BtcDailyDeathCrossBearRegimeShort

BTC Daily Death-Cross Bear Regime Short — Volatility-Confirmed Trend Reversal

A SHORT-ONLY strategy on BTCUSDT.BINANCE perpetual futures (Binance USD-M, ~0.10% RT) using DAILY bars. The strategy enters short only during confirmed bear-trend regimes identified by a triple-filter: (1) 50-day SMA below 200-day SMA (classic death cross), (2) current price below 200-day SMA (trend confirmation, not just an MA crossover), AND (3) 20-day realized volatility above its 90-day median (volatility expansion — bear markets historically expand vol, while sideways markets compress it). Exits on golden-cross reversal (50-SMA crosses back above 200-SMA), 10% adverse move stop, or 60-day time-stop. This is the FIRST short-only strategy proposed this session and adds genuine diversification to a pipeline of 11 long-biased strategies. Mechanism is structurally distinct from every prior session failure: NOT cross-asset (single instrument), NOT cross-sectional (no ranking), NOT mean-reversion (pure trend continuation in the bear direction), NOT sub-daily (daily bars), NOT calendar-based, NOT supp-data dependent, NOT accumulation, NOT a pair, NOT cross-venue, NOT options. While the technical primitives (SMA crossover) overlap with the Tier 1 trend-following category (which had 4 long-only failures), the SHORT direction is empirically the opposite regime: 2022 BTC -77%, 2018 BTC -84%, 2014 BTC -85% are all captured by death-cross + sub-200-SMA filters, regimes that NO long-only strategy can profit from. The volatility-expansion filter is the critical addition that prevents whipsaws during sideways markets — the same SMA crossover without it would fire repeatedly in chop. Trade frequency is naturally very low (~2-5 entries per 7 years) so fee drag is minimal.

Iteration 2 fixes a harness ROUTING failure, not a code defect: the prior backtest ran on BTCUSD.HYPERLIQUID-1-MINUTE over 5 days with empty parameters {} — impossible for a 50/200-day-SMA daily strategy (needs ~205 bars warmup) to ever trade. The strategy code is byte-identical to the version that already passed Layers 1-3 (imports clean, continuous bear-regime signal preserved, short-only sell-to-enter on BINANCE futures). The only change is making the config unambiguous and self-contained so the re-run cannot silently drop it: instrument_id pinned to 'BTCUSDT.BINANCE', bar_type to 'BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL', leverage 1.0, and every parameter emitted explicitly (rather than left to .get() defaults) so an empty-parameters override cannot occur. On a correct BTC daily run over the full history, the triple filter should place short entries only inside genuine death-cross + sub-200-SMA + vol-expansion bear regimes (2018, 2022). Given the tiny expected trade count, walk-forward sample sufficiency should be confirmed before optimizing.

Fails every robustness and forward gate decisively on optimization attempt 1. Deflated Sharpe DSR=0.0 (vs 0.95 bar) with the optimized Sharpe 1.56 FAR below the 225-trial expected-max luck bar of 5.95 (is_significant=false, PBO=0.8294>0.5, sharpe_ci_low=-0.336 straddling 0) — after multiple-testing correction the selected config is statistically indistinguishable from best-of-N noise; the PSR=0.9913 / DSR=0.0 split is the textbook tell. The walk-forward is is_overfitted=TRUE with avg IS Sharpe 8.40 (windows [2.84, 1.84, 20.51], one absurd) collapsing to avg OOS Sharpe 0.0 across all three windows, and the untouched time-ordered holdout FAILED and is NEGATIVE (holdout_sharpe -0.573). Sensitivity FAILED with 7 of 10 parameters flagged as cliffs — a knife-edge surface with no robust region to tune toward. The apparent strength of the optimized full-refit (Sharpe 1.56, PF 5.20, win 78%) is an artifact of refitting on just 9 trades whose edge is concentrated in the single 2022 bear regime (kurtosis 93-120). Not iterate: the binding failures are a negative holdout, all-zero OOS, and a 7-cliff minefield on a tiny single-regime sample — there is no robust parameter region, so a second best-of-225 sweep re-overfits the same 2022 trades. Not revise_hypothesis: this is overfitting on a thin single-regime sample (per the overfit-vs-dead-premise rule, overfit → abandon), not a proven mechanism stranded on a dead target — BTC is fine and there is no promoted short/SMA-crossover sibling; the SMA-crossover trend primitive already had 4 long-only failures this session and its short variant reaches the same end. FAILURE PATTERN: a structurally-distinct, faithfully-implemented single-asset daily death-cross short whose entire edge is the 2022 bear captured by ~9-15 trades cannot clear best-of-225 deflation (DSR 0.0, Sharpe 1.56 vs 5.95 luck bar, PBO 0.83); a flattering 9-trade full-refit (PF 5.2, win 78%, PSR 0.99) does NOT rescue an edge whose walk-forward OOS is zero, whose forward holdout is negative, and whose sensitivity surface is 7-of-10 cliffs.

