|
|
|
|
Our most recent focus in our research has been the application of a formal Design of Experiments (DOE) to testing and optimizing the parameter settings for automated trading systems. We design the experiments using Gosset, the ultimate experiment design program created by Neil Sloane of Bell Labs. The experiments are then run, sometimes by us but more often by the client. This is where the majority of work and cost occurs in any experiment or testing process. The main advantage of using DOE and a program like Gosset is that we can dramatically reduce the number of trials that are necessary for a robust test. In some cases, a few dozen trials can provide as much information as thousands of trials put together using the Brute Force approach. After the trials have been run, we use a variety of tools to analyze the results and can provide you with a picture of the sweet spot (ideal combination of parameter settings), the relative importance of one factor versus another, and the interactions between the factors. Note that we can do all of this without having to know the factors you are analyzing or the numerical settings you are using. All we need to know is how many factors you have, whether the settings are continuous or discrete, and if discrete, how many discrete states you are looking at. We provide you with the design with all of the factor settings noted in the plus one/minus one scale, you convert those for each factor and trial, and you run the experiment. After you run the trials, you can send the result to us for analysis using the plus one/minus one scale. We can do the analysis and report the results of that analysis without the slightest knowledge of what you are doing. Naturally, we can help even more when you let us know what you are doing, but we understand why that is not possible in many cases. Security is critical when your system may be a huge breakthrough. If you can't tell us what you are up to, we are not going to let that get in the way of getting you the best results. Note that we also can rapidly and inexpensively develop trading bots for almost any strategy that trades on an exchange on a public market. Of course, we haven't figured out how to do this without finding out what you are doing, but there's a cost to everything. |