The core idea
Predicting whether Bitcoin will go up or down tomorrow is extraordinarily difficult. Even the best machine learning models barely exceed 50% directional accuracy on short timeframes (Fischer & Krauss, 2018). Markets are noisy, non-stationary, and adversarial.
A more tractable question: what type of market are we in right now? Trending, consolidating, overheated, or in capitulation? This matters because different strategies have known failure modes in the wrong environment. Momentum strategies bleed in sideways markets. Mean reversion gets destroyed in strong trends. The value of regime detection is avoiding the wrong strategy at the wrong time.
Three broad regimes
Bitcoin markets tend to cycle through three recognisable phases:
| Regime | Characteristics | Typical on-chain signals |
|---|---|---|
| Accumulation | Prices below long-term averages, subdued sentiment, long-term holders acquiring | MVRV below 1.0, SOPR below 1.0 |
| Expansion | Prices rising above averages, increasing volume, new participants entering | MVRV between 1.0 and 3.0, SOPR above 1.0 |
| Euphoria / correction | Prices far above historical norms, speculative activity peaking | MVRV above 3.0, supply in profit above 95% |
These are not rigid categories with clear boundaries. Markets transition gradually, and identification is always partly retrospective.
Detection methods
On-chain metrics are the most accessible approach. MVRV ratio, SOPR, and realised price provide direct measures of aggregate holder behaviour. Glassnode’s research shows MVRV has been below 0.8 for approximately 5% of all trading days and above 3.2 for approximately 6%, making these thresholds statistically meaningful cycle markers (Glassnode Insights, 2023).
Hidden Markov Models (HMMs), originally applied to economics by Hamilton (1989), model markets as switching between unobservable states, each with distinct return and volatility properties. A three-state HMM fitted on Bitcoin daily returns typically identifies: a low-volatility uptrend, a high-volatility downtrend, and a range-bound consolidation state.
The Hurst exponent measures whether a time series is trending (H > 0.5), random-walking (H ≈ 0.5), or mean-reverting (H < 0.5). This directly answers whether a momentum or mean reversion strategy fits current conditions.
What regime detection does for investors
For most people, regime detection means checking a handful of on-chain metrics monthly and adjusting behaviour accordingly:
- During accumulation phases (low MVRV, capitulation signals): increase purchase amounts or deploy capital more aggressively.
- During expansion: maintain standard strategy.
- During euphoria signals: reduce exposure or pause purchases.
No timing precision is needed. The goal is to avoid running full exposure during statistically overheated conditions and to increase exposure during historically undervalued periods.
Limitations
Small sample size. Bitcoin has completed roughly four full market cycles. Any model fitted to four data points warrants scepticism.
Structural change. ETFs, institutional custody, and regulated derivatives are changing Bitcoin’s market structure. Past on-chain relationships may weaken as more activity moves off-chain (CCN, 2025).
False signals. Regime transitions are ambiguous. Signals can indicate a change that reverses within weeks. Predefined rules prevent emotional reaction to false positives.
Not predictive. Regime detection describes current conditions based on observable data. It does not forecast future price movements.
Sources
- Fischer, T. & Krauss, C. (2018). “Deep learning with long short-term memory networks for financial market predictions.” European Journal of Operational Research, 270(2), 654-669.
- Hamilton, J.D. (1989). “A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle.” Econometrica, 57, 357-384.
- Glassnode Insights (2023). “Mastering the MVRV Ratio.” Statistical analysis of MVRV threshold distributions.
- CCN (2025). “How to Use the MVRV Z-Score to Spot Bitcoin Tops and Bottoms.” Discussion of ETF impact on on-chain metric reliability.