EthBtcLeadershipTrendBetaNeutralPairLS

ETH-vs-BTC Beta-Neutral Leadership-Trend Pair, Long-Short Market-Neutral (Two USD-M Perps ETHUSDT + BTCUSDT.BINANCE, Daily Bars, Trade the PERSISTENT TREND of the ETH/BTC Ratio — Long the Leader / Short the Laggard, Flip on Confirmed Leadership Change, Dollar-Neutral, Low-Turnover, 3-Parameter)

A DOLLAR-NEUTRAL (beta-hedged) LONG-SHORT pair on ETHUSDT and BTCUSDT USD-M perps that trades the PERSISTENT TREND of the ETH/BTC ratio: it goes long the currently-leading major and short the lagging one, capturing relative alpha with crypto market beta hedged out. This is deliberately the OPPOSITE POLARITY of the dead ETH/BTC ratio MEAN-REVERSION (which faded the ratio and died 'other'); here we FOLLOW the ratio's trend, because the ETH/BTC ratio exhibits well-documented multi-month leadership regimes (ETH-season vs BTC-dominance) that trend rather than revert on the horizon traded. It is structurally distinct from every dead/in-pipeline family: NOT a single-name directional fade/reversal (those all lost money — this is a hedged relative trade with no net market exposure), NOT 8-major cross-sectional RANK rotation (L32 — this is a focused 2-leg beta-neutral cross of the two deepest, longest-history majors, not a rank-and-rotate over a wide basket), NOT funding-carry (L33 — no coupon harvested), NOT options, NOT liquidation/OI/premium/taker crowding signals, NOT a rare-event trigger (leadership is a continuous state). It fills THREE under-quota dimensions simultaneously — long-short (13.9%), market-neutral, and pairs scope — while low turnover (flips only on confirmed leadership change, a few times per year) keeps two-leg USD-M fees well below the relative moves captured (ETH/BTC regimes routinely run 15-40%).

Implements the ETH/BTC leadership-TREND edge as a market-neutral pair — the deliberate opposite polarity of the dead ETH/BTC ratio mean-reversion (which faded the ratio): here we FOLLOW the ratio's multi-month leadership regime (ETH-season vs BTC-dominance), long the leader and short the laggard, so crypto market beta is hedged out and the P&L is relative alpha. It is structurally distinct from the dead/in-pipeline families: not a single-name directional fade/reversal (this is hedged, no net market exposure), not 8-major cross-sectional rank rotation (L32 — this is a focused 2-leg beta-neutral cross of the two deepest, longest-history majors, no ranking/top-K), not funding-carry (L33 — no coupon), not options, not OI/premium/taker/liquidation signals, and not a rare-event trigger (leadership is a continuous state). It is pure OHLCV (ratio from prices), so there is no supplementary-feed data wall. The hysteresis band is the key design choice: the leadership state only flips when momentum crosses beyond +/-confirm_band, so the pair holds a regime for months and flips only a few-to-~dozen times/year, and a wide 25% rebalance no-trade band means day-to-day equity drift does not churn the legs — only a real flip trades, keeping two-leg USD-M fees negligible against the 15-40% relative moves ETH/BTC regimes run. Execution trades both perps via an on_bar override that computes signed target notionals and submits per-leg deltas with _submit_entry_instrument (netting flips a leg from long to short in one order); the base per-instrument hooks are inert stubs as in the proven multi-instrument pattern, and I evaluate off the primary (ETH) daily bar with a freshness tolerance so a 1-day leg mismatch (irrelevant to a slow leadership signal) never stalls it. Risk is bounded and neutral: each leg is 0.40x equity (gross ~0.8x, net ~0 dollar-neutral), leverage 1.0 with sizing that never references leverage, so the position cannot liquidate and there is no directional beta to blow up (avoiding the single-name-perp deaths). It fills three under-quota dimensions at once (long-short, market-neutral, pairs). Kept to exactly 3 tuned parameters (trend_len, confirm_band, leg_gross) to resist overfitting; the rebalance band, staleness, and min-notional are fixed structural choices. Note: dollar-neutral (equal-notional) legs are used per the design; ETH's higher beta makes this slightly ETH-tilted rather than perfectly beta-neutral, an intentional simplicity/robustness trade-off that keeps the parameter count minimal. Trade count is measurable — each flip is 4 leg-trades (close 2 + open 2) and confirmed flips recur through the window." }

Catastrophic near-total loss: total_return -99.9%, max_drawdown 99.93% (CI [99.90, 100.00]), profit_factor 0.287, avg_trade_return_pct -3.79%, expectancy -$306/trade on 322 well-sampled trades, with EVERY year deeply negative (2020 -81%, 2021 -55%, 2022 -51%, 2023 -49%, 2024 -34%). Do not be misled by the reported Sharpe 0.81 / Sortino 14.99 / Omega 5.84 / PSR 1.0 — these are broken-metric artifacts of a pathological return distribution (skew 27.4, kurtosis 836) driven by a couple of outlier up-days, not evidence of edge; the ground truth is a 99.9% wipeout. The 'dollar-neutral / beta-hedged' claim is also false in the engine: a genuinely market-neutral ETH/BTC pair cannot lose -32.4% / -19.4% / -16.0% in single days (Mar 2020 crash), so the leg sizing/rebalance is taking large un-hedged directional losses rather than a hedged spread. The core thesis fails on its merits too — the ETH/BTC ratio whipsaws rather than trending at this horizon, so 'follow the leadership trend' bought every reversal. Optimizing trend_len/confirm_band/leg_gross cannot rescue a mechanism that lost 99.9% across all years with broken neutrality; it would only overfit the handful of outlier up-days. This is the trend-following polarity of the already-dead ETH/BTC ratio mean-reversion family, now confirmed catastrophic in the opposite direction. Abandon.

