DOGE Beta-Hedged Idiosyncratic – Backtest

Hypothesis

A MARKET-NEUTRAL (beta-neutral) LONG-SHORT mean-REVERSION strategy that fades the IDIOSYNCRATIC 3-day overshoot of DOGE after removing its BTC-driven component. This is deliberately the OPPOSITE sign of the residual-MOMENTUM idea that died on XRP (that traded continuation of the residual trend; this fades a short-horizon residual overshoot), and it is mechanistically distinct from the dead cross-sectional BASKET reversals (USEQ short-term reversal, majors 2-day reversal) because it is a SINGLE-NAME residual trade……Show moreShow less

A MARKET-NEUTRAL (beta-neutral) LONG-SHORT mean-REVERSION strategy that fades the IDIOSYNCRATIC 3-day overshoot of DOGE after removing its BTC-driven component. This is deliberately the OPPOSITE sign of the residual-MOMENTUM idea that died on XRP (that traded continuation of the residual trend; this fades a short-horizon residual overshoot), and it is mechanistically distinct from the dead cross-sectional BASKET reversals (USEQ short-term reversal, majors 2-day reversal) because it is a SINGLE-NAME residual trade whose BTC hedge strips out exactly the systematic market drift that made those single-name reversals lose. Economic basis: high-idiosyncratic-vol meme majors like DOGE experience sentiment-driven overreactions that are NOT explained by the crypto market factor (BTC); the systematic (BTC) component of a move persists/trends, but the residual (DOGE-specific) component of a SHORT-horizon overshoot mean-reverts as the overreaction fades and liquidity providers are compensated. By hedging out BTC beta we isolate and fade only the idiosyncratic overshoot, not the trend. Chosen to (a) fill the under-represented long_short bucket in a momentum-saturated portfolio, (b) be a cross-asset/pairs trade, and (c) target a symbol (DOGE) with genuinely large idiosyncratic vol so the reversal capture comfortably clears two legs of fees.

Implements the hypothesis literally: (1) beta hedge - rolling beta_lookback_days OLS beta of DOGE on BTC, hedge leg sized at beta * DOGE notional in USD so systematic exposure nets out; (2) idiosyncratic 3-day overshoot - residual returns eps = r_doge - beta*r_btc, summed over resid_horizon_days and standardized to a z-score; (3) REVERSAL sign - the position is the OPPOSITE sign of the residual overshoot (short after a positive residual run), which is the deliberate inverse of the residual-momentum idea; (4)……Show moreShow less

Implements the hypothesis literally: (1) beta hedge - rolling beta_lookback_days OLS beta of DOGE on BTC, hedge leg sized at beta * DOGE notional in USD so systematic exposure nets out; (2) idiosyncratic 3-day overshoot - residual returns eps = r_doge - beta*r_btc, summed over resid_horizon_days and standardized to a z-score; (3) REVERSAL sign - the position is the OPPOSITE sign of the residual overshoot (short after a positive residual run), which is the deliberate inverse of the residual-momentum idea; (4) ~5-day hold via a daily-bar time stop plus a reversion exit. The signal is continuous (the raw z-score is returned every bar, not gated behind the entry condition) and thresholded in should_enter. Venue is Binance USD-M futures because the strategy must SHORT one leg every trade; leverage 2.0 is genuinely consumed in position_size (equity * per_leg_pct * leverage) to fund the two-leg book with margin headroom. Fee viability: a 1.5-sigma 3-day residual on DOGE is roughly 8-10%, and even a partial fade clears the ~0.25% of DOGE-leg notional round-trip cost of both legs. Per learning L59, both legs are sized to matched USD notional (hedge = beta x DOGE notional) off a FIXED equity fraction rather than a compounding one, and either both legs go on or neither does, so the book cannot drift into naked directional exposure.

Negative expectancy on a decisive 266-trade / 6-year sample: profit_factor 0.623 (<1, gross-NEGATIVE before costs — commission is only 4.39% of gross, so this is a losing signal not fee-fragility), avg_trade_return_pct -0.83% (below ZERO, far under the 0.15% Binance USD-M floor), expectancy -$227.3/trade, total_return -60.7%, Sharpe -0.346 with CI [-1.18, 0.39] straddling zero, max drawdown 63.7%. Optimization cannot lift an edge that is gross-negative before costs. The code faithfully implements the beta-hedged……Show moreShow less

