> overestimating low-probability events is not irrational — it's optimal

It can't possibly be more optimal than estimating them correctly.