Skip to main content

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.

Macro by MarkMacro by Mark
  • Home
Loading…
Loading…
← Models
MacroConceptsPolicyModels

TVP-VAR Model

Time-varying-parameter VAR that lets transmission coefficients and shock effects drift across samples.

OverviewGraphMathReferences

Graph workspace

Time-varying impulse responses (3D surface or stacked IRF snapshots)

Graph workspace

The defining output of a TVP-VAR is how the impulse response changes across time. A 2D panel showing IRF snapshots at three selected dates (e.g., 1975, 1995, 2015) reveals structural change directly.

Chart axis metadata
xHorizonquarters · 0 to 20yResponse of inflation to monetary policy shockpercentage points · -1.5 to 0.5

Shaded ranges come from assumed parameters in a simulated example. They are not estimated from observed data.

Coefficient path over the sample static chart preview showing Coefficient path over the sample
Response in the early window static chart preview showing Response in the early window
Response in the late window static chart preview showing Response in the late window
xHorizonquarters · 0 to 20yResponse of inflation to monetary policy shockpercentage points · -1.5 to 0.5

Current results

Persistence change

Explain: Persistence change
βˉ1,late−βˉ1,early\bar{\beta}_{1,\text{late}} - \bar{\beta}_{1,\text{early}}βˉ​1,late​−βˉ​1,early​

Mean autoregressive coefficient over the last quarter of the sample minus its mean over the first quarter. Positive means the process became more persistent.

-0.0313

coefficient units

Volatility ratio (late/early)

Explain: Volatility ratio (late/early)
σˉlate/σˉearly\bar{\sigma}_{\text{late}} / \bar{\sigma}_{\text{early}}σˉlate​/σˉearly​

Mean shock standard deviation over the last quarter of the sample divided by its mean over the first quarter. Below 1 indicates a Great Moderation pattern.

3.74

ratio

Peak response (early)

Explain: Peak response (early)
max⁡h∣IRFearly(h)∣\max_h |\text{IRF}_{\text{early}}(h)|maxh​∣IRFearly​(h)∣

Maximum absolute impulse response in the early window (the first quarter of the sample), with the window's mean coefficient and shock size. For a stable AR(1) the peak is the impact, which is the window's mean shock size; persistence shows in how slowly the response decays, not in the peak.

0.94 pp

percentage points

Peak response (late)

Explain: Peak response (late)
max⁡h∣IRFlate(h)∣\max_h |\text{IRF}_{\text{late}}(h)|maxh​∣IRFlate​(h)∣

Maximum absolute impulse response in the late window (the last quarter of the sample). The late-over-early ratio of the two peaks equals the volatility ratio; compare the decay of the two responses to gauge the change in persistence.

3.53 pp

percentage points

Controls

Shocks fade steadily
Explain: Shock memory at the start

Initial AR coefficient: 0.85

How long shocks last at the start of the sample; higher values make early shocks fade slower.

Higher initial persistence produces a more slowly decaying early-sample IRF.

Slow, steady drift
Explain: How fast shock memory changes

Coefficient drift variance: 0.001

How much the model's shock memory can wander from one period to the next.

Larger q produces more divergent IRF snapshots across time periods, showing greater structural change.

Shock size drifts slowly
Explain: How fast shock size changes

Volatility drift variance: 0.005

How much the size of shocks can wander from one period to the next.

Larger www produces more variation in the impact response (height of the IRF at h=0h=0h=0) across time snapshots.

No built-in trend
Explain: Trend in shock size

Volatility trend: 0

Whether shocks get steadily smaller or larger over the sample, beyond their random wandering.

Negative values shrink the impact of the late-window response relative to the early one; positive values enlarge it.

Enough to show drift
Explain: How many periods simulated

Sample length: 200

How many time periods the sample covers; longer samples let the model drift further.

With longer samples, the late-period IRF diverges more from the early-period IRF if q>0q > 0q>0.

20
Explain: How far ahead to trace

How many periods ahead the response to a shock is traced. It changes the chart only; the readouts do not depend on it.

Longer horizons reveal whether the IRF returns to zero at each snapshot date.

Keep going

Read next

Continue with a related concept, model, or example.

Theory model

IS-LM Model

a well-connected theoretical model two steps from here

New angle: theoretical model →
Macro by Mark
Help
Privacy PolicyTerms of UseEthics and Compliance

© 2026 Mark Jayson Nation LLC