a trading strategy based on Elliott Wave Theory - page 238
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
https://c.mql5.com/mql4/forum/2007/01/RNDUSD1m.zip
And please print the value which can be compared directly to spread, i.e.
nt-2H - it will make your results look more palatable. We expect identical zero over the whole range of H values.
And please print the value which can be compared directly to the spread, i.e.
nt-2H - so the result will look more digestible. We expect identical zero over the whole range of H values.
OK, will do.
Sergey, you're setting tasks faster than I have time to do them. :-))
I hope this zip is different from the one posted for comparison with EURUSD 2006 ticks.
Because for that zip I did it in parallel with ticks. I got the expected result.
I don't know if zero accuracy is satisfactory, but in the whole range Hvol fluctuates around 2.0
In absolute value it turned out sko=0.0132, in relative = 0.657%.
PS.
Had a look at this file. Different, of course, from those ticks. But, alas, not much.
This data is not suitable. The candlestick is represented via OHLC. Out of those 4 values, to construct
to build a zigzag is required 2: High and Low. Construction of a zigzag (I suppose any zigzag) by Open or Close has
The construction of a zigzag (I assume, any Open or Close) is meaningful only as a programming exercise.
To be able to compare results in an interesting and meaningful way, I think we should
minutiae to build from that 1 million ticks Wiener process that was posted earlier. After all
the minutes of the real process have a direct link to the ticks. We need to maintain that relationship
for the Wiener model as well. And it is desirable that the number of ticks fit into
in one candle should be distributed in the same way as for real ticks.
Can you do it ?
As a result (because there are twice as many real ticks as model ones), it will give minutes for half a year. But the structure quality of model bars will correspond to the real process much better. It seems to me that such a toughening of conditions of the experiment will make its results even weightier.
One more thing, Sergei,
Stay awake!!!
Stay awake!!!
:_)) Good!
I've already started preparing adequate minute bars for the arbitrage-free case.
The first thing I did was to adjust the Wiener ticks used to construct the minute bars. Now they are two millions and their distribution function (DP) better fits the distribution of first difference amplitudes of EURUSD 2006 tick array (see fig. (DP for a Wiener process has been forcibly multiplied by 0.5, because now there are 2 times more VP ticks than for EURUSD 2006)
An archive with updated Wiener ticks is available here:
http://www.filefactory.com/file/050ece/
Let's build EURUSD 2006 tick array FR by the number of ticks in one minute bar and cut the tick series into pieces with the same length distribution law.
The picture below shows the FR of the number of ticks in a 1 min. bar of EURUSD 2006 and the FR of the number of ticks in a 1 min. bar of the EP series.
We may state a satisfactory coincidence of distributions.
Now, let's select segments with the length of at least 1 minute in the EP tick series and form one-minute bars. The first and the last member of this segment will correspond to opening and closing prices (Open & Close), and the maximum and minimum values of the segment - High & Low values of the minute bar. For the resulting series of EP mini we will plot FRs based on open prices and compare the distribution of amplitudes with the same for EURUSD series in 2006, see fig.
The apparent difference between FRs can be explained by the presence of a noticeable negative correlation between samples in a series of EURUSD ticks that leads to the effect of negative feedback, i.e. the time series "tries" to keep the current price value to a greater extent than the EP series. Something can be written off as not strictly matching the FR for ticks see above.
You can take the arbitrage-free series of minutes which is generated from the EP ticks here:
https://c.mql5.com/mql4/forum/2007/01/USDRND1min.zip
The file format is as follows: Open High Low Close
Alexey, if you do not mind, share your thoughts (not predictions, not assertions): what is happening now with the pound according to Elliot's wave theory or its follower? Is a repeat of the 1.99 level possible? And in general, is an upward rush possible. I remember that predictions are a thankless business, but still, it's opinion, not assertion, that matters to me!!!
I ask for feedback from supporters of Elliot Wave Theory and get back on topic, as this thread has long failed to live up to its name. I respect, truly respect, mathematicians, statists and programmers (probably future millionaires or already:)), but it's still muddying the waters.
