Martingale is not evil at all, it brings profits - page 9

New comment
 
Reshetov: If the probability of failure is 0.3, then 14 consecutive failures will occur with probability 0.00000004782969

At what length of test series? Yura, you have to understand that on a long enough Bernoulli test series, this probability will be as small as one :)


P.S. On a series length of 14 trials, you're probably right (0.3^14). But 14 deals isn't serious.
 
Mathemat:
Reshetov: If the probability of failure is 0.3, then 14 consecutive failures will occur with probability 0.00000004782969

At what length of test series? Yura, you should understand that on a long enough Bernoulli test series this probability will be as small as unity :)


P.S. At a series length of 14 trials, you're probably right. But 14 deals is not serious.

And besides - you would need unreal deposit to cover 14 failures... And the 15th successful deal should have a sufficiently large rollback to cover the previous ones.

In my considerable experience of Martingale trading - more than 7 trades is a very big risk. Ideally 3-5 trades with a multiplier of 2.
 
If we are talking about determining the maximum number of losing trades, we can draw an analogy to the martingale based on averaging. If we use averaging (i.e. if we catch the bounce), we can find the maximum length of a non-bounce period on the history and use it as a calculation one. And on its basis, determine the risks by calculating the lot sizes of each following deal. That is why this martingale system is no more risky than the usual MM system
 

Continuing the topic of calculating price consolidation levels...

Below is a picture for the pair Funtjen on TF15... Time frame 1000 bars...

Maximum at 200.10... Minimum - 192.61 and 202.29...






Now a time frame of 10,000 bars...



 
Meat:
When we start talking about determining the maximum number of losing trades, we can draw an analogy to the martingale based on averaging. If we use averaging (i.e. if we catch the bounce), we can find the maximum length of a non-bounce period on the history and use it as a calculation one. And on its basis, determine the risks by calculating the lot sizes of each following deal. That is why this martingale system is no more risky than the usual MM system.
Finally, at least one trader has considered the issue without leaving the market. If MTS contains means of determining the direction of movement at the moment, it already gives statistical advantage at least in that this direction will be kept for some time, tests show that this scheme takes maximum 3 steps with doubling of a lot to get out of a drawdown. Hence a simple conclusion - if in three steps of doubling the system did not come out of the drawdown - it is time to do RESET.
 
Feller rules. But it's not easy, a whole theory of recurrence events was required. Here is part of page 337 of volume 1 with the results:


Here a "trial" is a trade that has two values - profitable/losses. The return time is such a number of Bernoulli trials (trades), at which the event "series of successive successful outcomes for the first time reached length r" occurs.

As we see, if we consider a losing trade as a success, its probability is 0.6 and the needed series is at least 15, we need 5400 trials (trades) on average to meet such a nasty series there. And if the parameters are different, then we apply formulas (7.7).

Martingale, relax early: for example, with a more realistic estimate of probability of a loss for the classical martingale (with doubling of a bet), equal to 0.75 (3 losses per 1 profit), applying the first formula of (7.7), we obtain (p=0.75, q=0.25, r=14) an average number of deals equal to about 0.982/0.25*0.75^14 ~ 220 deals.

2 Yura Reshetov: I did not understand your martingale exactly. Maybe it's not so aggressive.

2 Yuraz: Maybe I overestimated the probability of a loss (0.75)? What is the value of this figure in the system you made?

P.S. Analysis of a modified MoneyRain system:

Total trades 386 Short positions (% of winners) 146 (28.77%) Long positions (% win) 240 (42.92%)

 Profitable trades (% of all) 145 (37.56%) Profitable trades (% of all) 241 (62.44%)
Largest profitable trade 9317.00 losing deal -5555.00
Average profitable deal 1130.56 losing trade -262.14
Maximum continuous wins (profit) 6 (5598.29) Continuous losses (loss) 11 (-3376.82)


p=0.6244, q=0.3756, r=11. Average number of trades according to Bernoulli's scheme, giving the same series of 11, by the same formula turns out to be 473. You have met it a bit earlier, Yura. So far it turns out that the tests are sort of like independent...

 



Remarkable picture... It's a pity no one has commented... Maybe, I'm in my wave and I don't notice something, which is obvious for everybody...

At the time interval of 10000 bars we have 4 zones of price consolidation, and we can see their borders ...

Example of a strategy that is not difficult to implement as MTS ...

Averaging towards the nearest boundary ... Exit by Takei (standard value) or Stop Loss on the border of moving from one zone to another ...

Waiting for comments...

 
Mathemat Bulashev in his book "Statistics for Traders" has a topic

13.12. Probability of making a loss in a series of successive trades .

of successive trades .

How does it apply to the subject we are considering?

Files:
doc1.rar  173 kb
 
Thanks, lovova, but the archive hasn't opened (if it's Bulashev, I have it). I looked it up. 13.12 is actually a statement of the binomial distribution and calculations for a small number of trades, while 13.13 seems to be numerical modelling of the loss series. Yet Feller gives it all in analytical form, while Bulashev only says that there are analytical difficulties involved and suggests a numerical procedure.

P.S. The archive has been opened.

 
Mathemat:
Thanks, lovova, but the archive doesn't open (if it's Bulashev, I have it). I looked it up. 13.12 is actually a statement of the binomial distribution and calculations for a small number of trades, while 13.13 seems to be numerical modelling of the loss series. Yet Feller gives it all in analytical form, while Bulashev only says that there are analytical difficulties involved and suggests a numerical procedure.
But Feller what seemingly unfeasible to implement in MQL
12345678910
New comment