SolAvaxL1RelativeValueConvergencePairsLS

SOL/AVAX Same-Sector L1 Relative-Value Convergence, Market-Neutral Pairs (Long the Laggard / Short the Leader When the Daily Log-Ratio Z-Score Deviates ≥2σ — SLOW, WIDE-Threshold, Held for WEEKS to Amortize the 4-Fill Cost that Killed the 4h ETH/BTC Version — Hard Cointegration-Break Stop, 2-Parameter)

A MARKET-NEUTRAL (dollar-neutral) PAIRS convergence trade on two same-sector high-throughput L1 perps — SOLUSDT.BINANCE (primary) and AVAXUSDT.BINANCE (hedge, USD-M) — that deliberately fixes the two reasons the deployed-graveyard ETH/BTC ratio reversion died fee_edge: it trades on DAILY bars at a WIDE ±2σ threshold and holds for WEEKS, so a single convergence captures multiple percent of the ratio against the ~0.20–0.40% four-fill round trip, instead of the 4h/2σ version that churned tiny reversions below cost. SOL and AVAX are chosen because they are genuine economic substitutes (both high-TPS smart-contract L1s driven by the same alt-L1 narrative, flows and beta), so their LOG PRICE RATIO mean-reverts around a slowly-drifting equilibrium far more reliably than an unrelated pair. When the ratio stretches wide (one name has run relative to the other on no fundamental divergence), we short the rich leg and long the cheap leg in equal notional — net crypto beta ≈ 0, so the trade profits from RELATIVE convergence regardless of whether the whole market is up or down. This directly targets the direction quota (long_only is 85% over-represented) and the market-neutral bucket (2% — the single most under-represented by survival). This is NOT the 4h ETH/BTC ratio family (different pair, daily horizon, wider threshold, hard cointegration-break stop), NOT a rank/rotation dollar-neutral BASKET (L12 — this is a single 2-leg pair, not an N-name rebalanced sort), NOT carry/basis (dead), NOT options (L23), NOT liquidation/funding-directional (L3/L13). 2 core parameters (ratio lookback N, entry Z).

Implements the hypothesis exactly and fixes the two failure modes of the dead 4h ETH/BTC ratio family: (1) horizon — DAILY bars with a WIDE +/-2sigma entry and weeks-long holds, so one convergence captures multiple percent of the ratio against the ~0.20-0.40% four-fill round trip, instead of churning sub-cost 4h reversions; (2) pair quality — SOL and AVAX are genuine economic substitutes (same alt-L1 narrative/flows/beta), so ln(SOL/AVAX) mean-reverts around a slow equilibrium. calculate_signal returns the RAW z-score in its natural units (not clipped to [-1,1]), and should_enter thresholds it against the matching-unit entry_z, per the signal-range guidance — the value varies every bar. Market-neutral construction (equal-notional opposite legs) directly targets the under-represented long+short / market-neutral buckets. Data verified: both SOLUSDT and AVAXUSDT have continuous BINANCE daily history 2020-09 -> 2026-07, so the single extra leg cannot trigger data_unavailable and the two same-timeframe legs are always contemporaneous (no sync-barrier stall). Trade frequency checked on a mean-reverting AR(1) proxy: ~10 pair entries/year at these thresholds, so the 365-day sandbox clears the >=1-trade gate with margin and yields a measurable sample. _process_primary is overridden (not on_bar/on_start) to open/close both legs atomically while preserving the base alignment barrier — the same proven structure as prior 2-leg strategies; a single 2-leg pair (NOT an N-name rebalanced basket, L12/L24). leverage=2.0 is genuinely consumed: each leg is sized equity*per_leg_frac*self.config.leverage/price, so gross exposure scales with leverage while net delta stays ~0, satisfying the leverage-used gate. HARD cointegration-break stop at |z|>=3.5 caps the loss when the relationship genuinely breaks (per-trade risk control, L15). NOT carry/basis, NOT options, NOT liquidation/funding-directional.

