Please tell us your opinion. - page 6
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
It is difficult to tell anything about the robustness of the system from the shape of the curve. Robustness is in the idea. For NS, the idea is the correct preprocessing of the input data, with topology etc. taking second place. The input data should most accurately and unambiguously describe the market process being used, and for that you need to have an idea of it. And the right criteria for system rejection rule.
You have written it right. I agree. But it's all too general to find the right solution. As they say "about everything and nothing". I want to be specific.
In addition to the curve shape, you can see a lot of other useful information. Profit, number of trades, mathematical expectation, etc...... It seems to me this information may be useful when estimating working capacity of my TS in the future. Or am I mistaken?
Although the last sentence is not clear.......
The point is that it makes no sense to give results without disclosing the whole system. If you disclose it, then those who successfully trade similar systems will be able to help you. NS is not a system, the point for them is almost entirely what's in the input. Stability is in the idea, in its nuances, but not in the indicators.
About the failure criteria. You have decided to use the system, then come the results of real trading. When and under what conditions is the system discarded? This is what the success of trading depends on. There are no eternal systems and any system can crash. Universal adaptivity is a myth, a dream of the Grail, the dream of a smart machine that thinks for us))))
For example, the criterion of failure is the average profit per trade for a certain period and comparison with the reference (historical) one. Rejection if it has become lower. In fact, it is a trend following on the equity. There are other methods, but the efficiency depends on the specifics of the system.
The point is that it is pointless to give results without disclosing the whole system. If you disclose it, those who have successfully traded similar systems can help you.
Are you suggesting that you post the code here?
The criteria for rejection include, for example, estimating the average profit per transaction over a certain period and comparing it with the benchmark (historical). Rejection if it has become lower. In fact, this is trend-tracking on equity. There are other methods as well, but the efficiency depends on the specifics of the system.
Comparison of average trades for profit is interesting. What other methods are there?
So we talked and I decided to see: How would the probability behave in general, if it is deprived of all neuro in our 2008? I made a simple Expert Advisor that looks at the last some bars and calculates the probability of the next bar (up/down), the bars passed during testing are being "improved", i.e. they also participate in calculating the probability of the next bars.
I took the same pair, the timeframe is the same, the period is the same 2008 present day (2001 is in the picture, don't believe your eyes, period limitation in the Expert Advisor has been implemented for availability of historical data). Parameters are almost nothing, probability of opening, probability of closing and number of bars for probability calculation. Parameters are taken according to my own intuition. See the full attribute for the full picture. Result:
The year 2008 is so predictable for EURUSD that the EA of 200 lines written in an hour without any learning/optimization shows tolerable results. No wonder that your Expert Advisor shows good results. If you feed this "bare" probability to input even the most primitive neural network, the results will be comparable.
Z.I. Checked it on 2007, up to mid 2007 it's growing at the same rate as in 2008.
The point is that it is pointless to give results without disclosing the whole system. If you disclose it, those who have successfully traded similar systems can help you.
Are you suggesting that you post the code here?
You can figure out for yourself what you have got as a result, though with NS it is not quite easy.
IMHO, NS must use 1 type of trading, momentum trading, patterns, sessional etc. Adapt the method to a changing market, not a constant search for an incomprehensible composite. The input data, the learning criteria, should take this into account. It is necessary to limit the NS to a particular method and not give it as much freedom as possible. If several methods must be used within one system (TS portfolio), then several independent NS must be used. Otherwise, there will be no stability of the results. There will be a constant adjustment, though occasionally the robust variants can appear. But it's temporary, because there will always be some adjusted variants.
The criteria for rejection include, for example, estimating the average profit per transaction over a certain period and comparing it with the benchmark (historical). Rejection if it has become lower. In fact, this is trend-tracking on equity. There are other methods as well, but the effectiveness depends on the specifics of the system.
Comparison of average trades for profit is interesting. What other methods are there?
For example, using several averages. If average profit per trade for a small period N1 became less than 0, then the lot is decreased by one third, if it then crosses even slower, then another one third. And in the opposite direction the increase.
Using the MIDD and stopping the system trading if it is reached, usually multiplied by a factor. In fact, it is an analogue of the trailing stop on equity.
Statistical analogue of the previous one, but based on calculation of sigma and exit of equity over 3sigma.
Equity should be trending (positive average profit per trade), trend and tracked. Resumption of trading on the system is another issue. Again it is difficult for NS in case it uses not one method but a composite. You can always resume trading at the peak of the fit.
Not a fact. It is even more likely to be worse.....The probabilistic network does not work as an input enhancer. It calculates probability based on input data. And probability from probability, what you are suggesting is something too complicated. So it is impossible to say with certainty that it will be unambiguously better.
So we talked and I decided to see: How would the probability behave in general, if it is deprived of all neuro in our 2008? I made a simple Expert Advisor that looks at the last some bars and calculates the probability of the next bar (up/down), the bars passed during testing are being "improved", i.e. they participate in the calculation of the probability of the next bars too.
I took the same pair, the timeframe is the same, the period is the same 2008 present day (2001 is in the picture, don't believe your eyes, period limitation in the Expert Advisor has been implemented for availability of historical data). Parameters are almost nothing, probability of opening, probability of closing and number of bars for probability calculation. Parameters are taken according to my own intuition. See the full attribute for the full picture. The result:
I just don't understand, is it a period of optimization or no optimization at all?
The year 2008 is so predictable for EURUSD that the EA of 200 lines written in an hour without any training/optimization shows quite tolerable results. No wonder that your Expert Advisor shows good results. If you feed this "bare" probability to input even the most primitive neural network, the results will be comparable.
Z.U. Checked it on 2007, up to mid 2007 it's growing at the same rate as in 2008.
Well personally my opinion unlike yours is that TC can't work always and everywhere...... So for me it's ok.....