Assume we have the time series $X_{1,t}$ and $X_{2,t}$ where $X_{i,t}=(x_{i,1},...,x_{i,T})$. We build two models as (Assume the time series are weakly stationary): \begin{align} X_{1,t}=\sum_{\tau=1}^{p_1} a_{\tau} X_{1,t-\tau}+\epsilon_{1,t} \end{align}

\begin{align} X_{1,t}=\sum_{\tau=1}^{p_1} a_{\tau} X_{1,t-\tau}+\sum_{\tau=1}^{p_2} b_{\tau} X_{2,t-\tau}+\epsilon_t \end{align}

The first one is an $AR(p_1)$ model and the second one is also an $AR(p_1)$ model but with the exogenous variable $X_j$ (These are the two models that is built to check the Granger-causality). We obtain variances for first one, let's call it $\sigma^2_{unexp}$, meaning the unexplained variance and the second one $\sigma^2_{exp}$, the explained var.

  1. Now, we want to see if $\sigma^2_{unexp}$ is smaller than $\sigma^2_{exp}$, meaning we have an improved model. The question is: Is there a difference in performing F-test or Wald test in this case? Is it the same as one way ANOVA? How is the sample variances calculated for the F-statistics?
  2. Secondly, we can asses if there's improvement, also by looking at the coefficients $b_{\tau}$. If the coefficients $b_{\tau}$ are jointly significantly different from zero, then we have an improvement. What is the test for this (I think it's still F-test) and what are the steps?

I apologize if it's a lot to ask but I just want to get a clear image of the whole process, which seems to be found nowhere!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.