Thanks for the link, unfortunately it is firewalled for me.
It would be interesting to know the methodology he uses. Of course the grand caveat in all such models, "correlation does not imply causation". It's fairly easy to "retcon" a model to fit past circumstances and say "this will predict the future", but that is not always the case.
That, and circumstances now are much different than 4 months ago, and indeed much different than in any other past presidential election. There are simultaneous unprecedented (or at least exceedingly rare) crises shaping public opinion. The candidates' positions and responses to those crises will drive the voting results far more than any of the "traditional" motivations in 2020. So I wonder if his modeling takes these highly unusual circumstances into account.