How to handle collinearity with LightweightMMM

[openhubber]

I have the click data from different marketing channels, and some of them have high correlations between the pairs of channels.My understanding is that LightWeightMMM is a linear model so the collinearity will impact the explanation or the attribution analysis. Please confirm if my understanding is correct.

I read through some online suggestions to remove collinearity:

1. Remove correlated variable
2. Merge correlated variables
3. Principal analysis
4. Regularization with Ridge (this seems to be what Robyn uses)

For 1~3, I don't think it fully solve the problem I have, as I have two channels with most of the spend, and they are highly correlated. And one goal is to re-allocate the spend between these two channels. So either removing or merging may not work for this case.

Thinking about how we can apply regularization with Ridge to lightweightMMM model. Is it feasible to do?

Also, any other thoughts on solving this problem? Thanks!