Saltar al contenido principal

Cookies en Macro by Mark.Usamos cookies esenciales para inicio de sesión, idioma y seguridad. Con tu permiso, también usamos cookies opcionales de preferencias, para analítica y para medición publicitaria con el píxel de Meta (Facebook). Lee nuestra política de cookies.

Macro by MarkMacro by Mark
  • Inicio
Cargando…
Cargando…
← Modelos
MacroConceptosPolíticaModelos

Structural VAR (SVAR) Model

Identified vector autoregression that turns reduced-form residuals into economically named structural shocks.

ResumenGráficoMatemáticasReferencias

Understand the mathematics

Follow the same equations and assumptions one section at a time or in a continuous reading.

Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.

The Identification Problem

Start from the reduced-form VAR(p)\text{VAR}(p)VAR(p):

The reduced-form residuals utu_tut​ are composites of the underlying structural shocks εt\varepsilon_tεt​. Define the structural model as B0yt=Γ0+Γ1yt−1+⋯+Γpyt−p+εtB_0 y_t = \Gamma_0 + \Gamma_1 y_{t-1} + \cdots + \Gamma_p y_{t-p} + \varepsilon_tB0​yt​=Γ0​+Γ1​yt−1​+⋯+Γp​yt−p​+εt​, where B0B_0B0​ is invertible and E[εtεt′]=DE[\varepsilon_t \varepsilon_t'] = DE[εt​εt​′]=D (diagonal, typically InI_nIn​). Pre-multiplying by B0−1B_0^{-1}B0−1​ recovers the reduced form with ut=B0−1εtu_t = B_0^{-1} \varepsilon_tut​=B0−1​εt​.

The covariance restriction is:

This system of equations has n(n+1)/2n(n+1)/2n(n+1)/2 unique entries on the left (since Σu\Sigma_uΣu​ is symmetric) and n2n^2n2 unknowns in B0−1B_0^{-1}B0−1​ on the right. The gap is n2−n(n+1)/2=n(n−1)/2n^2 - n(n+1)/2 = n(n-1)/2n2−n(n+1)/2=n(n−1)/2 restrictions. For a bivariate system, that's 1 restriction. For a 4-variable system, 6. These must come from outside the data.

Without additional restrictions, the structural parameters are not identified. Any orthogonal matrix QQQ produces an observationally equivalent decomposition: if B0−1B_0^{-1}B0−1​ satisfies the covariance equation, so does B~0−1=B0−1Q\tilde{B}_0^{-1} = B_0^{-1} QB~0−1​=B0−1​Q for any Q′Q=InQ'Q = I_nQ′Q=In​. The set of all such rotations forms the orthogonal group O(n)O(n)O(n), which has dimension n(n−1)/2n(n-1)/2n(n−1)/2 - exactly the identification deficit.

yt=c+A1yt−1+⋯+Apyt−p+ut,E[utut′]=Σuy_t = c + A_1 y_{t-1} + \cdots + A_p y_{t-p} + u_t, \quad E[u_t u_t'] = \Sigma_uyt​=c+A1​yt−1​+⋯+Ap​yt−p​+ut​,E[ut​ut′​]=Σu​
Σu\Sigma_uΣu​

The reduced-form VAR: observable variables, estimated coefficients, composite residuals.

ut=B0−1εt,E[εtεt′]=Inu_t = B_0^{-1} \varepsilon_t, \quad E[\varepsilon_t \varepsilon_t'] = I_nut​=B0−1​εt​,E[εt​εt​′]=In​
B0−1B_0^{-1}B0−1​εt\varepsilon_tεt​

The structural mapping. Each column of B0−1B_0^{-1}B0−1​ is the instantaneous effect of one structural shock on all variables.

Σu=B0−1(B0−1)′\Sigma_u = B_0^{-1} (B_0^{-1})'Σu​=B0−1​(B0−1​)′
Σu\Sigma_uΣu​B0−1B_0^{-1}B0−1​

n(n+1)/2n(n+1)/2n(n+1)/2 equations, n2n^2n2 unknowns. The gap of n(n−1)/2n(n-1)/2n(n−1)/2 is the identification problem.

