DogeShibMemecoinPairsSpreadMeanReversionLS4H

DOGE-SHIB Memecoin Pairs Spread Mean Reversion

A market-neutral (dollar-neutral) pairs trading strategy on the two most liquid memecoin futures on Binance: DOGEUSDT and 1000SHIBUSDT. Both are top-tier memecoins driven primarily by retail sentiment, social media trends, and crypto news cycles, making them ~0.70-0.85 correlated on daily returns but with frequent short-term divergences as one momentarily catches retail attention over the other (e.g., Elon tweet about DOGE, exchange listings for SHIB, ecosystem launches like Shibarium). The strategy maintains simultaneous offsetting positions sized to equal dollar notionals, so net beta to crypto is approximately zero. Entry uses a single-condition z-score signal on the DOGE/SHIB price ratio. Architecturally identical to the two production pairs strategies (Sharpe >6) — only 4 core parameters to minimize overfitting risk and avoid the frozen-signal failures that have plagued multi-condition strategies. Adds memecoin/retail-sentiment exposure to the portfolio, complementing the existing L1, majors-ratio, and DeFi pairs trades.

Implements the hypothesis exactly: a market-neutral memecoin pairs mean-reversion using a single-condition z-score on the DOGE/1000SHIB price ratio, architecturally identical to the two production pairs strategies (Sharpe >6). It uses the base class's same-timeframe cross-leg alignment barrier so the z-score and both fills use contemporaneous DOGE and SHIB prices, and keeps the log-ratio in one synced buffer (appended once per processed primary bar) so rolling stats stay index-aligned — avoiding the independent-trimming garbage-alignment trap. calculate_signal returns the continuous raw z-score every bar (varies bar-to-bar, no gating) and should_enter/should_exit threshold it in matching z-units, avoiding the frozen-signal failure mode. Only 4 core signal params (lookback, z_entry, z_exit, z_regime) minimize overfitting. Equal 40%-equity notional per leg makes it dollar-neutral (~zero net beta), and per-trade edge (2-sigma spread reversion on volatile memecoins, typically >1%) comfortably exceeds the ~0.10% round-trip taker cost across both legs. Futures venue (BINANCE) is required because the strategy shorts one leg; leverage stays 1.0 since gross exposure is ~80% of equity.

Not worth optimizing: with a large, statistically adequate sample (738 trades over 6 years) the strategy shows a persistent, catastrophic negative edge — total_return -82.8%, Sharpe -1.18 with the entire CI below zero (sharpe_ci_high -0.29), profit_factor 0.79, expectancy -$112/trade, max_drawdown 85.4%, PSR 0.0009 — and loses in every regime (2021 -54%, 2022 -23%, 2024 -43%, 2025 -16%). This is not an outlier problem but a premise problem: the DOGE/SHIB log-ratio is not stationary/cointegrated. Memecoins are highly return-correlated (~0.7-0.85) but decouple and TREND apart (skew -8.5, kurtosis 138, -38.6% blow-out days, largest_loss -$36.8k), so a fixed-window z-fade systematically shorts the winner and longs the loser — the exact non-stationarity failure the hypothesis said it wanted to avoid (ETH/BTC) but never tested for on DOGE/SHIB. The only tunable levers are the 4 z-thresholds, and no entry/exit threshold turns a trending spread into a mean-reverting one, so 2 hours of optimization would only confirm the negative edge. The pairs mechanism is validated on the production siblings because THOSE pairs are actually cointegrated; the fix here is a genuinely cointegrated pair, which is a hypothesis-level retarget for the Research Lead, not a developer parameter iteration. FAILURE PATTERN: high return-correlation is not cointegration — a dollar-neutral z-fade on a correlated-but-trending memecoin ratio bleeds persistently across a large sample (PF 0.79, Sharpe CI entirely negative) because the spread trends rather than reverts; the pairs template only clears the bar on empirically stationary pairs, and DOGE/SHIB is not one.

