bmgarch: An R-Package for Bayesian Multivariate GARCH models

Abstract

Generalized autoregressive conditional heteroskedasticity models (GARCH) and their multivariate extension (MGARCH) are part of the economists’ toolbox ever since their introduction in the early 1980s (Bollerslev, 1986; Engle, 1982; Tsay, 2013). Typically, the goal is to generate forecasts of volatility and covolatility for the next day or the near future in time series of assets or market indices. While GARCH models are primarily used in the econometric context, they can be used to capture and forecast heteroskedasticity in any time series. In fact, Rast et al. (2020) presented a parameterization for predicting and capturing within-person variability in human behavior in intensive longitudinal designs.