Total account liquidation on 2023-11-10: total_return -100%, max_drawdown 100% (CI high 3007%). A market-neutral pairs trade should never liquidate, so the risk model failed structurally — avg_position_pct 90.8% stacked near-full gross exposure, and the stop_z=3.5 cointegration-break stop did not prevent a catastrophic loss (largest_loss -$537k, 2023-11-12 daily -24.95%). The root cause is a de-cointegration: during the Oct–Nov 2023 SOL rally, SOL structurally decoupled from AVAX, so ln(SOL/AVAX) stretched to +2σ and the strategy shorted rich SOL while it kept ripping — the 'equilibrium' the trade assumed had permanently shifted, which is a premise failure, not a mean-reverting stretch. Even before the blow-up the edge was marginal (profit_factor 1.12, Sharpe 0.34 with CI [-0.52, 1.28] straddling zero, only 76 trades). Optimizing ratio_lookback/entry_z cannot restore cointegration to a de-cointegrated pair or undo a 100% drawdown; tuning would only shift the liquidation date. Note for the research lead: the market-neutral pairs MECHANISM is a legitimately under-represented bucket and may be worth retrying on a still-cointegrated same-sector pair with far smaller per-leg sizing and a hard per-trade loss cap — but THIS pair (SOL/AVAX) has de-cointegrated and liquidated, so it is dead. Not worth 2 hours of optimization.

Market-neutral (dollar-neutral) pairs convergence on two same-sector high-throughput L1 perps: SOLUSDT.BINANCE (primary) long/short against AVAXUSDT.BINANCE (hedge), daily bars. The signal is the z-score of the daily log price ratio ln(SOL/AVAX) about its rolling N-day mean/std. When the ratio stretches to >= entry_z (default +2sigma, SOL rich vs AVAX) it SHORTs SOL and LONGs equal-notional AVAX; at <= -entry_z it LONGs SOL and SHORTs AVAX. Both legs equal USD notional so net crypto beta is ~0 and the trade profits from RELATIVE convergence regardless of overall market direction. It holds for weeks and exits on convergence (|z| <= exit_z=0.4), a hard cointegration-break stop (|z| >= stop_z=3.5), or a 45-day time stop. Alignment of the two legs is done by timestamp (not array index) so a gap in either series cannot desync the ratio; the base cross-leg sync barrier ensures the AVAX bar is contemporaneous with the SOL bar before a signal is acted on. Sizing is equity*per_leg_frac*leverage per leg. 2 core parameters: ratio_lookback N, entry_z.

Advance, but analyst/optimizer must weight the holdout heavily, confirm the edge is not carried by 1-2 convergences, and verify full-history pair count and per-window WF density.

Thin independent sample — the key risk in the encouraging result. total_trades 20 is ~10 distinct PAIR convergences over the 363-day sandbox; over the full ~6-year SOL/AVAX history only ~60 pair bets. Sharpe 0.80 has CI [-0.77, +2.21] straddling zero, so the +57.3% / profit_factor 1.46 is promising but NOT statistically established — could substantially reflect a favorable window (avg_win $16.5k, largest_win $35.5k, return_skew +1.65). Inherent to a daily/weeks-hold pairs trade, not a defect, but the walk-forward/holdout is where the edge must be confirmed and each WF window will contain very few trades.

Risk officer to confirm single-trade loss magnitude at 2x gross is acceptable and stop_z=3.5 is survivable; consider lower per_leg_frac/leverage for a breakable mean-reversion pair.

2x gross leverage amplifies the tail: per_leg_frac(0.5)*leverage(2.0)=1.0x notional per leg -> 2x gross. Net delta ~0 (dollar-neutral, correct), but gross amplifies single-trade PnL (kurtosis 15.6, largest_loss -$31.9k, cvar_95 7.9%). With stop at |z|>=3.5 and 2x gross, a genuine cointegration break can cost a large fraction of equity in one pair, though overall max_drawdown is a contained 18.7%.

