|
PureBytes Links
Trading Reference Links
|
All,
I am aligned to Hermans approach - what you guys are talking about
seems rather complex to me - the end result doesn't allow for a KISS
approach, which is what I like - I agree with H on the time
limitation and how "inline metrics" offer us RT system feedback plus
we are free to do some extra things there that we can't do under the
constraints of the BT (no offence to AB - the constraints are there
for a good reason and we all want to use the BT at some time or
another).
I am optimistic about the path I am going down with RT portfolio
emulation (as I discussed a little bit before under "Inline
Backtester" etc.
I have been playing with it a little (weekends only) and I believe I
can get code to:
- emulate a balanced portfolio
- integrate multiple systems into that portfolio (including
long/short)
So far my approach is very basic but I have to start somewhere.
(I will post some basic code examples if I get them to work OK)
Theoretically I have the base code for multiple systems working but
the practical limitation might be scaling up to many symbols (so far
I am looping from a watchlist of 3 tickers and the Help manual says
this is very slow in RT - still it is better to get it working in
theory first and worry about large scale application later).
I guess I didn't sell any of you on the idea so far:
- our rule based systems are like number generators (not random but
slightly biased - hopefully)
- trade series are generated when we present data to our systems
- over sufficient samples the trade series has a distinctive profile
(frequency distribution)
- the distribution is tight if our rules are tight
- the order of the trades is 'random' OR at the least not under our
control ???
So what we have is a series of trades, whose range is limited (by
rule) but whose order varies (according to the whim of the market).
THE TRADE SERIES PROFILE DOES NOT VARY (IF WE HAVE DESIGNED OUR
SYSTEMS WELL)
Nominally, the system can produce many different eq curves - based on
the order of the series (note that the above is a nominal model -
perhaps not an exact model - of the real world of trading but it is
close enough to enable us to make trading predictions).
HOWEVER, ONCE WE START TO ALLOCATE CAPITAL TO THE SYSTEM, IN VARYING
AMOUNTS AND AT DIFFERENT TIME INTERVALS, THE NUMBER OF POSSIBLE EQ
CURVE OUTCOMES ESCALATES.
This allocation of capital, is done by the BT in ways that take quite
a bit of effort to follow (I am not saying that the BT has any faults
only that Portfolio Backtesting can obscure the real UNDERLYING
performance of the system).
Further to that, the BT doesn't allow multiple system analysis, at a
Portfolio level.
THE FACT IS THAT THE TRADE SERIES CARRYS (MATHEMATICALLY) ALL OF THE
INFORMATION ABOUT THE SYSTEM THAT WE NEED - WITHOUT HAVING TO SUBMIT
IT TO THE BT.
In short:
the trade series, expressed as growth factor, can be used to emulate
different portfolio outcomes, after the fact i.e.
IT IS EASIER TO UNDERSTAND THE PORTFOLIO RAMIFICATIONS IF YOU GET THE
TRADE SERIES FIRST AND THEN TRIAL VARIOUS PORTFOLIO APPROACHES
INDEPENDENTLY.
Inline metrics (indicators) is one place in AB that lends itself to
this - no COM/OLE/CBT?.
Portfolio emulation example (for a balanced portfolio):
- take the trades as Gf e.g. 3% == 1.03 (in an array)
- repeat for several symbols
- average the GF, for all symbols, bar by bar, as an array to produce
the Portfolio GF
- start with any initial eq and multiply by the Portfolio GF
>From there we can change capital allocation at any bar if we want to,
provided we have the programming skills.
So far I have got close to doing the above:
- I am not sure if AB limitations OR my limitations will stump me
eventually
- I am optimistic it can be done
- initially, the method is very limited (no scaling in out etc,
closed trades only, no multiple trades per symbol) but that might
change later
Note: I understand it is not a BT, I am not trying to build a BT - I
am trying to get the same understanding of my systems that the BT
offers but do it in RT and with a lot less code and hand waving.
(less sophistication == less code but not necessarily less
understanding and usefulness).
The only problem I have encountered so far is that what I need to
share it with others is more difficult than what I need for myself -
so I have to work a bit harder there.
The reason behind that is that I have jumped the over the fence to
RootCauseEvaluation, which is a mindset, and hence I don't see a need
for all of the bells and whistles that BT's provide (all the noise of
the whistle blowing jsut distracts my concentration on what really
counts).
I will share some base code as soon as I can (it is rather simplistic
code but it helps to argue my case).
