Regime-Conditional Value-at-Risk (Regime-CVaR)

Unlike static CVaR, Regime-CVaR models the loss distribution conditional on the current market state. Each regime—such as high-volatility bear or low-volatility bull—has its own distinct probability density function. This prevents the underestimation of tail risk during crises by ensuring the distribution used for CVaR calculation reflects the prevailing volatility clustering and correlation breakdown dynamics. The model captures the stylized fact that asset returns exhibit heteroskedasticity and skewness that differ markedly across regimes.

Regime-CVaR does not assume the current state will persist indefinitely. It uses the transition probability matrix to compute a forward-looking risk measure that weights potential future regimes by their likelihood. For example, if the model infers a bull regime but the transition probability to a crisis regime is elevated, the CVaR estimate will be higher than a naive static model would suggest. This forward-looking property is critical for preemptive risk management and margin setting.

Regime-CVaR inherits the coherent risk measure properties of standard CVaR while adding state-awareness. It satisfies:

• Sub-additivity: Diversification benefits are preserved within each regime
• Positive homogeneity: Scaling position sizes scales risk proportionally
• Translation invariance: Adding cash reduces risk by that amount
• Monotonicity: Portfolios with systematically worse outcomes have higher risk

This mathematical coherence ensures Regime-CVaR is suitable for regulatory capital calculation and portfolio optimization under the Basel framework.

Regime-CVaR serves as the objective function in state-dependent portfolio optimization. The optimizer minimizes the expected shortfall conditional on the inferred regime, producing allocations that adapt to:

• Regime-switching correlations: Assets that typically diversify may co-crash in crisis regimes
• Time-varying tail dependence: Captured via regime-switching copulas
• Regime-switching betas: Systematic risk exposure shifts across market states

The resulting risk parity or minimum Regime-CVaR portfolios are inherently more robust to correlation breakdown than static mean-variance optimized portfolios.

Validating Regime-CVaR requires regime-aware backtesting to avoid misleading conclusions. Standard Kupiec or Christoffersen tests for VaR violation clustering must be applied within each regime. Key diagnostics include:

• Conditional coverage tests: Do violations cluster within specific regimes?
• Regime-switching Sharpe ratio: Is risk-adjusted performance stable across states?
• Ergodic probability calibration: Does the long-run state frequency match empirical observations?

A well-calibrated Regime-CVaR model should exhibit uniform violation rates across all inferred market states, confirming that tail risk is accurately captured regardless of prevailing conditions.

A stochastic matrix defining the probabilities of switching between regimes. Key properties include:

• Persistence: Diagonal entries near 1.0 indicate stable regimes (e.g., a 0.95 probability of staying in a bull market)
• Ergodic probability: The long-run fraction of time spent in each state
• Absorbing states: Regimes from which escape is impossible (rare in financial models) This matrix directly impacts the forward-looking horizon of Regime-CVaR calculations.

The foundational quantile risk measure that Regime-CVaR extends. Standard VaR answers: What is the minimum loss at a given confidence level? A 95% 1-day VaR of $1M means losses will exceed $1M on only 5% of days. However, VaR is not coherent—it fails sub-additivity and ignores the shape of the tail beyond the threshold. Regime-CVaR addresses both limitations by conditioning on the state and averaging tail losses.

The coherent risk measure at the heart of Regime-CVaR. Unlike VaR, CVaR asks: Given that we've breached the VaR threshold, what is the average loss? Mathematically, it is the conditional expectation of losses in the tail. The Basel III framework mandates CVaR over VaR for market risk capital. Regime-CVaR sharpens this by computing CVaR from a state-specific loss distribution rather than a static unconditional one.

The empirical phenomenon motivating regime-switching risk models. Mandelbrot observed that large price changes cluster together and small changes cluster together, violating the i.i.d. assumption of simple models. This clustering creates distinct low-volatility and high-volatility regimes. A Regime-CVaR model captures this by estimating separate tail parameters for each volatility state, preventing the underestimation of risk during calm periods.

Markov-Switching GARCH combines regime detection with time-varying volatility dynamics. Unlike a static regime model, MS-GARCH allows volatility persistence within each regime. A crisis regime might exhibit both a higher baseline variance and stronger autoregressive volatility effects. Regime-CVaR built on MS-GARCH captures both the current state and the path-dependent volatility within that state for more accurate tail forecasting.