Fractal Analysis?
- Thread starter youngtrader
- Start date
got to this website- http://fractalmarketreport.com. I have no idea how he comes up with his indicators, but I will tell you he's pretty good at determining resistant/support levels on the sp 500 (which is his primary trading vehicle). It's hard to make money on it bc, in my opinion, you need to be a full time trader with the ability to devote 6.5hrs a day looking at a screen. If you can't devote this kind of time to it, it might turn out to be costly. But regardless, go to his website. He's good.
As an aside, if anyone knows anything else about David I would be interested to hear your comments. Bc I've been pretty impressed in the 9 or so months I've read him.
As an aside, if anyone knows anything else about David I would be interested to hear your comments. Bc I've been pretty impressed in the 9 or so months I've read him.
'
"The other fractal approach is used to determine trending or trading periods
using hurst exponents (but has little use in predictability of future IMO)."
Reading his FAQ, it looks like he's using this approach, FDI. I haven't seen much use in it as a forecasting tool. Although, the methodology and derivations are readily available online for free.
The FDI calculation is objective, however, the future trajectories are subjective, like most other ta.
Certainly, there will be plenty of wrong calls, strung among the correct ones.
True enough, but he did provide the basics of reversals, momentum trend trading and breakouts, not a bad starting point for a newb.
He also said, study the damn charts until you get it, because nobody else can do that part for you, also not bad advice.
Charlatan? Rip off merchant may be more apt.
Hmm. his track record (http://www.fractalmarketreport.com/fmr-track-record.php) is just a few days in 2005 using well chosen examples. I think you have to subscribe to see the rest? That is usually a warning sign. I don't pay money or join someone's mailing list to see someone's past results!!!
Did you subscribe (or figure out how to see the results?)??
I came across this - his own IB - claimed results, but it is only 4 months:
http://www.precisionfutures.com/page/pf/fractal-program.html
Could you be a little more informative than "he's good"?
Does he make you regular money or provide you with just interesting reading?I had some interest in fractals and chaos theory in the past, but profits always seemed to be theoretical, like the profitunity guy LOL
Does he trade indexes? futures? stocks? The site was somewhat threadbare in specifics, especially the results page.
dtrader 98 wrote:
"Oh yeah, I did see some of those overlays. Only, I"m not certain which part was predicted and which is post data overlay. Is it only the period where the light blue continuous line is the prediction? Because that's only about a quarter of a cycle. How about like a whole month or longer overly, or are you saying only the short quarter cycle length is expected to be predictable before divergence?"
The future projection is the continuous cyan line, where the vertical lines shows the date of highs and lows. On that GIF file, the vertical lines on the prices are from a previous projection (I think it was the July 24th projection you were talking about, based on data ending on July 20th). And the vertical lines to the right of the prices shows where the modified FFT thinks the future high and low point would be.
1 cycle is considered to be a trough and a peak (or vice versa). Another way to look at is one red line followed by 1 green line (or vice versa) is 1 cycle.
So, in the picture, more than 1 cycle is shown, not a quarter cycle, but I consider that the projection is only "accurate" enough for 2 cycles at most.
Hence, my focus on swing trading.
"Oh yeah, I did see some of those overlays. Only, I"m not certain which part was predicted and which is post data overlay. Is it only the period where the light blue continuous line is the prediction? Because that's only about a quarter of a cycle. How about like a whole month or longer overly, or are you saying only the short quarter cycle length is expected to be predictable before divergence?"
The future projection is the continuous cyan line, where the vertical lines shows the date of highs and lows. On that GIF file, the vertical lines on the prices are from a previous projection (I think it was the July 24th projection you were talking about, based on data ending on July 20th). And the vertical lines to the right of the prices shows where the modified FFT thinks the future high and low point would be.
1 cycle is considered to be a trough and a peak (or vice versa). Another way to look at is one red line followed by 1 green line (or vice versa) is 1 cycle.
So, in the picture, more than 1 cycle is shown, not a quarter cycle, but I consider that the projection is only "accurate" enough for 2 cycles at most.
Hence, my focus on swing trading.
Here's how I'm interpreting it (graphic attached). If only the yellow line is prediction vs. actual, that's only 3 days (8/3 to 8/6?). Whereby, actual data is green bars and prediction is light blue (cyan) line. Am I interpreting that as you intend?
Also, signal definition of a cycle is pk to pk, or trough to trough (like one sine wave).
But looking back at your definition (pk to trough), that is about a half cycle in basic signal terminology. Although, as you mentioned the data and predictions are aperiodic, the standard zero crossing or pk to pk terminology should still apply IMO.
"So, in the picture, more than 1 cycle is shown, not a quarter cycle, but I consider that the projection is only "accurate" enough for 2 cycles at most."
More than 1 is shown, but how much is actual data overlaid on prediction. From what I outlined in yellow, (only place where green bars and cyan overlap), that's only 1 cycle (per your def) of actual data vs. prediction. It does look like prediction matched actual for the cycle attached. Was the cyan portion predicted before the real data was overlaid? If so, do you see this occur more often than not?
If you get a strong correlation, even to 1 cycle (per your def), that may well be very useful (I'm doubting a strong correlation, but you did much more work on it than me
)
You mentioned your fft is calibrated somewhat? How far does it deviate from the ideal fft on a window of data?
(for instance, do snr or fundamental tones track?). How long a length of data (days?) do you sample to create your projected waveform?
Again, interesting work. Assuming you created your predicted waveform BEFORE you overlaid the actual data, it's worth investigating. I like it better than the curve fit approach I've seen others post.
Also, signal definition of a cycle is pk to pk, or trough to trough (like one sine wave).
But looking back at your definition (pk to trough), that is about a half cycle in basic signal terminology. Although, as you mentioned the data and predictions are aperiodic, the standard zero crossing or pk to pk terminology should still apply IMO.
"So, in the picture, more than 1 cycle is shown, not a quarter cycle, but I consider that the projection is only "accurate" enough for 2 cycles at most."
More than 1 is shown, but how much is actual data overlaid on prediction. From what I outlined in yellow, (only place where green bars and cyan overlap), that's only 1 cycle (per your def) of actual data vs. prediction. It does look like prediction matched actual for the cycle attached. Was the cyan portion predicted before the real data was overlaid? If so, do you see this occur more often than not?
If you get a strong correlation, even to 1 cycle (per your def), that may well be very useful (I'm doubting a strong correlation, but you did much more work on it than me
)
You mentioned your fft is calibrated somewhat? How far does it deviate from the ideal fft on a window of data?
(for instance, do snr or fundamental tones track?). How long a length of data (days?) do you sample to create your projected waveform?
Again, interesting work. Assuming you created your predicted waveform BEFORE you overlaid the actual data, it's worth investigating. I like it better than the curve fit approach I've seen others post.