BnbDailyWilliamsVixFixCapitulationRecoveryLong

BNB Daily Williams VIX-Fix Capitulation Recovery Long

A LONG-ONLY synthetic-volatility-based reversal strategy on BNBUSDT.BINANCE perpetual futures (Binance USD-M) using DAILY bars. Implements Larry Williams' VIX-Fix indicator (Williams 2007 'Trading without Fear'), originally designed to give equity traders an options-free panic gauge: VIXFIX(t) = 100 × (highest_high(22, t) - low(t)) / highest_high(22, t). The indicator spikes when current bars print lows far below the 22-bar high — a structural definition of capitulation that requires no funding rates, no taker ratios, no supplementary data feeds (all of which have been falsified this session). Entry mechanism: detect that VIX-Fix has crossed ABOVE its 90th-percentile threshold (a 90-day rolling percentile) → confirms panic regime is active. Then enter LONG when VIX-Fix subsequently crosses BACK DOWN below its 80th-percentile threshold → confirms panic is subsiding and stabilization has begun. This sequence (panic → stabilization) historically precedes 4-8% bounces over 5-10 days in trending and ranging markets alike, with documented effectiveness across stocks, ETFs, futures, and crypto. BNB is selected because: (1) only ONE pipeline strategy currently uses BNB (the 4H volume breakout) — daily-bar VIX-Fix adds within-instrument diversity; (2) BNB has 5+ years of clean Binance USD-M data, no MATIC-style deprecation; (3) BNB sits in the precision map (standard major); (4) BNB exhibits especially clean capitulation patterns due to its Binance-token reflexivity (exchange-token holders panic-sell during exchange-FUD events, then resume buying when the news clears — a fundamental driver that produces particularly clean VIX-Fix signals). Structurally orthogonal to all 13 pipeline strategies and all 23+ session failures: it is NOT a moving-average crossover, NOT a fixed-percentage breakout, NOT a volume-based signal, NOT a pullback to a moving average, NOT a 3-day count reversal, NOT a multi-timeframe trend filter, NOT a calendar effect, NOT a death-cross short, NOT cross-asset, NOT supplementary-data-dependent (the indicator is built entirely from OHLC), NOT accumulation. The mechanism class — synthetic-volatility capitulation recovery — has NOT been tested in this session.

Faithful implementation of the hypothesis's synthetic-volatility capitulation-recovery mechanism, orthogonal to the pipeline's MA-cross / breakout / volume / supplementary-data strategies. VIX-Fix is computed purely from OHLC (highest-high variant per the hypothesis formula). The core is a panic->stabilization two-stage crossing encoded as a _panic_active latch: stage 1 arms when VIXFIX exceeds the rolling upper-percentile (90th over a 90-day window), stage 2 fires a LONG when it crosses back below the lower-percentile (80th), so the entry is a recovery signal that only triggers AFTER a genuine panic and fires once per episode. Rolling percentiles use linear interpolation over the trailing window. calculate_signal returns a raw decision variable: the explicit stabilization cross returns >= entry_threshold (1.0, plus a graded strength term from prior panic depth), while every other bar returns 0.9*tanh(vixfix/20) which is continuous, varies bar-to-bar, and stays strictly in (0, 0.9) below the threshold — satisfying the Layer-2 frozen-signal guard while making it impossible for a non-cross bar to leak an entry. Helper functions (_percentile, _vixfix) are module-level rather than @staticmethod to avoid the known Layer-2 proxy self-binding crash. Exits capture the documented 4-8% / 5-10 day bounce (tp/stop/time) with an optional re-panic exit for protection. Venue is BINANCE USD-M futures per the hypothesis's clean 5y+ BNB daily data path; the strategy is long-only and single-position, so a MARGIN account keeps get_account_equity() constant for correct 50%-of-equity sizing, with negligible funding over short holds, and leverage=1.0 is wired into sizing so it is not a set-but-unused field.

Failed deflated Sharpe and every forward gate. DSR=0.0007 (vs 0.95 bar) with the optimized Sharpe 2.38 FAR BELOW the expected-max best-of-225 luck bar of 5.99 (is_significant=FALSE, sharpe_ci_low=-0.036 straddling 0) — after multiple-testing correction the result is indistinguishable from selection noise. The high probabilistic_sharpe (0.9823) is the classic PSR-vs-DSR trap: PSR ignores the 225-trial count that DSR corrects for. Two independent forward gates concur: the HOLDOUT is NEGATIVE (holdout_sharpe -2.113, ratio 0) and the walk-forward is is_overfitted=TRUE with avg IS 5.59 collapsing to avg OOS -0.42 (windows [-5.67, +3.87, +0.55], one catastrophic). Sensitivity FAILED with 5 cliffs on the core signal parameters, and the optimizer's 'best' config is degenerate — it pushed upper_pctile to 1.0035 and lower_pctile to 1.167 (>1.0, clamped to the max), effectively disabling the panic-ARM stage that defines the hypothesis. Not iterate (attempt 1 of 2): the base config is a marginal, non-significant edge (Sharpe 0.88, CI low -1.72, negative alpha, IR -0.77), and a 225-trial sweep could only inflate in-sample Sharpe while OOS went negative — there is no robust region to tune toward, so a second sweep re-finds the same overfit/degenerate fit. Not revise_hypothesis: the VIX-Fix capitulation-recovery mechanism has no promoted sibling; it is a weak single-asset edge that fails deflation, not a proven mechanism stranded on a dead target. FAILURE PATTERN: single-asset daily VIX-Fix panic->stabilization recovery on BNB produces a modest, non-significant base edge whose best-of-225 optimization overfits catastrophically (avg OOS -0.42, negative holdout -2.11) and whose 'winning' params degenerate to >1.0 percentile thresholds that disable the hypothesized ARM stage — a high PSR (0.98) with DSR ~0 and a Sharpe below the luck bar (5.99) is decisive evidence of best-of-N inflation, not a real capitulation-recovery edge.

