|
PureBytes Links
Trading Reference Links
|
Hello List,
I'm continually battling with a problem maybe you could help solve?
I trade futures exclusively and for this reason I'm interested in
"points-only" testing in Metastock. I do believe that a valid idea has to
be properly tested, and that means tested against as much data of all kinds
and going as far back into history as possible.
However, with this approach I'm up against a very real problem. I usually
test against stock index data and it obviously comprises quite a few
incompatible price ranges. If Nikkei used to be above 30,000, then Dow was
below 200 for quite a while back in history.
Inevitably I have a problem with commissions. With points-only test I have a
points-only commissions as well, and it would be ridiculous to apply the
same points to Dow in 1925 and to Nikkei in 1989. I would rather need
percentage commissions, but that's impossible to have in Metastock.
Right now the only options I'm aware of are:
1/ Truncate my data into periods with similar price ranges (eg., 200-400;
400-800; etc.) and apply Metastock points-only different commissions to each
price range individually. This is what I'm actually doing.
One obvious disadvantage of this is that I'm practically buried under piles
of charts, notes etc. Moving all this stuff manually is what takes most of
the time, and what could be done in days (with % commissions) takes weeks
:-((
Another serious disadvantage is that this assumes testing in neverending
sideways (trendless) periods. So usually the amount of work is doubled by
the necessity to run the results against validation data sets with strongly
trending markets. For example, with Taiwanese index losing almost half her
value in a year time you'd have to use just a few months of data for
testing, but then see what happens to your ideas in a sharp transition
period.
2/ Use no commissions at all - hoping for a good win/loss ratio - if it's
high then probably commissions wouldn't matter much.
Except, of course, when they WOULD matter. Second problem: sometimes a
useless win/loss ratio doesn't preclude the overall profitability of a
system :-((
Much less work here anyway, and if some of these approximations captures
your attention then you can always process the work as usual (point1). Gosh,
how I love to hear about "approximations" in testing context :-((
3/ Use Omega TradeStation Strategy Performance Report which is easily
exported to Excel, trade-by-trade. Commissions set to 0 of course. In Excel
you can do the usual work (and so much for many of TradeStation advanced
testing capabilities:-(( }
However, TS IMO would be even slower than Metastock {in fact, to the point
of uselessness) in handling the amount of charts required in tests. Also
optimization speed makes it inferior {no offence here, Omega fans!
It's superior in just about every other respect... In fact I code everything
in TS when stuff has been tested in MS}.
Frankly, all this situation with testing impossibilities is quite annoying
and it seems nobody really cares... Most futures testing is done on 1-2
contracts and no more than 1.5-2 years back (if not even less) - which is
completely suicidal IMO.
Any ideas please? Testing across a BASKET of contracts would really speed
everything up and I immediately got excited when I heard of it... There was
talk here about MSBT addin. I've checked their website, it does have %
commissions alright, but it doesn't allow points-only test. Maybe there is
some software... AmiBroker? (Supposedly runs a test on multiple contracts?)
OmniTrader? etc.
Thanks, and all the best
Yarroll
PS. Talking about backtesting as described above... If it all worked as I
just hope it one day would, there would be different pitfalls to
encounter... Like 2-3 winning trades from the 20-30,000 range would
inevitably skew the results of a test which would have 20-30 losers in the
200-300 range..
So maybe the Metastock "idea" (or, necessity) to test in separate price
ranges shouldn't be so annoying after all ;-))
Could data somehow be normalised? So as TA indicators would just go nuts?
Or, maybe something else should be constructed, percentage wins or losses
vs. initial investment for every entry, just forsaking Tester's "Net Profit"
completely? (Just like Metastock "percentage" tests, only without the
brainless compounding:-(( }
|