RCE - it's not the holy grail of system design and evaluation but it
is a different approach that can expand our knowledge somewhat.
brian_z
--- In amibroker@xxxxxxxxxxxxxxx, "Barry Scarborough" <razzbarry@xxx>
wrote:
>
> I'm not sure what you mean but I often put multiple systems on one
> page. I watch how they perform compared to the one I am actually
auto
> trading. I also add other indicators at times to see if they will
add
> anything to the logic.
>
> Barry
>
> --- In amibroker@xxxxxxxxxxxxxxx, Dennis Brown <see3d@> wrote:
> >
> > Would it work to use multiple virtualized systems on the same
> physical
> > machine?
> >
> > Dennis
> >
> > On Aug 18, 2008, at 10:47 AM, Barry Scarborough wrote:
> >
> > > I think it is safer to run multiple instances of AB on separate
> > > systems. I keep getting mixed or confusing results when I try
to
> use
> > > two instances on one system. I was trying to capture 5 second
data
> > > and run auto trading on the same system. I opened the auto
trading
> > > first and then the 5 second. The data in the 5 second instance
> gets
> > > lost and I can't figure out why. I even tried to save it a few
> times
> > > during the day. I run it all day and when I looked at it later
it
> > > isn't there. Well some of it is which is even more confusing,
the
> > > most recent part. I have the database set to 100,000 bars so I
> should
> > > be able to capture about a month of data. Ain't working. So I
> started
> > > collecting the 5 second data on another system. I will see how
> that
> > > works. If using multiple instances on one system is supposed to
> work
> > > then someone needs to define the parameters for doing so.
> > >
> > > Barry
> > >
> > > --- In amibroker@xxxxxxxxxxxxxxx, "Paul Ho" <paul.tsho@> wrote:
> > >>
> > >> I think the easiest is to run multiple instances of AB, one per
> > > system. am i
> > >> missing something?
> > >>
> > >>
> > >> _____
> > >>
> > >> From: amibroker@xxxxxxxxxxxxxxx
> [mailto:amibroker@xxxxxxxxxxxxxxx]
> > > On Behalf
> > >> Of Barry Scarborough
> > >> Sent: Monday, 18 August 2008 9:46 PM
> > >> To: amibroker@xxxxxxxxxxxxxxx
> > >> Subject: [amibroker] Re: About Automatic Analysis
> > >>
> > >>
> > >>
> > >> Herman he is talking about automatic analysis. How can you run
> more
> > >> than one formula at a time? AA defines the formula that is
being
> > >> tested.
> > >>
> > >> You can test more than one system at a time by "forward
> testing". I
> > >> put my formula in a specially designed bar replay indicator
that
> > > will
> > >> keep track of the number of positions/shares and the price
when
> the
> > >> trade is made. I run BarReplay and feed 5 second data into the
> > >> formula. It tells where the trade is made, the conditions of
all
> > > the
> > >> indicators at the trade, whatever the designer wants to track,
> and
> > >> the price at the trade. I calculate the gain as each trade is
> > > closed.
> > >> I post all of that in the interpretation window. Then you can
> click
> > >> on each formula and see what the stats are. I also dump this
data
> > >> into _Trace so that I can go back later and focus in on a
> specific
> > >> area using DebugView when it doesn't trade as expected. Tracing
> > > more
> > >> than one formula is a pain and I typically use this for debug
> only.
> > >>
> > >> If you want to check longer periods you can capture hour or 15
> > > minute
> > >> data and feed that into a EOD chart or weekly chart to see
what
> is
> > >> going on in them. The shorter the period you feed into your
> formula
> > >> the more accurate the results. Let your imagination run away
and
> > > you
> > >> can test/tweak almost anything this way.
> > >>
> > >> IMO that is more accurate than back testing since you are
> tracking
> > >> the almost true performance of the system. Anyway, that is how
I
> > >> handle multiple formula analysis at a time.
> > >>
> > >> Barry
> > >>
> > >
> > >
> > >
> > > ------------------------------------
> > >
> > > 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
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker/
<*> Your email settings:
Individual Email | Traditional
<*> To change settings online go to:
http://groups.yahoo.com/group/amibroker/join
(Yahoo! ID required)
<*> To change settings via email:
mailto:amibroker-digest@xxxxxxxxxxxxxxx
mailto:amibroker-fullfeatured@xxxxxxxxxxxxxxx
<*> To unsubscribe from this group, send an email to:
amibroker-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|