Whether there is a process whose analysis of one part does not allow predicting the next part. - page 16
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
That's exactly what I meant by pattern, not Head-Shoulders, flags or triangles. Although your definition of a pattern already contains the repeatability property, I would still add that the same set of events (= pattern) must be present several times in the history, and its number must be many times greater than the randomness. For example, 5 tails in a row has probability 0.5*0.5*0.5*0.5*0.5=0.03125. That is, with 800 flips, 5 tails in a row will happen 25 times on average. But if they happen 100 or 200 times, we have a "pattern" and not a random set of events. There are many patterns all around us. And our brain is designed so that it recognizes patterns or structure in data quite quickly (I could go on and on about that). But the structure of price quotes is difficult to detect, even with a "suited for the task" brain. I've been struggling for a month with the algorithm of detecting patterns in price quotes. The same algorithm is very fast at identifying recurring patterns in any other type of information (images, speech) - probably because there is less noise and more structure. But in quotes there seems to be little structure and a lot of noise. But I will continue the battle. There is still hope, and the topic is interesting.
Great post, couldn't help noting it for myself!
about " But there seems to be very little structure (repeating statistically important patterns) and a lot of noise in the quotes.", or maybe there is no information in the quotes?
Hypothetically there is a pouring strategy at a constant lot, i.e. pips MO is positive, but at a spread of 0.
1. Is it possible to select such an MM (and Martin's derivatives as well) that the system would also fill up at a non-0 spread, and what would such an MM depend on?
2. What would the formula look like, by which we may calculate the maximum value of spread, at which the system would not be able to fill?
1 - You cannot, if the trades are independent.
2 - this is done by simulations.
If you do not like the spread - trade limit orders without it. In this case there is a problem related to variable "fill rate" - when the direction is correct - a small volume will be executed; when wrong - the whole volume.
If profitable trades happen, then it is theoretically possible to pull the system into profit by multiplying the lot.
There are such considerations.
If we believe that the system may be pulled out to a larger payoff with the help of a betting system, then we automatically must accept the statement that there are unrecorded regularities in the system. Consequently they have to be caught somehow. There may be two options here.
1. If the sequence of transaction results (detrended by the payoff value) is not, so to speak, white noise, i.e. if there are correlations between transaction results. In this case, we need to find these correlations and use them (see below)
2. If there are no correlations in the sequence of deals, then it would be good to look for correlations with behavior of price prior to deals or with other factors, at least with the time of day (filters). Well, the field of imagination is unlimited, in fact, we can create a new TS.
(It is necessary to take into account that the found correlations must be checked, i.e. the training and test sets must be distinguished in the set of deals).
In both cases, the result of the work must be a correspondence between the initial condition and the shift of the MO of the concrete trade. And then the linear programming comes into effect. Total MO of the system is calculated as MO = Sum(MO_at_those_conditions_i* probability of conditions_i* lot_i), this MO must be maximized by selection of parameters - lots for each variant. Restrictions - maximal risk size or maximal average lot per entry, here too one can fantasize.
If profitable trades happen, then it is theoretically possible to pull the system into profit by multiplying the lot.
I guess if such an obvious fact hasn't caught on, then you can't!
You can think as much as you like, it won't make 2x2 into 5.
Great post, couldn't help noting it for myself!
About " But there seems to be very little structure (repeating statistically significant patterns) and a lot of noise in the quotes. ", or maybe there is no information in the quotes?
I've given up on patterns so far, at least on higher timeframes (H1 and larger). It seems to me more and more that trading on H1 and above is essentially guessing the direction of the news. If consistent patterns exist, it's only intra-hours, on M1-M5 timeframes, that is, patterns of traders' reactions to news that has already come out. That's where you have to dig. After spending a lot of time with higher mathematics in Forex (complicated formulas, regression, Fourier, neural networks, etc.) I have come to disappointment: it is not suitable for Forex. Pipsing with simple tools is much easier and gives more reliable results.
There are such considerations.
If we believe that the system may be pulled out to a larger payoff with the help of a betting system, then we automatically must accept the statement that there are unrecorded regularities in the system. Hence they have to be found somehow. There may be two options here.
1. If the sequence of transaction results (detrended by the payoff value) is not, so to speak, white noise, i.e. if there are correlations between transaction results. In this case, we need to find these correlations and use them (see below)
2. If there are no correlations in the sequence of deals, then it would be good to look for correlations with behavior of price prior to deals or with other factors, at least with the time of day (in other words, filters). Well, the field of imagination is unlimited, in fact, we can create a new TS.
(It is necessary to take into account that the found correlations must be checked, i.e. the training and test sets must be distinguished in the set of deals).
In both cases, the result of the work must be a correspondence between the initial condition and the shift of the MO of the concrete trade. And then the linear programming comes into effect. Total MO of the system is calculated as MO = Sum(MO_at_those_conditions_i* probability of conditions_i* lot_i), this MO must be maximized by selection of parameters - lots for each variant. Restrictions - maximal risk size or maximal average lot per entry, here too one can fantasize.