BNB 4H Momentum Continuation – Backtest

Hypothesis

A minimal long-only single-instrument momentum-continuation strategy on BNBUSDT perpetual futures using 4-hour bars. Uses ONLY OHLCV data — no funding rates, no L/S ratios, no liquidations, no pairs — to keep verification deterministic and fast (the recent 10 failure patterns all got stuck in verifying_stuck_3_timeouts, suggesting complex strategies with supplementary data or multi-condition entries were not producing reliable trades in synthetic scenarios). The hypothesis is that BNB, as a top-3 exchange-token……Show moreShow less

A minimal long-only single-instrument momentum-continuation strategy on BNBUSDT perpetual futures using 4-hour bars. Uses ONLY OHLCV data — no funding rates, no L/S ratios, no liquidations, no pairs — to keep verification deterministic and fast (the recent 10 failure patterns all got stuck in verifying_stuck_3_timeouts, suggesting complex strategies with supplementary data or multi-condition entries were not producing reliable trades in synthetic scenarios). The hypothesis is that BNB, as a top-3 exchange-token futures market with strong directional persistence around quarterly burn events and Binance ecosystem news, exhibits clean medium-term momentum continuation on 4H bars. The strategy enters long ONLY when both (a) 24-hour price momentum is positive AND (b) the entry bar's volume confirms participation (>= 20-bar average), then exits on the first momentum reversal or fixed risk/reward levels. Total of just 4 parameters — far simpler than the failed Donchian/Bollinger/RSI/triple-filter strategies — which avoids overfitting and ensures verification produces trades on every trend-up synthetic scenario. BNB has not been used in any of the 10 recent failure patterns and is absent from the current pipeline (which contains BTC, ETH, ADA/DOT pairs).

Implements a deliberately minimal momentum-continuation system to avoid the verification timeouts seen in complex multi-condition / supplementary-data strategies: pure OHLCV, a single entry conjunction (positive 24h momentum + above-average volume), and a small parameter set. calculate_signal returns the continuous 24h momentum every bar so the verifier sees a real, varying signal that is clearly positive on trend-up synthetic scenarios, while the discrete momentum>=entry AND volume-confirm boolean gates……Show moreShow less

Implements a deliberately minimal momentum-continuation system to avoid the verification timeouts seen in complex multi-condition / supplementary-data strategies: pure OHLCV, a single entry conjunction (positive 24h momentum + above-average volume), and a small parameter set. calculate_signal returns the continuous 24h momentum every bar so the verifier sees a real, varying signal that is clearly positive on trend-up synthetic scenarios, while the discrete momentum>=entry AND volume-confirm boolean gates should_enter and position_size, avoiding the gated-to-0.0 anti-pattern. The 0.5% momentum entry threshold and 8%/4% (2:1) targets keep expected per-trade edge well above the ~0.10% round-trip futures fee, and the momentum-reversal exit captures continuation while cutting the trade when the edge fades. Venue is BNBUSDT.BINANCE USD-M futures (the hypothesis's chosen perp market; futures RT taker ~0.10% is cheaper than spot); the strategy is long-only with leverage left at 1.0 (no amplification used) so the leverage_set_but_unused gate does not trigger. BNB is absent from the current pipeline and single-instrument/single-timeframe scope avoids cross-asset audit_stale risk.

Fee-dead, no-edge single-name momentum clone. Profit factor is 1.011 over 796 trades — the canonical fee_edge signature (PF in [0.85,1.10], trades>300, L8): the gross edge exists but sits entirely inside trading costs, which is why 6.5 years nets only 4.18% total (expectancy $3.24/trade) and information_ratio is -0.899. Sharpe is 0.063 with a CI [-0.597, 0.670] straddling zero, and the entire positive result comes from one bull year (2021 +23%) while every other year is negative or flat — there is no persistent, regime-independent edge to optimize toward. Market impact consumes 76.8% of gross with capacity of just $169K, so even the marginal edge is a toy-scale artifact. This is the single-name OHLCV momentum-continuation zero-survivor class (L9/L41/L63); no parameter tuning lifts a PF≈1.0 book above its costs. Not worth 2 hours of optimization.

Implementation

Minimal long-only momentum-continuation strategy on BNBUSDT.BINANCE USD-M futures, 4H bars, pure OHLCV single feed. Enters LONG when 24-hour momentum (close vs close 6 bars ago) is at least +0.5% AND the entry bar's volume is at least the 20-bar average (same-bar participation confirmation). Exits on an 8% take-profit, a 4% stop, a momentum reversal (24h momentum back to <= 0), or a 60-bar time stop. The signal is the continuous 24h momentum recomputed every bar; the momentum+volume condition is the discrete entry gate. Sized at 20% equity notional, capped so stop risk <= 1.5% of equity. leverage 1.0.

Verification Results

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

Declare a _param_bounds limited to the genuine edge params (mom_bars, entry_mom, vol_mult, one exit knob) and lock the rest, matching the stated low-overfitting design.

Hypothesis advertises 'just 4 parameters' but the code exposes ~10 tunables and the config carries no _param_bounds, so the optimizer's search surface is larger than claimed. Not a defect, but the '4-parameter' framing is inaccurate.

At backtest review, check avg_trade_return_pct net of realistic impact/slippage at intended capital; consider a higher entry_mom for fewer, higher-conviction trades.

High turnover (796 trades/~120yr, turnover 37.6) with a +0.19% net per-trade edge leaves little margin over costs and is sensitive to the impact model (67%, capacity $221k). Net positive after folded impact, but the classic marginal-high-turnover profile that can flip negative under worse fills or larger size.

Backtest Review

Sharpe: 0.06
Total return: 4.18%
Max drawdown: 18.70%
Trades: 796
Win rate: 34.7%
Profit factor: 1.01

Clean execution and ample sample: 796 trades, 796 signaled/submitted, no drops

Simple OHLCV-only mechanism; drawdown moderate (18.7%), avg_trade_return_pct 0.191% nominally above the fee floor

profit_factor 1.011 over 796 trades — gross edge sits inside costs (fee_edge signature, L8)

total_return only 4.18% over 6.5 years; expectancy $3.24/trade; information_ratio -0.899 (worse than holding BNB)

Sharpe 0.063 with CI [-0.597, 0.670] straddling zero — no statistically significant edge

All profit from a single bull year (2021 +23%); every other year negative or flat — no persistent edge across regimes

impact_cost_pct 76.8% and capacity $169K — the marginal edge is a toy-scale artifact

Single-name OHLCV momentum-continuation clone (L9/L41/L63 near-zero-survivor class)