A0809
Title: The importance of the tails in high dimensional macroeconomic forecasting
Authors: Anna Simoni - GENES - CREST (France) [presenting]
Matteo Mogliani - Banque de France (France)
Luca Rossini - University of Milan (Italy)
Abstract: Forecasting beyond the central tendency of future values of macroeconomic and finance series is important for policymakers, as it allows for quantifying the likelihood of tail (upside or downside) risks. Nowadays, policymakers have available rich datasets that potentially contain valuable information for predicting tail events but that are difficult to exploit due to the large dimension. A large-dimensional quantile regression model is proposed to forecast the tails of the conditional distribution of future values of the target variable, and at the same time, the large dimension is dealt with by exploiting a bi-level sparse group structure as well as a regularization scheme. A Bayesian procedure is proposed based on the asymmetric Laplace distribution, which is shown to have optimality properties. While the group structure is assumed to be known, the sparsity structure is not, and it is shown that the approach learns adaptively about which groups and which variables are active. Interestingly, the sparsity structure is quantile-dependent. Frequentist asymptotic properties of the procedure are studied. Finite sample properties are illustrated through Monte Carlo experiments. Finally, the performance of the procedure is shown with real macroeconomic data.
