Structural VAR (SVAR) toolbox for bayesian VAR estimation and a range of identification methods
VAR toolbox that allows a bayesian estimation of the VAR coefficients with changeable priors (edit in BVAR in "Functions" folder).
Allows multiple identifications:
- Spectral and Limited Spectral restrictions - identifying the shock that maximizes the share of variance in a desired frequency band - see Dieppe, Francis, and Kindberg-Hanlon 2021 (JEDC) and Dieppe, Francis, and Kindberg-Hanlon 2021 (ECB Working Paper)
- The long-run restriction of Blanchard & Quah 1989/Gali 1999
- The Max Share approach of Francis et. al. 2014
- A Cholesky identification.
- Sign and zero restrictions
- Restrictions on the share of FEVD explained by the different shocks.
The bayesian VAR function outputs:
- Impulse responses (IRFs)
- Forecast error variance decompositions (FEVDs)
- Historical decompositions
- Structural shock series
Full examples are provided.
The file RunMain.m estimates a VAR consisting of US labor productivity, employment, investment and consumption as a share of GDP, inflation, and long-term bond yields. It then demonstrates how to estimate a technology shock using the Spectral, Limited Spectral, long-run, and Max Share restrictions and plots them side by side.
The file RunMain_signandzero.m estimates a VAR consisting of US labor productivity, employment, inflation, and long-term bond yields. It then demonstrates a method of identifying a technology shock, demand shock, monetary policy shock, and a supply shock using sign and zero restrictions. It also demonstrates how to impose FEVD magnitude restrictions. For example, technology shocks are assumed to have a larger share of the variance of labor productivity at the 5 year horizon, while demand shocks are assumed to dominate the FEVD of labor productivity in the first year. This is for demonstrative purposes rather than for any particular theoretical reason.