Optimized Trend Trading | Page 38 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Optimized Risk vs Reward Equation 335 replies

Loss Less Optimized Strategy 63 replies

Optimized Parameters in Metatrader Testing? 3 replies

"Optimized Parameter" feature in EA? 0 replies

Strategy with Optimized parameter 1 reply

Attachments: Optimized Trend Trading
Exit Attachments

Optimized Trend Trading

  • #741
  • Edited 2:43pm Apr 20, 2010 1:58pm | Edited 2:43pm
Quoting jamjamjam
mbkennel,
I tend to agree with a lot of your assessments, and find your background very interesting....
I personally don't find the climateaudit.etc sites to be particularly informative. It is true that climatologists could use some help from professional statisticians, but in that field getting the physics right is more important than getting fine details of the statistics right, and the mainstream climate people have the physics right and most of the skeptics do not.

Quote
Nevertheless, I'll be the first to admit I'm probably the only poster that doesn't have metaquotes, so hopefully I'll be able to add something useful regarding your code. As limited as my understanding is, I looked over your code briefly, and as I interpret it, you are essentially conditioning some of the HULL parameters adaptively as a function of a local SNR value....
It's not exactly the orthodox SNR but it functions the same way. I like abs a little better vs squared because returns are fat-tailed.

Compare to what you would get for a random walk.

Quote
Signal is usually extracted as the power of dominant component of the spectrum. That is the common signal processing approach at least. Your approach, If I'm understanding, sounds a bit like fractal type interpretation (like hurst, D, or R/S range type measurement). I don't know, there are a few ways to express this, but let me know if I'm on the right page first.
Yes. I don't think any predictively useful conventional "dominant component of a spectrum" exists in financial time series in time scales slower than than ultra fast (milliseconds) things which only the experts can exploit.

Quote
Did you come up with these adaptive equations on your own, or is there some reference to where they are shown elsewhere (for instance, using hyperbolic tangent as part of the scaling param).
I made them up from intuition. The tanh is to soft-limit the SNR statistic between [0,1).

Quote
Do you expect strong autocorrelation in SNR?
Yes!

If there isn't then there's nothing useful to do.

Firstly it's very well known that returns in financial series are strongly heteroskedastic, meaning that the first order model of price is a random walk with a variance parameter which fluctuates itself with time. High volatility is more likely to be followed by high volatility and vice versa. Also, financial time series are of course causal in contrast to some models, news releases are preceded by low volatility followed by high-but-decaying volatility---the time-reversal does not happen. (conventional stochastic random walk processes are statistically time-reversible after detrending, nonlinear dynamics is not).


Quote
Also, I like the idea about multifractals requiring multiple time frames to express, as it was mentioned earlier in this thread, but I'm not quite sure how you incorporate that concept here.
Thanks,
Jam

P.S. Regarding the power weighted kernel function, could you write out a simple explanation of what that is doing and why it's useful? thanks.
I don't have a simple explanation, it's only intuition. An EMA has an effective FIR impulse response with an exponential decay. If you imagine you have market participants with a wide variety of timescales and they're all using EMA's, and you average them up, you get power law correlations. Thermodynamic phase transitions show power-law correlations, and its widely believe markets are in some kind of self-organized criticality in the statistical-mechanics sense because of the multiple-participant diversity and self-learning agents.


Quote
P.S.S. As of yet, no one has expressed whether or not any of these functions can be reduced or expressed as a simple linear weighted coefficient FIR.
The HullAMA, no, it is not a linear operator because of the adaptive statistic depending nonlinearily.

The powerlaw kernel is nothing but FIR filter---the difference being that correlations (up to the cutoff) decay more slowly than typical filters used in conventional signal processing which always want to band-pass.

As I said, this is just raw material, but an unexplored direction. I'm here to see what this stimulates in others. I don't have answers.
  • #742
  • Edited 2:43pm Apr 20, 2010 2:23pm | Edited 2:43pm
Quoting mbkennel
....
I do like the way you think.

I'll have to tear apart the code a bit more to try to understand it. I'm still lost on the h1, h2, coefficients and the meaning of the buffer dependency on them. You wouldn't happen to be familiar with R by chance?

And I certainly see your background in non linear dynamics shines. I have been toying with the idea of algorithmic complexity as a measure which seems to be better able to differentiate between randomness and order, and think that although fin time series are not chaotic in the deterministic sense, that they are somewhere at the peak of complexity, much like EEGs.
Like yourself, I'm not sure where to take that, but have found some fascinating ideas within.

