Hi Utku
You need to understand that the 7.X and 9.1
system testers are very different animals and there are many traps for the
unwary EST user. Simply migrating code and parameters is rarely going to
give the same result as the old System Tester. Often the results will not even
be close.
For example, If you're buying with a 1-bar
delay (on next open say), the EST will probably abort a significant
number of trades. Why? Because when you set the EST system to use
100% of capital the number of shares to be purchased is calculated based on
the signal bar prices. However, when the EST discovers on processing the
execution bar that it cannot afford to buy the pre-calculated number of shares
it will abandon the trade rather than recalculate the number of shares that
can be purchased. This happens when the "actual" buy price is higher than the
signal bar buy price.
The EST attempts to buy a pre-calculated number
of shares whereas the old ST used 100% of money available.
There are a number of ways to get around this
problem. One that I use is to double the amount of capital available and then
specify that only 50% of that capital be used per trade. This will give a
starting equity line that is double that of the old ST, but net profit/loss
should much more closely resemble the active part of the old ST equity
curve.
There should be a document by Tom Sprunger in the
Files section that covers a number of EST issues - I suggest you read it. It
won't tell you all of the answers but it does explain a few things that
you should be aware of. Also, a number of past MSTT newsletters have
articles covering various aspects of the EST. Unlike Tom's document,
however, these are not freely available.
Ensuring that the same data is used
by various test tools is an important factor in achieving consistent
results, as Preston and MC have already mentioned, but it's far from
being the only factor to consider. Loading more data than is needed and
then using a "date filter" to restrict each test to the required date range is
something that works for me. Typically the default values of a date filter
would be set as required and then "AND Fml(Date Filter)" appended to your Buy
code.
As the author of the Trade Equity series of
indicators and explorations I've spent a lot of time finding out why the old
ST, the EST and Trade Equity results can be and often are very different. By
understanding the causes and making appropriate code and/or parameter
adjustments I've been able to test systems across hundreds of securities and
consistently achieve totals that are within a few cents of each other.
Obviously this cannot be the case when trying to match "100% of capital"
against "number of shares" (Trade Equity can use either method), but even then
you should be able to get results that are well within the same
ballpark.
Regards
Roy
----- Original Message -----
Sent: Wednesday, August 27, 2008 10:27
PM
Subject: [EquisMetaStock Group]
Systemtest version problem
Hi to all,
I was using MSpro 7.0 (i know it's old ..) And wrote
down various EXPERSTS, SYSTEMTESTS, EXPLORATIONs.
Now I am using MSpro 9.1 and imported the old formulasto nw
version. And the results of same SYSTEMTESTs' differ from each
other.
I think it is not only because of price field because the equity curves
of the systems are unrelated.
Does anyone have an idea ?
“That which hath
been is what will be,
that which is done is what
will be done,
and there is nothing new, under the
sun.”
__________ NOD32 3393 (20080827) Information
__________
This message was checked by NOD32 antivirus system.
http://www.eset.com