SHORT-ONLY BTCUSDT.BINANCE USD-M perp, DAILY bars, OHLCV-only. Enters short only in confirmed bear regimes via a triple filter: 50-SMA below 200-SMA (death cross), close below the 200-SMA (trend confirmation), and 20-day realized vol above its 90-day median (vol expansion, which suppresses sideways-chop whipsaws). Exits on golden-cross reversal, a 10% adverse-move stop, or a 60-day time stop. Naturally very low frequency (~2-5 entries/7yr) so fee drag is minimal. Requires the full multi-year history (2018/2022 bear markets) to test the thesis; the 50/200 SMAs need ~200 bars of warmup.

Trades faithfully implement the hypothesis: short-only (0 longs, 15 shorts), entries fire in confirmed bear regimes, and the bulk of profit is the 2022 crash (+6.97 that year) — exactly the mechanism claimed.

Clean execution plumbing: 15 signaled = 15 submitted, zero size/notional/cash drops; signal is not frozen or degenerate.

15 trades over ~6.4 years is enough sample to make optimization meaningful (more than the ~2-5 the hypothesis feared), with positive expectancy (338/trade), PF 1.41, and low fee drag.

Genuine diversification value — first short-only, market-neutral-ish (beta -0.03) strategy, positive alpha (+0.016).

Thin, single-regime edge: Sharpe only 0.36 with sharpe_ci_low -1.13 straddling 0; profits are concentrated in 2022 with high kurtosis (120) and skew (7.77) — outlier-carried.

Win rate 47% and information_ratio -0.67; the edge may not survive multiple-testing/deflated-Sharpe correction after the sweep.

15 trades is workable but on the thin side — optimization must guard against overfitting the one bear regime.

Correct, faithful implementation: short-only (0 longs, 15/9 shorts), entries fire in bear regimes, clean plumbing, benchmark_meaningful=true with negative beta (genuine diversification profile).

The optimized full-refit looks superficially strong (Sharpe 1.56, PF 5.20, win 78%) — but only because it is refit on 9 trades.

Fails deflated Sharpe decisively: DSR=0.0 (vs 0.95 bar), optimized Sharpe 1.56 FAR below the 225-trial expected-max luck bar of 5.95, PBO=0.83 (>0.5, selection more likely than not overfit), sharpe_ci_low=-0.336 straddles 0, is_significant=false. PSR 0.9913 vs DSR 0.0 is the classic multiple-testing trap.

Walk-forward is_overfitted=TRUE: avg IS Sharpe 8.40 (windows [2.84, 1.84, 20.51]) collapses to avg OOS Sharpe 0.0 across all three windows.

Holdout FAILED and is NEGATIVE: holdout_sharpe -0.573 — the untouched recent window loses money.

Sensitivity FAILED: 7 of 10 parameters are cliffs (fast_sma, slow_sma, vol_window, max_hold_bars, stop_loss_pct, vol_median_window, vol_expansion_mult) — knife-edge, no robust plateau to tune toward.

