Whether there is a process whose analysis of one part does not allow predicting the next part. - page 14
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
And I don't have any future quotes, and I don't think you do either.
But I can offer the grid as a test quotes from the past that the grid doesn't know. Look at the dates.
Sample 2011.10.01-2012.03.01 8890 examples, Forward 2012.03.01-2012.05.25 4828 examples.
but the market could do worse:
a) the spread, about which the tongue is already muzzled) but even if it is not, then
b) as far as I understand it, the direction of bars of smaller spreads is better predicted than the direction of sharp movements. Hence the moral: even a direction prediction with a positive MO can give a negative result on points if we are more successful in predicting the direction of small bars and less successful in predicting the direction of large ones. I.e. a bad forecast of outflows can literally eat up excellent results on a flat. Apparently, we need a clear work with stops, and "smart" stops, stupidly put a stop at one level, probably, will not work...
alsu:
1. straight 18...
but on the market it could get worse:
1. a) the spread, about which the tongue is already muzzled) but even if it does not exist, then
3. b) as far as I understand it, the direction of bars of smaller spreads is better predicted than the direction of sharp movements. Hence the moral: even a direction prediction with positive MO can give a negative result on points if we are more successful in predicting the direction of small bars and less successful in predicting the direction of large ones. I.e. a bad forecast of outflows can literally eat up excellent results on a flat. Apparently, we need a clear work with stops, and "smart" stops are needed here. Stops at one level probably will not work...
1. Well, it certainly is not a grail, the market is constantly changing (today patterns that worked a few days ago do not work). Requires training at least once a day.
2. Spread is taken into account. The grid has been trained not to react to movements less than 2 maximum spreads for an instrument.
3. there is a letter. The direction of bars of a smaller spread is predicted better (partly because there are more of them in Sample, and partly because they contain the main information about the process, imho) than the direction of sharp movements, and the resulting direction of several bars at once is even better. The colour of particular bars is very poorly predicted.
In general I think to put SL for 3sco of the possible movement size of the predicted area. We perform training without SL and TP, and trade with SL but without TP, in this way the main possible loss from fat tails will be smaller (it would be anyway if not correctly recognised abnormally big movement colour). This way one can additionally increase the MO to what the grid can give.
In general, purely hypothetically, although there are already some developments, predicting process system should have two states - "know" and "don't know". In the "know" state the system makes a prediction. In the 'don't know' state it refrains from making a prediction, either the system does not know the current state of the process or the system knows that in this case it is 'better to refrain' from making a prediction. Over time, if the process changes its characteristics and internal relationships, the system is more and more often in a state of "don't know" and finally stops, in fact, forecasting being in a stable-constant state of "don't know". Such a system is valuable in any case - it is enough to tune/learn it once and you can "forget" about its existence, because the worst that can happen is that the system goes from "don't know" state.
All is well and good, but there is a BUT. Pattern reversals occur in financial markets, when one and the same pattern of the current state of an instrument becomes the cause of reversed effects, earlier one had to buy - and now one has to sell in such cases. Hence, it is necessary to train the system continuously to keep up-to-date with the knowledge of such recent reversals of causal patterns.
As far as I know, all modern forecasting systems are focused on a constant state of "know", so the slightest change in the characteristics and internal connections of the process leads to erroneous forecasts. This is expressed in reduced profitability of systems outside the Sample area.
Opinions please, colleagues.
I have a purely theoretical 2*sqrt(2) RMS ))
This ratio is obtained when the likelihood ratio of the Laplace and Gaussian distributions passes through a critical point and makes sharp outliers towards Laplace, i.e. just thick tails. The problem is in calculating the RMS prediction, but here too the grid can be used, just remove the diurnal seasonality.
In general, purely hypothetically, although there are already some developments, the system that predicts the process should have two states - "I know" and "I don't know". In the 'know' state the system makes a forecast. In the 'don't know' state it refrains from making a prediction, either the system does not know the current state of the process or the system knows that in this case it is 'better to refrain' from making a prediction. Over time, if the process changes its characteristics and internal relationships, the system is more and more often in a state of "don't know" and finally stops, in fact, forecasting being in a stable-constant state of "don't know". Such a system is valuable in any case - it is enough to tune/learn it once and you can "forget" about its existence, because the worst that can happen is that the system goes from "don't know" state.
All is well and good, but there is a BUT. Pattern reversals occur in financial markets, when one and the same pattern of the current state of an instrument becomes the cause of reversed effects, previously one had to buy - and now one has to sell in such cases. Hence, it is necessary to train the system continuously to keep up-to-date with the knowledge of such recent reversals of causal patterns.
As far as I know, all modern forecasting systems are focused on a constant state of "know", so the slightest change in the characteristics and internal connections of the process leads to erroneous forecasts. This is expressed in reduced profitability of systems outside the Sample area.
Opinions please, colleagues.
When we say that the same pattern now gives the opposite signal and it is necessary to retrain the system, are we not engaging in self-deception? Maybe there was no pattern linking this pattern to the signal? For example, we flip a coin and notice that after three tails most times heads appear. Is it a regularity or because of the lack of a large number of experiments (which allows evaluating statistics more accurately) we've come to the wrong conclusion? I've been torturing myself with patterns for a long time now and am always thinking about this question.
By the way, what is the depth of price history that allows to estimate the market condition?