Determine the future operability of the vehicle. - page 7
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 can be the other way round
Almost 100% match with the exception of a few pips of requotes.
Also understandable. But that's not exactly what I'm asking. These are the principles of building an un-optimized TS, preferably without optimization. And I am trying to find out how from the reports from the period of optimization and OOS period to determine the capability of the TS in the future.
To gain sufficient OOS statistics try the following strategy:
You have an optimization period of 8 months and an OOS period of 1 month.
Choose 10 optimization and validation periods. For example:
1. From 01.01.2007 - 31.08.2007 optimize. Then from 01.09 - 30.09.2007 check.
2. from 01.02.2007 - 31.09.2007 we optimise. Then from 01.10 - 30.10.2007 we check.
...
10. 01.11.2007 - 30.06.2008 we optimise. 01.07-31.07.2008 we check.
By checking these statistics you will be able to make statistically reliable judgements about the system and its parameters.
To gain sufficient OOS statistics try the following strategy:
You have an optimisation period of 8 months and an OOS period of 1 month.
Choose 10 periods of optimisation and validation. For example:
1. From 01.01.2007 - 31.08.2007 optimize. Then from 01.09 - 30.09.2007 check.
2. from 01.02.2007 - 31.09.2007 we optimise. Then from 01.10 to 30.10.2007 we check.
...
10. 01.11.2007 - 30.06.2008 we optimise. 01.07-31.07.2008 we check.
Having collected these statistics you will be able to judge about system and its parameters statistically reliably.
I agree, "there is a letter in the word". But there is one "but". Those patterns that were found 8 months ago, for example, may not work at the moment. And there are a lot of such examples. I, on the other hand, try to find some confirmation of TC working in the near future as I do not believe in the eternal existence of TC.....
I agree, 'there is such a letter in the word'. But there is a "but". Those patterns that were found 8 months ago, for example, may not work at the moment. And there are a lot of such examples. I try to find some confirmation of TC in the near future, because I do not believe in the eternal existence of TC.....
>> Bring me an example, if possible, where forward analysis was done as it should be (for example the way it is suggested above), and the system was profitable, but as soon as the time of its use has come it all went to hell...
I agree, 'there is such a letter in the word'. But there is a "but". Those patterns that were found 8 months ago, for example, may not work at the moment. And there are a lot of such examples. I try to find some confirmation of the TS work in the near future, because I do not believe in the eternal existence of TS.....
Each individual trade has its own level of risk and probability of winning. The same applies to a series of N trades. If we know and control these characteristics, we can control the profitability of the system. In fact, the purpose of forward testing is to estimate how close the system's risk level and probability of winning is to the one we set it to during the optimization period, and how stable these indicators are to market changes.
In order to assess it, one monthly forward testing is clearly not enough. We need the statistics of a series of tests. If the results of a series of forward tests (for example, 10 one-month tests close to the time of the real trade) give consistently similar results by the level of risk and probability of winning, we can confidently assume that the same parameters will characterize the system in the real trade.
We use the system as long as the real results correspond to those shown in the forward test.
But there is a "but". Those patterns that were found 8 months ago, for example, may not work at the moment. And there are a lot of such examples.
This suggests that the number of patterns is small, there is no generalization between them, and the network just remembers the input sample
I am trying to find some evidence of TC in the near future because I do not believe in the eternal existence of TC.....
The greater the statistical number of instances when the system worked on the history, the higher is the probability that the system will continue working successfully in the future and as a rule the system will work in the area where it was trained after 8 months and a year with a slight equity correction.
The greater the statistical number of cases when the system works on history, the higher is the probability that the system will continue working with the same success in the future, and as a rule, this system will work in the area where it was trained after 8 months and after a year, with a slight equity deviation.
Do you have any examples? Everything is clear in theory, you are saying the right things. But the practical application?
Any examples? It's all clear in theory, you say the right things. And the practical application?
I don't have the actual results of application in real trading even on demo account yet, now I'm busy transferring the system to MQL in the terminal.
but i have shown the test sketches on the alpari forum .
no actual results from real trading even on a demo account yet, currently busy transferring the system to MQL in the terminal.
But I showed the test sketches on the Alpari forum .
What I saw on Alpari is not a very good example. Firstly, the drawdown is large, equities are not smooth and secondly, I am not interested in the optimization (or training) period but in OOS in the first place, this is three. The optimization period is shown here, not the OOS. Everyone can make a good fit during optimization but what will happen at OOS is a big question and how long will it work at OOS is also a big question. That's what we're talking about here. The more statistical cases the system works on history, the higher is the probability that such a system will work successfully in the future and as a rule such a system will work in 8 months and a year with a slight change of equity. Everyone here knows that. I'm trying to find out and understand the specifics.
What I have seen on Alpari is not a good example. Large drawdown is one, flat equity is two, and we are not interested in the optimization period (or training) but in the OOS in the first place is three. The optimization period is shown here, not the OOS. Everyone can make a good fit during optimization but what will happen at OOS is a big question and how long will it work at OOS is also a big question. That's what we're talking about here. The more statistical cases the system works on history, the higher is the probability that such a system will work successfully in the future and as a rule such a system will work in 8 months and a year with a slight change of equity. Everyone here knows that. I'm trying to find out and understand the specifics.
and you don't know how to poke your nose in the mud, you don't know how to????
I'm not bumsy, it's bad here, that's one, two.... to understand the specifics, read carefully what people write!
each system has its strengths and weaknesses, why not talk about the strengths of systems to strive for and not look for their own weaknesses? not about my drawings, but in general.
these drawings, february march april,
i'm either contradicting myself or i really don't understand anything, "what i've seen at alpari is not a very good example", but if we're talking about the number of events in the statistics, and how the number of events affects the probability of future performance, i have a higher probability density of profitable trades! even though your ratio of profit/loss is less than 1.27, you have 9 it seems.
Moreover, I had a 15-minute timeframe, while you have an hour one, it means you have 4 times less history, what's wrong with my drawing?
about the OOS, I wrote on alpari that the figures are not all history, the data that the network has not seen only the last 2-3 months (in the figure above the last 4 months), but it would be pointless to prove it if 90% of the forum people there do not have a primary education, the result shows one equity chart, because this system is built in a network of neurosolutions and processing occurs in a connected dll with the network, and it does not distinguish. I used ns to display the result and it is the same for dll, it doesn't change the final result for me, of course it was the best I had at that time. If you do not know what to do with it, then you will have to do it with a good result. So I will have to sacrifice either increased fidelity and stability of the system in the future but less equal equity; or more equal equity but less fidelity and less stable equity in the future, though it is also a rhetorical question, I can see it as a flaw in my TS. It is therefore silly to point out the weaknesses of the system when it has advantages that are more important than any other parameters. If the system was trained on 12 months of 15 minutes, the probability is higher that it will work with the same eequity angle in the future even though one of the months was with drawdown, The same drawdowns have been observed on the other side of the history, although if you look at the parameters that have been used to set the system, this month has only been 7% of the whole history that has been used to optimise the system, which I have seen on the following 2 months, March-April equity was growing and the angle will be saved, I just need to shift the history every month. which actually can be seen on the next figures.