Acceptable as-is; a true break-detector would be a separate parameter reflected in the hypothesis.

The hypothesis names a 'hard cointegration-break stop' but implements it as a z-score threshold (|z|>=3.5), not a statistical cointegration/half-life test. A |z|>=3.5 exit is a defensible proxy and keeps to 2 core params, so acceptable — noting only that the stop is deviation-based, not a true cointegration test.

Genuinely under-represented bucket (market-neutral pairs) with low benchmark correlation (0.04, beta 0.03) — the neutral construction itself works

Daily/wide-threshold/weeks-hold design correctly amortizes the 4-fill cost that killed the 4h ETH/BTC version

avg_trade_return_pct 2.53% clears the fee floor when trades win

ACCOUNT LIQUIDATED on 2023-11-10 — total_return -100%, max_drawdown 100% (CI high 3007%); a 'dollar-neutral' trade should never liquidate, so the risk model failed structurally

The SOL/AVAX ratio DE-COINTEGRATED in the 2023 SOL rally: shorting rich SOL while it kept ripping produced largest_loss -$537k; the stop_z=3.5 cointegration-break stop did not prevent the wipeout

avg_position_pct 90.8% — near-full gross exposure made a single de-cointegration event fatal; return_skew -3.3, kurtosis 143

Marginal edge even pre-blowup: profit_factor 1.12, Sharpe 0.34 with CI [-0.52, 1.28] straddling zero, only 76 trades

This strategy correctly diagnosed why the 4h ETH/BTC ratio reversion died on fees and fixed it structurally — daily bars, a wide 2σ threshold, and weeks-long holds on two genuine economic substitutes to amortize the four-fill cost — and the neutral construction itself worked (beta 0.03, correlation 0.04). But the core assumption failed: during the Oct–Nov 2023 SOL rally SOL permanently decoupled from AVAX, so the strategy shorted 'rich' SOL as it kept ripping, took a -$537k loss, and liquidated the account on 2023-11-10 despite its cointegration-break stop, with near-full 90.8% gross exposure making the event fatal. Even before the blowup the edge was marginal (PF 1.12, Sharpe 0.34, 76 trades). The analyst abandoned it at review — this pair has de-cointegrated and is dead — while noting the market-neutral pairs mechanism itself may be worth retrying on a still-cointegrated pair with far smaller per-leg sizing and a hard per-trade loss cap.

The market-neutral pairs mechanism is a legitimately under-represented bucket, but pairs convergence assumes a stable cointegrating relationship — SOL/AVAX de-cointegrated and no cointegration-break stop can restore it — and a 'dollar-neutral' book must never run near-full gross exposure without a hard per-trade loss cap, or a single premise failure liquidates the account.

It was abandoned at the pre-optimization backtest-review gate for total liquidation of a supposedly dollar-neutral trade: the SOL/AVAX ratio de-cointegrated during the Oct–Nov 2023 SOL rally — SOL structurally decoupled and kept ripping while the strategy shorted it as 'rich' — so the assumed equilibrium had permanently shifted (a premise failure), the stop_z=3.5 break did not prevent the wipeout, and near-full 90.8% gross exposure made the single de-cointegration event fatal.

A market-neutral, dollar-neutral pairs convergence trade on two same-sector L1 perps — SOLUSDT.BINANCE and AVAXUSDT.BINANCE — fading ≥2σ deviations of the daily log-ratio z-score (long the laggard, short the leader), held for weeks with a wide threshold and a hard cointegration-break stop, explicitly designed to amortize the 4-fill cost that killed the 4h ETH/BTC ratio version.

It liquidated the account on 2023-11-10: total return -100%, max drawdown 100% (CI high 3007%), largest loss -$537k on a single -24.95% day; even pre-blowup the edge was marginal — profit factor 1.12, Sharpe 0.34 (CI straddling zero), 55% win rate over only 76 trades, with avg_position_pct at 90.8% of equity and a catastrophic left tail (skew -3.3, kurtosis 143).