LtcBtcRatioZScoreMeanReversionLongSpreadPair4H

LTC/BTC Ratio Z-Score Mean Reversion Long-Spread Pair (4H)

A DOLLAR-NEUTRAL CROSS-ASSET RELATIVE-VALUE PAIR strategy: simultaneously LONG LTCUSDT and SHORT BTCUSDT in equal dollar notional, triggered by extreme NEGATIVE z-scores of the LTC/BTC ratio. RATIONALE driven by the now-DECISIVE session finding (10 strategies analyzed, TIER 1 path-dependent trend-following EMPTY post-walk-forward, all long-only OHLCV directional alt-perp strategies abandoned with 100% failure rate, AND the empirical falsification of cross-symbol mechanism generalization via AVAX Supertrend's 1.72-Sharpe gap vs DOT): the only viable strategy class remaining per the analyst's explicit recommendation is cross-asset relative-value pairs. Pipeline already has 4 pairs: ETH/BTC (4H), BNB/ETH (4H), SOL/ETH (4H), LINK/ETH (1H). Adding LTC/BTC extends pair coverage to a FIFTH ratio with structurally different characteristics: (a) LTC/BTC is one of the longest-history cointegrated pairs in crypto (Litecoin launched 2011, 5+ years of Binance perp data on both legs), (b) historically REMARKABLY stationary — LTC has hovered in the 1-3% of BTC price range for 8+ years despite massive absolute moves in both, (c) LTC's 'digital silver' narrative (smaller blocks, faster confirmations, payments-focused) is structurally distinct from BTC's 'digital gold' store-of-value narrative — these are independent flow drivers, (d) LTC halvings (every 4 years) create unique relative-performance pulses that mean-revert. DESIGN CHOICES that ELIMINATE every session failure mode: (a) Dollar-neutral construction, (b) MR on stationary spread, (c) ZERO supplementary data, (d) Not directional/contrarian/oscillator/trend-following, (e) MAXIMALLY SIMPLE (2 params). DIFFERENTIATION FROM EXISTING PAIRS: ETH/BTC = smart-contract vs SoV; BNB/ETH = exchange vs DeFi; SOL/ETH = throughput-L1 vs mature-L1; LINK/ETH = oracle vs L1 — adding LTC/BTC = payments-coin vs SoV creates a fifth independent cointegration source with low cross-pair correlation.

Implements the analyst-endorsed cross-asset relative-value pair class as a fifth, low-correlation cointegration source (payments-coin vs store-of-value), reusing the landed SyncedBuffer pairs architecture. SyncedBuffer guarantees the ratio is computed on contemporaneous LTC/BTC 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 4-HOUR) ensures calculate_signal fires only when the BTC bar for the current LTC timestamp is present, so one aligned sample commits per bar. min_bars_required is 2 so the buffer starts filling immediately, and zscore_last returns None until 180 aligned samples accrue, gating warmup without dead bars. Unlike a symmetric pair this is single-direction (long-spread only) per the hypothesis: entry strictly on z <= -z_entry, exit on z >= z_exit (reversion) or time stop, keeping exactly two headline tunables (lookback_bars, z_entry). Entry is dollar-neutral and atomic -- each leg sized to equal USD notional (equity*leg_frac/price, not equal qty, so the position is delta-neutral and P&L tracks the ratio reverting), and 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). The returned z-score is a genuinely continuous per-bar signal so Layer-2 frozen-signal detection passes while entries stay threshold-gated. Both legs are liquid majors with 5+ years of Binance perp daily/4H history, so no leg risks data_unavailable; the should_enter/should_exit/position_size hooks are inert because the base entry path manages only the primary instrument.

Failed deflated Sharpe: DSR=0.0078 (vs 0.95 bar), expected-max=2.53 over 225 trials — the optimized Sharpe 1.00 is far below the best-of-N luck bar, indistinguishable from selection noise (is_significant=false, sharpe_ci_low -0.3447 straddling zero). Every multiple-testing/forward gate fails: PBO 0.75 (75% overfit probability); walk-forward is_overfitted=TRUE with avg IS 1.84 collapsing to avg OOS -0.622 (windows [-4.45, -0.69, 3.27], two negative and propped by one); and sensitivity FAILED with 3 cliffs on the strategy's CORE defining parameters (z_entry, lookback_bars, time_stop_bars) — the edge is knife's-edge on every knob. The single passing gate (holdout 0.349) and the improved optimized point metrics (PF 1.25, tail_ratio 9.3) are single-window/in-sample-selection artifacts; the PSR 0.9423 / DSR 0.0078 split is the diagnostic tell of best-of-225 inflation, not a generalizing edge. The marginal pre-optimization edge (Sharpe 0.53 with a CI straddling zero) was the warning, and it materialized: the LTC/BTC mean-reversion is genuine but too thin and fragile to clear honest deflation. Not iterate (attempt 1 of 2): with all three core parameters as cliffs, an overfit walk-forward with negative average OOS, and DSR ~0, there is no robust region to tune toward — a second sweep re-finds another knife's-edge, sub-significant config. Not revise_hypothesis: this is a one-sided pairs-MR with no proven sibling stranded on a dead target; the prior ETH/BTC and LINK/ETH pairs already failed (de-cointegrated), so the family is not a reliable redirect. FAILURE PATTERN: a one-sided dollar-neutral ratio mean-reversion on a modestly-cointegrated crypto pair can show a positive base backtest and even pass a single holdout window, yet still fail deflated Sharpe (0.0078, far below a 2.53 luck bar) with PBO 0.75, an overfit walk-forward with negative average OOS, a CI straddling zero, and cliffs on all of its core z-entry/lookback/time-stop parameters — the reversion edge is too thin and parameter-fragile to survive best-of-225 selection. A passing holdout and a clean dollar-neutral build do not rescue a pair whose Sharpe is indistinguishable from selection noise.

