a trading strategy based on Elliott Wave Theory - page 17
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
А даёт ли стратегия прибыль при тестировании по ценам открытия (быстрый метод) ?
It does too. The difference of profit with M1 (all ticks) is 5-10%. I just think that all ticks give more reliable results and that's why I don't use M1(fast method).
That's very good, that's what I expected.
The difference in testing methods affects if the program uses programmed closing or opening of explicit orders. And since in your case SL and TP close, the testing method shouldn't affect it.
The difference in testing methods affects if the program uses programmatic closing or opening of explicit orders. And since in your case the closing is by SL and TP, the testing method shouldn't affect it.
And of course we should not forget that apart from the fact that with different testing methods (all ticks) and (fast method) you will have differences in values of co-optimized parameters in addition to differences in the strategy profitability! And there is no way to fix it :o). If we can somehow reconcile with the difference in profits of different testing methods, then it is very problematic to live with differences in values of optimized parameters, at least for me :o))).
Vladislav, could you help and suggest what literature (in electronic form) is worth reading to deal with the above message? Or do you mean the Bulashev textbook, recommended by you? That is, by a quadratic form you mean approximation of a price series by a sum of terms consisting of a constant, a first-order term and a second-order term? Or am I missing the point? What about converting the quadratic form I assume to potential energy? How is this done? I haven't encountered it yet either :o(
And you also said that you have posts about your problem statement on Spider in the Santiment thread. But I searched badly and couldn't find your VG's post. Please give me the link, if it is not difficult.
Thanks in advance for detailed reply!
1. When constructing a linear regression channel, do you use a straight line equation, or do you approximate the price series by an equation that contains a second-order term and then reduce this second-order equation to a linear straight line equation after mathematical transformations as described in Bulashev's book? Please give your opinion on the expediency of applying the first-order and second-order approximation equations to the price series. Is there any perceptible difference in different equations in terms of results (trading itself)?
2. You said that you use standard deviation in your strategy. Could you explain how you use it?
Thank you in advance for your answers!
Vladislav, can you help and suggest what literature (in electronic form) is worth reading to deal with the above message? Or do you mean the Bulashev textbook recommended by you? That is, by a quadratic form you mean approximation of a price series by a sum of terms consisting of a constant, a first-order term and a second-order term? Or am I missing the point? What about converting the quadratic form I'm assuming into potential energy? How is this done? I haven't encountered it yet either :o(
And you also said that you have posts about your problem statement on Spider in the Santiment thread. But I searched badly and couldn't find your VG's post. Please give me the link, if it is not difficult.
Thanks in advance for detailed reply!
Regarding quadratic forms ( F(x,t) = A*x^2+B*t^2 + C ) - this is mataphysics, field theory and optimization theory in the mathematical sense of those terms. What is meant by optimisation of system parameters is just a consequence of a wide enough class of mathematical methods to obtain an extreme solution that satisfies a contradictory set of constraints. I haven't found it in electronic form, though I'm sure it exists. There is a lot of literature - I can not even say where to start.
As for the thread about a centiment - too lazy to look for it now - it did not get to the equations then: it was decided not to pay attention :).
I can outline the main points here:
1. Markets are managed by people (even if they have a lot of capital, it does not matter).
2. people with the same interests have the same "attraction zones" (for example, people with similar psycho-types prefer to trade certain instruments, which basically provides the peculiarities of the markets - hypothesis)
It seems to be a deadlock (which many people start with), but if we make some more assumptions, then there is hope:
People tend to act the same way in the same situations (the presence of repetition in similar decisions).
Let us assume that the actions of any group of managers in the market stem from their desire to maximise profit. Let us also assume the existence of some manager (an ideal system) which ALWAYS achieves an extreme result. Then the action must come from something stronger than a simple desire to move the market in one direction or the other. Example - a couple of years ago, Japan's intervention in support of the quid was quite successful. After a couple of attempts Japan announced that they were not playing such games anymore. And they were throwing a lot of money in five-ten minutes trying to stop the trend of the Euro.
Further, accordingly, it is possible to assume the presence of some external force, moving the market or forming prerequisites for decision-making of market managers. It remains to assume (quite logical in my opinion), that this force is the result of many constituent factors and it will be possible to try to set the task and evaluate the solution.
Actually such system, which ALWAYS gets the correct prognosis we will get in an ideal result (it is like a Carnot cycle - theoretically it exists, practically it is possible to approach it better or worse). And in reality of course there are some ranges of uncertainty.
And another thing - all this comes from the fractal nature of the market (this hypothesis develops in opposition to the efficient market hypothesis) - that is, that there are periods of non-random prediction in the market. That is, to go looking for a black cat in a dark room, one must assume the presence of a certain number of black cats in at least some part of the dark rooms :).
Good luck and trailing trends.
Vladislav, about the Carnot cycle, I can suggest that your strategy uses the calculation of the work done by the external force, based on summing up the white and black candles, for example by opening and closing prices. So, I understand that if you sum up separately the bodies of white and black candlesticks, then we will have a presumptive ratio of how much more work was done downwards than upwards or vice versa. So, we can suppose from this data that the system is at one of its two extremes based on the history analysis, for example? Then, if it is not a secret, on the basis of what timeframe do you make these calculations? And what is the optimal number of bars to calculate? Although I certainly may suppose that it is not a matter of which timeframe we use for calculation and how many bars we need. Then which timeframe do you think should be used for the calculation? Because depending on the time interval that we take for the calculation, do ALL the results depend on it? Perhaps, you take a time period corresponding to P=64 in the Murray indicator you are using? That is, it is better to take a period of 64 trading days for the calculations?
