Tools Beta Program

 Our Process

We first generate ideas for trading strategies either based on prior direct experience with the market, or by very carefully looking for anomalies or patterns in tick data.  We are looking to be able to predict a one to three increment move in any particular contract with a very high degree of accuracy.  We want to be in and out of the market as fast as possible to limit exposure to randomness that we cannot predict.  More often than not, we have found that this type of predictability is more commonly found in spreads or relationships between contracts rather than in the outrights directly.

 Our Tools

Successful high-frequency trading requires some very specialized tools.  Traditional tools commercially available might store five minute, one minute, or even one second data.  Most high-frequency trades are in and out in less than one second...so tick data, estimation of latency, and a realistic simulation-matching algorithm are indispensable to properly test whether a particular strategy will actually make money. 

All trading strategies are going to have some parameterization.  So after finding a strategy that seems like it might work, the next step is to find the 'best' parameterization for that strategy that is stable and consistent over an adequate time frame.  There are many ways to accomplish this.  Our method is to run a very large distributed genetic optimization on our parameter space to narrow in on the stable clusters of parameters.  From this, we prune our parameters to eliminate ones that have little effect, then re-optimize.

Technical Difficulties

It doesn't take a Computer Science degree (or maybe it does) to imagine what types of challenges you might face. 

First, how to economically store years of tick data at every level of the book which is being pumped off the futures exchanges at a rate of about 600 million ticks per day.  That's about 150 BILLION ticks per year.  Even when highly compressed, the resulting database is several terabytes in size and suffers from a phenomena called 'bit rot' where (due to cosmic rays or other sources) a few bits of data are corrupted every day.

Second, if we are going to use this database for back-testing, it must be optimized for very fast reading and seeking.  At this level , letting the computer processor touch the data would be way too slow so you need to have a very fast alternative to get data from the disk straight into memory. 

Third, simulation and live trades should provide appropriate feedback in the form of charts, pre and post-trade analysis, and statistics in order for the trader to see if there is anything that needs adjusting, or if there is any flaw in the implementation of the strategy.

Fourth, the path between simulating a strategy and running a strategy live should involve no code change and should be as simple as flipping a switch from simulation to live.  Your simulated matching engine and latency need to match reality pretty closely.

Beta Program

We are not a software company.  We are a prop trading company making millions of dollars a year from software that we have spent a long time developing.  We started off very small and have never taken a dime of outside money.  Whether your goal is just a few thousand a year—or a few million—we have found trading, done correctly, can be an intellectually challenging and rewarding way to make a living.

Of course we would never make our full suite of tools available (ok, we would for the right price) but we believe that even things like very accurate tick data graphs of outrights and spreads, or some limited ability to back test a strategy based on full book tick data, would be useful for a broad audience.

If you are interested in being involved in our Beta Program send an email to describing your current trading experience, what tools you currently use, and what you'd want to use our tools for.