a trading strategy based on Elliott Wave Theory - page 232
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
The author had the sole purpose of defending his dissertation. And it does not guarantee anything.
Sergei, you are wrong. Such a substitution is quite adequate.
We'll see who's right. :о)
Take your advice and derive the proof yourself. We would appreciate it.
Yesterday I did. So far it hasn't worked, in general, that's why I asked.
PS: I'm getting the impression that no profitable TC will work on the basis of this dissertation.
It works, but only under certain H. :)
...North Wind in one of his messages, mentioned that in this thesis there is no definition of arbitrability ( http://forum.fxclub.org/showthread.php?t=32942&page=9 18.12.2006, 10:46), i.e. the criterion according to which one can unambiguously determine: one can or cannot get the stable income from this instrument with the existing brokerage companies commission.
See the paper on p.64, Assertion 2.1.1.
Obviously the strategy is profitable if the right hand side of the inequality is greater than zero. By neglecting the last term in the right side of the inequality due to its smallness, we obtain the arbitrability condition:
|nt-2H|/Spread>1, where nt is the total length of the zig-zag (in points) related to the number of links (breaks), or the average length of a link. H - discreteness of partitioning (in points). Spread is the DC commission (in points).
For example, if nt-2H>0, we should use H+-strategy (open towards price movement), if nt-2H<0, we should use H-strategy (open against price movement).
All of the above is also true for the renko-build....
In general yes, on average each trade should yield more than
spread, but that's not what I meant. It would be interesting to know the confidence
boundaries for 2H. For normal distributions you can calculate such things,
but water for 2H is difficult. Although, there are reports that the same Student test
is good enough to calculate, numerically, but not methodologically.
It works, but only with certain H. :)
Do you, despite your lack of free time, still do it? Secretly. :-))
I also get that H-volatility is different from 2. But so far I've only calculated on a small amount of data.
And you, despite your lack of free time, still do it ? Secretly. :-))
I also get that H-volatility is different from 2. But so far I've only calculated on a small amount of data.
These are old results.
H-volatility, for "large" intervals usually tends to 2,
irrespective of H value, as it should be in theory. On "short" intervals,
as well as H-Hurst may show anything. Since the data is quite
"random" the result (calculation of H-volatility) is also "random".
The task, in principle, is stated by Pastukhov - to find "markets" with abnormal
H-volatility. Long term.
X[i+1]=X[i]+sigma, where sigma is a random variable that has a distribution identical to the generating series.
Thus, we have a Wiener process(VP) with zero arbitrability. According to the thesis, the value nt-2H, for such a series, must tend to zero. This is what we are going to check!
See the figure.
It shows the distribution functions(PDF) of the series EURUSD ticks 2006 and EP on the left. Integral values of FR for both distributions are 10^6 - exactly this amount of ticks was used for the simulation. A small discrepancy in the shape of FR, is associated with the inaccuracy of the selected coefficient in the sigma construction, responsible for the "width" of the FR wings for the EP. The absence of samples with zero amplitude, is due to the absence of adjacent ticks with the same amplitude, in the initial series.
To the right, correlograms for both rows are shown. Let me remind you that the correlogram shows the degree of correlation between
Y[i] and Y[i-k] of the original series (sample Y[i], or the first difference: Y[i]=X[i]-X[i-1]), where k runs from 1 to the desired value (in my case up to 100). As you would expect for EP, the correlation coefficient between any samples tends to zero. So the series is "correct".
The answer to the question "how close to zero should the obtained value be so that it can be considered zero" can be found in a textbook on matstatistics. As far as I remember, the value must lie in the corridor +-A*3/SQRT(n), where A is the modulus of the maximum value that takes our function (1), n is the number of samples, in our case 10^6. Thus, GP can be considered to be ACTUALLY a random walk if its correlogram will lie in the corridor of +-0.3%. This is indeed true (see fig.), we have a case of arbitrage-free market!
Not of little interest for an inquisitive mind, is a view of the correlogram for the series of ticks USD. We look. Draw conclusions (if you have a head)!
The EP series I use can be found here:
https://c.mql5.com/mql4/forum/2007/01/RNDusd_1.zip
To be continued.
