|
PureBytes Links
Trading Reference Links
|
Thomas Stridsman has written some good articles dealing with
this in Futures mag, in particular one titled "Truth be
told" in the Jan 99 issue. He discusses the advantages of
using percentage based performance statistics and also
recommends testing on "Ratio-Adjusted Data (RAD)" that keeps
the percentage relationship constant over time. This means a
big move in the past has the same percent change as a big
move in the present.
>From the article - "As a trader, it's important to use
percentage rather than point-based stops. This is especially
true in a market like the S&P 500 that has been trending
heavily over an extended period. For instance, when we
traded the S&P 500 bwith our Dynamic Breakout System (DBS)
model and a $5000 money managment stop and a $10,000
trailing stop, only six out of 11 trades got stopped out
from 1988 to 1989. From 1996 to 1997, 31 out of 32 trades
were stopped out."
The main point to consider, and I'm paraphrasing Stridsman
here, is that how a system might have performed in the past
is of much less interest than how it will perform in the
future. Therefore the performance statistics should
represent how the system would perform if it were applied to
the market *right now*, when it is forced to deal with the
current market price levels, volatility, etc.
Dave
> From: Phil Lane [mailto:accumulator@xxxxxxxxxxxxxx]
[snip]
> Here's the problem - what if the max dd in the
> backtest occured way back
> when. Trading one contract, as i the backtest,
> you would have drawn down by
> about 14K. But IF the test reflected a realistic
> position size (it doesn't)
> your drawdown would have been 8 times that big, a
> whopping $112,000.
>
> Remember that 8 contracts back in the 80's is
> equivalent to just 1 now. So
> who's to say you won't draw down AGAIN by
> $112,000??? Or $224,000, since we
> wanted to double things just to be on the safe
> side. Your $50,000 account
> is going to have a problem!
>
> Sincerely, and I hope this is not totally confusing,
> Phil
>
>
|