Non-Gaussian SVAR: Non-Gaussianity

Impact of Non-Gaussianity on Estimation

This page illustrates how the degree of non-Gaussianity affects the estimation of structural shocks in our SVAR model.

Shock Distributions

In the simulation below, the first structural shock $\epsilon_{1,t}$ is independently drawn from a standard normal distribution. The second structural shock $\epsilon_{2,t}$ is independently drawn from a mixture of normal distributions i

$$ \epsilon_{2,t} \sim 0.9 \mathcal{N}(0, 1) + 0.1 \mathcal{N}(s, 1). $$

The location parameter $s$ of the second component can be chosen using the slider above:

Interactive Simulation

Move the $\phi$ and $s$ sliders to see how the rotations and non-Gaussianity affect the dependencies of the innovations displayed in the table below the scatter plot.

Interactive Loss Calculation

The loss value in the table again sums up all squared co-moments to get an overall measure of the dependencies.

$$ loss(\phi) = \text{mean}(e_{1,t}(\phi)e_{2,t}(\phi))^2 + \cdots + \text{mean}(e_{1,t}(\phi)^2 e_{2,t}(\phi)^2 - 1)^2 $$

The plot below shows the loss for all rotation angles $\phi$.

Takeaway: Gaussian shocks lead to a flat optimization landscape without a unique minimum at $\phi_0$. This illustrates why non-Gaussianity is crucial for identifying structural shocks in SVAR models based on independent shocks.

More Resources