recent-regime-exploration-summary.md 4.7 KB

Recent-Regime Strategy Exploration Summary

Local data end: ETH/BTC 15m 2026-05-17 06:45 UTC. 3m cache ends 2026-04-29; 5m cache ends 2026-05-10, so live-relevant recent exploration should prefer 15m until lower timeframes are refreshed.

Market Read

  • Last 30d: ETH -6.26%, BTC +4.25%, ETH/BTC -10.08%: ETH is materially weaker than BTC.
  • Same-bar BTC/ETH correlation remains high around 0.88-0.90, but BTC lead/lag edge is weak, so BTC should be a regime filter at most, not the main alpha.
  • Current ETH BB squeeze is active, but recent squeeze follow-through quality is low; breakouts need fast confirmation and tight exits.

Recent 120d Candidate Scan

family name 90d_total_return 30d_total_return 14d_total_return 7d_total_return max_recent_drawdown recent_trades
squeeze_breakout squeeze-q0.2-th0.004-sl0.008-tp0.016 -0.80% 10.23% 5.50% 0.84% 4.18% 27
squeeze_breakout squeeze-q0.15-th0.004-sl0.008-tp0.016 -1.32% 3.93% 2.62% 0.84% 4.18% 19
squeeze_breakout squeeze-q0.2-th0.007-sl0.006-tp0.012 -3.83% 3.05% 3.18% 0.79% 3.09% 21
squeeze_breakout squeeze-q0.25-th0.007-sl0.006-tp0.012 -8.05% 3.22% 2.34% 0.79% 2.13% 24
squeeze_breakout squeeze-q0.2-th0.004-sl0.006-tp0.012 -11.10% 2.98% 3.42% 0.27% 4.58% 37

Read: the top squeeze-breakout row is attractive over 30d/14d/7d, but this scan only used a recent 120d candidate window. Treat as a tactical observation candidate, not robust strategy evidence.

3y Squeeze Refinement

name total_return max_drawdown trades profit_factor return_90d return_30d return_14d return_7d
sqref-q0.2-tb4-tr0.004-sl0.008-tp0.016-h48-long-none-cd8 -62.88% 66.80% 617 0.74 -4.26% 8.29% 3.84% 0.37%
sqref-q0.2-tb4-tr0.004-sl0.006-tp0.012-h48-long-none-cd8 -64.51% 66.70% 691 0.70 -5.92% 3.79% 2.33% 0.17%
sqref-q0.15-tb4-tr0.004-sl0.008-tp0.016-h48-long-none-cd8 -40.53% 45.70% 475 0.82 -2.90% 1.64% 1.58% 0.37%
sqref-q0.2-tb4-tr0.004-sl0.008-tp0.016-h48-long-directional-cd8 -32.54% 36.18% 248 0.75 -5.83% 0.52% 0.16% 0.00%
sqref-q0.2-tb4-tr0.004-sl0.006-tp0.012-h48-long-directional-cd8 -40.68% 42.35% 278 0.64 -2.17% 0.33% 0.16% 0.00%

Read: extending the same family to 3y makes it unattractive. Even the best 30d rows have large full-period losses and profit factor below 1.0.

Mean-Reversion / False-Breakout Search

name symbol bar side_mode total_return max_drawdown trades win_rate profit_factor
eth-5m-fbmr-short-rl48-cw288-cq0.2-sw0.0016-sl0.006-tp0.008-h18 ETH-USDT-SWAP 5m short 0.32% 3.65% 79 55.70% 1.03
eth-5m-fbmr-bidir-rl48-cw288-cq0.2-sw0.0016-sl0.006-tp0.008-h18 ETH-USDT-SWAP 5m bidir -0.51% 6.78% 152 57.24% 0.99
eth-5m-fbmr-long-rl48-cw288-cq0.35-sw0.0016-sl0.006-tp0.008-h18 ETH-USDT-SWAP 5m long -0.71% 4.91% 146 57.53% 0.98
eth-5m-fbmr-long-rl48-cw288-cq0.2-sw0.0016-sl0.006-tp0.008-h18 ETH-USDT-SWAP 5m long -0.82% 5.26% 73 58.90% 0.95
btc-5m-fbmr-long-rl24-cw144-cq0.2-sw0.0016-sl0.006-tp0.008-h18 BTC-USDT-SWAP 5m long -1.54% 4.04% 37 59.46% 0.76

Read: this direction is weak. Best result is near flat over 36 months and loses over recent 30/90d.

BTC Regime Router v2

name total_return max_drawdown min_recent_total_return trades win_rate profit_factor
regime-router-v2-ts60-ml18-vol0.03-mom0.02-rz0.25-h12 -60.02% 79.57% -3.21% 643 38.72% 0.96
regime-router-v2-ts60-ml18-vol0.03-mom0.02-rz0.25-h24 -60.02% 79.57% -3.21% 643 38.72% 0.96
regime-router-v2-ts60-ml18-vol0.04-mom0.02-rz0.25-h12 -64.27% 81.75% -3.21% 649 38.52% 0.95
regime-router-v2-ts60-ml18-vol0.04-mom0.02-rz0.25-h24 -64.27% 81.75% -3.21% 649 38.52% 0.95
regime-router-v2-ts120-ml18-vol0.03-mom0.02-rz0.25-h12 -32.74% 63.59% -14.87% 557 39.14% 1.00

Read: no live candidate. It trades often but loses after fees; short regimes are the main drag.

Decision

  • Do not replace the current live strategy with mean-reversion, BTC router, or the 3y squeeze-refine candidates.
  • Keep the 120d squeeze-breakout candidate as a read-only observer candidate only: squeeze-q0.2-th0.004-sl0.008-tp0.016.
  • Next useful exploration path: refresh 3m/5m through today, then test whether the 120d squeeze edge exists on lower timeframes and whether execution delay/slippage destroys it.