![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
blinking and working on the last bar is not an error of MT4, it is an error in the logic of the algorithm! MT4 will fairly give you Close[0], but you understand that at the beginning of the bar it is almost the same as Open[0], but at the actual close it will be completely different. And if your system makes a trading decision taking Close[0] into account, it means that this decision will change with every new tick :(
This is not an error, this is a feature of the platform. Some do not give Close until it is really formed as Close of the used frame. And they plot indicators only on formed bars.
.... And if your system decides to trade with Close[0] then with every new tick this decision will change :(
Why so sad? :)))
Because it only shows that in this case your current system does not work, not about all of possible systems.
About the importance of systematic strategies and the lack of importance of the forward test, I almost completely agree with faa1947.
Thank you, at least someone agrees. The list at the beginning of the post is mostly a lot of soothing: a scratch - apply green, a bruise - apply iodine. When you write "stay away from fights" it's too general. Besides my skepticism concerning the forward testing I would like to once again draw the attention of forum visitors to the fact that a proper TS should take into account BP characteristics - non-stationarity in the first place, not testing on the rise, fall and flatness - triviality.
I am attaching the EURUSD spectra for two consecutive periods. My question to the author: what "grains of concrete knowledge" may help to create the TS for the period 12.08 - 12.09, and then to use it for the next period?
Thank you, at least someone agreed. The list at the beginning of the post is mostly a lot of soothing: a scratch - apply green, a bruise - apply iodine. When you write "stay away from fights" it's too general. Apart from my skepticism concerning the forward testing, I again want to draw the attention of forum visitors to the fact that a proper TS should take into account BP characteristics - non-stationarity in the first place, not testing on the rise, fall and sidewall - triviality.
I am attaching the EURUSD spectra for two consecutive periods. The question to the author: what "grains of concrete knowledge" he collected, will help to create the TS for the period 12.08 - 12.09, and then to use it for the next period?
Such recommendations may be correct, but they are not universal. Everything depends on the trading methods, symbols and timeframes used. Non-stationarity is not a property of a series, but the property of the method of its observation. Of course, the non-stationarity of some observations of EURUSD does not mean that all systems have implemented this non-stationarity in their results.
The hypothesis of patterns and the use of neural networks to identify (and search for) them in relation to a time series of market quotes is comparable in its effectiveness to doing the same on random data. As an example, apply the same method to predict the digits in the notation of Pi. Neural network based strategies are a form of fitting (adherents of NS call it learning) to a story with the ability to fit on the fly - auto-optimize (adapt). The haphazard approach is a black box approach: "I'll make it up and see if it works". A lot comes from rumours about the use of neural networks by the strategies of Simons' most profitable hedge funds. But in reality there are no neural nets there.
Simons' black box is trivial statistical arbitrage (spread trading) that uses huge computing power and input stream. Simons was originally at the forefront of applying market inefficiency calculations, which is why he is still the leader in billion-dollar arbitrage. His use of exclusively highly liquid trading instruments is dictated by the need for this condition for guaranteed arbitrage.
Arbitrage is a systematic approach. Using patterns in "noise" is a systematic approach...
I don't understand what arbitrage is at our level. Arbitrage does not refer to trading at all, it seems to me. BP is not noise in the sense of normal law. Systematicity starts with the fact that we take into account the characteristics of BP, the basic non-stationarity. I attach the spectra of two consecutive months as proof of non-stationarity. If we follow Hirst, BP is in three states: noise (0.5), trend (>0.5), and unsteady state (<0.5). This is only the beginning of systematicity, but not systematicity. Then, bit by bit, we start to introduce systematicity. For example, we select the orthogonal indicators (one cannot be obtained from the other), such as the trend indicator and the volatility indicator. We apply some mathematical methods, e.g. we start to use spectrum analysis to identify the trends, etc. The positive result of such design is always the same: the TS is able to identify the beginning and the end of the trend, a good system is also able to detect a sideways trend. Unfortunately, the creation of TS is an art. You cannot teach how to create TS. But you can set some requirements which will help you to create your TS.
The topic of this thread is interesting, but it constantly slips to the level of znakharstvo - "it is not good to create TS". Let someone make a system to the attached charts without using adaptation.
Such recommendations may be correct, but they are not universal. Everything depends on the trading methods, instruments and timeframes used. And non-stationarity is not a property of the series, but a property of the way it is observed. Of course, the non-stationarity of some observations on EURUSD does NOT mean that all systems contain this non-stationarity in their results.
