duration, up to 8 consecutive losses). Root cause (symptom, not independently verified): the SOL-HL momentum edge is concentrated in the 2023-2024 bull (annual 2024 +232%) and decays/reverses thereafter (2025 -7.8%, 2026 -29.8%), on only ~3 years of Hyperliquid history with 37 trades — the OOS windows are empty/edgeless-to-negative and the recent holdout is negative, so the sample is too short and too regime-concentrated to survive forward-testing or multiple-testing deflation. Not iterate (attempt 1 of 2): the failure is negative OOS + negative holdout + deflation, not a tunable parameter region — sensitivity already passed, so a second sweep would re-find the same IS-positive / OOS-negative pattern on the same thin data; there is no robust region to tune toward. Not revise_hypothesis: the TS-momentum long/short mechanism is ALREADY validated and captured by the promoted SOL sibling (SolTimeSeriesMomentumLongShortSingleAsset1D reached paper_stage on another venue), so its value is not lost; this HL instance fails purely because HL's ~3-year history is too short to validate a low-frequency daily-trend edge — an engine/data limitation, not a premise a different instrument reframes. This is the FOURTH HL daily-trend instance this session (BTC-HL Donchian, ETH-HL Donchian, SOL-HL Donchian, and now SOL-HL TS-momentum) to die on the identical signature: all/negative walk-forward OOS, a negative holdout, DSR 0.0, PBO >0.5, fat-tail single-trade headline. FAILURE PATTERN: single-instrument daily trend/momentum strategies on Hyperliquid majors cannot be validated on HL's ~3-year history — even the historically strongest trender (SOL) with a sibling-promoted mechanism collapses to all/negative walk-forward OOS, a negative holdout, DSR 0.0 and PBO >0.5 because the edge concentrates in the 2023-2024 bull and a fat-tail single-trade headline defeats forward-testing and multiple-testing deflation. PRESERVED FINDING (the genuinely useful result): HL sizes the LONG/SHORT flip-via-flat path cleanly — avg_position_pct 97.3%/67.6%, metrics_reliable=true, 0 dropped entries, balanced long+short legs, no impossible engine-driven equity jumps — so the BINANCE USD-M over-exposure defect does NOT reach the HYPERLIQUID short/flip path; the pre-registered HL long-short sizing probe PASSES. Only the trend edge is unvalidatable on this venue's short data, not the sizing.

Long/short daily time-series (absolute) momentum strategy on SOLUSD.HYPERLIQUID, pure OHLCV. The signal is the lookback (30-day) return vol-normalized into a Sharpe-like t-stat score: goes long when score > +0.5 and close > 50-day SMA, short when score < -0.5 and close < 50-day SMA. Exits on momentum reversal (score crossing the exit band or close crossing the SMA against the position), a 3x-ATR hard stop frozen at entry, or a 15% percentage stop. Flips are routed through flat (exit fully, then re-enter the opposite side on a later flat bar) so the engine never nets opposing fills. One fixed-fractional position (95% of equity notional) per entry, no vol-target rescaling, leverage 1.0.

Pre-registered LONG-SHORT sizing probe PASSED: avg_position_pct 97.26% (<=100%), metrics_reliable=true, end_unrealized_pct 14.6% — the HL short/flip path sizes correctly (distinct from the long/flat siblings, which never exercise shorts). No over-exposure on the two-sided book.

The +262.9% single-period return is a legitimately captured SOL Q4-2023 trend leg, not an over-exposure artifact (confirmed by clean sizing).

Validated TS-momentum LS mechanism (promoted on the SOL sibling elsewhere); sound bug-avoiding construction (flip-via-flat, one fixed-fractional entry, leverage 1.0).

SINGLE-TRADE DEPENDENCY: the entire return is one Q4-2023->Jan-2024 trade (+262.9%); 2023/2025/2026 are all negative (-9.3%/-7.8%/-29.8%). Strip that trade and profit_factor falls below 1 (net loser).

The LS/short addition DRAGS rather than helps: profit_factor 1.16 here vs 2.0 on the long/flat SOL Donchian sibling; shorts (19 trades, 31.6% win) bleed in 2025-2026 downtrends.

Deep max_drawdown 61.3%; Sharpe 0.87 with CI -2.17 to +1.87 straddling 0; return_skew 12.9 / kurtosis 183 / tail_ratio 0.013 — one fat-tail outlier dominates.

Redundant with the SOL Donchian long/flat sibling already in optimization — both SOL daily-trend books are carried by the same Q4-2023 trade; optimizing both double-spends compute on one event.

37 trades dominated by one winner is too thin/concentrated for a 3-window walk-forward + 225-trial deflated Sharpe to clear.

