Strategies

Binance USD-M Alt/BTC Ratio-Momentum Pairs (Trend-Follow the Relative-Strength Spread, Dollar-Neutral, Daily, Established Majors)

Outcome: Abandoned

AltBtcRatioMomentumPairsTrendFollowNeutral

Outcome Summary

This strategy reasoned that since fading crypto alt/BTC ratios lost because the ratios trend rather than revert, trend-following the same spread should be the validated sign, applying the factory's proven daily-momentum edge to relative value while staying dollar-neutral and inside the data-safe envelope. The implementation was correct and genuinely market-neutral, but the backtest exposed no realized edge — profit factor 0.97, negative expectancy, a near-zero Sharpe with a CI through zero, and a +14.36% headline that was almost all unrealized mark-to-market masking a roughly -14.8% realized return. The apparent upside came from a handful of 2021 alt-season tail days (kurtosis 322), offset by a -48.3% 2023 and a 1,701-day drawdown. After two iterations the reviewer abandoned it before optimization, judging it the same no-edge, outlier-carried signature as the ratio and basis pairs families and not a tunable region but a dead premise.

Strategy report

Hypothesis

A dollar-neutral, multi-pair RATIO-MOMENTUM strategy that TREND-FOLLOWS the alt-vs-BTC relative-strength spread across a small set of established-major ratios (ETH/BTC, SOL/BTC, BNB/BTC, XRP/BTC) on Binance USD-M. For each alt, it computes the ratio R = alt/BTC and goes LONG the alt + SHORT BTC (equal notional) when R is in an UPTREND, and SHORT the alt + LONG BTC when R is in a DOWNTREND — riding relative-strength regimes ('alt season' = alts trend up vs BTC; risk-off = alts trend down vs BTC). This is the deliberate CORRECT-SIGN INVERSION of the refuted cross-asset ratio mean-reversion: the analyst empirically proved that crypto ratios are DRIFT-DOMINANT and NON-STATIONARY — they TREND with relative-performance regimes (SOL structurally outperformed AVAX), which is exactly why fading the ratio (MR) lost. Trend-FOLLOWING the ratio is therefore the validated sign, applying the factory's one proven edge (daily trend persistence, the +392% / Sharpe 2.59 basket) to the relative-value spread instead of to outright price. It stays strictly inside the validated envelope: DAILY cadence (intraday/4H single-bar directional on majors is now refuted in both signs at ~-98%), established majors only (data-safe — broad-alt and HL baskets die on data_unavailable), pure OHLCV (no funding/basis/options/COIN-M/liquidations dependency), and SYMMETRIC (short-only failed standalone; this takes both relative directions within a net-edge book). It fills the under-represented pairs scope (14.9% vs single 72.4%), is market-neutral (each pair dollar-neutral — improving the 87%-long skew), and is distinct from the cross-sectional majors basket (explicit per-pair ratio-trend vs cross-sectional ranking). Few parameters resist overfitting.

Iteration 2 fix: the only Layer-2 failure was `_bar_ts() takes 1 positional argument but 2 were given` across all six synthetic scenarios. Root cause: `_bar_ts` was a `@staticmethod`, but the Layer-2 strategy proxy rebinds methods such that `self._bar_ts(bar)` passed `self` as a positional arg into a 1-arg static function. Converting `_bar_ts` to a normal instance method (`def _bar_ts(self, bar)`) resolves the call signature while leaving all other logic — imports, signal computation, rebalance construction, sizing — byte-identical, so the earlier-passing Layer 1 static checks remain green. Strategy still implements the validated correct-sign edge: trend-FOLLOWING the drift-dominant alt/BTC ratio (the inversion of the refuted ratio mean-reversion), strictly inside the validated envelope (daily cadence, established majors, pure OHLCV, dollar-neutral, symmetric).

