Optimized Trend Trading

I didn't understand you. What do you mean saying "is this you?" You suspect me? I just showed you a screenshot that i've found.

Thanks for restate the suscepted scam related to this thread! Just in case newbies accidentially found this thread. We should all help to fight against scam on FF.

The TudorGirl/CodeBreaker case is just restated on Jacko's thread
http://www.forexfactory.com/showthre...96#post3973096

hi asm8086

do you have DESA indicator for MT4 platform? can you share it for us?
thank you.

if others have the indicator or the links for MT4,please share.
thank you very much.

yourspace

In deed, this is very impressive threads and I have to say that some guys are here really competetive.

Just one point,
Hodrick-Prescott filter is based on Band Pass filters and generally band pass filters are adaptive which means end point keep changing their mind as new price information arrives.


But then a guy who was doing a lot of works on A trous wavelet algorithm had same problem and he fixed the problem. The new algorithm called Non decimated haar wavelet, which is quite famous time serise analysis tool.

As the name suggest (i.e. non decimated) , the algorithm does not change even if new bar arrives. You might look at the his work.

HP is generally great tool to see what is current trend direction!!!

Hımm... Thanks for non decimated haar wavelet information , it is avaliable for Tradestation users as an add on product: http://www.traderssoft.com/sys/wt/ts/

Hi,Pls send the formula for the above indicator., I am also working on the filter similar to CB filter

Hi,

I'm working on IGIndex at the moment. Have had success polling in C# with HttpWebrequests, but no luck getting the Lightstreamer C#.Net Client to point to IgIndex's server. All the viewstates, cookies: no problem. I've also looking at some betting companies feeds as well.

Anyone interested in some collaboration? Please send me a PM.

Thanks.

Here are two indicators I developed some time ago combining a HA applied on gently averaged data with already known low lag smoothing techniques.

The code is quite simple.

The goal was to use the data from one of those two as input in a more elaborated math forecasting algorytm.

Please feel free to let me know your opinion.

Thank you

Dan

Have recently read the entire thread.

Am a degreed physicist and get most of it as I am/have been a student of Ehler's work. I haven't had graduate work in the DSP arena, but had enough to be dangerous during undergrad. I use Ehler's stuff successfully in the 5 min and 15 min timeframes. (Somebody had posted a few pages back asking if anybody was familiar with his work). I have his books and all of his papers and seminar presentations (well over 700 pages). After reading all of it more than once I found that he seems to protect a little nugget. He talks about fisher transforms, zero lag moving averages, etc etc etc. In only two places does he mention "what he uses" (read that "what his software uses"). He prefers a band pass filter. The calculation of the band pass filter preserves the relative amplitude of the input signal - which is very useful when trying to determine if the current cycle has enough meat in it to trade it profitably. So, I use a BPF as well as a Fisher Transform.

Somebody else just recently posted asking about a "clutter filter". I use two of them and they work great: first, I simply require the BPF amplitude to be greater than a certain threshold which tells me I have room to trade - otherwise the market is in a tightly ranging market not worth messing with. second, i use Ehler's "trend vigor ratio". This simply compares the trending component of the current move with the cyclical component of the move. If the trend component is larger than the cyclical component then I only take signals in the direction of the trend. Otherwise, cyclical signals are ok and I avoid trend based signals.

Hope this helps.

Oh, sorry - almost forgot, the above mentioned tools/formulas are all available for free on the MESA web site (Ehler's site) - You have to read through a few of the papers to find them, but they are all there.

I am not posting my results because if I do - half the people won't believe them and the other half will say "just because i post them doesn't prove anything". but, let me just say this: those that say "dsp analysis is pointless" or "dsp analysis of financial time series doesn't work" simply haven't taken the time to dive in and figure it out --- because it DOES work.

The CB/TG thing is a shame and I have no idea if he was blowing smoke or not. My guess is that his stuff did/does work AND that he was also blowing smoke to make it look bigger and better for whatever his ultimate motives were. He made a statement at some point about hitting 77 trades in a row. I, or course, cocked an eye brow at that one, but my EA hit 19 out of it's last 20 trades last week (which doesn't happen all the time, of course - it also missed 6 in a row last tuesday - and NO, my stuff is not for sale and I am not starting a fund (except a "college fund" as my daughter will be in college in 2 years)), so I just figured he probably IS that much better than me - whatever, no big deal. The reason I have concluded his stuff works is that if I can make Ehler's stuff work with only a small fraction of CB's high level math proficiency - then surely he "was" (and maybe "is") making some pips.