Edge is a single-regime artifact: essentially all profit is 2022 (base annual +6.97; every other year flat-to-negative), on a tiny sample (15 base / 9 optimized trades) with return_kurtosis 93-120.

The backtest was run on the WRONG instrument and timeframe: config shows BTCUSD.HYPERLIQUID-1-MINUTE-LAST-EXTERNAL over 5 days (2026-05-03 to 2026-05-13) with empty parameters {}, but the hypothesis and strategy code target BTCUSDT.BINANCE on 1-DAY bars over ~7 years. This is the FOURTH strategy in a row misrouted to this exact dataset (identical instrument/timeframe/dates/bars_processed=5159 with empty parameters) — the backtest harness is overriding instrument_id/bar_type and dropping the parameters dict for every strategy this batch. Fix that routing first. Then re-run with the correct FactoryStrategyConfig: instrument_id='BTCUSDT.BINANCE', bar_type='BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL', and the default parameters, over the full multi-year history (the 50/200-day SMAs require ~1 year of warmup alone, so the run must span the 2018/2022 bear markets to test the thesis at all). Verify metrics_reliable=true and that the short entries land in genuine bear regimes. Only after a valid BTC daily run can this be judged; given the tiny expected trade count (~2-5 entries/7yr), also confirm there is enough sample for walk-forward before optimizing.

BtcDailyDeathCrossBearRegimeShort introduced the session's first short-only strategy, fading BTC in death-cross + sub-200-SMA + vol-expansion bear regimes, and its 15-trade backtest faithfully did what the hypothesis claimed — market-neutral-ish (beta -0.03), positive alpha, profit factor 1.41 — with nearly all gains from the 2022 crash. The analyst let it advance to optimize, but the 3-phase optimization exposed the thinness: the best-of-225 config posted deflated Sharpe 0.0 against a 5.95 luck bar, PBO 0.83, a walk-forward that collapsed from an 8.40 in-sample Sharpe to 0.0 out-of-sample, a negative -0.573 holdout, and 7-of-10 sensitivity cliffs. The apparent 9-trade refit strength (Sharpe 1.56, 78% win) was judged an artifact of overfitting a single 2022 regime, so on iteration 2 the analyst abandoned it rather than re-sweep the same trades. The verdict framed it as classic single-regime overfitting on a tiny sample — the SMA-crossover primitive's short variant reaching the same dead end as its four long-only siblings.

A faithfully-implemented, structurally-distinct short strategy still fails when its entire edge is one bear regime captured by ~9–15 trades — a flattering low-sample full-refit (PF 5.2, PSR 0.99) cannot rescue an edge whose walk-forward OOS is zero, whose forward holdout is negative, and whose parameter surface is a cliff minefield.

The backtest review passed it to optimize, but the analyst abandoned it after optimization for failing every robustness and forward gate: deflated Sharpe 0.0 (vs 0.95 bar) with optimized Sharpe 1.56 far below the 225-trial luck bar of 5.95, PBO 0.83, walk-forward is_overfitted=TRUE (avg IS Sharpe 8.40 collapsing to OOS 0.0), a negative holdout (-0.573), and sensitivity cliffs on 7 of 10 parameters.

A short-only BTCUSDT.BINANCE daily strategy that shorts only in confirmed bear regimes identified by a triple filter — 50-SMA below 200-SMA (death cross), price below the 200-SMA, and 20-day realized vol above its 90-day median — exiting on a golden-cross reversal, a 10% stop, or a 60-day time stop, as the session's first short-only diversifier.

The initial backtest took 15 short trades (0 longs) over ~6.4 years with a 47% win rate, profit factor 1.41, total return 4.48, Sharpe 0.36 (sharpe_ci_low -1.13), max drawdown 8.5%, and near-zero beta (-0.03) — with essentially all profit concentrated in the 2022 crash (+6.97 that year) and heavy kurtosis (~120). The optimized refit looked strong on the surface (Sharpe 1.56, PF 5.20, 78% win) but rested on just 9 trades.