Black people have more heart attacks, white people are more likely to develop Seasonal Affective Disorder in the same geographic region, etc. It’s nothing to do with predicting behaviour, it’s just differing physiologies, causing different levels of risk of certain physiological conditions irrespective of the choices we make or how we’re raised.
As well, different races respond better or worse to different drugs. There might be a cheap way to treat hispanic people for a condition but only an expensive way to treat any other race. In that situation, if there was actuarial data saying that Hispanic person A and non-Hispanic person B had an equal chance of developing that condition, it would make sense for person B’s premiums to be higher, no?
If you can predict, per person, not only the chance of an event causing a condition requiring an insurance-covered response; but also predict, per person, the size of the total insurance pay-out from a given event, then people who will need more expensive treatments for the same condition will need higher premiums. It’s like flood insurance: insuring a house closer to the coast costs more, but insuring a larger house also costs more. Because the pay-out will have to be higher to fix a larger house, even though the flood was the same. Same damage—more expensive solution required.