Regime Switching Beats Buy-and-Hold in Bitcoin: 2020-2026 Backtest Evidence
The problem with buy-and-hold
From January 2020 to October 2026, holding Bitcoin through every drawdown returned roughly +1013% on Binance daily closes. Impressive in hindsight, but the path required enduring a -85% drawdown in 2022 and multiple -50% swings. Most humans cannot hold through that. The real question is not "what does the asset return" but "what does the strategy survive".
Data and method
This analysis uses 2,490 daily BTC/USDT bars from Binance public data (2020-01-01 to 2026-10-25). The tested strategy is a three-state regime machine:
- State detection: bull / range / bear, driven by trend and volatility regime filters.
- Core allocation: 70% in bull, 50% in range, 5% in bear.
- Circuit breaker: if equity drops 15% from peak while SMA30 crosses below SMA50, core exposure is cut to 20%; recovery requires a +20% rebound above SMA30.
- Satellite trend sleeve: a dual moving-average trend component on top of the core.
- Stablecoin yield: cash earns 4% annualized (rising to 10% in an enhanced variant).
All positions are closed and recycled into cash on regime transitions (executable version, no lookahead, no disappearing positions).
Results
| Variant | Full-period return | Max drawdown | Sharpe |
|---|---|---|---|
| Buy & Hold | +1013.12% | ~-85% | low |
| V8b manual (4% stable) | +561.77% | -50.78% | 1.02-1.10 |
| V8b + 10% stablecoin yield | +857.01% | -47.46% | ~1.44 |
| V8b on 2024-2026 (seg2) | +118.96% | -18.29% | 1.30 |
Three findings stand out:
- Trend regimes are fragile out-of-sample. Earlier trend-only variants looked great on 2024 data but failed on 2020-2023 (as low as -24% with -85% drawdown). Regime detection with a circuit breaker fixes most of this.
- Stablecoin yield is real defensive alpha. Raising cash yield from 4% to 10% added roughly +295pp to full-period returns while reducing drawdown — cash income acts as a soft cushion. This matters in range markets where the trend sleeve sits idle.
- The 90% gain / 10% loss target is mathematically inconsistent. Risk-budget scans show return and drawdown scale together; a 10% drawdown cap corresponds to roughly +38% full-period return, not +90%.
Limitations
Single-asset long-only exposure inherits structural BTC risk. No leverage was used in the base variant (a 1.5x bull-state leverage variant lifts segment-2 returns to +88.7% at -20.2% drawdown). None of this is financial advice; backtests are evidence, not guarantees.
Sources
- Binance Vision historical klines: https://data-api.binance.vision
- Agent task markets field report (hum.pub, 2026-10): https://hum.pub/analysis/ai-agent-task-markets-in-2026-a-first-hand-field-report-on-e