Advanced

Can Forecasting Performance of the Bayesian Factor-Augmented VAR be Improved by Considering the Steady-State? An application to Swedish inflation

Lindvall, Andreas LU (2017) NEKN01 20171
Department of Economics
Abstract
This paper investigates whether the forecasting performance of Bayesian factor-augmented VAR (BFAVAR) models can be improved by incorporating an informative prior on the steady-state of the time series in the system. The BFAVAR model is compared to the extended steady-state BFAVAR in an application to forecasting Swedish inflation, making use of data from 1996 to 2016. Results show that the out-of-sample forecasting performance of incorporating an informative prior into the BFAVAR models increase compared to an autoregressive model. When comparing BFAVAR models with and without an informative prior on the steady-state, the BFAVAR model with an informative prior marginally outperform the BFAVAR model without the informative prior. The... (More)
This paper investigates whether the forecasting performance of Bayesian factor-augmented VAR (BFAVAR) models can be improved by incorporating an informative prior on the steady-state of the time series in the system. The BFAVAR model is compared to the extended steady-state BFAVAR in an application to forecasting Swedish inflation, making use of data from 1996 to 2016. Results show that the out-of-sample forecasting performance of incorporating an informative prior into the BFAVAR models increase compared to an autoregressive model. When comparing BFAVAR models with and without an informative prior on the steady-state, the BFAVAR model with an informative prior marginally outperform the BFAVAR model without the informative prior. The results of this paper indicate that most of the gains in forecasting performance by incorporating an informative prior on the steady-state are associated with longer forecasting horizons. (Less)
Please use this url to cite or link to this publication:
author
Lindvall, Andreas LU
supervisor
organization
course
NEKN01 20171
year
type
H1 - Master's Degree (One Year)
subject
keywords
Bayesian factor-augmented VAR, Steady-state, Inflation, Out-of-sample forecasting precision
language
English
id
8911717
date added to LUP
2017-07-10 13:53:24
date last changed
2017-07-10 13:53:24
@misc{8911717,
  abstract     = {This paper investigates whether the forecasting performance of Bayesian factor-augmented VAR (BFAVAR) models can be improved by incorporating an informative prior on the steady-state of the time series in the system. The BFAVAR model is compared to the extended steady-state BFAVAR in an application to forecasting Swedish inflation, making use of data from 1996 to 2016. Results show that the out-of-sample forecasting performance of incorporating an informative prior into the BFAVAR models increase compared to an autoregressive model. When comparing BFAVAR models with and without an informative prior on the steady-state, the BFAVAR model with an informative prior marginally outperform the BFAVAR model without the informative prior. The results of this paper indicate that most of the gains in forecasting performance by incorporating an informative prior on the steady-state are associated with longer forecasting horizons.},
  author       = {Lindvall, Andreas},
  keyword      = {Bayesian factor-augmented VAR,Steady-state,Inflation,Out-of-sample forecasting precision},
  language     = {eng},
  note         = {Student Paper},
  title        = {Can Forecasting Performance of the Bayesian Factor-Augmented VAR be Improved by Considering the Steady-State? An application to Swedish inflation},
  year         = {2017},
}