I am wondering what is going on "under the hood" or intuitively of what the implication of modelling outliers as dummy's in a VAR-model.
To make this question more clear I will provide an example.
Let Oil-price, Unemployment rate, GDP , Inflation rate and repo rate be quaterly timeseries from 2000-2019.
Using these variables we estimate a VAR(p)-model. In the process we test for autocorrelation and normally distributed residuals, and find that the later is not satisfied. We explore the residuals using QQplots and see that we have a outlier in the left tail.
We use this info to create a dummy variable where this outlier is stored as OIL.minimum. We see that this outlier correspond to observation 2008 Q3.
We now estimate a new VAR(p) model and see that we now have no autocorrelation and normally distributed residuals. So we are satisfied with our model and proceed to estimate Impulse/response functions(IRF's).
We observe that the IRF have changed some and show other effects than previously.
So now my question is, how has our dummy been used in this new model? If someone could explain the process, I would be very greatful,
Many thanks.
