One of the hardest things about insurance is figuring out the probabilities of very unlikely events. Hurricane Andrew was a moderately large storm that directly hit three major population centers - Miami, Ft. Myers and New Orleans. That circumstance is infrequent enough that modeling it statistically leaves a wide range of uncertainty.
I would think about insurance as more of a smoothing mechanism of inherently uncertain outcomes. When something happens that's unexpected and causes a larger loss, that is typically recouped by the industry over a few years of higher rates. That way the industry is still absorbing volatility, which is valuable to their customers, but doesn't require that they be 100% correct about the probabilities of infrequent events.