|
PureBytes Links
Trading Reference Links
|
If any one is interested in exploring his/her system's performance
shifts when parameters jitter, use the built-in random() function and
scale your parameter by shifting each of them say, +/- 20%, run them
1000 times and you will have some good insight into your system, I
think. Be sure to attache one random() to each parameter, do not
share.
It is easy to run 1000 times, just set a dummy 1000 loops 'optimize'
and there you have it. Export your AA optimization outputs to Excel
and you can generate all the interesting statistics to 'estimate' your
systems performance when the parameter values shift.
Thomas
--- In amibroker@xxxxxxxxxxxxxxx, "quanttrader714" <quanttrader714@xxx
.> wrote:
> I did the simulations for the example with XLSim ($125). I also
have
> and recommend resampling stats ($179). Well worth the $ if you want
to
> explore this stuff. BTW I have *no* connection with either of
these.
>
> --- In amibroker@xxxxxxxxxxxxxxx, "tchan95014" <tchan95014@xxxx>
> wrote:
> > Hello,
> >
> > MCS is a powerful tools indeed.
> >
> > It might be easy for some to DIY but it might be difficult for
> others.
> >
> > Go to www.tickquest.com to get a free EquityMonarco program to do
> some
> > of these sometime confusion tasks.
> >
> > Search for XLSim, it used to be a free spreadsheet program, but I
> am
> > afraid www.analycorp.com does not offer it for free any more, but
> > there are many floating around in the Internet somewhere. It is
> > limited to 100 rounds though. But some interesting tutorial should
> be
> > still available for free.
> >
> > Also, www.unicorn.us.com/trading, Alex offers a low cost
> spreadsheet
> > program as well as pointers and info to other similar programs and
> > tutorials.
> >
> > Hope this helps...
> >
> >
> > Thomas
> >
> >
> >
> >
> >
> > --- In amibroker@xxxxxxxxxxxxxxx, "Dave Merrill" <dmerrill@xxxx>
> > wrote:
> > > Mark, thanks a huge amount for moving this forward so quickly,
> > yeoman work.
> > >
> > > that said, uh, sorry I'm thick, but I'm pretty lost here. are we
> > supposed to
> > > be able to do something like this ourselves without your
> forthcoming
> > > example, which I assume would be a spreadsheet? this is where we
> use
> > > simulations to estimate the probability of profit and estimate
> > future
> > > drawdowns, right? that I got from your earlier really-quick
> summary,
> > but how
> > > that translates into stuff to do I don't get yet.
> > >
> > > some specific things I don't understand:
> > >
> > > - "simply set up a structure in the worksheet that corresponds
to
> > the
> > > criterion and run the simulation"
> > >
> > > - "randomly select 10 trades with replacement"
> > >
> > > - "set up a spreadsheet structure that will calculate max dd
from
> a
> > string
> > > of x trades, run 1000 simulations and graph the output"
> > >
> > > what criterion? run what simulation? what's "replacement"? run
> 1000
> > > simulations how?
> > >
> > > I assume this will all get clearer with example in hand, which
> > you've said
> > > is coming soon. so really, I'm just checking to make sure I'm
not
> > missing
> > > the boat completely. am I?
> > >
> > > dave
> > >
> > >
> > > > Here's where it starts to get labor intensive. We need to go
> back
> > and
> > > > find the best behaved issues with good PFs, with best behaved
> here
> > > > meaning fairly actively traded *and* consistent on *both* %
> profit
> > per
> > > > trade and % profit per bar graphs. So run the system on a
> group
> > of
> > > > stocks, sort by # trades and go down the list. There are
> > obviously
> > > > many mathematical ways to rank order the consistency but it's
> > better
> > > > to eyeball each and every one if you intend to perform the
> > criteria 4
> > > > and 5 simulations.
> > > >
> > > > We'll be using something called the nonparametric bootstrap
> which
> > is a
> > > > name for resampling with replacement. Although it's an old
> idea,
> > it
> > > > hasn't been very well researched until fairly recently and
even
> > today,
> > > > it's still fertile ground for much additional research.
> There's
> > not
> > > > surprisingly then a lot of controversy over when it is and
isn't
> > > > appropriate to apply it. There are also very few diagnostics
> for
> > it
> > > > unlike most other statistical procedures but these, frankly,
are
> > > > advantages because it's an incredibly powerful procedure if
you
> > apply
> > > > it properly and sometimes that means doing your own research.
> > > >
> > > > We'll use an Excel based simulation package (with its own
random
> > > > number generator). Copy and paste the % profit/trade column
> (from
> > to
> > > > a well behaved issue traded with a robust system) into a
> > worksheet.
> > > > To perform the simulations for criteria 4 & 5, simply set up a
> > > > structure in the worksheet that corresponds to the criterion
> and
> > run
> > > > the simulation. To *estimate* the probability of profit after
> 10
> > > > trades for criterion 4, randomly select 10 trades with
> > replacement,
> > > > sum them and note the answer (the software will do this
> > > > automatically). Then repeat 999 or 4999 or however many more
> more
> > > > times. Usually a total of 1000 is sufficient. Take the
output
> > and
> > > > display as a histogram and a cumulative distribution. To
> > *estimate*
> > > > future max dd for criterion 5, set up a spreadsheet structure
> that
> > > > will calculate max dd from a string of x trades, run 1000
> > simulations
> > > > and graph the output. Standards for robustness on criteria #
4
> > and 5
> > > > -- whatever you feel comfortable with. Mine are based on my
> > > > benchmark system and are proprietary.
> > > >
> > > > Cautions: Your past data must be representative of how the
> system
> > > > will perform in the future, however, simulating with output
from
> > > > robust systems traded on well behaved issues is a bet worth
> > taking.
> > > > But even if your data *are* representative, this still treats
> > trades
> > > > as if they're independent and they're not necessarily. This
> also
> > > > cannot account for single extreme events that haven't happened
> yet
> > > > (like price shocks). There are a few more esoteric concerns
> but
> > all
> > > > except price shocks are reasonably well mitigated by applying
my
> > > > proprietary adjustment. Actually, I have several and am
> willing
> > to
> > > > share two that work well *but offline*, just for the asking,
to
> > those
> > > > interested enough to actually buy simulation software to
> explore
> > this.
> > > > Or to the core members of the team that pursues the robustness
> > > > challenge if there ever is one.
> > > >
> > > > I consider this to be keeping my promise but will still try to
> > post an
> > > > example tonight, tomorrow morning at the latest.
------------------------ 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/I3w.vC/hP.FAA/3jkFAA/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/
|