a trading strategy based on Elliott Wave Theory - page 212

 
I think we are in the process of developing an excellent, statistically sound mathematical apparatus for escorting our brokers to the nearest loony bin. <br / translate="no">
PS: What I mean is that this criterion will work every once in a while...or, in a tick....o about that...

Well, well, we already know how to deal with high transaction frequency :-) - Just plug in one more "independent" criterion (for example, by the connection: previous jump - expected jump), and the frequency of trades will fall several times, while the forecast reliability will only increase!
Clearly, the market deflection under small disturbances is shown by the distribution function of the amplitude of the market reaction in response to a +2 point (e.g.) EURCHF 2004 perturbation 1 min:

For comparison, here is the unperturbed distribution function of the same instrument:
 
Right, it is possible to insert a criterion, even two. But I still have my doubts about this approach. Let's see.... :о)


"
Sergey:
The interpretation is as follows: if we saw a +10 pips perturbation, it is more likely to expect a -10 pips pullback on the next bar (see fig). Of course, the pullback may be any, even "to the wrong side", but statistically, the amplitude of the rollback is equal to the disturbance amplitude. Errors are not senile, they are equal probability and will absorb, with an increase in the number of trades, each other, but the statistical advantage will remain on our side!"


But there is no such thing!!! If you look at the minute charts the price jumps anywhere! I haven't found any confirmation (with my eyes very visible!) of this anywhere! The same bounce will happen but a day, a month, a year later.... We will ONLY be leaking with this super stat advantage!!!!!


According to the statistics (most likely), the price should stay in place all the time (if it's +10 now it's probably -10), but that's not happening!!!! Exactly because we don't look at the price, we look at the deviations...

Or maybe I don't understand anything about stat benefits either...quite possibly.

PS: Not that I'm distracting, but I'd like to remind you that you promised to give your thoughts on trend definition ...
 
<br / translate="no"> But there is no such thing!!! If you look at the minute charts the price jumps anywhere! I haven't found any confirmation (with my eyes very visible!) of this anywhere! The same bounce will happen but a day, a month, a year later.... We will ONLY drain with this super-stat advantage!!!!!

You're wrong.
If we are going to drain, it's for one reason - FAC module on volatility, less spread!
So you don't need to look anything up with your eyes, but sit down at your computer and run a number of instruments on different TFs for evaluation on this parameter.

PS: Not that I'm distracting, but I'd like to remind you that you promised to give your thoughts on trend definition ...

Uh-huh...
 
<br / translate="no"> If we are going to lose, it's for one reason - the module of the FAC on volatility is smaller than the spread! So we don't need to look anything up with our eyes, we need to sit down at the computer and run a number of instruments on different TFs to assess this parameter.


Why do we need deviations then? Ah-ah-ah-ah I see, for them we consider FAC (still I do not like this abbreviation...).

Let's take EURUSD:

Spread - 3 or 0.0003?
For it, the FAC should be [0:1].
Volatility on average, in what range???
 
...What happens on the ticks is easy to imagine. <br / translate="no"> Since the price is moving slowly, and the ticks are ticking fast, there must be a very strong negative autocorrelation. And understandably: up and down and up and down ...
So what follows from this? After every tick upwards open downwards and vice versa ? :-)))
...

a little on this subject
http://forum.fxclub.org/showpost.php?p=618349&postcount=297
http://forum.fxclub.org/showpost.php?p=624720&postcount=326

http://forum.fxclub.org/showpost.php?p=622143&postcount=310

http://forum.fxclub.org/showpost.php?p=626115&postcount=334
 
Sergey, I can't understand the "nature" of FAC's work on volatility equals around the spread. Please explain. Is it empirically derived or scientifically intuitive?

PS: tell me the limits of volatility values for eurosd. I just don't count volatility at all. And right now I can't do such calculations.
 