Long-only daily reversal on BNBUSDT.BINANCE perpetual futures using Larry Williams' VIX-Fix, an OHLC-only synthetic panic gauge: VIXFIX = 100*(22-bar highest high - low)/highest high. A two-stage state machine arms when VIX-Fix spikes above its rolling 90th-percentile (panic) and enters LONG when it falls back below the 80th-percentile (stabilization). Exits on a 6% take-profit, 5% stop, a 10-bar time stop, or a fresh VIX-Fix panic spike. No funding/taker/supplementary data; single instrument, leverage 1.0, 50%-of-equity sizing.

Genuine base edge: Sharpe 0.88, profit_factor 1.23 (>1.2 bar), positive expectancy (+$406/trade, avg_win 4.4% > avg_loss 3.4%)

Mechanism correctly implemented and trades match the hypothesis — two-stage panic->stabilization latch fires on real capitulation episodes, capitulation-recovery not knife-catching

Broadly consistent across regimes (5 up years / 2 modest down years), max drawdown a tolerable 17.4%, not one-year-carried

Low fee/impact drag (commission 2.7% of gross, impact 4.9%) and deployable capacity ($41.7M); 9.2% exposure = capital-efficient episodic edge

Structurally distinct mechanism class (synthetic-volatility capitulation) not previously tested this session

43 trades over 6 years is the low end for walk-forward (~14/window) — overfitting risk if the optimizer over-selects on a thin sample

Alpha slightly negative (-0.012) and recent regime soft (2026 -4.1%, latest rolling Sharpe -0.19) — watch the holdout for decay

Sharpe CI straddles 0 (-1.72 to 3.06), so significance must be earned in optimization/deflation, not assumed

OHLC-only mechanism with no dependence on falsified supplementary feeds; low exposure (~8-9%) leaves capacity

Base config has positive skew (1.29) and a coherent two-stage panic->stabilization latch

PBO 0.4166 is below 0.5 — the one robustness sub-metric that does not outright fail

Failed deflated Sharpe: DSR=0.0007 vs 0.95 bar; optimized Sharpe 2.38 is FAR BELOW the expected-max luck bar 5.99 over 225 trials (indistinguishable from best-of-N noise)

is_significant=FALSE and sharpe_ci_low=-0.036 (bootstrap CI straddles 0)

Holdout FAILED with a NEGATIVE Sharpe -2.113 (ratio 0) — the untouched recent window loses money

Walk-forward is_overfitted=TRUE: avg IS 5.59 collapses to avg OOS -0.42, with one window at -5.67

Sensitivity FAILED: 5 cliff parameters (stop_pct, lower_pctile, vix_lookback, max_hold_bars, pctile_window)

Optimizer selected degenerate percentile params >1.0 (upper 1.0035 / lower 1.167) that clamp to the max, effectively disabling the hypothesized panic ARM stage

Base config is a marginal non-significant edge (Sharpe 0.88, CI low -1.72, negative alpha -0.012, IR -0.77)

This strategy implemented Larry Williams' options-free VIX-Fix panic gauge as a two-stage capitulation-recovery long on daily BNB futures, relying only on OHLC so it sidestepped the session's falsified supplementary feeds. Its initial backtest was coherent but marginal — Sharpe 0.88, profit factor 1.23, positive expectancy across 43 trades — enough to clear the backtest-review gate with an 'optimize' verdict. The 225-trial optimization then inflated in-sample Sharpe to 2.38 while every forward gate collapsed: a negative holdout (-2.11), overfitted walk-forward (avg OOS -0.42), five sensitivity cliffs, a failed deflated Sharpe, and degenerate winning parameters that clamped above 1.0 and disabled the very panic-arm stage central to the thesis. The analyst abandoned it on attempt 1 as a non-significant single-asset edge whose optimization was indistinguishable from best-of-N noise.

A high probabilistic Sharpe (0.98) paired with a near-zero deflated Sharpe and an optimized Sharpe below the best-of-N luck bar is decisive evidence of selection-noise inflation, not a real edge — especially on a thin ~43-trade sample.

The analyst abandoned it at the post-optimization ANALYZING stage: it failed the deflated Sharpe gate (DSR 0.0007 vs 0.95 bar, optimized Sharpe 2.38 far below the best-of-225 luck bar of 5.99), the holdout was negative, walk-forward was flagged overfitted, sensitivity showed 5 cliffs, and the 'best' params degenerated to >1.0 percentile thresholds that disabled the panic-arm stage.

A long-only, OHLC-only Williams VIX-Fix capitulation-recovery strategy on daily BNBUSDT Binance USD-M futures, using a two-stage panic-arm then stabilization-fire latch to buy bounces after capitulation episodes.

The base backtest was a marginal edge: Sharpe 0.88, total return 16.2%, profit factor 1.23, 48.8% win rate over 43 trades with +$406/trade expectancy and 17.4% max drawdown. Optimization raised in-sample Sharpe to 2.38, but the untouched holdout printed a negative Sharpe of -2.11 and walk-forward out-of-sample averaged -0.42.