Pre-registered HL long-short sizing probe PASSED: avg_position_pct 97.3% (initial) / 67.6% (optimized), metrics_reliable=true, leverage 1.0 honored, 0 dropped/size-zero entries across 37/47 trades with both long (18/22) and short (19/25) legs — the BINANCE USD-M over-exposure defect does NOT reach the HYPERLIQUID short/flip path. The flip-via-flat construction sizes cleanly two-sided.

Mechanism is sound and simple: pure-OHLCV single-instrument TS-momentum, flips routed through flat (no netted opposing fills), fixed-fractional single-entry, no vol-target restacking. Trades implement the stated long/short momentum mechanism.

Sensitivity surface is smooth (0 cliffs across all 9 parameters).

Walk-forward out-of-sample Sharpe is [0.0, 0.0, -1.50] (avg -0.499, NEGATIVE) vs in-sample [8.14, 4.40, 4.26]; is_overfitted=TRUE. The edge does not merely vanish OOS — it loses money.

Holdout FAILED: holdout_sharpe -0.68, ratio 0 — the once-tested recent window is negative.

Failed deflated Sharpe: DSR=0.0, is_significant=FALSE, PBO=0.6151 (>0.5). Optimized Sharpe 1.48 is far below the expected-max-under-noise bar of 7.51 over 225 trials — indistinguishable from best-of-N selection luck. sharpe_ci_low -0.615 straddles 0.

Headline is a single fat-tail trade: initial +109% return rests on the 2024-01-21 +262.9% parabolic-SOL trade (skew 12.87, kurtosis 183.36, tail_ratio 0.0132 — gains are ~1% of losses by magnitude). Profit_factor only 1.157; annual returns 2024 +232% but 2023 -9.3%, 2025 -7.8%, 2026 -29.8%.

Severe risk profile: max_drawdown 61% (initial) / 38% (optimized) with a 764-day drawdown duration and 5-8 max consecutive losses; CVaR_95 12.6%.

Thin/regime-concentrated sample: 37 trades (initial) / 47 (optimized) over HL's ~3-year history — too short for the daily-trend edge to forward-test.

SolHyperliquidDailyTSMomentumLongShort aimed to close the factory's two biggest quota gaps at once by porting a SOL-promoted TS-momentum long-short mechanism to Hyperliquid, while serving as the dedicated HL long-short sizing probe. That probe succeeded — balanced long and short legs sized cleanly at avg_position_pct 97.3%/67.6% with no over-exposure — but the trading edge did not: the +109.5% headline was one parabolic 2024 SOL trade, profit factor was just 1.16 (the short side dragging versus the 2.0 long/flat sibling), and 2023/2025/2026 were all negative. Optimization improved the in-sample numbers and passed sensitivity, yet the generalization gates collapsed harder than its siblings — walk-forward OOS averaged negative (-0.499), the holdout was -0.68, DSR 0.0, PBO 0.62. The analyst abandoned it as the fourth identical HL daily-trend failure, noting the TS-momentum mechanism is already validated and captured by a promoted SOL sibling elsewhere, so only the HL expression fails on short data — while the clean two-sided sizing on Hyperliquid is the preserved, useful result.

Adding a short side does not rescue an HL daily-trend edge unvalidatable on ~3 years of history — this is the fourth HL daily-trend instance to die on the same all-zero/negative-OOS, negative-holdout, DSR-0 signature, and the shorts actively bled in 2025–2026 downtrends; the genuinely durable finding is that HL sizes the long/short flip-via-flat path cleanly.

Despite passing sensitivity and the sizing probe, the analyst abandoned it (iteration 1 of 2) because every generalization gate failed: deflated Sharpe 0.0 vs an expected-max-luck bar of 7.51, PBO 0.62, walk-forward OOS of [0.0, 0.0, -1.50] (avg -0.499, negative) against in-sample [8.14, 4.40, 4.26] with is_overfitted=TRUE, and a negative holdout (-0.68); the short side also dragged (PF 1.16 vs 2.0 on the long/flat SOL sibling).

A long/short, single-instrument, OHLCV-only daily time-series momentum strategy on SOLUSD.HYPERLIQUID (flip-via-flat, fixed-fractional entries, leverage 1.0) intended to fill the two largest quota gaps — long_short direction and the HL venue — by running a sibling-promoted TS-momentum mechanism on the one venue where shorts size cleanly, while doubling as the pre-registered HL long-short sizing probe.

The default backtest returned +109.5% but with weak quality — Sharpe 0.87, profit factor 1.16, 29.7% win rate over 37 trades (18 long, 19 short), 61.3% max drawdown — and the entire return rested on one Q4-2023→Jan-2024 trade (+262.9%, skew 12.87, kurtosis 183, tail_ratio 0.013); optimization raised it to Sharpe 1.48, PF 1.53, +245% with sensitivity passing (0 cliffs), and the long-short sizing probe passed throughout (avg_position_pct 97.3%/67.6%, metrics_reliable=true).