BnbSpotDrawdownAccumulationLong

BNB Spot 30-Day Drawdown Accumulation with Recovery Exit (Daily)

A long-only single-instrument accumulation strategy on BNBUSDT spot (BINANCE_SPOT) using daily bars and OHLCV-only data. Architecturally mirrors the highly successful BTC Spot 30-Day Drawdown Accumulation (Sharpe 3.33, paper_stage) and the same pattern on ETH spot already in the pipeline — but applied to BNB, which has fundamentally different demand drivers (Binance exchange ecosystem flywheel: BNB-paid trading fee discounts, Launchpad participation, BNB Chain gas, quarterly BNB burns). These structural buyers create a different (and complementary) drawdown-recovery dynamic than BTC's institutional ETF flows or ETH's staking/DeFi flows. BNB's diversification value is high precisely because its drawdown character is exchange-ecosystem-driven rather than macro-driven. Single-dominant-filter design (200-day SMA + 30-day drawdown threshold) — explicitly NOT a multi-condition AND-gate (the recurring failure mode of AVAX Calm-Regime / DOT Pullback / SOL TOM, all abandoned for sparse-signal + outlier-dependency reasons). Calibrated for 5-10 entries/year (the empirically validated success zone of the BTC/ETH versions), NOT the 2-3/year sparsity zone that has killed multiple recent attempts. Uses ONLY OHLCV — no funding rates (which can structurally disappear, killing BTC Negative Funding), no proxies (which produce bait-and-switch failures), no calendar anomalies (which decay in 24/7 markets). BNB perp at 4H is the only existing BNB strategy; this spot/daily/DCA construction is mechanism-orthogonal so adds genuine portfolio diversification.

Directly replicates the highly successful, landed spot drawdown-accumulation pattern (BTC Sharpe 3.33; ETH/AAVE siblings in pipeline) and applies it to BNB, whose exchange-ecosystem structural buyers (fee-discount holders, Launchpad, BNB Chain gas, quarterly burns) give an exchange-driven rather than macro-driven drawdown-recovery dynamic — genuine diversification vs the BTC (ETF) and ETH (staking) accumulators, and mechanism-orthogonal to the only existing BNB strategy (a 4H perp). Venue is BINANCE_SPOT (CASH, leverage 1.0) exactly per the venue rules and the hypothesis: long-only, unleveraged, no funding component. It is deliberately a SINGLE-dominant-filter design (drawdown threshold + one 200-SMA trend gate), explicitly NOT the multi-condition AND-gate that caused the sparse-signal abandonments (AVAX Calm-Regime, DOT Pullback, SOL TOM). Reusing the proven sibling structure means its behavioral correctness and return profile are already established; the only open question for the new instrument is trade frequency, which I verified directly on real BNB daily history: drawdown_threshold=0.15 with the 200-SMA filter yields 31 entries over 6.3 years = 4.9/year — in the calibrated 5-10/year success zone, with 31 round-trips comfortably clearing the >=5-per-walk-forward-window floor (closed-position count, no harvest workaround needed). BNB spot daily OHLCV has full multi-year history via on-demand collection, so there is no data-gap risk. Verified locally: Layer 1 static (clean), Layer 2 synthetic (all 6 scenarios pass with a non-frozen SMA-deviation signal), and the real-data entry-frequency check above.

Failed deflated Sharpe: DSR=0.406, expected-max=7.96 over 225 trials — the optimized Sharpe of 7.13 is below the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO 0.546 > 0.5). Overfit by over-selectivity, exactly the SOL-DCA failure predicted at pre-optimization: walk-forward is_overfitted=true with avg IS 8.30 collapsing to avg OOS 0.0 and ALL three OOS windows exactly [0.0, 0.0, 0.0] — the optimizer raised drawdown_threshold 0.15->0.205 and cut peak_lookback_days 30->15, thinning the strategy to 19 trades (from 51 baseline) so the chosen config produces ZERO qualifying trades in every out-of-sample window, and the holdout is a degenerate fail (holdout_sharpe 0.0 against a 0.0 WF-OOS reference, ratio 0). The favorable full-period sharpe_ci_low (0.67) and PSR (0.979) are bootstrap statistics that do not correct for the 225-trial selection that DSR corrects for, and are swamped by the all-zero OOS and degenerate holdout — a config that produces no out-of-sample trades has not generalized regardless of its in-sample CI. Sensitivity passing (cliff_count 0) is necessary but not sufficient. This is not a tunable-region case: the 225-trial sweep already established that its best config does not trade out-of-sample, so a second pass over the same space hits the identical wall — there is no robust region to tune toward. Not revise_hypothesis: the drawdown-accumulation mechanism is already landed on its BTC spot sibling (Sharpe 3.33, paper_stage), so BNB is a dead target for the family rather than a premise needing reframing. FAILURE PATTERN: porting the 7-parameter spot drawdown-DCA family to BNB overfits in optimization — the optimizer maximizes in-sample Sharpe by raising drawdown_threshold and shortening the peak lookback to over-select, yielding zero walk-forward OOS trades and a degenerate holdout, failing DSR/PBO; this is the second DCA port (after SOL) to fail with this identical zero-OOS signature.

