This is why AI will not be able to conquer the financial market. Unlike Chess or Go, where AI can be trained on past games, it has no winning strategies to train when it comes to trading--unless RenTech is willing to share its cash cow (yeah, right, keep dreaming).
Doesn’t it take a long time to come up with a workable model by limiting yourself to forward testing? I would think that at least some quick and dirty back testing coupled with some out of sample testing before doing forward testing in real time could save you some time.
Hello schizo, Can you please re-explain or simplify? I am do not understand. I thought the AI can go back in time and look for winning algos?
Training AI models require a huge set of data. We've had IBM Watson beat the Chess master and Google's AlphaGo beat the Go Master. They were first trained using thousands of games that were available (eg. previous games played by humans). Likewise, to come up with a moneymaking AI in the financial markets, you'll need not only large data sets, but profitable strategies to train it on. But where will ya get that from? It ain't like you can buy it on the web.
When trading with own money it is not actually forward testing, It's a euphanism because the intent is to profit whilst flying from the seat of your pants, with money management to save you. I actually wanted to backtest my current method but I no longer have the tools, my software and hardware long gone, besides I'd struggle to code it as I'm now rusty in that software. My previous backtesting endeavours never produced any meaningful results. I generally can eyeball several charts and determine whether something will work. The key to trading imo is this. What is the common denominator that the majority are looking/referring to in trading. There are several common denominators and the more of them which line up results in greater probability.
I get what you’re saying. But even that is the result of a form of back testing, formal or informal, in a manner of speaking, no? You can spot something on a chart that resulted in an favourable outcome, and you can then check other such instances on a chart to see if it is loosely repeatable with some reliability by checking both backward and forward. Over time, you refine your criteria, and you then check again, both backward and forward. It’s an iterative process. Screen time. As Yogi Berra once noted, “You can observe a lot just by watching.” While it may or may not be as systematic or as substantial as a computer back test, the argument can be made that you get a much better feel for the landscape around such setups. And that is not without value. Again, just my opinion.
Yes, I think that's right. As we know when traders speak of backtesting, it's basically software they're talking about which is rapidly number crunching via filters. Running your eye over some charts is not quite as quick and more difficult when you have several criteria. I think software crunching is more accurate than a discretionary (eyeballing) approach. When I mentioned...... If one uses common sense it becomes not very difficult. The difficulty is finding something simple in a sea of noise. Traders are pulled in every direction with (media) noise. Trading would have more liars and scammers than anything, but this time I won't raise the subject of religion.
There are times one should be trading and times when one should not be trading. Knowing the difference makes a big difference to the bottom line.