SolPremiumIndexExtremeMeanReversionLS4H

SOL 4H Premium Index Extreme Mean Reversion (Long & Short)

A long-short single-instrument microstructure mean-reversion strategy on SOLUSDT perpetual futures using 4-hour bars. The strategy uses the perpetual premium index — the spread between mark price and the underlying spot index — as the sole entry signal. When the premium becomes extremely negative (perp trades meaningfully below spot), longs are entered; when extremely positive (perp trades above spot), shorts are entered. The thesis is microstructure-grounded: funding rate mechanism + cross-venue arbitrage flows reliably pull the premium back toward zero on a multi-hour horizon, so extreme premiums offer a high-probability mean-reversion trade with the funding rate itself acting as a tailwind on the long side. Strategy is intentionally simple (3 core parameters: long_threshold, short_threshold, exit_threshold) and uses only OHLCV + premium_index data — no funding rate, no L/S ratios, no liquidations, no pairs. Distinct from existing portfolio: SolDailyAdxDonchianBreakoutLong is daily trend-following, SolAvaxPairsSpreadMeanReversion is cross-instrument; this is intra-instrument microstructure reversion.

Implements the hypothesis exactly: the premium index is a direct microstructure observable, and the funding mechanism plus cross-venue arbitrage flows reliably pull it back toward zero on a multi-hour horizon, so extreme premiums are a high-probability mean-reversion setup — with negative premium also implying negative funding, paying the long to wait. Venue is the BINANCE SOLUSDT perp (MARGIN) because the strategy is genuinely long AND short (a spot CASH account cannot short the positive-premium leg), the premium_index data is itself perp microstructure, and the engine accrues the funding tailwind on held positions; gross exposure is ~0.5x equity at leverage 1.0, so no leverage is needed and the unused-leverage gate is avoided. The three core parameters are exactly long_threshold, short_threshold, exit_threshold as specified; stop-loss and max-hold are safety nets, not signal logic. I confirmed on the REAL SOLUSDT premium_index parquet that coverage is complete and continuous (339,916 klines at 15-min from 2020-09 to 2026-03, zero >48h holes — unlike sparser supp series), and that over 5.5 years of 4H bars the signal fires 199 long-trigger and 239 short-trigger bars (~40-70 round-trips/year, balanced both-sided), giving ~15-30 OOS trades per walk-forward window — well above the 5-trade floor. The window-mean over the 4H bar smooths 15-min premium noise into a representative bar-level premium. Verified locally: Layer 1 static (clean, divisions guarded), Layer 2 synthetic (all 6 scenarios pass, finite signals), and the real-data trigger-frequency check above.

The SOL premium-index extreme mean-reversion LS is decisively unprofitable: Sharpe -0.77, profit factor 0.82, negative expectancy (-$256/trade), total return -59.9%, max drawdown 74.8%, Sortino -1.09, omega 0.84, with a 38% win rate and a 1863-day (never-recovering) drawdown. This is NOT a data-coverage issue — premium_index data is present across the full 2020-2026 window (annual returns exist for every year); the strategy simply rarely triggers (130 trades, exposure 6.9%) and loses when it does, dominated by 2021 (-53.6%). The mechanism is falsified: on a 4H horizon, perp premium-index extremes coincide with violent directional flow rather than reversion opportunities — the perp trades at a premium during pumps (shorting it is run over) and at a discount during crashes (longing it is run over), so fading the premium is effectively fading momentum and bleeds. Optimizing the three thresholds over 225 trials cannot flip a negative-expectancy mechanism into a promotable one (best-of-N over a losing signal would fail holdout). This is the 'results so poor no optimization can help' case, consistent with the session's repeated failure of microstructure/MR fades. Not revise_hypothesis: the premium-MR mechanism has no proven sibling to redirect to and the failure is absence of edge, not a structurally-dead target for a working mechanism. FAILURE PATTERN: intra-instrument perp premium-index extreme mean-reversion on a 4H horizon has negative edge — premium extremes mark active directional flow, so fading them is dominated by momentum continuation, producing PF 0.82, negative Sharpe, and a 2021-concentrated catastrophic drawdown.