Anyway, for those of you still working on this (as I continue to) it would be great to see this thread survive and continue. One of the reasons I believe this stuff works so well is that so few people will actually put in the effort to hash it out to the point they get it working. (At one point I had 8 demo accounts all running a different twist until I started to zero in on what worked best for me.)

Holy Grail? No.
Able to make some pips? Yes.

Perfect system with no draw down? No.
Able to make some pips? Yes.

Worth CB trading his integrity to start his own fund? No.
Able to make some pips? Yes.

And, based on that - I, too, would like to hear something from the reported 30+ that have (again reportedly) successfully implemented what was learned from this thread. Even if you don't want to say boo about what you are doing - if you are, in fact, out there at least let everybody else here at FF know of your success or failure. It will help the rest us to keep the faith in pursuing DSP as a pip making method.

You got my interest, what kind of time periods have you tested your DSP method, I have personally found methods that work with some markets but usually not longer term. What I'm really interested in have you backtested (with good data) your system during the last 7-10 years and did it make consistent results all those years?

mikkom -

Nice to see that someone is still around on this thread.

I typically don't believe in backtesting. That's one of the reasons I always have 4+ demo accounts running tests (always forward testing). However, I do believe that ALL markets are cyclical in nature and, therefore, can be successfully modeled with DSP analysis at any given time.

I haven't had much luck with 30m and H1 time frames. The stops have to be too wide for my taste and risk management.

I think, however, that while I firmly believe DSP analysis for the time/price series of financial data is absolutely where the action is --- we CANNOT EVER forget that it is NOT the series that drives price action (the series only helps us attempt to predict it's next location). What drives it is human behavior. And, because it is human behavior, that is why I believe DSP works better on shorter time frames - i.e. when the cyclic/trending behavior of we humans can be reasonably predicted. (Not that someone else hasn't or will figure out DSP analysis on longer time frames - I just haven't yet and this explanation is just my hypothesis as to why). The longer the time frame the more [opportunity for/probability of] fundamental issues that can/will drive human behavior which makes it's prediction (by any analysis) more difficult - which is also why the longer time frames need wider stops because there is much more movement within the time frame.

So, the way I view the markets and their analysis is simply this: lagging indicators LAG (how's that for rocket science?) and, as such, can't help us much with predicting "price movement" (read that "behavior"). Which means whoever uses a lagging indicator doesn't have any kind of an edge.

What DSP analysis brings to the table is very low lag which is just enough of an edge that the probability of a successful trade goes up. It is probably worth noting that the big money guys (and I am talking really BIG money guys) own casinos. (I have a friend who is the controller of a large casino and he tells me "things".) Their "edge" is between 1% and 2% (depending on the game) and they make millions and millions and millions. If we can get to 60% winners and 40% losers our "edge" is either 10% or 20% depending on how you look at it. Most traders cringe at a loser while casino owners smile because that is just part of the equation which they KNOW, over time, they will win.

For me DSP analysis is the antidote to lag and gives us an edge. IMHO everybody makes this whole "trading thing" way too difficult. Losing is part of the game and is no big deal as long as you have an edge. While the DSP math is pretty high level (even for me - BSc in Physics) - it is still just math applied to some price data. (Ehler's even presents his formulas in 8th grade notation so even the most mathematically challenged can implement them.)

I don't know if it is ok for me to directly post any of Ehler's stuff here or not. I will, however, direct you (and anybody else that has interest) to his website (just google "Ehler MESA" and you'll find it). He has a bunch of stuff available for free (he has even communicated via email with me a few times when I had questions - very nice man). Look specifically for his paper/seminar/powerpoint/pdf (whatever form it is in currently) called "Anticipating Turning Points". While I have invested hours and hours and hours reading all his books and over 700 pages of his work - 99% of what I am doing is in that one paper.

One final thought (i have written way too much already so a little more won't hurt). My brother is a hockey player and we always talk about the really good hockey players (that, by extension, are incredible skaters) ABSOLUTELY TRUST THEIR "EDGE". This is referencing the edge of their skate blades not letting them down. While we traders have a different kind of edge the analogy is the same. We have to trust our edge - which for me is DSP.

Hope this helps.

.

Thank you for posting jefffxtrader.

I also think that the DSP is quite useful. The Corona system is the best free system available from Ehlers.

But I think that the DSP is just a part of the "equation" if that exists of course.

