DislikedWhy are you keen on using a non-standard time period of 12 minutes 30 seconds?Ignored
Good question. It was just a suggestion that somebody made to me a while back; and I subsequently found (to my great surprise) that it did seem to work well on EURUSD at specific times of the day. This was all gut feeling though; I could only look at the M1 chart in Metatrader, and attempt to visualise what the 12.5M bars "might" look like. Now that I've got the tick data streaming in, I'm going to test the theory. I'm building my own system from scratch, so I can make it a trading robot and back-tested at the same time.
If you go back in this thread, you will find someone posted a method to use non-standard TF's in MT4, which might make your robot development much quicker.I would be interested in seeing your sniffer. You offered to post it, please do.
If anyone is interested in Level II and tick data in real time, you can get them using MBT's Navigator SDK. A demo account will give you a feed to develop any analysis/trading software. However the demo has some data glitches not present in their live account.
The paper pointed out by FAJST shows a way to use tick price, tick volume, and LevelII quotes to estimate (forecast,predict) the future tick moves. That seems to me to be useful, and I would like to join FAJST and anyone else interested in following this trail.
I have some ideas that might simplify the computation a lot; however there's a lot to compute between ticks, so some form of parallel computation may well be necessary.
MadCow
... bad news. Well I have it now, and won't forget. Thanks my friend.
- #698
- Edited Jan 13, 2010 11:44am Jan 12, 2010 5:16pm | Edited Jan 13, 2010 11:44am
You can get my coefficients from the MQL4 indicator I made.
Philippe, your asm8086 filter is very impressive! And I like the mechanism you used to calculate your ullma.
I'm wondering if you might share your design methods for your asm8086 filter. As I understand, you've only posted the filter co-efficients. It would be FANTASTIC if you cared to share MatLab m code for designing asm8086 filters! (Pretty please
Did you use any of the standard (or toolbox) MatLab filter design functions to help? design.equiripple?
Thanks so much for sharing your incredible work.
- #699
- Edited 12:10pm Jan 15, 2010 11:30am | Edited 12:10pm
Hi novatago,
Here's the Matlab code to design the "asm8086" filter (posts #526 & #527) :
You can play with the first 4 parameters...
However it's possible do even better (and more simple) :
Here's the pseudo code to produce this tweaked ullma :
As you can see, the lag is lower than the original implementation while having the same smoothing property and less overshoot.
Enjoy,
Philippe
Here's the Matlab code to design the "asm8086" filter (posts #526 & #527) :
Inserted Code
%---------------------------------------------------------------------------- N=39; BIN=1; WIN=3; ITP=4; xi=flipud(1-((0:1/N:1)'.^ITP)).*N+N; a=(1:-1/(N+1):1/(N+1))'.^WIN; h=flipud(fftshift(ifft([ones(BIN+1,1);zeros(N*2-(BIN*2+1),1);ones(BIN,1)]))); ir=interp1((1:N*2)',h,xi,'spline'); ir=ir.*a; H=dfilt.dffir(ir./sum(ir)); %----------------------------------------------------------------------------
You can play with the first 4 parameters...
However it's possible do even better (and more simple) :
Here's the pseudo code to produce this tweaked ullma :
Inserted Code
%--------------------------------------------------------------------------- HA = heikinHashi H0 = CB's ullma type 0 H1 = CB's ullma type 1 x = (HA.High + HA.Low) / 2 ' = derivative symbol y = H0(x) + (H1(H0(x)') * 2 - H0(H1(H0(x)'))) %---------------------------------------------------------------------------
As you can see, the lag is lower than the original implementation while having the same smoothing property and less overshoot.
Enjoy,
Philippe
Here's the Matlab code to design the "asm8086" filter (posts #526 & #527) :
BTW, you might try looking at Matlab's GA or Direct Search tools to find best model or even direct filter co-efficients. You do this by creating an objective function that compares the output of your filter to the output of a target filter such as a perfect non-causal filter or one of the filters you're trying to model from this forum. MSE (Mean Squared Error) works ok as tool for comparison.