a trading strategy based on Elliott Wave Theory - page 270
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
500 counting on the chart corresponds to 09.01.2004 (19:00), clockwise. The data is either Alpari or Metakvotes. Can't remember anymore.... Hearst is pretty robust to different DCs.
This is probably the most important thing. All that's left to do is use it. I haven't tried it yet.
500 отсчет на графике соответствует 09.01.2004 (19:00), часы стало быть. Данные или альпари или метаквотес. Не помню уже…. Херст довольно устойчивый к разным ДЦ.
This is probably the most important thing. The only thing left to do is to use it. I haven't tried it yet.
I have a slightly different opinion. We should not talk about Hearst's absolute stability, but only about relative stability. That is, its robustness to work within a particular strategy. For example I gave results last year when depending on different data of different brokerage companies I got results that have serious quantitative differences for my strategy. But grasn said he is using digital filtering of this indicator and in his strategy (estimation methodology) the use of Hearst indicator gives stable results. He may be right. Excessive noise is better filtered even in Hearst Ratio. Although in this case additional system parameters appear (for example, filter cutoff frequencies), which is an inevitable payment for "comfort" of getting negligibly small difference of Hearst figure in different DC, but again within the limits of a particular strategy.
(0<H<0.5) High probability that the "movement" will change its direction
(0.5<H<1) High probability that the "movement" will keep its direction
You can see that channels up to the 300th datum have values much lower than 0.5 and therefore the "movement" is more likely to change its direction.
Here, in one case, it talks about the direction of "movement", and in the other just about "movement" (and both times in inverted commas). I would like to understand what is meant by "movement". Because it is just the opposite with the trend direction in this picture - after a small Hurst the trend seems to get stronger, but after a large one it breaks
Alas, the calculation should be done for 100% of cases :). For 80% of cases the prediction should be correct. But how to select these cases is a separate question. Each of the selected cases will be an entry into the position.
That's right, Hurst is neutral to data of different brokerage companies. Checked.
To Solandr
I have a slightly different opinion. We should not talk about the absolute stability of Hearst, but only about the relative stability. That is, its robustness to work within a particular strategy. For example I've posted results last year where depending on different data of different brokerage companies I got results that have serious quantitative differences for my strategy.
But grasn said that he applies digital filtering of this indicator and in his strategy (estimation methodology) the use of Hearst indicator gives stable results. He may be right. Excessive noise is better filtered out even in the Hearst Ratio itself. Although in this case additional system parameters (e.g. filter cutoff frequencies) will appear which is an inevitable payment for "comfort" of getting negligibly small difference of Hearst figure in different DC, but again within the limits of a particular strategy.
We must speak only about the absolute stability. And the strategies have nothing to do with it. At least I do so when I test a component separately. If Hirst should show the probable movement change by H value, he must do so without any strategies using data of different brokerage companies.
Indeed, Hirst's curve turns out to be somewhat noisy (it shouldn't be) and requires further processing. To get filtering parameters (and not take them from the ceiling) we need to "ask" them from the data itself and they will suggest it (frequencies will remain frequencies for different datums). And it's not about filtering, it's about the calculation algorithm, that's the main thing. The methods described in books are not quite correct here. I.e. for example Peters virtually rewrites the methodology of the founder, and it (my humble opinion, may be wrong, although I have already made conclusions for myself, confirmed by studies) simply is not applicable to forex series (compare the so-called "inflow" and series of quotes, their properties are different). The valuation turns out to be highly biased. And after reading the authorities many traders go this way...
And these articles are different. In some of them he is American, in others - Englishman, or Egyptian. Some people even manage to write that (almost verbatim) "Hirst didn't understand what he'd discovered till the end of his life. Man, I'm tempted to write my own article, but I can't write. Just for the record, Hirst published several books, one of which is called "The Nile" and was translated into Russian in 1954. The main thing is not even in the algorithm, but in the phenomenon that Hirst discovered and which he perfectly understood until the end of his days.
to Candid
Here in one case it talks about the direction of "movement" and in the other just about "movement" (and both times in inverted commas). I would like to understand what is meant by "movement". Because the opposite is true for the trend direction in this picture - after a small Hirst the trend seems to get stronger, but after a large one it breaks
.
