I am curious if anyone has considered the problem of an exhaustive pairs trade screen? As I understand, Quantopian has N ~ 10^3 symbols in its database. Screening all possible pairs would require 0.5N(N-1) ~ 10^6 screening iterations.
Is this approach something Quantopian could support now or in the future?
Naively, it would seem that the problem is tractable, so long as the screening algorithm is efficient. For example, on the Nvidia website I see that desktop GPUs can have thousands of cores, so perhaps the total screening time could be cut down by at least three orders of magnitude with parallel computing.