Machine learning in trading: theory, models, practice and algo-trading - page 2204
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 idea of looking for a rebound/reversal is interesting.
1) You make a forecast ahead. How do you trade it? Place orders? Like I said earlier,
2) the further forward the forecast is, the closer it is to random.
1) wait for confirmation that the level has triggered, I described it in fyle
2) If you imagine that the rebound is a cluster of stop orders or takeovers, it does not work here, orders may stand for a long time, there is no linear decreasing dependence on time reliability...
2) Instead of your normalization, you can just do a price delta from Y (for tree-like models, you don't have to normalize it). The result will be the same.
You could, but it would be rougher, and the result according to my research would be much worse, normalization kills the difference in volatility, and that's a good thing
3) I think you can't pick individual rules out of the forest. You have to take exactly the average. Otherwise you get a fit. Maxim and I discussed this in the thread on his last article. He did some experiments and confirmed that averaged predictions are worse on test/train, but better on new data (exams). Read the discussion there.
You and Max didn't do anything with the levels, it's the accuracy of the value that matters here, not the generalization, I explained it in the file, that it should be so, and why exactly so.
2) if to imagine that the rebound is some accumulation of orders of stops or takeovers, then it does not work here because the orders can stand for a long time, there is no linear decreasing dependence on the reliability of time.
I agree, the orders will stand for a long time.
But if a strong event happened between your pattern and the point "Y" and the price went far down, stops of many Buy trades triggered and TPs of Sell trades triggered. As a result our expected TPs and SLs at some level have strongly decreased, even if by half. The remaining TP may not be enough to reverse the price and create a rebound.
So it's better to look at the price behavior on the whole area from your pattern to the Y. The probability of a bounce is higher if the price hasn't moved down strongly.
You can, but it will be more rough and the result according to my research will be much worse, normalization kills the difference in volatility and this is good
Volatility is not important? Do you think that a pattern with a 15 pt swing has the same consequences as the same pattern with a 150 pt swing? I don't.
You and Max didn't do anything with the levels, it's the accuracy of the value that matters here, not the generalization, I explained it in the file, that it should be so, and why it should be so.
It will be a fit for 1 random result. Hopefully experiments on the exam sample will change your mind.
But, if a strong event occurred between your pattern and the Y-point and the price went far down, many Buy trades had their stops triggered and Sell trades had their TPs triggered. As a result our expected TPs and SLs at some level have strongly decreased, even if by half. The remaining TP may not be enough to reverse the price and create a rebound.
So it's better to watch the price behavior on the whole area from your pattern to the Y. I.e. the rebound probability is higher if the price hasn't moved down strongly.
The confirmation is trying to solve this problem, if it is there, it means that someone else is sitting there.
Volatility is not important? Do you think a pattern with a 15 pt swing has the same consequences as the same pattern with a 150 pt swing? I don't.
Don't give me absurd examples.
if you have a pattern with a 10 candlestick size, the spread won't be large, but due to the different volatility the model will consider even identical patterns different!
You'll need to increase the number of trees to describe the same data volume... You'll increase the "curse of dimensionality" effect of market data even more
It's going to be a fit for 1 random result. Hopefully experiments on an exam sample will change your mind.
Don't compare your experience in classifying ZZ by stochastics , with regression to predict accurate values...
argue less, ask more, I know exactly what I do, and why I do it this way and not that way.... And believe me, how else I know...
Let's not give absurd examples.
This is a normal example. If you approximate it, the model will see them as exactly the same.
Discretization of predictors into 10-20 plots or rough quantization, not into 255 quanta, but into 10-20 would rather help here.
And though... the forest can figure it out on its own, even without that.
You'd have to increase the number of trees to describe the same amount of data...
50-100 is enough. The number only for a good averaging is needed.
To account for different patterns, you need to increase the depth of the tree.
Well try it, I'm all for it!
This comrade has been gnawing at the crumbs that have fallen from the table of the smart man (Semen Semenych's cluster analysis) for more than a decade. In fact, these clusters are just an attempt to "invent" risk-neutral portfolios for currencies. That is, just another homemade bicycle that won't ride because the markets aren't stationary. So, it quite logically follows the intention of this character to make money by selling fakes for the sake of which the whole forum is covered with all kinds of trash. The administration does not prevent it for obvious reasons.
I looked up history... And indeed. It turns out that this mammoth here is so bad already in terms of mental health. And the other one is already tired to sink his marmosets in cent accounts, which are masquerading as dollar accounts in that brokerage house.
You're a weakling, Mahim. You're back. You gave your word and you took yours back. You don't owe anyone))))
Rinat has a big deposit and he is not afraid.
You won't get anything unless you start building yours from scratch.
Take off your glasses and see that MO is nothing more than an automatic fit with history.
And you have to look one step ahead to make money.
Even a fool will last long with a big depo, but to grow a big one from a small one is an art.
Good luck, hero of your word.
I looked through all the potential MoM futures contracts on Moex and made this list for myself
They have a history from 2014 and transactions take place regularly.
And this is what I discarded
I guess I'll have to look at the stock - not many tools...Ten years is a long time... Gentlemen, maybe we should already start thinking whether it is worth climbing this Everest. Of course nobody will suffer from hypoxia, but those who have gone cuckoo already begin to separate smoothly from the main group. I, probably, will wait for a while at a neutral altitude. I'm scared.
Don't be afraid)) they are always near and everywhere. You can't see them from every angle... that's good, not so scary))))