LtcTimeSeriesMomentumLongShortSingleAsset1D

LTC USD-M Time-Series Momentum, Long-Short (Single-Instrument Perp, Daily Bars, Pure-Price Trend Sign, EMA/Trend-Cross Exit, Wide ATR Stop, Low-Parameter — Validated-Family Extension to an Uncovered Deep-History Asset)

A LONG-SHORT, SINGLE-INSTRUMENT time-series (absolute) momentum strategy on LTCUSDT.BINANCE (USD-M perpetual, DAILY bars). This is a deliberate, conservative extension of the factory's ONE repeatedly-validated edge — time-series momentum/trend, whose SOL sibling (SolTimeSeriesMomentumLongShortSingleAsset1D) reached paper_stage (~0.69) — onto a LIQUID, DEEP-HISTORY asset the validated book does not yet cover (the existing TS-momentum-LS siblings are BTC/ETH/SOL only). After a long run of clever-but-dead ideas this session (cross-venue funding fee-dominated, COIN-M funding unbacktestable, Deribit options harness-broken, HL short-window non-generalization, microstructure verification failures), the highest-EV move is to broaden the one edge that demonstrably survives the full pipeline, on data that structurally avoids the dominant failure cause. LTC has multi-year Binance USD-M history (since 2019/2020), so walk-forward windows and the held-out recent tail each span real bull/bear/chop regimes — directly addressing the recent-regime NON-GENERALIZATION that killed the 6-month HL impulse expression. It keeps the VALIDATED trend payoff geometry (let winners run via a trend/EMA-cross exit, many small losses, fewer large winners) — NOT the inverted small-TP/large-stop mean-reversion geometry that broke the COIN-M Bollinger fade. It is pure-price (no supplementary-data coverage wall like the HL OI strategies), low-parameter (3) to resist the overfitting/sensitivity cliffs that recur, and long/short — filling the under-weight direction bucket (14.7% vs the 85.3% long-only over-weight). It is NOT any blacklisted family: not funding/carry, not options, not HL, not microstructure, not a fade.

A deliberate, conservative extension of the factory ONE repeatedly-validated edge -- time-series momentum/trend, whose SOL sibling (SolTimeSeriesMomentumLongShortSingleAsset1D) reached paper_stage (~0.69) -- onto a liquid, deep-history asset the validated book does not yet cover (existing TS-momentum-LS siblings are BTC/ETH/SOL only). After a run of clever-but-dead ideas (cross-venue funding fee-dominated, COIN-M funding unbacktestable, Deribit options harness-broken, HL short-window non-generalization, microstructure verification failures), the highest-EV move is to broaden the one edge that demonstrably survives the full pipeline on data that structurally avoids the dominant failure cause. LTC has multi-year Binance USD-M history (since 2019/2020), so walk-forward windows and the held-out recent tail each span real bull/bear/chop regimes -- directly addressing the recent-regime NON-GENERALIZATION that killed the 6-month HL impulse expression. It keeps the VALIDATED trend payoff geometry (let winners run via a trend/EMA-cross exit, many small losses, fewer large winners) -- NOT the inverted small-TP/large-stop mean-reversion geometry that broke the COIN-M Bollinger fade. Pure-price (no supplementary-data coverage wall like the HL OI strategies), low-parameter (3) to resist the overfitting/sensitivity cliffs that recur, and long/short -- filling the under-weight direction bucket (14.7% vs the 85.3% long-only over-weight). Not any blacklisted family: not funding/carry, not options, not HL, not microstructure, not a fade. leverage=1.0 and sizing uses risk_frac/notional cap not leverage, so no unused-leverage gate triggers.

Not worth the 2 hours: the validated TS-momentum mechanism is correctly built but the edge is absent on LTC specifically. The metrics that don't depend on return scale are decisive and unambiguous: profit_factor 0.69 (<1 means gross losses exceed gross wins — definitionally a losing strategy), Sharpe -0.48, Sortino -0.79, win_rate 24.4%, expectancy -$101/trade, probabilistic_sharpe 0.13. The payoff geometry is correct (avg_win $936 vs avg_loss $435 = 2.15:1, let winners run) and the trades match the hypothesis (balanced 83 long / 77 short, trend-cross exit) — so this is NOT a code bug to iterate. The problem is the asset: at a 2.15:1 payoff the breakeven win rate is 31.7% but LTC delivers only 24.4%, because LTC is a chronic laggard/chopper that ranged and bled through the 2024/2025 bull while the validated siblings (BTC/ETH/SOL) actually trended — so the EMA-cross trend exit gets whipsawed (many small losses, few payoffs). This is a PERSISTENT multi-regime loss (negative in 5 of 7 years: 2020/2022/2023/2024/2025; rolling Sharpe parked at -4 to -7 for years), not a single-window miss a parameter sweep can rescue — optimizing a PF-0.69 base would only fit an overfit in-sample region that fails DSR/holdout. The premise ('LTC trends like the validated majors') is falsified by the data. REDIRECT (Research Lead, new hypothesis): the TS-momentum-LS family is proven on BTC/ETH/SOL; if extending coverage, point it at a genuinely TRENDING uncovered deep-history asset, not a chronic range-bound laggard like LTC. FAILURE PATTERN: porting the validated daily TS-momentum-LS trend mechanism onto a low-trend/chronic-chopper alt (LTC) produces correct geometry but a sub-breakeven win rate (24% vs 31.7% needed), profit_factor <1, negative Sharpe/Sortino, and losses in 5 of 7 years — trend-following requires an asset that trends; the family does not generalize to laggard alts.

