Best way is to build up a simple model using just 1 instrument and try it out. I'm sure a best-fit algorithm can be found from the analysed signals. Of course one would try all available tools in one's arsenal, ie. indicators, oscillators, all the data of the instrument, and of course some maths... And also considering the data of the sector and market indices, and of similar instruments that could have a significant impact on the traded instrument... ie. correlated movements etc... In sum one would do solve an optimization problem by doing some machine learning and curve-fitting... The idea of using the local lows and highs (can also be of MAs) as signals, and from that finding an algorithm (ie. developing a system) is IMO brilliant, but I think we all at some time in our research tried this already... The core of the problem is simply "the next bar problem", as it is simply impossible to predict correctly the direction of the next (ie. future) bar or tick... But recently I read in an article that someone is able to predict the next x seconds or even minutes. I should check that again, but I'm more a statistical type, so I don't believe one can correctly predict the direction of such random events... OTOH correctly predicting even just some seconds or minutes would be a big plus in this game. I'm currently researching correlation analysis, maybe it can help with this problem when used with strongly positive and strongly negative correlated stocks etc...
I tried this with AI. Wanted to do this for a few instruments and at the end combine them all. I don't remember how many calculations this computer did in an hour, but after weeks of testing there was not a single indicator that could generate profits in ALL marketconditions. And if I ran the program again with data from the next week all values from indicators changed after optimalization. Conclusion was that continues optimalization was necessary to survive, not even making big money. The ultimate prove that building a stable and profitable system in ALL market conditions is not "simple".
Did you guys throw Eureqa at the problem...like I did long ago? As well as multiple ML algos? http://www.nutonian.com/products/eureqa/
Congratulations, you are being unbelievably pedantic. No kidding that they don't "steal" but "copy". Yes, I understand how data works. My real job is as a software engineer. My point still stands.
How is your experience with it? If a trader programmer would come up with such a model, people would demand to show results of the profits it makes, and would ask him why he sells it if it is so good, why not apply it yourself and get rich. I wonder why that company isn't using it themselves at the stock exchange... ;-)
Exactly. It is good at what it does, but it doesn't forecast the markets well at all, imo. It basically curve fits...like most other ML.
I'm not a programmer, but just curious how much work would be involved converting 700-1000 lines of EasyLanguage strategy code into a personal custom built platform? In my case, charts and accountability wouldn't be necessary, since I am able to see that in the broker interface. Basically just a pure on/off automation tool with inputs for multiple instruments