nonlinear granger causality test

[Chiraz KARAMTI]

Hi,
Do anyone have an idea how to run nonlinear granger causality test for panel data? I know the pvargranger command, however it's a linear test.
thanks

[Carlo Lazzaro]

Chiraz:
unknown beast for me, but the following paper covers the subject you're interested in (no Stata flavour, though) https://papers.ssrn.com/sol3/papers....act_id=2608388

[Jeff Wooldridge]

Frankly, that literature is a mess. It doesn't distinguish between causality in mean, variance, or some other uninteresting part of the distribution. In my view, you're better off just adding nonlinear functions of lags, as well as interactions of lags, and doing a joint test. So, take your linear model for Y(i,t) as a function of lagged Y(i,t) and lagged X(i,t). Then add things like X(i,t-1)^2 and X(i,t-1)*Y(i,t-1). This must be estimated using an IV approach, similar to Holz-Eakin, Newey, and Rosen and Arellano-Bond. Then do a joint test on the linear as well as nonlinear terms involving X(i,t-1).

You should tell us much more. How big are N and T, for example?

[Chiraz KARAMTI]

Great idea! So you mean like White heteroscedasticty test we add non linear terms (interactions, powered variables...). My panel is a short one with 31 countries and only 15 years.