Determine the future operability of the vehicle. - page 8
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
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, on the other hand, am trying to find out and understand the specifics.
Let's think about it. What do we have?
First, we have indicators and performance indicators of the TS (Profit, Depth, Profitable trades, etc.), on which the developer focuses during optimization.
Secondly, we have the results of working of TS on the period of optimization by the selected indicators.
Third, we have standard deviation of performance indicators (if we do measurements on one-month periods of optimization, we will get 8 one-month results of the optimization period and we will know the standard deviation by performance indicators)
Fourth, we have the OOS results for the selected indicators.
What can we see from this data?
1. How much the OOS results correspond to the optimization results.
2. Whether the deviation of the OOS results from the optimization results is within the standard deviation.
3. whether the system will meet our requirements by simultaneously selecting the most unfavourable results within the standard deviation.
If the answer to all these three questions is affirmative, it is very likely that the system will behave the same way in real trading as long as the market does not change significantly in comparison to the period of optimization and OOS.
and you don't know how to poke your nose in the mud, you don't know how to????
I'm not a bumsie, it's bad here, it's bad one, two.... to understand the concept, read carefully what people write!
I'm not talking about my drawings, I'm talking about my drawings in general.
Oh, you're a piece of work. Did you mention the screenshot on Alpari? - I gave my opinion. What's the problem? If you hadn't mentioned it, I wouldn't have said anything. I would have posted a different screen, a better one. Maybe the opinion would have been different. If an opinion isn't good, it doesn't mean "sticking your nose in the mud". I don't see any dirt in my statement..... just an opinion. ))) Anyway, it's a good drawing, of course. Don't worry so much. )))
You've got to be kidding me. Did you mention the screen on Alpari? - I gave my opinion. What's the problem? If you hadn't mentioned it, I wouldn't have said anything. I would have posted a different screen, a better one. Maybe the opinion would have been different. If an opinion isn't good, it doesn't mean "sticking your nose in the mud". I don't see any dirt in my statement..... just an opinion. ))) Anyway, it's a good drawing, of course. Don't worry so much. )))
>> you give and I give, you have answered, I have answered and what's the problem? what do you do not like, if you had not answered I would not have answered, but I mean that you do not understand what you are writing, first read statistics and probability theory.
You won't :)
I have read both the topic and the forum (and not just this one). No one anywhere has come to a consensus.
Generally speaking, there are three main ways of determining the efficiency of a machine, so as not to fall into a fitting.
1. Optimization on a small sample by all the parameters and subsequent runs of OOS on still smaller periods with a shift. If after 10-15 such runs the system continues to make a profit, in my opinion - it works.
There are options here, as well as in the next paragraphs: optimize for all parameters or slightly modify the ones you've already obtained. Again from experience: the second variant will give 0 in the end (if we use the optimization tab limitations).
2. Optimization over the entire (almost entire) sample with the corresponding limitations and a small OOS period (for assurance). Such optimization makes sense only when the Expert Advisor uses some global properties of the market which are constantly present in it. Such optimization does not give a large variety of parameter variants and "grail" profits, but when it does not change its characteristics at OOS, then apparently it can be trusted.
3. So called "sequential" optimization. It is carried out for 1/3 of a sample (variants are possible), then obtained variants are run through OOS (not by hands of course ))))) and sequentially eliminated. I don't know, but in my uneducated opinion it's the best way to kill any TS.
I've had experience. Optimization for a period of 1 year. OOS run - same (almost the same) performance for 1.5 years.... Then, strange as it may seem, it goes down, moreover in July and October-November 2007 (????), and 2008 with the same parameters again in profit and without slippage. Not a pipsqueak.
Now I am trying to compare 1 and 2. I will write something about results after 2-3 days
About OOS, I am speaking only about shift forward, but not backward..... so it is more logical, and what is the sense to watch the past history?
IMHO, of course. Stats, links will be posted if anyone is interested.
