How do regime-dependent correlation models work?

Question

AcadiFi · Accepted Answer

Regime-dependent correlation models recognize that correlation parameters shift across market states. Three practical approaches: DCC-GARCH (Dynamic Conditional Correlation) lets correlation evolve continuously via a GARCH-like update rule Q_t = (1-a-b)*Q_bar + a*u_(t-1)*u_(t-1)' + b*Q_(t-1), then normalize to correlations. Captures gradual shifts. Markov-switching correlation models assume a hidden state S_t takes values 1..K with transition probabilities, and correlation matrix differs across states. Typical spec: K=2, state 1 is 'calm' with rho=0.25, state 2 is 'crisis' with rho=0.85, and transition probabilities p_11=0.98 and p_22=0.92 (crisis is persistent once entered). Fit via EM or Bayesian MCMC. Copula regime models combine a marginal fit with a regime-switching copula, allowing the dependence structure (not just correlation) to change. During the calm regime use Gaussian copula; during crisis use Student-t with 4 DoF for tail dependence. For a $3.2B multi-asset fund, Markov-switching found calm-to-crisis transition probability 2%/month (avg crisis every 4 years), aligning with observed 2008, 2011, 2020 events. Risk metrics computed under posterior-weighted mixture of regimes give more realistic VaR than static correlation. ```mermaid graph LR A[Calm
rho=0.25] -->|2%/mo| B[Crisis
rho=0.85] B -->|8%/mo| A ```

How do regime-dependent correlation models work?

Master Part II with our FRM Course

Related Questions

Practice Questions