LinkEthOneHourRatioZScoreMeanReversionLongSpreadPair

LINK/ETH Ratio Z-Score Mean Reversion Long-Spread Pair (1H — Sub-Daily)

A DOLLAR-NEUTRAL CROSS-ASSET RELATIVE-VALUE PAIR strategy on 1-HOUR bars: simultaneously LONG LINKUSDT and SHORT ETHUSDT in equal dollar notional, triggered by extreme NEGATIVE z-scores of the LINK/ETH ratio. RATIONALE driven by the now-definitive session finding (9 strategies analyzed, TIER 1 path-dependent trend-following EMPTY, all long-only OHLCV directional alt-perp strategies abandoned): the analyst's explicit recommendations now reduce to TWO viable paths: (1) cross-asset relative-value pairs (already proven by ETH/BTC/BNB/ETH/SOL/ETH pairs in pipeline) and (2) SUB-DAILY timeframe strategies (NOT yet explored). This proposal explicitly combines BOTH recommendations: a cross-asset pair on a NEW ratio (LINK/ETH) AT a NEW timeframe (1H, faster than the existing 4H pairs). DESIGN CHOICES that ELIMINATE every session failure mode: (a) Dollar-neutral by construction (eliminates beta-to-crypto exposure that killed all 9 directional alt-perp strategies), (b) MR on a CO-MOVING ratio (not absolute price — failed), (c) ZERO supplementary data dependency, (d) NOT contrarian short on a single instrument, (e) NOT oscillator entry, (f) NOT trend-following on alt-perp, (g) NEW pair direction (LINK as DEX-oracle vs ETH as L1 — different sectors, different flow drivers, different beta profiles than ETH/BTC, BNB/ETH, SOL/ETH already in pipeline), (h) NEW timeframe (1H captures FASTER MR cycles than the 4H pairs — addresses the analyst's 'sub-daily' recommendation directly), (i) MAXIMALLY SIMPLE — only 2 decision parameters (z-score lookback=240, entry threshold=-1.5). KEY DIFFERENTIATION FROM EXISTING PAIRS: ETH/BTC, BNB/ETH, SOL/ETH all live at 4H — a 1H pair captures different reversion frequencies (faster, more trades per year) and is structurally orthogonal to the 4H sleeve. LINK is the oracle/DeFi-infrastructure sector — uncorrelated with smart-contract-platform (ETH), exchange (BNB), throughput-L1 (SOL), and store-of-value (BTC) drivers already represented.

Implements the hypothesis with the repo's SyncedBuffer pairs helper (used by landed pairs strategies) and the centralized multi-leg pattern. SyncedBuffer guarantees the LINK/ETH ratio is always computed on contemporaneous closes (it commits an (a,b) sample only when both legs report the same ts_ns), eliminating the independent-array-trim alignment bug; the base cross-leg barrier (both legs 1H) ensures calculate_signal fires only when the ETH bar for the current LINK timestamp is present, so the buffer commits one aligned sample per bar. The strategy is one-sided per the hypothesis: it enters only on extreme negative z (z <= entry_z = -1.5) by going long LINK / short ETH, and exits both legs on reversion (z >= exit_z = 0.0) or a 240-bar time stop, keeping the decision surface to the two stated tunables (lookback_bars, entry_z). Entry is dollar-neutral (each leg sized to equity*leg_frac/price, not equal qty) so P&L tracks the ratio reverting rather than crypto beta, and it is atomic -- if either leg fails its qty/min-notional check, neither opens, preventing a naked directional bet. leg_frac=0.45 keeps gross ~90% within the leverage-1.0 futures margin ceiling (leverage not set >1, so the unused-leverage gate does not apply). min_bars_required=2 lets the buffer fill immediately while zscore_last returns None until 240 aligned samples accrue, gating warmup without dead bars. The returned z-score is a genuinely continuous per-bar signal so Layer-2 frozen-signal detection passes while entries stay threshold-gated. 1H bars on a new LINK/ETH ratio give faster reversion cycles orthogonal to the existing 4H ETH/BTC, BNB/ETH, SOL/ETH pairs; both legs are liquid majors with multi-year 1H history so no data_unavailable risk. The should_enter/should_exit/position_size hooks are inert because the base entry path manages only the primary instrument.