Respect!
I read the thread with great interest. There's a lot of flooding, of course, but apparently that's the order of things... North Wind, don't you have that material published on the site, presented in the form of an article? I would be grateful to you. Also, I am interested in what you are doing now, what is the most perspective direction in trading? I will be glad if you join us as a critic and generator of ideas concerning the subject discussed here.

grasn 10.01.07 15:33
Let's take EURUSD:
Spread - 3 or 0.0003?
For it the FAC should lie [0:1].
Volatility, on average, in what range???
Sergey, I cannot understand the "nature" of the FAC's product of volatility equal to the spread. Explain please. Is it empirically derived or scientifically intuitive?

Sergey, the dimension of volatility and spread should be the same. If it is in metres - then in metres, and if it is in kilometres - then everything is in kilometres :-).
I use the "ideal" TS model in estimation calculations which comes to predicting only one parameter - the direction of the expected jump in price. The amplitude of this jump can be assumed equal to the volatility of an instrument in a selected timeframe or its standard deviation, which is almost the same. Taking into account that FAC can be interpreted as a relative value of prevalence of one type of price movement over the other (the opposite and counter-directional jumps), then we can state, without loss of accuracy, that the TS, based on the "ideal forecasting indicator", will NOT make a mistake when choosing the direction of the opening position, with the probability proportional to the absolute FAC value, underlying in the "ideal forecasting indicator". Profit or loss in pips from each trade is reasonable to estimate the value of the standard deviation of the instrument. Then the profit of TS on a sufficiently long time interval can be estimated as the difference of all successful trades and unsuccessful ones, each of which is multiplied by the volatility. Further, let's relate the obtained gross return to the number of executed trades and get the average estimation s of the "ideal" TS return per one trade:
s(TF)=Volatility(TF)*{(n+)-(n-)}/N=FAC(TF)*standard deviation(TF), where (n+) is the number of deals with positive balance, (n-) is the number of "negative" deals, N is total number of deals.
Which was required to prove.
PS: tell me the limits of volatility values for eurosd. I just don't calculate volatility at all. And right now I can't do such calculations.

If you can't estimate volatility, estimate standard deviation.) There will be no difference.
PS: Not that I'm distracting, but I'd like to remind you that you promised to give your thoughts on trend definition ...

Let's go...

