Don’t be fooled by those expensive servers for high performance HFT computing from Oracle Sparc, Cray, etc. This Youtube video show its the programming
I just sent this to someone:
It seems you are wanting to spend money on some high end servers. My experience with CUDA is it can only be used for backtesting, not real time analysis trading. Also, these expensive servers are just that, expensive. The performance techniques all done in your programming:
You will only get this performance in C++ or .NET, Java is quite limited to achieve this.
I use Matlab Simulink to generate code either into C, C++, or FPGA ready if need be. You don’t need to blow your budget on expensive hardware like you listed.
I am learning this from one of the best quants in London UK. I talk with him for many hours per week. He does not hold back on what he knows.
Hope this helps
My question and answer part 2 on Strategies for those without servers at the exchanges
More answers for my Youtube viewer on this:
When you say mid to low frequency, what time period would you define those as? Tradebot Systems was quoted in 2008 stating that their average hold time was 11 seconds which I would consider high frequency. Are you talking minutes, hours, or days?
–> I am talking subsecond like in tick data
What do you mean when you say stay away from platforms as they are limited to one data connection? If someone is trading an automated system using a datafeed like IQFeed or Nanex wouldn’t they be inputting the data into their trading platform via the databroker API, processing it and then sending the order to the broker via the broker API? What is the disadvantage of having only one data connection?
–> I will focus on multiple execution order routing services like LMAX, IB, etc. If you have a multi amount of execution choices within your platform, you can develop a way to get access to the broker’s order book to build more intelligence execution for your orders. Popular platforms like Metatrader or Tradestation cannot do this. Commercial black boxes fail as well. As for data, focus on better providers like Bloomberg or Reuters but they are pricey. I only use IQFeeds for now as they are affordable.
What specific strategies is your site focusing on (statistical arbitrage, pairs trading, technical analysis, fundamental analysis, event trading)? Ernie Chan’s Quantitative Finance book advocated running Matlab before the market opens and screening for setups using statistical arbitrage. He may have been using pairs trading (like GLD vs GDX) or basket trading and was using significant leverage via a proprietary trading firm.
Jane Cralle who wrote An Introduction to Algorithmic Trading recommended streaming tick data into excel on your home computer and trading based off of that. Her optimal algo target is 8 out of 10 winners with 2 stopped out at 40 basis points and her profit target was 25-50 basis points net of commission per trade.
–> See my survey to see what is popular. I will be focusing on those with propietary custom ones extended of the ones I focus. As those prove to be more profitable, I will be raising the membership rates exponentially because of this.
My question and answers on strategies for those without servers at the exchanges
Here is a Youtube email message from someone:
Like your channel, looks like you have put a lot of thought into it. 2 questions: 1. I was thinking about joining but is your site focused on people who are trying to get jobs at HFT firms or for retail traders trying to develop their own algos within the limitations of not having servers co-located at the exchanges?
–> I would say the membership focuses on the second. I have many institutional traders who are looking for alternative ways to develop models and strategies. 2. What do you think are the best quantitative strategies for individual investors who do not have servers at the exchanges?
–> You can still do mid or low frequency with the same strategies. Having servers on the market will be regulated at some point. The SEC already noted they are investigating this. As a retail trader without access to a server on the exchange, the best I can do is to put a server in a New Jersey data center. But even here there’s a 2+ ms delay when routing orders straight to the broker. If a server on the exchange is at <65 microseconds latency, the HFT guys can get 30+ Immediate or Cancel orders off in the time it takes for a retail guy to do one. It seems that this time disadvantage would eliminate many of the most profitable, sure-fire HFT strategies. So what strategies are left?
–> It is a distinct disadvantage and noted by the SEC. I am sure something will be addressed on this in the future. ALso, stay away platforms as they are limited to one data connection and brokers. Your broker could be corrupt if dealing with forex.