Skip to content

Regime Detection

Difficulty expert

Overview

Market regimes determine which strategies work. A strategy that excels in a trending market will fail in a ranging market. Detecting regime changes is critical for adaptive trading.

Regime Types

Regime Characteristics Best Strategies
Bull Low Vol Rising prices, low VIX Trend following, momentum
Bull High Vol Rising prices, high VIX Momentum with stops
Bear Low Vol Falling prices, low VIX Short selling, put buying
Bear High Vol Falling prices, high VIX Cash, volatility longs
Range Low Vol Sideways, low VIX Mean reversion, options selling
Range High Vol Sideways, high VIX Mean reversion with wider bands
Crisis Crash, extreme VIX Tail hedges, cash

Network Structure Across Timeframes

Most regime detectors look at one asset (or one market aggregate) at a time. A separate body of work treats the cross-asset correlation graph itself as the regime variable: when the shape of the correlation matrix changes, regime has changed — even if any single-asset volatility looks stable.

The Epps Effect

A long-documented stylized fact: measured correlations between assets decrease as sampling frequency increases. At daily horizons two assets may show ρ = 0.7; at 1-minute bars the same pair often shows ρ = 0.2 — not because the relationship is gone but because asynchronous trading, micro-noise, and partial information dispersion erode the linear comovement signal in short windows.

Sample horizon   Measured ρ(A, B)
─────────────────────────────────
Monthly          0.75
Weekly           0.68
Daily            0.55
Hourly           0.35
1-minute         0.18
1-second         0.05

Implications:

  1. The "true" correlation regime is horizon-dependent. A daily-correlation-based portfolio risk model overstates intraday diversification benefit.
  2. Reported by venue: 2022 crypto-market work shows the Epps effect flattens during stress episodes — cross-asset linkage strengthens at all horizons simultaneously, indicating regime transition.
  3. Use horizon-matched correlations for the decision being made. Hourly risk decisions need hourly correlations, not daily.

Centrality in Asset Networks

Build a graph where nodes = assets and edge weights = some correlation- or transfer-entropy-based linkage. Compute centrality measures (PageRank, eigenvector, betweenness) per node.

Empirical finding (crypto, 2022 sample): at daily horizons, ETH is increasingly central in the crypto correlation network — its volatility transmits to alts more reliably than BTC's. At certain shorter horizons, BTC's transmission role appears to shift toward mid-cap tokens (LTC, then certain exchange tokens), suggesting the "leader" varies by horizon and regime.

Putting it together

A practical multi-horizon regime monitor watches:

  1. Per-asset vol (existing methods above) — the "y-axis"
  2. Centrality of each asset across two horizons (e.g. daily and hourly) — the "x-axis"
  3. Epps gap = ρ_daily − ρ_intraday for major pairs — a narrowing gap is the cleanest single-number signal of cross-asset regime stress

A transition in any one of these warrants a re-check of strategy applicability before sizing changes. A transition in two is a strong signal to de-risk until the new regime stabilizes.

References for this section: 1. Network structure of cryptocurrency markets across timescales (2022). 2. Epps, T. W. (1979), "Comovements in stock prices in the very short run", JASA.

Practical Guidelines

  1. No Perfect Detector — All methods have lag and false signals
  2. Multiple Signals — Use several regime indicators
  3. Transition Periods — Regimes don't change instantly
  4. Backtest Each Regime — Validate strategy per regime
  5. Position Size by Regime — Reduce size in uncertain regimes
  6. Cash Is a Position — When in doubt, reduce exposure
  7. Monitor Continuously — Regimes can change quickly

Next Steps