Long-only BNB spot (BINANCE_SPOT, CASH) 30-day drawdown-accumulation strategy on daily bars, OHLCV-only, mirroring the proven BTC/ETH/AAVE siblings (BTC at paper stage, Sharpe 3.33). It buys the dip inside an established uptrend: enter LONG when BNB has fallen >= drawdown_threshold (15%) from its rolling peak_lookback_days (30) high AND the close is above the trend_sma (200-day, adaptive to available warmup), the single-dominant filter that avoids catching falling knives in a downtrend. It exits when price recovers to within recovery_threshold (5%) of the entry peak (the reversion target), on a stop_loss_pct (10%) catastrophic stop, or after max_hold_days (60). Each dip->recovery is a round-trip, calibrated to ~5/year (the BTC/ETH success zone, not the 2-3/year sparsity-failure zone). Position size is 20% of equity per entry. calculate_signal returns a continuous SMA-deviation diagnostic (non-frozen in all scenarios) while the discrete drawdown+trend gate lives in should_enter. Single instrument, long-only, leverage 1.0, zero supplementary data.

Proven mechanism family (BTC spot drawdown-accumulation sibling at paper stage, Sharpe 3.33); clean execution (51 signaled/submitted, 0 dropped)

Healthiest sample of the DCA ports: 51 fully-realized trades over ~8.5 years (~6/yr), in the 5-10/yr success zone and far above the failed SOL port's ~19 — end_unrealized only 0.6%

Positive edge above threshold: profit_factor 1.24, positive expectancy $550/trade, in-market only 24.5% of the time (capacity to scale/combine)

Front-loaded edge: 28.7% total return dominated by 2017-2018 BNB explosion; recent years barely positive (2024 +1.8%, 2025 +0.6%, 2026 +0.6%) with 2019/2021/2022 negative; CAGR only ~3.1%

Distorted dispersion: annualized_vol 75% and Sortino 10 are per-period-attribution artifacts, so headline Sharpe 1.94 is unreliable; sharpe_ci_low -2.95 straddles 0 widely

7 parameters on 51 trades (~7/param) — the same over-parameterization that let the optimizer curve-fit the SOL sibling (relaxed drawdown_threshold -> zero OOS trades)

Clean execution and sensitivity passed (cliff_count 0); mechanism landed on the BTC sibling (Sharpe 3.33, paper_stage)

Favorable full-period sharpe_ci_low 0.67 and PSR 0.979 in isolation — but these do not correct for the 225-trial selection

Failed deflated Sharpe: DSR 0.406 (<0.95), optimized Sharpe 7.13 below expected-max luck bar 7.96 over 225 trials; is_significant=false, PBO 0.546 > 0.5

Walk-forward is_overfitted=true: avg IS 8.30 collapses to avg OOS 0.0 with ALL three OOS windows [0.0, 0.0, 0.0] — the chosen config produces zero out-of-sample trades

Optimizer over-selected: drawdown_threshold 0.15->0.205, peak_lookback 30->15, thinning to 19 trades (from 51) — curve-fit in-sample on a thinned base

Holdout degenerate fail (holdout_sharpe 0.0 vs 0.0 WF-OOS reference, ratio 0) — no out-of-sample activity to validate

BnbSpotDrawdownAccumulationLong was the strongest-looking of the spot drawdown-DCA ports — a faithful transfer of the landed BTC sibling (Sharpe 3.33) to BNB, with a healthy 51-trade sample in the 5-10/year success zone and a positive baseline edge that earned an 'optimize' verdict. But its edge was front-loaded into the 2017-2018 BNB explosion with recent years barely positive, and the over-parameterized optimizer curve-fit exactly as predicted: it raised the drawdown threshold and halved the peak lookback to maximize in-sample Sharpe, thinning to 19 trades and producing zero out-of-sample trades across all three walk-forward windows, failing DSR (0.406) and leaving a degenerate holdout. The analyst ruled it overfit-by-over-selectivity — the second DCA port after SOL to fail with this identical zero-OOS signature — and abandoned it. It ended after one iteration as abandoned, having reached optimization and analysis but never risk review.

Porting the 7-parameter spot drawdown-DCA family to BNB overfits in optimization the same way the SOL port did: with only ~7 trades per parameter, the optimizer maximizes in-sample Sharpe by raising the drawdown threshold and shortening the peak lookback until the strategy produces zero out-of-sample trades — a front-loaded, 2017-2018-dominated baseline edge does not generalize, and full-period bootstrap CI/PSR do not rescue a config that has no out-of-sample activity.

Although it passed the backtest-review gate with an 'optimize' verdict, the analyst abandoned it at the post-optimization analyzing stage: it failed deflated Sharpe (0.406, below the luck bar, is_significant=false), and the optimizer over-selected by raising the drawdown threshold and shortening the peak lookback until its chosen config produced zero qualifying trades in all three walk-forward OOS windows plus a degenerate holdout — so it reached optimization and analysis but never advanced to risk review.

A long-only single-instrument drawdown-accumulation strategy on BNBUSDT spot (daily bars, OHLCV-only) that bought when BNB fell >=15% from its 30-day rolling peak while price was above the 200-day SMA (a single-dominant-filter dip-buy in an uptrend), exiting on recovery toward the entry peak, a stop-loss, or a time stop — a port of the proven BTC spot drawdown-accumulation sibling (Sharpe 3.33, paper_stage) to BNB.

The baseline was the healthiest of the DCA ports — 51 fully-realized trades over ~8.5 years (~6/yr), profit factor 1.24, positive expectancy $550/trade, only 24.5% time in market — but front-loaded (28.7% total return dominated by the 2017-2018 BNB explosion, CAGR ~3.1%, recent years barely positive, headline Sharpe 1.94 unreliable with sharpe_ci_low -2.95); optimization then over-selected (drawdown_threshold 0.15→0.205, peak_lookback 30→15) thinning to 19 trades, and DSR collapsed to 0.406 (below the 7.96 luck bar), PBO 0.546, walk-forward in-sample 8.30 falling to 0.0 OOS with all three windows producing zero out-of-sample trades and a degenerate holdout (ratio 0).