In midtown Manhattan one recent morning, Ryan Sheftel and a handful of math whizzes at New York-based Global Trading Systems were debating how to profit from rapid-fire changes in the world’s most-liquid market, US treasuries.

Their quandary: how to gain an edge on dozens of firms seeking out trading signals by buying and selling milliseconds ahead of rivals. Hours later, after one suggested tracking treasury moves by simulating atoms colliding inside metal, a bid to predict how other traders would respond to market moves, they began tweaking their proprietary computer codes to react faster.