I don't know what you mean by "inferred empirically." If you mean "statistical inference," there are tons of unstated assumptions that basically assume the observed object is benign or indifferent. There is work in machine learning on learning in adversarial settings, but it's a much harder problem. Markets are super adversarial, and in addition there are incentives against publishing sensible analyses (why give away money to hostile/competing interests?)

edit: Sorry, I should say "tons of assumptions." People state them, and it's clear they are benign, e.g. samples are i.i.d.