BtcFourHourVolExpansionCloseReversalLongShort

BTC 4H Volatility Expansion + Close-Position Reversal Long-Short (BINANCE USD-M Futures, 4H, OHLCV-Only, Price-Action Mechanism)

Single-instrument long-short price-action contrarian strategy on BTCUSDT.BINANCE. Mechanism reads INSIDE each bar's structure rather than across bars: when a single 4-hour bar exhibits unusually large range (volatility expansion) AND closes near one extreme of its own range, this signals exhaustion of the directional move that produced the expansion — sellers have absorbed buying pressure at the top, or buyers have absorbed selling pressure at the bottom. The closing-position-within-range encodes WHO won the last trades of the bar in a way that volume alone cannot. Trade contrarian to the implied exhaustion direction on the next bar's open. DELIBERATE DATA-SAFETY DESIGN: my last THREE hypotheses (ETH/BTC Pairs, SOL HL Funding, ETH Spot-Perp Carry) all failed at Layer 3 with `data_unavailable: refill pending`, and the data appears to fail most often on 6H/8H bar types AND on supplementary streams (funding_rate, taker_ratio). This hypothesis uses ONLY BTCUSDT.BINANCE-4-HOUR (the proven path of landed BtcFourHourVolumeBreakoutLong) with ZERO supplementary data — purely OHLCV. Three under-represented portfolio buckets filled: (1) long-short direction (currently 9.8% vs ≤55% long-only ceiling — most-violated quota); (2) price-action mechanism class (every prior hypothesis used a derivative data signal — funding, OI, taker ratio, IV, liquidations, ratio spreads — this is pure OHLCV bar-structure reading, fundamentally different signal class); (3) intra-bar microstructure interpretation (reading close-position-in-range is a candle-pattern technique not yet represented in the portfolio's volume-breakout / trend / momentum / mean-reversion strategies). Expected ~100-200 entry cycles per year × 6 years of BTC 4H data = 600-1200 trades, dramatically above the walk-forward sample-size floor (ADA failed at 62). The mechanism is BTC-specific (calibrated to BTC 4H typical range distribution), structurally immune to AVAX-style cross-asset analog overfitting since there's no claim of family-portability.

Implements the hypothesis directly: a pure intra-bar microstructure read (volatility expansion x close-position-in-range) that is a distinct signal class from the portfolio's volume-breakout, trend, momentum, and derivative-data (funding/OI/taker/IV/liquidation) strategies, and fills the long-short and price-action buckets. The signal is continuous and varying — a graded contrarian magnitude scaling close-extremeness by the expansion ratio — while the precise entry gates live in should_enter via _armed_side, satisfying the 'continuous signal, gated decision' rule; the contrarian sign convention is explicit (close-near-high => SHORT, close-near-low => LONG). The cp zero-range guard returns 0.5 so flat bars never produce NaN. Data-safety is honored by using ONLY the proven BTCUSDT.BINANCE-4-HOUR OHLCV path with zero supplementary streams and zero extra instruments, avoiding the data_unavailable failures seen on 6H/8H and supplementary feeds. Exit economics (1.2% TP vs ~0.10% round-trip futures fee) clear costs, and the tight stop + 4-bar time stop bound the risk of fading a bar whose move keeps extending. Routed to BINANCE USD-M futures because the strategy is genuinely long+short (impossible on a spot CASH account); leverage stays 1.0 since 15%-per-direction sizing uses no margin amplification, avoiding the leverage_set_but_unused gate.

Dead premise with no edge -- not a code or sparsity problem. The 'fade volatility-expansion bars that close at an extreme' contrarian strategy traded correctly as specified (901 entries, 426L/475S on BTC 4H) but LOSES money: profit_factor 0.796 (<1), Sharpe -1.41, Sortino -1.79, total_return -23.66%, expectancy negative, and negative in 5 of 7 years. It shows the inverted fade payoff (win_rate 0.495 with avg_win $202 < avg_loss $249): a wide bar closing near its high/low more often CONTINUES than reverses, so the fade captures small reversions and gives them back on continuation; both legs lose (long_win 0.491 / short_win 0.499). The Sharpe CI is entirely negative (-2.38 to -0.51) with probabilistic_sharpe 0.0018, so the true Sharpe is near-certainly negative, and commission is 13.7% of gross (the 901-trade frequency makes fees dominant). The parameters (range_mult, upper/lower_cp, expansion_cap, tp/stop) cannot convert a sub-1 profit factor that loses every year into an edge. Not iterate (code correct, matches the hypothesis; the premise is the flaw). FAILURE PATTERN: intra-bar close-position contrarian fades of volatility-expansion bars on a crypto major have no edge -- this is the identical no-edge signature as the just-abandoned streak-exhaustion long-short (same developer, same long-short-quota motivation, same BTC 4H contrarian fade) and the broader Keltner/RSI(2)/Bollinger/SUI-reversal fade family: directional expansion persists rather than mean-reverts, yielding avg_win < avg_loss, sub-1 profit factor, and net-negative returns across every regime and both legs. The long-short quota is real but cannot be filled with a money-losing fade; close-position-in-range carries no contrarian predictive information on BTC 4H.

