THE DCC-CHOLESKY MULTIVARIATE STOCHASTIC VOLATILITY MODEL

Back to Page Authors: Benny Hartwig

---

Keywords: Bayesian methods, Multivariate stochastic volatility, Dynamic correlations, Gibbs sampling

---

Abstract: It is well-known that the estimate the stochastic covariance matrix of the Cholesky multivariate stochastic volatility model is not rotationally invariant. This paper shows that the procedure induces a nonlinear and asymmetric bias in the estimate, which exhibits a systematic pattern across orderings. This paper introduces the DCC-Cholesky multivariate stochastic volatility model which fixes the associated bias and renders rotational non-invariance an insignificant problem. The method can easily be implemented in existing routines as it requires only one additional step of sampling the states of volatility. This paper presents one example from the empirical literature in which this bias has a major impact on the conclusions.

---