My hypothesis is that it is in the same time wrong and useful to analyse through DSP the time price series.

The best mathematical instruments so far come from the Science of chaos.
In practice what we have under metatrader same indicators to measure the fractal dimension: ivar, fdi, fgdi.

You can find them easily in the official mql tread. Even Ehler recently posted how to use his version of fractal dimension index to distinguish trend from range. This article is in the Stocks and Commodities magazine.

In fact I prefer to use DSP instruments in range bound market.
So I use time filter in that purpous avoiding the European session and the US session.

1. So I measure the cycle length with the cycle period of the corona system.
2. After that I make a linear regression with length measured of the dominant cycle period. That is my probabilistic space. I really like range bound trading in that space.
3. I put a 5 degree polynomial with the same period. When the polynomial bends that is a signal.
4. It is necessary to have a good signal to noise ratio
5. As an oscillator confirmation there are plethora of choices:
One of my favorites is the Fisher Transform of Ehlers.
As a matter of fact I prefer to use neural nets in those kinds of environments (a neural net cannot predict any break out but works great in range bound markets). But a simple oscillator works.

If you haven't tested your method with longer term data how can you know it works longer term?

The key question then is, how long have you tested it live. It's easy to find methods that work with certain kind of market types but when regime changes, the method no longer works.

can u please share the indicators with us

A modern market approach consists in not trying to predict based on the gut feeling, technical analysis or even fundamental analysis.

The idea is to try to predict not the market direction but the market state.

And you need a system for every particular market state:

1. Cluster the market states:

All the successful traders have in common that they try to distinguish the market state.
A common approach is to use the relationship between the time of the day in the forex markets and the corresponding volatility.

2. Appropriate Use of an adapted system for the period

So when a market state is identified it is necessary to use an adapted system for this market state.

Traditionally there are:

2.1 Intra-day Trend following system

A Moving average strategy is the archetype of those systems.

Neural nets perform very bad as a trend following systems (with some exceptions for Neural Nets specially tuned for those conditions).
Digital filters with genetic optimization do much better


2.2 Intra-day Range Bound system

Oscillator based strategy is the archetype of those systems. Modern oscillators appears to be much smoother but still all the oscillators shares a collinearity.

The neural nets perform well in those conditions.
Statistical indicators performs very well too. In fact the phase space is so big under such conditions that it is more appropriate to model it statistical instruments. The Bollinger bands as a rudimentary statistical instrument is a classic player in the Intra-day range bound systems.


2.3 Break - out system after a contraction of the volatility

Those systems are particular. For example a Neural Net cannot never ever predict a break-out in the forex market.

How to distinguish between those market conditions:

Well the obvious reason is practice, practice, practice and experience, experience, experience
The human mind and the human perception abilities are much more suited to distinguish those states than any machine. For example an image recognition, the machines are still not capable of doing it for now.

There are some methods and some new ones that will be covered later.

My opinion is that the human mind is appropriate to distinguish between the market states. We believe that when a market state is identified it will continue for a while.

Second we use statistical instruments and methods to help us.

And third once a state is identified we use the appropriate tested methodology. For me a tested algorithmic system appropriate for a particular market state outperforms the average human trader.

So we should not try to predict the future. Even if we try to predict this will destroy our capacity of observing. This phenomenon happens because we select and choose to observe only what conforms our hypothesis.

Thanks for great tip.I am using Juriks' Fractal dimension version instead of Ehlers.I could not work out exactly how to use ivar indicator yet.I am using volatility indicators (smoothed by dsp) with great success.All the best.

http://picasaweb.google.com/lh/photo...eat=directlinkThe FGDI is the fractal dimension graph indicator (slight upgrade of the FDI fractal dimension indicator).

If the FGDI is lower than 1,5 level that means that the time series are persistent

If the FGDI is higher than 1,5 level than means that the times series are antipersistent

If the FGDI is equal to 1,5 there is a 50/50 probability.

The FGDI estimates the fractal dimension of the times series.
The fractal dimension = 2 - Exponent of Hurst (H)

So if the exponent of Hurst is equal to 0.5 we have a FGDI equal to 1,5.
Exponent of Hurst of 0.5 means that the movement is random without long term memory processes.
Exponent of Hurst > 0.5 - persistent
Exponent of Hurst < 0.5 - antipersistent

This is explained elsewhere and I just remember that basic stuff. The FGDI (FDI) according to me is better than the fractal dimension indicator published by Ehler, and moreover its mt4 implementation is not free but in the elite section of TSD.

