Non-Gaussian SVAR: Maximizing Non-Gaussianity

Maximizing non-Gaussianity

This page demonstrates that minimizing the innovations' dependencies (as shown on the previous page) is equivalent to maximizing the shocks' non-Gaussianity. i Consequently, we can identify $\phi_0$ by finding the most non-Gaussian innovations.

Measuring Non-Gaussianity

The non-Gaussianity of the innovations $ e_t(B(\phi)) $ can be measured using the skewness and excess kurtosis:

$$ \begin{matrix} \text{Skewness} & \; E[e_{1,t}(B)^3] \\ \text{Skewness} & \; E[e_{2,t}(B)^3] \\ \text{Excess kurtosis} & \; E[e_{1,t}(B)^4-3] \\ \text{Excess kurtosis} & \; E[e_{2,t}(B)^4-3] \end{matrix} $$

Interactive Simulation

Move the $\phi$ slider to see how the rotation angle affects the non-Gaussianity of the innovations displayed in the table below the scatter plot.

Interactive Loss Calculation

The loss value in the table above simply sums up all squared higher-order moments to get an overall measure of the non-Gaussianity.

$$ loss(\phi) = \text{mean}( e_{1,t}(\phi)^3 )^2 + \text{mean}( e_{2,t}(\phi)^3 )^2 + \text{mean}( e_{1,t}(\phi)^4-3 )^2 + \text{mean}( e_{2,t}(\phi)^4-3 )^2 $$

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

Takeaway: Choosing the angle $\phi$ that maximizes the innovations' non-Gaussianity allows us to estimate $\phi_0$. i

Next Steps

The next page illustrates how the degree of non-Gaussianity of the structural shocks affects the estimation.