From theory to practice - page 340
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
If the signal is stationary and periodic, there is no point in killing it, as it is a grail, adjusted for residual noise
if the signal is stationary noise - even better, no need for any NS at all
What kind of signal it actually is - the author is not interested at all, his aim is to extract the noise component from it, hoping that the noise will accurately predict the next values of the signal. )))
What the hell is this nonsense? Such rows are easily predictable
By the way, here is also a review article on BP forecasting that recommends isolating and analysing the noise component after removing trend, cyclical and other signal components...
https://pokrovka11.files.wordpress.com/2011/12/emetrix_time_series.pdf
What is the common sense, it is not clear...
Turning BP into RNG will make it impossible for you to earn anything at all, in principle).
You contradict yourself,"You can predict mean and variance for RNG (stationary value), for this purpose study distributions" You have answered the question yourself, while the variance and mean will be constants.
A worthy conclusion to the transition from theory to practice: no signal, no sane model, no pamm account. They even called us dowsers)).
Waiting for Alexander III - he's not going to fail)
You contradict yourself,"For a RNG (stationary value) you can predict the mean and variance, for that you study distributions" You have answered the question yourself, while the variance and the mean are constants.
There is no contradiction.
For increments on the CLO, predicting the mean increment is possible, it is a constant equal to 0. For a series of prices by LFO (integral of a series of increments, aka "random walk") this means a 50/50 prediction of the direction of the trade, and there is no possibility of earning. The prediction of the nearest current increment (which A_K2 wrote) is also 0 plus minus a few RMS, which is of no practical use.
But the GSF increments are a simplified price model, a basic one, a zero approximation.
Unlike GSF, in real price increments there is a dependence of the next price on the previous one, i.e. there are situations when the transaction direction forecast differs significantly from 50/50. Your task is to find such situations in price increments, and by definition they do not exist in RNG.
Faced with blatant downism, in the form of the claim that you cannot predict series of random numbers with Erlang distribution, I am forced to leave the forum forever.
Alexander, personally, for example, I am trying to save you the cost of many years of research time by drawing your attention to deliberately false paths, which I have been down even before you knew about them.
So by "leaving the forum forever" you are only doing yourself a disservice.
Waiting for Alexander III :)
There is no contradiction.
For LFO increments, predicting the average increment is possible, it is a constant of 0. For a series of prices on the LFO (integral of a series of increments, aka "random walk") this means a 50/50 prediction of the direction of the trade, and there is no possibility of earning. The prediction of the nearest current increment (which A_K2 wrote) is also 0 plus minus a few RMS, which is of no practical use.
But the GSF increments are a simplified price model, a basic one, a zero approximation.
Unlike GSF, in real price increments there is a dependence of the next price on the previous one, i.e. there are situations when the transaction direction forecast differs significantly from 50/50. Your task is to find such situations in price increments, and by definition they do not exist in the RNG.
The probability that the deviation in absolute value under the normal distribution will be less than 3*SCO is 0.9973. In other words, the probability that the absolute value of the deviation will exceed triple the RMS is very small and is 0.0027=1-0.9973. This means that in only 0.27% of cases it is likely to occur.
In practice: if the distribution of the random variable under study is unknown, but the condition holds, then the variable under study is normally distributed, otherwise it is not normally distributed, which is what BP proves, but if we achieve normality from BP in consequence of some non-linear transformation, then why not use this rule?
The probability that the deviation in absolute value under a normal distribution would be less than 3*SCO is 0.9973.
Not the deviation, but one increment. But A_K2 does not trade a single increment, it trades just the deviation from the SMA, which consists of many consecutive increments. For these deviations we have to construct our own distribution and calculate our own probability. Besides, the SMA itself shifts during a trade so it is a big question whether the closing price will be in profit. Good idea is to draw a distribution of deviations over time from the entry price, and I suppose it will be much closer to the uniform than to the normal.
In short, all this spam about streams and distributions is pure scientific water without the slightest understanding of what's going on) Normality for our purposes means... well, nothing at all, except that it's normal.)
It means that only 0.27% of the time it can happen.
what if you think about it? the residuals are analysed to see if the model has selected all the information. If they are noise, it's OK. Trend and severity are killed for homoscedasticity, cycles remain for forecasting. No periodic cycles - no forecast (except for expected payoff).
In the market cycles are non-periodic, so ARIMA does not work, but try to apply GARCH for variable variance (heteroscedasticity), when the process memory cannot be completely killed, and the next volatility values depend on the previous ones
Alexander proposed a way to kill the ARCH effect (process memory, markoving, fat tails) and his narrative can in no way be called incorrect or absurd
If the market is constantly in flux: changes in amplitude, phase and frequency, a de-trend will not help, as it is all possible on the history of the process, by fitting, but the inertia present in the market is believed to give the opportunity to work out such an approach, in the short run, but when the rules change, the historical fit diverges from reality. The possibility of obtaining alternative approaches to address the non-stationarity of the process, in the form of thinning BPs by Erlang flows (Palm flows) is possible, and other alternatives are also possible.