Do not optimize — this is the third pairs-mean-reversion strategy this session with the identical de-cointegration failure (after the abandoned SOL-AVAX and ETH/BTC pairs), and it is a clear money-loser. Over 366 trades it returns -57.6% with profit_factor 0.86, Sharpe -0.75 (sharpe_ci_low -1.50), omega 0.70, Sortino -0.56, expectancy -$146/trade, PSR 0.037, and negative annual returns in 5 of 7 years (2020 -18%, 2021 -26%, 2022 -23%, 2023 -11%). The 71.6% max drawdown on a supposedly dollar-neutral book, together with return_skew -7.4 and largest_loss -$26.8k vs largest_win $10.4k, is the signature of a NON-stationary, trending spread: LINK structurally underperformed ETH across 2020-2023, so the one-sided long-spread (long LINK / short ETH on negative z) repeatedly caught a falling knife. The one-sided design compounds this — it can profit only if the ratio rises, with no defense against the secular LINK/ETH downtrend. This is a clean, correctly-sized, dollar-neutral build (beta -0.015, no blow-up) with a healthy trade count, so it is NOT a sizing bug or sparsity — it is a genuine absence of edge because the spread is not mean-reverting over the test period, and no change to lookback_bars or entry_z makes a trending ratio revert. The pairs-MR mechanism may work on a genuinely cointegrated pair, but LINK/ETH over this window is not one (the same lesson as the abandoned ETH/BTC and SOL-AVAX pairs); abandon rather than optimize, and any future pairs attempt should be cointegration-tested and two-sided rather than a one-sided long-only ratio bet.

Dollar-neutral cross-asset pair mean-reversion on LINKUSDT (leg A) and ETHUSDT (leg B) Binance USD-M futures, 1H bars. On each timestamp-aligned 1H bar it commits the (LINK, ETH) close pair to a SyncedBuffer, forms the LINK/ETH ratio, and z-scores it over a 240-bar (10-day) rolling window. When z <= -1.5 (ratio cheap) it opens the long spread: long LINK and short ETH at equal USD notional. It closes both legs when z reverts to >= 0.0 or after a 240-bar time stop. One-sided (long-spread only) per the hypothesis. Pure OHLCV, no supplementary data, leverage 1.0 with ~90% gross (45% per leg). Two decision parameters (lookback_bars, entry_z).

Clean dollar-neutral construction (beta -0.015, exposure 77%, avg_position 46%) with no sizing/exposure bug, healthy frequency (366 trades), pure OHLCV

Novel sector pairing (oracle vs L1) at a new 1H timeframe

Consistent money-loser: total_return -57.6%, Sharpe -0.75 (sharpe_ci_low -1.50), profit_factor 0.86, omega 0.70, expectancy -$146/trade, PSR 0.037; negative in 5 of 7 years

71.6% max drawdown on a supposedly market-neutral pair, with return_skew -7.4 (largest_loss -$26.8k vs largest_win $10.4k) — the spread is non-stationary and trended down (LINK structurally underperformed ETH)

One-sided long-spread design can only profit if the ratio rises — no defense against the secular LINK/ETH downtrend; z-score MR keeps buying a declining spread

No parameter change makes a trending/non-stationary spread mean-revert

LinkEthOneHourRatioZScoreMeanReversionLongSpreadPair explicitly combined the session's two surviving recommendations — cross-asset relative-value and sub-daily timeframes — into a dollar-neutral 1H long-spread on a novel LINK/ETH (oracle vs L1) pairing. The execution was sound and genuinely beta-neutral with 366 trades, but it lost 57.6% with a profit factor of 0.86 and a 71.6% drawdown, because the LINK/ETH ratio trended down rather than reverting as LINK underperformed ETH, leaving the one-sided long-spread repeatedly catching a falling knife. This was the third de-cointegration failure of the session after the abandoned SOL-AVAX and ETH/BTC pairs. The analyst abandoned the run at the backtest-review gate, reiterating that the pairs mechanism needs a cointegration-tested, two-sided spread rather than a one-sided ratio bet.

Combining sound design recommendations (dollar-neutral, cross-asset, sub-daily) cannot manufacture an edge when the chosen ratio isn't actually cointegrated — a one-sided long-spread has no defense against a secular downtrend, so pairs attempts must be cointegration-tested and two-sided rather than one-sided z-score bets on an arbitrary ratio.

The analyst abandoned it at the backtest-review gate before optimization because — for the third pairs strategy this session — the spread was non-stationary: LINK structurally underperformed ETH across 2020-2023, so the one-sided long-spread logic repeatedly bought a declining ratio (a falling knife), and the 71.6% drawdown on a dollar-neutral book confirmed the de-cointegration with no parameter change able to make a trending ratio revert.

A dollar-neutral, OHLCV-only relative-value pair on 1-hour bars — going long LINKUSDT and short ETHUSDT in equal dollar notional when the LINK/ETH ratio's z-score over a 240-bar (~10-day) window fell to -1.5 (the ratio looked cheap), exiting on reversion to z=0 or a time stop — a one-sided long-spread bet combining the analyst's cross-asset-pair and sub-daily-timeframe recommendations on a novel oracle-vs-L1 sector pairing.

The build was clean and genuinely market-neutral (beta -0.015, no blow-up) over a healthy 366 trades (183 long, 183 short), but it lost heavily: -57.6% total return (CAGR -11.3%), Sharpe -0.75 (CI low -1.50), profit factor 0.86, per-trade expectancy -$146, PSR 0.037, a 71.6% max drawdown, and a fat left tail (skew -7.4, largest loss -$26.8k vs largest win $10.4k), negative in 5 of 7 years.