|
PureBytes Links
Trading Reference Links
|
I was thinking a similar thing...drawdown (pain) vs average dollars/trade
(gain) means much more to me than the other stats.
Also, I think the notion that "30" trades represents a statistically
significant sample is bogus. Thousands of trades in both the backtest and
out of sample data would inspire confidence in one's system...the more
trades is better, and over decades (if possible) of data. The idea that one
should not use long data histories because the market has drastically
changed is not supported by the historical record (as TJ suggested in
another thread). If I recall correctly, Mark Brown has indicated that
Oddball was developed in the mid-80's and has performed profitably since
then (correct me if I'm wrong, Mark, as I know you will ;-)). As others on
this list have shown or demonstrated from personal experience and real time
performance stats is that systems/methods persist over long periods
(decades, even centuries) of time. Large scale market structural changes
would torpedo even the most robust system out of the water. But we know
that that's not the case. End of soapbox :-)
MT
Gary Fritz wrote:
> > BBBO VARIATIONS TEST FRIDAY 12/7/01
> > Test No. % ProfitablSharpe RaProfit FacAvg. W/L Net Max DD
> > 1 48 4.37 2.00 2.17 146k 14.8k
> > 2 46 5.90 2.23 2.56 171k 19.1k
> > 3 46 5.86 2.22 2.52 167k 19.5k
> > 4 49 5.97 2.28 2.37 174k 18.5k
> > 5 45 6.21 2.28 2.70 176k 14.7k
> > 6 46 6.31 2.31 2.75 178k 12.6k
>
>In every measure except %profitable (which is not all that important,
>especially for such a small difference), #6 is the clear winner. I
>would also look at average $/trade, but I'd bet it's highest (or
>close to it) for #6.
>
>Before I put much weight on it, though, I'd have to see how long of a
>test this was and how many trades it encompassed, and how the system
>behaved on other stocks. It would also be useful to see how each of
>the variants behaved on out-of-sample data. Is #6 more curve-fit?
|