OptionsImpliedFearExtremeReversalLS

BTC/ETH Options-Implied Fear-Extreme Reversal, Long-Short (Two-Instrument BINANCE USD-M Perps Driven by Each Coin's DERIBIT ATM Implied-Vol vs Realized-Vol Extreme — Buy Capitulation When Options Price Panic, Fade Complacency, 3-Parameter)

A LONG-SHORT strategy that trades BTCUSDT and ETHUSDT USD-M perps on BINANCE, but whose SIGNAL comes from the DERIBIT options market: the ratio of each coin's at-the-money implied volatility to its own trailing realized volatility (IV/RV). This is a cross-market sentiment gauge — the crypto analogue of the VIX 'fear' signal — and it is NOT an options TRADE, so it sidesteps the structural <100-trade death that has killed every put-write/straddle/condor in this factory (we trade the liquid perp continuously, not monthly-expiry options). The ATM-IV series is proven-available and usable in this factory (the surviving Deribit BTC put-write is explicitly 'IV>RV gated'), which de-risks the data-assembly wall that killed the long-straddle attempt. Mechanism: when options suddenly price extreme PANIC (IV spikes far above realized vol) while price is in a drawdown, that is capitulation — go LONG the perp and ride the recovery; when options price extreme COMPLACENCY (IV collapses below its range) while price is stretched up, the market is under-hedged and fragile — fade SHORT. Distinct from every dead/in-pipeline family: NOT options premium harvest, NOT funding-carry (L33), NOT OI/premium/L-S/taker crowding fades (different feed — options-implied vol), NOT liquidation-feed dependent (multi-year Deribit IV, no 90-day wall), NOT breakout/trend, NOT cross-sectional rotation. Uses two independent per-coin signals (not a relative ranking, so L32 does not apply) to reach a measurable trade count.

Implements the options-implied fear-extreme edge with IV/RV as the cross-market sentiment discriminator: a spike in implied vol relative to realized vol during a drawdown is the options market pricing panic at a low (capitulation -> buy the perp and ride the recovery), while IV collapsing below its range during a stretched rally is complacency/under-hedging (fragile -> fade short). This is distinct from every dead/in-pipeline family: it is NOT an options trade (so no monthly-expiry <100-trade wall -- we trade the liquid perp continuously), NOT premium harvest, NOT funding-carry, NOT OI/premium/L-S/taker crowding (different feed -- options-implied vol), NOT liquidation-fed (multi-year Deribit DVOL, no 90-day wall), and NOT breakout/rotation. Critical data decision: deribit_dvol is a flat {ns:iv} series with no coin field, so the supp loader reliably injects only the primary underlying's DVOL; rather than risk a multi-leg supp wall (the failure mode that killed prior baskets), I use BTC DVOL as the market-wide crypto fear index (ETH IV is highly correlated) and differentiate the two legs by each coin's OWN realized vol and price drawdown/stretch state -- yielding two genuinely independent per-coin signals (absolute triggers, not a relative ranking, so L32 does not apply) from one reliably-loaded feed, which is exactly how the surviving Deribit put-write consumes the IV series. Per lesson L5 the trigger is a rolling z-score of IV/RV (scale-invariant), not a fixed constant, so it fires at a measurable rate across vol regimes on two coins x 4h bars. The DVOL gate is mandatory (missing IV -> no trade) so the strategy always tests the options-fear hypothesis rather than degrading to price reversion, and there is no look-ahead (DVOL read at/before the bar close, nearest-preceding with a staleness guard; DVOL values normalized from percent to fraction). Two instruments are traded via an on_bar override (each coin acts on its own bar, no cross-leg alignment lag) using _submit_entry_instrument/_submit_exit_position; the base's per-instrument hooks are inert stubs as in the proven multi-instrument pattern. Every trade is hard-bracketed (symmetric ATR target/stop) with conservative per-coin sizing (1% risk, 0.4x-gross cap each so combined <= 0.8x, leverage 1.0 with sizing that never references leverage) so a stop-out costs a small fraction of equity and cannot liquidate, and shorts are legal on the perp. Fee headroom: a 2-ATR move on 4h BTC/ETH is typically 2-5%, an order of magnitude above the ~0.10% USD-M round-trip, with multi-day holds keeping turnover low. Kept to exactly 3 tuned parameters (ivrv_z, price_move, exit_atr) to resist overfitting; the vol windows, ATR period, hold cap, and risk caps are fixed structural choices." }

No fee-clearing edge: avg_trade_return_pct 0.071% is below the 0.15% futures fee floor (L22, disqualifying at review on its own) and the dollar-weighted expectancy is negative (-$26.5/trade). On 291 well-sampled trades the result is a net loser — profit_factor 0.947, Sharpe -0.157 (CI [-0.82, 0.44]), total_return -9.2%, PSR 0.31 — with inverted reward/risk (53.6% win rate but avg_loss $1,088 > avg_win $892). This is the classic fee_edge signature: gross edge is ~0 and costs push it negative, so the IV/RV-timed reversal has no directional edge on BTC/ETH perps at this configuration (net-negative across the window; 2024 -22.6% dominates the scattered positive years). Optimizing ivrv_z/price_move/exit_atr cannot lift a 0.07%-per-trade, PF-0.95 mechanism above the ~0.10% round-trip — it would only overfit the few favorable windows (e.g. the 2026 uptick). The options-fear feed is a genuinely novel signal, but the measured edge is a below-fee-floor net loser with no visible parameter region that clears costs. Abandon rather than spend 2 hours optimizing.