Negative expectancy on a decisive 266-trade / 6-year sample: profit_factor 0.623 (<1, gross-NEGATIVE before costs — commission is only 4.39% of gross, so this is a losing signal not fee-fragility), avg_trade_return_pct -0.83% (below ZERO, far under the 0.15% Binance USD-M floor), expectancy -$227.3/trade, total_return -60.7%, Sharpe -0.346 with CI [-1.18, 0.39] straddling zero, max drawdown 63.7%. Optimization cannot lift an edge that is gross-negative before costs. The code faithfully implements the beta-hedged idiosyncratic residual-reversal hypothesis (133 signaled = 133 submitted, balanced 133 long / 133 short, atomic dual-leg, point-in-time rolling OLS beta, residual z-fade + stop + time exit) so this is a fair test, and QA's concern is borne out on full history: the premise is FALSIFIED regime-wide — the book loses in the calm (-9.3%), normal (-16.5%) AND stressed (-49.6%) vol terciles and in 5 of 7 calendar years (2021 -42.7%, 2022 -14.7%, 2024 -12.5%, 2025 -8.5%). The DOGE-specific short-horizon residual overshoot does not mean-revert; hedging out BTC beta does not create edge. This is the beta/dollar-neutral single-name long-short reversal family (L45 multi_instrument 0/84, L55 sub-1.15-PF fade) that dies negative-expectancy. Tuning 3 parameters cannot rescue a PF-0.623 signal negative in every regime and most years — not worth 2 hours of walk-forward.

Implementation

Market-neutral DOGE-vs-BTC beta-hedged idiosyncratic short-term reversal on DAILY Binance USD-M bars. A rolling 60-day OLS beta of DOGE log-returns on BTC log-returns strips the systematic crypto-market component out of DOGE's moves; the residual (DOGE-specific) returns are summed over a 3-day window and z-scored against their own overlapping-sum distribution in the same lookback. That z-score is the continuous signal, recomputed every bar. |z| >= 1.5 means DOGE has idiosyncratically overshot: the strategy FADES it, shorting DOGE and buying beta-weighted BTC notional (or the mirror image), so the book carries ~zero net BTC beta and P&L is pure residual convergence. Exits when the residual reverts inside |z| < 0.4, when the hedged residual PnL hits the 8% stop, or after 5 daily bars. Legs are sized off a fixed 15% of equity times leverage 2 for DOGE and beta times that for the BTC hedge (never compounded off unrealized PnL), giving ~0.7x gross notional at ~2x margin headroom.

Verification Results

No code change required; confirm at full-backtest that per-position duration reports correctly for the two-leg book.

Reported avg_holding_bars is 0.0 while exposure_pct is 30.8% and avg_position_pct is 33.7% over 2248 days — the holding-period metric is almost certainly a multi-instrument reporting artifact (the DOGE leg is genuinely held ~days given the exposure figure), not a same-bar open/close bug. Flagging only so the analyst does not misread it as instant flat-flat churn.

Sandbox smoke test is deeply negative (total_return -60.5%, Sharpe -0.34, PF 0.62, avg_trade_return_pct -0.83%) on 266 trades, and critically it LOSES in all three vol terciles (calm -14%, normal -21%, stressed -44%) — the mechanism is not carried by any single regime. This is a single-parameter unoptimized draw, but the uniform-across-regime loss suggests the idiosyncratic-residual-reversal premise may simply be the wrong sign / no edge on DOGE (echoing prior single-name reversal deaths). Worth the analyst checking on full history whether the beta-hedged residual z-fade is gross-positive before any tuning; if PF stays sub-1.0 with commission only ~4% of gross, it is a falsified premise, not a fee-fragility issue.

Backtest Review

Sharpe: -0.35
Total return: -60.72%
Max drawdown: 63.67%
Trades: 266
Win rate: 50.4%
Profit factor: 0.62

Code faithfully implements the hypothesis: beta-hedged residual z-fade, 133 long / 133 short balanced, atomic dual-leg entry/exit, point-in-time rolling OLS beta, entries=signals (133 signaled = 133 submitted). Fair test.

Decisive sample: 266 trades over 2249 days (6+ years), metrics_reliable=true.

Gross-negative signal, not fee-fragility: profit_factor 0.623 (<<1) with commission only 4.39% of gross — the beta-hedged residual reversal LOSES before costs. QA's concern confirmed on full history.

avg_trade_return_pct -0.83% (below ZERO, far under the 0.15% Binance USD-M floor); expectancy -$227/trade; total_return -60.7%; Sharpe -0.346 with CI [-1.18, 0.39] straddling zero.

Premise falsified regime-wide: loses in calm (-9.3%), normal (-16.5%) AND stressed (-49.6%) vol terciles, and in 5 of 7 calendar years (2021 -42.7%, 2022 -14.7%, 2024 -12.5%, 2025 -8.5%). The idiosyncratic residual overshoot on DOGE does not mean-revert — it is the wrong sign / no edge.

This is the beta-neutral single-name long-short reversal family (L45 multi_instrument 0/84, L55 mean-reversion PF<1.15) that dies negative-expectancy; sits far below any survivor tier.

Outcome Summary

Analyst 'abandon' verdict at the pre-optimization backtest-review gate; later stages never ran.

A market-neutral long/short strategy that fades DOGE's idiosyncratic 3-day residual overshoot after hedging out its BTC beta...

Over a decisive 266-trade, 6-year backtest it lost money: total_return -60.7%, Sharpe -0.346, profit_factor 0.623, avg_trade_return_pct -0.83%, 63.7% max drawdown — gross-negative before costs, losing in all vol regimes and 5 of 7 years.