Single-instrument long-short price-action contrarian strategy on BTCUSDT.BINANCE USD-M futures 4H bars, OHLCV-only. On each closed bar it measures volatility expansion (bar range vs the trailing 20-bar average range) and the close's position within the bar's own range, cp = (close-low)/(high-low). When a bar's range is an expansion (>= 1.6x average) AND it closes near an extreme (cp >= 0.75 near the high, or cp <= 0.25 near the low), the directional move is treated as exhausted: it SHORTs after a wide bar closing near its high and LONGs after a wide bar closing near its low. Exits on a 1.2% take-profit, 1.5% stop-loss, or a 4-bar (~16h) time stop. calculate_signal returns a continuous contrarian score (-(cp-0.5)*2 * expansion, positive=>long, negative=>short) that varies every bar; the binary expansion/extreme gates are captured in an _armed_side flag checked by should_enter. Sizes 15% of equity per direction, capped by 1.5% risk over the stop.

Implemented correctly and trades as specified: 901 contrarian fades (426 long / 475 short), clean entry_diag (901 signaled = 901 submitted, 0 dropped), metrics_reliable=true, large sample; addresses the long-short quota gap

No edge / money-losing: profit_factor 0.796 (<1), Sharpe -1.41, Sortino -1.79, total_return -23.66%, expectancy -25.6 (negative)

Inverted fade payoff: win_rate 0.495 with avg_win $202 < avg_loss $249 -- expansion bars continue rather than revert; both legs lose (long_win 0.491 / short_win 0.499)

Negative in 5 of 7 years; Sharpe CI entirely negative (-2.38 to -0.51), probabilistic_sharpe 0.0018; commission 13.7% of gross (fee-dominated at 901 trades)

~0.5+ (pre-opt)

~62 WF floor

2 of 7 (trivially)

majority

negative (-25.6)

BtcFourHourVolExpansionCloseReversalLongShort was a carefully engineered attempt — using only the proven BTC 4H OHLCV path to sidestep the Layer-3 data failures that killed the developer's prior three ideas, and reading intra-bar close-position as a fresh signal class to fill the long-short quota. The data engineering worked (901 clean trades), but the premise was backwards: wide bars closing at an extreme continued more than they reversed, producing the fade family's inverted payoff (avg_win $202 < avg_loss $249), profit_factor 0.796, Sharpe −1.41, and a −23.66% return negative in 5 of 7 years, with fees consuming 13.7% of gross at this frequency. The analyst abandoned it at backtest review without optimizing — there is no parameter that turns a sub-1 profit factor losing every year and both legs into an edge. Arriving right after the streak-exhaustion fade from the same developer, it reconfirms that BTC 4H contrarian fades of directional intensity — whether read across bars or inside one — simply have no edge, no matter how novel the lens or how clean the data path.

Intra-bar close-position contrarian fades of volatility-expansion bars on a crypto major carry no predictive information — like the streak-exhaustion sibling and the broader Keltner/RSI(2)/Bollinger/SUI fade family, directional expansion persists rather than mean-reverts, yielding avg_win < avg_loss, sub-1 profit factor, and losses across every regime and both legs — and sound data-safety engineering plus a real quota gap cannot rescue a premise the market disproves, especially when high trade frequency makes fees dominant.

The analyst issued an abandon verdict at backtest review on a dead-premise basis: a wide BTC 4H bar closing near its high/low more often continues than reverses, so the fade shows the inverted payoff, a sub-1 profit factor, and net-negative returns in every regime and both legs; the Sharpe CI is entirely negative so the true Sharpe is near-certainly negative, the 901-trade frequency makes fees dominant (13.7% of gross), and no parameter set can convert a money-losing fade into an edge — the code matches the hypothesis, so the premise is the flaw and there is nothing to iterate.

A single-instrument, OHLCV-only long-short price-action contrarian strategy on BTCUSDT.BINANCE 4H that reads intra-bar structure: when a bar's range is an unusual expansion (≥1.6× its trailing 20-bar average) AND it closes near one extreme (close-position ≥0.75 or ≤0.25), treat the move as exhausted and fade it — short after a wide bar closing near its high, long after one closing near its low — with a tight take-profit, stop, and ~16h time stop, deliberately using only the proven 4H OHLCV path to avoid prior data-availability failures and to fill the long-short quota.

It implemented and traded correctly with a large sample — 901 entries (426 long / 475 short), clean entry diagnostics, metrics_reliable=true — but had no edge and lost money: profit_factor 0.796, Sharpe −1.41, Sortino −1.79, total_return −23.66%, expectancy −25.6, max drawdown 26.1%, negative in 5 of 7 years. The payoff was inverted (win_rate 0.495, avg_win $202 < avg_loss $249), both legs lost (long_win 0.491 / short_win 0.499), the Sharpe CI was entirely negative (−2.38 to −0.51) with probabilistic_sharpe 0.0018, and commission was 13.7% of gross.