Application of mathematical analysis and higher mathematics - page 6

 
Drop me a link to Kelly, please. I'm as good as an elephant at lamb. Or is it from the people.
 
Catch ! http://konkop.narod.ru/Files/Kelly.zip
Very interesting...

Man... Sometimes I think we are tolerated just to keep some maths school in the country... :(
 
 
eugenk1 писал (а):
Heh... That's the definition of these very boundaries is the Trader's Philosopher's Stone or Grail... :) About the muva's, I'm just giving an example. I just think that the original system can be very, very simple, if it has an adaptation unit attached. But to make this block is very, very difficult. This is where strong mathematics is needed.

Dear Black Cat (sorry, I do not know your patronymic),
It is not that difficult to make an automatic system for adapting an EA to current market conditions. But you are unlikely to succeed. When I read this post of yours
eugenk1 wrote (a):
Turing's tests don't matter to me, I'm not interested at all in creating a mechanism that's as dumb as the average sucker. My task is more modest. To pick out some stable features from the market, which can be used to optimize the system quickly enough. It's easier said than done... About the ternary logic I actually said only in relation to the cited link. I liked the idea of price patterns too much. I strongly disliked timeframes and everything related to them from the very beginning of Forex. By the way, I wrote my first program on mql4 on this very topic. I've watched some interesting conversations about it on the spider, but I don't understand it yet. I don't have enough knowledge yet. I'd like something simpler, Yang-Mils fields, gauge invariance, superstrings etc... :))))))))))))


I wanted to advise you, as a physicist to a physicist, to use methods of continuum integration and quantum statistics. Things simple enough to suit your level. However, then I remembered another post
eugenk1 wrote (a):
dmitry, phoenix is too complicated, I've looked at it already. Interested in something very, very simple. For example on muwings or stochastic. I agree, the picture is very nice. You can clearly see what happened on the 13th... But it would be better to start such researches with something simple.

and realised that if Hendrick's advice doesn't suit you due to complexity, then my advice probably won't either.
Therefore, instead of advice, I want to ask a simple and direct question: are you going to do the auto-adaptation of the EA before the EA itself? I understood it that way. You have written a lot about this auto-adaptation, but nothing about the strategy it is supposed to adapt.

It seems that you have never optimized or even tested anything. I think so, because the script that tests the EA on history differs from the EA itself by 2-3 dozen operators. And if there is such a script, it takes only 10 minutes to turn it into a primitive optimizer. We do not need dll, C or even C++ for that. Of course, we may create a more complex optimizer that will use mathematical methods instead of the simplest search of variants. But is it really necessary? In order to confirm or disprove your idea about smoothness of changes in market characteristics it is enough to take the simplest system of 2 muwings that you mentioned 10 times already and optimize it on different, not overlapping market segments. The method of direct search is quite suitable here. 2 moving averages have only 2 parameters whose values change discretely in a limited range. Even 1000x1000= only 1 million passes. It will take you 1-2 hours, if we take a 1-month period of history, which fully corresponds to your desire to tune the EA according to the latest data and for a short period of time.

As a person who has 20 years of programming experience, this is a one-day task at most. But perhaps after completing it, you will lose some of your illusions (which is good) and stop speculating about what it should be like, and start working concretely in this direction (which is very good).

Because my words "you won't succeed" only mean that it makes no sense to talk in the abstract about the system's auto-adaptation and how complex it is, until you have the system itself. Agree, it looks funny when a person talks about how he will spend his money earned on Forex, if he doesn't even know how to make a transaction.

PS If you still find the 2 muwings system difficult, take a system based on 1.
 
Yrixx, I've really only just started working with the optimiser. Before, I was just putting something in at random and it somehow worked. Phoenix is complicated by the fact that it has quite a lot of parameters, and I'm still just getting to grips with the optimizer. It's with the search for a prototype system that I've now started. The simplest systems (2 mouve, MACD, stochastic) unfortunately have failed to be optimized at all. They either fail to optimize at all, even for a month. They either fail or have huge drawdowns. So everything is not so clear with the system itself. And finally, your huge post has only a couple of lines of meaning. If you have something to say on the case - I'll listen. If you're writing just to hone your eloquence on me - better leave this venture. This game can only be long-lasting when it's mutually interesting. I'm not interested at all. So excuse me.
 
1) When I hear about "volatile market" and "fit as you go" I am reminded of my own experience on neural networks. That would be fitting. I've checked, though I may have been wrong.

See, my estimates are as follows - the ability of the system to fit the story (let's denote it by A) A ~ exp{P}/N^Alpha. P - number of parameters, N - number of deals, Alpha - some degree. Of course, behavior of the profit f-fi by each parameter is important here, but generally speaking, by experience, the formula is correct on average. So optimization is a dangerous thing!

