Colocating

So true. Colocation without a FPGA board and an expert knowledge of micro market and order book dynamics and the ability to express such in concise code nowadays makes zero sense. Did i forget a very deep pocket? Everything else is snake oil For the masses. I believe a total of 1 or 2 individuals on this whole site might possess such skillset

Retail collocation is a sales ploy. It's like the difference between walking from New York to San Francisco vs riding a bike. Sure, a bike is faster but it makes no practical difference because almost everyone else is flying or driving.

The PNL difference between 300 ms latency and 1 ms latency won't be measurable.

There are semi-retail ways of getting down to 60-80 mics that would have a measurable impact. That's like crossing America in a Ferrari. At least you are beating all the other car drivers.

Proximity hosting with a reliable exchange connection and power is worth doing.
 
The PNL difference between 300 ms latency and 1 ms latency won't be measurable.

Would partially agree with this.

Under normal trading conditions I agree the PnL difference is negligible. However, in liquidity squeezes the difference in marketable execution price between 1ms and 300ms can be several ticks. I currently have my Globex machines in 350 E Cermack and have been crunching the numbers to see if there is a valid business case to colo using these guys instead.

http://www.theomne.net/dedicated-servers/

It's not clear cut at the moment, but analysis of my simulation data suggests it could be worthwhile (if coupled with some tweaks to my strategy which would shorten its time horizon during such squeezes).
 
Would partially agree with this. ...
It's not clear cut at the moment, but analysis of my simulation data suggests it could be worthwhile (if coupled with some tweaks to my strategy which would shorten its time horizon during such squeezes).
What round trip times are you using to estimate your PNL for 350 E Cermack vs theomne.net?

As I see news events can trigger liquidity crunch and I suspect that HFT guys turn off their algos... if they do so then it should be for a good reason. You might still get away if your algorithm can reasonably predict how far the price can get squeezed either direction in a long time frame before hand and prepare to trade accordingly. Eitherway, this needs quite a bit of backtesting.
 
Last edited:
Unless the algo is specifically targetting the trading of news events, it would be prudent to turn it off. News events typically cause markets to go into a yo-yo set of moves that is targetted at true price discovery.
 
I trade futures and my near the exchange colocation server often fills a tick faster than my personal computer. It always executes exits in fast markets much faster sometimes by as much as 3-4 ticks. The trading system is not HFT but orders are inline and the earlier ones get priority on the order matching engine. I used to run game servers as a kid and know that there is considerable difference when talking 30ms+ differences in response times.
 
Back
Top