Machine learning in trading: theory, models, practice and algo-trading - page 1455
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
Interesting topic, btw... I want to learn how to model jumps so I can subtract them from the model later
http://stuartreid.co.za/interactive-stochastic-processes/
Some gaps happen at obviously non-random points in time (session opening, for example) and it's unlikely they can be modeled by some Poisson's process.
Some gaps occur at clearly non-random points in time (opening a session, for example) and can hardly be modeled by some Poisson process.
I have not yet understood how they use it all in real life. They built some graphs, had some fun and then what? )
let's say, I want to make a normal series from a tail series, where should I add/remove these jumps to get something similar from initial one, and what would be of value in the future
I have not yet understood how they use it all in real life. They made a few graphs, looked at them, got happy, and then what? )
Let's say, I want to make a normal series from the tail series, where should I add these jumps, so that I could get something similar from the initial one, and that would be of value in the future?
A possible option is that the continuous component has a trend, and the jumps do not, but they distort the simple average of the increments greatly. Within the framework of this model, you could probably somehow try to remove their influence on the definition of the trend (evaluation of the mu parameter).
A possible option is that the continuous component has a trend and the jumps do not, but they strongly distort the simple average of the increments. Within the framework of this model, we can probably somehow try to remove their influence on the definition of the trend (estimation of the mu parameter).
"Jumps" of prices up and/or down can be determined almost exactly the same way as any other typical movements (rollback/up/rise/reverse, etc.)
At least somewhere the MO works
I run java scripts in Chrome
If the zigzag is so well predicted, let's figure out how to trade it well.
Looked again at the article https://www.mql5.com/ru/articles/1103
It opens at the beginning of a bar and closes at the beginning of the next bar and "does not consider spread, slippage and other niceties of the real market".
I put together the balance (above) and price (below):
All the growth of the balance occurred in 2 zigzag knees. Perhaps NS found this solution - if the price moves in one direction 50-100 pts, then the ZZ is prescaled in the same direction. I.e. it obtains a good lag, like in simple prices. The fact that NS was moving in one direction can be seen by coincidence of the moments of small reversals and reduction of balance at the same time.
On these two strong movements the NS has earned. In small movements, according to the same algorithm the trade is in a good minus. And how soon such a strong movement will be repeated - no one knows.
In my experiments, I tried to train the model not on zigzags, but on TP/SL. The result was 50/50%.
But I doubt whether I should try it with ZZ. It looks very much like chart trading and it is not profitable as you know.
If the zigzag is so well predicted, let's figure out how it can be well traded.
With all due respect, but you're talking nonsense, you can "predict" for example one of the chips or their linear combination, everything will "predict well", but what's the use? With ZZ the same thing, in the direction of ZZ there is a large admixture of momentum(mov) of the chips, in one form or another, it is predicted, but NOT THE FUTURE, in these 65% of all 65% is from the admixture of the past, there is not even those 2-3% that can barely scrape out of ordinary data, but cool for article or sale grail sucker. It is better to forget about ZZ.
PS I heard one algotrader about supposedly working with ZZ, which at first glance could be true, but until the SB is not tested, can not say for sure, in general, the idea was to take the target for the next point ZZ value AFTER the LAST BLOCKS. That is, if there is a series of 100 lags and for example chips is just the price itself with 5 lags for point 10: chips will be {6,7,8,9,10}, and the target ZZ at the point after the 11th is looking for a knee and the next direction of leg ZZ, so the previous direction no matter what length it is mixed with the past. Do you want to check and let me know what comes out on the SB, maybe I'll check too, if I can.