Strategies

FMZ Dual Thrust: LINK/LTC/BCH Hyperliquid 4H Long–Short Basket

Outcome: Abandoned

HyperliquidThreeSleeveDualThrust

Outcome Summary

HyperliquidThreeSleeveDualThrust tested a perpetual-futures, completed-bar, 1x-leverage adaptation of the Dual Thrust breakout on LINK, LTC and BCH 4-hour bars. Its first backtest covered 721 trades on clean data and lost 71.7%, with a Sharpe of -1.98, profit factor 0.658, a 75.4% max drawdown and losses in every year and every volatility regime. The analyst abandoned it at backtest review, before optimization, because the edge was negative even before commission and impact (about -0.31% per trade). Optimization, risk review and promotion were never reached, and the verdict applies only to this adaptation, not to the original intrabar quarterly-futures version.

Strategy report

Hypothesis

Test the posted Dual Thrust range-breakout mechanism independently on LINKUSD, LTCUSD and BCHUSD Hyperliquid perpetuals using 4-HOUR bars. Exact catalog directories exist for all three. Use completed-bar decisions, prior completed bars for range formation, market orders and capital-relative sizing. This is an explicit adaptation: the source polls intrabar, trades quarterly futures at 10x margin and uses unspecified external parameter defaults. This experiment instead uses perpetuals, 1x account leverage, a symmetric coefficient, bounded exposure and protective exits. It does not claim to reproduce the original execution or returns.

The source contains executable breakout and reversal rules, so it qualifies as a concrete idea. Hyperliquid is underrepresented at 9.4% against a 20% target; a three-instrument long–short basket also addresses excessive single-symbol and long-only exposure. Options are more underrepresented but would change the user's mechanism. Corpus concentration is empty, so low historical usage of these particular symbols cannot be established. Dual Thrust already appears in the pipeline on AVAX; this remains a related source-directed experiment, with explicit differences in implementation, venue, portfolio construction and execution. It should not be described as a novel signal family.

negative_expectancy (fee_edge/no_edge): Fixed-default Dual Thrust (lookback 6, k=0.5) on 4H LINK/LTC/BCH Hyperliquid perps loses money on a reliable sample. avg_trade_return_pct is -0.834% against the +0.15% Hyperliquid floor. PF is 0.658, Sharpe -1.98 (95% CI [-3.39, -0.62]) over 721 trades, max drawdown 75.4% against the pre-registered 15%, and every year and every volatility regime lost money. The gross edge before commission and impact is still about -0.31% per trade, so the breakout signal itself is wrong-signed after whipsaws, not merely eaten by fees. The optimization plan only tunes lookback (3-12) and the coefficient (0.3-0.65), which cannot turn a uniformly negative, tightly estimated edge positive. This confirms QA's edge concern: the full history (-0.83% per trade, PF 0.66) matches the smoke test (-0.80%, PF 0.674). Because the result is a tight negative estimate, it measures absence of edge, not a coding defect. The verdict covers this completed-bar 4H perp adaptation only. It does not test the original intrabar quarterly-futures version. Separately, the avg_holding_bars=0 and exposure_pct=163.6% fields should be reviewed for reporting consistency.

Implementation

Independent long–short Dual Thrust sleeves on LINK, LTC and BCH Hyperliquid perpetuals. Each completed 4H close is compared with symmetric rails around that bar's open, using the preceding completed bars' Dual Thrust range. Market exits occur on the opposite rail, an entry-anchored protective stop, or 42 held bars. A full one-bar cooldown follows closure. Entry sizing risks at most 0.3% of equity against the modeled stop distance, capped at 20% notional per sleeve and 60% aggregate gross.

Implements the explicitly adapted perpetual, completed-close, 1x basket rather than claiming source execution parity. Range formation excludes the decision bar, and extra-leg timestamps must equal the primary timestamp. Stop distance is the greater of 1% of decision-close price and half the entry-time range; it is frozen when the position opens. Stops execute at completed closes, so gaps and intrabar excursions can exceed the sizing risk budget. Opposite breakouts close positions before cooldown-governed reverse entries. Only lookback_bars and range_coefficient are tunable under the supplied research plan. Position and cooldown state use persisted scalar attributes. No verification was run, as requested. The requested outbox file could not be written because this workspace is read-only; the complete bound JSON is returned here.

Verification Results

The sandbox lost 68.14% over 721 closed positions, with profit factor 0.674 and average net return -0.800% per trade. Full-history review should examine breakout follow-through, whipsaws and close-only stop overshoot. These results do not establish a coding defect.

Backtest Review

Sharpe: -1.98
Total return: -71.66%
Max drawdown: 75.44%
Trades: 721
Win rate: 25.2%
Profit factor: 0.66

The sample is reliable: 721 closed positions over 815 days on all three legs, with no missing or duplicate bars and metrics_reliable=true.

Accounting reconciles, nothing is left open at the end (end_unrealized_pct 0), and funding is credited.

Trades match the stated mechanism: 351 longs and 370 shorts, a symmetric two-sided breakout with no pyramiding.

avg_trade_return_pct is -0.834%, against a Hyperliquid floor of +0.15%. It is negative, not just below the floor.

Profit factor is 0.658. Win rate is 0.252 with a payoff ratio of about 1.95 (avg win $758 vs avg loss $389), which is not enough to offset the hit rate.

Sharpe is -1.98 with a 95% CI of [-3.39, -0.62]. The whole interval is negative, so this is a tight, wrong-signed estimate.

Every calendar year lost money (2024 -12.1%, 2025 -59.8%, 2026 YTD -19.9%), and so did every volatility regime (calm Sharpe -0.42, normal -3.09, stressed -2.44).

Max drawdown is 75.4% against the pre-registered 15%, and the drawdown lasted 778 days.

The edge is negative even before costs. Commission ($9.5k) plus modeled impact ($29.8k) comes to about $54 per trade. The net loss is about $99 per trade on roughly $14.6k of notional, so the gross result is about -$45 per trade, or about -0.31% of notional. Lowering costs or tuning parameters cannot flip the sign.

The signal diagnostic shows no information: IC 0.10, t=1.1.

Some reporting fields look inconsistent: avg_holding_bars is 0.0 and exposure_pct is 163.6%, although gross notional is capped at 60%. These are worth a separate look, but they do not affect the sign of the realized PnL.

Outcome Summary

Completed-bar 4H Dual Thrust breakouts on these mid-cap perps were wrong-signed before fees, because of whipsaws, and the full-history result matched the smoke test (-0.80%/trade, PF 0.674), so a negative smoke test with QA edge concerns is a reliable early stop signal.

At backtest review the analyst voted to abandon for negative expectancy. avg_trade_return_pct was -0.834%, far below the +0.15% Hyperliquid floor, and the gross edge before costs was still about -0.31% per trade, so the 3-12 lookback and 0.3-0.65 coefficient ranges in the optimization plan were judged unable to flip the sign.

An adaptation of the FMZ Dual Thrust range-breakout run as three independent long/short sleeves on LINK, LTC and BCH Hyperliquid perpetuals with 4-hour bars, completed-bar decisions, 1x leverage, a symmetric coefficient (lookback 6, k=0.5) and protective exits.

Over 815 days and 721 closed trades it returned -71.7% (CAGR -43.2%) with a Sharpe of -1.98 (95% CI [-3.39, -0.62]), profit factor 0.658, win rate 25.2%, avg_trade_return_pct -0.834% and max drawdown 75.4%. Every calendar year and every volatility regime lost money.

Analysis

New version of Dual Thrust Trading strategy

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