Regime Sensitivity & Volatility Dependency
Trading Intelligence
11 min read
regimeSensitivityvolDependency
Quantify how much your strategy depends on specific market regimes and volatility levels.
Loading
11 min read
Quantify how much your strategy depends on specific market regimes and volatility levels.
A strategy that performs brilliantly in trending markets but collapses in choppy conditions is not robust — it is regime-dependent. Understanding Regime Sensitivity and Volatility Dependency helps you quantify this fragility.
This is lesson 1 of 3 in Robustness Metrics. Prereqs: Edge Degradation, Variance & Standard Deviation, Biases in Backtesting. Next: Autocorrelation of Returns.
A market regime is a sustained period where price behavior follows a consistent pattern:
Regimes are not discrete categories — they exist on a spectrum and can overlap (e.g., high-volatility trending, low-volatility mean-reverting).
Regime Sensitivity measures how much your strategy's performance varies across different market conditions.
Calculation:
RS = StdDev(expectancy_per_regime) / |Mean(expectancy_across_regimes)|
Regime Sensitivity Score interpretation bands. Highlighted band triggers investigation.
| Score | Interpretation |
|---|---|
| <0.3 | Low sensitivity — performs consistently across regimes |
| 0.3 - 0.6 | Moderate — measurable regime drift; tradable but expect uneven equity |
| 0.6 - 1.0 | High — significant regime dependency |
| > 1.0 | Extreme — strategy only works in specific conditions |
A score above 0.6 should trigger investigation. Either your strategy needs regime filters, or you need to accept that it will have extended drawdown periods when the regime shifts (see Equity R-Squared for a complementary stability metric, and Biases in Backtesting for the upstream cause). Caveat: a filter chosen because it would have avoided the bad regime in-sample is curve-fitting in a costume — validate any filter on out-of-sample data with at least 30 trades per regime.
Volatility Dependency specifically measures the correlation between your strategy's returns and market volatility (typically measured by ATR or realized volatility).
Calculation:
VD = Spearman_rho(r_t, sigma_t)
Compute on vol-normalized returns; otherwise any fixed-notional sizing rule will leak σ into the numerator and inflate the correlation mechanically.
Volatility Dependency interpretation. Highlighted band is the ideal target for most strategies.
| Correlation | Meaning |
|---|---|
| +0.5 to +1.0 | Strategy thrives in high volatility |
| +0.2 to +0.5 | Mild positive dependency |
| -0.2 to +0.2 | Volatility-neutral (ideal for most strategies) |
| -0.5 to -0.2 | Mild negative dependency |
| -1.0 to -0.5 | Strategy suffers in high volatility |
Trend-following strategies typically show positive volatility dependency — they need big moves to generate profits. Mean-reversion strategies often show negative dependency — high volatility blows through their levels.
Regimes cluster — vol begets vol, trends persist. A 5-year sample is rarely 5 independent years; it is often 2–3 macro regimes stacked. If your sample over-represents one regime (the 2020–2021 crypto bull run is the canonical example), your expectancy is conditioned on that regime and tells you almost nothing about the next one.
The textbook mitigation is volatility targeting — size positions so that ex-ante portfolio σ is constant (e.g., 15% annualized). This neutralizes the most mechanical part of vol-dependence; what remains is genuine regime-edge variation.
If you know your strategy has high regime sensitivity, you can anticipate drawdowns when the regime shifts. This is not a prediction — it is risk management through self-awareness.
Combining a high-vol-dependent strategy with a low-vol-dependent strategy can smooth your overall equity curve. This is the foundation of strategy diversification.
During unfavorable regimes, you can reduce position size rather than stop trading entirely. This keeps you in the market for when the regime returns, while limiting damage.
Define clear rules: "If ATR drops below the 20th percentile of its 60-day range, reduce size by 50%." This is regime-adaptive position sizing.
Per-regime trade statistics. Highlighted row has N<30 and is statistically unreliable.
| Regime | N trades | Expectancy (R) | Win % | Avg win | Avg loss | MaxDD |
|---|---|---|---|---|---|---|
| Trend / High-vol | 84 | +0.42 | 47% | 1.9R | -1.0R | -8% |
| Trend / Low-vol | 31 | +0.05 | 41% | 1.2R | -0.9R | -11% |
| Range / High-vol | 22 | -0.18 | 35% | 1.4R | -1.0R | -19% |
| Range / Low-vol | 67 | +0.31 | 58% | 0.9R | -1.0R | -6% |
Flag any cell with N < 30 as statistically unreliable. 4. Compute the Regime Sensitivity Score: Use coefficient of variation across regime expectancies. 5. Plot returns vs volatility: A scatter plot of weekly returns vs weekly ATR immediately reveals dependency.
Many traders backtest across a single regime (e.g., the 2020-2021 crypto bull run) and believe they have an edge. When the regime changes, the strategy fails — not because the backtest was wrong, but because it was incomplete.
The antidote is simple: test across all available regimes, and measure how much performance varies. A truly robust strategy should be at least marginally profitable in every regime, not just spectacularly profitable in one.
Use the Monte Carlo simulator to see how the same edge parameters produce different equity paths. High regime sensitivity means some paths thrive while others collapse.
The coefficient of variation of expectancy across regime buckets. Higher = more fragile.
At least 30 — fewer and per-regime expectancy is dominated by noise. Flag any bucket with N < 30 as statistically unreliable.
It removes the mechanical part of vol-dependence by holding ex-ante portfolio σ constant. Structural regime-edge variation — where your signal itself behaves differently across regimes — remains.
No. Beta is correlation to a benchmark return; vol-dependence is correlation to the magnitude of returns (i.e., σ itself).
Regime Sensitivity and Volatility Dependency are not exotic metrics — they are survival metrics. They answer the most important question in strategy development: "Will this still work when market conditions change?" If you cannot answer that question with data, you are trading blind.