Some models incorporate real world data. Some models draw analytical conclusions from [slight modifications] to relationships widely agreed to be true.

And some are pretty purely abstract. With this one you can largely guess the conclusion from the stated assumptions despite the complexity of the algebra but... there isn't much reason to believe the stated assumptions approximate reality. Yes, if you model political polarization as economic interactions in which people gain more from riskier trades with people from a different race or political party, pick their political party according to these economic aspirations, and are risk averse in proportion to their income, then the logic of those assumptions is that inequality will make people more likely to sort themselves into parties which match their races and consider those important to their decisions. But I'm not sure that's really how US racial politics works, or that the complexity of the model's dynamics make it more rather than less plausible.

There are better reasons to believe reducing inequality can reduce political polarisation than this model, as well as better reasons to believe it won't make much difference