_Market description - page 3
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
I think the target selection approach is flawed.
Well, you have chosen 100 points as a target (now it's 1000 points ;) ), and the move was 300 points, so you have lost 200 points. Or, maybe it was 80 points and after correction you will get all of it back without trailing stop, and with fixed trailing stop you will save only part of it.
It is a little more complicated than that, these issues are touched upon here periodically, but no one is likely to reveal their secrets without a counter interest.
It is not so important, when new data comes in the forecast has to be corrected, the target may change, but that's all secondary... The main task I'm trying to solve is to formalize, to make an approximate list of parameters you need to analyze when you make a forecast. Which parameters are important and which ones aren't, etc. It's obvious that this set of market characteristics should be somehow tied to the goal, moreover, it's hardly constant, i.e. a fast market - some data is decisive, "rotten" market - absolutely different data... That's all. I don't need anybody's secrets, this is a branch of semiphilosophical investigation, but not a search for other people's strategies, I have enough of my own.)
In general, at the moment I think Slutsky's model is the closest to reality...
I read about Slutsky's model once, but, on the one hand, it's probably correct, cyclicality, periodicity, but on the other hand, it's not very clear how to use it in practice, i.e. with profits when trading.
This caught my attention, because it most accurately corresponds to what I've always seen in practice - if you adjust to a certain area an indicator, working on, say, a divergence - for some time it works for 5 points, but then the market changes its characteristics and everything "breaks"...
And this is closer to practice, in fact any indicator on the final period can be adjusted to work for 5, and then it will break. Right. Even a simple crossover of wands can show excellent results, if the wands periods correspond to the current market situation. And what if we reveal the dependence of these periods on the current significant characteristics of the market, and make the periods their function? I've already written once, that even simple replacement of constant wave periods by linear function of H-L of some pre-bars, it increases productivity on both training period and OOS (I'll demonstrate it when I have time). But H-L is my intuitive "unbiased", the "correct" dependence is probably a bit more complicated. Finding such dependencies, among other things, is one of the goals of this little study. Maybe I'm just an unimportant "speaker", it's hard to convey a poorly formulated idea even for myself)
on the other hand it is not very clear how to use this in practice, i.e. with profit in trading.
Read here posts of ANG3110 ... He expresses some ideas about using Fourier transform to calculate some sinusoids... I think we should work roughly in this direction...
making the period of the indicator a function of some market characteristics is a great idea... But the main question is what kind of characteristics these characteristics are and what kind of function should depend on these variables... To be honest, I have a very vague idea about the concrete implementation of such a technique... Maybe these variables should be the results of optimizations of this indicator at different periods?
...maybe these variables should be the results of the optimisations of this turkey at different periods?
Perfect! But I suspect it won't work: it's too simple, it doesn't work that way in life.
Perfect! But I suspect it won't work: it's too simple, it doesn't work that way in life.
I agree... I've never tried it myself, but I don't think it will work... and it's not clear how to "extrapolate" the next value of such a function... ...with Fourier? That's how many of these values you'd have to get to look for harmonics between them... well, so far it's clear that the case is dark...
I promised to show you about "off the market" mash-ups, I promise not to renege on my promises).
The experiment is mega simple. Two almost identical simple Expert Advisors, no stop-losses or takeovers, opening and closing/reversing by crossing the waving. In the first Expert Advisor, wavelet periods are selected directly in the optimizer. From 1 to 500 with step 1 (250 000 variants), optimization has been performed for 9, 10 month 2008, forward 11 month 2008. I used an unrealistic period, I was working on it with another Expert Advisor. For forward, the best result of the optimizer was obsolete.
1st Expert Advisor: 3219 Profit:2892.18 Deals:9 Profitability:14.19 MO:321.35 Drawdown:886.16 6.54% FASTPERIOD=93 SLOWPERIOD=27 Lots=0.1 StopLoss=0 TakeProfit=0
Forward:
Conclusion, no money can be raised by crossovers, 70-80s are gone forever....
