HlSolTimeSeriesMomentumVolTargetedLS1D

Hyperliquid SOL Daily Single-Asset Time-Series Momentum Long-Short (Cross-Venue Validation, Moskowitz-Ooi-Pedersen 2012)

Cross-venue validation of the single-asset time-series momentum (TSM) mechanism on SOL-USD perpetual at HYPERLIQUID using daily bars. This is the explicit 4th-asset / 1st-venue-extension of the proven mechanism family: BtcTimeSeriesMomentumLongShortSingleAsset1D, EthTimeSeriesMomentumLongShortSingleAsset1D, and SolTimeSeriesMomentumLongShortSingleAsset1D (all in pipeline, with the SOL Binance version at paper_stage Sharpe 2.18). The hypothesis: if the Moskowitz-Ooi-Pedersen 2012 TSM mechanism captures a real cross-sectional risk premium on SOL on Binance, that premium should persist on the same asset traded at a different venue with a different participant mix (HL's retail-momentum-chaser dominant flow vs Binance's institutional MM-heavy flow). If HL SOL TSM also produces a positive walk-forward Sharpe with avg OOS > 0.5, the venue-independence claim is validated and the mechanism class can confidently be extended to additional HL assets (BTC, ETH on HL). If HL SOL TSM fails despite Binance SOL TSM succeeding, that's a meaningful signal that the SOL TSM edge is Binance-microstructure-specific (e.g., comes from a participant subset that doesn't trade HL). Either outcome is informative — there is no 'fallback mode' that would muddy the test. Mechanism: pure 2-parameter directional TSM. Daily timeframe ensures sufficient walk-forward trade frequency (~30-40/year expected based on Binance SOL benchmark = ~10-13 per 12-month OOS window — well above the 5-trade floor). OHLCV-only with NO supplementary data dependency means no infrastructure gap can block evaluation. Volatility-targeted position sizing eliminates capital-fraction discretion.

Targeted fix for the iteration-1 optimization failure: the optimizer sweeps every numeric key present in config.parameters, and the prior config exposed 8 keys, which inflated the Deflated-Sharpe expected-max luck bar to 6.09 and collapsed DSR to 0 -- a multiple-testing artifact, since the walk-forward OOS was genuinely all-positive [7.56, 2.76, 2.08]. SMALLEST CHANGE applied exactly per analyst guidance: (1) config.parameters now contains ONLY the three documented core knobs -- lookback, entry_threshold, exit_threshold -- so the search is a 3-trial space; (2) stop_loss_pct (6.0), take_profit_pct (18.0), vol_floor (0.005), max_notional_frac (1.0), and target_vol_annual (0.35) are hardcoded in __init__ as fixed constants at the proven SOL-Binance benchmark values, removing them from the search; (3) all signal, entry/exit, and sizing logic is byte-identical to the previous version (which passed all earlier verification layers and produced a not-overfit walk-forward), so no earlier layer regresses and no imports change. The three remaining tunables keep their robust-region clamps. Cutting the free-param count 8->3 lowers the deflation penalty so DSR can reflect the real OOS edge; the unchanged stop/TP/vol-target logic still runs every bar. Per the decision rule, if the 3-param re-run now clears DSR>=0.95 / is_significant with a non-negative holdout it should promote; if it still fails, the HL ~3-year sample is too short to clear significance and the strategy should be abandoned.

