Short story, is not easy at all.
Recently I’ve been approached for many startups looking for setup their own HFT shop because they’ve heard how profitable this business is. And they are right! I know for a fact that serious firms are making fortunes (look what Virtu Financial or Citadel are doing).
One firm called me asking me for consultancy and join their team, in order to add more manpower in their operation and to improve their models/strategies, since some of them were under-performing, and month after month they were experiencing a decay in their overall profit. So, we agreed on the terms, and I was on board.
At first, everything looked good, but several weeks we found out that they were under-performing because of bad execution response. After deeper analysis, I realized that their software platform was producing a lot of latency.
My real work just started… I measure all incoming/outcoming messages, and all the trades’ logs, benchmarking processes and communication with all the exchanges. The result was that trades were being placed late. The system was reacting slowly. And not just a couple milliseconds!
So, I decided to create (as many times) their platform from scratch, taking care of each process and benchmarking everything we do. I put latency down, and the overall system was working as expected.
There were one more thing to do: move the FIX Engine to FPGA. But that is another story for another chapter.
I’ve seen this situation over and over again with different clients
Is not enough to have a great model. Network and system latencies will throw away all your profits if you are not carefully taking care of it.
Network latency is the easiest: make sure that you have the proper hardware and you are collocated.
System latency will depend on in your engineers (or your platform vendor), and if you don’t have a control over it, will give you headaches with your overall performance.