PureBytes Links
Trading Reference Links
|
> It's not the double digit gains of some of our more
>active strategies, but given the limitations it seems like quite good
results on
>both a total return and risk/adjusted return basis.
>How do you define "working"?
Jerry
Working = execution issues + design issues.
Execution:
The portfolios were bought on Jan 3, while model testing was done based on
buying the closing price of prior month. In other words, it bought the
opening pop-up highs of early Jan, lost about 7% due to that simple gap
between Dec 31 close and Jan 3's open - and thus would be up double digits
and thus much better.
The producer at the website realized the mistake ex-post-facto and decided
to leave it alone since a lot of retail investors were/are following it at
the time for investing or for benchmarking.
Design:
This system can do better. The historical deviations at individual stock
level have not been as high as the ones you see today. The return and the
risk-return are both below historical norms. The creator knows it and
acknowledges it.
Typically, YTD June, this system would be up - say - 25% with the SPX up 5%.
Typically, this system would lose 40% on an individual stock at its worst
point in the portfolio's lifecycle. These are backtested results from 1986,
even though the public portfolio was first created in 1998. Typically the
difference in returns of the best performers and the worst performers
wouldn't be greater than 50% (and therefore the whole portfolio would swing
about in sync). Today it exceeds 150% with some stocks down 70% and some
others up 80% - and the intra-portfolio correlation has vanished even though
the system parameters went untweaked. Today's bottomline looks better purely
due to June/July, for much of the year the SPX kicked butt. Individual
portions did outperform, for eg the MVP Growth or Flare YearTrader each had
individual periods of good performance - absolute/relative.
So while at an absolute level it has outperformed, the people who
participated in its creation feel kind of dissatisfied and have worked on
improving the risk-return profile and will probably modify the public models
around Dec 31, 2000. Most of the modifications are money-management related,
for eg - quitting a stock based on historical 2 std devn of MAE as opposed
to letting it ride the year out down 80%. One modification will revert to
the 1998 and 1999 practice of buying the close - model testing will thus be
compatible with real life execution.
Still, the official portfolios are what they are.
I will dig up in my archive the discussions we've had re its issues, and
email them to you later - incl backtest results etc, if you're interested.
Regards
Gitanshu
|