I specifically qualified it "with indicators" but in general ML is glorified curve-fitting. I do not promote APS, I talk about it with other users. Actually I do not recommend it to anyone with less than 10 years of system testing experience. Why are you getting so sensitive about it? Any trading monkey knows that ML with indicators results in curve-fitting. APS by the way is not ML. I will keep promoting it more now that it is not available for sale. If you have a problem with me talking about software you do not have vested interest let me know. Please let me know what to talk about, how and when. Also, please let me know whether you ever traded a single contract or share and whether you have ever used a ML program. That would be interesting. Let me know which and I will ask you specific questions about its use if I have the time of course. We will have fun.
Maybe, but James Simons, who ran Renaissance Technologies, which was mentioned in the same post, did pretty well also. Ok, Ok, I slipped on that one, but I"m sure you understood my message; I'm up in Tahoe at the moment, and shouldn't even be on the internet much.
You can talk as much about software as you like, that's what we are all here for, to exchange ideas. It's just that when you disparage something, and simultaneously promote it elsewhere, it seems a bit odd to me. Just wanted to make sure you understood what ML was, before disparaging it. I see you don't. Cheers.
Totally agree, except for the part on neural networks. We use them extensively for classification and they work fine (even better than SVM in our case). But, many people that try them seem not to understand what they do and how to train them (not that I try to say that you don't).
I should have said they <b>easily can be</b> a complex form of curve fitting if not used an intelligent manner. In the wrong hands, they are a dangerous tool. In responsible hands, it is just another powerful tool.
Totally with you on this one. No matter how you come up with a system with a perceived edge it can be by chance. Many seem to think that just because they found a system "manually" it is not "fitted". What makes it harder for us that rely on ML is that we need to understand not only the likelyhood that our performance results from "chance" (some part of it ususally does since curve fitting/selection bias is part of most ML algorithms - that's part of how they "learn") but also to what degree the system will deteriorate when we walk-forward test them. If I find a system with a Sharpe Ratio of 5 on the training data I'm pretty certain that it will not hold in walk-forward testing. The big question is the probability that the ML algorihm have found something that's still useful (like real trading SR >1).
Thank you for the persmission to talk freely. You still do not get it. APS is not based on ML. My question stands: have you ever traded a single share or contract and which ML program have you used?
hi, Would like to know how you know that these hedge fund managers use ML ? Where is your source ? thanks