This repository is associated with the article Fasano and Rebaudo (2021). Variational inference for the smoothing distribution in dynamic probit models. The key contribution of the paper is outlined below.
We develop a variational Bayes approach, extending the partially factorized mean-field variational approximation introduced by Fasano, Durante and Zanella (2020) for the static binary probit model to the dynamic setting exploting the theoretical results in Fasano, Rebaudo, Durante and Petrone (2021).
This repository provides codes to implement the inference methods associated with such a new result. More precisely, we provide the R code to perform inference on the smoothing distribution under dynamic probit models with three different methods:
- i.i.d. sampling from the exact unified skew-normal distribution
- partially factorized mean-field variational Bayes (PFM-VB) approximation
- mean-field variational Bayes (MF-VB) approximation
Structure of the repository:
- the functions to implement the above methods can be found in the
Rsource fileFunctions.R - the financial dataset analyzed in the illustration can be found in
Financial.Rdata, while the original entire dataset is publicly available at Yahoo Finance - the code and a tutorial to reproduce the results in the paper is available at
Illustration.md