BayES is a software designed for performing Bayesian inference in some popular econometric models using Markov Chain Monte Carlo (MCMC) techniques. Bayesian inference traditionally requires skills and a lot of eﬀort from the part of the researcher, both in terms of mathematical derivations and computer programming. BayES provides canned procedures for Bayesian inference for a set of models, thus avoiding the time-consuming process of deriving complete/full conditionals and coding the samplers.
BayES is primarily menu driven, providing an intuitive user interface that enables ﬁrst-time users to run a model in a matter of minutes. However, it also features a rich scripting language that allows fast and eﬃcient communication with the program, while facilitating reproducibility of the analysis. Users familiar with other popular matrix languages will ﬁnd the transition to BayES’ scripting language a breeze.
Modern Bayesian inference is based on very computationally intensive techniques. However, MCMC methods are “embarrassingly parallel". BayES has built-in multi-threading support, thus making estimation much faster: when estimating a model, BayES will start as many threads as the number of chains requested by the user. No additional coding or knowledge of multi-threading processing is required.1
BayES is not simply another econometrics software package. Its emphasis is in models that are hard or impossible to estimate using classical (frequentist) inference and, thus, it is not a substitute for mainstream and well-established econometric software packages. On the other side of the spectrum, BayES is not competing with general or statistical scripting languages either. BayES is designed for the user who wants to perform Bayesian inference in a computationally involved problem, but who does not want go through the process of learning a new programming language for doing so.
The current version of BayES (version 2.5) supports the following models:
- linear models, including models for cross-sectional and panel data (random eﬀects, random coeﬃcients)
- stochastic frontier models for cross-sectional and panel data, using an array of possible speciﬁcations
- discrete-choice models, including models for cross-sectional and panel data and for binary or multi-response dependent variables
- simple models for ordinal data (ordered Probit and Logit)
- simple models for count data (Poisson and negative Binomial)
- type-I and type-II Tobit models
- linear seemingly unrelated regressions
- vector autoregressive models for time-series data
Version 2.5 of BayES runs on both 32-bit and 64-bit Microsoft® Windows® (Windows 7, 8, and 10) and Linux systems2 and on 64-bit macOS systems. Appendix A provides installation instructions under these systems.
1Future development plans include taking advantage of Graphical Processing Units (GPUs), which are present in most personal computers. Such a feature could increase speed by orders of magnitude.
2BayES has been tested on CentOS 7, Fedora 28, Debian 9.4, Ubuntu 18.04, openSUSE 15, and Linux Mint 19. It may be able to run on other distributions as well, but use at your own risk.