Machine learning in trading: theory, models, practice and algo-trading - page 2181
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
The notion of a trend change alone is not enough. At least it didn't work for me. 5-7 states. I hope to reduce, but at least I have to increase the number of state parameters, which complicates things a lot.
A sharp change in trend is not the same as a smooth one, etc. Basically, the number of types of possible deviations of price behavior from the SB is infinite. If we take all possible variants of change of one kind by another, we will get infinity squared).
In my opinion, a working model cannot have a large number of parameters. So it cannot describe the price as a whole, but only some of its small pieces or individual aspects. Another thing is that such simple partial models can be obtained by simplifying more complex and all-encompassing ones.
A sharp trend change is not the same as a smooth one, etc. In principle, the number of types of possible deviations of price behavior from the SB is infinite. If we take all possible variants of change of one kind by another, we will get infinity squared).
In my opinion, a working model cannot have a large number of parameters. So it cannot describe the price as a whole, but only some of its small pieces or individual aspects. Another thing is that such simple partial models may be obtained by simplifying more complex and all-embracing ones.
I stopped at 3 types of trends (a flat is a trend with zero speed), narrowing and widening of the channel, and a fence, when the width of bars is higher than the average width of the high-low and the channel edges are not correlated with each other and are not constant. Depending on the previous state and the current algorithm for determining the appearance of signal points. They may or may not appear.
sys signal points, change of state points.It doesn't draw well. It's better to just draw in piecemeal lines.
It doesn't draw well. It's better just with a piecewise line.
Reminds me of some kind of Ishimoku.
I already answered - a certain Demco was forming equal-tick bars (100 ticks per bar according to Alpari) and worked with OPEN prices of such bars.
I had collected some ticks with 100 ticks interval (on the demo), of course there isn't enough data, but it doesn't seem to be like those pictures you've posted.
Most likely this interval should be chosen for a particular account.
If someone wants to get more data I'll attach the ticks collector for mql4
I stopped at 3 types of trend (a flat is a trend with zero speed), narrowing and widening of the channel, and a fence, when the width of the bars is higher than the average width of the high-low and the channel edges are not correlated with each other and are not constant. Depending on the previous state and the current algorithm for determining the appearance of signal points. They may or may not already be there.
sys signal points, state change points.It is a good approach for visual analysis. If you try to do it with methods of standard matstat, it will be hard to count distributions of range type (high-low) for SB with unknown variance (when it's also estimated from a sample).
I picked up some ticks with 100 ticks interval (on the demo), of course there is not enough data, but it doesn't look like those pictures you posted.
Most likely this interval should be chosen for a particular account.
If someone wants to save more data I will lay out the tick collector for mql4.
With 300 ticks at 100 - it turns out that instead of 1 minute bar, there will be 3 100 ticks.
The idea is not universal. In one brokerage company one, in another one... On the fifteenth ...
For visual analysis, this is a good approach. If you try to do it with standard matstat methods, it would be hard to count distributions of values like spread (high-low) for SB with unknown variance (when it is also estimated from the sample).
Stopped at the return logic, If the value is outside the corridor, then dial the mean from it, if the mean has changed, then the change is significant, if the return to previous values, then the outlier is eliminated. On velocities, on trend changes, it's fine (more or less visible and accurate). on more complex patterns... working on it. I estimate variance as a ratio of average differences of minima of maxima to average high-lows or opener closes on a segment.
I don't see any way to determine the point of variation yet.
I picked up some ticks with an interval of 100 ticks (on the demo), of course there is not much data, but it doesn't look like those pictures you posted.
Most likely this interval should be chosen for a particular account.
If someone would like to save more data I will lay out the tick collector for mql4.
Well, I don't know...
Here, specifically:
Data format: Time; Open; High; Low; Close; Real VolumeThe data was converted from real Dukascopy ticks for two years, from the beginning of 2016 to the end of 2017
Sliced bars with 100 ticks per bar. Bar timings were taken from the timing of the first tick, unfortunately without microseconds because the format of time storage in MT does not allow to display microseconds on the chart.
If you check it, everything is correct, I got stationarity and bimodality.
Perhaps, there is something I have not said. Well, never mind.
Alpari real gives out 300 ticks a minute or so - it combines them from 3 quote/liquidity providers. Demo they have many times less. Other DCs will give a different number too.
With 300 ticks for 100 - it turns out that instead of 1 minute bar will be 3 hundred ticks.
And in general I think the idea is not universal. One day one day, another day another day,... On the fifteenth ...
I support, filtering (conversion) liquidity DB / DC can change at any time! If we even compare M1 data before there was a huge difference between DB/DC, even the difference between M1 of one DC on MT4 and MT5 terminals!