Advent of Alpha Day 24: Order Execution Latency

About six months ago I was watching some youtube videos about HFT trading, looking for ideas, and some guy popped up and talked about a patent for a technique that NASDAQ and other exchanges had to explicitly ban,

The patent basically boiled down to this: you might want to long or short something very quickly, based on a new piece of information. Let’s suppose you knew interest rates were about to get announced at 9am local, and knew if they went up, went down, or held, how you would need to be on a market position was known to you. You want an algorithm to get the news and then execute: buy, sell, or do nothing.

But let’s suppose you wanted to optimise this situation to within milliseconds. How do you beat everyone else on the street?

The technique they patented was holding back the last byte: at 8:59:59 they’d send all the orders in but not send the last byte of the TCP packet. On the news being announced, they’d only have to send one byte through to the exchange for the one order they wanted to complete, and send a FIN on a corrupt packet (or drop the connection), to all the other orders.

This is a lot of work, but if you feel your work is latency sensitive - latency arbitrage, or situations where you have best price execution turned off, you may need this to move you from flat to +EV.