|
PureBytes Links
Trading Reference Links
|
I have seen that a certain type of "scaling out"
is beneficial in my backtesting. In case this result
does not apply in other circumstances (I don't know,
I haven't tested all the budzillions of possible others),
I'll spell out the boundaries of what I've tried and
where I've seen scaling out to be helpful. Then if
you decide to extrapolate beyond these boundaries
to the vast regions of the unknown and untested, well
I suppose I can't stop you, so I wish you good luck.
What is this kind of scaling out? Entering a position
with full size ("N" contracts or "N" lots or "N" shares
or "N" bushels) and then exiting a portion of the position
on dayX at priceX, exiting a second portion of the position
on dayY at priceY, exiting a third portion of the position
on dayZ at priceZ, etc.
In what circumstances have I tested it and found it
beneficial compared to exiting the position all at once?
Circumstances possessing all 7 of these characteristics:
* Fully mechanical trading system
* Applied to a portfolio of >50 futures markets
* Using fixed fractional position sizing
* Same parameters and same rules for all markets
* Large account value, big enough to support simultaneous
positions in 50 markets
* Longer term trading: Approx 1 - 2 round trip trades
per year in each market of the portfolio
* Loose initial stop such that even the huge winner trades
are "only" 5-10 Rmultiples. (as opposed to "tight"
initial stop systems that have substantial numbers
of 20, 30, and 40 Rmultiple winning trades)
In these cases, I've found that scaling out rules similar
to the following, actually improve "risk adjusted returns"
(gain-to-pain ratios like Sharpe, MAR, Ulcer, Lake, RRR,
Rsquared, Kratio, etc.) compared to exiting all at once.
These are nothing more than Profit Target Exits, applied
to a portion of the position rather than the full position.
SO1: The first time a trade's profit hits 1.5 Rmultiples,
exit 30% of the position
SO2: The first time a trade's profit hits 3.0 Rmultiples,
exit 30% of the remaining position
SO3: The first time a trade's profit hits 4.5 Rmultiples,
exit 30% of the remaining position
SO4: Whenever the trading system's natural exit is
signalled, exit 100% of the remaining position
For this type of scaling out at profit targets, the key
seems to be "use it only on systems with loose stops".
Its benefit apparently is, to take profits before they
disappear. Rather than "Let your profits run", this
method says "Let SOME OF your profits run, but ring the
damn register too! Ka-ching!" Another side effect is,
the percentage of winning trades goes up, which is
economically irrelevant but a tremendous psychological
boost, especially to outsiders such as investors.
My testing software counts up wins and losses in an
interesting manner: if it enters a position with 81 contracts,
then exits 1/3rd of them (27) on day 5 (leaving 54 cars), then
exits 1/3rd of the remainder (18) on day 10 (leaving 36 cars),
then exits 1/3rd of the remainder (12) on day 15 (leaving 24
cars), then gets an exit-everything signal on day 20, my testing
software counts this as 4 trades. All four trades happened
to enter on day 0, the first one exited on day 5, the 2nd one
exited on day 10, the 3rd one exited on day 15, and the 4th
exited on day 20. Trades 1, 2, and 3 would be counted as
winners (since they did exit at profit targets after all)
and trade 4 would either be a winner or a loser depending
on final exit price.
This testing software says that the base systems WITHOUT
SCALING OUT have approx 40% winning trades. When I enable
scaling out, wining percentage jumps to about 75%. It's
not economically significant but I do find it intriguing
nevertheless.
I've gotten this same basic result "scaling out at profit
targets is beneficial" on a half dozen different long term
commodity trading systems. This isn't enough data to
"prove" anything, but I do find it reassuring that scaling
out helps several systems and not just one.
By the way, for those who enjoy inductive reasoning,
I leave you with the following open ended suggestion to
be creative. Turning it over in your mind for half an hour
may suggest some interesting tests for YOU to perform.
If (systems with LOOSE stops) get better when
(scaling out) is added, maybe (systems with TIGHT stops)
might get better when _____(XYZ)_____ is added.
Best wishes,
Mark Johnson
Incidentally, I get omega-list in "digest" form which is
24-48 hours later than "email" form. So, if there are
any replies to this post, I won't see them for a while.
|