MetaTrader does not reflect reality ! How can I fight this ? - page 18

 
Mathemat писал(а) >>
Yes, there aren't many trades. However, with so many trades, the profit factor is quite encouraging to be able to bank on the variability of the system's results.
How much more? This is OOS. Regularities that were found on that period are not working anymore and this TS gradually starts to give losses. So while this way (300 trades) evaluate the TS - the TS will stop working.....))))) And in reality there is already another TS, with other parameters....))))
 

Yeah, well, now I'm thinking that with so many trades, that profit factor probably isn't enough to decide it's profitable.

 
Mathemat писал(а) >>

Yeah. Now I'm thinking that with so many trades, that profit factor probably isn't enough to decide if it's profitable.

Isn't half a year of TS work enough? On hourly timeframe it's about 3168 bars.

 
LeoV писал(а) >>

Privat, please evaluate the TC. Optimisation until 01.09.2008. From 01.09.2008 OOS+real. Tested with 0.1 lot.

The result is perfect, of course. But there is not enough information to estimate it. We cannot say that the obtained result is better/different in comparison with the results obtained after optimization... For me the key parameter is maximum drawdown... If it has increased, it's a reason to revise TS parameters....

 
kharko писал(а) >>

The result itself is great... But there is not enough information to evaluate... We can't say that the result obtained is better/different from the results obtained after optimisation... For me the key parameter is maximum drawdown... If it has increased, this is a reason to revise parameters of TS....

Here is the optimization plot -

The symbol EURUSD (Euro vs US dollar)
Period 1 Hour (H1) 2008.02.01 00:00 - 2008.08.29 22:59 (2008.02.01 - 2008.08.31)
Model By open prices (only for Expert Advisors with explicit bar opening control)
Parameters -----
Bars in history 4591 Modelled ticks 8182 Simulation quality n/a
Chart mismatch errors 0
Initial deposit 10000.00
Net profit 27539.00 Total profit 69599.60 Total loss -42060.60
Profitability 1.65 Expectation of winning 140.51
Absolute drawdown 408.00 Maximum drawdown 8017.20 (19.05%) Relative drawdown 37.84% (6414.10)
Total trades 196 Short positions (% win) 98 (54.08%) Long positions (% win) 98 (59.18%)
Profitable trades (% of all) 111 (56.63%) Loss trades (% of all) 85 (43.37%)
Largest profitable trade 4490.30 losing transaction -2018.00
Average profitable deal 627.02 losing trade -494.83
Maximum number continuous wins (profit) 6 (5353.60) Continuous losses (loss) 6 (-4063.40)
Maximum Continuous profits (number of wins) 5353.60 (6) Continuous loss (number of losses) -4063.40 (6)
Average continuous winnings 2 Continuous loss 2

 
LeoV >> :
How much more could it be? >>This is an OOS. Regularities that have been found on that period are not working anymore and this TS is gradually starting to produce losses. So while in this way (300 trades) evaluate the TS - the TS will stop working.....))))) And in reality there is already another TS, with other parameters....))))

It turns out that a black box is created, which is not clear why it works profitably on the OOS, but the main thing is that it works. So it turns out that one can write some nonsense in terms of logic or play with different coefficients, see a plus on optimisation, see that the plus is maintained for a week or two and run it for real. It is the same as the chance.

In my opinion, we should understand the logic of the system and not try to get the odds right just because somebody somewhere is calling it "neural networks".

If there is an understanding of the causes of the system performance, you can already judge a lot about it. It is not a black box. You can analyze the algorithm. And how do you analyse a black box?

 
mql4com писал(а) >>

It turns out that a black box is created, which is not clear why it works profitably on the OOS, but the main thing is that it works. So it turns out that one can write some nonsense in terms of logic or play with different coefficients, see a plus on optimisation, see that the plus is maintained for a week or two and run it for real. It is the same as the chance.

In my opinion, one should understand the logic of the system and not try to guess the odds just because someone somewhere is calling it a neural network.

If there is an understanding of why the system works, then you can already judge a lot about it. It is not a black box. You can analyze the algorithm. And how can you analyze a black box?

Well, firstly, nobody has written that it is a neural network and it is not a neural network. Second - of course, the logic of how the system works is clear. Thirdly - nobody said that he wrote nonsense in terms of logic - no need to exaggerate. Fourthly - it was a question of assessing the quality of TC - good or bad. Therefore your tirade is not clear. What do you mean? I'm talking about evaluating the quality of TC, whether with or without a neural network...))))

 
LeoV писал(а) >>

Here is the optimization plot -

With which lot was the optimisation carried out? The expectation...? Too big a difference of more than 4 times....

I suspect the value is equal to 1 lot....

Now we can compare....

TC's profitability has more than doubled....

Specifically by the numbers.... Here are my ratios as a measure of profitability

3.43519521 - at optimization

7.346351815 - at RP

Conclusion: TC with these parameters on the takeoff... no reason to worry...

 
kharko писал(а) >>

With which lot was the optimisation carried out? The expectation...? Too big a difference of more than 4 times....

Oops, sorry, got confused - there's lot 1, here's lot 0.1 -

Symbol EURUSD (Euro vs US Dollar)
Period 1 Hour (H1) 2008.02.01 00:00 - 2008.08.29 22:59 (2008.02.01 - 2008.08.31)
Model By open prices (only for Expert Advisors with explicit bar opening control)
Parameters ------
Bars in history 4591 Modelled ticks 8182 Simulation quality n/a
Chart mismatch errors 0
Initial deposit 10000.00
Net profit 2753.90 Total profit 6959.96 Total loss -4206.06
Profitability 1.65 Expected payoff 14.05
Absolute drawdown 40.80 Maximum drawdown 801.72 (6.07%) Relative drawdown 6.07% (801.72)
Total trades 196 Short positions (% win) 98 (54.08%) Long positions (% win) 98 (59.18%)
Profitable trades (% of all) 111 (56.63%) Loss trades (% of all) 85 (43.37%)
Largest profitable trade 449.03 losing transaction -201.80
Average profitable deal 62.70 Deal loss -49.48
Maximum continuous wins (profit) 6 (535.36) Continuous losses (loss) 6 (-406.34)
Maximum Continuous Profit (number of wins) 535.36 (6) Continuous loss (number of losses) -406.34 (6)
Average continuous winnings 2 Continuous loss 2

 
kharko писал(а) >> Conclusion: TC with these parameters on the takeoff... no reason to worry...

In general, the most pressing question for me is how to assess the performance of the TS in the future....... There is no problem with the TSs, the only problem is how to understand how this TS will work in the future....))))