Sample Size: $T$
T represents the sample size or the number of observations in the dataset.
Rotation angle: $\phi_0$
In the bivairate SVAR $u_t = B_0(\phi_0) \epsilon_t$, the reduced form shocks $u_t$ are a rotation with an angle of $\phi_0 (180 / π)$ degree of the structural form shocks $\epsilon_t$.
Rotation angle: $\phi$
In the bivairate SVAR $u_t = B_0(\phi_0) \epsilon_t$, the reduced form shocks $u_t$ are a rotation with an angle of $\phi_0 (180 / π)$ degree of the structural form shocks $\epsilon_t$. The rotation $\phi $ allows to calculate the innovations $e_t(B(\phi)) = B(\phi)^{-1} u_t$ as a rotation of the reduced form shocks $u_t$ with an angle of $\phi (180 / π)$ degree.
Introduction
Proxy variables are among the most popular tools to identify Structural Vector Autoregression (SVAR) models. This short interactive guide illustrates how endogenous proxy variables lead to biased SVAR estimates and it illustrates the test for proxy exogeneity in Bruns and Keweloh (2024).
The structure of the guide
The guide is structured as follows:
- The first page introduces the SVAR model and the identification problem.
- The second page illustrates how a proxy allows to identify the SVAR.
- The third page illustrates how the proxy exogeneity assumption can be tested if two proxies are available.
- The fourth page introduces the test for proxy exogeneity in Bruns and Keweloh (2024).
The SVAR model
In a SVAR a vector of time series $$ y_t = [y_{1t},...,y_{nt}]'$$ is explained by its own past and a vector of structural shocks $$\epsilon_t =[\epsilon_{1t},...,\epsilon_{nt}]'$$ using
The following assumptions allow an easy visualization of the identification problem.
Simplifying assumptions
The VAR(p)
Assumption 1: No Lags
The SVAR has zero lags:
To further simplify, let's consider an SVAR with only two variables and shocks such that each of the two reduced form shocks is equal to an unknown linear combination of the two structural shocks.
Assumption 2: Bivariate SVAR
The SVAR contains only two variables and shocks:
Finally, assume that $B_0$ is orthogonal. Without the orthogonality assumption the matrix $B_0$ contains four unknown parameters. However, an orthogonal matrix $B_0$ essentially only depends on a single unknown parameter, which simplifies the visualizations.
Assumption 3: Orthogonal $B_0$
The matrix $B_0$ is orthogonal and can be written as:
The relationship between reduced and structural form shocks
In the simplified SVAR model, the reduced form shocks $u_t$ can be thought of a rotation with an angle of $\phi_0 (180 / π)$ degree of the structural form shocks $\epsilon_t$:
- Move the $\phi_0$ slider to adjust the angle of the rotation and observe how the reduced form shocks $u_t$ change in the interactive simulation at the bottom of the page.
Structural shocks and innovations
In a given application, we neither know the rotation matrix $B(\phi_0)$ nor do we observe the structural shocks $\epsilon_t$. Instead, we only observe the reduced form shocks $u_t$ (in practice we only observe the macroeconomic variables $y_t$ and need to estimate the VAR to get the reduced form shocks $u_t$. However, with the simplifying no-lags assumption it holds that $u_t=y_t$). Consequently, the question is how can we identify $B(\phi_0)$ and recover the structural shocks $\epsilon_t$ from the reduced form shocks $u_t$?We can always choose an angle $\phi$, calculate $B(\phi)$:
and get the corresponding innovations $ e_t(B(\phi)) = B(\phi)^{-1} u_t $ such that
These innovations represent the shocks that would yield $u_t$ if the true angle of the data-generating process $\phi_0$ were equal to the chosen angle $\phi$. Specifically, for $\phi=\phi_0$ we get $e_t(B(\phi_0)) = \epsilon_t$.
- Move the $\phi$ slider to adjust the angle of the rotation and observe how the innovations $e_t(B(\phi))$ change in the interactive simulation at the bottom of the page.
Interactive simulation
Note that all rotations $\phi$ lead to uncorrelated innovations. Therefore, the assumption of unccorelated shocks can not be used to identify the SVAR. Meaning the assumption of uncorrelated shocks is not sufficient to use the reduced form shocksto recover the rotation $\phi_0$ and the structural shocks $\epsilon_t$.
Next steps
The next page illustrates how a proxy allows to identify the SVAR.