Obtaining a stationary BP from a price BP - 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
I'm not talking about you. I meant it in a general sense. I was simply expressing my view on the topic of market stationarity, based on my humble life experience, so to speak). I was trying, so to speak, to bring a little bit of common sense, in my opinion, into the whole discussion.
Well, this is some kind of unproductive approach. There must be an error correction work - errors in the method, not in the code. Can we at least know the timeframe, sample size and grids? ;-)
What do you mean? What kind of mistakes?
It's much simpler, because no matter how you twist and turn it, you can't squeeze out perfect stationarity. After all, stationary BP by definition is such a series, on which if we impose Bolinger Bands, we obtain three strictly horizontal lines. I.e., neither a simple moving average (expectation) must not bend, nor channels (RMS) must not narrow - go away.
Whenever I see or intend to apply a transformation to a price, I am haunted by the spectre of Dub's martingale stopping theorem (I do not know its exact formulation myself, judging it only by a well-known layman's paraphrase). The ghost asks: "Why the fuck would you bother with a martingale if it's the same bastard you'll get anyway?".
My latest mouvee grimace has turned out to be a fake, apparently. Trying to make the prediction sustainable eventually led to a trivial prediction (m.o. prediction price equals last price).
Before we make fun of prices, we should ask ourselves: which price property are we torturing - roughly speaking, martingale or non-martingale? It should be the non-martingian one, i.e. the one that gives us a principal opportunity to go beyond the martingian hopelessness.
I know at least one non-martingian property of price. It is the sacred cow "if a trend started, it will continue". This property doesn't exist in a Wiener process, but the real price does, since it turned out to be so robust. This property tells us directly that the future price readings at least sometimes appear to be dependent on history (the dependence of readings is a violation of martingness).
The only thing left to do is to catch the moment "the trend has begun". This is another global problem that many here are also trying to solve.
The second likely non-marting property of price is the fat tails of the return distribution. According to Peters this also indicates a correlation.
The third candidate for price martingality is the presence of peculiar catastrophes in any forecast process, which often completely break the fairness of the forecast. These catastrophes usually coincide with overly fat tails. After such catastrophes we have to change, if not all logic, then at least some parameters of the price analysis.
Anyway, we can go on and on about all this, but the ghost of The Oak theorem will still hang over me :)
Well, I swore a little bit - and it is easier now...
Whenever I see or intend to apply a transformation to a price, I am haunted by the spectre of Dub's martingale stopping theorem (I do not know its exact formulation myself, judging it only by a well-known layman's paraphrase). The ghost asks: "Why the fuck would you bother with a martingale if it's the same bastard you'll get anyway?".
My latest mouvee grimace has turned out to be a fake, apparently. Trying to make the prediction sustainable eventually resulted in a trivial prediction (m.o. prediction price equals last price).
+1 :)
This is the essence of fitting.
Glad for your success. But I dont understand some details, i.e. how DT defines the minimal size of price movement, but i am sure you understand what you are saying about it :o(
I was referring to the minimum trading horizon (MT), at which the DC commission does not exceed the positive MO of the TS. However, it is not rational to increase TF too much because of the constant increase in market efficiency (the same martingale that Alexey has been talking about for so long). Thus, we have an optimum on the only TC parameter.
Everybody does it, don't be embarrassed and don't get all that nonsense in your head.
>> OK. I won't.
You don't have the slightest idea of when the mismatch will occur.
You're right - there is no idea - I'm not a predictor! But there are statistics and they show that, on average, the characteristic time of existence of the identified quasi-stationary process (QSP) is such-and-such, and this can and should already be exploited
Once a month? Can you simulate the market for a month ahead?
I can estimate with some certainty the duration of the detected pattern on which you can make money. Right now that horizon averages 10 to 20 working days, on which I make an average of 1 transaction per day.
Once again. I am not talking about predicting the price (which is the only thing you can earn on), but about identifying the PSC and estimating its capacity. And I have always predicted the price only one step ahead (in TC events countdown). It seems to be reasonable, because confidence of BP forecast of price type is very low (at 1-5% level), and n-step-ahead forecast has (%)^n value, i.e. already at the second step it tends to zero (P=0.01^2=0.0001->0), which makes the procedure of drawing of different curves at the right border of price series (in future) absolutely meaningless! Unless, of course, you consider it from the point of view of artistic value. But that is a matter of the "artist's" taste and playfulness.
Mathemat wrote (a) >>
I am aware of at least one non-martingian property of price. It is the sacred cow of "if a trend has started, it will continue".
It doesn't.
Any method of statistical analysis reliably shows that the series of the first price difference plotted on different TFs or for different trade horizons is characterized by the property of antipersistence (there are some rare exceptions) or, in other words, by "reversal".
Alexey, I don't have to tell you that BP obtained by integration of such a SV (analog of the price one) will most likely not continue the trend that started it, but will reverse and go the other way!
The holy cow is rather a different colour - "if a trend has begun, it will reverse".
Alexey, I don't need to tell you that BP obtained by integrating such a NE (analog of the price one) will most likely not continue the trend started, but will turn around and go in the opposite direction!
Found an analogy, a pattern???
The holy cow is rather a different colour - "if the trend started, it will reverse".
It depends on what you call a trend. The problem here is your definition of a trend.
Just give me the main parameters you can use to say - that's the way it is, a trend.
It's much simpler, because no matter how you twist and turn it, you can't squeeze out perfect stationarity. After all, stationary BP by definition is such a series, on which if we impose Bolinger Bands, we obtain three strictly horizontal lines. I.e., neither a simple moving average (expectation) must not bend, nor channels (RMS) must not narrow - diverge.
..................................................
etc.
Many notions of stationarity/non-stationarity have been expressed. But no matter what you call any phenomenon in animate or inanimate nature, it does not become different and does not change its essence. Thus, it is probably important how we understand these characteristics of the phenomenon (price time series) to be able to use these characteristics for our purposes.
I will try to give my definitions of the stationarity of the process, IMHO.
A stationary time series has a maximum and minimum in its values, so the graph of a stationary BP lies in a strictly horizontal corridor. It can be divided into two types:
1) Independent Stationary BP . Each successive value is independent of the preceding one. It cannot be approximated (example - white noise, and bold in Reshetov's quote).
2) Dependent Stationary BP. There is a dependence between time-varying values, not necessarily between consecutive values. It can be approximated.
Obviously, if the price series belonged to either the first or the second definition of stationarity, it would be very easy and unencumbering to trade by placing the pending orders on the channel boundary in the first case or by approximating the series in the second case. The price series does not belong to either one or to the second definition of stationarity (my definition)
So, it is necessary to use the price series properties, which refer to one of the stationarity definitions. For example increment. As Matemat said "It only remains to catch the moment "the trend has begun". And no one forbids us to trim thick tails.
That's how my idea works.
While writing this, Neutron had time to say something of my considerations
Did you see my post on page 15 of the sub?
Did you see my post on page 15?
Yes I did, your point is clear to me and I agree with it. I was just trying to somehow formulate a definition of stationarity for myself, one that we can use.