Runtime (MCMC)

Markov chain Monte Carlo (MCMC) is a family of algorithms for sampling from posterior distributions.

In risk modeling, MCMC is typically used for calibration (inferring uncertain model parameters from data) rather than for the forward Monte Carlo simulation of losses.

---

Conceptual outline

Given parameters $\theta$ and observed data $D$, Bayesian inference defines:

$$P(\theta \mid D) \propto P(D \mid \theta)\,P(\theta)$$

MCMC constructs a Markov chain whose stationary distribution is $P(\theta \mid D)$.

---

Reference engine status

The reference engine in this repo does not implement a general MCMC calibration pipeline today.

If you need calibration, start with:

• Empirical lognormal calibration from single_losses (supported): see Runtime (Severity)

More advanced calibration workflows would be engine/tool-specific.