okx-codex-trader/reports/strategy-expansion

Strategy Expansion Summary

Date: 2026-05-08

This directory records the non-ultrashort strategy exploration round.

Conclusion

The only branch worth continuing is the low-risk BTC/ETH rotation strategy:

• trend_basket-1h-lb720-tr1440-bt2880-rb168-top2-mm0.000-bm0.000-vw336-vc0.0300-lev0.75-exp0.50-vt0.2000
• Total return: 152.02%
• Annualized return: 15.64%
• Max drawdown: 7.72%
• Calmar: 2.03
• Trades: 45
• Win rate: 57.78%
• Profit factor: 7.38
• 3y return: 54.98%
• 1y return: 15.91%
• 6m return: -0.28%
• 3m return: -0.28%

Audit status: passed. The equity curve uses next-bar execution and no same-row lookahead was found. Trade statistics were repaired to use the same executed turnover fee basis as the equity curve.

Branch Decisions

• Rotation: keep. The raw rotation search had excessive drawdown, but the risk-refined variant is plausible and passed audit.
• Mean reversion: keep only as secondary research. ETH 4H RSI2 works historically, but recent 6m/3m were negative. The eq_days30 self-disable rule avoided the recent drawdown but did not improve portfolio Calmar versus rotation alone.
• Trend swing: exclude for now. The attractive daily EMA candidates have only 12 trades; 4H variants have more samples but fail recent-window checks.
• Rotation + mean-reversion portfolio: exclude for now. 70/30 improves versus 50/50, but still underperforms standalone rotation-risk by Calmar.

Next Work

Promote rotation-risk to read-only signal observation before considering any live execution. Required checks:

• Build current signal intent for the selected rotation-risk strategy.
• Compare local and remote candle alignment for 1h aggregated BTC/ETH data.
• Observe live signals without orders for at least several rebalances.
• Add explicit execution design only after signal state and rebalance timing are proven.