...The visible difference of FRs can be explained by the presence of a noticeable negative correlation between samples in a number of EURUSD ticks, which leads to the negative feedback effect, i.e. the time series "tries" to hold the current price value to a greater extent than the EP series. Something can be put down to not strictly matching the FR for ticks see above....
Have written a bit about it here
http://forum.fxclub.org/showpost.php?p=599756&postcount=60
http://forum.fxclub.org/showpost.php?p=594214&postcount=6
[added later] a brief conclusion I made for myself. At first glance, the distribution of ticks
is almost similar to a normal distribution. But this is a deceptive similarity. If you look closely
consider the mechanism of the distribution of tics, it turns out that if it were not acted upon by
by external forces, the distribution would be triangular, but it's not. Essentially,
two forces are at work, one tends to return the ticks to their previous level, they are either the mechanisms of
market liquidity, or the mechanism of ticks formation in a particular DTZ. The other force, of a higher
order, tends to change this level. As a result, we have a distribution, which is more like
to the Cauchy distribution.
Another conclusion. In old books on mathematical statistics, early to mid last century, you can find
that a random process (in this case, tics) can be completely characterised by
by its distribution. Then, when this assertion was not confirmed, it was argued that
can be characterized by a bivariate distribution. More recently, attempts to assert something
to argue something like that have ceased. Where I'm going with this? I'm saying that to model with a normal distribution
tics are inherently flawed because the mechanisms are different. The external coincidence of the shape of the curves...
of the curves does not allow us to identify these two different processes.
A little self-citation.
I have decided to be generous to my heart and to help, so to say, biological science. I decided to make my own contribution.
I collected banknotes of one denomination from my entourage under the threat of physical violence
. The banknotes were of different age. It gave me confidence that they didn't come from the same wad
. Not mine. What else can I say, the people around me are clearly not rich.
I hope they're honest.
Following the method of Dr. G. Rosenberg, a biologist, supported by Prof. S. Ruderman
, I copied all the numbers from the notes in a row. I looked at the expensive drawing with interest.
Here one has to be diligent and extremely attentive. A widespread mistake,
you should copy numbers, not denominations, otherwise the picture turns out to be interesting, but
about something else.
The numbers, from 0 to 9, are supposed to be evenly distributed. On the histogram (not shown here
) no hints to normality or any other regularity could be found.
In the resulting series (blue with small circles), the noble naturalist suggests to look at
the size of the waves, count them and come to the conclusion, as he once did, that "...every third wave,
naturally, on average will be slightly bigger than its neighbours...". The way of defining the "wave" itself and
defining "wave size" remained a mystery to me. Of course I understand that waves - the same
is a perfectly obvious thing for Doctor of Biology, but still I would like some clarity for modest
admirers of the talent.
But! it was easy to calculate the number of extremum points (in Kendall Stewart's sense), on
a smoothed arithmetic average with a period of 5 (reddish on the chart), they turned out to be
exactly 32. Thus, every third (3.09375 to be exact) point of this smoothed series is an extremum
. Based on these results, I believe that geometers from the sect of PI number witnesses simply
must immediately declare gazavat to Doctor of Biology G.Rosenberg. Especially since the damned
vivisector promised that there will be no less than 4.5 waves per first level maximum, and 11
waves per second level maximum. That said, the thunderer of neighborhood frogs and subtle connoisseur of the microscope as
a tool for assessing the sex life of amoebas, angrily rejects the connection between the number 11 and periods of solar
activity. Heliobiologists are alarmed. Just a spit in the soul of an entire scientific branch -
"This example seems to illustrate quite convincingly one of the typical errors in the use of
statistics in biology - taking the coincidence of curves as binding evidence of a causal link
one phenomenon to another." - he tells them.
On the other hand, if you change the word "biology" in the quote to "speculation" you get quite
not even bad.
Summing up, I, as a matter of foresight, decided not to send donations to biologists. The
was a very convenient tool for analysis, although it is a bit riddled. Until they repent and
telegraph urgently why this their Dr. Duremar used a moving average period of 5, and
not 6 or even 7 - they will not receive money.
