Author's dialogue. Alexander Smirnov. - page 26

New comment
 

to GODZILLA

At least something cleared up. Ү knew that JMA is resource-intensive!!!

 
GODZILLA:
In general, the most ideal averaging algorithm in the Expert Advisor is not the one that shows the maximum profitability in the optimizer, but the one that is just more or less stably profitable beyond the right border of the optimization period.
+1.
 
Mathemat:
I don't understand how you can talk about indicator trading parameters outside the context of EA that uses them. ANG3110, please clarify what strategy. And second: what is Risk, as a formula (everyone understands it differently)?


I've described how the experiment is constructed a couple of pages before. But it's not hard for me to repeat it. The indicators are taken. For noise immunity, a threshold condition is inserted to make them less twitchy.

For example:

extern int d = 2;

Then we put di = d * Point in init() and add variable si = Close[Bars - period -1];

And then in the loop

if (d>0)

{

if ( MathAbs( ma[i] - ma[i+1] ) >= di ) si = ma[i];

ma[i] = si;

}

There can be other variants not necessarily stepwise.

For the experiment, the Expert Advisor is made reversible, so that if ma changes direction, it will flip in reverse:

if (t0==Time[0]) { map = ma; t0 = Time[0];}

ma = iCustom(......., 0);

fss = 0;

if (ma > map) {if (fs==2) fss=1; fs=1;}

if (ma < map) {if (fs==1) fss=2; fs=2;}

By doing this, we guarantee a single trigger.

Then if (fss==1) { close the Sell order and enter a Buy order}

if (fss==2) {close Buy order and include Sell order}

That is all.

Risk is the number of lots

AFM=AccountFreeMargin();

LC=MarketInfo(Symbol(),MODE_LOTSIZE)/AccountLeverage();

Ls=MathFloor( 0.01 * Risk * AFM / (Lots * LC) ) * Lots;

If Risk = 0, then Ls = Lots; i.e. one lot.

That's all, further we substitute in the optimizer and go.

The Expert Advisor itself is not difficult for me to lay out, but it has various additional features, like statistics, information output on the screen, trailing bars, etc., up to the fact that the colour of lines of deals on the chart in my system. Then I have either to exclude all this from the code or attach all this pile, but I think I have described everything in such details, that it may take 20-40 min. of work if someone wants to repeat the experiment.

 
I see now. Thank you, ANG3110.
 
GODZILLA:
I don't take time formats less than an hour seriously, in general, and I'm racing experts on every conceivable currency chip, by

M15 is very suitable for a quick experiment. In the clockwise mode it is very long, especially for JMA, you may be waiting for a long time when you need to optimise, say, 30 - options, plus thresholds and risks. Hourly charting by opening prices - too crude, there will be a big error. M1 - M5 is also too long. That's why M15 is sort of an optimum in terms of speed and acceptable accuracy. We need to get approximate comparative information, not to make a working expert.
 
ANG3110:

I wrote the currency pair GBPUSD15 - so the period is naturally M15. If we are talking about period T3, then just use optimizer, I have T3 of my own making and it is slightly different from the one that is available on the Internet. Not only the period is optimized, but also the b-coefficient. Same with JMA, I don't know what variant you might have.

I was just wondering which T3 you used. From your work from 2005 or is it a more recent release? I didn't post it here without your permission and sent you an email with this question. Maybe I messed up the address?
 
ANG3110:

I wrote the currency pair GBPUSD15 - so the period is naturally M15. If we are talking about period T3, then just use optimizer, I have T3 of my own making and it is slightly different from the one that is available on the Internet. Not only the period is optimized, but also the b-coefficient. Same with JMA, I don't know what variant you might have.





It's absolutely unclear, what are we talking about? I think I was just misunderstood! The optimization period is not a chart period and not a muving period! I think everything is clear from the picture:



Most people here are experienced in this business and usually understand everything, but I will try to be more clear. We take an Expert Advisor, load it into the tester and optimize it, for example, for the period from 01.01.2006 to 31.03.2006. We obtain a long list of optimized parameters and profitability of the Expert Advisor. We select ten optimized parameters by any suitable criterion, fix virtual profitability of each variant and the virtual monthly deposit increment ratio in the table. After that we test the Expert Advisor with each of the optimized sets of parameters, but for a period from 01.04.2006 to 30.04.2006. Then we determine a real, not imaginary, benefit we have from these optimized parameters and fix in the table the real profitability and the real deposit increment ratio of each variant of ten. After this procedure is completed we move the optimization period one month forward (from 01.02.2006 to 34.04.2006) and repeat all the procedures described above again. In this way, we create a table for each month of 2006 and 2007. What to do with a table like this, I think no misunderstandings should arise! Anyway, such an approach to analysis of the Expert Advisor completely nullifies the desire to cause a massive stir in public consciousness in the forums and fool people around, because with this approach you very quickly come to a true understanding of the value of such things! But, of course, this is the only correct way and it requires much more work, and not everyone will like the sometimes rather murderous results of such research!
 
GODZILLA:

You've got the wrong place, the origin of the question is right here 'Author's Dialogue. Alexander Smirnov.', you have read it. In this thread we discussed ironing and lagging, various approximating and smoothing algorithms, and the professor asked the question with his indicator. I was simply asked a question by someone not yet very experienced in these matters. Now we are going in a different direction. Testing and optimization techniques, building debugging and working experts, statistics and so on are beyond the scope of this question. I was working on the analysis system and needed to add some smoothing algorithms into it that were more or less adequate and noise-proof. At this stage there was no aim to get a working Expert Advisor, because such methods are not suitable for real trading. I just made such a quick experiment. I was not interested in anything else at this stage. For myself I chose either T3 or LWMA, though during the further work on the analysis system it became less critical which algorithm to use - it depends on the market form, number of oscillations, ratio of forward and backward waves, the trend slope angle or its absence, presence of fast or slow harmonics, etc.
 
VBAG:
I was just wondering which T3 you used. Is it from your work from 2005 or do you have more recent releases? I didn't post it here without your permission and sent you an email with this question. Maybe I messed up the address?

I'll have a look in my mailbox, it's flooded with spam and I rarely look through it now. Yeah, there's a letter. I'll reply to it later.
 
ANG3110:
... The presence of fast or slow harmonics and so on and so forth.

If it is not difficult to explain it in details, how you isolate it, on the basis of what transformation, Fourier, Walsh, MME or other. I interesuet your estimate how many oscillations exist at the same time and how these oscillations were isolated. Thank you.
1...192021222324252627282930313233...44
New comment