Bayesian Methods for Empirical Macroeconomics with Big Data
Bayesian econometric methods are increasingly popular in empirical macroeconomics. They have been particularly popular among macroeconomists working with Big Data (where the number of variables under study is large relative to the number of observations). This paper, which is based on a keynote address at the Rimini Centre for Economic Analysis' 2016 Money-Macro-Finance Workshop, explains why this is so. It discusses the problems that arise with conventional econometric methods and how Bayesian methods can successfully overcome them either through use of prior shrinkage or through model averaging. The discussion is kept at a relatively non-technical level, providing the main ideas underlying and motivation for the models and methods used. It begins with single-equation models (such as regression) with many explanatory variables, then moves on to multiple equation models (such as Vector Autoregressive, VAR, models) before tacking the challenge caused by parameter change (e.g. changes in VAR coefficients or volatility). It concludes with an example of how the Bayesian can address all these challenges in a large multi-country VAR involving 133 variables: 7 variables for each of 19 countries.
The Review of Economic Analysis is committed to the open exchange of ideas and information.
Unlike traditional print journals which require the author to relinquish copyright to the publisher, The Review of Economic Analysis requires that authors release their work under Creative Commons Attribution Non-Commercial license. This license allows anyone to copy, distribute and transmit the work provided the use is non-commercial and appropriate attribution is given.
A 'human-readable' summary of the licence is here and the full legal text is here.