Fast Fourier Transform - Cycle Extraction - page 92
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
Jerry, there is nothing stopping you from using fft or any other method to create adaptive indicators, but you need to know how these different methods work on non stationary data in order to understand their limitations. There are many reasons why you might run into problems. That is why reading the thread is essential. I am sorry but you came across as very unwilling to dig into the details. All of your questions have been answered at some point in this thread so what are you waiting for?
As far as testing your models....back tests are worthless. Forward tests are not. The sooner you can make a realistic evaluation of your model the better. If you think you have a model that has performed well over years of historical data then whats the problem? Put it to work in the real world and see what happens. Why not? If you see that after three weeks that your account is zero then you know you have a problem that no amount of further back testing is going to solve. If the opposite is true, then great!
I hope you do succeed Jerry, it is simply not a direction that I would take personally. I have looked at SQ and other similar systems and I know how they work. Throwing out the losing models and keeping the winners works great in the past. Computers are terrific at cycling through parameters for hours until you have reasonable equity curves. What you don't know is the lifespan of your model when faced with real markets.
Please don't take this as negative. This is just my personal view.
regards,
AlexThanks Hughes, glad we're not doing personal stabs anymore. ;-) Happy that you as altitude non-official representation of senior-ness on forex-tsd is not one of ousting antagonisms.
There are two major factors that have been added to such computer runs-amuck, bar-open-only trading which makes for tickless sensitivity stable strategies, and MonteCarlo robustness tests, pointing toward lack of fx world sensitivity strategies.
Do I know if its all up to the the task yet? Of 'course not! Lol. But coming up with 12yr 1hr bar strategies that use only std indies so far which show an overall pattern with good robustness maybe is it. BTW I do use tiny live real $ accts at ibfx to verify things. I don't admit to success yet, have heard of couple folks doing a replacing EA's routine that carries them forward, so working on it. Maybe I never find, thanks for your wish for me instead of insinuations. I did seem to y'all as unwilling to tread thru much, but that's because of many factors, age, peering thru a smartphone screen, not knowing where to start, and terminologies unknown or confused, not to mention rattled by seeming unknown links to whole fft packages software downloads.
I like mladen's last reference to get me off the ground, it immediately reminded me of electronics math engineering class (graduate of SJSU Jan '73). I also have contact with a math Phd Tomasz of fxcraft.com so there might be a beginning journey for me. I've also been keeping a private forum of StrategyQuant updated of this journey in case SQ folks or someone reading might like to go forth with, so since this is a vital arena something may come of that.
I'm glad forex-tsd deleted yours & mine spat as it is of no use.
I'm happy for y'alls wisdom and any help on this journey is not just for me but may be of immense value for those after my life. Albeit one camp (manual) usually disbelieves or discounts the fully auto route...........
.....I've had concepts that foretells that possibly replacing EA results is a balance mid-point between these two camps (not accidentally expecting an EA is right forever, as EA vendors expect you to believe).
Okydoky,
Best,
Jerry
Cycle still holding and Spectre projections seems to be still in effect, which I personally find amazing given its programming. Price came very close to 3360 although I was expecting it to hit a bit higher before it turns over, but, oh well, that is trading. Update chart below.
theory is scary but the indicator seems to do just fine on daily chart
theory is scary but the indicator seems to do just fine on daily chart
Theories like this can, by defaut, not be scary.
It's people that make them scary.
Here is one of the books that can be found on MIT open course ware.
It describes what is known as "Fastest Fourier Transform in the West". Maybe somebody finds it helpful
____________________
PS : the fftw library itself can be found here FFTW Installation on Windows
Mladen,
Is there any difference from discrete wavelet transforms(haar wavelet etc..) and FFT algos in practice ,do you know?
Mladen, Is there any difference from discrete wavelet transforms(haar wavelet etc..) and FFT algos in practice ,do you know?
There is a notable difference in how they are calculated and used. I made a script in Rapidminer using MODWT to compare Haar and Daub-4 wavelets. I think I posted a picture here somewhere of what it looks like. In the end and with enough work, you can extract a period length from a time series just like you can with FFT and Mesa. The problem is however not the accuracy of the extraction, it is what happens next. The projection is the problem and it assumes that the cycle will extend into the future with a high degree of similarity.
I will try and illustrate the problem because we have a text book case on EURUSD Daily. I will use a sinewave indicator that has been very kindly uploaded by MrTools recently. Lets say for example that your cycle extraction tool returned 32 bars. It really does not matter how you came up with that number but you will see the problem. If we were following the cycle then we would have peaked on 31/07/2013. We fell the next day and then reversed on the the third all while your cycle is indicating down. Obviously, something has changed in the market that we could not anticipate with this alone. This is just in the context of time. If we look now at price, you can see that we have just moved above the channel and it has become support so Monday is a pretty important day and not just for EURUSD but also for the AUDUSD. Perhaps you could just say that Monday is an important day for the Dollar.
If the market is not cycling then it is in a trend in the short term with that trend being just part of another larger cycle that we have not identified. We will have to wait and see what happens.
I agree with Alex, according to the cycle the down move was due on July 31st but price continued going up, this is where PA comes into play. we forecasted that 3400 was a reliable target long in EU and we have now hit that and price started to descend. 3187 is an important support level and I believe EU will test that level before going up again, but prior to going all the way down to that level we may see a spike back up to 3400. If 3400 is broken then EU will continue up from there.
There is a notable difference in how they are calculated and used. I made a script in Rapidminer using MODWT to compare Haar and Daub-4 wavelets. I think I posted a picture here somewhere of what it looks like. In the end and with enough work, you can extract a period length from a time series just like you can with FFT and Mesa. The problem is however not the accuracy of the extraction, it is what happens next. The projection is the problem and it assumes that the cycle will extend into the future with a high degree of similarity.
I will try and illustrate the problem because we have a text book case on EURUSD Daily. I will use a sinewave indicator that has been very kindly uploaded by MrTools recently. Lets say for example that your cycle extraction tool returned 32 bars. It really does not matter how you came up with that number but you will see the problem. If we were following the cycle then we would have peaked on 31/07/2013. We fell the next day and then reversed on the the third all while your cycle is indicating down. Obviously, something has changed in the market that we could not anticipate with this alone. This is just in the context of time. If we look now at price, you can see that we have just moved above the channel and it has become support so Monday is a pretty important day and not just for EURUSD but also for the AUDUSD. Perhaps you could just say that Monday is an important day for the Dollar.
If the market is not cycling then it is in a trend in the short term with that trend being just part of another larger cycle that we have not identified. We will have to wait and see what happens.
Just as a matter of interesting (or not interesting?
) note. Had one eurusd StrategyQuant result whereby if there was no long trade open from a formula then open on Sunday's opening.......nice in backtest buuuut in reality w/ 8-10 pips early on upon opening (so I was advised by knowlergeable fx-EA guy) it was unreal strategy so dumped it. In fact when I start to add a short side to a eurusd strategy with SQ I begin w/ "not Sunday".
Thx,
Jerry
As it turned out, the sine wave did eventually produce a valid cross. What's nice about the sine wave--the uniformity of the waves are "readable" to the human eye. Forgetting cycles for a moment, we can see the wave pattern is left intact at least to some extent. Yes, I'm always a genius in retrospect.
@Mr Tools, my belated thank you for this indicator. The sine wave I was previously using is junk.
@hughesfleming, do you mind if I ask... What are the dots located the price extremum on your chart? MTF ZigZag/Semafor, perhaps? I'm researching sine wave confirmation (in addition to raw PA) and not finding much.