I raised the question about the FIR reduction, because if it is true, we should be able to find an optimal solution using other areas outside of the common signal processing/statistical methods (think GA/GP). Our goal here is not so much to find the most smooth parameter or best fit in the conventional regression sense, rather, it is to define the most optimal fitness itself (sharpe/sortino/betfreq/etc) and derive the parameters around such constraints. It's a rather non conventional approach to the problem statement.

P.S. The problem with looking at prediction from a heteroskedasticity clustering approach is that it only determines strong autocorr in absolute magnitudes, not so much direction. I will say, however, that I've seen strong autocorrelation in hurst parameters over sliding windows; that might be a more useful observation towards your goal as it directly relates to your SNR hypothesis.
  • #743
  • Apr 20, 2010 2:35pm Apr 20, 2010 2:35pm
Quoting jamjamjam
I do like the way you think.

I'll have to tear apart the code a bit more to try to understand it. I'm still lost on the h1, h2, coefficients and the meaning of the buffer dependency on them. You wouldn't happen to be familiar with R by chance?
I'm better at MATLAB.

Hull moving average. This is what I got it from:
http://www.istockanalyst.com/article...icleid/3019836

I used exponential instead of a FIR WMA. EMA's are great because they're the simplest recursive IIR filter. The EMA version of the Hull MA is hence a IIR filter, if you have fixed coefficients.


Quote
And I certainly see your background in non linear dynamics shines. I have been toying with the idea of algorithmic complexity as a measure which seems to be better able to differentiate between randomness and order, and think that although time series are not chaotic in the deterministic sense, that they are somewhere at the peak of complexity, much like EEGs.
Like yourself, I'm not sure where to take that, but have found some fascinating ideas within.

I raised the question about the FIR reduction, because if it is true, we should be able to find...
I think that last approach is pretty dangerous and prone to massive overfitting. Once you have a robust method and understanding thereof then it might be able to help refine a method, but there is no substitute for hard human intuition deriving some essential "physics" at the start. A colleague who worked for a hedge fund said that useful ideas had to work with plain linear regression at the start. Fancy methods might get you a bit more lift but any phenomenon robust enough to be exploited should be visible---with the right representation---with simple methods.
  • #744
  • Apr 20, 2010 2:41pm Apr 20, 2010 2:41pm
Powerlaw seems similar to an MACD plot.
Time flies like an arrow; fruit flies like a banana.
  • #745
  • Apr 20, 2010 2:46pm Apr 20, 2010 2:46pm
Quoting mbkennel
I'm better at MATLAB.
Did i hear MATLAB??

Well in that case: http://www1.cs.columbia.edu/~gms2155.../Project1a.pdf

I've been trying this out and it has a little ways to go.. SL isn't incorporated yet, but the author (with whom i am in correspondence with), has laid out the code so it is easy to follow..

Tried it on a demo and it seemed to do alright for a while, but i haven't figured but how the TP is calculated.
Time flies like an arrow; fruit flies like a banana.
  • #746
  • Apr 21, 2010 9:22am Apr 21, 2010 9:22am
Quoting jamjamjam
So is it safe to say that it is (or can be replicated by) a linear weighted moving average filter with only real number coefficients (and does not use forward looking data)? The reason I ask is because there might have been a more interesting approach to finding the optimal coefficients.

Not so much to find the best smoothed fit, but perhaps the most profitable.
Yes, that is exactly what I do. I don't care if my signal line is the best fit, just if it generates trades with a low risk:reward ratio and a high success percentage.

Personally, I don't understand much of the DSP, but I've found the discussion must interesting.

In a nutshell, I now use weighted moving averages, with a few tweaks (Std deviation and daily cycles) to give me a "reasonable" signal line. I then place trades when the price is against this signal, making sure I'm not entering just after a MAJOR news announcement.

