|
PureBytes Links
Trading Reference Links
|
I once heard from someone that some totally wrong data (way out of range
ticks) improved his system's performance... Why would you really want to
test this data thing:
1) To see which data source to use in order to get best returns from your
system?
2) To prove that your system is robust enough to work with just about any
dirty data out there?
If 2) is the reason for this, here are some crazy ideas that I have
experimented with myself, and think that can be quite good in making sure
the system is robust:
1) Open your data up in Excel and add a percentage of random noise to it.
Experiment with several different percentages to see how far you can go
before your system falls apart.
2) Again in Excel, flip the data around without changing the dates (so that
data for the latest bars comes first and then continues on to the former
earlier dates). Also you can cut the data into separate pieces and shuffle
them around (leaving the dates/times, of course).
3) Experiment with various different back-adjustment methods (easy with
Unfair Advantage).
4) Make up your own data. For example take a sine wave, and add a
percentage of random noise to it. Give it a tick value in $ when opening it
in TS, and run your system on it. Of course, it is important not to be very
far from the actual market numbers. Then make up some more complex data -
triangles, squares, etc.in various combinations, and see how your system
would have performed in these conditions.
5) Use a crazy slippage amount. For example, calculate the average range
for all the bars, and divide it by two. Convert this number to a $ value
and use as slippage. If your system is still *consistently* profitable with
something like $500 in slippage in the case of daily Crude Oil (something
I'm working with), it obviously should not be a very bad one.
Hope this helps. At least it should sound bad enough to relieve some of the
bad-data worries. ;-)
Ivo
-----Original Message-----
From: Mark Baze [mailto:mark_baze@xxxxxxxxx]
Sent: Tuesday, August 07, 2001 1:47 PM
To: omega-list@xxxxxxxxxx
Subject: Data swap/comparison
Hello,
Id like to see how using a different data source
affects the optimization/backtesting results for my
trading system. Would someone who is similarly
interested like to swap data (allowing you to see
how my data affects your systems results)? I have,
from a reputable provider, S&P futures intraday (not
EOD) tick data in ASCII (text) format as individual
contracts for March 1990 (SPH90) through June 2001
(SPM1). The data is directly importable into
TradeStation and can also be converted with provided
software into continuous contracts that are also
directly importable. If you would be interested and
have similar data, please contact me privately by
e-mail to check our sources, etc.
I think this is a worthwhile experiment, especially as
a comparison between two reputable data vendors to see
just what sort of an effect that data differences can
have on backtesting and the optimized parameters that
are obtained. To be more specific, I already know
that the effect can sometimes be large, but Id like
to further see how blending the optimized parameters
from optimizations performed on data from different
sources might affect the future or out-of-sample
results that can be obtained. Ill publish my results
once completed.
Id also like to see how my system performed prior to
1990 (especially the 1985-1990 time period to see how
it did before, during and after the 87 crash), but
dont have any data from this time period to provide
for a swap. If someone would be generous enough to
donate intraday tick data for this time period I would
really appreciate it.
Thanks,
Mark
|