|
PureBytes Links
Trading Reference Links
|
Hi nand,
I mean a system which can do better than
the 50:50 (or 48:52) which is the "usual"
random outcome for UP or DOWN.
A simple example: say the price of a stock is
below the lower Bollinger Band. Here the probability
becomes much greater for an UP than if it were
in the middle or above the upper band.
My current studies try to predict the outcome of
the next 1 bar only, for more bars much more work
would be necessary. But, IMHO 1 bar is fully sufficient,
because the game starts over again the next bar...
BTW, C++ is definitely not necessary for this,
any language where you can sort a table (ie. array
of arrays or structures) would be sufficient, for
example simple C. It also could be that eventually ABs
scanner and backtester will be enhanced for creating
and testing of such types of systems too.
UM
----- Original Message -----
From: "nkis22" <nkishor@xxxxxxxxxx>
To: <amibroker@xxxxxxxxxxxxxxx>
Sent: Wednesday, April 30, 2003 6:53 PM
Subject: [amibroker] Re: Systems for indices
> UM,
>
> yes, hit ratio of prediction is worth folowing. IMO
> predicting systems are worth the effort. It would so nice to know
> for example, after MA cross over, what it the probability for
> next 3 bars to close up or down etc. THen not only we get the
> signal from the crossover, but we also have some idea of
> most probable outcome. I like it. In fact I look at this kind
> of anlaysis too but not having much progress - I think I have
> to learn C++.
>
> ANthony is right in pointing out that t he overall upward
> bais is only p=.52, this is from a massive research report
> several years ago.
>
> nand
>
>
> --- In amibroker@xxxxxxxxxxxxxxx, uenal.mutlu@xxxx wrote:
> > Anthony,
> >
> > I was meaning that the prediction should be done
> > using a real world system which takes care of indicator
> > values etc. Simply said, using a "normal" system, but which
> > is specialized on one index or ticker only and gives for
> > each day an up/down forecast. If this is done for all trade
> > days and over a longer periode then the statistics (hit ratio)
> > will give a good indication on the overall quality of the system.
> > So, what is the difference of this approach to "normal" systems:
> > - here the system has to predict the outcome of every trade day
> > (this is mostly not the case with usual systems; they enter a
> > position and stay longer than a day in the trade)
> > The chance is of course 50:50, but I'm a believer that this should
> > be improvable using TA,FA, maths etc.
> >
> > I'm currently testing such a system which has a hit ratio (correct
> > prediction of the next days outcome as up OR down for all in N100)
> > of about 56%. Somewhere I had read (I think it was in Jeff Cooper's
> > book) that he on average has a hit ratio of more than 60%. Such a
> > beast I would like to have.
> >
> > BTW, the 56% hit ratio translates to about 851 to 495% profit since
> > 10/1/2001 (that's in 19 months; for simplicity without counting
> commissions)
> > depending on the number of securities one trades per trade (1 to 5
> tested).
> > The risk is about only 2.8% of the initial cash.
> >
> > Using a smaller period in the indicators used within the code
> increases
> > the chances but at the same time also the risks (what else? :-). It
> brings
> > even 1039 to 393% (again for 1 to 5 securities per trade), but the
> risk now is
> > about 11.6% of the initial cash.
> >
> > So, if the results are good also for other and longer time frames,
> then
> > I think I've found the Holy Grail (HG), don't you think so? :-)
> >
> > UM
> >
> > ----- Original Message -----
> > From: "Anthony Faragasso" <ajf1111@xxxx>
> > To: <amibroker@xxxxxxxxxxxxxxx>
> > Sent: Wednesday, April 30, 2003 3:23 PM
> > Subject: Re: [amibroker] Systems for indices
> >
> >
> > > No....I believe studies have been done...I have also prepared my
> own
> > > explorations in AMI which shows
> > > Generally that it is 50 / 50 ( with a slight bias to
> up ....ex..52 / 48..
> > > etc.) where the next day will close ...up
> > > or down....
> > >
> > > looking back at the past data....we then are looking for
> Patterns ( I have
> > > created explorations in AMI to do this also ) which gives the
> probability of
> > > the following.
> > >
> > > Example:
> > >
> > > Will Monday close up if Friday is less than thursday, or Friday
> is greater
> > > than Thursday..
> > > Will Monday close Down if Friday is less than Thursday, or Friday
> is greater
> > > than thursday.
> > >
> > >
> > > HIT RATIO;
> > >
> > > Simplified example:
> > >
> > > 9 wins * $ 1.00 = $ 9.00
> > > 1 loss * $ 10.00= (- $ 10.00 )
> > >
> > > TOTAL...............= ( - $ 1.00 )
> > >
> > > It should be the Quality of the Wins in relation to the Losses.
> > >
> > > 1 win * $ 10.00 = $ 10.00
> > > 9 loss * $ 1.00 = ( - $ 9.00 )
> > >
> > > TOTAL..............= $ 1.00
> > >
> > > Anthony
> > >
> > > -------Original Message-------
> > >
> > > From: amibroker@xxxxxxxxxxxxxxx
> > > Date: Wednesday, April 30, 2003 6:25:25 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: Re: [amibroker] Systems for indices
> > >
> > > BTW, reducing (atomicing) any system development, and also
> > > the backtesting, to this absolute main goal (predict the next day
> > > as up or down) makes IMHO everything much simpler and also
> > > more reliable. Ie. the quality of any system should be measured
> > > in how its "hit ratio" on average is, based on daily data of the
> past.
> > > Not less and not more is required to develop a good system.
> > > One should concentrate all efforts on this single issue only, ie.
> > > trying to achieve the highest possible hit ratio on average.
> > > Isn't it?
> > > UM
> > >
> > >
> > >
> > > ----- Original Message -----
> > > From: <uenal.mutlu@xxxx>
> > > To: <amibroker@xxxxxxxxxxxxxxx>
> > > Sent: Wednesday, April 30, 2003 11:54 AM
> > > Subject: [amibroker] Systems for indices
> > >
> > >
> > > > Hi,
> > > > I'm looking for systems which can predict on a daily basis
> > > > whether a specific index will rise or fall the next day.
> > > > It would be good if the hit ratio were >= 57 % on average
> > > > for past data.
> > > >
> > > > For example for any of the following indices:
> > > > Nasdaq composite (^IXIC)
> > > > Nasdaq100 (^NDX)
> > > > Standard&Poors500 (^SPX)
> > > > Dow Jones Industrial (^DJI)
> > > > or for any other index (fe. ^SOXX, ^BTK, ^XAU ...)
> > > >
> > > > Thx,
> > > > UM
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Rent DVDs from home.
Over 14,500 titles. Free Shipping
& No Late Fees. Try Netflix for FREE!
http://us.click.yahoo.com/z3tQYA/hP.FAA/AG3JAA/GHeqlB/TM
---------------------------------------------------------------------~->
Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
|