Not worth optimizing: the alt/BTC ratio-momentum trend-follower has no realized edge. Profit factor 0.97, negative expectancy/avg_trade_return, and Sharpe 0.124 with a CI straddling 0; the positive +14.36% headline is almost entirely open-position mark-to-market (end_unrealized_pct 29.13 → realized return ≈ -14.8%). What little upside exists is pure outlier luck concentrated in a handful of 2021 alt-season tail days (return_kurtosis 322, skew 12.76, June 2021 +107%), offset by catastrophic regime years (2023 -48.3%) and a 32% drawdown lasting 1701 days. With only two tunables (lookback, rebalance_days) and the edge resting on ~5 tail days, a best-of-225 sweep would overfit the 2021 spike and fail deflated Sharpe. This is the same no-edge/outlier-carried signature the cross-asset ratio and basis pairs families produced repeatedly this session; trend-following the ratio (the correct-sign inversion) does not convert a drift-dominant spread into a deployable edge. Abandon rather than iterate — the mechanism is implemented correctly; the premise simply yields no robust edge to refine.

Implementation

Dollar-neutral multi-pair ratio-momentum trend-follower on Binance USD-M daily bars. For each alt (ETH, SOL, BNB, XRP), computes ratio R = alt/BTC and goes LONG alt + SHORT BTC when R is in an uptrend, SHORT alt + LONG BTC when downtrend. BTC short/long contributions across pairs are netted into one BTC target. Equal per-pair notional keeps total gross <= gross_cap*equity (1.0x, no leverage); each pair is dollar-neutral. Rebalances every rebalance_days on aligned daily bars.

Verification Results

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['btc', 'alts', 'lookback', 'gross_cap', 'resize_frac', 'min_notional', 'qty_precision', 'rebalance_days', 'default_qty_precision'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000) - steady_downtrend: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000) - flat_ranging: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000) - volatility_spike: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000) - zero_volume: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000) - price_gap: TypeError: AltBtcRatioMomentumPairsTrendFollowNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735692660000)

Backtest Review

Sharpe: 0.12
Total return: 14.36%
Max drawdown: 31.95%
Trades: 308
Win rate: 46.1%
Profit factor: 0.97

Mechanism executes the hypothesis correctly: dollar-neutral, balanced long/short (158/150), reasonable trade count (308) for a daily pairs strategy

Genuinely market-neutral (beta 0.03, benchmark_correlation 0.09); low commission drag (1.36% of gross)

No edge: profit_factor 0.97 (<1), expectancy -47.9, avg_trade_return negative, Sharpe 0.124 with CI straddling 0 (-0.64 to 0.68)

The +14.36% headline is open-position MTM, not realized: end_unrealized_pct 29.13 implies realized closed-trade return is ~-14.8%

Extreme outlier dependence (return_kurtosis 322, skew 12.76, tail_ratio 0.13) — the only gains come from ~5 2021 alt-season tail days (June 2021 +107%); strip them and it loses

Violent regime dependence (2023 -48.3%, 2020 -16%, 2022 -8%) with a 32% drawdown lasting 1701 days

Outcome Summary

Inverting the sign of a refuted ratio-mean-reversion idea does not manufacture an edge: a drift-dominant, non-stationary spread whose only profits come from a few outlier alt-season days is not deployable in either direction.

It was abandoned at the pre-optimization backtest-review gate (optimization never ran) because there was no realized edge to refine — the upside rested on roughly five 2021 tail days, so a parameter sweep would merely overfit the spike, and trend-following the spread did not convert a drift-dominant ratio into a deployable edge.

Trend-follow the alt/BTC relative-strength spread dollar-neutral across four established-major ratios (ETH/BTC, SOL/BTC, BNB/BTC, XRP/BTC) on daily Binance USD-M bars — going long the alt and short BTC when the ratio is in an uptrend and vice versa — as the correct-sign inversion of the refuted ratio mean-reversion.

The mechanism executed cleanly and market-neutral (308 trades, 158 long / 150 short, beta 0.03, 1.36% commission drag), but showed no realized edge: profit factor 0.97, negative expectancy (-47.9), Sharpe 0.124 with a CI straddling zero (-0.64 to 0.68), and the +14.36% headline was almost entirely open-position mark-to-market (end_unrealized 29.13% implies realized return ≈ -14.8%). Gains were concentrated in a handful of 2021 alt-season tail days (kurtosis 322, skew 12.76, June 2021 +107%) against catastrophic regime years (2023 -48.3%) and a 32% drawdown lasting 1,701 days.

Backtest and paper results are hypothetical. Trading involves risk of loss.