I've spent a lot of time thinking about using various automated methods to discover trading patterns/strategies and I'm coming to the conclusion that it's mathematically impossible. When you apply some automated discovery mechanism you either have no preconceived notion or only a very general notion for what you are looking for. You then use some search technique to look at huge number of possible relationships and indicator/price permutations. If you are using a dumb random search approach you can easily search millions of possibilities. If you use a smart directed search algorithm (ie genetic algorithm), you can search the equivalent of billions or tens of billions of permutations. When you look at so many possibilities, you are guaranteed to find quite a few methods that work incredibly well simply by chance. Even worse, some of these methods will pass any statistical test you throw at it (sharpe ratio, t-test, edge test, correlation test, etc.) since all these test compare the observed results vs what you'd expect randomly. But when you look at a 1,000,000 random samples, you will obviously have some that perform better than 99.999% of random, thus passing any possible test you throw at it. So even if there are meaningful patterns that your search discovers, they will be intermingled with numerous patterns that work simply by chance. And there is absolutely no way to actually separate them out. I'd be curious to hear if anyone sees a flaw in this reasoning. -bulat
I think datamining is useful. I am developing trading system based on data mining techniques. Which data mining programs do you use? There is a lot of publication about data mining & trading.
Are you saying this because you've actually achieved meaningful results in real trading through a system developed via data mining or is this purely theoretical? -bulat
Sorry I have a only backtesting results yet. I am not a good programmer, It take me a lot of time to write somethink... First I tried probabilistic neural network and had a resoult about -3000US$ loss each testing month.. hehe.. Now I did another system based on decision trees and the system first month earned about 500US$, second month lose -300US$ I am playing with forex minilots. In spite of the fact I have used simple stupid pattern recognition technique and silly crappy decision rules, the system do not lose a lot.. it is enough for the beginning. I know more powerfull techniques but haven't implemented yet. I think the datamining is a future of trading. Computers are better in decision making and pattern extraction than man.
Will you please tel us what is your degree in? I am looking for a topic for a Phd desertation, mechanical/industrial, I was thinking it will be great if I can find a topic releated to trading! any thoughts?
flackac, You have advantage to use data mining and pattern recognition methods in searching for a profitable pattern. Do you have any exposure to knowledge base engineering, multi agent systems, learning and artificial intelligence?
I think so too! p.s. can I recommend a couple of books that address these imponderables? 1. "Market Models" by Carol Alexander (tough read) 2. "Fooled by Randomness" by Nassim Taleb (easy read) These texts will explain to you that no matter how much you back-test, your results will be susceptible to the "ever shifting sands" of the market. The geek term is "non-stationarity" but I'll spare you the detail. Still, can and must back-test. The key is that you must continually update your back tests using up to date data to incorporate the constantly changing structure of the market (especially volatility and correlation) acrary and others before him tell you that the ideal is to find multiple uncorrelated strategies and trade them together
bulat, Maybe I'm just ignorant (and I'm not kidding), but if a strategy you've discovered survives the statistical tests and performs just as well in an out of sample test then isn't that as much proof as you're likely to get? I guess I thought the patterns that worked merely by chance would not usually work in a good out of sample test.