Obtaining a stationary BP from a price BP - page 19
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
joo писал(а) >> В чем я путаюсь?
joo , and great! So I've been slacking off.
Here, continued dabbling with trend and counter-trend modelling of price formation:
On the left are the increments, on the right is a characteristic view of the price dynamics. At the top for a trending market, at the bottom for a flat market. The correlation is only for the neighbouring counts in the first difference (FDR). In fact the dependences are of course more complicated, though trivial in nature. Thus, for price series, there exist reliable dependences only between neighboring samples in the FPR, but they are complicated by multiplicity of trade horizons. For example, a series of ticks can be divided into minutes and require correlation between neighboring bars. Split into five-minute ticks and also require correlations, etc.
How to model this? Any ideas?
P.S. By the way, take a closer look at the first difference series for the trend market... There appears to be non-stationarity for MO! MO is positive for an uptrend and negative for a downtrend!
The non-stationarity seems to be more related to the nature of the market than it may seem at first glance. And it is an invariable characteristic of a trending market... Look also, a flat market is more stationary at this point, and the price series is consequently more "calm".
In short, Sklifosofsky!
The point is that stationarity is not necessarily predictability and predictability is not necessarily stationarity.
Stationary BP is strictly a sideways trend with clear horizontal channels. Even if there is white noise in the channel, i.e. it is unpredictable by definition, even a fool may trade profitably in it using simple tactics for rebound from channels (even with the most sophisticated martin it will be very difficult to fail). It is possible to make forecasts in a sloping trend if its channels are clear. It is possible to predict on a seasonal component. It is possible to forecast and ... etc., etc. on everything, if it has the memory. I.e. it is important that the level of BP amnesia is not too high.
In general it is not necessary to bring VR to strict stationarity for forecasting. And normality - non-normality of distributions in general has no effect on predictability, because this property is most necessary for the corresponding PRNGs.
The most important thing is that the predicted BP should have the property of reverse transformation, i.e. it should be possible to restore unambiguously the initial BP from which it was obtained. Otherwise, what's the point if you don't get paid anything for successful predictions on all sorts of nerdy crap?
A stationary BP is strictly a sideways trend with clear horizontal channels. Even if the channel is white noise, so it is unpredictable by definition, it is still profitable even for a fool who uses the elementary tactics for rebound from channels (even with the most sophisticated martin it is very difficult to lose).
Yura, you've got to use your brain.
1. "A stationary BP is strictly a sideways trend with clear horizontal channels" - but a price BP is always dangling any way and only sometimes in a sideways channel.
2."White noise is unpredictable by definition, BUT (that) will still be able to trade profitably in it" - turns out a logical contradiction!
Why the hell did you write that? Once again, for those who are in the tank: when they talk about stationarity/non-stationarity, they are referring to the series of the first difference of the price. When they say that one cannot make money on SP with zero MO (like white noise), they mean that one cannot make money on BP built by integrating this SP (analog of price series)! You are really confusing (by a number of definitions and parameters) the price series with its first difference and thus misleading forum members.
Sorry, of course, for the tone of the message, but I tried to stick to your style of communication :-)
Yura, you've got to use your brain.
1. "A stationary BP is strictly a sideways trend with clear horizontal channels" - but a price BP is always dangling any way and only sometimes in a sideways channel.
2."White noise is unpredictable by definition, BUT (that) will still be able to trade profitably in it" - turns out to be a logical contradiction!
Why the hell did you write that? Once again, for those who are in the tank: when they talk about stationarity/non-stationarity, they are referring to the series of the first difference of the price. When they say that one cannot make money on SP with zero MO (like white noise), they mean that one cannot make money on BP built by integrating this SP (analog of price series)! You really confuse (by a number of definitions and parameters) a price series with its first difference and thus mislead forum participants.
Sorry, of course, for the tone of the post, but I tried to stick to your style of communication:-)
Yes, understand the definitions and how they differ from those used in physics or classical probability theory, and then you can debate, dummies. Hell, they teach this SCIENTIFIC approach in any discipline in any institute! Has anyone here ever been to any institute?
Yura, you've got to use your brain.
1. "A stationary BP is strictly a sideways trend with clear horizontal channels" - but a price BP is always dangling any way and only sometimes in a sideways channel.
2."White noise is unpredictable by definition, BUT (that) will still be able to trade profitably in it" - turns out to be a logical contradiction!
