Forecasting with high-dimensional bayesian vector autoregression
- Authors
- Mar?n Nicolalde, Eddson Eduardo
- Format
- MasterThesis
- Status
- publishedVersion
- Description
This document presents a brief recapitulation on Bayesian Vector Autoregression (BVAR) priors which serve as basis for High-dimensional models. For different prediction horizon, a forecasting exercise has being carried out using the Dummy-observations prior and Conjugate Stochastic Search Variable Selection. Five models with different size and deliberately large number of lags have been speci?ed. A comparison among the proposed models and a typical OLS estimated Small VAR, with BIC driven lag selection, shows that larger models generally outperform the benchmark in the sense of Mean Square Forecast Error.
El presente estudio utiliza t?cnicas de Machine Learning para grandes cantidades de datos (Big Data) con el objetivo de precedir en el desempe?o de 7 variables econ?micas en el corto y mediano plazo. Se emplearon t?cnicas provenientes de la escuela bayesiana de la estad?stica as? como simulaciones MCMC (Muestreo de Gibbs).
- Publication Year
- 2016
- Language
- eng
- Topic
- ECONOMETR?A
ESTAD?STICA BAYESIANA
ESTAD?STICA MULTIVARIADA
VECTORES AUTORREGRESIVOS
CADENAS DE MARKOV
- Repository
- Repositorio SENESCYT
- Rights
- openAccess
- License