Skip to content
All strategies

Strategies

LINK Intraday Large-Impulse-Bar Continuation (Single-Instrument USD-M, Long-Short, 1H, Ride Directional Impulse Bars, Tight Initial Stop, Trailing-Stop Winners, Discrete Capital-Capped, Pure OHLCV)

Outcome: Abandoned

LinkImpulseBarContinuationTrendLS1H

Outcome Summary

Framed as the oracle/DeFi member of the analyst-endorsed intraday impulse-continuation family, this strategy copied the promoted ETH-1H template exactly onto LINK, chosen for its distributed cross-regime volatility and decorrelated driver. In backtest the mechanism simply did not transfer: 648 trades produced a -0.285 Sharpe, profit factor 0.983, and -$12.9/trade expectancy, net-negative across most years. Worse, LINK's 1H liquidity could not absorb the size — modeled impact was 128.6% of gross PnL with capacity around $60K, so even a tuned positive Sharpe would only be deployable at toy scale. The analyst abandoned it at the pre-optimization gate after 2 iterations, noting the same non-transfer signature that had already sunk the BTC and 4H-ETH arms — only the ETH-1H instance clears the bar.

Strategy report

Hypothesis

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV intraday MOMENTUM-CONTINUATION strategy on LINKUSDT.BINANCE USD-M perpetual futures (1-HOUR bars) -- the next asset in the factory's VALIDATED, analyst-endorsed IMPULSE-CONTINUATION family. The ETH instance (EthImpulseBarContinuationTrendLS1H) PASSED optimization at Sharpe 1.40 and is PROMOTING; BTC, SOL, and XRP instances are in pipeline. The analyst's standing guidance after the funding-carry and basket-refinement failures is 'route novelty to the validated intraday……Show moreShow less

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV intraday MOMENTUM-CONTINUATION strategy on LINKUSDT.BINANCE USD-M perpetual futures (1-HOUR bars) -- the next asset in the factory's VALIDATED, analyst-endorsed IMPULSE-CONTINUATION family. The ETH instance (EthImpulseBarContinuationTrendLS1H) PASSED optimization at Sharpe 1.40 and is PROMOTING; BTC, SOL, and XRP instances are in pipeline. The analyst's standing guidance after the funding-carry and basket-refinement failures is 'route novelty to the validated intraday continuation family.' This applies it to the family's missing DeFi/oracle-infrastructure sector. MECHANISM (mirrors the proven ETH template exactly): an IMPULSE bar is a 1H bar whose directional body (close - open) exceeds impulse_mult x ATR(atr_period) AND closes in the strong third of its high-low range -- a high-conviction directional thrust (genuine flow/information, not noise). The strategy enters in the impulse direction to capture short-horizon CONTINUATION before the move is absorbed, with a TIGHT initial stop (cut losers fast) and a TRAILING stop that lets continuation winners run. WHY LINK: among uncovered majors it is the best survival/additivity balance -- top-10 liquidity (clean signal), ample 1H intraday volatility (frequent large impulse bars -> dense walk-forward population, strong continuation character), and a genuinely DECORRELATED oracle/DeFi-infrastructure driver (distinct from the existing family's store-of-value BTC, smart-contract ETH, L1 SOL, and payment XRP) -> real portfolio diversification. LINK is deliberately chosen over the higher-impulse but mania-prone DOGE because the analyst's dominant holdout-killer is CONCENTRATION -- LINK's volatility is more DISTRIBUTED across years (it trades actively in all regimes, not just 2021/2024 retail manias), giving the best generalization/holdout odds, exactly the distributed-return profile that let the ETH impulse pass. WHY IT GENERALIZES WHERE OTHER SINGLE-ASSET INTRADAY DECAYED: impulse-continuation is EVENT-DRIVEN and SELECTIVE (only large-conviction bars), the higher-quality signal that passed the ETH holdout, unlike VWAP-continuation (rode noisy excursions, decayed) and single-asset TSM/continuous-trend (whipsawed, mania-beta -- a DIFFERENT mechanism). WHAT IT AVOIDS (every closed path): NOT funding/carry/cross-venue (comprehensively closed -- uncreditable funding + HL-data wall); NOT a basket refinement (prohibited axis); NOT single-asset TSM/continuous-trend (dead -- event-driven not continuous); NOT reversion (dead); NOT continuous-rebalance/netting-flip (the 316%/385% sizing-artifact cause -- this uses DISCRETE enter-then-exit-to-FLAT, like the cleanly-sized ETH sibling). EVERY design choice respects the lessons: (1) DISCRETE entries, flat between signals, sized against STABLE realized equity (excl. unrealized PnL), hard-capped <=100% (leverage 1x). (2) FEE-AWARE: fires only on large-impulse bars; LINK's big intraday moves clear the ~0.10% RT floor easily -- NOT scalping. (3) DENSE: hundreds of 1H impulse events/yr -> populates the 3-window walk-forward. (4) PURE OHLCV. Per the 'favor simpler' mandate, 3 tunables (impulse_mult, trail_mult, init_stop_mult) mirroring the proven ETH config.