I started programming for Metaquotes, but after finding several bugs, I have now abandoned it in favour of my own hand-carved trading platform (based on OpenQM). MT4 is still nice for it's graphing ability, but I would never trust it with my children's inheritance. :-)

Flea
  • #747
  • Edited Apr 22, 2010 2:24am Apr 21, 2010 9:36pm | Edited Apr 22, 2010 2:24am
Quoting mbkennel
I think that last approach is pretty dangerous and prone to massive overfitting....
Point taken, although, by the same token, in this case, I'm not quite sure why hullma, or any of the ma family of models/parameters are any more 'physics based' or robust than a search optimized approach. Both can use the same criteria in sample for performance comparison, and neither is necessarily any more robust than the other, as they both use the same template model to begin with, and are dependent on arbitrary subjective parameters. The very page you referenced said he had no idea why the parameters were used. From my limited gathering, it seems that hull arrived at his equation via trial and error; the only sense of optimal I've seen is arbitrary itself, as most filter based methods (incl DSP) are a result of part art/part science, based upon the application (some like sharp rolloff, others want smooth settling). I noticed a lot of arbitrary coefficients like Beta and fast/slow parameter scaling that you used to scale the SNR.

Anyways, I started to digest it (HEMA) a bit more, and first replicated my interpretation of the (non adaptive) ema version of Hull (using 5 and 10 period EMAs as the base functions). From what I can see, it appears more jagged than the smoothed intention of WMA based hull, even with the double smoothing;Is that your impression? and what was your intention behind replacing the WMAs from the standard HULL filter with EMAs?

Also, I'm puzzled as to your choice of slow and fast parameters. As I understand the HullMA, it is specifically designed to slice the averaging window into two even regions, then double smooth 2xfast(n/2)-1Xslow(n) over sqrt(n) periods. Yet, you use 2 and 30 day ema periods in the ema arguments. Apologies if I'm slow here, as I haven't had much interest in TA indicators until this thread, but I'm not sure why you use 1:2 ratios of the ema periods, like the hull WMA? Although I understand the linear weighting function and the exponential rolloff are not identical, I'm missing some translation between the region splitting there, as I'm not sure how the 1:2 linear region ratio scales to 2:30 exp. Is there a simple translation equation?

Lastly, I can intuitively understand the need to soft limit the ema coefficients, as anything much greater than 1.2 or so produces severe ringing. Also, I can understand scaling alpha as a function of SNR in general (more noise/sig should be less tight to allow some leeway, less noise/sig should be tighter?) but I am still trying to understand the whole relationship of the adaptivealpha = ah+SNR*(al-ah), and parameter arguments for each of the emas (slow/fast/combined). I assume you have some reason other than trial and error to produce those equations, but I still need to dig in more to understand it. Maybe you can give some pointers there.

Also, I have to say that I really like your innovative approach to measuring SNR; although it does seem to fluctuate all over the place with a small rolling window (once again seemingly subjectively chosen), I very much like the fact that it's not a spectral based measure for the same reasons you asserted. I wonder whether it might be helpful to be log transformed for different vertical price scales (or do the signal and noise relation both scale fairly well over various price ranges); but just thinking out loud there.

And I guess, ultimately, I'm wondering what exactly is your (or anyone else can chime in) measure of optimality? I would use some type of profitable based measure rather than smoothness (not that the two are mutually exclusive depending on your approach). If we go over to the jurik page, there aren't many arguments made other than that the signal might be smoother with less overshoot than other filters, etc.. etc.. but, that should not necessarily be the driving criteria, IMO.

When I get a moment, I'm going to build the full adaptive model, and see if it helps any more on the intuitive understanding.

That's all for now, thanks for any thoughts and/or references to your derivations of the adaptive ema mods.
  • #748
  • Apr 22, 2010 12:28pm Apr 22, 2010 12:28pm
Quoting jamjamjam
Point taken, although, by the same token, in this case, I'm not quite sure why hullma, or any of the ma family of models/parameters are any more 'physics based' or robust than a search optimized approach.
The HullAMA isn't.

If you're going to do a search, what evaluation/boostrapping crtierion are you going to use? You'll have to make some kind of cross-validation sets, deal with autocorrelation in trade results, etc.

Quote
Both can use the same criteria in sample for performance comparison, and neither is necessarily any more robust than the other, as they both use the same template model to begin with, and are dependent on arbitrary subjective parameters. The very page you referenced said he had no idea why the parameters were used....
I agree. It's entirely heuristic.

Quote
Anyways, I started to digest it (HEMA) a bit more, and first replicated my interpretation of the (non adaptive) ema version of Hull (using 5 and 10 period EMAs as the base functions). From what I can see, it appears more jagged than the smoothed intention of WMA based hull, even with the double smoothing;Is that your impression? and what was your intention behind replacing the WMAs from the standard HULL filter with EMAs?
To allow non-integer time scales which is necessary in this application.

