Please use this identifier to cite or link to this item:
Casarin, Roberto
Grassi, Stefano
Ravazzolo, Francesco
van Dijk, Herman K.
Year of Publication: 
Series/Report no.: 
Tinbergen Institute Discussion Paper 15-084/III
A Bayesian nonparametric predictive model is introduced to construct time-varying weighted combinations of a large set of predictive densities. A clustering mechanism allocates these densities into a smaller number of mutually exclusive subsets. Using properties of Aitchinson's geometry of the simplex, combination weights are defined with a probabilistic interpretation. The class-preserving property of the logistic-normal distribution is used to define a compositional dynamic factor model for the weight dynamics with latent factors defined on a reduced dimension simplex. Groups of predictive models with combination weights are updated with parallel clustering and sequential Monte Carlo filters. The procedure is applied to predict Standard & Poor's 500 index using more than 7000 predictive densities based on US individual stocks and finds substantial forecast and econ omic gains. Similar forecast gains are obtained in point and density forecasting of US real GDP, Inflation, Treasury Bill yield and employment using a large data set.
Density Combination
Large Set of Predictive Densities
Compositional Factor Models
Nonlinear State Space
Bayesian Inference
GPU Computing
Document Type: 
Working Paper

Files in This Item:
2.73 MB

Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.