ZigZag
This indicator cannot be used for trading. It effectively looks into
the future. Say you set its threshold at 4%. It will show tops and
bottoms that are 4% or greater difference. If you get a high today,
but a higher hi 2 weeks from now, it will move the reversal to the new
higher high.
It is most often used to calculate theoretical performance of a series
of price movements. Then you can run your indicators over the same
period and calculate a "capture ratio". Good for comparing indicators
- not trading.
-----------
Signal contains information, it is coherent. Noise is random, it is
incoherent. Noise is generated from the random arrival of orders at
the exchange. Also from other types of trading activity such as
options and futures hedging, spread trading etc - totally unrelated to
merits of the security.
Since noise is random it can be averaged out at the expense of having
to collect more data. Pure noise decreases as the square root of the
number of samples taken. Taking 10 times as much data decreases noise
by a factor of about 3 - the square root of 10.
The ROC indicator only uses 2 data points - the current close and the
close 'n' bars ago. It is reported in points or percent - AB uses %
which allows comparing securities.
You can improve ROC by averaging O+H+L+C/4 today and n bars ago. Plot
this and you will see that 8 estimates price have less noise than 2.
Or you can average the price n bars ago and compare it to the close
today. This is called Most Anchored Momentum.
Or you can use RSI which averages all of the gains and all of the
losses. Quite a bit more data than just 2 closes. Plot that and you
will see that it is quieter still.
Or calculate a regression over the number of bars, then use the
resulting equation to calculate the starting and ending points of the
regression line - taking a % of course so you can compare securities.
And there are others - but this should get you started.
--- In amibroker@xxxxxxxxxxxxxxx, "Louis Préfontaine" <rockprog80@xxx>
wrote:
>
> Hi Ed,
>
> Can you tell more about the 'signal to noise" ratio and how to build it?
>
> In the same way, I am looking for a way to integrate a zigzag function
> related to ETF in a formula. Is it possible to automatically set
the zigzag
> function and select, let's say the 100 best results (that is,
reacting like
> the ETF) in the last 252 days which would then be automatically
traded? I
> want to avoid having to redo a new watchlist all the time...
>
> Thanks,
>
> Louis
>
> 2008/5/20 Ed Hoopes <reefbreak_sd@xxx>:
>
> > 1. 200 symbols sounds like a lot. Spend some time on CAREFULLY
> > selecting a universe of NON-correlated ETFs. You can find an .afl
> > that produces a correlation matrix either here or at the AB website.
> > Only include low correlated ETFs in your list say < 0.6 and > -0.6 (or
> > whatever bounds you like).
> >
> > This should drop your list down to several dozen.
> >
> > 2.(with all due credit to Howard Bandy) Think of a reversion to mean
> > system rather than a trend following system you have outlined below.
> >
> > 3.Think about the signal to noise ratio of your ranking system. ROC
> > only uses 2 values to compute the ROC. This produces a very noisy
> > signal and lots of buys / sells based solely on random noise - not
> > good. Think about all of those estimates of the value of the ETF in
> > between.
> >
> > Reef Break
> > "Headed for Acid Drops in a few days - can hardly wait"
> >
> >
> > --- In
amibroker@xxxxxxxxxxxxxxx <amibroker%
40yahoogroups.com>,
> > "upsidetarget" <digimax@> wrote:
> > >
> > > I use a simple ETF RS system (on 200 symbols) which I want to
put into
> > > Amibroker. Here is what I have so far:
> > >
> > > PosQty = 14;
> > > SetOption("MaxOpenPositions", PosQty );
> > > PositionSize = -100/PosQty;
> > > PositionScore = ((ROC(C,63)+ROC(C,252))/2);
> > > Buy=ROC(C,63)>0 AND ROC(C,252)>0 AND Percentile(ROC(C),63,69) AND
> > > Percentile(ROC(C),252,69);
> > > Sell=???;
> > >
> > > I don't think this will do it...
> > >
> > > What I want is to take a percentile ranking of ROC(C,63)
> > > and a percentile ranking of ROC(C,252), add them and divide by 2.
> > >
> > > Example: ROC(63) rank = 70, ROC(252) rank = 90
> > > ... result (70+90)/2=80 [average rank]
> > >
> > > Buy: Average Rank =>80
> > > Sell: Average Rank <80
> > >
> > > Problem is, I don't think I can use percentile ranking to get this
> > > done. Is there another way?
> > >
> > > Thank for any ideas...
> > >
> > > Michael
> > >
> >
> >
> >
>