The basic objectives of time series analysis.
The basic objective of a statistical analysis of a time series is to:
1. Determine which non-random functions are present in the decomposition, i.e. determine the type of indicators;
2. construct "good" estimates for those non-random functions present in the expansion;
3. to select a model adequately describing behaviour of non-random residuals and statistically estimate parameters of the model.
Successful solution of these problems determined by the basic purpose of statistical analysis of time series is the basis for achieving the final applied goals of the study and, first of all, for solving short- and medium-term forecasting problems of time series values. The main elements of econometric analysis of time series are briefly described below.
- Most mathematical-statistical methods deal with models in which observations are assumed to be independent and equally distributed. The dependence between observations is most often seen as a hindrance to the effective application of these methods. However, a variety of data in economics, sociology, finance, commerce and other fields of human activity come in the form of time series in which the observations are mutually dependent, and the nature of this dependence is precisely the main interest of the researcher. The totality of methods and models for studying such dependent observation series is called time series analysis. The main objective of econometric analysis of time series is to construct as simple and econometrically parameterised models as possible, adequately describing the available observation series and providing the basis for solving, in the first place, the following problems:
(a) discovering the genesis mechanism of the observations comprising the analysed time series;
(b) construction of the optimal forecast for the future values of the time series;
(c) working out the management and optimization strategy for the processes under analysis.
- When discussing the genesis of the observations forming a time series, one should keep in mind (and, if possible, model) the four types of factors under the influence of which these observations may be formed: long-term, seasonal, cyclical (or opportunistic), and random. It is not necessarily the case that all four types of factors need to be involved simultaneously in the formation of the values of a particular time series. A successful solution to the problems of identification and modelling of these factors is the basis, the basic starting point for achieving the final applied goals of the study, the main of which are mentioned in the previous paragraph.
- In starting the analysis of a discrete series of observations arranged in chronological order, it should first be ascertained whether factors other than purely random ones were indeed involved in the formation of the values of this series. The term "purely random" refers only to those random factors that generate sequences of mutually uncorrelated and equally distributed random variables with constant (time-independent) means and variances. The answer to the given question is obtained by performing a statistical test of the corresponding hypothesis, for instance, with the help of one of the "series tests", the Abbe criterion, the Box-Pierce test and the Ljung-Box test.
If such a statistical hypothesis test shows that the available observations are mutually dependent (and possibly unequally distributed), then an appropriate model for the series is fitted. The model set within which this selection is carried out is usually restricted to the following classes of models:
(a) the class of stationary time series (which are mainly used to describe the behaviour of "random residuals"),
(b) the class of non-stationary time series that is a sum of deterministic trend and stationary time series,
(c) the class of non-stationary time series that have a stochastic trend, which can be removed by successive differentiation of the series (i.e., by moving from a level series to a first-order or higher order difference series).
In the framework of econometric analysis of time series, we combine the series in classes (a) and (b) into one class, which, following the recently accepted practice [see, for instance, Maddala, Kim (1998)], we call the class of TS-series (trend stationary series, stationary relative to the deterministic trend). An adequate method for residualisation of time series belonging to class (b) is subtraction from the deterministic trend. On the contrary, for series belonging to class (c), an adequate method of residualization of a series is a transition from a series of levels to a series of differences (of first or higher order).
- Stationary (in the broad sense) time series are characterized by the fact that their mean, variance and covariance do not depend on the time for which they are computed. Interdependencies existing between the members of a stationary time series can usually be adequately described within autoregressive models of order p (AR(p)-models), moving average models of order q (MA(q)-models) or autoregressive models with moving averages in residuals of order p and q (ARMA(p, q)-models).
- A time series is called integrated (reintegrated) of order k, if consecutive differences in this series of order k (but not of lesser order!) form a stationary time series. The behaviour of such series, including those containing the seasonal component, is quite successfully described in econometric applied problems by using autoregressive models&#61485; integrated moving average of order p, k and q (ARIMA(p, k, q)-models) and some of their modifications. This class also includes the simplest stochastic trend model - the random walk process (ARIMA(0, 1, 0)). The random walk increments form a sequence of independent, equally distributed random variables ("white noise"). Therefore, the random walk process is also called "integrated white noise".
- To fit a model to a particular time series means to identify a suitable parametric family of models as an admissible set of solutions, and then statistically estimate the model parameters from the available observations. This whole process is commonly referred to as the model identification process, or simply identification. For proper identification of a time series model, it is necessary to decide whether the time series under study is stationary, stationary with respect to the deterministic trend (i.e. the sum of the deterministic components and the stationary series), or whether it contains a stochastic trend.
 
I keep reading Northwind's excellent posts at http://forum.fxclub.org/showthread.php?t=32942&page=1
At times, I die of laughter when confronted with the remarks of militant floodlords! It's like a circus, they should be in kindergarten for maths, but no, they're the wrong age! Generals, man.
What is interesting, the situation that takes place on most forums (ours, perhaps, is a rare exception), unequivocally indicates that the main contingent of "traders" participating in forums, are illiterate and, as a rule, flawed people. Who entered the market, perhaps, only out of despair and selective dementia.
Sorry, I couldn't resist.
 
Neutron 11.01.07 07:58
...North Wind, do you not have that material published on the site presented as an article? I would be grateful to you. I'm also interested in what you are doing now, what line of research you consider the most promising in trading? I will be glad if you join us as a critic and generator of ideas on the subject discussed here...

No, I don't have any material in the form of an article and unlikely will. I've been doing everything little by little.
I deal with everything bit by bit, but mainly, of course, from stochastic methods point of view. The same problem about decomposition, but apparently not in a pure form, as it is formulated by the classics.
I have read this topic in full, with interest, at least because I myself have followed this path. Personally I liked the "caterpillar" from time analysis methods. But again, I could not use the pure method of An.Vremena.
 
Neutron 11.01.07 09:41
... Sometimes, I die of laughter when confronted with the remarks of militant flooders! It's just a circus, they should be in kindergarten - to learn maths, but no, not that age! Generals, man...

:) Pay no attention, this is a test sent down to us from above, the grace of our father and his son and his spirit, as a test of the strength of our faith :::)))