Not Mashka's business! - page 11
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 initial results may look 'fantastic', but in reality it is a 'harsh' reality. Nothing in the future is peeked at. But I have found a "flaw" in my optimiser. The situation when the system cannot find a stable MA process (i.e. sliding window length) is not handled correctly, and the window length is assigned a value of zero (I didn't think such situations were possible). If the window is 0, the result is not the MA, but the price series itself. Such an area appeared on the investigated range:
Let me remind you: EURUSD, hours,(H+L)/2, we searched for a stable MA in the range of sliding window from 24 to 600.
Forecast results (first 500 samples)
Forecast one bar forward
Error graph between actual value and forecasted value. You can see the segment where the system fails to find a correct solution and assigns the moving window length zero (the forecast should not be generated):
At this very section there is a "chattering", significant errors appear - the Burg method predicts the source - price series itself, which confirms the practice and recommendations for its use (BP should be smooth)
The dynamics of the sliding window length selected by my optimizer:
Full predictive model
Error between fact and predicted value.
Minimum number of forecast samples - 2
Maximum number of forecast samples - 3
Preliminary conclusion
This really works (Burg's method and my optimizer) and can actually be used. All predicted values fall within the range of +/- 2 points (if you exclude live-row prediction, which is elementary). About eighty percent of the errors sit in the +/- 1 point range. I will finish the study andpossibly publish in a separate topic in more detail, so as not to trash the host, it is possible that someone will be interested in this approach.
Preliminary conclusion
All projected values fall within a range of +/- 2 pips
2p, not 20p? The error value on the chart is 0.002
Preliminary conclusion
All projected values fall within a range of +/- 2 pips
2p, not 20p? The graph has an error value of 0.002
A plot of 500 counts is under study. The red horizontal lines are drawn +/- 0.0001 and most of the forecast points fit into this range. The maximum dispersion (very large indeed) is observed in the area of about 50 samples where a stable MA is simply not found and the price is directly predicted. I didn't count figures (later), but by eye we can see that about 450 points fit into the range of +/- 0.0002..:
(just in case:).
It's easy to exclude such points in Expert Advisor, all you should do is not to make a forecast when optimizer returns value of sliding window length equal to "0". Just don't make predictions, wait :o)
Figuratively, your method boils down to this analogy.
You throw stones into the water and try to predict by the rise of the water level in the world's oceans which stone will be thrown in the next instant. That's instead of looking at the rocks:-)
Figuratively, your method boils down to this analogy.
You throw stones into the water and try to predict which stone will be thrown in the next instant by the rise in the water level of the world's oceans. That's instead of looking at the rocks:-)
(cheerfully so) Seryoga, you're "goofing off" or being clever, which is essentially the same thing. My optimizer and Burg work, it's practically a fact, there are still a few more samples to calculate. And what to compare it figuratively with is a matter of generalisation. It's not like I'm invoking old Freud to make a figurative comparison between your methods and models. :о)
And if one is familiar with the general model of linear prediction, one cannot understand even more about stones, water and levels. Although levels - as a conceptual notion, are very important. Here, for example, consider the level of fluctuating beer in a glass. Can you tell how predictable this process is?
You're being ironic!
Sitting on the edge of the pool in the sauna and wiggling your foot in the water, look closely at the patterns on the bottom - at first glance a jumble of light reflections. But if you're looking at the exact geometry of the pool, you'll find the position and phase of your foot with the precision of the period of the shortest wave. So, the process is predictable, but it is unclear why dilute the useful signal in a sea of noise, so that it can be reconstructed (predicted) by indirect signs?
You're being ironic!
Sitting on the edge of the pool in the sauna and wiggling your foot in the water, look closely at the patterns on the bottom - at first glance a jumble of light reflections. But if you're looking at the exact geometry of the pool, you'll find the position and phase of your foot with the precision of the period of the shortest wave. So, the process is predictable, but it is unclear why dilute the useful signal in a sea of noise, so that it can be reconstructed (predicted) by indirect signs?
I'm not being ironic, I just don't understand you, where is a useful signal and why do you need a swimming pool with legs? Anyway, I'm going on another business trip, I'll be there in a couple of weeks, so let's continue if I'm in the mood.
Please ignore the picture. It's not there - it's a glitch.
Neutron
good glitch, what does a=10 and b=0.9 + analysis window depth mean ?
I'm telling you - there's no picture! If you can see it, then you need help :-)
Seriously, it's the most usual casual recursive filter, of the form:
y[i]=a*x[i]+b*y[i-1], where a=0.1, b=0.9