Quote
Also, I'm puzzled as to your choice of slow and fast parameters. As I understand the HullMA, it is specifically designed to slice the averaging window into two even regions, then double smooth 2xfast(n/2)-1Xslow(n) over sqrt(n) periods. Yet, you use 2 and 30 day ema periods in the ema arguments. Apologies if I'm slow here, as I haven't had much interest in TA indicators until this thread, but I'm not sure why you use 1:2 ratios of the ema periods, like the hull WMA?
I do. The adaptive estimator chooses the effective period to be between a fast and a slow limit (2 and 30) according to an estimation of a SNR-like (Kaufman efficiency ratio as it turns out) quantity. Call that L. In a regular Hull MA L is constant. Here, it changes with time. L=L(t).

Then I do a a one bar update at t with recursive EMA updates of effective period L, 2L, and sqrt(L) and apply the Hull formula.

Quote
Although I understand the linear weighting function and the exponential rolloff are not identical, I'm missing some translation between the region splitting there, as I'm not sure how the 1:2 linear region ratio scales to 2:30 exp. Is there a simple translation equation?
See above, they are different things.

Quote
Lastly, I can intuitively understand the need to soft limit the ema coefficients, as anything much greater than 1.2 or so produces severe ringing. Also, I can understand scaling alpha as a function of SNR in general (more noise/sig should be less tight to allow some leeway, less noise/sig should be tighter?) but I am still trying to understand the whole relationship of the adaptivealpha = ah+SNR*(al-ah), and parameter arguments for each of the emas (slow/fast/combined)....
No, it is 100% heuristic. The alpha is constrained to be between [al,ah] and moves from one end of the range to another based on the SNR(ER).

Quote
Also, I have to say that I really like your innovative approach to measuring SNR; although it does seem to fluctuate all over the place with a small rolling window (once again seemingly subjectively chosen), I very much like the fact that it's not a spectral based measure for the same reasons you asserted. I wonder whether it might be helpful to be log transformed for different vertical price scales (or do the signal and noise relation both scale fairly well over various price ranges); but just thinking out loud there.
Turns out it's not so new, it's called Kaufman efficiency.

Quote
And I guess, ultimately, I'm wondering what exactly is your (or anyone else can chime in) measure of optimality? I would use some type of profitable based measure rather than smoothness (not that the two are mutually exclusive depending on your approach). If we go over to the jurik page, there aren't many arguments made other than that the signal might be smoother with less overshoot than other filters, etc.. etc.. but, that should not necessarily be the driving criteria, IMO.
I agree.

Typically any optimality criterion is based on some kind of underlying assumed dynamical/statistical data-generating model (e.g. slow modulated sinusoid plus Gaussian noise, e.g.), somebody else *asserts* a desired optimality criterion, like "minimum mean squared error subject to a smoothness constraint X" and then shows that a particular filter design is optimal with respect to that criterion assuming the data generating procedure with certain underlying parameters and assumptions about distributions on those underlying parameters.

In this circumstance there isn't any useful notion of optimality with the DSP type of methods. We don't know the data generating procedure or any mathematically appropriate optimality criterion.
  • #749
  • Edited 6:09pm Apr 22, 2010 5:55pm | Edited 6:09pm
Quote
Also, I'm puzzled as to your choice of slow and fast parameters...2:30
Quoting mbkennel
I do. The adaptive estimator chooses the effective period to be between a fast and a slow limit (2 and 30) according to an estimation of a SNR-like (Kaufman efficiency ratio as it turns out) quantity. Call that L. In a regular Hull MA L is constant. Here, it changes with time. L=L(t).

Then I do a a one bar update at t with recursive EMA updates of effective period L, 2L, and sqrt(L) and apply the Hull formula.
...
I certainly maintain the impression that you are more astute than the average poster, however, I was under the impression you had developed these ideas on your own. It would have been easier to reference the sources you borrowed from up front, rather than leave us new viewers scrambling to understand how "you" came up with some of the parameters (2/30)/ideas expressed in your code.

http://www.investopedia.com/articles...g-averages.asp

I've found references to everything outside of the hull nested EMA concept
so far.

Did you come up with the nested hullEMA on your own, or is there an exact reference other than the basic hullWMA that you know of (and can point us to)? Again, I want to carry on with the analysis, but make sure we are all on equal starting footing rather than spin wheels over already established ideas.

