Hi
see below -
jayram ($101,883.00)
3/18/2007 4:33 PM Wealth-Lab Developer 4.0 Build 3
---------------------------------
Tried this and seemed to work quite well...
{$I 'LNRet'}
AnalyzeSeries( lnretseries(getexternalseries('DX_0000$', #close),250),
'LR250_DX');
Analyzeseries( lnretseries(getexternalseries('SP_0000$', #close),250),
'LR250_SP');
analyzeseries( lnretseries(getexternalseries('FC_0000$', #close),250),
'LR250_FC');
//analyzeseries( lnretseries(getexternalseries('ND_0000$', #close),250),
'LR250_ND');
AnalyzeSeries( lnretseries(getexternalseries('SF_0000$', #close),250),
'LR520_SF');
Analyzeseries( lnretseries(getexternalseries('GC_0000$', #close),250),
'LR250_GC');
//analyzeseries( lnretseries(getexternalseries('LC_0000$', #close),100),
'LR100_LC');
analyzeseries( rsiseries(#close,14), 'AD_RSI14');
The inputs are based on reaearch into correlated markets over 10 years +
indicators (only rsi seemed useful), and of course several different runs with
various other inputs. Played around with several network structures to get the
simplest possible - ended up with 1 hidden layer and 3 neurons. Other info:
0.2 training rate
0.6 momentum
trained on 8 years of aussie dollar (cash series) - note all external market
inputs are also cash series).
last 2 years out of sample gave good seperation between win/loss on the next
50bars (output is the log return over the next 50 days).
Anyone else like to try?
J
[Edited by jayram on 3/18/2007 4:39 PM]
brucev ($100,000.00)
3/19/2007 1:30 AM Hi Jayram,
where did you get your data from?
Cheers,
Bruce jayram ($101,883.00)
3/19/2007 2:43 AM Brucev,
Got data from CSI... any experience with them?
J brucev ($100,000.00)
3/20/2007 2:44 PM Hi Jayram,
I don't have any experience with CSI data. I trade in the UK using data from
Findata, and in Australia using data from Norgate.
I think this data should be available in the aussie feed, but just to be sure,
can you describe each of the input series - just a few words on each to make
sure I choose the correct ones!
Cheers,
Bruce jayram ($101,883.00)
3/21/2007 2:27 PM Inputs are simply the 250bar log returns recalculated each
day (so oldest bar removed, latest one added to recalculate 250bar log
return)of each of the cash series mentioned above + the 14day rsi of the Aussie
Dollar's cash series. I have not removed outliers. all of the cash series
inputs were selected based on correlations over 10 years and some research into
the correlation. I'm still tweaking as i think the correlation between the DX
and the SF is too high to include both as inputs...
here's the output also
{ Predict Log return after 50 bars }
{$I 'LNRet'}
var T, Bar, start, stop, OutSer: integer;
var x: float;
OutSer := CreateSeries;
for bar:= 250 to barcount-51 do
begin
x:=lnret(bar+50,#close,50);
@OutSer[Bar]:= x;
end;
AnalyzeSeries( OutSer, 'Output' );
brucev ($100,000.00)
3/22/2007 5:14 AM Hi Jayram,
sorry I only trade stocks, so I meant what is the real name for "DX_0000$",
etc, so I can match them up to something meaningful in my feed. My data
supplier doesn't have any symbols named "DX_0000$", "SP_0000$", etc...
Cheers,
Bruce jayram ($101,883.00)
3/22/2007 11:53 AM Oh ok, sorry.
If you use Norgate (i used to use premiumdata) you should be able to get the
data easily for computed contracts, not sure if they do the cash series
though...
try CME for symbols 'FC', 'LC', 'AD', 'SF', 'SP', and 'ND'
use NYBOT for symbol 'DX'
FC = Feeder Cattle
LC = Live Cattle
AD = Aussie Dollar
SF = Swiss Franc
SP = S&P 500
ND = Nasdaq 100
DX = US Dollar Index
If no cash series i would try the computed contract series.