It is beautiful!
It can be noted that the number of terms in the kagi series and the renko-constructions do not coincide. That's how it should be. Somewhere in the expanse of the dissertation, Pastukhov pointed out that the length of the kagi row will be greater than or equal to the length of the renko-row and proved it.
Next step, it is necessary to check correctness of the constructions. For this purpose, we construct distribution functions for lengths of sides of corresponding constructions. Obviously, we should not see lengths less than H=5. For the kagi constructions, the side lengths lie in the range from H=5 to infinity, in steps of 1 point. This is understandable, because an extremum can form at any time. For renko-constructions, side lengths lie in the range from H=5 to infinity in steps of H points. Which is also obvious, because the sides are formed only at multiples of H levels.
Let's see what we have got:
Everything is like in a drugstore! (Unless, of course, to the contrary) Integral on FR gives the number of members of series in corresponding constructions.
Now we can look at the behaviour of the value f(H)=nt-2H, for a Wiener process. We expect zero over the whole range of H values.
to grans
Sergey, pay attention to the picture of the Wiener process graph (the first figure in this post). It's proved that it is impossible to make profit on it in principle (the example is arbitrage-free), but the eye can see trends! Look, there are trends but it is impossible to earn!
To be continued.
Basically the same, but without the logarithmic scale.
In theory it should be,
2H appears in about 25% of cases. That's what it's all about.
Yay!!!
We see that the theory does not lie and the value f(H), for a random process, "hangs around" near zero in the whole range of the presented descratization values (1-30 points). The answer to the question "how close to zero should the obtained value be in order to be considered zero" will be given by the visual analysis of the obtained data. We built the arbitrage index f(H) for the non arbitrage market, it is clear that when analyzing the real market time series of the same length as the model one, we have the right to expect the greater index f(H).
Let's consider that f(H) statistically reliably signals of arbitrage if its value exceeds the corresponding range for the model nonarbitrage series with the same number of members.
No one but Northwind knows better what such fiddling is related to.
Now it's turn for the analysis of real time series...
Let's examine the series of EURUSD ticks.
On the vertical axis, the average return from one trade is plotted in points. The positive value indicates the necessity to open along the market, and the negative one - against it. It can be seen that in this case the greater profitability is provided by the renko-building. The reliability of the obtained result can be considered satisfactory. At the discreteness of 30 points the confidence boundary lies in the area of 2 points (see fig. above), while in fact we have a yield of 4 points per trade. As of today it is possible to operate on this instrument with 1 pip commission. Net profit for each transaction is 1-2 points in the area of 25-point rent decomposition.
The strategy showed us the possibility to get arbitrage profit with this instrument, 1-2 pips for every 25 pips of price movement. Price moves on the average 2-3 times per day, 200 working days per year (in MTS trading). So, we have 3*200*2=1200 points per year - optimistic variant, 2*200*1=400 points per year - pessimistic variant. All this on condition of stability of the criterion.
The question requires further study.
Results of construction on EURCHF.
Minimal spread for this pair is 2 points. Margin trading is possible with 1-2 pips profitability per trade, with a 15 pips rent breakdown. The 15-point range is passed by the instrument 4-5 times a day on the average. Thus, 4*200*2=800 points per year.
Construction results for EURGBP.
Minimal spread on this pair is 1-2 points. Margin trading is possible with 1-2 points profitability per a trade with 13 point spread. The tool passes the range of 13 points on the average 3-4 times a day. Thus, 3*200*2=600 points per year.
Also we can note the higher profitability of the rent decomposition. Maybe this situation is typical for all currency pairs?
We can note the divergence of results for the tick and minute series for 2006. The analysis shows that the difference is due to the dynamics of the respective constructions due to the ignoring of the intra-bar history. Consequently, one would expect different results for kagi and renko constructions on other TFs. The issue requires further investigation.
1. Let us consider the fact of satisfactory temporal stability of the return criterion to be proven.
2. Estimation of yield shows that it is in principle possible to obtain marginal profit in some currency instruments.
It's time to write an emulator of trades and make sure the results of obtained estimations correspond to the TS operation.