The spectra I have provided are a model of the original time series and of course there are discrepancies between the model and the original series. But this model states that BP is non-stationary. And we all know that the distances between two market ups and downs change all the time on all timeframes. I think the model given is much closer to any other models that try to convert non-stationary BP to stationary, or just forget about it when reasoning about the fit.
The list at the beginning of my post is mostly soothing: scratch - put green, bruise - put iodine. >> this is specific.
that's exactly why i started this compilation! if there's something wrong with the system, put a band-aid on it.
Question to the discoverer: which of his "gleanings of concrete knowledge" will help to create a TS for the period 12.08 - 12.09, and then use it for the next period?
Why do you always want to see the answers to your questions in my thread? :))))
It's not gonna help you build a new system!!!!!!
You can use it to check if your system has any bugs that others have already stepped on. This is a list of the pitfalls, not a recipe for how to avoid them or build a bug free system!
I don't understand what arbitrage is at our level. Arbitrage does not refer to trading at all, it seems to me. BP is not noise in the sense of normal law. Systematicity starts with the fact that we take into account the characteristics of BP, the basic non-stationarity. I attach the spectra of two consecutive months as proof of non-stationarity. If we follow Hirst, BP is in three states: noise (0.5), trend (>0.5), and unsteady state (<0.5). This is just the beginning of systematicity, but not systematicity. Next, bit by bit, we start to introduce systematicity. For example, picking up Orthogonal indicators (it is impossible to obtain one from the other), for example the trend indicator and the volatility indicator. We apply some mathematical methods, for example, we start to use spectral analysis to identify the trends, etc. The positive result of such design is always the same: the TS is able to identify the beginning and the end of the trend, a good system is also able to identify the sideways movement. Unfortunately, the creation of TS is an art. You cannot teach how to create TS. But it is possible to set some requirements, which will help to create TS.
The topic of this thread is interesting, but constantly slips to the level of znakharstvo - "it's not good to tailor the TS". Let someone make a system to the attached charts without using adaptation.
From my point of view this is a haphazard approach: "what if something works". Again, you might as well predict the numbers in the Pi entry.
With such haphazard approaches patterns will not be born. It is possible to make a profit without the luck factor. Because you can predict equity (not price) when using a systematic approach. The haphazard approach is a chance.
P.S. About the mandatory use of orthogonal indicators - I agree. This is the reason why indicators should not be a derivative of price, as price itself is an indicator.
The spectra I present are a model of the original time series and, of course, there are discrepancies between the model and the original series. But this model states that BP is non-stationary. And we all know that distances between two market ups and downs change all the time on all timeframes. I think the model given is much closer to any other models that try to convert non-stationary BP to stationary, or just forget about it when reasoning about the fit.
I can't open your file properly, but I will repeat that non-stationarity is relative to some system of observations. For example, successive observations at equal intervals of time. The resulting series is non-stationary. This does not mean that any observations of the original process will be a non-stationary series. In fact, any TC is one of the ways to observe/convert the initial nonstationary series. And the new series can be stationary. Moreover, this is one of the main tasks of systems engineering - to get a series of stationary returns.
Therefore, I agree that a continuous and sufficiently long series of discrete-time price observations is non-stationary. But that does not mean that prices are non-stationary in principle. To paraphrase: "It is our task to extract non-random profits from a random process!" random to stationary.
And how will you look for moments when a price series behaves quite stationary - this problem has no universal solution.
I can't open your file properly, but again, non-stationarity is relative to some system of observations. For example, consecutive observations at equal time intervals. The resulting series is non-stationary. This does not mean that any observations of the original process will be a non-stationary series. In fact, any TC is one of the ways to observe/convert the initial nonstationary series. And the new series can be stationary. Moreover, this is one of the main tasks of systems engineering - to get a series of stationary returns.
Therefore, I agree that a continuous and sufficiently long series of discrete-time price observations is non-stationary. But that does not mean that prices are non-stationary in principle. To paraphrase: "It is our task to extract non-random profits from a random process!" random to stationary.
And how will you find the moments when the price series behaves quite stationary - this task has no universal solution.
I have included the file in Word 2003 into the archive. Maybe because of this it could not be opened.
However, let us distinguish between the source TP - the one that goes into the terminal and its modification (processing). I discuss the initial BP and argue that it is the one that has the property of non-stationarity. The proof of non-stationarity is already some models of this initial BP. Any TS deals with this initial BP, transforms it into another form (for example, into a wrecker) and decisions are made on this transformed BP. The problem is in the lifetime of this transformation: everything is ok for history data, but the market has changed and in reality we have another BP and apply to it a transformation that was good in the past.