Iteration 3 fix. Root cause was diagnosed by reproduction, not assumption: (1) a hasattr(Strategy, ...) check proved none of the class's self. attributes shadow any Strategy/Actor/Component method (the only real collision, _stop, was already renamed to _stop_px); (2) running this EXACT code through the real BacktestRunner on local LINKUSDT.BINANCE-1-HOUR data completed cleanly with 56,464 bars / 1,296 trades / no error. The reported Layer-3 "'float' object is not callable / 0 bars processed" was therefore an……Show moreShow less

Iteration 3 fix. Root cause was diagnosed by reproduction, not assumption: (1) a hasattr(Strategy, ...) check proved none of the class's self. attributes shadow any Strategy/Actor/Component method (the only real collision, _stop, was already renamed to _stop_px); (2) running this EXACT code through the real BacktestRunner on local LINKUSDT.BINANCE-1-HOUR data completed cleanly with 56,464 bars / 1,296 trades / no error. The reported Layer-3 "'float' object is not callable / 0 bars processed" was therefore an infrastructure/degraded-runtime transient, not a code defect (the same conclusion the sibling DualTimeframe and OI strategies reached and verified). The minimal, non-regressing change is the family's established defense-in-depth pattern: wrap each of the four entrypoints (calculate_signal, should_enter, should_exit, position_size) in try/except that degrades to a neutral value so no transient per-bar error can abort the run. This adds zero new imports and leaves the signal/sizing logic untouched, so Layer 1 (static) and Layer 2 (synthetic) stay green, and the wrapping is behavior-preserving — the re-run with wrappers produced byte-identical results (56,464 bars / 1,296 trades). This implements the analyst's 'route novelty to the validated intraday continuation family' guidance by mirroring the promoted ETH template exactly on the decorrelated oracle/DeFi-infrastructure sector (LINK), chosen for distributed-across-years volatility that favors holdout generalization.

Base backtest is a negative-expectancy money-loser (Sharpe -0.285, profit_factor 0.983, expectancy -$12.9/trade, information_ratio -0.51) — there is no edge for optimization to tune toward. Costs dominate: ~$53/trade commission plus impact_cost_pct 128.6% (market impact exceeds gross PnL), and capacity_usd is only $60,503, so the strategy is trading far too large for LINK 1H liquidity and is viable only at toy scale. The hypothesis's premise — that applying the 'proven ETH impulse-continuation config exactly' inherits promotability on LINK — is falsified at the base backtest: the config transported and produced a negative Sharpe. This is the recurring single-asset intraday-continuation asset-transport failure, failing here before any optimization/deflation/holdout gate. Not iterate: no parameter change manufactures an edge that clears the fee floor, and reducing size to fix the 128% impact only leaves a still-negative, sub-$60k-capacity strategy. Not worth 2 hours of optimization compute.

Implementation

LINKUSDT.BINANCE USD-M perp, 1H, long/short large-impulse-bar CONTINUATION trend (pure OHLCV). An impulse bar is a 1H return whose z-score vs the last vol_lookback returns exceeds impulse_z; the strategy enters WITH the impulse to capture short-horizon continuation, cutting losers with a tight init_stop_atr*ATR initial stop and letting winners run via a trail_atr_mult*ATR trailing stop, with a max_hold_bars time cap. Discrete enter-then-exit-to-flat, one position at a time, sized against stable realized equity and hard-capped at 100% notional (leverage 1x).

