|
PureBytes Links
Trading Reference Links
|
I've been studying position sizing / money management for about a
month or two now, and have been following most of the recent postings
on Money Management with great interest. My understanding is
increasing, but I have been having some inexplicable results in some
backtests and explorations that I have been running as learning tools
regarding position sizing, and I sure could use some help or input:
Like some others of you out there, I noticed that whenever I add
position-sizing code to any of my experimental systems, the percent
gain drops WAY down.
For example, in one particular test using a Dip-Buying algorithm
translated from Wealth-Script, the results were +18.54 with no
position size statement in the code and only +0.07% with Tharp's
volatility based stops set at a generous 5% of total equity.
Lowering the % of TE to the recommended 1 or 2% resulted in further
degradation of results, (1% : -0.02%loss and 2%: +0.00% negligible
gain.) This did not surprise me as I had already read that many of
you had the same or similar experiences with severely impacted gains
in your own backtests that also were resultant from adding position
sizing.
However, I then determined to try to get an idea at what percentage
of total equity risked with Tharps volatility, WOULD the results come
near to those obtained with no position sizing. That's when I
discovered that it made almost no difference whether %TE risked was
0.1%, 0.2% or whatever, even at 10%, 25%, 33%, 50% and 100% of total
equity risked, the results were still enormously degraded compared to
those without explicitly coded position-sizing. That just didn't
sound right to me.
Growing suspicious, I then proceeded to test a wide range of fixed-
dollar-amount sizings, and fixed-share-amount sizings, all with
similarly degraded overall results as the Volatilty-based sizings.
I found it made no significant difference whatsoever what position
sizing scheme or formula I employed -- What I found was...
---As soon as the PositionSize variable is added to the code, the
percentage gain results decrease dramatically. All of which leads me
to wonder...
1) What is the default for position size used in AmiBroker's backtest
calculations when the variable is NOT declared? Anybody know?
--I don't think it's total equity from Settings because I tested for
that -- when I use the PositionSize variable using "capital/close"
(total equity as position size) the results are still way down (over
95% down) from teh non-declared sizing results, and if that is so,
how can the non-declared sizings result in such better gains? It
doesn't make sense.
2) Why can't these non-position sizing results be approximated using
ANY position-sizing formula when the variable IS declared? I have
been wracking my brain on this, to no avail.
Anybody else tried anything like this? If so, did you notice the
same thing. I mean, I can ceratinly see where capping ones risk to a
small percentage of total equity, and the resulting inevitable
decrease in position sizes, especially for very volatile issues, must
have SOME negative impact on profits and backtest results; Less Risk-
Less Profit; it's the price of insuring against catastrophic loss,
and would be more pronounced the shorter the hold-time of the system,
but this effect should not occur with fixed-share or fixed-dollar
sizings, at least not to the same degree. And, it seems to me, that
at SOME parameter range, the results using the "PositionSize"
variable should approach and eventually approximate the default(no
explicitly declared position size variable) gain percentages. But
apparently, judging from the results of my own testing anyway, this
is not occuring, so how are the better results without position size
declaration even possible?
I'm baffled. Please anybody, help, comment, question, share insights.
If anyone out there gets interested enough to take the time and
trouble to test some of their systems - not necessarily with Tharp-
style volatility based position sizing - but with a similar wide
variety of simpler, more "vanilla" share-based or dollar-amount-based
sizings, I'd sure love to hear what you find out.
I'm hoping that this is some error or misunderstanding on my part and
not a problem inherent in the the PositionSize variable
implementation, but on the off-chance I've inadvertantly uncovered a
glitch of some sort, it'd be a good thing to get to the bottom of it.
Respectfully,
Nick Molchanoff
|