How to build a High Frequency Trading system

Since reading this, I've been doing a fair bit of digging around on HFT development - it seems like an incredibly interesting area to be working in.
I found this especially interesting:

Standard computer clocks are generally terrible at maintaining an accurate time over long periods. This is because a clock on a computer is just a counter, and the counter needs to be updated at exact time intervals to be accurate; computers are incapable of doing this, however, because they only perform operations at discrete times defined by their chip clock frequency, the period of which will never be an exact multiple of the required time interval. Here is a link to a more complete description of this and related issues.

Unfortunately the link is broken which is a shame.(Edit: just tested it again and it works)

I wonder how you get around this, there was mention of a GPS clock, but they can set you back a good $2500.

Has anyone else on here encountered this issue, and how did you fix it?

Btw this is an interesting read:

http://www.beyonddiscovery.org/content/view.article.asp?a=458
 
I imagine a mere 2.5k isn't a lot if you're bent on HFT - your pool is certainly higher anyway.
 
I imagine a mere 2.5k isn't a lot if you're bent on HFT - your pool is certainly higher anyway.

Yep I agree, it is a small amount in the grand scheme of things. However even a GPS clock isn't perfect it would seem and I would imagine people boot strapping their own system at home would probably put off (rightly or wrongly) splashing out on hardware until it was necessary.

Maybe someone with the experience can answer his, but is jitter and losing time on the clock still a problem faced by HFT platforms - or has throwing money at hardware largely made this a redundant issue?
 
MIT released a paper titled Developing high-frequency equities trading models

Abstract: The purpose of this paper is to show evidence that there are opportunities to generate alpha in the high frequency environment of the US equity market, using Principal Component Analysis (PCA hereafter) as a basis for short term valuation and market movements prediction. The time frame of trades and holding periods we are analyzing oscillate between one second to as high as 5 minutes approximately. We particularly believe that this time space offers opportunities to generate alpha, given that most of the known quantitative trading strategies are implemented in two different types of time frames: either on the statistical arbitrage typical type of time frames (with valuation horizons and trading periods in the order of days or weeks to maybe even months), or in the purely high frequency environment (with time frames on the order of the milliseconds). On the latter strategies, there is really not much intention to realize equity valuations, but rather to benefit from high frequency market making, which involves not only seeking to earn profit from receiving the bid/ask spread, but also from the transaction rebates offered by the numerous exchanges to those who provide liquidity. We believe that there are more opportunities to capture existing inefficiencies in this arena, and we show how with very simple mathematical and predictive tools, those inefficiencies can be identified and potentially exploited to generate excess returns. The paper describes our underlying intuition about the model we use, which is based on the results of short term PCA's on equity returns, and shows how these results can predict short term future cumulative returns. We randomly selected 50 of the most liquid equities in the S&P 500 index to test our results.

http://dspace.mit.edu/handle/1721.1/59122
 
are there any blogs or articles written by quants in the field who have developed HFT systems?? most of this stuff is proprietary, so very hard info to find externally
 
are there any blogs or articles written by quants in the field who have developed HFT systems?? most of this stuff is proprietary, so very hard info to find externally

Never heard about this. The only blog which is close that I'm aware about it's a Matt Davey's Tales from the Trading Desk. Not HFT's quant but he covers a lot essential things in his blogs from order management and communication with exchanges through appropriate data strucutres, algorithms and tecnhologies to general trends in algo trading.
 
I wonder how you get around this, there was mention of a GPS clock, but they can set you back a good $2500.

Has anyone else on here encountered this issue, and how did you fix it?

I haven't had to deal with this much myself, but I just wanted to chime in that you can easily get a GPS time signal for next to no cost. All you need is a GPS receiver of some kind (standalone, the one in your phone, etc.) that'll let you directly access it's output. All GPS chips output their data in a format called NMEA 0183 (and/or NMEA 2000 for newer chips); you can parse this yourself or use a pre-existing library to extract the time signals from the satellites and sync your server's clock to the signal at some fixed interval (say, 10 minutes).

It may not be quite as accurate as that $2500 system, but it should certainly be good enough for testing/research purposes until you want to spend the money for a better setup.

Edit: NMEA 0183 on Wikipedia
 
Never heard about this. The only blog which is close that I'm aware about it's a Matt Davey's Tales from the Trading Desk. Not HFT's quant but he covers a lot essential things in his blogs from order management and communication with exchanges through appropriate data strucutres, algorithms and tecnhologies to general trends in algo trading.

I'll second this. "Tales from the Trading Desk" covers a lot of great technical material useful for quants and algo traders. From what I've seen, it also tends to lean a bit towards Windows and .NET-based development.
 
thanks guys, I thought as much. I read the article that Andy Nguyen posted about but it's a thesis project, so its validity remains to be seen, unless of course you come from within...
 
Back
Top Bottom