From theory to practice - page 470
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
There is, unfortunately, no way to make predictions. In essence, it is similar to the definition of a trend change in classical thechanalysis. Only some kind of statistical model is added, within which you can count confidence intervals, probabilities, etc.
What is the theory of random spaces, what do you mean? In MO I use transformations of feature spaces to find the best separability of points
When TS starts to crash it's understandable that the market is broken )
didn't you give me that link?http://www.mathnet.ru/links/8df13a62a9bf8f035cadc19d1d7ad057/at1472.pdf
can't remember where it's from, waiting for it to be wailed on )
What is the theory of random spaces, what do you mean? In MO I use transformations of feature spaces to find the best separability of points
And when the TS starts to break down it's clear that there is a discontinuity in the market.)I beg your pardon, pr-in is processes )
Probably discontinuity, but maybe too much risk in the trades. You have to consider the probability.
3) switch from non-stationary to stationary time series and use existing time series analysis or neural networks - specific price values are not important for trading, it is only important to know the trend of further movement
Unfortunately it is not always possible. A typical example is a sharp change of trend (top/trough)
Unfortunately, this is not always possible. A typical example is a sharp change in trend (top/trough)
That is correct, that is why I have remembered about transitions to stationary series. No one forbids considering a price as a set of increments - a positive increment has become a negative one
I now want to decompose the ZigZag by this principle, by the increments of the ZigZag angle
But if you analyze the price movement, there is nothing better than the standard indicators; their formulas were developed by nobody knows when and by whom, but they operate better than the classical mathematical apparatus.
3) switch from non-stationary to stationary time series and use the existing mathematical apparatus for time series analysis or neural networks - the specific price values are not important for trading, it is only important to know the trend of further movement
You're thinking right, my friend!
But it's a very, very hard way... I've been working in this direction for a year now... The result, so far, is this:
On the left chart we see the sum of tick quotes increments in the sliding window = 24 hours for the 30th order Erlang flow (averaging 1 every 30 seconds)
The right one shows a graph of ACF values averaged over the set in that window.
The non-stationarity of the process is:
1. the uneven volume of tick quotes during the day - it is different, during night and day hours. Therefore, the only true solution is to work in a sliding window = 24 hours. Only in this case we have a pseudo-Poisson flow of tick quotes.
2. I thought 1) was the only reason for non-stationarity... Wrong... It happens... The second reason of nonstationarity - existence of "memory" of the process, i.e. when ACF >0 on the right chart. As you can see it is at those moments when process dispersion begins to increase sharply (red-blue lines on the left chart). Especially, it is visible for EURJPY, isn't it?
I VERY much expect to "smooth out" the variance by switching to working with higher order Erlang flows.
I'm still on my way.... On the road that leads to the Grail...
I hope you can do it faster...
For now, everyone - gentlemen. I will be away from the forum for the next few days.
Good luck everyone!
The non-stationarity of the process is:
1. the uneven volume of tick quotes throughout the day - it is different during night and day hours. Therefore, the only true solution is to work in a sliding window = 24 hours. Only in this case we have a pseudo-Poisson flow of tick quotes.
Either equitic bars, I've written many times before. Actually equitical ones are not enough; there should be more ticks per bar at night, less at daytime, and only one at all during the news.
The second reason of non-stationarity is presence of "memory" of the process, i.e. when ACF >0 on the right chart.
The only "reason" for non-stationarity is the pricing principle itself, price is an integrated series, i.e. I(1) and not I(0).
Stationary (to a rough approximation) would be the derivative of the price series, i.e. the first differences.
Either equal-tick bars, as I've written many times before. In fact, equal ticks are not enough, at night there should be more ticks per bar, during the day less, and on the news none at all.
I don't analyze ticks now, I used to make my history by ticks density, if there are few ticks I would write them in one bar, if there are many ticks then many bars would be formed, it would result in few bars at night, about one bar in 15-30 minutes, while during the day I would record up to 3 bars in a minute on ticks on the news
i can't say that such a conversion is of much value, anyway the market entry analysis even on М1 has a profit around the spread, i currently analyze charts from М5
You're thinking right my friend!
I hope you'll do it faster ...
I don't think so, no one does, all that can work is a trading strategy with high expectation of winning, and if a continuous gain is greater than a continuous loss by several times - definitely not going to work, only in the tester
How to explain the problem, which everyone understands, but is afraid to say out loud for themselves... I like simple examples: here's a pack of cards, here we are playing poker, can you find a way to play so that you always win? - Even if you take a poker beginner as your opponent, there will still be a time when you don't have a good card and you have to lose. So it is with market strategies, and please note that there are no beginners in the market, only professionals - we try to trade based on their past actions - the price moves on the chart. This is not an illusion that one can sit in front of a PC and find a continuous profit on the chart, it cannot and should not happen. Alas, money management is more important than the strategy itself, missed profit is also bad, big losses after a series of small wins are also bad .....
Alas, everything has to be calculated and it's not about the probability of the future price direction, but the probability of taking a big profit or a big loss.
Alexander_K2 all the same you have to transfer your calculations to MT, your analysis in a third-party program will not let you calculate the probability of winning
The only "reason" for non-stationarity is the pricing principle itself, price is an integrated series, i.e. I(1), not I(0).
Stationary (to a rough approximation) would be the derivative of the price series, i.e. the first differences.
It is perfectly correct to say that it is correct to speak of stationarity of increments. It's just that on this forum the accuracy of wording somehow gets lost over time)
But my point is that prices are essentially non-stationary. That is, no algorithmic transformations (taking differences of any order, replacing a variable, etc.) can bring it to a stationary series over a sufficiently long time interval. The same tops/troughs are a good example of that. But it is possible to approximate it with some accuracy by piecewise processes with stationary increments. To do this, one can use methods of the decomposition problem.
Quite right that it is correct to speak of the stationarity of the increments. It's just that in this forum the accuracy of wording somehow gets lost over time)
But my point is that prices are essentially non-stationary. That is, no algorithmic transformations (taking differences of any order, replacing a variable, etc.) can bring it to a stationary series over a sufficiently long time interval. The same tops/troughs are a good example of that. But it is possible to approximate it with some accuracy by piecewise processes with stationary increments. You can use methods of the decay problem to do this.
Why can't it be done...? :) clickable
What's wrong with that...? :) clickable
Are these differences? If so, how do you count them at the edges?