Hi Rien - re using real stocks
I have been thinking about this, but why not just pick real stock prices instead? Yeah, no reason not to. Real stocks would be backtesting, and would be interesting but different. You can think of it as the artificially generated data (AGD) being your experimental control, and the real stock data (RSD), whether backtesting or realtime, as the subject data. With the AGD, they contain only the information that you put into them, ie the std dev and the length of the series in the simplest case I mentioned. The RSD represents the information input of the entire market; necessarily more difficult to understand.
If the algorithm closely mimics real stocks, then we can just as well use real stocks. The reason for the artificial stocks is to be able to quantify just how well an algorithm performs on a stock with certain characteristics. I think that's the gist of the excercise - connecting the AGD and RSD through some characteristics, whether aspects of the stocks or the AIM algo.
This should give you a better "feeling" for the algorithm. I'd call this an intuitive understanding of the process; that's a subjective goal.
The other approach would be to take the whole universe of stocks and let the algorithm loose upon it. That would require quite some computing power, and you would still face the selection problem. Mm, I think this is both a less formidable task computing-power-wise than you think, and maybe less informative too. In terms of CPU cycles, tha AIM algo isn't very demanding, and there's only a finite world of stocks you can test. I'd guess, say, 10K to 100K different stocks? The download and storage problems might be more formidable. There's so much information in all those histories that you may end up with a mush for results. Maybe a handfull of competing companies in the same sector over the same period would be more manageable.
Anyway I will use CSV files as the stock-database so, anyone could create any stock data he likes and play with that. I suppose that most spreadsheets are able to generate CSV files. Uhhuh. You can always call your AIM program from a master program that generates the test data. The function is called system() in C/C++, and an analog surely exists in Java. You'll need to communicate through files, which is a performance hit, but you shouldn't need a lot of performance for this app.
hth Tim
AI