Dollar-neutral LTC/BTC ratio z-score mean-reversion pair, long-spread only, on Binance USD-M futures, 4H bars. Each timestamp-aligned 4H bar commits the (LTC, BTC) close pair to a SyncedBuffer, computes the LTC/BTC ratio, and z-scores it over a rolling 180-bar (30-day) window. When z <= -z_entry (LTC cheap vs BTC) it goes LONG LTC and SHORT BTC in equal USD notional, expecting the historically stationary ratio to revert up; it closes both legs when z reverts up through z_exit (~0) or after a 60-bar time stop. Pure OHLCV, no supplementary data, leverage 1.0 with ~90% gross (45% per leg). Two edge parameters: lookback_bars and z_entry.

Genuinely profitable, correctly-built dollar-neutral pair: total_return +29.8%, PF 1.12, positive expectancy (+$198/trade), tail_ratio 5.18, beta 0.012, no sizing/exposure bug

Positive in 4 of 7 years including recent regimes (2024 +30.5%, 2025 +10.5%) — the LTC/BTC ratio retained mean-reversion where the ETH/BTC sibling de-trended and lost

Healthy frequency (140 trades), pure OHLCV, clean construction

Marginal Sharpe 0.53 with sharpe_ci_low -0.60 (CI straddles 0) — deflated Sharpe is the binding optimization risk after best-of-N selection

Liquidity/capacity wall: impact_cost_pct 11.66% with capacity_usd only $7.35M — the thin LTC leg lets market impact eat a meaningful share of the edge; deployable only at small scale

2023 -20.1% shows the ratio can trend against the long-spread for a year; holdout must confirm recent-regime persistence

Clean dollar-neutral construction (beta 0.009, no sizing/exposure bug); the only positive-expectancy pair of the session's pairs cohort, with a passing holdout (0.349) and improved optimized point metrics (PF 1.25, tail_ratio 9.3, max_DD 11%)

Recent-positive profile (2024 +28.6%, 2025 +10.2%) — the LTC/BTC ratio retained some mean-reversion where ETH/BTC and LINK/ETH de-cointegrated

Failed deflated Sharpe: DSR 0.0078, is_significant=FALSE; optimized Sharpe 1.00 far below the expected-max luck bar 2.53 over 225 trials

PBO 0.75 (high overfit probability); sharpe_ci_low -0.3447 (CI straddles 0)

Walk-forward is_overfitted=TRUE: avg IS 1.84 collapses to avg OOS -0.622, OOS windows [-4.45, -0.69, 3.27] (two negative, propped by one)

Sensitivity FAILED with 3 cliffs on the core defining parameters (z_entry, lookback_bars, time_stop_bars) — knife's-edge on every knob

PSR 0.9423 vs DSR 0.0078 split confirms the headline is best-of-225 selection inflation; capacity is also modest (impact 5.4%, capacity $34M)

LtcBtcRatioZScoreMeanReversionLongSpreadPair4H extended the pairs sleeve to a fifth ratio, betting on the historically stationary LTC/BTC spread, and it stood out as the session's only positive-expectancy pair — +29.8% with a 1.12 profit factor and positive recent years — earning an 'optimize' verdict where the ETH/BTC and LINK/ETH siblings had de-cointegrated and lost. But its marginal Sharpe of 0.53 with a CI straddling zero proved the warning: optimization produced a deflated Sharpe of just 0.0078 against a 2.53 luck bar, PBO 0.75, an overfit walk-forward with negative average OOS, and sensitivity cliffs on all three defining parameters. The single passing holdout and the improved point metrics were best-of-225 selection artifacts, confirmed by the PSR 0.94 / DSR 0.008 split. On its first iteration the analyst abandoned it, concluding the LTC/BTC reversion was real but too thin and parameter-fragile to clear honest deflation.

A genuine but thin mean-reversion edge — even the one cointegrated pair that retained reversion where ETH/BTC and LINK/ETH de-trended — can still be too fragile to survive honest deflation: a marginal base Sharpe with a CI straddling zero is the warning, and a passing single holdout cannot rescue a config that is knife's-edge on every core parameter with negative average OOS.

Despite a passing holdout (Sharpe 0.349), the analyst abandoned it post-optimization because the deflated Sharpe was 0.0078 (vs the 0.95 bar) with the optimized Sharpe 1.00 far below the 2.53 luck bar, PBO was 0.75, walk-forward was_overfitted=TRUE (avg IS 1.84 collapsing to avg OOS -0.622 with two negative windows), and sensitivity FAILED with cliffs on all three core parameters (z_entry, lookback_bars, time_stop_bars).

A dollar-neutral, OHLCV-only relative-value pair on 4H bars — going long LTCUSDT and short BTCUSDT in equal dollar notional when the LTC/BTC ratio's z-score fell to -2 (the ratio looked cheap), exiting on reversion to z=0 or a time stop — a one-sided long-spread expression of the 'digital silver reverts to digital gold' thesis on what was argued to be crypto's most stationary cointegrated pair.

Unlike its de-cointegrated pair siblings, the base backtest was genuinely profitable: +29.8% total return over 140 trades (70 long, 70 short) with profit factor 1.12, positive expectancy (+$198/trade), tail ratio 5.18, and positive recent years (2024 +30.5%, 2025 +10.5%) — but only Sharpe 0.53 with a CI low of -0.60, and an 11.66% impact cost against just $7.35M capacity. Optimization improved the point metrics (Sharpe 1.00, PF 1.25, max DD 11.1%) but the forward windows broke.