Skip to content

Study Guide

External events

Adaptive Subspace Shrinkage with Mixture Functional Horseshoe priors

research seminar

The seminar will be held in presence and online via teams: please click here 

abstract:  "Striking a balance between model complexity and parsimony is of crucial importance for producing accurate forecasts and structural inference. In this paper, we propose a non-parametric VAR model which relies on splines to approximate the unknown function between the response vector and its lags. This spline-based VAR is highly flexible and potentially overfits the data. To circumvent issues related to overfitting we introduce a prior which shrinks the non-parametric model towards a simpler model."