Let's go over it again. Firstly, everything is correct and secondly, it is simple. We have the initial series, we need to determine the general dynamics of the movement.
Calculated Hirst:
I have already written, I chose the linear regression channel as an estimate of "movement" (in principle an estimate (dependence in the form of a formula) of the direction of movement can be chosen, but I have not managed to calculate Hirst for anything yet). In other words, I estimate the coefficient b in the formula y=a+b*x.
Let's distinguish two channels.
(1) The longest one (500 samples) having the minimum Hirst value of 0.34
(2) The channel starting from 400 samples, which has the maximum Hirst value of 0.62.
These are the channels:
Let's estimate the movement by eye. A channel of length 500 should go bust, while channels after 300 counts may still pretend to live in them (it means that the price will hang in there for some time, don't forget about probabilistic nature of the index, it's statistics). If we mentally (without having a picture ready) prolong channel borders (that is, roughly speaking, range) we may notice that for some time the area of the current price (if we trust the range and b) will belong to both channels. Thus, we can conclude that the channel with length of 500 samples will be "alive" for some time, the price will go out of it in the direction of the small channel's b-coefficient.
Let's make one more check. Let's take a closer look at the picture below. The dead zone of 100 samples is not chosen randomly, it is calculated. I just want to note that if you take a smaller value, the already calculated Hurst will not change. In this case, I'm interested in "macro movement". You can see how the reversal is going and you can see where the price should go.
Now let's see what happened in the future:
The long channel "collapsed", the short one has also collapsed, but a bit later. So, what's so special here? Why did the channel of 0.6 according to Hirst collapse? Not much in particular, except for the fact that quotes are multifractals and therefore there is a dependence of Hirst on time (what has been shown is the dependence on the number of samples). And also the lifetime of the channel from Hearst, the spread and length of the channel and some other things. But what is good about Hearst is that it gives (regardless of the strategy) a good signal that the system will rebuild. And how it will do that is a separate conversation.
The above described has also "energy" confirmation, the ripple model is a good proof of that. The point of my strategy is not to build a bunch of channels at all. You don't need a whole bunch of them. What is needed is one, the most reliable channel and calculation of pulsation structures of this channel. Maybe I will tell you about it sometime.
I tried to estimate the trend in different ways, and even made some tables using different methods of technical analysis and got a result similar to the one described by the North Wind with flipping a coin (by the way, he is nowhere to write articles for). Hurst has real prognosticating qualities, unlike all other methods. I made similar experiments with all statistical methods of trend assessment. Noise is noise in Africa, even if it is sometimes directional. I hope to guide you on the right path with my sloppy presentation of the material. :o)))
Alas, it should be calculated for 100% of cases :). For 80% of cases the forecast should be correct. But how to select these cases is a separate question. Each of the selected cases will be an entrance into the position.
Oh, that's what you mean. I see what you mean. That's what I'm doing now.
When I was speaking about relative stability of Hearst coefficient for a specific strategy I meant the transition zone equal to 0.5 only. This zone is a watershed for price movement states. In my strategy, for example, it was the basis for trade decisions. And exactly because different brokerage companies may have values close to 0.5 at one and the same time, but "only" from different sides, there was a significant difference (several times!) in the final balance. I do not know how exactly the decision to enter and maintain a position is made in your strategy, but even with the use of filtration only, the allowable difference between the Hearst values in different brokerage companies can only be reduced, but not eliminated entirely, since we formally have different initial data and one calculation algorithm. Though, decrease of differences alone should increase stability of results for different DCs.
Of course, if we consider areas far from a watershed, for example 0.8, then it makes absolutely no difference that the difference between values of different DCs will differ by 0,001, because the information will be absolutely the same from the point of view of my strategy, for example. And from this point of view one can consider Hurst indicator to be absolutely stable, because it provides absolutely the same information.
