Machine learning in trading: theory, models, practice and algo-trading - page 2335
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
A good example would be 10 hours with an average of 10 minutes, which is systemic.
Why should the average position holding time necessarily be a characteristic of the system, and not just a statistic?
The explanation - the TS holds a position if certain conditions are met. The conditions are such that they often occur for a few minutes, but once a year they may hold for hours.
Will this explanation do?)
Why should the average time of holding a position necessarily be a characteristic of the system, and not just a statistic?
Explanation - the TS holds a position if certain conditions are met. The conditions are such that they often occur for a few minutes, but once a year they may hold for hours.
Will this explanation do?)
You are engaged more in finding a theoretical counter-example. Let's get closer to practice. On smartlab I made a more detailed post (out of righteous anger I won't publish the link).
So a non-systemic result is a result obtained outside the market pattern used by the TS.
It's stupid to consider the luck as a systematic one just because the algorithm traded.
You are more concerned with finding a theoretical counter-example. Let's get closer to practice. On smartlab I made a more detailed post (out of righteous anger I will not publish the link).
TC makes a profit if it exploits some market inefficiency. If we are not talking about the occasional profit, of course.
So a non-systemic result is a result obtained outside the market pattern used by the TS.
It's silly to consider luck as systematic only because the algorithm traded.
In this formulation, it is different. Here I agree, you are partly right. Partly because it comes down to a philosophical question about the impossibility of separating luck from regularity. And if it comes down to mathematics, it comes down to the question that we don't know the whole sample because most of it hasn't happened yet.
If a good poker player wins a big tournament once in two years (100%), and his average place is 65%, can we say that winning a tournament is luck, and not a pattern...
You can set a condition when marking up data for the teacher that if a trade hangs for more than 30-60 minutes (each task has its own, long term traders may need a week) - consider it unsuccessful. Thus, the algorithm will be adjusted for fast deals, which will suit your system.
There is no need to refuse from throwing out unsystematic deals, of course. The question is what to do with it on the real?
It is not necessary to refuse to throw out unsystematic trades, of course. Here the question is, what to do on the real with such?
In the real world to wait, if not thrown out of the training.
If they were thrown out, then close it after that time, so that everything was in accordance with the system. If the system did not learn on them, their success rate will be 50%, ie long term will be 0-th result from them. I.e. closing them will not bring any profit or loss, so we will not lose money, but the system will be stable with their random winning or losing.
It is not necessary to refuse to throw out unsystematic trades, of course. The question is, what should we do with such a thing in the real world?
In my opinion, the concept of "market inefficiency" should be specified somehow. Is it supposed to be always inherent in the market and to a constant degree? Or does it manifest itself occasionally and/or has a variable level? Can it be determined in some other way than simply by the behavior of equity?
Since the notion of "market inefficiency" is basic in your reasoning, only its formalization may result in the formalization of the algorithm for certain trades.
In my opinion, we should somehow specify the concept of "market inefficiency. Is it supposed to be always inherent in the market and to a constant degree? Or does it manifest itself sometimes and/or has a variable level? Can it be determined in some other way than simply by the behavior of equity?
Since the notion of "market inefficiency" is the basic one in your reasoning, then only its formalization may lead to the formalization of the algorithm for certain deals.
I suck at formalization.
I'm not very good at formalization.
Well, surely you have some internal idea, if it helps you to create TS.
Beautiful and convenient formalizations are good only for lectures and monographs, which you will hardly please us).
In my opinion, we should somehow specify the concept of "market inefficiency. Is it supposed to be always inherent in the market and to a constant degree? Or does it manifest itself sometimes and/or has a variable level? Can it be determined in some other way than simply by the behavior of equity?
Since the notion of "market inefficiency" is the basic one in your reasoning, then only its formalization may lead to the formalization of the algorithm for certain trades.
Googled what they write about "market inefficiency" - 99% of information is the usual sophistry, written to order from copywriters' exchanges.
but there is a more or less definite concept of "an efficient market" - in general they give the definition that "the price takes into account everything" and "the market price reflects the real value of the asset.
But if we go in the opposite direction - to add "not" prefix to "efficient market" and to receive a certain antonym using the definition "efficient market", then in my opinion we will not be able to receive a formalized description of "market inefficiency" - so it's another riddle with the formalization of the problem
If the average time of position holding for a TS is 10 minutes. And the current position hangs for 10 hours, then its result is a random (completely non-systematic)?
- Entirely systematic, provided your TS does not take position holding times into account, i.e. it uses some patterns, channels, returns to the average
- totally non-systemic, if your TS has strict limitations on the trading hours, ie it is tied to trading sessions or to the style of trade in relation to certain time ranges - intraday, short, medium term
The most interesting question is what to do? - imho, try to formalize such TS in a few TS - one TS with the maximum time to keep the position + TS with the remaining trades, then check on one TS and evaluate the statistics of trade, if separately does not work any TS, imho initially was either trade at chance Or this TS may work only as a composition of the TS - here everything is complicated, or the hung order in the market compensates for the new orders on the drawdown and / or the rate of profit - if the trade is a set of orders. Or the order has become a delay in terms of time for the next order - if trading on one order
All in all you can not rewrite here, but the problem is like in games with incomplete information - either you have an optimal strategy or you have mixed strategies which your algorithm selects in some probabilistic way.