Here I intend to strictly follow the example of the poet A.S. Pushkin. "And to familiarise myself with the personality of
Alexander Sergeyevich (my favourite Russian poet, by the way) I recommend reading Pushkin's
correspondence with friends. And not so much to giggle about how Pushkin described Anna Kern, but to rejoice at the
fact that Pushkin was a very, very thrifty man. He married his sister, but did not give her a dowry
. Though he promised to do so. Then he corresponded with his sister's happy husband for two years. At first,
he wrote politely: "I have no opportunity. Then he sent a challenge to a duel.
No money - let's shoot." (c) John Polikarpovich Mozgovoy
...I ask supporters of Elliott Wave Theory to respond and get back on topic, as this thread has long failed to live up to its name. I respect, truly respect, mathematicians, statists and programmers (probably future millionaires or already:)), but it's still muddy.
The right thing to do is to start a new thread, about VTE, it's useless here, there have already been attempts.
Hi Sergei!
It's not working out well for the new modelling minutiae file. I believe it has to do with the quality of the modelling.
Here are two pictures that illustrate what actually causes doubts.
и
The first one shows dependencies of number of zigzag nodes (or kagi-building) for real TICs (usd), and for
of model Wiener process (vnr), which you gave earlier (the same 1 million ticks).
The curves are very nice and close in spirit. :-)
I cannot show the analogous curves for bars of the minute chart, because the model minute charts do not give such a picture.
Instead, I give you the graph with the red line showing the correlation of the number of zigzag nodes for real ticks to
the number of zigzag nodes of model ticks. And with the blue line - the same ratio of numbers of zigzag nodes, but for minute real and
model graphs.
As can be seen, for ticks the ratio tends to const, while for minutiae it tends to star.
This is a fundamental difference in the behaviour of the model data, the source of which is unknown to me.
I can tell by eye that there is a VERY large number of outer bars on the modeled minute chart,
Their size is VERY large (>100 pips - hardly ever on real 1-minute charts).
As a result, for real candlesticks ATR = 2.19, while for model ones ATR = 3.96!
In reality about 350000 bars are obtained from 1969732 ticks,
and for the model process only 106831 bars are obtained out of 2200000 generated ticks.
I think there is something wrong with the model minute generation process itself.
I'm posting a series of ticks for EURCHF and EURGBP 2006 for the corresponding H-builds.
http://www.filefactory.com/file/8aa7b3/
The archive volume is 6.5 Mb.
As promised, I have built a time series of ticks "fully" identical to EURUSD 2006 ticks.
https://c.mql5.com/mql4/forum/2007/02/fakUSD1tik.zip
The following parameters are the same: FR for the first differences and the correlation coefficient between the first differences see fig,
Standard deviation (almost) and FAC (correlation coefficient for adjacent differences at different lags) see fig.
Based on these ticks, a series of minutiae have been modelled according to the scheme described in the post above. The FRs for the series of EURUSD1m 2006 and the model are shown in the figure below.
I think the FRs coincide very well.
You may check the archive with artificial data points here:
https://c.mql5.com/mql4/forum/2007/02/fakUSD1min.zip
Sergey, I do not undertake to judge about significance of correlations between first differences, FR and other statistical aspects for this case.
However, I would like to report the following detail. I have calculated H-volatility on model ticks of a new series (i.e. 2200000 ticks).
The results are as follows. Apart from H=1 for which Hvol=2.168, the other values fit quite well to the value of 2.0.
Max(Hvol-2)=0.088, Min(Hvol-2)=-0.018 - this is without taking into account this first value.
For the entire series (50 points), it is sko = 0.0292 or 1.45%.
That is, the difference from the previous series in 1 million ticks is insignificant.
On the other hand, 2200000 ticks per 106831 minute bars is on the average 20.6 ticks/min.
This is not at all consistent with an average of 5.5 ticks/minute of real data flow.
This, plus what I wrote earlier, leads me to believe that it's about the process of bar formation after all,
and not the shape of the FRs, first differences, or anything like that.
The research needs to be followed through, of course. Abandoning it halfway is not very serious. It's no big deal!
But it is better not to detach ourselves from reality and not to sink into abstraction. And reality is a trading strategy,
which in this case is more than simple. The only question is whether it is real ! That can be found out independently of
of the study of the Wiener process.
By the way, then explain what a Wiener process is and how it differs from a Markov process.
PS I will repeat the experiment with new minutes, but I stand by my IMHO.