Actually, you can fairly comfortably build a second-tier latency setup on the cheap (like 1-2k a month) and third-tier for even less (~500/month), for either futures or equities. There are companies providing all the components - shared colocated servers and/or proximity hosting, retail-priced trading APIs and fairly cheap data feeds. The real question is finding alpha that requires something like that.
ES Algo 10K - 20 K positions / day
- Thread starter Futures_Trader
- Start date
Interesting. I didn't know that...
Speaking of latency. I noticed that my cloud based servers has a ping response of less than a 1ms on the Internet. That's LAN-type speed. However, and perhaps, I don't have a complete view of AWS' infrastructure.
Well, what you care about is latency from your server to the exchange; e.g. what do you think is the ack time from the moment you send the order? Most of that will be flight time from you to the matching engine and back to you.
what is the definition of second tier latency and which companies provide this for 1-2k? please list all the names you can
I don't know exactly. However, I reckon the flight time from a VPS in AWS to the exchange is significantly faster than the flight time from my desktop computer.
Tangent: Flight time is important. However, unexpected outages and a lack of redundancy are equally, if not more, important. For example, what if Verizon, Comcast, etc decides to do maintenance or a construction crew severs a telco line?
Personally, if I were a retail algo trader, I would never run a trading robot from my home computer/network.
Just forget it. If you need speed to be competitive, you have already lost. The high-frequency firms are impossible to compete with unless you have similar resources.
So go for Python. It is fast enough and will be a zillion times easier to learn and quicker to implement. Python does also have a few tricks up its sleeves to speed up. But that doesn’t matter. You cannot compete on speed regardless
Shallower learning curve is a better bet as it will ultimately come down to confidently running the strat. All criticisms aired on this post are legitimate. Even if you are flipping 20s or 500s. Mistakes are costly and will cost you time and money but you know that. Automation and market impact are a completely different game.
A lot of prop firms have remote traders and top tier data feeds. DM if you are curious I only post because that is how I connect.
C++ is faster than Python. C++, Java, and C# are all well suited for fast code. Well written they will easily handle your needs. If you need your algorithms to run even faster the general solution is to modify (or find a modified) the Linux kernel.
If you are only trading the ES Python would probably be fine.
Take into account your data provider, broker, etc. when picking a language.
Keep in mind that the data that any retail trader sees is really old. The time it takes your algorithms to calculate and issue buy/sell orders is insignificant compared to the time between a trade on the exchange, and when you get the data. Add to that the time between sending your order and receiving the broker confirmation.
Will you be co-located? Consider TT co-located with ADL to code your strategy unless your strategy is really complex. Its actually pretty cheap for lightning fast exection.
If rithmic isn't even co-located, you are asking the wrong question because whatever ultra fast code you setup would still be shit slow if you are not co-located at the CME exchange.
If rithmic isn't even co-located, you are asking the wrong question because whatever ultra fast code you setup would still be shit slow if you are not co-located at the CME exchange.