____________________________________________________________________________________________________________
2-nd Expert Advisor: here we select coefficients of a linear function the result of which will be a period of a wopper for crossover search, coefficients from 0.05 to 25 with step 0.05 (the same 250 000 options) the second multiplier is the difference between the maximum and minimum of the last 2 bars in pips, the best option (with one caveat, I was not satisfied with half of passes, very long, because of the multiplicity of wops searched):
3146 Profit:3067.51 Deals:61 Profitability:3.37 MO:50.29 Drawdown:496.00 4.76% FASTPERIODK=4.9 SLOWPERIODK=13.75 Lots=0.1 StopLoss=0 TakeProfit=0
Forward:
Hm, quite a working variant) 70-80 back and the wizards work)
Hooking and experts, those who wish may experiment, repeat, check and maybe correct errors
Experts:
read here posts from ANG3110... there he expresses some ideas of using the Fourier transform to calculate certain sinusoids... I think we should work roughly in this direction...
making the period of the indicator a function of some market characteristics is a great idea... But the main question is what kind of characteristics these characteristics are and what kind of function should depend on these variables... To be honest, I have a very vague idea about the concrete implementation of such a technique... Maybe these variables should be the results of optimizations of this indicator at different periods?
Of course I read this thread, but I think everything should be a little simpler, all these transformations are taking us very far, you can lose touch with the market, and forget what it was all about. Although probably to each his own, for me what is easier, to feel to see, to understand why. Resonances, harmonics are for smart people. I'm for "primitive trading", wands and CP maximum, this is "my atmosphere", transparent, understandable and in principle with acceptable results.
And as for the practical implementation, just an example above...
I am a supporter of "primitive trading", wizards and TF maximum, here is "my atmosphere", transparent, clear and in principle with not the worst result.
I'm inclined to the same... the more I sink into all this "batanika", the stronger the desire to give it all up and go the purely empirical way... but you can't... You have to find a middle ground (between theory and practice)...
You have to find the middle ground (between theory and practice)...
Of course, I'm not trying to look for profit in the crossing of the wagons, I'm just trying to establish meaningful market characteristics. There's a great passion for neural networks, what hasn't been implemented, including me, the result is a little more than zero... Why? Pitiful primitive inputs that do not mean anything for analysis. Whatever you want to analyze. No one has invented a better sequence of normalized price increments with increasing step. We cannot see the forest for the trees... It's a Fourier transform, too. Does one have to take it apart to drive a car? No, you have to know where the gas and the brake are and know how to drive. The market is a car, gas and brake are significant features that have an impact on future movements and the ability to drive is the same as PNN (best suitable for time series prediction IMHO).
Enough of me "going off the deep end", I'm afraid I'll be considered locally insane as it is (from some replies to private messages on this subject it would seem so:)) I think you can figure out what I'm looking for and, again, join the search if you wish.
Of course, I'm not trying to look for profit in the intersection of the wagons, I'm just trying to establish meaningful market characteristics. There's a great fascination with neural networks, which have not been implemented, including by me, but it's a little more than zero use... Why? Pitiful primitive inputs that do not mean anything for analysis. Whatever you want to analyze. No one has invented a better sequence of normalized price increments with increasing step. We cannot see the forest for the trees... It's a Fourier transform, too. Does one have to take it apart to drive a car? No, you have to know where the gas and the brake are and know how to drive. The market is a car, gas and brake are significant features that have an impact on further movement, the same PNN will do for knowing how to drive (most suitable for time series prediction IMHO).
Enough of me "spreading", I'm afraid I'm already considered a local madman (from some replies to private messages on this subject it would seem so:)) I think you can figure out what I'm looking for and, again, join the search if you wish.
If you do not know anything about the physics, you may use AO and AC, but I do not know anything about maths, physics is my thing. I've noticed that in the morning (Asian session) every single indicator shows a more or less decent volume or speed.