|
PureBytes Links
Trading Reference Links
|
Leslie Walko wrote:
>
> MT:
>
> Here is the quick answer:
> History does not re-occur in the same sequence in
> the future as it occurred in the past. Thus, correlation,
> between markets change constantly.
>
> The "Worst Case Analysis" in TR is the worst case IN HINDSIGHT,
> --historically. This is equivalent to basing your trading
> account size on Max DD in the TS reports.
>
> If in the future the loosing runs occur in a different sequence
> than in the past, your Max DD and your "Worst Case Analysis" would
> change.
> As a matter of fact, your portfolio composition should change
> also.
Some people think that cherry-picking markets is just another form of
curve-fitting. One year's dog might be next year's hero. On the other
hand, traditionally some markets have a tendency to trend better (e.g.
due to government intervention). There are several schools of thought on
markets selection, whether you give them equal weight (risk) etc.
Not knowing which "school of thought" PortfolioMCS belongs to, I still
don't quite understand what it does. Is it that it helps a trader to
decide on 1) portfolio composition and 2) weighting of each "revenue
generating processes" (market/system) in the portfolio ?
But in historical testing, we're always doing things in hindsight with
probabilities, hoping that the future will somewhat resemble the past,
aren't we? i.e. we hope market correlation won't change totally. In the
case of TR, I can test alternate histories using WCA and also test
different sequence of loosing runs by doing MCS on daily returns of the
portfolio.
> Both TR and Portfolio Stream (RINA) have another significant
> error. They compound during the run. Therefore, your 'raw' data
> is not raw data at all, but the profit stream is curve fitted to
> the the total net equity that existed at that moment in time, in
> HISTORIC time.
> (This is the reason why some net losing markets seem to 'help'
> portfolios. They make one big win at the beginning of the run
> and loose all the time after that, yet products like TR and
> P.Stream keep them due to the historic timing of their single big
> win.)
> :-(
I don't know about PS, but TR certainly doesn't "compound during the
run", unless you specifically ask it to in your position-sizing code,
e.g. by implementing a fixed-fractional betsizing. If you do compound,
then ofcourse it affects future results. In fact I recall some papers on
TR, back in 1999, pointing out this as "end years' bias". To avoid it
you can just bet a fixed amount, e.g. $20k per trade, from start to
finish.
But you shouldn't light-heartedly present mere opinions (imho) as
irrefutable absolute FACTS ... using such strong statements "XXX and YYY
do not work", "have significant errors", "ZZZ is only software in
existence that does so-so". Especially as long as it's obvious to me
that you haven't actually used at least one of the software programs you
offer opinions on.
Regards, M
|