PS. And to get closer to the market, especially if several orders are processed by one signal, during testing and optimization, it would be better to prescribe in the code, instead of the pause for return, after processing of each order, so that the next order would be processed on the next tick. Of course, this would be a pain in the ass, especially if they are interrelated ))
PS2 There are so many topics of the same type that I do not know where to write..... to the admins - should we merge?
From experience. The machine resource is there, so are the possibilities for expire.
I have read both the topic and the forum (and not just this one). No one has come to a common opinion.
Generally speaking, there are three main ways of testing serviceability, so as not to fall into a fitting.
1. optimization on a small sample by all parameters and subsequent runs of OOS on even smaller periods with a shift. If after 10-15 such runs the system continues to make a profit, in my opinion - it works.
There are options here, as well as in the next paragraphs: optimize for all parameters or slightly modify the ones you've already obtained. Again from experience: the second option in the end (if you use restrictions on the optimization tab) will produce 0.
2. Optimization over all (almost all) sample with appropriate restrictions and with a small OOS period (to be sure). Such optimization makes sense only when an Expert Advisor uses some global properties of the market, which are present in the market permanently. Such optimization does not give a large variety of parameter variants and "grail" profits, but when it does not change its characteristics at OOS, it seems to be trustworthy.
3. so called "sequential" optimization. It is carried out for 1/3 of a sample (variants are possible), then obtained variants are run through OOS ( not by hands of course ))))) and sequentially eliminated. I don't know, but in my uneducated opinion it's the best way to kill any TS.
I've had experience. Optimization for a period of 1 year. Run on OOS - same (almost the same) performance for 1.5 years.... Then, strange as it may seem, it goes down, moreover in July and October-November 2007 (????), and 2008 with the same parameters again in profit and without slippage. Not a pipsqueak.
Now I am trying to compare 1 and 2. I will have to wait 2-3 days and then I will write something upon the results.
I`m speaking about OOS, I`m speaking only about shift forward, but not backward..... it`s more logical, and what`s the sense to look past history?
IMHO, of course. Stats, links will be posted if anyone is interested.
PS. And to get closer to the market, especially if several orders are processed by one signal, during testing and optimization, it would be better to prescribe in the code, instead of the pause for return, after processing of each order, so that the next order would be processed on the next tick. Of course, this would be a pain in the ass, especially if they are interrelated ))
PS2 There are so many topics of the same type that I do not know where to write..... to the admins - should we merge?
It would be nice to read the whole thing in an article.
It would be nice to read the whole thing in an article.
you have to make up your own mind first.... I'm at a crossroads :)
I had an experience. Optimization for a period of 1 year. Run on OOS - same (almost the same) characteristics for 1.5 years.... Then, strangely enough, plum, and in July and October-November 2007(????), and 2008 with the same parameters again in the plus and without drawdowns. Not a pipsqueak.
The interesting thing is that I had the same thing. Well, maybe the dates are a little different. Also - the whole of 2008 is on the plus side, but August is on the minus side, so I wonder if it will work further.
The interesting thing is that I had the same thing. Well, maybe the dates are a little different. And also - the whole of 2008 is on the plus side, and August is on the minus side, so I think - will it work further?
I like this chart most of all :) ..... I still have equity, but it depends on the Expert Advisor. I don't want it to be without drawdowns, as long as the general trend is positive.
I'm trading with such trades, though on micro trades, but one should develop "confidence" not program it.
At work, I cannot download the picture, "someone else's" proxy does not let me in.... in the attache..... 563 deals it on optimization (2005), then OOS, plum - end of 2007, 2008, unfortunately, did not save, and parmeters destroyed out of spite :).
Criticism accepted )), especially since I gave up my watch anyway )
And this is a preliminary result of USDCAD Day 2005-2007 optimization inclusive:
profit - 1725.59 trades-60 profitability 1.74 MO - 28.76 drawdown - 895.82 drawdown % - 0.09% MinProfit (condition for modifying a stop or closing a single position) =5
SumProfit=65 (condition for closing a combined position in the direction) TakeProfit=0 StopLoss=400....
I agree in advance that profit is small and the number of deals does not go far enough, but as usual ......., imagine that "it works" on all currency pairs :), after all there is also TF240)