Long-short time-series (absolute) momentum on LTCUSDT.BINANCE USD-M perpetual, DAILY bars, pure price. calculate_signal returns the trailing momentum = close/close[-momentum_lookback]-1 (continuous, sign = trend direction). It enters in the trend direction confirmed by the EMA side: LONG when momentum > 0 AND close >= EMA(ema_period), SHORT when momentum < 0 AND close <= EMA. The primary exit is a TREND/EMA-CROSS: a long closes when the daily close drops back below the EMA, a short closes when it rises back above (let winners run while the trend persists, cut quickly when it breaks). A WIDE ATR stop (stop_atr_mult x ATR, default 4x) beyond entry is a reversal backstop, not the main exit -- preserving the fat-tailed winner distribution momentum needs. Position sizing risks risk_frac (2%) of equity over the wide-stop distance, capped at max_notional_frac of equity (no leverage). Single position, flat between signals; on a trend flip the EMA-cross exit closes and the next bar re-enters the opposite direction. 3 core parameters (momentum_lookback, ema_period, stop_atr_mult).

Mechanism correctly implements the validated TS-momentum family: balanced long-short (83/77), correct trend payoff geometry (avg_win $936 vs avg_loss $435 = 2.15:1, let winners run), trend-cross exit + wide ATR backstop, 3 params, deep history (1861 data_days, 160 trades) — trades match the hypothesis, so no code bug.

Healthy trade count and clean pure-price design that avoids the session's data/venue landmines.

Losing strategy regardless of return scale: profit_factor 0.69 (<1, gross losses exceed gross wins), Sharpe -0.48, Sortino -0.79, expectancy -$101/trade, probabilistic_sharpe 0.13.

Win rate 24.4% is below the 31.7% breakeven for a 2.15:1 payoff — negative expectancy by construction; the asset doesn't trend often enough at these params.

Persistent multi-regime loser: negative in 5 of 7 years (2020/2022/2023/2024/2025), rolling Sharpe at -4 to -7 for long stretches, max_consecutive_losses 13, recovery_factor -0.73 — not a single-window miss.

Premise falsified: the TS-momentum family is validated on BTC/ETH/SOL (trending majors); LTC is a chronic laggard/chopper that ranged through the 2024/2025 bull, so trend-following gets whipsawed.

negative 5 of 7 years

positive across regimes

24.4% vs 31.7% breakeven

above breakeven

LtcTimeSeriesMomentumLongShortSingleAsset1D conservatively extended the factory's one repeatedly-validated edge — daily TS-momentum — onto LTC, a liquid deep-history asset the BTC/ETH/SOL siblings didn't cover, keeping the proven let-winners-run geometry and a clean 3-parameter pure-price design. The mechanism was implemented correctly (balanced 160 trades, 2.15:1 payoff), but LTC simply doesn't trend enough: a 24.4% win rate against a 31.7% breakeven produced profit factor 0.69, negative Sharpe/Sortino, and losses in 5 of 7 years. The analyst ruled the premise falsified by the data — trend-following needs a trending asset, not a chronic chopper — and redirected the family toward a genuinely trending uncovered asset. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Trend-following requires an asset that trends: porting the validated TS-momentum mechanism onto a chronic laggard/chopper like LTC (which ranged through the 2024/2025 bull) yields correct payoff geometry but a 24.4% win rate against a 31.7% breakeven, profit factor <1, and losses in 5 of 7 years — extend the family to a genuinely trending uncovered deep-history asset, not a range-bound alt.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) because the edge is absent on LTC specifically — a sub-breakeven win rate and profit factor below 1 across multiple regimes, not a code bug or single-window miss — so optimizing a PF-0.69 base would only overfit, and optimization and all later stages were never reached.

A long-short, single-instrument time-series (absolute) momentum strategy on LTCUSDT.BINANCE USD-M daily perps — going long when trailing 50-day momentum is positive and price is above its EMA, short when negative, riding until a trend-cross EMA exit with a wide ATR stop backstop — a conservative pure-price, 3-parameter extension of the factory's validated TS-momentum family (SOL sibling at paper stage) onto a liquid, deep-history asset the book did not yet cover.

Over ~1,861 days and 160 trades (83 long / 77 short) it was a persistent loser despite correct trend geometry (avg_win $936 vs avg_loss $435 = 2.15:1): profit factor 0.69, Sharpe -0.48 (CI [-1.58, 0.23]), Sortino -0.79, expectancy -$101/trade, win rate 24.4% (below the 31.7% breakeven for its payoff), PSR 0.13, with losses in 5 of 7 years and max 13 consecutive losses.