The essence of optimisation - page 8

 
toxic:

Hmmm, I used to think so too, but lately I've been having doubts and the "truth" has been slipping away.

I don't want to produce too many "letters", I'll try an example.

For example: Random entry, TP and SL.

Let's optimize TP and SL.

For simplicity, we may fix the proportion ofTP / SL and optimize one parameter (their multiplier).

Does the result of such optimization make any logical or practical sense, and the main thing is WHY?

What does it depend on?

Random Entry - once wrote a code with a random entry (I ran it through the optimization of the optimum stop and take)

Stop was very short take was several times bigger

there is only 2 parameters optimized - stop and take - and everything comes down to a good MM - even lot

---

as a matter of fact - if the strategy has an entry and exit according to EA, the optimization is very useful and necessary

 
YuraZ:

is not a true statement...

For a start - if you look closely at the charts - even not particularly armed eyes can see repetitions

there are patterns that repeat themselves

If you stare at graphs for a long time, you can only see an optical illusion with the naked eye. A stick, a stick, a pickle and now one sees a recognisable image of a head with shoulders. There are repeatable patterns, there are not very repeatable ones, but the trick is that you can never tell which one of them works at the moment on the market with this approach to optimization. The second one is black swans - like black cockroaches, there are a lot of them and they are different, it is such a "dark matter" of the market universe, which occupies all the space, but you can not see it.

The most popular way of optimization nowadays - it's in the nearest interval of history. So if I want, for example, not to optimize January-February-March, but, say, March2011-March2012-March2013 with a forward to March2014, I simply cannot do it with MT. That's how a drunk looks for lost keys under a streetlight, as you can't see anything elsewhere. The most popular way to predict - a simple extrapolation, they say, if it worked yesterday, it will work tomorrow. Like inertia rules.

But there's more.

-If it hasn't happened for a long time, it's more and more likely to happen again,

-if it's never happened, why wouldn't it happen?)

If I show you any strong spike on a chart, taken at random-you will never be able to tell where it came from-the Fed's decision, a nuclear plant accident, or a computer glitch. But I don't see brokers actively offering a story reflecting at least regular news (if there is such a thing at all). Because as long as the myth "all the necessary information is in the price" exists - their income will not become scarce.

Often one hears - TS has ceased to be profitable, because "the market has changed". And what exactly has changed? What are the characteristics of the market? The volatility? How to calculate it in figures, not by eye? Has the market changed only on this instrument or has it changed globally? And what was the M1 aggregate? Etc. etc. Real optimization cannot avoid these questions, and people cannot talk enough about stop-losses.

 

Paragormon: 

The most true pattern I know about the market - it fluctuates....


Well said! Respect.

the fact that the market is fluctuating is the real law (and not a mirage and ghosts that abound in the market).

i would add to the real consistent patterns that the market volatility also fluctuates

this leads to the third pattern: trends, big and small, are formed in the market accidentally/not accidentally.

we have to rely on these regularities when choosing a strategy.

 
toxic:

OK, I'm half-joking about the ratings too, I don't really care.

In general, one of the ideas is that we analyze "window testing" chart, "walk forward" as it is also called, but in this case it is "walk backward", i.e. for example, having a million bars of the tested symbol we increase it by 100 000, shift it by 10 bars and watch dynamics of parameters.


It appears to be something similar:


You can clearly see the regularity of the "migration" of extrema in the process of window shift, this regularity being quite "slow" in the sense of "inertial", i.e. predictable for some strategies and quite random for others.

Dynamics of extremum shifts for "qualitative" types of strategies I have figured out how to approximate and predict ahead, though frankly speaking it is not as trivial as it may seem at first sight.

If this kind of research makes sense to someone, I can continue thinking, if not, I can't.

It seems to me that we should not shift by bars as it is not clear what we obtain, for example, we have tested for 100 000 and before the end a position was opened and immediately closed not according to the algorithm implemented in the EA but because the testing had been completed.

During the next testing shifted by 10 bars, we actually get the dynamics of price movement depending on the open position.

A similar mess happens at the beginning.

It would be logical to shift by fully processed trades.

For normal testing, obviously, it does not play the decisive role, because a couple of incorrect deals (not closed according to the algorithm given in the Expert Advisor) will not spoil the picture, provided the number of deals is large enough or the testing period is long.

 

Apparently, it really is a fake.

Generally, from the beginning the task was to make a self-adaptive system implementing the "dummy theorem".

That is, by means of sliding optimization for each window we obtain coefficients for the dummies, and then by smoothing and approximating the obtained series of coefficients into the future (the nearest one) we make profitable dummies forever.

But it does not work. Besides, it is not easy to implement.

But I'm not responsible for my words yet, I wouldn't swear by blood, because I'm correcting logical mistakes again and again. Maybe this time it's a mistake too.

 
toxic: Maybe this time it's a mistake too.
Nah, it's usually the other way around :) if it shows a profit, you have to look for an error.
 
toxic:

Apparently, it really is a fake.

Generally, from the beginning the task was to make a self-adaptive system implementing the "dummy theorem".

That is, by means of sliding optimization for each window we obtain coefficients for the dummies, and then by smoothing and approximating the obtained series of coefficients into the future (the nearest one) we make profitable dummies forever.

But it does not work. Moreover, it is not easy to implement.

But I'm not responsible for my words yet, I wouldn't swear by blood, because I'm correcting logical mistakes again and again. Maybe this time it's all wrong too.

What's the mash theorem?


I have no idea what to do further on. If you may give me a link to read about it. What the sliding optimization is and how it should be used in the Expert Advisor.

If I'm talking about walking forward, how should I use it in my EA?

 
toxic:

Apparently it really is a fake.

It happens.

You should at least show us what is fake and what is not.

papaklass:

To be sure whether there is an error or not, you need to post the results of your research. The collective intelligence will find the error, if there is one, much faster than searching alone.

Unfortunately, this is not customary among traders. Therefore, we 'stew' in our own juice. :)

+1

perepel:

Whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa, whoa!

It's the fuckin shit!

Look it up on spider if you like nonsensical reads about "can you trade SBs" and such.

It's about how at any given time you can get the most out of a price with mashmas if you know the right parameters, which you can only find out in hindsight.

 
nowi:



I would add to the real laws that the market volatility also fluctuates.

This leads to the third pattern: trends, big and small, are formed in the market accidentally/not accidentally.

we have to rely on these regularities when choosing a strategy.

The topic is actually not about trends, but about the essence of optimisation. For some reason, it seems to be taken for granted here that the task of optimization is to catch these very trends.

My idea was exactly that the essence of optimization, as most people see it, is adaptation of variables to continuous and more or less recent history and the only question is how deep we go into this history. At the same time we can eat up this history in small portions or, on the contrary, consume it for decades - one way or another everyone proceeds from the postulate of market inertia. My point is that no one talks about other properties of the market and about optimisation according to some other principles.

 
Paragormon:

The topic is actually not about trends, but about the essence of optimisation. For some reason, it seems to be taken for granted here that the task of optimization is to catch these very trends.

My idea was exactly that the essence of optimization, as most people see it, is adaptation of variables to continuous and more or less recent history and the only question is how deep we go into this history. At the same time we can eat this history in small portions or, on the contrary, consume it for decades - one way or another everyone proceeds from the postulate of market inertia. My point is that no one talks about other properties of the market and about optimisation according to other principles.

what do you mean "nobody"?
What did I write about? :)