Dollar-neutral long-short pairs trade on the two most liquid memecoin perps (DOGEUSDT primary, 1000SHIBUSDT hedge) on Binance USD-M futures, 4H bars. Each aligned close computes the log-ratio R=log(DOGE/1000SHIB) and a 40-bar rolling z-score. z>=+2 shorts DOGE / longs SHIB; z<=-2 longs DOGE / shorts SHIB, each leg at 40% equity notional (~80% gross, leverage 1.0). Exits on reversion (|z|<=0.5), divergence stop (|z|>=3.5), 30-bar time stop, or 12% per-leg drawdown stop. Both legs are held in lockstep so the book has ~zero net crypto beta and isolates the idiosyncratic DOGE/SHIB spread.

Clean, low-parameter (4 core z-params), correctly dollar-neutral construction with proper same-timestamp cross-leg alignment; ample trade sample (738 trades)

Advances the market-neutral and pairs quotas the portfolio wants — the mechanism itself is validated on the production cointegrated-pairs siblings

Persistent, catastrophic negative edge across a large sample: total_return -82.8%, Sharpe -1.18 with the ENTIRE CI below 0 (sharpe_ci_high -0.29), profit_factor 0.79, expectancy -$112/trade, max_drawdown 85.4%, PSR 0.0009

Losing in every regime: annual returns 2021 -54%, 2022 -23%, 2024 -43%, 2025 -16% (only trivially positive 2023/2026); rolling Sharpe negative for most of 6 years — a consistent bleed, not an outlier

The DOGE/SHIB log-ratio is not stationary/cointegrated: it trends as memecoins decouple (skew -8.5, kurtosis 138, largest_loss -$36.8k, -38.6% blow-out days), so z-fade systematically shorts the winner and longs the loser — the exact non-stationarity failure the hypothesis claimed to avoid but never tested for this pair

The only tunable levers are 4 z-thresholds, which cannot convert a trending spread into a reverting one — this is a pair-selection/premise failure, not a parameter issue

DogeShibMemecoinPairsSpreadMeanReversionLS4H tried to add memecoin exposure to the portfolio by porting the production dollar-neutral pairs template to a DOGE/1000SHIB 4H z-score fade, betting the two correlated memecoins would revert after short-term divergences. On its first iteration the backtest produced a catastrophic, consistent negative edge across 738 trades — a -82.79% return, Sharpe of -1.18 with a fully negative confidence interval, 85.4% drawdown, and losses in nearly every year (2021 -54%, 2022 -23%, 2024 -43%, 2025 -16%). The analyst abandoned it at the backtest-review gate, concluding the DOGE/SHIB ratio trends rather than reverts (skew -8.5, kurtosis 138), so the strategy structurally shorted the winner and longed the loser — a pair-selection/premise failure no z-threshold tuning could repair. Optimization and all downstream stages were skipped, and the fix was flagged as a hypothesis-level retarget to a genuinely cointegrated pair rather than a developer parameter change.

High return correlation (~0.70-0.85) is not cointegration — a mean-reversion z-fade only works on an empirically stationary spread, and correlated-but-trending pairs like DOGE/SHIB must be tested for stationarity before assuming the pairs template will transfer.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate on an 'abandon' verdict, failing every criterion (Sharpe, max drawdown, total return, profit factor, sharpe_ci_high); the analyst judged the DOGE/SHIB log-ratio non-stationary/non-cointegrated, so the z-fade systematically shorted the winner and longed the loser, and the four z-thresholds could not fix a premise-level flaw — the analyst, risk, and optimization stages were never reached.

A dollar-neutral pairs trade on Binance USD-M futures that faded a rolling 40-bar z-score of the DOGE/1000SHIB log-price ratio on 4H bars, shorting the rich leg and longing the cheap leg to isolate the idiosyncratic memecoin spread with near-zero net crypto beta.

Over 986 data days and 738 trades it lost money persistently: total return -82.79%, Sharpe -1.18 with the entire confidence interval below zero (CI high -0.29), profit factor 0.79, expectancy -$112/trade, 85.4% max drawdown, ~50% win rate, and a probabilistic Sharpe of 0.0009.