|
PureBytes Links
Trading Reference Links
|
At 7:00 AM -0500 6/19/98, Timothy Morge wrote:
<clip>
>Now I have a little system I want to play with. I want to know if setting
>some stops and trailing stops and all those neat system testing things
>mean a thing if I test it in TradeStation. And if they do, what should I
>test and optomize, and what do I look out for? Last, if I am looking at a
>system that uses tick data for the S&P, how much data should I test it on
>and where can I get something other than just the tick contract data I
>have in my server [something like a big chunk of continuous S&P tick data
>maybe?]??
The key thing to remember is that Omega treats historical bars in an
"interesting" way. Since they only have OHLC data on each bar, they have to
assume certain things about the sequence of the price within the bar when
evaluating stops. This is well described in the on-line help
("Understanding How TradeStation Simulates Market Activity with Historical
Data").
This causes more serious errors if the duration of a trade is less than a
few bars since the pattern of the price within each bar then becomes very
significant in evaluating the stops. But if the duration of a trade is at
least several bars long, then it becomes less significant since the pattern
of prices within a trade is more readily apparent. The extreme case is tick
data which replays what happened in real time (including bad ticks!). But
you will get pretty good results if the duration of a trade is, say, over
five bars long. So daily bars are pretty good if your trade last five or
more days; one minute bars are good if your trades lasts five or more
minutes, etc.
This is an emotionally charged issue with many people. Needless to say,
this is a "simulation" of the true facts (as Omega clearly states). Using
pure tick data would be perfect but not too practical for long-term testing
with TradeStation (13000 bar limits, etc.) But since the data in the future
will never exactly repeat the data in the past, and since you are only
trying to gain confidence in the ability of your system, it isn't all that
bad a "simulation" if you understand the constraints.
In the extreme case where the trade is "zero" bars long, (enter and exit on
the same bar), you get ridiculous results which have been used by
unscrupulous people to make systems they sell look very good.
Unfortunately, it also causes the historical ("simulated") trades to differ
from the real-time trades which can be very disconcerting if you don't
understand what is happening. (Trades seem to "appear" or "disappear" when
you switch from using real trades on real-time tick data to using
"simulated" trades on historical data.)
This is my understanding. Perhaps others will correct any of my
misunderstandings.
Bob Fulks
|