B~0−1=B0−1Q,Q′Q=In  ⟹  B~0−1(B~0−1)′=B0−1QQ′(B0−1)′=Σu\tilde{B}_0^{-1} = B_0^{-1} Q, \quad Q'Q = I_n \implies \tilde{B}_0^{-1} (\tilde{B}_0^{-1})' = B_0^{-1} Q Q' (B_0^{-1})' = \Sigma_uB~0−1​=B0−1​Q,Q′Q=In​⟹B~0−1​(B~0−1​)′=B0−1​QQ′(B0−1​)′=Σu​
B0−1B_0^{-1}B0−1​QQQΣu\Sigma_uΣu​

Any orthogonal rotation of a valid decomposition is also valid. This is why identification requires restrictions beyond the covariance matrix.

Derive the reduced form from the structural model

Pre-multiply the structural model by B0−1B_0^{-1}B0−1​ to isolate yty_tyt​. The reduced-form coefficients are Aj=B0−1ΓjA_j = B_0^{-1} \Gamma_jAj​=B0−1​Γj​.
B0yt=Γ0+∑j=1pΓjyt−j+εt  ⟹  yt=B0−1Γ0+∑j=1pB0−1Γjyt−j+B0−1εtB_0 y_t = \Gamma_0 + \sum_{j=1}^p \Gamma_j y_{t-j} + \varepsilon_t \implies y_t = B_0^{-1} \Gamma_0 + \sum_{j=1}^p B_0^{-1} \Gamma_j y_{t-j} + B_0^{-1} \varepsilon_tB0​yt​=Γ0​+j=1∑p​Γj​yt−j​+εt​⟹yt​=B0−1​Γ0​+j=1∑p​B0−1​Γj​yt−j​+B0−1​εt​
εt\varepsilon_tεt​B0−1B_0^{-1}B0−1​

Compute the covariance of reduced-form residuals

Orthogonality and unit variance of structural shocks simplify the expectation.
Σu=E[utut′]=E[B0−1εtεt′(B0−1)′]=B0−1In(B0−1)′=B0−1(B0−1)′\Sigma_u = E[u_t u_t'] = E[B_0^{-1} \varepsilon_t \varepsilon_t' (B_0^{-1})'] = B_0^{-1} I_n (B_0^{-1})' = B_0^{-1} (B_0^{-1})'Σu​=E[ut​ut′​]=E[B0−1​εt​εt​′(B0−1​)′]=B0−1​In​(B0−1​)′=B0−1​(B0−1​)′
Σu\Sigma_uΣu​B0−1B_0^{-1}B0−1​εt\varepsilon_tεt​

Count equations vs. unknowns

Σu\Sigma_uΣu​ is symmetric, so only n(n+1)/2n(n+1)/2n(n+1)/2 of its entries are distinct. B0−1B_0^{-1}B0−1​ has n2n^2n2 free entries.
Equations: n(n+1)2,Unknowns: n2,Deficit: n(n−1)2\text{Equations: } \frac{n(n+1)}{2}, \quad \text{Unknowns: } n^2, \quad \text{Deficit: } \frac{n(n-1)}{2}Equations: 2n(n+1)​,Unknowns: n2,Deficit: 2n(n−1)​

Show observational equivalence under rotation

Orthogonality of Q guarantees the rotated decomposition produces the same reduced-form covariance.
Σ~u=(B0−1Q)(B0−1Q)′=B0−1QQ′⏟=In(B0−1)′=Σu\tilde{\Sigma}_u = (B_0^{-1} Q)(B_0^{-1} Q)' = B_0^{-1} \underbrace{Q Q'}_{= I_n} (B_0^{-1})' = \Sigma_uΣ~u​=(B0−1​Q)(B0−1​Q)′=B0−1​=In​QQ′​​(B0−1​)′=Σu​
B0−1B_0^{-1}B0−1​QQQΣu\Sigma_uΣu​