One question: I understand that you are using a sliding window for Hearst indicator, i.e. Hearst as an indicator in real time mode?
PS: and perhaps a formal different calculation algorithm
My strategy is slightly different:
1. determine the sample that is the minimum in terms of the strength of the relationship between the elements, calculated on the basis of the correlation criterion. We get the total sample on which it makes sense to look for something.
2. a matrix is generated for dynamic analysis.
3. one, the most reliable channel is identified.
4. The structure pulsation is calculated for it (working title).
5. the channel rhythm is determined and the future swing, drifts and finally the reversal zones are calculated by analogy with Murray levels (just by analogy, not exactly by it)
6. trading decisions are made
7. the beginning of the channel is fixed and all basic parameters are monitored later (something like quality control)
PS: Hearst types - is it if more than 0.5 or less than 0.5? Hmmm, in general 0.5 is not a good value for decision making. Noise however.
In general, I assumed that "movement" meant following a channel, but they were not drawn, so I had to go for a little provocation :). The other purpose was to demonstrate the subjectivity of picture perception. As for Hearst, in my realization of trading by linear regression channels I collected statistics (about 5.5 years) by various indicators (about 40 as a result), including Hearst (though calculation methods may not coincide). Among this heap of indicators there is not a single one which allows to judge more or less confidently whether the channel will survive or break up.
And in your example there is no dependence on the Hearst value after all. True, as I understood you are actually focusing on its dynamics.
By the way, this picture can be interpreted in different ways. First, from my experience I've come to a conclusion that the appearance of a new channel of the same direction but less steep may be considered as a kind of a pattern. As a rule, it means the end of a trend soon. Secondly, it is nothing else but a well-known effect of a weak break before a correction :)
By the way, this picture can be interpreted in different ways. First, from my experience I came to the conclusion that the appearance of a new channel of the same direction but less steep may be considered as a kind of a pattern. As a rule, it means the end of a trend soon. Secondly, it is nothing else but a well-known effect of a weak break before a correction :)
I purposely chose a difficult area that has a rather steep reversal. I deliberately did not reduce the blind spot, as if "took away the turbulence at the stern" and deliberately moved away from the turn itself. Why did I do that? Now let's look really objectively again:
Long Channel, the "trend" has a Hurst value of 0.3. Did it survive? - No, it hasn't, it's a medical fact. The small channel has a value of 0.6. It survived - no, it didn't. And why is that? Precisely for the simple reason that a reversal has begun. In the current structure, before the red vertical line, there is no new movement, look closely (I chose that moment on purpose, no LR will show it in the right direction). In other words, at the start of the reversal, there are simply no long-lived linear regression channels yet and there just won't be any for a while!!!! That's obvious if you're building a LR channel and not a parabola or something else. And there won't be until a new movement, a new channel is born. Another thing is that we cannot yet say with 100% certainty what it is, a reversal or a sideways movement or something else. But Hearst will point in the right direction for some time and say for sure that it is not worth orienting to the long channel.
And now let's "live" with the eurusd a little bit more and measure the Hurst. Let's fix the beginning of the old long channel and calculate the new Hearst values in 100 counts, just where the exit outside the old long channel occurs.
Old calculation:
New calculation. Another 100 new samples are added to the sample. Note the overall shape of the curve. It is just to the point of Hearst stability (transition through 0.5 has remained approximately at the old place considering new 100 samples), it still shows that long channels are undesirable.
And here are the new, currently stable channels, with counts 427 and 485 according to the new measurement (Hearst maximums):
Don't forget that this is (H+L)/2.
You can look further:
Dynamic analysis is used, but not as described. It's a little more complicated than that.
Apparently I'm not as technically savvy and don't know what overhang or overlap are. Or rather, I don't know them both. And what I have, I hardly think it's a correction (I've given a picture of the calculation place).
Anyway, if you don't like Hirst, don't use it. Sorry for wasting my time. :o(
But then at least share your achievements in trend detection.