a trading strategy based on Elliott Wave Theory - page 202
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
In this regard, I suggest discussing this topic, besides this part may become the basis or a good complement for the strategies under development. If you have any thoughts or who did such studies, please share.
PS: Great hope for Sergei(Neutron). Surely he has a couple, three interesting criteria.
Merry Christmas to all! And upward trends in all endeavours!!! :о))))
:) No, not in the sense that everything went in a cultural way, no werewolves were seen :)
It's just that my research has gone a little off the direction of the branch... So decided to take a closer look at Fibo try to score some statistics to see what's what... something like this... ...though it's raw. I don't know if I've done it, but I'm too lazy to look for it and look through other people's code, the indicator on which I'm writing this looks terrible, but the program understands what and where.
I meant, for example, statistically reliable criteria of trend detection. I.e. I suggest discussing alternative methods to technical analysis to detect it. And the issues you raise are just as important and interesting. But here we should probably go sequentially, first we find the trend, then we predict the lifetime.
PS: By the way, from page 90 to, I think, page 93, I "presented" my Hearst and it was on, I think, page 30. But, in fact, the first mention of it is dated page 4. :о)
As for the duration of the "structure", I practically learned (well it seems so to me, at least experiments confirm it so far) to determine by Hearst. Why I needed a "reasonable" trend is briefly stated, somewhere from 90 - 92 pp.
In this regard I suggest we discuss this topic, moreover this part can become a basis or a good complement for the strategies we develop. If you have any thoughts or who has done such research, please share.
PS: Great hope for Sergei(Neutron). Surely he has a couple, three interesting criteria.
Merry Christmas to all! And upward trends in all endeavours!!! :o))))
I join in the good cause of Grasn. Realising that we, as amateurs, are capable of turning this topic into a kitchen talk of housewives, I encourage time series specialists to participate in the attempt to apply the accumulated scientific knowledge to the forex currency market.
I join in the good cause of Grasn. Realising that we, as amateurs, are capable of turning this topic into a kitchen talk for housewives, I urge experts in the field of dynamic time series to participate in the attempt to apply the accumulated scientific knowledge to the Forex currency market.
Sounds like a call for help... :o))) joke. The help of professionals in dynamic rows would help, but that's just where to get them. Thanks for the support anyway. :о)
By the way Sergey, they say in America housewives "overturned" some stock exchange or some major player (I can't remember the details exactly, and it was a relatively long time ago).
Now I'm trying to "cook" a simple dish (let's consider it a frosting) on the basis of the commutative sum criterion. Somewhere in the depths of the Internet. The essence of the criterion is simple: the sum is calculated
V(i)=SUM{f(y(i) - median(y))}
where
f(i)=1 if (y(i) - median(y))>=0
f(i)=-1 if (y(i) - median(y))<0
The statistic is the number of transitions R through zero for a given confidence probability alpha. If R1(alpha)<R<R2(alpha) is satisfied, then the series is considered random. Here is a small example:
Original series
Parameter V(i)
In this case, the number of transitions for this number of samples, is near the R1 boundary but formally there is not enough 1 more transition for this sample to be considered random. Still experimenting...
Can anyone else share their research?
Grasn, can you be more specific about what "trend detection" means? An existing trend can be detected by a simple MA crossing, for example. The estimation of its stability and life time is another matter (i.e. how many points the price changes before it turns)... This is really a question of questions...
example, I meant statistically reliable criteria of trend detection. I.e. I suggest discussing alternative methods to technical analysis for its detection. And the issues you raise are also very important and interesting. But we should probably be consistent here, first we find the trend, then we predict the lifetime.
I do not think that in this matter even with a stretch, even with a very stretch, one can rely on MA crossing. It says NOTHING about the market phase. That is why all systems based on MA crossovers fail. In addition, 2 MAs are 2 parameters. This is an arbitrary behavior. And where there is arbitrariness, there can be neither science, nor objectivity.
Unfortunately, I don't know what alternative TA methods are. You don't mean FA, Sergey? :-)) If you don't mind explaining what you mean.
I, for example, can't imagine how it's possible to do it: "first we find the trend, then we predict the lifetime", if I don't know what it is at all. A directional price change within 5 minutes is a trend ? Within 5 days ?
So in order to go consistently, we first have to agree on what a trend is. I mean its formal definition.
I believe that the issue with a formal definition is a matter of principle. And not because without it we cannot identify trends. Each of us does it, though each of us does it differently. And most likely everyone does it by eye and not with the help of any criteria. But we have gathered here because everyone is trying to create an expert, i.e. a trading program. And the programme does not understand eloquence and gesticulation.
One more thing. I do not think that experts in the field of mathematical statistics, graphical analysis, dynamic systems, artificial intelligence, etc., etc. are less valuable in this matter than experts in the field of dynamic time series. However, you should not count on them. It is also a principle: in Forex one must rely only on oneself. No good man will come and bring you neither a ready-made expert, nor a strategy, nor even a theory. But a good idea can be born as a result of discussion here on the forum. And then everyone (who has understood it and appreciated it) can (independently) try to implement it in their programme.
I 'm not suggesting to squeeze anything out, I'm just curious, is there any positive result when using this Hurst method? Just curiosity? If so, great!!! :)))))))))))))))
Complementary, you say :0) Well, if so, then advise, some literature. Just for general information...
I'd be very grateful to you. :)))
I don't think you can rely on MA crossing even by a stretch, even a very stretch. It says NOTHING about the market phase at all. That is why ..........