And for those lurkers watching, please chime in the discussion. I don't think we can constructively make much progress if no one wants to share (if it's good or bad, please state your experience).

from the reference I sourced:
Quote
Conclusion
Robert Colby tested hundreds of technical-analysis tools in The Encyclopedia of Technical Market Indicators. He concluded, "Although the adaptive moving average is an interesting newer idea with considerable intellectual appeal, our preliminary tests fail to show any real practical advantage to this more complex trend smoothing method." This doesn't mean traders should ignore the idea. The AMA could be combined with other indicators to develop a profitable trading system. (For more on this topic, read [i][url="http://www.investopedia.com/articles/technical/03/021903.asp"]Discovering...
P.S. Hope I don't sound disparaging, I fully assume you weren't intentionally hiding anything (otherwise, why would you even bring up kaufman), but hopefully you understand my POV.
  • #750
  • Apr 22, 2010 6:02pm Apr 22, 2010 6:02pm
Quoting jamjamjam
I certainly maintain the impression that you are more astute than the average poster, however, I was under the impression you had developed these ideas on your own. It would have been easier to reference the sources you borrowed from, rather than leave us new viewers scrambling to understand how "you" came up with some of the parameters (2/30)/ideas.
The parameters are totally arbitrary, choose what you want.

Quote

http://www.investopedia.com/articles...g-averages.asp

I've found references to everything outside of the hull nested EMA concept
so far.

Did you come up with the nested hull on your own, or is there an exact reference other than the basic hullWMA that you know of (and referenced)?
Again, I want to carry on with the analysis, but make sure we are all on equal start footing rather than spin wheels over already established ideas.
I've seen the Hull MA formula a number of times. I implemented it with EMAs using my own intuition how it should work but probably somebody's done the same before. Othertimes I have seen adaptive moving averages like the Kaufman stuff and many other variants.

But I never saw an adaptive Hull MA before, so I put 1 and 1 together.

BTW this thing only took about an hour of programming it's not that interesting. Long run, there's something in power laws but it's not clear what yet.
  • #751
  • Edited 1:23pm Apr 23, 2010 12:50pm | Edited 1:23pm
[quote]why use nested ema instead of conventional wma?[/qutoe]

Quoting mbkennel
To allow non-integer time scales which is necessary in this application.
Ok, I'm a bit more familiar with some of the basics now. Could you elaborate
a bit on the above? Better yet, give a concrete example of a non-integer time scale and why a WMA would not work on it?

I also noticed in most of the definitions floating around that the adaptive parameter: a=al+snr(ah-al) (i may have written the subscripts backwards), is typically squared when added back into the EMA coefficients. You didn't; any reason?

I completely agree that markets follow a power law, that's pretty much accepted and established. How using an ema vs any other ma connects to that; not so familiar. Also, CB seemed to be keen on fractal properties of different time scales, yet, he was staunchly against using IIRs vs FIRs.

I tend to agree (with CB) that markets may look smoother on different time scales due to heteroskedasticity clustering, the reason being, IMO, is that sampling at different rates in and of itself is a type of filtering. If you sample at a much slower rate, you can skip clustered noisy periods; there is some research I've seen, where they resample in such a way (asynchronously) that the resultant resampled distribution is purely gaussian. A few quants have used similar approaches.

I'm surprised that no one seems to have much to say on this thread, as a basic search turned up several other threads on kaufman's AMA. Regardless of whether CB was self promoting, he had some good ideas and that's what lured me in.

As I go back over his postings, he seemed to mix two different approaches on this thread; 1) cointegration.. being stat arb 2) fast smoothed filter crossovers
It seems to me that he made his trading based income on 1.
Kind of odd to whet appetites with the crossover filter from nirvana theory, then jump into cointegration properties.
  • #752
  • Apr 24, 2010 7:15pm Apr 24, 2010 7:15pm
[quote=jamjamjam;3667040]
Quote
why use nested ema instead of conventional wma?[/qutoe]



Ok, I'm a bit more familiar with some of the basics now. Could you elaborate
a bit on the above? Better yet, give a concrete example of a non-integer time scale and why a WMA would not work on it?
A WMA has a linear kernel which say for a timescale of 5 may have relative weights 5, 4, 3, 2, 1 (divided by their sum) for successive values in the past.

Suppose you wanted 4.3. The built-in MT4 functions won't allow it. You could construct an approximating kernel (and ensure coefficients sum up to 1), but that's a pain.