Why the hell did you write that? Once again, for those who are in the tank: when they talk about stationarity/non-stationarity, they are referring to the series of the first difference of the price. When they say that one cannot make money on SP with zero MO (like white noise), they mean that one cannot make money on BP built by integrating this SP (analog of price series)! You are really confusing (by a number of definitions and parameters) the price series with its first difference and thus misleading forum members.
Sorry, of course, for the tone of the post, but trying to stick to your style of communication:-)
Do you think Reshetov is such a dumbass that he confused the first difference with the cumulative sum - you call it integration. He wrote correctly, there is no contradiction. (white noise), then its first differences are unpredictable, because ACF=0, but the cumulative sum of white noise itself is predictable, and this is important, because the variance is finite, and the MO does not float in time, and in general the series as a whole has static parameters without any "adaptivity" and games with probability.
Reshetov писал(а) >> Суть в том, что стационарность - вовсе не обязательно прогнозируемость, а прогнозируемость - вовсе не обязательно стационарность.
Well, hello. We talked, talked, talked, talked, talked, and now we have a deal.
Yes.
Do you think Reshetov is such a dumbass that he confused the first difference with the cumulative sum - you call it integration. He wrote it correctly, there is no contradiction. series(white noise),then its first differences are unpredictable because ACF=0, but the cumulative sum of white noise is predictable,and this is important because the variance is finite and MO does not change with time and in general the series as a whole has static parameters without any "adaptivity" and probability games.
it is not so. Only a theoretical "ideal" series is unpredictable by definition. If, for example, HP with mo=0 and some dispersion is obtained in practice, it does not mean that this series cannot be earned and is unpredictable. It can even contain deterministic dependences but their occurrence is rare enough to affect the distribution over the entire series. The average hospital temperature isn't relevant here.
to Neutron
Apologies for the delay. Sergey, I find many conceptual "inconsistencies" in your approach. On the one hand you say that stationarity is impossible, and on the other hand you assure that statistically the system remains stable for about a month (no parameter changes required). But in that case, who prevents you from adjusting the steady-state parameters once a month?
Ещё раз. Речь идёт не о прогнозе цены (на чём собственно только и можно заработатьт), а об выявлении КСП и оценки его мощности. А цену я всегда предсказывал и предсказываю только на один шаг вперёд (в отсчётах событий ТС). Это кажется разумным, т.к. достоверность прогноза ВР типа ценовых крайне низка (на уровне 1-5 %) и достоверность прогноза на n-шагов вперёд имеет ценность порядка (%)^n, т.е. уже на втором шаге стремится нулю (P=0.01^2=0.0001->0), что делает процедуру рисования рвзличных кривулек на правом краю ценового ряда (в будущее) совершенно бессмысленной! Если, конечно, не рассматривать её с точки зрения художественной ценности. Но это уже дело вкуса "художника" и его игривости.
This is if you use some kind of AR model head-on, yes. But if you think about it a bit? For example, my artistic curves give very good results (and these are stochastic control systems with random structure combined with probabilistic neural networks) :o) I've actually come to the opposite result applying fractal analysis. Predicting by one count is a dead end. Time lag of one count is total Chaos, you can never predict anything there.
No problem. - Positive correlation coefficient between neighbouring samples in a series of first price differences.
Oh how!!!! And you write that you don't see the feasibility of converting to stationary? But let me see, what you're doing x(n)-x(n-1) isn't one of the ways of converting the series to stationary? Really and here you have to be extremely careful in applying the well-known formula. As you know very well, autocorrelation is a limit, which in the general case cannot be taken (sometimes it is possible to estimate it). Only by introducing significant restrictions on the properties of the series (one of them being stationarity) was it possible to obtain this formula. The price does not satisfy these conditions at all, the difference almost does:
Here is the interval of the whole series
This is what its correlation looks like (R(n)=R(-n))
And in fact, the autocorrelation estimate is as follows (and even then, very imprecise, without calculating confidence intervals and stuff)
PS: small mistake, the correct example is here https://forum.mql4.com/ru/27563/page22
it is not. Only a theoretical "ideal" series is unpredictable by definition. If in practice, for example, an HP with mo=0 and some variance is obtained, it does not mean that this series cannot be earned and is unpredictable. It can even contain deterministic dependences but their occurrence is rare enough to affect the distribution over the entire series. The average temperature in the hospital is not indicative here.
I mean the white noise, which is practically corrected for the volatility. The signs of white noise values are unpredictable.
What kind of series are we talking about in the highlighted case or is it just the prices of the last deals of the instrument?