Yuri, I'll start with a question, though it's not tactful, what is FA?
Totally agree with your opinion about MA, that's the reason why I decided to bring up this important topic as trend-finding by alternative methods. By alternative methods, I have so far modestly meant at least applied mathematical statistics, perhaps the professional areas you have listed would also provide a useful basis for ideas.
In terms of mathematical statistics (as I understand it), the concept of trend is based on some abstract strength of relationship between values of a series. And apparently all criteria reveal this relationship in one way or another. In my post on this page ("grasn 06.01.07 23:20") I just gave an example of such a criterion (I assume you were writing a reply at that time :o). The criterion in this case is the number of zero crossings. For a line, there would be no zero crossings at all. In other words, the number of zero crossings is an indirect estimate of the connectivity of the data.
I agree that the definition of a trend is very important, but I can't give it (I don't mean an intuitive definition, although you can start with them), for the simple reason that I don't know at the moment what a trend is. If there were a rigorous applied mathematical definition (not something like, trend is a directional movement of price, the presence of links between data, or the subsequent value must be greater than the preceding one, or similar), then finding it would be much easier, and there would be no need to discuss the issue.
You may be right that the notions of trend lifetime and trend itself should not be disconnected. In this case, the task would become significantly more difficult, as it would be necessary to search for a trend with such a lifetime. It is complicated. As I wrote earlier (I must be bored already :o), it is possible to identify the lifetime of a "structure" based on Hearst data and some additional parameters, but the Hearst indicator cannot find a trend, though it may be possible to provide it with such a valuable ability - if values different from 0.5 are considered a trend (how much different, plus "twitchiness" of the indicator). It seems to me more logical and easier to first find a trend based on some criteria, and then predict its duration.
PS: By the way, while I was writing, I had an idea of defining a trend (terminology... define means find): a series is divided into segments of equal length, all values on these segments are summed up and normalized. If there is a dependence of the following kind: each successive value is greater than the previous one or vice versa, this indicates the presence of a trend. It is more a search for a trend than a definition of it, though. :о)
Correction, I didn't mean exactly MA. The normalized sum is assigned to the middle of such a segment, without a sliding window, for example... :o)
Alex Niroba:
grasn I do not want to force anything, I am just curious, is there any positive result from using this Hearst method? Just curiosity? If so, great!!! :)))))))))))))))
Complementary, you say :0) Well, if so, then advise, some literature. Just for general information...
I'd be very grateful to you. :)))
Can't speak for everyone, but I have what I call the most positive results.
Regarding the addition of Hirst's wave theory, I cannot advise anything. It is so far my own ideas and my own research. If the results are successful, I will certainly write. I will write and if it is not at all. :о)
V(i)=SUM{f(y(i) - median(y))}
where
f(i)=1 if (y(i) - median(y))>=0
f(i)=-1 if (y(i) - median(y))<0
The statistic is the number of transitions R through zero for a given confidence probability alpha. If R1(alpha)<R<R2(alpha) is satisfied, then the series is considered random. Here is a small example:
In this case, the number of transitions for this number of samples, is near the R1 boundary but technically there is 1 more transition missing to declare this sample random. Still experimenting...
Can anyone else share their research?
Grasn, the expression you gave for the "comulative sum criterion" is, in fact, an inaccurate expression for the autocorrelation coefficient for the residuals series. Indeed, if the latter can be defined as the ratio of the number of co-directional price jumps to the total number of jumps in a selected section of the time series (defined by the TF) as:
r[i]=SUM{(Open[i+1+k]-Open[i+k])*(Open[i+k]-Open[i-1+k])}/SUM{|Open[i+1+k]-Open[i+k]*(Open[i+k]-Open[i-1+k]|}, where summing is done on window k=0...100 (for example).
We will get almost the same result. But autocorrelation coefficient is known to all, while "comulative sum criterion" is known only to a small circle of housewives :-) Properties of FAC are described in textbooks on matstatistics and properties of the "criterion of commutative sum" are not studied, but, of course, reduced to the latter. Therefore I call everyone to refrain from reinventing the wheel and use the mathematical apparatus that has been tested for centuries. Let us achieve OPTIMAL behaviour, not only in the market but also in research!
Back to our sheep.
I have some scattered material on time series from scientific conferences and interesting pieces of dissertations in this area approved by VAK. Unfortunately, I did not save the references to authors of works, so if I quote, I will do it without cuts.
When building models of long-run relationships one should take into account the presence or absence of a stochastic (nondeterministic) trend in the analyzed time series. In other words, it is necessary to decide whether each of the series under consideration is attributed to the class of series which are stationary as compared to the deterministic trend (or just stationary) - TS (trend stationary) series, or to the class of series which have a stochastic trend (possibly, together with the deterministic trend) and which lead to the stationary (or stationary as compared to the deterministic trend) series only by a one-time differentiation of series - DS (difference stationary) series.
The fundamental difference between these two classes of series is that in the case of TS series, deduction of the deterministic component from the series makes the series stationary, whereas in the case of DS series, deduction of the deterministic component leaves the series non-stationary due to the presence of a stochastic trend.
Note that the trend identification procedure is possible only in the case of deterministic trends in the series. This is an important point. Our task, therefore, is reduced to defining the class which the analyzed time series belongs to and only then to discussing the trend identification method. I have already done some preliminary work in this area and the result is negative: forex time series on the Forex market contain ONLY nondeterministic (stochastic) trends. There is NO way in nature to detect them in time. But maybe I was wrong :-)
Shall we work on it?