The output is a linear convolution of input with the filter coefficients. For long time scales (hundreds to thousands of past bars) this is very slow in MT4 (though I do it elsewhere because there's no alternative), whereas the IIR recursive formula of an EMA is very fast.

Quote
I also noticed in most of the definitions floating around that the adaptive parameter: a=al+snr(ah-al) (i may have written the subscripts backwards), is typically squared when added back into the EMA coefficients. You didn't; any reason?
In some codes I've seen that as a parameter called "G", you can put it in. It adds a nonlinearity so that higher SNR results in more than linear changes in the timescales of the filter.

Quote
I completely agree that markets follow a power law, that's pretty much accepted and established. How using an ema vs any other ma connects to that; not so familiar.
Most linear filter theory is not particularly appropriate then, because as I stated in the beginning, they have finite impulse response or exponentially decaying (thin tailed) response.

I believe that in effect many timescales need to be considered as potential influences all at once.

Quote
Also, CB seemed to be keen on fractal properties of different time scales, yet, he was staunchly against using IIRs vs FIRs.

for what purposes and wny? I personally believe that the classic DSP (i.e. conventional linear systems) is likely to be inadequate, FIR vs IIR is a minor issue.

Quote
I tend to agree (with CB) that markets may look smoother on different time scales due to heteroskedasticity clustering, the reason being, IMO, is that sampling at different rates in and of itself is a type of filtering. If you sample at a much slower rate, you can skip clustered noisy periods; there is some research I've seen, where they resample in such a way (asynchronously) that the resultant resampled distribution is purely gaussian....
Yeah, there was a whole bunch of teasers but not so much meat.
  • #753
  • Apr 27, 2010 7:41pm Apr 27, 2010 7:41pm
Quote
You might be interested in the packet sniffing that I've done to connect to IGIndex (UK).

After using Wireshark and Fiddler to see what goes up and down the wire, I built a tick logger using Python and mechanise. That worked fine until the 19:00 to 22:00 busy period, when the Lightstreamer servers are overloaded. I also wanted to make it completely standalone, and not require me to manually startup/login to the IG platform.

I've now re-written using a multivalue system (U2), which allows me to easily recover from timouts and disconnects. Last...

I'm also interested in accessing feeds from other spreadbetting companies to have a variety of prices to trade against from time to time. My approach to automated trading is to connect to as many prices as possible, monitor them in real-time and sit in the long grass until mis-pricing arises.

I've done a lot of reading on SSL and have started getting to grips with Wireshark, OpenSSL and tcpdump etc. I've had limited success looking at how to drag Ajax from pages, but can poll with HttpRequests and all that to get 'nearly' real-time access to prices on some financial betting websites. I can log in and execute trades programmatically this way as well. But if there's SSL or port 443 involved, I'm bunched. .

I'd rather make the connections myself over SSL and monitor prices in real-time over a socket or by hooking onto the Ajax in a webpage, rather than polling, but am stumbling in the dark. I'd appreciate any pointers as to what to read up on to approach this. If I could get a few hints on the likes of ElephantFlea's approach to IGIndex, that would speed me on my way and I could reverse-engineer exactly which requests I needed to make for the protocols of the other sites I want to create a real-time feed for.

Any help at all would be great.
  • #754
  • Apr 28, 2010 2:33am Apr 28, 2010 2:33am
Quoting tarbhjohnnyd
I'm also interested in accessing feeds from other spreadbetting companies to have a variety of prices to trade against from time to time. My approach to automated trading is to connect to as many prices as possible, monitor them in real-time and sit in the long grass until mis-pricing arises. .
This is in effect taking advantage of errors in the broker/spreadbetters systems. Whilst it's possible, I think most brokers would find it unacceptable over the long term. Particularly with large bet sizes. In fact a friend tried exactly this, and got most of their trades cancelled by the broker.

FYI, my approach is to use the signal lines and clutter filter that are described in this thread, and enter a trade when the price is away from the trend. This works for me, and I'm now in the midst of automating it.

Quoting tarbhjohnnyd
I've done a lot of reading on SSL and have started getting to grips with Wireshark, OpenSSL and tcpdump etc. I've had limited success looking at how to drag Ajax from pages, but can poll with HttpRequests and all that to get 'nearly' real-time access to prices on some financial betting websites. I can log in and execute trades programmatically this way as well. But if there's SSL or port 443 involved, I'm bunched. .
First, get to grips with Fiddler (windows only). That's far simpler for disassembling HTTPS.

