Cookies on Macro by Mark.We use essential cookies for sign-in, locale, and security. With your permission, we also use optional cookies for preferences, for analytics, and for advertising measurement with the Meta (Facebook) Pixel. Read our cookie policy.
VAR with regime-dependent dynamics, where impulse responses change after a threshold variable crosses a specified boundary.
Graph workspace
The core value of a Threshold VAR is showing that the same shock produces qualitatively different responses depending on the regime. A side-by-side IRF panel - one for each regime - makes this asymmetry immediately visible. The threshold variable's time path with the estimated cutoff overlaid provides regime classification context.
Peak response (low regime)
Maximum absolute response in the below-threshold regime to a unit (1 percentage point) shock. The low-regime response decays from impact, so its peak is the impact itself, 1 percentage point, whatever the persistence: how fast shocks dampen shows in the half-life gap, not here.
1.00 pp
percentage points
Peak response (high regime)
Maximum absolute response in the above-threshold regime. It also peaks on impact, at the volatility ratio set on the page, so comparing it with the low-regime peak measures the asymmetry in shock size; the half-life gap measures the asymmetry in persistence.
1.50 pp
percentage points
Half-life gap
Difference in half-lives across regimes. Positive values mean shocks are more persistent in the high regime.
3.27
periods
Regime 1 fraction
Share of observations classified into the low regime. Values below 0.15 or above 0.85 signal a poorly balanced split.
0.4025
proportion
AR(1) coefficient, low regime: 0.5
How long a shock keeps working while the threshold variable stays below its cutoff.
Controls how quickly the left-panel IRF decays to zero. Higher values produce slower decay.
AR(1) coefficient, high regime: 0.85
How long a shock keeps working while the threshold variable stays above its cutoff.
Controls how quickly the right-panel IRF decays. Values near 1.0 produce near-permanent responses.
Threshold percentile: 40
Where the dividing line sits in the ranked data, which sets how often each regime occurs.
Shifting the threshold changes the fraction of observations in each regime. Extreme values shrink one panel's effective sample.
Volatility ratio (high/low regime): 1.5
How much bigger the random shocks are above the cutoff than below it.
Widens the confidence band in the right panel relative to the left panel.
How many periods after the shock the response is traced. It changes the chart only; the readouts do not depend on it.
Extends or shortens the x-axis in both panels.