Failed deflated Sharpe: DSR=0.0 (vs 0.95 bar), expected-max=6.677 over 225 trials — the optimized Sharpe 1.44 is far below the best-of-N luck bar, indistinguishable from selection noise (is_significant=false, sharpe_ci_low -0.0639 straddling zero). The decisive forward gate also fails: the HOLDOUT FAILED with a NEGATIVE Sharpe -0.970 (ratio -0.289). The expected-max luck bar is so high because the strategy's cross-window Sharpe variance on the short 3-year Hyperliquid sample is enormous (trial Sharpes span -1.5 to +10.6), so the honest full-sample Sharpe cannot clear it; the encouraging-looking inputs (PBO 0.41, is_overfitted=FALSE, avg OOS 3.36) are misleading because the OOS average is propped by a single 10.61 window with another window negative (-1.12), all drawn from the same short bull-dominated HL history, and there is a sensitivity cliff on the core lookback parameter (1.45 at 24/26 -> 0.68 at 29). Recent performance has decayed (2026 -18.2%). This was an explicit cross-venue validation test: the TSM mechanism is already validated and landed on the Binance SOL sibling (paper stage, Sharpe 2.18), and the hypothesis pre-committed that an HL failure would be a meaningful signal the edge is Binance-microstructure-specific. That is the outcome — HL SOL TSM does not validate (DSR 0.0, negative holdout, CI straddling zero). Not iterate (attempt 2 of 2, at the limit): the failure is intrinsic to the short, high-variance HL sample plus a negative holdout and a core-parameter cliff, not an under-explored region; no parameter change manufactures statistical significance. Not revise_hypothesis: the mechanism is not stranded on a fixable target — it already works on Binance; the venue extension simply did not generalize. FAILURE PATTERN: extending a validated single-asset TSM mechanism to a new venue (Hyperliquid) with only ~3 years of history fails deflated Sharpe because the short, volatile sample produces extreme cross-window Sharpe variance that inflates the expected-max luck bar (6.68) far above the honest selected Sharpe (1.44), and the time-ordered holdout is negative (-0.97) — a modest real edge on a short cross-venue sample cannot clear best-of-225 deflation, the same DSR-failure-on-high-variance-short-sample signature as the abandoned XRP volume-Donchian (DSR 0.0004 vs a 4.21 luck bar). A clean non-overfit walk-forward and positive OOS average do not rescue an edge whose chronological holdout loses money and whose Sharpe CI straddles zero.

SOL-USD Hyperliquid daily single-asset Time-Series Momentum (Moskowitz-Ooi-Pedersen 2012), long/short, with volatility-targeted sizing. Cross-venue validation of the proven Binance SOL TSM mechanism. Score = trailing lookback-day return / realized daily vol; signal = tanh(score/2) in [-1,1]. Enters long/short on |signal| > entry_threshold while flat, exits on direction flip past exit_threshold (hysteresis), or on a hardcoded 18% take-profit / 6% stop-loss safety net. Position sizing targets 35% annualized vol, capped at 1.0x equity notional (leverage 1.0). Iteration 2 reduces the optimizer-tunable surface from 8 parameters to exactly 3 core knobs (lookback, entry_threshold, exit_threshold), hardcoding the safety-net and vol-sizing constants to the Binance benchmark values.

Sufficient data and healthy frequency for a daily walk-forward: ~3 years (1,100 bars), 101 trades (~34/yr), long+short (60/41) — unlike the data-starved HL 1H attempts

Distributed edge: Sharpe 1.44, max_DD 21.7%, tail_ratio 2.1, positive 2023/2024/2025; not single-outlier-dependent (kurtosis 11.2 moderate)

Cross-venue test of a proven mechanism (Binance SOL TSM sibling at paper stage, Sharpe 2.18); deployable capacity $54.6M with impact only 4.3%, minimal 3-parameter surface

sharpe_ci_low -0.0026 — CI just touches zero; modest edge, deflated Sharpe likely the binding optimization gate on the shorter 3-year sample

2026 -14.8% (partial year) — recent softness the holdout must clear

3-year history yields ~1-year walk-forward OOS windows (~10-13 trades each) — adequate but thin; OOS trade counts should be confirmed non-empty

Non-overfit walk-forward inputs (is_overfitted=FALSE, PBO 0.41) and a clean absolute profile in-sample (Sharpe 1.44, alpha 0.225, capacity $47M, impact 4.6%)

Mechanism is genuinely validated on the Binance SOL sibling (paper stage, Sharpe 2.18); this was a well-designed cross-venue test

Failed deflated Sharpe: DSR 0.0, is_significant=FALSE; optimized Sharpe 1.44 far below the expected-max luck bar 6.677 over 225 trials (luck bar inflated by extreme cross-window Sharpe variance on the short 3-year HL sample)

HOLDOUT FAILED with a NEGATIVE Sharpe -0.970 (ratio -0.289) — held-out forward window loses money

sharpe_ci_low -0.0639 (CI straddles 0); full-sample Sharpe not distinguishable from zero

Sensitivity cliff on the core lookback parameter (1.45 at 24/26 -> 0.68 at 29)

Recent decay (2026 -18.2%); walk-forward OOS average (3.36) propped by a single 10.61 window with another window negative (-1.12)

