Our Masha! - page 7
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 trick is that any super-duper can be matched with a close counterpart from SMA or EMA...
Is it worth the trouble then?
Think, do, kill a lot of time and effort - and end up with an EMA 4 - "miracle machine"))
Well the MA that will go ahead of the market and predict - how to make it? .....))))
Make a time machine, fly to the future. Come back to our time with quotes from the future for about 10 years in advance))), and then you can use these data to draw a "MA that goes before".
Any other options?...))))
This is kind of embarrassing, but I have to report that on EURUSD M1 Ideal_MA 0.02 is completely the same as MA 50 Smoothed Close.
And if we also take into account that the mysterious SMMA is algorithmically identical to EMA, we get another super property of EMA, prophetically expressed by Neutron as a wish in the first post of the thread. Man, what a great thing this exponential smoothing is after all!
Make a time machine, fly to the future. Go back to our time with quotations from the future about 10 years in advance)) and then you can use that data to draw a "MA that goes ahead".
This option will not work.
The matter is that when you start trading with 10 standard lots, you will start to have a small, but finite effect on the market. Moreover, this influence is characterized by comulativity, i.e., it accumulates over time and as soon as in the second week of trading you will understand that the quotes from the future do not correspond with the current reality, i.e., there is a real division of worlds into a "real" - which we observe, and "alternative" - which was (you saw it yourself) and is now, but "there". Your splendid trade will stall before it has even unfolded!
The trick is that any super-duper swing can be matched with a close SMA or EMA counterpart, just with a smaller period. For example, Jurik with a period of 10 and a phase of +100 is almost identical to EMA4 or SMA5. Yes, Jurik is smoother - but that smoothness will not give a dramatically higher profit. In order to make a really different MA, you need to make a MA that goes ahead of the market - that is, predicts it. Then yes, we will not find an analog to it among existing MAs. All known to us МАшекшек go behind the market as consider the data from the past (history), and not from the future and varying the period of MAHs, you can always plus or minus fit one to another - the difference will not be substantial (the impact on the profit - minimal) ..... But the MAH that will go ahead of the market and predict - how to make it? .....))))
Going back to what I wrote above, I can give you an example. If you take and compare the conventional MAH, for example JMA, with notorious Hodrick-Prescott Filter (I call it Hondrick - it redraws on last bars) on the history when it has successfully redrawn, we see that when JMA goes up, Hondrick goes down and vice versa at the turning points of the market. This is the concept of "staying ahead of the market" - i.e. predicting that the market will eventually go down and not up, and vice versa. But Hondrick is overdrawing, that is why it is based on history. But how to make such an MA without re-drawing - this is the very super MA, the task to strive for.....)))))
And even with this variant, you can see significant drawdowns, when Hondrick goes up and MA goes down and vice versa - and all this because of the strong smoothness of Hondrick.....
... I guess so.
Now we can do some coding!
Um, imho, no. What you end up with could be called a "greedy" algorithm -- since you're essentially minimizing only the current value.
In order to get a match, you have to compute the derivatives of the whole sum. That's why I mentioned the function type.
If you take and compare the usual MA, for example JMA, with the notorious Hodrick-Prescott Filter (I call it Hondrick - it redraws on the last bars) on the history when it has already successfully redrawn,
There really is a big difference in the name, cf. "Band Rock" or "Vocal Instrumental Ensemble"
I remember the difference was the reward ))))
the same with HP - if you try to milk it under the name "MAshkf" you will be as useful as a goat,
but don't be offended by HP, it is NOT a mashka, and certainly not a djurik.
it is a variant of polynomial regression
there really is a big difference in the name, cf. "Band Rock" or "Vocal and Instrumental Ensemble"
I remember the difference was in the reward))))
the same with HP - if you try to milk it under the name "MAshkf" you will be as useful as a goat,
but don't be offended by HP, it is NOT a mashka, and certainly not a djurik.
it is a variant of polynomial regression.
Korey , brilliant as ever......)))))
In order to get a match, we need to compute derivatives of the whole sum. For this I actually mentioned a kind of function.
Can we really put neurons on it?
A standard 3-layer won't do, we need at least a recurrent layer (at least a neuron). It would be nice to have multiplying neurons as well...
Or build a neuromodel in general?
Um, imho, no. What you end up with could be called a "greedy" algorithm -- since you're essentially minimizing only the current value.
In order to get a match, you have to compute the derivatives of the whole sum. That's actually why I mentioned the function type.
There seems to be a misunderstanding here. Yes, at each step I minimize only the current value and, as a result, the whole smooth VR is optimal... Seems logical - by acting optimally at each step, we go all the way optimally. Besides, look at Bulashov's work (see file on previous page), method we use is considered there more or less strictly.