Econometrics: one step ahead forecast - page 98

 
avtomat:

So create an adequate model, use econometric techniques, and harness the power of econometrics!!! what's stopping you?!

The only question is, where is the power of econometrics...

Oleg, we have also been waiting for something sensible from you using the TAU. And you amuse us with pictures and incomprehensible formulas. You would at least set forth a concept that is accessible.

Well, let's say, at least like this (I do not remember where I heard about it, but I heard it exactly): in the market at any given time there are 2-3 players, and they determine all the processes. The market does not immediately assimilate all the incoming information, and therefore there is a certain... uh... relaxation, with different transients superimposed on each other. And so this relaxation is our potential bread.

Why - a clear and concise concept, also meaningful. Everything else (diffusions, and the estimation of their parameters by quotients and the construction of "regressions" approximating the behaviour of the market) is already a technique. And by the way, one cannot do without econometric methods (or rather statistics) at the final stage of model evaluation anyway. It's a random process anyway...

Something similar once tried to do about 15 years ago, when I've never heard of Fora. But this model was a bit different, although it was also reduced to a diphurk - but parametric.

 
tara:

Oleg, the model is adequate, so is the author. It's just that you have different goals, ambitions and mass-size characteristics :)

;) I just really liked this passage:

for the stated model, which is primitive, poorly justified and doesn't use a hundredth of an econometric

 
Mathemat:

Oleg, we have also been waiting for something intelligible from you using TAU for a long time. But you entertain us with pictures and formulas that are incomprehensible. You should at least set forth the concept at an accessible level.

The pictures concentrate the information in a concentrated form -- clearly, understandable, without unnecessary words. For example, the last pictures I showed here are actual actual results of the system built on the basis of TAU. Are these results not clear as well? However, it has not aroused any interest. And what complex and incomprehensible formulas did you see there? It is enough to penetrate a little, and everything becomes clear and understandable -- the goals and objectives set, as well as the effectiveness of the system - current and prospective.

Well, let's say at least this...

Maybe on long winter evenings.... maybe I will.... I remember you hinting about the article...

Everything else (diffusers, and the estimation of their parameters by quotients and the construction of "regressions" that approximate the behaviour of the market) is already a technique.

It is the technique used that determines the final result.

And by the way, one cannot do without methods of econometrics (or rather statistics) at the final stage of model evaluation anyway. It's a random process anyway...

I'm not saying that they should be unconditionally thrown away. I am saying that statistics has its own well-defined range of tasks that it deals with and is designed for. Attempts to extend statistics to areas that are not under its control will not lead to positive results. Here, for example, an evolutionary model for balanceA and balanceB processes will be included from Monday, using elements of statistics; but by no means is statistics the basis for this model.

 
I have an interesting question. If we can use some well-known method for analysing open and close values, then what about the analysis of high and low or tick values, because they are not distributed at equal intervals in time. For example, the same prediction for one step of the author of the topic, how to do it, take the number of values per time interval or determined as now, how to deal with forecasts, for a discrete-time series it is known that the confidence interval is calculated for one time interval, but for non-discrete ones - for how long? Are there any statistical methods for dealing with such series?
 
avtomat: The pictures concentrate the information in a concentrated form - clear, understandable, without unnecessary words. For example, the last pictures I presented here are the actual actual results of the system built on the basis of TAU.

I didn't ask for the results of the system.

You tell me about the concept ofthe system itself . Some hint of it is right there. That's it.

If you don't want to do it, don't do it. But you created your branch for a reason...

 
-Aleksey-:
I have such an interesting question. If we can use a well-known method for analysis of open and close values, then what about analysis of high and low values or ticks, because they are not distributed at equal intervals in time. For example, the same prediction for one step of the author of the topic, how to do it, take the number of values per time interval or determined as now, how to deal with forecasts, for a discrete-time series it is known that the confidence interval is calculated for one time interval, but for non-discrete ones - for how long? Are there any statistical methods for dealing with such series?

In my mind, it's much worse than that. I have repeatedly raised this issue in the thread, but no one has responded. The point is this. I immediately stipulated a verbal model of the market: kotir = trend + season + periodicity + outliers + noise.

I am modelling two components in this topic: trend + noise. There is no season on forex, spikes are not interesting. The most interesting part is periodicity, which I understand as a wave with a variable period. We have sliced the continuous kotir into equal pieces and it appears that periodicity with a variable period was sliced wherever it occurred. And we're trying to predict! I have no approaches, no one has responded. C-4 gave a link where the author claims that periodicity is subject to the logarithmic law. But I do not understand how it can be verified. In fact, how to detect this periodicity?

 
faa1947:

In my mind, it's much worse than that. I have repeatedly raised this issue in the thread, but no one has responded. The point is this. I immediately stipulated a verbal model of the market: kotir = trend + season + periodicity + outliers + noise.