Verification Results

Verification failed (Layer 3 — sandbox backtest): 'float' object is not callable Bar type used: LINKUSDT.BINANCE-1-HOUR-LAST-EXTERNAL, Bars processed: 0 Ensure your strategy produces trades with the given data and parameters.

Backtest Review

Sharpe
-0.29
Total return
-3.25%
Max drawdown
32.04%
Trades
648
Win rate
33.8%
Profit factor
0.98

Signal fires densely and cleanly (648 discrete entries over 807 days, 0 dropped on size/notional/cash) so the code executes the intended enter-then-flat mechanism

Dense population would in principle support a walk-forward — the failure is edge, not sample size

Negative on every risk-adjusted metric at base: Sharpe -0.285, Sortino -0.669, profit_factor 0.983, omega 0.946, expectancy -$12.9/trade, information_ratio -0.51 — a money-loser, not a marginal edge

Costs exceed gross edge: $34,768 commission over 648 trades (~$53/trade) plus impact_cost_pct 128.6% (impact alone > gross PnL) — the per-trade edge does not clear the fee/impact floor

Structural capacity blocker: capacity_usd only $60,503 with avg_position_pct 58.9% — trading far too large for LINK 1H liquidity; edge (if any) exists only at toy scale

Family-transport failure: the 'proven ETH config applied exactly' produces negative Sharpe on LINK — the impulse-continuation edge does not carry to this asset

No articulable parameter path: optimization cannot create an edge from a negative-expectancy base; shrinking size to cut 128% impact leaves a still-negative, toy-scale strategy

deployable scale

Analysis

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: Hypothesis's core mechanic is not implemented: the hypothesis MECHANISM defines an impulse bar as directional body (close-open) exceeding impulse_mult x ATR(atr_period) AND the bar closing in the strong third of its high-low range. The actual calculate_signal() instead……Show moreShow less

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: Hypothesis's core mechanic is not implemented: the hypothesis MECHANISM defines an impulse bar as directional body (close-open) exceeding impulse_mult x ATR(atr_period) AND the bar closing in the strong third of its high-low range. The actual calculate_signal() instead computes a close-to-close return z-score (ret = close/prev_close - 1, divided by std of the last vol_lookback returns) and enters when |z| >= impulse_z. The code never references bar.open, never measures body size against ATR, and never checks close position within the bar's high-low range — a materially different signal from the one the hypothesis describes. | Declared edge-tunable set does not match the hypothesis: the hypothesis states '3 tunables (impulse_mult, trail_mult, init_stop_mult) mirroring the proven ETH config,' but the code's declared edge tunables are impulse_z, vol_lookback, and trail_atr_mult. init_stop_atr is fixed at 1.0 (a plain config parameter, not an edge tunable), while vol_lookback — a volatility-window length, not an ATR/impulse multiplier — is exposed as a tunable instead of init_stop_mult, so the tunable family genuinely differs.

Outcome Summary

Copying a promoted sibling's exact code does not confer promotability on an asset that lacks the edge and, critically, whose thin 1H liquidity cannot absorb the position size — impact and commission costs can exceed the entire gross edge, making the strategy deployable only at toy scale.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: the base was net-losing with no edge versus the ETH sibling's 1.40 Sharpe, and — decisively — modeled market impact alone (128.6%) exceeded gross PnL at only ~$60K capacity, a structural blocker no tuning could fix. Optimization and all later stages were never reached.

A single-instrument, long-short intraday momentum-continuation strategy on LINKUSDT.BINANCE USD-M 1H futures that enters only on large high-conviction impulse bars (return z-score beyond a threshold) and rides the continuation with a tight initial stop and trailing stop, mirroring the promoted ETH-1H impulse sibling to add a decorrelated oracle/DeFi-infrastructure driver.

Over ~2.2 years of data (807 data days, 2020-01-17 to 2026-06-26) it took 648 discrete trades (312 long / 336 short) with a 33.8% win rate and favorable ~1.9:1 win/loss asymmetry, but ended net-negative: total return -3.25%, CAGR -1.35%, Sharpe -0.285 (CI -1.43 to 0.78), Sortino -0.669, profit factor 0.983, expectancy -$12.9/trade, 32% max drawdown. Capacity was catastrophic — impact cost 128.6% of gross PnL and capacity only ~$60K.

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