Re-optimize over ONLY the documented 2-3 core parameters and FIX the rest at the proven Binance-benchmark values. The hypothesis specifies a 2-parameter directional TSM, but the optimizer swept 8 parameters (lookback, entry_threshold, exit_threshold, stop_loss_pct, take_profit_pct, vol_floor, max_notional_frac, target_vol_annual). Searching 8 free knobs over a short ~3-year HL sample inflated the expected-max luck bar to 6.09 and collapsed DSR to 0 — a multiple-testing artifact, NOT proof the edge is fake (the walk-forward is genuinely not-overfit with all-positive OOS [7.56, 2.76, 2.08]). SPECIFIC CHANGES: (1) keep tunable: lookback, entry_threshold, exit_threshold only; (2) hardcode stop_loss_pct, take_profit_pct, vol_floor, max_notional_frac, target_vol_annual to the proven SOL-Binance benchmark constants (remove them from the search). Cutting the free-param count 8->3 lowers the deflation penalty so DSR reflects the real OOS edge. (3) Verify the narrowed config does not collapse in the most-recent window (holdout was mildly negative, -0.146, in the weak 2026 regime). DECISION RULE: if the 3-param re-run gives DSR >=0.95 / is_significant=true / non-negative holdout with the walk-forward still not-overfit, promote; if DSR still fails or the holdout stays negative, abandon — it means the HL ~3-year sample is too short to clear multiple-testing significance and/or the SOL TSM edge is venue/regime-specific. Detailed feedback written to workspace/discussions/c91e9ca1-39f7-4edd-81c0-d9208f1a86ce/iteration_1_feedback.md

HlSolTimeSeriesMomentumVolTargetedLS1D was an explicit cross-venue test of the single-asset TSM mechanism that landed on Binance SOL at paper stage (Sharpe 2.18), applying an identical 3-parameter directional momentum core to Hyperliquid SOL daily bars. It earned an 'optimize' verdict on a respectable base (Sharpe 1.44, 101 long/short trades, 21.7% drawdown), but optimization exposed an edge that could not clear best-of-225 deflation on the short HL sample: deflated Sharpe 0.0 against a 6.677 luck bar inflated by enormous cross-window Sharpe variance (trial Sharpes -1.5 to +10.6), a negative holdout (-0.970), a CI straddling zero, and a lookback sensitivity cliff. The hypothesis had pre-committed that an HL failure would be a meaningful signal the SOL TSM edge is Binance-microstructure-specific — and that was the outcome. At its second iteration, the analyst abandoned the venue extension, noting the mechanism still works on Binance but simply did not generalize to Hyperliquid.

Extending a validated mechanism to a venue with only ~3 years of history can fail deflated Sharpe because the short, volatile sample produces extreme cross-window Sharpe variance that inflates the best-of-N luck bar far above the honest selected Sharpe — and per the hypothesis's own pre-commitment, an HL failure where the Binance sibling succeeds signals the edge is Binance-microstructure-specific.

The analyst abandoned it post-optimization (its second and final iteration) because the deflated Sharpe was 0.0 against an inflated 6.677 luck bar, the holdout FAILED with a negative Sharpe of -0.970, the CI low was -0.0639 (straddling zero), and sensitivity FAILED with a cliff on the core lookback parameter — even though the walk-forward was non-overfit (PBO 0.41), its positive OOS average (3.36) was propped by a single 10.61 window with another window negative.

A cross-venue validation of the Moskowitz-Ooi-Pedersen 2012 single-asset time-series momentum mechanism on Hyperliquid SOL-USD daily bars — a long/short, OHLCV-only, 3-parameter directional TSM (score = trailing return / volatility, with hysteresis entry/exit thresholds) and volatility-targeted sizing, held identical to the proven Binance SOL sibling (at paper stage, Sharpe 2.18) to test whether the edge is venue-independent.

Over ~3 years (1,100 bars) the base backtest took 101 trades (60 long, 41 short) for a 161.8% return (CAGR 39.9%) with Sharpe 1.44, profit factor 1.36, and 21.7% max drawdown, but the CI low (-0.0026) just touched zero. Optimization barely changed the headline (Sharpe 1.44, PF 1.30, 162.7% return over 103 trades), while the forward windows broke down.