|
PureBytes Links
Trading Reference Links
|
What you are saying then is that backtesting large samples is fruitless ???
If # of trades are not important, then what is your criteria for a "good"
system ?
If that's the case, I would have traded the Demark Sequential for the past 3
years......because "it was proven" and the claims "were 60%+ wins"......even
as high as 90% !
I would now be bankrupt.
MY Back-testing indicates it is a "streaky" methodology with HIGH, HIGH
variances of returns.
> -----Original Message-----
> From: Rich Estrem [mailto:estrem@xxxxxxxxxxxxx]
> Sent: Saturday, July 15, 2000 2:19 PM
> To: omega-list@xxxxxxxxxx
> Subject: RE: A question about system design.
>
>
>
> FWIW, I have come to believe that in the world of
> trading systems, # of trades in a backtest has very little
> correlation to how well the system will do going forward,
> until the sample size becomes larger than is often
> possible. Over the past 9 years, we have tested many
> many hundreds of systems, and traded over 100 different
> ones. All the ones we trade/traded had excellent backtest
> results. Some had many hundreds of trades, a few had less
> than 20. I believe that there is probably little or no
> positive correlation between the number of trades in the
> backtests and the realtime success of those 100+ systems.
> We do trade a very simple stock system which tested
> extremely well on hundreds of stocks and continues to hold
> up very well on probably 500+ trades realtime.
> Backtest runs on this method included thousands of trades.
>
> regards,
> rich
>
>
>
> At 11:48 AM 7/15/00 -0500, Simms wrote:
> >My experience is that 100 is minimal.....300+ trades is better.
> >
> >> -----Original Message-----
> >> From: Brian [mailto:bnm03@xxxxxxx]
> >> Sent: Friday, July 14, 2000 11:21 PM
> >> To: List, Omega
> >> Subject: RE: A question about system design.
> >>
> >>
> >> The magical number of 30 comes from sadistics (statistics) where 30 is
> >> considered a minimum sample size for any valid (statistical)
> >> analysis. You
> >> really should try to acquire at least 30 trades. I've seen
> systems that
> >> backtest great across 10-15 trades but then quickly fall apart in
> >> real time.
> >>
> >> > -----Original Message-----
> >> > From: Ross S Bond [mailto:ross.bond@xxxxxxxxxxxxx]
> >> > Sent: Friday, July 14, 2000 6:54 PM
> >> > To: VBatla@xxxxxxx; omega-list@xxxxxxxxxx
> >> > Subject: Re: A question about system design.
> >> >
> >> >
> >> > From my reading to date 30 to 40 trades over the portfolio
> is considered
> >> > statistically relevant.
> >> >
> >> > Ross
> >> > ----- Original Message -----
> >> > From: <VBatla@xxxxxxx>
> >> > To: <omega-list@xxxxxxxxxx>
> >> > Sent: Saturday, July 15, 2000 6:30 AM
> >> > Subject: A question about system design.
> >> >
> >> >
> >> > > In testing a system I understand that you need at least 30
> - 40 trades
> >> > before
> >> > > you can accurately evaluate it.
> >> > >
> >> > > What if you have a system that only trades 10 - 15 times over
> >> a 15 year
> >> > > period? The one I'm testing holds close to that frequency over
> >> > 38 markets
> >> > > that I tested it on. About 75% of those markets show a
> profit. The
> >> > question
> >> > > is: should the system make 30 - 40 trades per market or
> per portfolio
> >> > before
> >> > > it can be taken seriously?
> >> > > [Of course all of the parameters are the same for
> >> each market.]
> >> > >
> >> > > Thanks in advance for any opinion,
> >> > >
> >> > > Vince
> >> > >
> >> >
> >> >
> >>
> >
> >
> >
>
>
|