Long-short single-instrument microstructure mean-reversion on SOLUSDT perpetual futures (4H bars). The sole entry signal is the perpetual premium index (mark vs spot index, a small fraction) read from the premium_index supplementary klines and averaged over the trailing 4H bar window. When the window-mean premium is extremely negative (perp trades below spot, <= long_threshold = -0.0015) it enters LONG; when extremely positive (>= short_threshold = +0.0015) it enters SHORT. It exits when the premium reverts into a tight band around zero (long exits once premium >= -exit_threshold, short exits once premium <= +exit_threshold, exit_threshold = 0.0003), with a catastrophic stop-loss (4%) and a 12-bar (2-day) max-hold as safety nets. calculate_signal returns the continuous window-mean premium; the thresholds are applied in should_enter/should_exit. A 24h staleness guard makes the strategy go inert where premium data is missing rather than trading blind. Single instrument, long+short, futures venue, leverage 1.0, ~0.5x equity notional per position.

Correct no-fallback design and clean two-sided execution (58 long / 72 short); premium_index data present across the full 2020-2026 window

Genuinely novel intra-instrument microstructure signal class

Decisively unprofitable: Sharpe -0.77, profit factor 0.82, negative expectancy (-$256/trade), total return -59.9%, max DD 74.8%

Mechanism falsified: fading 4H premium-index extremes = fading directional flow during volatile moves, which gets run over (2021 -53.6% dominates)

38% win rate; avg_loss frequency overwhelms the larger avg_win; max_drawdown_duration 1863 days (never recovers)

Very low exposure (6.9%) and sparse triggers — even with more trades the per-event edge is negative

SolPremiumIndexExtremeMeanReversionLS4H tested a genuinely novel intra-instrument microstructure signal — fading perp premium-index extremes on SOL 4H — on the thesis that funding and cross-venue arbitrage pull the premium back to zero. The data was present and the no-fallback design clean, but the mechanism was falsified: premium extremes coincided with violent directional flow, so the fades were run over (PF 0.82, Sharpe -0.77, -59.9% total, 74.8% drawdown concentrated in 2021). The analyst judged this an absence of edge rather than a data or tuning issue — premium extremes are momentum, not reversion — and abandoned it at the backtest-review gate, consistent with the session's repeated microstructure/MR-fade failures. It ended after a single iteration as abandoned, never reaching optimization, analysis, or risk review.

Intra-instrument perp premium-index extreme mean-reversion on a 4H horizon has negative edge: the perp trades at a premium during pumps and a discount during crashes, so premium extremes coincide with violent directional flow rather than reversion opportunities, and fading them is dominated by momentum continuation — no threshold tuning flips a negative-expectancy fade positive.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as decisively unprofitable with a falsified mechanism — premium-index extremes mark active directional flow, so fading them is effectively fading momentum and gets run over — and with premium_index data present across the full window (not a coverage issue), optimization and all later stages were never reached.

A long-short single-instrument microstructure mean-reversion strategy on SOLUSDT 4-hour perpetual futures (OHLCV + premium_index data) that used the perp premium index (mark vs spot spread) as its sole signal — going long when the premium was extremely negative (≤-0.0015) and short when extremely positive (≥0.0015), exiting when it reverted toward zero (|premium|≤0.0003), with a 4% stop and 12-bar max hold.

Over ~5.3 years (2020-09-14 to 2026-05-17) it took 130 trades (58 long / 72 short) with a 38% win rate and a decisively negative edge: total return -59.9%, CAGR -7.4%, Sharpe -0.77, profit factor 0.82, expectancy -$256/trade, a 74.8% max drawdown lasting ~1863 days, dominated by a -53.6% 2021, at very low 6.9% exposure.