Machine learning in trading: theory, models, practice and algo-trading - page 2992
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
Nice ride...
IMHO, econometrics to a small extent and financial stochastic mathematics to a large extent. At a high level, these sciences are intertwined with each other and with machine learning.
IMHO, econometrics to a small extent and financial stochastic mathematics to a large extent. At a high level these sciences are intertwined with each other and with machine learning.
what happens if you decompose the bar chart into principal components
above is the original, then the principal component 1, 2, 3.
the sum of the first two principal components
cleaner chart without high-frequency fluctuations, no lags ...
filtering ))
As soon as the presentation of the theory of signals becomes mathematically meaningful, stationarity (in the broad sense) and energy spectrum appear at once. And nonsense like "a signal is any set of numbers" does not appear obviously.
The article from the linked site attempts some sort of generalisation of the concept of stationarity. Regardless of how useful it is for DSP, it doesn't make much sense for prices. Once again, I will point out that the stationarity of an asset price, or the price of a portfolio of them, is a trader's eternal paradise, because it means an eternal flat on which to trade forever, such as a return to the mean.
This is the end of the discussion of COCs in this thread.
Well, if you apply filters to this, having previously checked "it" for stationarity, then maybe something will give something to someone, although I doubt it
when "it" was fed into bousting as a feature, it turned out that the greater the differentiation, the more worthless the predictions on new data.
again, you could just take a few Mashas of different periods from this and not suffer nonsense
Everything is perfect in this paper: target selection via FF and retraining at least at every step, and the most informative feature in the form of fractional-differenced series. In general, everything that has been prayed for in the topic lately :) I messed up with filters, I agree, I should have added them for a complete stuffing :)
Can I see the results before I die.
https://perraudin.info/gsp.php
Live happily ever after
Well, can you describe the outline of the idea itself....
For me, the point is always to find deviations of price from SB. Econometrics differs from financial stochastics by modelling time - discrete in the first case and continuous in the second, which leads to rather different mathematics, but the essence is the same.
Here is a fairly standard example of such a search within econometrics article1 and article2. The approach is precisely related to the search for stationarity (in the asset price or spread) - that is, stationarity is assumed to be possible only sometimes and is defined as a deviation from a more typical SB, rather than being a constant property as in the study of signals in DSP.
For stochasticity, it is difficult to give a simple but meaningful example. My paper on gaps may serve as a hint in this direction, because the distribution studied there is easier to be considered at continuous time. And if we assume the dependence of this distribution on some features, we can develop the idea in the direction of MO.
For me, the point is always to look for deviations of price from SB. Econometrics differs from financial stochastics by modelling time - discrete in the former and continuous in the latter, which leads to rather different mathematics, but the essence is the same.
Here is a fairly standard example of such a search within econometrics article1 and article2. The approach is exactly related to the search for stationarity (in the asset price or spread) - that is, stationarity is assumed to be possible only sometimes and is defined as a deviation from a more typical SB, rather than being a constant property as in the study of signals in DSP.
For stochasticity, it is difficult to give a simple but meaningful example. My paper on gaps may serve as a hint in this direction, because the distribution studied there is easier to be considered at continuous time. And if we assume the dependence of this distribution on some features, we can develop the idea in the direction of MO.
Financial markets are NOT stationary, this should be taken out of brackets, accepted as an axiom and any evidence of stationarity should be considered null and void.
In reality, based on Soviet science, financial markets are not just non-stationary, they are undefined. A random process was once called undefined if a human being is involved in shaping that random process.
The most excellent example of the uncertainty property is the application of mass service theory to underground passenger flows. All the indices that the theory of mass service gives for a random process in the form of the flow of passengers in the underground are perfectly calculated and exist within rather narrow confidence intervals. But take a balloon, pop it in the crowd and shout "terrorists" - the whole theory of mass service goes into tatters. And a stationary process turns into an undefined one with a lot of maimed and trampled people. All this we see in financial markets, when news can do anything to the market.
Yes, you can take a time period and prove stationarity on that time period. You can take another time period and prove that price transformations are stationary. But there is no instrument modelling the impact of news on the market, after which a random process can be stationary, non-stationary or chaotic. and after news the characteristics of stationarity or non-stationarity are likely to be different compared to previous periods.
Financial markets are NOT stationary, This should be taken out of brackets, accepted as an axiom and any evidence of stationarity should be considered null and void.
In reality, based on Soviet science, financial markets are not just non-stationary, they are undefined. A random process was once called undetermined if a human being takes part in the formation of this random process .
The most excellent example of the uncertainty property is the application of mass service theory to underground passenger flows. All the indices that the theory of mass service gives for a random process in the form of the flow of passengers in the underground are perfectly calculated and exist within rather narrow confidence intervals. But take a balloon, pop it in the crowd and shout "terrorists" - the whole theory of mass service goes into tatters. And a stationary process turns into an undefined one with a lot of maimed and trampled people. This is what we see in financial markets, when news can do anything to the market.
Yes, you can take a time period and prove stationarity on that time period. You can take another time period and prove that price transformations are stationary. But there is no instrument modelling the impact of news on the market, after which a random process can be stationary, non-stationary or chaotic. and after news the characteristics of stationarity or non-stationarity are likely to be different compared to previous periods.
Totally agree. Market uncertainty has no probabilistic character unlike natural processes. If we talk about mathematical models of market uncertainty, the closest will be from game theory. But this science is still too underdeveloped to provide practically useful models.
From game-theoretic models it is possible to obtain, under some additional assumptions, probabilistic models. For example, Nash equilibrium in mixed strategies. We could be interested in models describing oscillations near such equilibria, like in mechanics. But so far I have not seen a sufficiently developed matapparatus for such studies.
For me, the point is always to look for price deviations from SB.....