From theory to practice - page 1704
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
By the way, all sorts of indicators of discord, entropy and so on only muddy the swamp and do not give any improvement
This is a strong assertion.
In 2-3 months I will either support it or refute it based on practical results. Specifically now, in my TS, I use the indicators of malfunction - kurtosis and asymmetry of increment distribution - then I will check (based on at least 100 trades on the real account) whether the results would be better/different without them.
This is a strong assertion.
In 2-3 months I will either support it or refute it based on practical results. Specifically now, in my TS, I use the indicators of mismatch - kurtosis and asymmetry of increment distribution - then I will check (based on at least 100 real trades), would the results be better/different without them.
Well, there is a martingale here, if it has already started to average, then you cannot stop it with any slippage and other things, otherwise the trade will freeze
Well there's a martingale, if it's already starting to average, it can't be stopped by any sort of malfunctions or else the trade will freeze
Hmmm... If you are going to attach a neural network there (obviously, for classifying entry points into deals), it will act as a malfunction indicator. No?
Hmmm... If you're going to stick a neural network in there (obviously to classify entry points into a trade), it will act as a malfunction indicator. No?
Yes, but only for the first trade to determine the direction, and then the averaging is still needed if you missed the direction
This is a strong assertion.
In 2-3 months I will either support it or refute it based on practical results. Specifically now, in my TS, I am using the indicators of discontinuity - kurtosis and asymmetry of increment distribution - then I will check (based on at least 100 trades on the real) whether the results would be better/different without them.
About three months ago you were running some super TS. Naturally, you launched it without monitoring, or rather you should have looked for it in somebody's wife's house. How did it end?
:))) Instead of $200 in my account, I ended up with $115. I was about to quit, but this post helped me:
Forum on trading, automated trading systems and trading strategies testing
From Theory to Practice
Vladimir, 2018.03.03 15:40
Maybe the point is that you are looking for a single "bell"? One for every time of day and day of the week. Take a look:
https://www.mql5.com/ru/forum/221552/page19#comment_6168925 activity changes during the day
https://www.mql5.com/ru/forum/193378/page49#comment_5239746 for Mondays, Tuesdays, ...Fridays
Why not make this "bell" a function of two more variables, the day and hour numbers, since you're not looking for a scalar (a number), but a distribution? Or parametrize the bell you are looking for, so that the parameters become functions (at least tabulated) of the day and hour numbers. After all, as I understand it, you don't need all of it, a description of behaviour around "heavy tails" is enough...
If this gives the real Grail, then Vladimir is a genius and I will give him my TS for nothing.
More ideas should be thrown around :-)
Where and how the price moves is not very important, but
But price often (almost always) gets into autocorrelation when there is a sharp counter-trend movement.
for the sake of fresh air :
you can open the chart and find the area where the ACF is almost 1. It is not far away and is quite specific.
Such events are infrequent, but they will ruin the distribution statistics because the "steps/bars/increments" are simply identical. It's just the same, that's how the market works.
That is, some small plots in the statistics on which everything is built will enter twice.
I decided to mess around with the martychs and bypass the CE algo, so far that's it. I'm thinking of putting some machine learning in there as well.
By the way, all sorts of indicators of degradation, entropy and so on only muddy the swamp and do not give improvements
What's wrong with that?
The simpler and fewer the filters, the more reliable and better (easier) it goes
More ideas should be thrown around :-)
Where and how the price moves is not very important, but
But price often (almost always) gets into autocorrelation when there is a sharp counter-trend movement.
for the sake of fresh air :
you can open the chart and find the area where the ACF is almost 1. It is not far away and is quite specific.
Such events are infrequent, but they will ruin the distribution statistics because the "steps/bars/increments" are simply identical. It's just the same, that's how the market works.
That is, some small areas in the statistics on which everything is based will be included twice.
In fact, it's precisely these kinds of areas that cause the histogram of increments to become overly spiky. From the point of view of conditions of the Glivenko-Kantelli theorem, the condition of independence of the sample is violated here. Positive dependence (positive correlation coefficient), obviously, leads to histogram peakedness, and negative correlation leads to flatness. In addition to this, sampling heterogeneity (violation of another condition of this theorem) may well lead to a double-top histogram.