I am a supporter of complex fundamental conclusions + optimization of theoretically justified parameters.
 
kniff, what do you mean by "theoretically justified" ? Suppose there are 10 parameters in the system, all needed for some reason. Which ones are reasonable and which ones are not? Or should it be known to the author, who knows the system from A to Z? In that case, the author can still be wrong. Sometimes I'm completely lost, seeing results of running my own code in tester... I haven't worked with neural networks yet, I just don't understand them well yet. What are your results? Several people are praising them.
 
eugenk1 писал (а):
Phoenix is complicated by the fact that it has quite a lot of parameters, and I'm still only dealing with the optimizer. It is with the search for a prototype system that I have now started. The simplest systems (2 mouve, MACD, stochastic) unfortunately have failed to be optimized at all. They either fail to optimize at all, even for a month. They either fail or have huge drawdowns. So everything is not so clear with the system itself.


Exactly! The matter is not about the optimizer and not about optimization! A normal approach of a scientist is to go deep into the phenomenon, understand its essence, its nature. Then on this basis build a model. Having a model, you can study its adequacy, validity of its predictions, etc. If they are sufficient, it can be implemented in the form of an Expert Advisor, the parameters of which have to be optimized.

There are other approaches called "I'll try this way", but they are the way of amateurs. For those who have at least "some mathematical school" it's not serious. Therefore I propose not to get lost in thought, and to discuss if not the nature and regularities of the market, then at least the ideas on the basis of which its model can be built. Or models, if there are any. And as an axiom, I propose to assume that apart from the price data flow, there is nothing else, and all information is embedded and displayed in this flow. I propose to abstract away from volumes, because they do not express anything in Forex.

One more thing. Usually, people of our age, as well as people who are simply cultured, address everyone, except friends and relatives, as you. I suggest that this tradition should not be broken.

 
I just refer to everyone on the internet as "you". From the first day I started using it. Real life is different. OK, let's use the first name, if that's more comfortable for you. As for the difficulty of phoenix for me, I've been on forex for less than half a year and my understanding of technical indicators is still very vague. I've lost 4 months of a half year, being addicted to lots, grids, news and other "tsar's ways". That's why the most I can do now is a couple of muves and an oscillator. Otherwise I run the risk of creating a monster that I may not be able to handle on my own. What is important to me here is not the profitability, but a clear understanding. I have already run into instability, even with just two muvs. Regarding the price flow. I totally agree with you. As for the volumes - I don't know. The volumes sometimes look very funny. Look for example at EURUSD 2006.11.24 10:30 Moscow timeframe. By the way, when I was studying MT4, I used to watch how volumes were changing during the price movement. And I was opening if it seemed that there was a trend and the volumes were increasing as the price was moving in the direction of the trend. It seems to be working... So, it is debatable, but let's leave it for now for the sake of simplicity. Now the patterns. I understand you want to see some physics behind all this ? So far I believe in the crowd and tehanalysis (exactly I believe, I can't prove anything, but it sounds plausible). The bottom line is this. There is a crowd. And there are generally accepted (and even imposed on the crowd) methods of tehanalysis. All see charts approximately the same, with an accuracy to the noises. And they believe in the same thing. And therefore they perform approximately the same actions. For example, I haven't met anything more absurd than Elliott Waves and figure analysis. Nevertheless, it makes the crowd to act in approximately the same way. And it means the truth for the market. Of course, not all act in the same way. The same waves with figures can be interpreted very differently. Everything is stricter with the indicator values. But there is a certain randomness here as well. Therefore, complex order accumulations can occur on the market. Now if by chance or by intention of the game masters the price gets into one of such congestions, it causes a response. Multiple orders are triggered, the volumes grow (by the way, this is another reason why I consider the volumes important) and eventually the price either accelerates or slows down. The market model is spreading of disturbances in a highly non-linear and it is unknown how distributed the environment is. Unfortunately such a model gives little practical help. The properties of the medium are unknown. Moreover, it is not quite clear how they can be known. We can only guess where there are areas of strong non-linearity - order accumulations. For this purpose one should know classical tehanalysis, even if it seems to be an absolute rubbish. To predict the actions of the crowd, you need to know its beliefs. Unfortunately, this model does not explain at all and does not even try to explain temporary drifts, that is what interests me now. Alas, I have nothing better to offer. Can you suggest something else ?

By the way, here in forex, my 20 years of programming is worth almost nothing. Almost, because I'm still able to write code by myself and I'm not afraid of debugging it via Print(). It's a completely different problem to solve. Not the logical construction of a system solving a well-defined task, but fighting with the unknown. So hardly knowledge of C/C++, even a flawless one, can help in UNDERSTANDING a simple system containing 3-5 different indices.

Yes, Yrixx, I also wanted to ask you. It seems that you are a beginner here too. Have you worked out some methods of system decomposition? I now try to think of it (and do) as sets of enabling and disabling signals. For example, two mooves crossed. The signal for a deal. But stochastic is hovering somewhere in the middle. It seems to be easier to think this way. But images in the Strategy Tester sometimes confuse me. What's up with that?
 
Neurons make you a millionaire on history and drain your depo on reality.

They have the same problem as systems with many degrees of freedom (parameters). They "remember"
a specific time series rather than fundamental patterns. This is why everyone complains about the short "lifetime" of the systems.