Factor-Augmented Vector Autoregression (FAVAR)
Factor-Augmented Vector Autoregression · Also known as: factor-augmented VAR, FAVAR model, Faktör Artırımlı VAR (FAVAR)
FAVAR is a multivariate time-series model that first compresses information from a very large set of variables into a few common factors, then includes those factors alongside the observed variables in a vector autoregression. It was introduced by Bernanke, Boivin and Eliasz in 2005 to study monetary policy using hundreds of macroeconomic indicators at once.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use FAVAR when you have a large number of continuous, stationary time series (potentially p far exceeding the number of periods T) and want the explanatory richness of all of them inside a compact VAR. It suits forecasting, relationship analysis, and explanation of macroeconomic dynamics. It needs a credible factor structure in the data, stationarity, and a reasonable sample (at least about 60 periods). It is not appropriate when only a few series are available or when no common factor structure underlies them.
Strengths & limitations
- Brings the information of hundreds of variables into a small, estimable VAR, avoiding the curse of dimensionality.
- Produces impulse responses for many variables at once, giving a fuller picture of how shocks propagate through the economy.
- Mitigates the omitted-variable and 'price puzzle' problems that plague small VARs by conditioning on broad information.
- Requires that a genuine factor structure exists; if the variables do not share common factors the compression is meaningless.
- All series must be stationary and the sample reasonably long (at least about 60 periods).
- The factors are estimated, not observed, so they can be hard to interpret and add estimation uncertainty to the second stage.
Frequently asked
How is FAVAR different from a plain VAR?
A plain VAR can only include a few variables before it runs out of degrees of freedom. FAVAR first summarises a large dataset into a small number of estimated factors and then puts those factors into the VAR, so the system stays small while still reflecting hundreds of series.
How are the factors estimated?
Most commonly by principal component analysis on the large panel of standardised, stationary series. The leading components serve as the common factors that drive the whole dataset.
How many factors should I include?
There is no single answer; information criteria and the share of variance explained guide the choice. Too few factors throw away signal, while too many reintroduce the high-dimensionality problem the method was meant to avoid.
Do my series need to be stationary?
Yes. Both the factor extraction and the VAR require stationary inputs, so non-stationary series should be transformed (for example differenced) before estimation.
Sources
- Bernanke, B. S., Boivin, J. & Eliasz, P. (2005). Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach. The Quarterly Journal of Economics, 120(1), 387-422. DOI: 10.1162/0033553053327452 ↗
- Stock, J. H. & Watson, M. W. (2002). Macroeconomic Forecasting Using Diffusion Indexes. Journal of Business & Economic Statistics, 20(2), 147-162. DOI: 10.1198/073500102317351921 ↗
How to cite this page
ScholarGate. (2026, June 1). Factor-Augmented Vector Autoregression. ScholarGate. https://scholargate.app/en/econometrics/favar
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Markov-Switching ModelEconometrics↔ compare
- OLS RegressionEconometrics↔ compare
- Threshold and Smooth-Transition VAREconometrics↔ compare
- VAR ModelEconometrics↔ compare