Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 47
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
To speed things up, let me ask another question: there are a lot of graphical plots on the spider from .... / So you think that only graphical plots can give a forecast of further price movement?
There is one on the spider.
I don't think that only chart studies can give a forecast. In the arsenal of "..." there were also mathematical models (Protoforma). And what is in the arsenal of many other researchers I don't know. But there is probably more than that.
Why is that? I think you are a carrier of doubt that this type of alphabet is applicable to the market. Do you think it's fundamentally inapplicable? I'm just wondering.
Well I've already answered that it does. For example, bar representation or any other (Renko, XO, etc.), apart from deal flow or quotes. Partial loss information compression, also a useful task. But not directly applicable to forecasting, as described in the Russian example above.
I was desperate to see at least one developer in Minsk, but it turns out that's where they are hiding :)
In Minsk? Plenty.
I, by the way, want to go to Minsk in the next 2-3 months just for fun. We can cross paths.
Well I have already answered that it is applicable. For example, bar representation or any other representation (Renko, XO etc.) other than deal flow or quotes. Partial loss information compression, also a useful task. But not directly applicable to forecasting, as described in the Russian example above.
The point is not clear... Does it mean that we can't make calculations by selecting "only one price" from a candlestick? After all, "with partial loss" is "mashka". Or is that not what you mean? Moreover, the straight lines through the "centre of the candlestick's body" are very strong factors for price testing. They are as significant as the extrema. And in the case of a line built on the "doj-dodge" where "three in one" but "one price" is concentrated, it has even "the highest priority". This is to say that highlighting "one price on a candle" is ignoring (losing) the data.
I don't quite follow you anymore. I said "dangerous" in the context of working with non-stationary BP, and a direct consequence of neglecting this danger would be models that don't work in reality.
1. the source quotient is unsteady and you have to work with the source quotient.
2. The difference taking of the original quotient is one of the detrending methods. The word itself means that you have removed the trend. Taking differences is not the only detrending method. The one you used is the worst one - the information about the trend is lost. And you can take a smoothing and the residual from the smoothing and it will, if you follow you, be stationary.
3. Your statement that the differences are stationary is too strong a statement and generally not true. This statement is equivalent to the statement that all quotes have integration I(1), which is not true: on different timeframes and windows the same EURUSD may have a higher integration and the most unpleasant thing - the fractional one(Hurst index!= 0.5).
4. In fact you offer a kind of a surrogate for pair trading: because of stationarity there is always a return to the average. Here is the EURUSD incremental chart with a window of 50. This series is stationary (the probability of a unit root is 0).
And here is the tabulation of this chart:
Tabulation of D_EUR
Date: 10/13/12 Time: 14:17
Sample (adjusted): 2 50
Included observations: 49 after adjustments
Number of categories: 6
Cumulative Cumulative
Value Count Percent Count Percent
[-0.006, -0.004] 1 2.04 1 2.04
[-0.004, -0.002) 3 6.12 4 8.16
[-0.002, 0) 18 36.73 22 44.90
[0, 0.002] 21 42.86 43 87.76
[0.002, 0.004] 5 10.20 48 97.96
[0.004, 0.006] 1 2.04 49 100.00
Total 49 100.00 49 100.00
We see that the deviations below and above 20 pips (let's assume that these are spreads and stops) are 8% and 12%. This suggests the following TS: if the rise deviates by more than 20 pips, let us enter the position. We can wait, because the sample on the history is stationary.
I have such a TS - it does not work because the increments are not stationary in reality.
The point is not clear... Does this mean that by isolating "only one price" from the candle, you can't make calculations? After all, "with a partial loss" is "waving". Or is that not what you mean? Moreover, the straight lines through the "centre of the candlestick's body" are very strong factors for price testing. They are as significant as the extrema. And in the case of a line built on the "doj-dodge" where "three in one" but "one price" is concentrated, it has even "the highest priority". It's to say that highlighting "one price on a candle" is ignoring (losing) the data.
Of course it can. Otherwise why the bar and other representation at all. I was talking about forecasting like "*ussian" - you can easily replace the asterisk. It's tempting to invent a formal language and search for patterns in that way.
Who is thinking clearly... ;)
Thank you unknown (or guided?) friend.
My name is Nikolai. Very pleased to meet you. I have read your work, I bow to you. This model really works, predictions come true. Thank you.
Slave or not, I don't know, more likely not than not. I have been in close contact with Vadimcha for over two years, I know where the feet of his models come from. I know that he was in close contact with multipoint. Yours and his models are the only thing that works 100% in the market.
What prompted me to write this article was a discrepancy between the declared approach and its further implementation. An attempt to involve information theory in the analysis without understanding its methodology and capabilities. If using TI, there is only one approach - to investigate patterns and their probabilities of occurrence. (For anyone interested, read about synchronous signal reception. The bottom line is simple - multiplication of received by the link and the reference signal will give 100% recognition of the original signal under one condition - the probability of matching must be different from 50%, ie with a probability of 1% to49% the signal will be recognized as well as with a probability of matching 51%-100%. Interesting, isn't it?)
Thank you, Nikolai!
Just let's refrain from making a hundred and fifty percent statements. We all have things to work on and strive for;).
P.S. Vadim says hello.
Yes, if only we had a reference signal!
Obtaining it is no less of a challenge than deciding on the intended type of signal itself.(
I was prompted to write by the discrepancy between the declared approach and the subsequent implementation. The attempt to involve information theory in the analysis without understanding its methodology and possibilities. If we use TI, there is only one approach - to investigate patterns and their probabilities of occurrence. (For anyone interested, read about synchronous signal reception. The bottom line is simple - multiplication of received by the link and the reference signal will give 100% recognition of the original signal under one condition - the probability of matching must be different from 50%, ie with a probability of 1% to49% the signal will be recognized as well as with a probability of matching 51%-100%. Interesting, isn't it?)
how does this apply to the market?