Second, take a look at stunnel (linux). That allows you to connect to a https server as simply as a http one.

Quoting tarbhjohnnyd
Any help at all would be great.
This is getting a little off-topic. Feel free to send a PM to me (with your email), and we can continue this off-line. I'm in the UK too.

Flea
  • #755
  • Apr 28, 2010 7:23pm Apr 28, 2010 7:23pm
Thanks very much for the reply.

Quote
First, get to grips with Fiddler (windows only). That's far simpler for disassembling HTTPS. Second, take a look at stunnel (linux). That allows you to connect to a https server as simply as a http one.
I've taken fiddler as far as I can and can make requests and deal with responses in HTTP and also in HTTPS (programmatically logging in and submitting trades etc) - basically anything on port 80. But all this is done with HttpWebRequests, so it's really just polling. When it comes to picking up XML and prices on port 443, though, I'm lost and would appreciate some pointers. .

Quote
This is getting a little off-topic. Feel free to send a PM to me (with your email), and we can continue this off-line. I'm in the UK too.
Agreed. Just that I only realised that 'Junior members' can't send PM's. . Not great. Would you mind sending me one and I can send you an email address then?

Tarbh.
  • #756
  • Apr 29, 2010 1:51am Apr 29, 2010 1:51am
I see that indicator it's not bad if someone wants you can use it btw ... i prefer all newbies start learning technical fund analyses and start learn alot for this then make theyr own analyse , trendlines , pivots and more ... dont use indicators to invest your money ... dont invest in indicators invest in knowledge... for the newbies dont try to search for profitable indicator so you wont make with 1 click on mouse hundreds of dollars ..


1.learn analyses how to make them ... and such .
2.Very good trading system is THV Cobra im not pro trader i start dealing with forex before a year ... belive me everyday from 1 year i test learn or do something search for holy grails and such... i blow 2 accounts and i didnt find EA , or indicator that works .. i try alot of not free indicators and systems wich report great income .. but it's not true


so try THV cobra it's very good balanced system if you follow the rules and + make technical analyse and use it with cobra rules and indicators you wont regret ..

Realy THV Cobra is the best system i found from 1 year search ...

dont take that like adviresment or something .. just a good tip from person who do that allday and still didnt find something "holy"
  • #757
  • May 3, 2010 7:59am May 3, 2010 7:59am
Does anyone have an experience or knowledge of how percolation theory apllied to financial markets?Thanks.
  • #758
  • May 10, 2010 3:12am May 10, 2010 3:12am
Forgive me for dragging up old material. I've just implemented a sinc filter in my trading system, and I've found my implementation has horrible lag. I was under the impression that a sinc filter had low phase lag. Am I wrong?

I've included a picture and my implementation code. Am I doing something obviously wrong? It seems to have a periodlength/2 lag in the filtered output.
Attached Image (click to enlarge)
Click to Enlarge Name: Screen shot 2010-05-10 at 1.04.12 AM.jpg Size: 195 KB
Attached File(s)
File Type: txt sincfilter.txt   2 KB | 508 downloads
  • #759
  • Edited 4:35am May 10, 2010 4:33am | Edited 4:35am
Hi,

Looking at your code, I noticed that your impulse response is symmetrical, so in this case, yes, the phase delay is constant and equals to (length-1)/2.

However, you can transform this linear phase FIR into a minimal phase one. In Matlab it can be done using the 'polystab' function.
In this case the magnitude response will be exactly the same (so having the same smoothing property) but the phase response becomes non-linear and thus the lag will be approximately less or equal to (length-1)/4, which can be a big improvement.

Philippe

http://www.mathworks.in/access/helpd.../polystab.html
  • #760
  • May 10, 2010 7:15pm May 10, 2010 7:15pm
Ahh. I figured there was something I was missing. This looks like great "fun" to implement in java.

Thanks Philippe.

Quoting asm8086
Hi,

Looking at your code, I noticed that your impulse response is symmetrical, so in this case, yes, the phase delay is constant and equals to (length-1)/2.

However, you can transform this linear phase FIR into a minimal phase one. In Matlab it can be done using the 'polystab' function.
In this case the magnitude response will be exactly the same (so having the same smoothing property) but the phase response becomes non-linear and thus the lag will be approximately less or equal to (length-1)/4, which can be a big improvement.

Philippe...
3 traders viewing now