Long-short reversal trading BTCUSDT and ETHUSDT USD-M perps on 4h bars, timed by a cross-market options-fear gauge. For each coin it forms IV/RV = Deribit DVOL (30-day ATM implied-vol index, the crypto VIX) divided by that coin's own trailing realized vol, and z-scores it over a rolling 60-bar window. FEAR/CAPITULATION (ratio z >= ivrv_z=1.5 while the coin is in a drawdown, ret over ~2 days <= -price_move=3%) -> LONG the perp to ride the recovery; COMPLACENCY (ratio z <= -1.5 while stretched up, ret >= +3%) -> SHORT. Each trade uses a symmetric ATR bracket (target and stop exit_atr=2.0 x ATR from entry) plus a 30-bar (~5-day) time cap. DVOL is shared from BTC (the market-wide fear index, reliably loaded) while each coin supplies its own realized vol and price state, giving two independent signals. Per-coin sizing risks 1% of equity over the ATR stop, capped at 0.4x equity each (<=0.8x combined, leverage 1.0). 3 tuned params (ivrv_z, price_move, exit_atr).

Acceptable given the data constraint. If coin-keyed DVOL becomes reliably loadable, select the symbol's own IV series in _dvol_at.

Title/hypothesis says the signal uses 'EACH coin's DERIBIT ATM implied-vol', but the code uses the PRIMARY (BTC) DVOL as a shared IV series for BOTH legs (_signal_ccy derives from the primary=BTC; _dvol_at reads one _dv_ts/_dv_val series for every symbol). The ETH leg thus uses BTC IV / ETH RV, not ETH's own IV. This is explicitly disclosed and justified in the rationale (flat {ns:iv} series has no coin field; only primary underlying reliably injected; ETH IV highly correlated), so it is an intentional design choice, not a mislink. Warning only: ETH's IV is a BTC proxy, but each coin still supplies its own RV and price state, keeping the two signals distinct.

At backtest_review, verify total_trades over the full DVOL-covered window >= ~100; if DVOL history limits it well below that, treat as unmeasurable rather than a confirmed edge.

Sandbox produced only 39 total_trades over 363 days across TWO instruments (~20/coin/yr), below the ~100-trade measurability floor (L16). Trade count is hard-capped by DVOL coverage since the IV gate is mandatory. Not a code defect (metrics_reliable=true, Sharpe 1.32, PF 2.12, win 69%, avg_trade_return_pct 1.61%) and should scale on a multi-year window, so it does not block — but analyst must confirm full-window count clears the floor.

Novel cross-market signal (Deribit IV/RV) that loads and fires; mechanically clean (no liquidation, avg_position 24%, realized headline, benchmark_meaningful=false correctly flags the LS structure).

Well-sampled at 291 trades (117 long / 174 short), so the negative result is decisive not noise.

avg_trade_return_pct 0.071% is below the 0.15% futures fee floor and dollar-weighted expectancy is negative (-$26.5/trade) — no fee-clearing edge.

Net loser: profit_factor 0.947, Sharpe -0.157 (CI [-0.82, 0.44]), total_return -9.2%, PSR 0.31.

Inverted reward/risk: 53.6% win rate but avg_loss $1,088 > avg_win $892.

Net-negative across the full window; 2024 -22.6% dominates the scattered positive years.

Classic fee_edge signature — gross edge ~0, costs push it negative; single-name/pair reversal-fade class.

This strategy paired a genuinely novel cross-market signal — the Deribit implied-vs-realized volatility extreme as a crypto 'fear' gauge — with continuous long-short perp trading on BTC and ETH, deliberately sidestepping the sub-100-trade wall that had killed every options trade in the factory. Mechanically it worked cleanly and fired 291 well-sampled trades, but the measured edge was a below-fee-floor net loser: 0.071% per trade, profit factor 0.947, Sharpe -0.157, and a -9.2% total return with losses bigger than wins. The analyst recognized the fee_edge signature — gross edge roughly zero, costs tipping it negative — and abandoned it at the backtest-review gate rather than spend hours optimizing three parameters that could not lift the mechanism above trading costs. It never advanced to optimization, risk review, or paper trading.

A genuinely novel signal (Deribit options-fear IV/RV) is not the same as a tradable edge — if per-trade return sits below the fee floor with a sub-1.0 profit factor, tuning the parameters only overfits the few favorable windows rather than manufacturing edge that clears costs.

The analyst issued an 'abandon' verdict at the pre-optimization BACKTEST_REVIEW gate — the classic fee_edge signature: gross edge near zero pushed net-negative by costs, with per-trade return below the fee floor and no visible parameter region that could clear the ~0.10% round-trip, so it never reached optimization or analysis.

A long-short strategy trading BTCUSDT and ETHUSDT USD-M perps on Binance whose signal came from the Deribit options market — using each coin's implied-vs-realized volatility (IV/RV) extreme to buy capitulation when options priced panic during a drawdown and short complacency when IV collapsed during a stretched-up move.

Over 2,399 data days and 291 well-sampled trades (117 long / 174 short) it was a net loser: avg_trade_return_pct just 0.071% (below the 0.15% futures fee floor), expectancy -$26.5/trade, profit_factor 0.947, Sharpe -0.157 (CI [-0.82, 0.44]), total_return -9.2%, with inverted reward/risk (53.6% win rate but avg_loss $1,088 > avg_win $892) and 2024 (-22.6%) dominating the scattered positive years.