J jayram ($101,883.00)
3/22/2007 11:54 AM ..oh and GC = Gold, and can be found in NYMEX data
Ton Sieverding <[EMAIL PROTECTED]> wrote:
Thanks John. Most important part for me is to hear that you got AB
running a NN. Please let me know the Intermaket data you were using. Also
what's the learning procedure for the NN ? In other words, can you please give
us a little more background music ?
Regards, Ton.
----- Original Message -----
From: [EMAIL PROTECTED]
To: [email protected]
Sent: Saturday, October 06, 2007 11:05 AM
Subject: Re: [amibroker] NEURAL ANALYSIS IN AMIBROKER
PLease bare in mind that this is only the source code to the NN in C++
You would still need to create the function and feed data in.
Thsi network is based on an example network i built a while back to predict
the log return of the Aussie Dollar futures contract over the next 50 days. It
needs to be fed the log returns of 6 intermarkets to provide a relevant output.
the reason i posted the code was not so someone could use it as a NN, but to
emphasise teh fact that a NN built using a 3rd party tool COULD be integrated
into AB using ADK and used in a trading system.
John
Prashanth <[EMAIL PROTECTED]> wrote:
Hello Ton,
Attached is the AFL that a person by name John had posted in this forum on
23-09-07.
Cheers
Prashanth
----- Original Message -----
From: "Ton Sieverding" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Saturday, October 06, 2007 12:11 AM
Subject: Re: [amibroker] NEURAL ANALYSIS IN AMIBROKER
> Prashanth do you still know where to find the code for a NN ? I really
would
> like to see how this works because I just don't understand what AB has to
do
> with NN ...
>
> Regards, Ton.
>
> ----- Original Message -----
> From: "Prashanth" <[EMAIL PROTECTED]>
> To: <[email protected]>
> Sent: Friday, October 05, 2007 7:49 PM
> Subject: Re: [amibroker] NEURAL ANALYSIS IN AMIBROKER
>
>
> > Hello Vinay,
> >
> > I think this question of you rs was answered a while back. As far as my
> > belief and knowledge is concerned, No, you cant create a Neural Network
> > using just AB.
> >
> > Someone had posted a formula which he said that proved that NN can
infact
> be
> > coded in AB, but since my C++ files have a compliation error, I havent
> been
> > able to test out the same.
> >
> > There are a lot of NN based trading products in the market though I am
not
> > sure as to whether they really can make money for the buyer since if it
> > could, it woudnt be for sale in the first place.
> >
> > Cheers
> >
> > Prashanth
> >
> > ----- Original Message -----
> > From: "Vinay Gakkhar." <[EMAIL PROTECTED]>
> > To: "amibroker" <[email protected]>
> > Sent: Friday, October 05, 2007 3:40 PM
> > Subject: [amibroker] NEURAL ANALYSIS IN AMIBROKER
> >
> >
> > > Dear Prashanth,
> > >
> > > Can you please tell me whether there is any formula or add-on for
> > > Amibroker which analyses the available past data like a neural network
> > > does, finds the common pattern, and shows the result like neural
> network?
> > >
> > > Thanks,
> > >
> > > vinay
> > >
> > >
> > > Please note that this group is for discussion between users only.
> > >
> > > To get support from AmiBroker please send an e-mail directly to
> > > SUPPORT {at} amibroker.com
> > >
> > > For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
> > > http://www.amibroker.com/devlog/
> > >
> > > For other support material please check also:
> > > http://www.amibroker.com/support.html
> > >
> > > Yahoo! Groups Links
> > >
> > >
> > >
> >
> >
> >
> > Please note that this group is for discussion between users only.
> >
> > To get support from AmiBroker please send an e-mail directly to
> > SUPPORT {at} amibroker.com
> >
> > For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
> > http://www.amibroker.com/devlog/
> >
> > For other support material please check also:
> > http://www.amibroker.com/support.html
> >
> > Yahoo! Groups Links
> >
> >
> >
>
>
>
> Please note that this group is for discussion between users only.
>
> To get support from AmiBroker please send an e-mail directly to
> SUPPORT {at} amibroker.com
>
> For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
> http://www.amibroker.com/devlog/
>
> For other support material please check also:
> http://www.amibroker.com/support.html
>
> Yahoo! Groups Links
>
>
>
