Since yesterday evening I have been studying neural networks in the Neuro-Lab program to Velslab. Previously, there were periodic attempts to understand this issue and the case gradually moved forward.. The main obstacle was, of course, bias against the method of using neural networks in the game on stock exchanges and over-the-counter markets, since there is an opinion that this is just a fit to the investigated curve. But now I began to doubt such a one-sided approach. Though, a neural network is like a reasonable human head, who comprehends the world and learns from his mistakes. Likewise, here the hidden esoteric meaning of the pattern-weaving of a particular graphic is comprehended..
But first, certainly, you need to deal with the program itself — what where to click and what where to look. After intense study, the situation cleared up. Everything can be done without any programming skills at all, so I think, зря пугали в video из предыдущего поста что спекулировать на биржах без знания языка C# невозможно. Moreover, the program outputs the simplest indicator with values from 0 to 100 and using it, all that remains is to create a script through the strategy wizard. Well, or as a last resort, if there is a ready-made complex script, add indicator manually — this is a couple of lines in total. Here is an example of the entire system in the strategy wizard:
I tried to train RTS futures (Daytime) on patterns. First I took InSample first 50% graphic arts. That is, the training took place in the first half of the schedule.. Then he mastered the strategy — to 2011 the system earned, but then a small drain began. Then I trained the first 75% graphic arts. In this case, OutOfSample has already stopped merging, but it didn’t make money either. Then I trained the last 20%. In this case, on OOS (First 80%) it turned out to be a normal upward curve. That is, we can assume that before 2011 any systems earned on Ri for years, not only trendy, the main thing is to systematically.
Here's what happened for one RTS Index futures contract without slippage and commissions:
But this, of course the result is educational and theoretical. Why educational? Because just one trained indicator was applied without any filters and conditions, received for 2-3 symbolic workout minutes, just to see what comes of it. Why theoretical? because schedule futures on RTS, which we are using has the wrong bars, not corresponding to trading reality due to the fact that the OPEN of a new day always starts at the CLOSE level of yesterday, and sometimes it is impossible to open positions at this price. Therefore, it is impossible to accurately test on such a chart.. I.e, the graph is purely theoretical, а не реальным. Therefore, the result is also theoretical.. Just comprehending neuro-trading… :)