Dollar-neutral long-short pair on ETHUSDT and BTCUSDT USD-M perps, daily bars, that follows the leadership trend of the ETH/BTC ratio. Each day it forms ratio = ETH/BTC and its momentum vs a trend_len=50-day SMA; a hysteresis band (confirm_band=0.02) sets the leadership state — momentum >= +band -> ETH leads (long ETH / short BTC), <= -band -> BTC leads (long BTC / short ETH), inside the band -> keep the current leader, so flips are confirmed and rare. Each leg targets leg_gross=0.40 x equity notional with opposite signs (dollar-neutral, gross ~0.8x, net ~0), rebalanced via signed-position deltas with a 25% no-trade band so only a genuine leadership flip trades. Long+short on the perps; leverage 1.0 keeps a margin buffer. 3 tuned params (trend_len, confirm_band, leg_gross).

Lean abandon over re-tuning trend_len/confirm_band (which would curve-fit flip points to this window).

Net loser: total_return -23.13%, Sharpe -0.69 (CI [-2.41, 0.90]), PF 0.63, win_rate 0.33, avg_trade_return_pct -2.42% over 30 balanced trades (15L/15S). avg_win $4066 > avg_loss $3217 but the 33% hit rate makes expectancy negative — the ratio trend-followed but the window favored ETH/BTC MEAN-REVERSION. Not a code defect (no L17 signature; polarity correct). IMPORTANT: the sibling ETH/BTC mean-reversion already died, and now this trend-following polarity also loses — both directions of the daily ETH/BTC timing trade have failed, strong evidence of no exploitable edge.

If iterated, add a rolling beta estimate to scale the BTC leg — but only if the edge is first shown to exist.

Title says 'beta-neutral' but the code sizes DOLLAR-neutral (equal notional). ETH's higher beta leaves residual net crypto beta (sandbox beta -0.19, correlation -0.37) — not truly market-neutral. Disclosed by the developer as an intentional 3-param simplification, and the description's primary framing is 'dollar-neutral', so NOT critical. Part of the -23% may be uncompensated residual beta.

Confirm full-window flip/trade count before treating the negative edge as conclusive.

30 trades over 362 days (~7-8 flips) is below the ~100-trade measurability floor; Sharpe CI spans zero. Inherent to the low-turnover hysteresis design; needs a longer window. metrics_reliable=true.

Well-sampled at 322 trades with balanced long/short legs and benchmark_meaningful=false correctly flagging the intended market-neutral structure.

Near-total loss: total_return -99.9%, max_drawdown 99.93% (CI [99.90, 100.00]) — hard abandon.

Decisively negative economics: profit_factor 0.287, avg_trade_return_pct -3.79%, expectancy -$306/trade, win_rate 28.3%.

Every year catastrophically negative (2020 -81%, 2021 -55%, 2022 -51%, 2023 -49%, 2024 -34%).

'Dollar-neutral / beta-hedged' claim is false: single-day losses of -32.4%, -19.4%, -16.0% (Mar 2020) are impossible for a hedged pair — the legs take large un-hedged directional losses (kurtosis 836, skew 27.4).

Reported Sharpe 0.81 / Sortino 14.99 / PSR 1.0 are broken-metric artifacts of the pathological distribution, NOT edge; the real result is a -99.9% wipeout.

Thesis fails: the ETH/BTC ratio whipsaws rather than trends at this horizon, so the leadership-trend follow bought every reversal.

This strategy inverted a previously-dead ETH/BTC mean-reversion idea, instead following the ratio's leadership trend by going long the leader and short the laggard in a supposedly dollar-neutral perp pair. The result was catastrophic: a -99.9% near-total wipeout with a 99.93% max drawdown and every single year deeply negative, driven by the ETH/BTC ratio whipsawing rather than trending so that each flip bought a reversal. The headline Sharpe of 0.81 and Sortino of 14.99 were exposed as broken-metric artifacts of a pathological, outlier-driven distribution, and single-day losses above 30% during March 2020 proved the neutrality claim false — the legs were taking large un-hedged directional hits. The analyst abandoned it at backtest review, noting optimization could not rescue a mechanism that lost 99.9% across all regimes with broken hedging.

A claimed market-neutral structure must be validated against realized single-day moves before trusting it — un-hedged leg sizing turned a 'beta-neutral' pair into a directional bet — and following ETH/BTC leadership trend simply bought every reversal because the ratio whipsaws rather than trends at this horizon.

The analyst abandoned it at the pre-optimization backtest-review gate: it was a 99.9% wipeout, and the 'dollar-neutral/beta-hedged' claim proved false in the engine — single-day losses of -32.4%/-19.4%/-16.0% during the March 2020 crash are impossible for a truly hedged pair, so the legs were taking large un-hedged directional losses while the ETH/BTC ratio whipsawed rather than trended.

A dollar-neutral, long-short pair on ETHUSDT and BTCUSDT USD-M perps that followed the persistent leadership trend of the ETH/BTC ratio — going long the leading major and short the lagging one, flipping only on a confirmed leadership change via a hysteresis band, with crypto market beta hedged out.

It was a catastrophic near-total loss over 322 well-sampled trades (161 long / 161 short): total return -99.9%, max drawdown 99.93% (CI [99.90, 100.00]), profit factor 0.287, win rate 28.3%, expectancy -$306/trade, and avg_trade_return_pct -3.79%, with every year deeply negative (2020 -81%, 2021 -55%, 2022 -51%, 2023 -49%, 2024 -34%). Reported Sharpe 0.81, Sortino 14.99 and PSR 1.0 were dismissed as broken-metric artifacts of a pathological distribution (skew 27.4, kurtosis 836).