Machine learning in trading: theory, models, practice and algo-trading - page 2886
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 agree with everything.
from personal experience:
I tried, a long time ago, to predict a step ahead of the GCH, it turned out with results that something like 60%. took just the direction of increments for the last n samples. in fact, hacked the algorithm of the GCH.
I did the same thing innumerable times with time series - the total result is a little more than 50%. but this does not mean that there is no information here, there is, but practically no information between neighbouring bars. but if you take bars not in a row, but only those that satisfy "some conditions" - the results are already better than when predicting the GCH. Time series are very heterogeneous in the availability of information by ordinal counts, but this still cannot be called non-stationarity, and I don't know how to call it more correctly.
ss. I don't want to teach anyone, just thinking out loud.
Each bar has an internal structure. If the structures coincide, it may be some condition.
A bar can be considered as a complex structure.
The example shows 5 minute bars in 1 hour. The hours are not sliced by the beginning of the hour, but I think the point is clear, that there is a difference between looking at a bare hour bar or a structural one.
You still holding the Jew's sel?
1) Of course it is important, which does not negate the importance of increments. It is the increment that is traded.
2) There is a big difference between "I see levels with my eyes" and "analysis has shown the presence of levels and their importance for trading".
3) The price is not SB, but the important question is "in what sense and to what extent is the price not SB now?". Different variants of differences from the SB can lead to opposite ways of trading them.
4) If we remember the ancient Greek founders of atomism, they could have atoms of any size, even bigger than a planet (the main thing was indivisibility, not smallness). It is the same in the market - a participant-state, for example, can outweigh all other market participants.
Econophysics tries to adapt to markets the ideas of statistical physics, which studies matter as consisting of atoms. It looks interesting, but so far without much results.
5) A market is deterministic only with respect to all the information that determines it. This information is never fully available to anyone, so there is always uncertainty. There are currently two ways of modelling uncertainty: probability theory and game theory.
Makes sense) Regarding the 5th bullet point. NVIDIA has predicted the creation of a computerised twin of a factory, with people and equipment.)))))) To test and analyse manufacturing processes.)) I once expressed such an idea about game modelling. I read it in the feed of the merchant now.)))))))
You still holding the Jew's sel?
They're down.
They're hanging.
Do you have a goal? Or are you waiting for a pattern?
Do you always have such high quality entries? I'm surprised.
1) in order to trade something, it must first be properly analysed, increments are not suitable for analysis, because knowledge of past prices is lost.
1. The increments can be taken from the point of creation of the asset and then the absolute component will be accounted for.
2. The result can be seen visually, and the reasons can only be surmised, and possibly compared with the real, with FA, but it is far away. And just fixing the results gives little food for analysis.
3. These are just different approaches, one does not exclude the other. Comparing with SB and looking for complex movements on points separated in time is another approach.
4. Of course, we need to simplify by dividing into groups of identical participants and modelling the behaviour in each group. At the same time identical participants are not necessarily identical in their actions. This is also a difficult task.
5. Close to behavioural modelling. Many factors, many variants of results).
Do you have a goal? Or are you waiting for a pattern?
Do you always have such high quality inputs? I'm just surprised.
The markets are a mess. It's the beginning of the year. After the 10th I will look at the behaviour of the markets. Still waiting.
1. The increments can be taken from the point of creation of the asset, in which case the absolute component will be taken into account.
2. The result can be seen visually, and the reasons can only be surmised, and possibly compared with the real, with FA, but it is far away. And just fixing the results gives little food for analysis.
3. These are just different approaches, one does not exclude the other. Comparing with SB and looking for complex movements on points separated in time is another approach.
4. Of course, it is necessary to simplify by dividing into groups of identical participants and modelling behaviour in each group. At the same time, identical participants are not necessarily identical in their actions. This is also a difficult task.
5. Close to behavioural modelling. Many factors, many variants of results).
1. returns kill absolute price values
2. hypothesis - experiment - fixation of results - analysis - conclusion . Scientific method
3. xz
4. who said it would be easy?
5. still xz )
1. retourns kill absolute price values
2. hypothesis - experiment - recording of results - analysis - conclusion . Scientific method
3. huzzah
4. Who said it would be easy?
5. I still don't know.)
I can't understand what is the difference between increments and absolute price difference in the window. Besides, you can train not only on increments, you can train on relative changes from the absolute price, or on logarithmic changes also from the absolute price)))))
it is possible on relative changes from the absolute price, or on logarithmic changes from the absolute price too)))))
Doesn't make any difference, with transformations, the sort order of elements will not change, i.e. splits in trees will be in the same places. If you use neural networks, it should be the same, but I'm not sure.....
PS. It won't. Firstly, everything is scaled to the range 0...1. Secondly, if you logarithmise any series, the order will not change, but weights and offsets are used there. After logarithmisation of a series with the same weights and offsets it will have a different effect (perhaps by orders of magnitude). But this is more of a disadvantage of neural networks than a plus.