The IVAR is similar to the FDI but its scale is different. The center line is 0.5.

The interpretation is the same when it goes below 0,5 the movement is persistent and vice versa, it is antipersistent.

According to the statistical mechanics if the time series are random walk the H should be equal to 0.5.

Hurst has discovered that a lot of natural phenomena follow a "biased random walk" or trend with noise. The strength of the trend can be measured by how the Hurst exponent is above 0.5 (that means FDI and FGDI below 1,5 and Ivar below 0.5)

In fact all those formula are estimating the Hurst exponent. I think that the easiest is to understand the Rescaled range analysis as a method, to understand what is going on. I precise that I have no high education in mathematics and I was able to understand the logic.

The fundamental principle is if the time series are random their range will increase with the square root of time (this is original idea of Einstein in his paper for the Brownian motion). Einstein found that the distance a random particle travels increases with the square root of time used to measure it.

And Hurst decided to make a ration dividing the Range by the the standard deviation of the observations (R/S) R- range S - standard deviation.

So

R/S = (a*N)^H

R/S = rescaled range N = number of observations a = a constant H = Hurst exponent

So we expect to have H=0.5 if the price time series are random. But they are not folks, that does not mean that they are easily predictable either .

All this may seem very theoretical. The exponential moving average is either theoretical but easy to use. This is the same.

I will share now one strategy.


Fractal break -out

Basically the ideal price movement is in a price channel. The price channel can be horizontal (range channel) or directional (trend channel). In fact it does not matter the type. When a price breaks the borders of a chart channel we have a break - out of the channel. If that happens we have a probability that the market conditions has changed and the price will further go away from the channel.


The channel can be regular but it can be irregular. All the classifications of the channels is in fact the technical analysis.

A channel defines a trading range. We have up limits and down limits of the channel. You need at least two points from above and from below to draw a price channel. Then just draw a line connecting the up limits and the down limits of the channel.


Well, but when the prices goes beyond the channel and the gets back to the channel often the technical guys will tell you that you have drawn the channels improperly and you need more educations. When you get more education you will draw the lines properly but again you will see the prices go beyond the channel and get back to the channel. As in the avatar the pilot says before going to battle Ain't that a shit.


This is called a false break - out that happens more often that we want to admit (50 % of the cases ). Ain't that a shit again?


Anyway My idea is to get new details how to trade a Break - out.


1. Use of the volatility probability
We have a break - out probability in periods of increased volatility. So it is necessary to check statistically the volatility. This is well known by the professional players. On the net there are studies with the volatility. Usually we expect a break - out during the open of the European Session and the US session


2. Use of fractal indicators FGDI, IVAR, FDI
We use a fractal dimension graph index indicator (FGDI). When a breakout occurs it is accompanied with a change of the fractal state of the price series.

That means that the graph goes from a state of high fractal dimension to a state of low fractal dimension. We have a shift in the internal structure of the price time series.

2.1 Fractal Break - out
Conditions: We are at a blue zone with fractal dimension greater than 1.5. We move to a zone below 1.5. This is a fractal breakout


This is usually observed after a range (FGDI greater than 1,5)


2.2 Fractal Break - in
We are at a red zone, fractal dimension less than 1,5. The fractal dimension get lower.
I call it so because we are in a red zone of low fractal dimension and the fractal dimension gets lower.

This is usually observed in an established trend when we have a breakout in the direction of the trend


3. A peak in the Hurst Difference
Well this is a kind of measure of the change of the transition of the fractal dimension from one state to the other.
This is a new indicator find in the code base of mql. So the idea is to measure the rate of change of the Hurst Exponent. A high peak means that something is going on.

http://picasaweb.google.com/lh/photo...eat=directlink

I'm always keen to find out how Fractal dimension or Hurst Exponent can create an edge in trading but so far I cant find a good practical use.

I am really sorry but I am not yet allowed by the forum rules to show you. One picture and you will get the idea.

Try this:
picasaweb.google.com/lh/photo/Vjc4JLXzi4l-gadzj1A2P75q53yscd2gmWk5Uc6teQI?feat=directlink

John

How is your experience with neural network on indicators?How much difference do they make on indicators?I read that Neural networks are no good in noisy time series but how noisy is noisy?I am thinking to use them for signal filtering first (Neuroshell approach especially adaptivie net algos sounds simple and effective)then use the filtered indicators for classification pupose in an EA.Thanks.