Poor Man's Financial Supercomputing Analysis is here.

I am currently working on a personal project that takes advantage of the supercomputing capabilities of GPU processors that are available in high-end 3d video cards. Four of these video cards on a single motherboard with 4 PCI-Express slots effectively gives the user the power of about 300 intel core duo processors or a $4 million dollar supercomputer for about $4000.00. Using the OpenCL API I am developing a framework for financial analysis, decision support, and automated trading. I was able to build my current setup for about $3500.00 excluding the cost of software which includes MatLab and 3rd party GPU engine software that enables MatLab to use the GPUs.

I am currently using a simple EA to write out the tick data from the MT4 terminal into a CSV file which is then picked up by MatLab. I had intially started working on an EA that would call winsock2.dll to send a UDP packet to a simple daemon on the server but I decided to keep it simple for the time being. In the meantime I am thinking of new ideas on how I could leverage the computing power that I now have humming away in my spare bedroom.

I would like to know what you would do if you had a supercomputer?

AS it relates to trading? i cant think of anything right now.

But u could use it as a rendering farm-let or something. Convert videos for people online or soemthing...i dunno,

Well that's really not my intention nor my interest but obviously that would be something others could look into.

My interest is in its capabilities for statistical analysis as it pertains to forex and other forms of trading. Financial institutions have used multi-million dollar supercomputers for statistical analysis as an unfair advantage and with a $3500.00 I am hoping that the common folk could at least have a game changing arsenal on their side.

It would be lovely if that power could be used by the MT4 Backtester - CPU power is only useful for simulation (unless you strategy is soooo complicated that it takes forever to create a buy/sell and the speed would help you create better signals)

One obvious use, as far as trading is concerned, would be for trading system parameter optimisation. This area is typically very resource/time intensive, and would benefit most from extra computing power.

To use it, though, you would need a fairly advanced trading simulator with multi-parameter optimisation modules.

I am not too familiar with Matlab, so I don't know if there is something already available to do this.

Regardless, I think more and more people are beginning to explore this option (outside of the big algorithmic thinktanks).

Cheers

If I had a supercomputer, I would naturally run my NN on it. (After benchmarking to see the speed =D)

This is actually where my current line of thinking is going. If a good simulation is created it could then present these dynamic parameters to the EA/trading system. First it would simultaneously run multiple simulations (made possible by the computing power we have available) using historical data up to the current tick and compare those values to find the best ones that we could feed back to the EA.

I chose MatLab since it's something I am intimately familiar with back in my college studies and there are already 3rd party tools that are available to take advantage of the GPU. I have looked into OpenCL as another option but time constraints prevent me from further exploring that route at this time.

My original idea was spawned when I read an article about a scientist using a cluster of 8 Playstation 3 game systems to do some calculations because he was looking for a cheaper way to run his simulations. The key limitation with the Playstation 3 is that each unit is limited to 256megs of memory and that puts a big restriction with the amount of data that could be used at any given time. Then some guys were able to crack WEP security used in wireless networking devices using a single NVIDIA graphics card which got me interested in the idea again. Some researchers at the University of Antwerp were able to build their own as stated in this article. With the amount of memory you can put on a single motherboard running a 64-bit OS we have enough memory to store the data for the type of simulations I am talking about.