Scribd
Upload
421 views

Favar Package

This add-in allows users to estimate Factor-Augmented Vector Regression (FAVAR) models using a two-step principal component approach. The user specifies the number of lags, factors, and observed and unobserved variables. The model relates unobserved factors and observed variables to a large data set using factor loadings. The add-in estimates the FAVAR model and allows users to perform impulse response analysis and identify structural shocks. It provides options for estimation via a graphical user interface or command line.

Uploaded by

Trevor Chimombe
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
421 views

Favar Package

This add-in allows users to estimate Factor-Augmented Vector Regression (FAVAR) models using a two-step principal component approach. The user specifies the number of lags, factors, and observed and unobserved variables. The model relates unobserved factors and observed variables to a large data set using factor loadings. The add-in estimates the FAVAR model and allows users to perform impulse response analysis and identify structural shocks. It provides options for estimation via a graphical user interface or command line.

Uploaded by

Trevor Chimombe
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Package Name: FAVAR

Author: Davaajargal Luvsannyam


Date: 2015.11.18

Default Proc Name: FAVAR


Default Menu Text: Factor Augmented VAR
Interface: Dialog and command line

Description

This add-in allows you to perform the estimation of Factor-Augmented Vector Regression
(FAVAR) models by using a two-step principal component approach (see more details in Bernanke,
Boivin, and Eliasz 2005).

To illustrate the main idea of this add-in, consider the following transition equation (VAR model):

𝐹 𝐹
[ 𝑡 ] = Φ(𝐿) [ 𝑡−1 ] + 𝑣𝑡 ,
𝑌𝑡 𝑌𝑡−1

where Φ(𝐿) is a lag polynomial of finite order d. The error term 𝑣𝑡 is mean zero with covariance matrix
𝑄. 𝑌𝑡 is vector of observable economic variables and 𝐹𝑡 is vector of unobserved factors.

It is assumed that the informational time series 𝑋𝑡 are related to the unobservable factors 𝐹𝑡 and the
observable factors 𝑌𝑡 by

𝑋𝑡 = Λ 𝑓 𝐹𝑡 + Λ𝑦 𝑌𝑡 + 𝑒𝑡

where Λ 𝑓 and Λ𝑦 are matrix of factor loadings and error terms 𝑒𝑡 are mean zero and will assumed either
weakly correlated or uncorrelated. The package estimates FAVAR model by a two-step principal
component approach. The identification of structural shocks in the transition equation is recursive
(cholesky) where all factors respond with a lag to change in the variable (e.g., federal fund rate) ordered
last in 𝑌𝑡 . Other identification schemes will be implemented in the next version.

Dialog

Upon running the add-in from the menus, a dialog will appear:
The first box lets you specify a number of lags while the second box specify the group of informational
time series variables (𝑋𝑡 ) from which the unobservable factors 𝐹𝑡 is estimated using first k principal
component. The next box specify slow-moving variables. Slow-moving variables are not assumed to
respond contemporaneously to unanticipated changes in the variable which is ordered last in 𝑌𝑡 . On the
next box enter a group of selected macroeconomic variables for impulse response analysis. On the next
box enter a vector of transformation code of Y and selected macroeconomic variables: 1=no
transformation; 4=logarithm; 5=first difference of logarithm. On the next (6th) box just put names of 𝑌𝑡
and selected macroeconomic variables. It should be text object which contain columns of names. On the
next box enter 𝑌𝑡 variable. It should be series, not group. Other boxes specifies some optional inputs.

References:

Bernanke, B. S., J. Boivin and P. Eliasz (2005), “Measuring the Effects of Monetary Policy: A Factor-
Augmented Vector Autoregressive (FAVAR) Approach”, Quarterly Journal of Economics 120.1:
387-422

Command line:

Syntax: favar(options) lags x(group of data) xs(group of slow-moving variables) xir(group of


selected variables) tcode(transformation code) name(Y and selected variable) @ endogenous
variables(Y)

E.g. favar 13 xdata xslow xir tcode yx_name @ ffr

Options:

argument explanations
factor number of factor
horizon number of steps for impulse response function
rep number of bootstrap replication
sample sample size
ci percent of confidence interval

E.g. favar(factor=5, horizon=40, rep=2000, ci=0.95) 13 xdata xslow xir tcode yx_name @ ffr

You might also like