AFAICT car accident data is hard to come by at that degree of granularity. You could use median insurance quotes, but that averages over several different types of risk and is difficult to obtain at scale besides. One fair bet might be to contact the authors of this paper, who hired a firm to collect this data for a bunch of zip codes. (They probably wouldn’t release the data for exactly that reason, but if you had it, you could then hook the zip codes up with census data.
Of the ten zip codes in the US with the highest premiums, six are in cities with unusually concentrated black populations (Detroit, Brooklyn, Philly, New Orleans, Baltimore and Miami), two are in cities without large black populations (Providence and LA), one is probably a giant meth lab (Royalton, KY), and the last one has a large population of criminogenic Yale faculty or something (Woodbridge, CT).
But extreme value statistics tend to make for bad inferences. I’d much rather have zip code data for the whole country. So I guess my answer is ‘no’.