Granger causality is about linear inter-temporal dependence. Chatterjee correlation is about non-linear contemporaneous (or time-independent) dependence. They're tools for quite different applications.
Looks like you could do the opposite: use this measure as a proxy for correlation between input data X and Y, and seeing if increased correlation between X and Y is superior to just prediction of Y without X. Maybe model the correlation value going over a threshold as a sequence of Bernoulli trials
Granger causality is about linear inter-temporal dependence. Chatterjee correlation is about non-linear contemporaneous (or time-independent) dependence. They're tools for quite different applications.
Looks like you could do the opposite: use this measure as a proxy for correlation between input data X and Y, and seeing if increased correlation between X and Y is superior to just prediction of Y without X. Maybe model the correlation value going over a threshold as a sequence of Bernoulli trials