Time-varying parameters vector autoregressions with multivariate stochastic volatility

• Applicant: Professor Dr. Roman Liesenfeld
• Subject Area: Statistics and Econometrics
• Term: from 2018 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 394413895

Vector autoregressions (VARs) are widely used to provide a systematic way to capture richdynamics in multiple time series, and thus have become a widely used tool for modelingand forecasting in macroeconomics and nance. In recent years, VARs with time-varyingparameters and stochastic volatility (TVP-VARs) have received increasing attention becauseof an ever-growing body of evidence questioning the usual assumption of stable parameters,especially, in time series where data encompass several decades, increasing the probabilityof changes in the dynamic structure of the underlying stochastic process. Like standardVARs, TVP-VARs oer an approach to data description, forecasting, structural inferenceand policy analysis, and as such can be used to analyze a large number of economic problems.However, standard multivariate stochastic volatility (MSV) specications used for the timevaryingcovariance matrix of the TVP-VAR innovations are not invariant w.r.t. the orderingof the time series variables in the VAR system so that this ordering undesirably aects theinference results.In this project we consider ordering-invariant yet exible MSV models based on Wishartprocesses. They have become popular in nancial applications but the statistical inferencerequired for their empirical application remains to be a challenging task. We aim at developingMonte-Carlo based inference procedures for a maximum likelihood analysis of thoseWishart MSV models and then will extend the inference procedures for the analysis of TVPVARswith Wishart MSV specications for the covariance matrix of the innovations. Thedeveloped methods will be implemented for thorough nancial and marcoeconomic applications.

• DFG Programme: Research Grants
• International Connection: Brazil
• Cooperation Partner: Professor Dr. Guilherme Valle Moura