Dimension of the orthogonal group matches the deficit

The manifold of orthogonal matrices has exactly the same dimension as the identification deficit. This is not a coincidence - it is the geometric content of the identification problem.
dim⁡O(n)=n(n−1)2\dim O(n) = \frac{n(n-1)}{2}dimO(n)=2n(n−1)​

State the identification requirement

These restrictions must come from economic theory, institutional timing, long-run neutrality, external instruments, or inequality constraints.
Point identification requires n(n−1)2 additional restrictions on B0−1\text{Point identification requires } \frac{n(n-1)}{2} \text{ additional restrictions on } B_0^{-1}Point identification requires 2n(n−1)​ additional restrictions on B0−1​
B0−1B_0^{-1}B0−1​

Why Cholesky became the default

The lower-triangular Cholesky factor of a positive-definite matrix is unique. Setting B0−1=chol⁡(Σu)B_0^{-1} = \operatorname{chol}(\Sigma_u)B0−1​=chol(Σu​) imposes n(n−1)/2n(n-1)/2n(n−1)/2 zero restrictions (a zero upper triangle), which is exactly the deficit. This is why Sims (1980) used recursive ordering - it's the simplest way to fill the gap. But 'simplest' is not 'most credible.'

Set identification vs. point identification

Sign restrictions typically do not pin down a unique B0−1B_0^{-1}B0−1​. They define an identified set - all rotations QQQ satisfying the sign constraints. The IRF is then an interval, not a point. Researchers report the pointwise median and the 16th/84th percentile bounds of the identified set, but these are not confidence intervals in the frequentist sense. They reflect identification uncertainty, not sampling uncertainty.

Transition checks

Impact jump
x0=Bε0,xh=AhBε0(h=1,2,…)x_0=B\varepsilon_0,\qquad x_h=A^hB\varepsilon_0\quad(h=1,2,\ldots)x0​=Bε0​,xh​=AhBε0​(h=1,2,…)

In a linear state-space representation, B maps the innovation into the contemporaneous state. A delayed variable has a zero row in B and moves only after A propagates the shock. A jump is therefore a timing restriction, not a charting artifact.

A nonzero point at h=0 means the shock changes the plotted variable on impact. It is not the end of a smooth pre-shock segment.

Do not smooth through h=0 when the model says the shock enters contemporaneously.

Decay or hump
rh=aρhorrh=a he−λhρh,∣ρ∣<1r_h=a\rho^h\quad\text{or}\quad r_h=a\,h e^{-\lambda h}\rho^h,\qquad |\rho|<1rh​=aρhorrh​=ahe−λhρh,∣ρ∣<1
ρ\rhoρ

Stable eigenvalues shrink repeated applications of A. If a variable is reached through another state first, the response rises before the stable decay dominates, producing a hump-shaped path.

Post-impact movement follows the model recursion. A hump means delayed propagation, not interpolation by the chart.

Do not treat a hump as evidence of a linear ramp unless the model includes that ramp.

Variable timing

Innovation

ε0≠0,εh=0 for h>0\varepsilon_0\ne 0,\qquad \varepsilon_h=0\ \text{for}\ h>0ε0​=0,εh​=0 for h>0

The shock is dated at horizon zero. A response at h=0 is an impact effect, not the end of a hidden ramp.

State vector

xh=AhBε0x_h=A^hB\varepsilon_0xh​=AhBε0​

A state can move on impact only through the relevant row of B; otherwise it moves after A propagates the shock.

Reported response

rh=CAhBε0+Dεhr_h=CA^hB\varepsilon_0+D\varepsilon_hrh​=CAhBε0​+Dεh​

The plotted variable can jump directly through D or indirectly through the state vector.

Keep going

Read next

Continue with a related concept, model, or example.

DSGE model

Financial Accelerator (BGG)

a contrasting take on Structural VAR

Contrast it →
Theory model

IS-LM Model

a well-connected theoretical model two steps from here

New angle: theoretical model →
Macro by Mark
Ayuda
Política de privacidadTérminos de usoÉtica y cumplimiento

© 2026 Mark Jayson Nation LLC