Vladislav, about the Carnot cycle I can assume that in your strategy you use the calculation of the work done by the external force based on the summation of white and black candles for example by opening and closing prices. So, I understand that if you sum up separately the bodies of white and black candlesticks, then we will have a presumptive ratio of how much more work was done downwards than upwards or vice versa. So, we can suppose from this data that the system is at one of its two extremes based on the history analysis, for example? Then, if it is not a secret, on the basis of what timeframe do you make these calculations? And what is the optimal number of bars to calculate? Although I certainly may suppose that it is not a matter of which timeframe we use for calculation and how many bars we need. Then which timeframe do you think should be used for the calculation? Because depending on the time interval that we take for the calculation, do ALL the results depend on it? Perhaps, you take a time period corresponding to P=64 in the Murray indicator you are using? That is, it is better to take a period of 64 trading days for the calculations?
As for the Carnot cycle - it is just an example, as a limiting value.
As for the Murray's dimension - 64 is the recommendation of method's developers. I cannot judge if it is the best result, but I use the following estimation to determine the minimal point far enough for convergence of methods:
I do not remember the exact link, I looked through articles on analysis related to calculation of persistence (Hurst coefficient > 0.5). There were estimates on the fractal dimensionality of markets. Conclusions that were made: Hearst's coefficient for many types of markets lies in the area of 0.62-0.64, which in turn denotes the loss of initial conditions for time series in average of 90 days. That is, perturbations more than 90 days back in time will have a vanishingly small impact. Now, I set my starting point for reference to no more than half a year (180 days to be exact) - 90 days does not always give enough information for convergence, although perhaps this is the result of the implementation and with other algorithms and quality criteria 90 is enough - I do not know yet. When I was calculating periods using all available history, the result was not better - I simply spent more time on calculations.
The number of bars to be calculated is determined by the structure itself and the octaves are built for it - so I can't even say at what point in time they are calculated - the computer has been counting for a long time now :). The methods do not depend on TFs, so you can do it on any TF, as long as you have enough history for half a year.
I don't evaluate candlesticks, paternas and the like - you will get noise-dependent methods. That is, the result will depend on the quality of quoting, which, IMHO, is not good.
Good luck and good trends.
...we may make an assumption that a trajectory function may be adequately represented by a certain quadratic form - which is almost simple: searching for extremums of quality criterion functionals for such forms is a highly researched area. That is, one has to make a selection of samples that satisfy the quality criteria in an extreme way.
One and the same Murray reversal level for different channels will be in different confidence intervals - you need to cut it off somehow, don't you? And the quality criterion is potential energy - see about quadratic forms - nothing unusual.
I have looked through the literature on the subject. I assume that perhaps not everything can be found in the application to this particular case. But on the basis of what I have managed to look at the following assumptions regarding quadratic forms. First of all, let's begin with the method of finding this most approximating price series function. I suppose I can take a parabola function in the form
y(t)=A(t-t0)^2+B where y is a price, t is time, t0 is a point on the timeline where the parabola has an extremum and A and B are coefficients.
Then follows the problem of finding such optimal coefficients A and B that make the parabola optimal by criterion of potential energy minimum. As far as I understood from the reviewed sources, the essence of this optimization is as follows. We imagine the curve of the parabola as a line with the same field potential. Let it be zero for certainty. The gradient of such potential field will be directed pre-pendicular to the parabola line. Then the problem of the potential energy minimization reduces to the problem of finding such a parabola, in which the sum of the squares of the shortest distances between the points of the price series and the curve of the parabola will be minimal. So we need to optimize the parameters of the parabola to find the shortest distance between the points of the price series and the line of the parabola. The shortest distance is the distance along the straight line intersecting the parabola at right angles. So we need to solve the problem of finding those shortest distances. Could you share, at least methodologically, which method you use. I, for example, imagine the process of finding these shortest distances as follows.
1. We select (the algorithm of the calculation is not clear yet) more or less true parabola for the existing price series, which we want to approximate.
We approximate it by a polygon which has the equation of a line passing through each point tangent to the parabola. The equation of the line for point T Y(t,T)=a(t-T)+b, where a=2A(t-t0) and b=y(T).
3. Then for some selected point in the price series, we constrain the area of values along the t and y axes in which there exists a point of intersection of the perpendicular drawn from the point to the parabola with the parabola itself.
4. Iterate the equations of the segments of the polygon that lie in this region of the polygon for the intersection with the perpendicular. Do the required number of iterations and approximations to get the required error of perpendicular length calculation from the point to the curve.
5. Sum the squares of these segments and thus obtain the value of the target function.
6. Next change the parameters of the parabola and carry out the calculation in items 2 to 5 as many times as required. The smallest value of the target function corresponds to the value of parameters of the parabola that approximates the price series in the optimal way.
Then it is probably possible to calculate the parameters conditionally speaking "quasi-dispersion" and "quasi-SCO" from the obtained optimal value of the target function. On this basis, in addition to the existing parabola we can draw several more parabolas on the price chart having conditionally the numeric probability characteristics and embodying the potential field lines having the same probability of the trend reversal. For example the lines of 70%, 80%, 90% probability of reversal.
Vladislav, do you think I am moving in the right direction of understanding your strategy or I don't understand anything at all and have gone in a completely different direction?
I've forgotten the course of VM, I may be wrong, but you can try it this way:
The shortest distance from a point to a parabola would be the distance from the point along the line that coincides with the normal.
The normal to the parabola can be calculated through the first derivative, (the derivative is the tangent of the angle of inclination of the tangent).
therefore a system of equations can be constructed:
1. the equation of the parabola.
2. the equation of the straight line (normal) (knowing the derivative)
3. the point belonging equation of the straight line (normal)
If we solve the system, we obtain a rigorous solution.