In this topic I am modelling two components: trend + noise. There is no season in Forex, spikes are not interesting. The most interesting part is periodicity, which I understand as a wave with a variable period. We have sliced the continuous kotir into equal pieces and it appears that periodicity with a variable period was sliced wherever it occurred. And we're trying to predict! I have no approaches, no one has responded. C-4 gave a link where the author claims that periodicity is subject to the logarithmic law. But I do not understand how it can be verified. In fact, how to detect this periodicity?


Periodicity in cotier increments? The concept of periodicity is too strict - all phases of a process have a fixed duration in astronomical time. Where can this periodicity come from? From real economic cycles tied to astronomical time - tax periods, harvesting crops, etc. A broader concept is cyclicality. This is when development also consists of a sequence of defined phases, but their duration is not constant in astronomical time. Almost all speculative processes are cyclical, not periodic. They are not tied to astronomical time, the concept of internal time is often used here. What constitutes an effective internal time counter depends on the process in question. What this is all about - to understand in time which phase of the process we are in.

For example, let's take the process - the accumulation of unrealized profits by a certain part of the participants. For example, speculators that trade different variants of trend-following. We can distinguish 2 main phases - their accumulation of open positions up to a certain limit (since they don't have infinite money in total) and the phase of profit taking, which can turn into a collapse and many of them have to take a loss. The practical implementation of such processes is the inflation of speculative bubbles. What would be an effective time gauge? Logically, the volume of positions accumulated by these speculators. The time of the accumulation phase comes to an end as this volume approaches critical. Any indicator of accumulation/distribution, which has overbought and oversold zones, focuses on these processes. The internal time will be the value of this indicator relative to these zones.

I.e. for the cyclic processes it needs some conditional "internal" time to understand when to wait for the phase change and to correspond.

 
Avals:

Where might this periodicity come from? From real economic cycles tied to astronomical time - tax periods, harvesting crops, etc. The broader concept is cyclicality

I understand your concern about the term "periodicity". It is based on a periodic function with well established concepts. But I have no other word for it. Variable periodicity is clearly visible on ZZ

You can clearly see that the distance between the vertices is different all the time. I can predict the target, the direction, but I can't predict the duration. There are Fibo time ons, that of Gann, Elliott waves, but these are all patterns.

But there is the concept of a phase. Maybe here?

 
faa1947:

Where might this periodicity come from? From real economic cycles tied to astronomical time - tax periods, harvesting crops, etc. The broader concept is cyclicality

I understand your concern about the term "periodicity". It is based on a periodic function with well established concepts. But I have no other word for it. Variable periodicity is clearly visible on ZZ

You can clearly see that the distance between the vertices is different all the time. I can predict the target, the direction, but I can't predict the duration. There are time ounces of Fibo, that of Gann, Elliott waves, but these are all patterns.

But there is a concept of phases. How about here?



In principle, we can say that ZZ divides the series cyclically into two phases (swinging up/swinging down) and that these phases do not have a rigid period. But usually when we talk about phases we are talking about the real processes behind the formation of the series, and the final manifestations are already consequences of these phases.

When there is only one process affecting the series or others affecting much less, then the phases on the series itself will coincide with the phases of the process. For example, we measure the temperature outside and make a graph. Clearly the multi-year graphs will highlight the phases of the changing seasons, behind which is the process of the earth's rotation around the sun and the position of its axis in relation to the sun. But if there are many influential processes, and at least part of them are non-periodic, the phase markings on the series will not clearly correspond to the phases of the processes. Mixing will occur. It is like measuring the temperature near a geyser or a volcano, for example. In addition to the phases of the solar process, there will also be phases of volcanic activity and the resulting temperature graph will not clearly show only the seasonal phases - it will be a mixture of the results of the superposition of the phases of the various processes.

 
Mathemat:

I didn't ask for the results of the system.

You talk about the concept ofthe system itself . Some hint of it is right there. That's all.

If you don't want to do it, don't do it. But you created your own thread for some reason...

That "hint" --- "up to my neck" simplified presentation of the "system concept", presented from the very beginning, from the very first page.

Now, a year-long experiment (and I'll note in parenthesis, on a real account with real money and cash) has started, based on the TS tractor, which is based on the same "concept" --- already, isn't this enough reason to create a branch?

At the moment the third month of the experiment is over. The results of the first two months show an annual average return of 2332%. According to the accepted model, when we reach the operating point, annual efficiency of the system will be in the range of 70000% - 80000% p.a. --- checking of such modeling result is also a good enough reason to create a branch.

By the way, informativeness depends not only on presentation but also on perception ;)