Predicting the Future is a Reality - Using Mathematics and the Moon
- Thread starter Robert Gordon
- Start date
-
- Tags
- fibonacci stock market
CapelDodger
Penultimate Amazing
Frankly, I don't see it working on the average punter. The trick is to get them enthused, and wavelet analysis ain't gonna do it. Once you have them enthused you can throw in "And that's not even to mention wavelet analysis!" and it'll raise the cheers higher.
I go by wallet analysis - maximise flow from theirs to mine.
CapelDodger
Penultimate Amazing
Street cred, lost. I actually regard Star Wars with a loathing that passeth understanding.
There is no Dog but Dogbert.
Maybe you could call your set of investments made with the system your Karstein Regression Analysis Portfolio?
I'm not sure exactly what the waves are trying to do, but it looks like a gussied up version of curve fitting (sort of), with some mathematical lingo thrown in to make it sound more scientific (the name "Fibonacci" pops up a lot).
But mostly, it doesn't seem that complex. It looks like they just draw some trend lines on the stock graphs, and talk about market cycles and Fibonacci ratios.
Basically, the key is to be as vague as possible with regards to the actual modeling, so no one realizes it's a load of crap.
Try it, and you'll get a surprise.
If you extend the analysis past the end of your data, you get nada. The phases and amplitudes in your analysis reconstruct your data block - including the discontinuity at each end. Before and after the period covered by your data, you'll get basically zip - discounting noise from limitations in the numerical representation on the computer.
There's a (good) possibility I made a mistake when I tried it, but I really don't think I could have screwed up to the point that I got a reconstruction of the data plus a cut off back to zero.
Wavelets might be a better way to go. They're better at localizing time variations than the FFT is. When I do them, I use a gaussian function such that 90% of the signal amplitude is within 1/2 of the wave period of the wavelet. I've never tried them on stockmarket data, though.
Truthfully, I don't see much point in it. From a signal processing point of view, stock prices don't have much coherency. That I have tried, even to the point of searching for earlier time frames of a stock to see if there were an area much like the last (for example) five trading days. There are lots of periods where such five day stretches are similar (coherance >.8.) I tried to "retroactively" predict the sixth day by find stretches simliar to the five preceding the sixth. Got no where. The sixth day in all cases blew the coherency to crap. Whether the sixth day went up or down was pretty much even chance.
I did all of that using periods of various lengths, from just a couple of days up to a couple of weeks. With shorter periods, you get more similar stretches but they don't do much for prediction - pure random chance. With longer periods, you get far fewer similar stretches - and the ones you do get are generally not all that similar.
The data I had weren't all that great, either. To get any information about daily variations, you really need at least 2 stock quotes from each trading day - preferably lots more. The only data I could get were once daily - low, average, and peak. It didn't matter much which ones I used, either.
Does anyone know where I could get prices for a stock for each minute during the day - for free and going back at least twenty years?
Re: http://www.internationalskeptics.com/forums/showthread.php?postid=2547092#post2547092
MortFurd, you might be right, I have avoided fourier analysis for about 15 years now...
However isn't it just a summation of functions of Ansin(nx)? and thus periodic at a period of the beat frequency? (Aside: is this frequency infinite?) (Feeling thik now...)
MortFurd, you might be right, I have avoided fourier analysis for about 15 years now...
However isn't it just a summation of functions of Ansin(nx)? and thus periodic at a period of the beat frequency? (Aside: is this frequency infinite?) (Feeling thik now...)
slyjoe
Illuminator
There's two ways of looking at a fourier series. One is as a complex number (real and imaginary components) and the other is as phase and amplitude. What you are talking about doing is taking a group sine waves at the frequncies in your FFT analysis, at the same phase as given by the FFT, and summing them - and extending the wave past the end of your original data block.
Up until you extend it past the original data block, you are in effect doing an inverse FFT the hard way. An inverse FFT will reconstruct the original data block - including its boundaries. If you are using a square window, then the boundary will be square. If you are using any other kind of window, then boundary will be rounded somewhat depending on the characteristics of the window. The inverse FFT will reconstruct all of that.
Consider a short impulse in the middle of your data block. Do an FFT, and it will show you the frequency and phase of all of the components. Do the inverse FFT, and it will reconstruct the original impulse - and the flat signal around it.
Any data block is pretty much like that impulse. Your data block is just a long, ragged, impulse - and all it can do is reconstruct that impulse. As freaky as it seems, the sine waves (at the proper phase) will sum to zero out side the bounds of the original data block.
It's been probably six or seven years since I did this (FFT of stock data,) so I may have some of the details wrong.
You might download R and try it for your self. R has all of the functions built in that you will need, and can graph things as well. Just do an FFT, get the amplitude and phase for each frequency bin, then calculate the sum of the sines outside (t<0 or t>end of data) the range of your data. Below t=0, it'll sum to zero, above t=end of data it will also sum to zero.
Almo
Masterblazer
They've been watching the move Pi too much.
Ansin(wnx) + Bncos(wnx)
or
Ansin(wnx + phin)
(or as MortFurd notes the complex representations of these)
the FT is just a different ways of showing the N real numbers in your time series; and yes, if you extend it it will repeat exactly with a period equal to the length of your original time series, for all time.
forgot the aside question...
this frequency is not infinite (or infintiesimal): the corresponding period is just the duration of the observations.
this frequency is not infinite (or infintiesimal): the corresponding period is just the duration of the observations.
Michael C
Graduate Poster
If I could reliably predict stock market prices using the moon (or indeed any other method), I wouldn't set up an advisory firm. I'd keep very quiet about it and make millions on the stock market.