Chop vs. Trend

[ssrrkk]

Quote from slacker:

At the end of the day you will have results that are only backfitted to the data series.

There is nothing in Amp, phase or frequency that will help you trade the next bar. Any research to the contrary would be great news however... Have you had success applying signal processing techniques to market data so that you can trade the signals?

Thank you.

I'm new to algorithmic trading and am testing out several ideas at the moment using my homebrew c++ client to IB's api. I haven't tried any of these signal processing / time series analysis methods yet but will get to them soon.

I always hear about this "past is no indication of future", but I do think there has to be some correlation between the immediate past and the immediate future of the stock which fades away the further you get from your training period. Otherwise, why is it that our eyes can recognize certain features of an intraday timeseries of a stock that persists for more than a day? I think the autocorrelation time itself also varies as a function of time, and there might be certain recognizable indicators that are correlated to the appearance of an especially strong or longer autocorrelation time during which you might have a greater success rate in applying predictive algorithms...

 

[slacker]

Quote from ssrrkk:

...Otherwise, why is it that our eyes can recognize certain features of an intraday timeseries of a stock that persists for more than a day?

Yep, I hear ya... From my notebook of thoughtful quotes found on the web that I agree with:

1/25/2005
The Clustering Illusion, by GM Nigel

One of the main enemies of the chess player (and indeed mankind) is in attempting to associate outcomes of individual games with certain behavior, when in fact no pattern existed. So we end up jumping around nervously from one idea to another (often regarding the opening, lucky pens or shirts etc) when in fact it was just father random at work.

"The clustering illusion popularly refers to the natural human tendency to "see patterns where actually none exist." Since according to a branch of mathematics known as Ramsey Theory complete mathematical disorder in any physical system is an impossibility, it may be more correct to state, however, that the clustering illusion refers to the natural human tendency to associate some meaning to certain types of patterns which must inevitably appear in any large enough data set."

Recently described in a book titled "Mean Markets and Lizard Brains" by Burnham; he describes the natural desire to find patterns where no pattern exists.

...Professor Skinner gave pigeons food without attempting to reinforce any particular behavior. In fact, he gave the pigeons food 'at regular intervals with no references whatsoever to the bird's behavior.'

The outcome of this experiment was superstitious pigeons.... the pigeons attempted to make sense of the outcomes.... One bird was conditioned to turn couter-clockwise about the cage, making two or three turns between reinforcements......Each pigeon developed its own superstitious behavior..... The point is that our 'lizard brains' seek a logical pattern to illogical behavior.

Building a 'superstition' around signal analysis and cycles can be fun.

Check out the white papers on:

http://www.mesasoftware.com/

Then check the performance of those cycle systems on:

http://www.attaincapital.com/
(Why do these systems not do better, if they can identify trend and choppy periods accurately?)

The best (IMHO) Tradestation software that works (and available to the public) with cycles is done by Clyde Lee at

http://www.theswingmachine.com/

I like most of his stuff and have purchased one of his indicators. Quality work all around.

I have a C/C++ system also working with IB data and once had it working with the MESA and Fourier routines downloaded at

http://www.nr.com/

If you can program the system will go together quickly. You will find you can define cycles in the past, but trading the next bar remains difficult.

"Every trader is Dr. Professor of the last bar;
and Mr. Humble at the next bar."

Good luck!

slacker

 

[nononsense]

Quote from ssrrkk:

how about this. use your favorite spectral analysis tool to find the dominant wavelengths of your timeseries. ifyou do find long wavelength components that have much larger coefficients compared to your high-frequency components (i.e., you must screen for stocks with this property), then you might be able to choose the appropriate window for your moving average indicator... pretty basic idea that should be easy to back test.

You're a cute pink baby.
Go ahead with your "pretty basic idea(s)". Please, send us a postcard when you start making some money!

 

[science_trader]

Quote from ssrrkk:



I always hear about this "past is no indication of future", but I do think there has to be some correlation between the immediate past and the immediate future of the stock which fades away the further you get from your training period. Otherwise, why is it that our eyes can recognize certain features of an intraday timeseries of a stock that persists for more than a day? I think the autocorrelation time itself also varies as a function of time, and there might be certain recognizable indicators that are correlated to the appearance of an especially strong or longer autocorrelation time during which you might have a greater success rate in applying predictive algorithms...

There is short-term auto-correlation for stocks. But it fades after 15 minutes.

 

[Thunderdog]

Quote from nononsense:

(i) What makes you believe that there is a "math of trend identification"?

(ii) If such a "thing" would exist, what makes you believe that you could "figure it out"?

nononsense

Well, if you were to read my post again, you would see that:
(i) I don't; and
(ii) I don't.

Happy reading.

 

[thomas_spinner]

Usually the problem is that you don't know what kind of market it is, until it is already over. If you knew when to "switch" strategies, you would be rich!
-Thomas

Quote from Yag:

Hello all,

I am doing my first newbie attempts as far as creating (I'd even say goofying with) my own trading systems.

I have noticed that the principles that work during trending markets wont during the chop, and viceversa.

How could I express in conceptual terms (as opossed to computing terms, I know nothing about that) so a trading platform could apply one set of rules during a trending market and another set of rules during the chop?

The only idea I have come up with is requiring an initial filter, like price moving above over an EMA for a set number of periods in a row in order to apply the trend principles, otherwise apply the chop rule set. Any other ideas guys?


Cheers.

 

[Equalizer]

Quote from Thunderdog:

Well, if you were to read my post again, you would see that:
(i) I don't; and
(ii) I don't.

Happy reading.

I think it was a case of NN seeing something in your post that wasn't really there?!

Hmm... "Me thinks tis a weasel..."

 

[gnome]

Bottom line... you have to presume/guess whether the market is "trend" or "range" and trade accordingly.

 

[Thunderdog]

Quote from gnome:

Bottom line... you have to presume/guess whether the market is "trend" or "range" and trade accordingly.

Well, if I had to guess, I would think that there may be a small minority of people out there who can identify a trend while it is still exploitable with better than random odds. I do not know how, and I do not count myself among them, but there are enough people in the world that the possibility exists. However, I would be willing to bet any amount of money that they could not do it with the kind of mathematical certainty or even "confidence" that some people here may believe or suggest. The study of human behavior, either individual or en masse, is not as tidy or elegant as the study of pure (hard) sciences such as mathematics, physics and the like.

As for trading markets that are either trending or range-bound, I try not to rely on the distinction. In my feeble attempt to circumvent the need for a distinction, I scale out of trades. Sometimes it plays out well, and sometimes I get the worst of all possible worlds. But for now, it is the best I can do.

 

[gnome]

Quote from Thunderdog:

Well, if I had to guess, I would think that there may be a small minority of people out there who can identify a trend while it is still exploitable with better than random odds. I do not know how, and I do not count myself among them, but there are enough people in the world that the possibility exists. However, I would be willing to bet any amount of money that they could not do it with the kind of mathematical certainty or even "confidence" that some people here may believe or suggest. The study of human behavior, either individual or en masse, is not as tidy or elegant as the study of pure (hard) sciences such as mathematics, physics and the like.

As for trading markets that are either trending or range-bound, I try not to rely on the distinction. In my feeble attempt to circumvent the need for a distinction, I scale out of trades. Sometimes it plays out well, and sometimes I get the worst of all possible worlds. But for now, it is the best I can do.

In other words.... "guess".