S
T R A T E G I S T N E W S L E T T E R
==========================================
=========================================
Archive of
Model Observations
=========================================
This document contains
previous commentaries about model performance (e.g., the Voted QQQQ
Signal and the Grail Signals),
as well as notes about portfolios.
-
Model
Observations (Friday, 5/9/08):
The market 'held' decently on Friday, although it gave our recent
Long a very slight drawdown (-0.06%).
It seems clear that the market dearly wants to go up, and if the
price of oil ever starts to normalize, then a strong upward move is
quite likely.
-
Model
Observations (Thursday, 5/8/08):
The extremely high price of oil continues to act as a damper on the
market, and in fact it is surprising that the market can go up at
all under such conditions. Today at least got us out of our Short
with a decent gain, and we also ended up the day 'in the Black' on
our new Long ... so no complaints.
-
Model
Observations (Wednesday, 5/07/08):
We were pretty sure that the market would head downward on
Wednesday, but of course it is never possible to be 'certain'. It
was an excellent day for our model, and we 'believe' that the market
would have lost some ground even if oil had not spiked in price –
but we will never know...
-
Model
Observations (Tuesday, 5/6/08):
Although the signals for Tuesday were weak, and although our Model
flip-flopped back and forth several times as it further optimized
itself overnight ... moving to a SELL/S and then back to a BUY, the
official signal (TimerTrac) remained a BUY ... and Tuesday proved
our model right. Now, for Wednesday we have a fairly clear-cut
SELL/S (SHORT) signal, and so we will once again change directions.
-
Model
Observations (Monday, 5/5/08):
After switching from a Short to a Long at the Open price today, we
ended up with a slight drawdown of -0.29%,
although prior to the Open (say at 7-730am ET) it was possible to
make the transition while QQQQ was still depressed by $0.14 or so
(when we are holding a Short, and when our new signal is for a Long,
then we generally try to get out of the Short 'prior' to the market
Open – and especially when the pre-market period is weak, and
thus to our advantage).
-
Model
Observations (Friday, 5/2/08):
Our Short worked well most of the day on Friday, but then QQQQ moved
up just past the unchanged mark and increased our drawdown very
slightly (-0.16%). However, although the signals are a bit weak and
indecisive, things look 'Long' for Monday ... and so we are
switching to a Long position for QQQQ.
-
Our
current Voted QQQQ Signal model now back-tests at 85.0% accuracy for
the past 6 years (actually 5.75 years), and it wants to be Long
61.8% of the time – the PERFECT percentage (Fibonacci Ratio).
So, although it occasionally shorts more aggressively than some
people are comfortable with, it will go Long every time that it
looks reasonable.
-
Model
Observations (Thursday, 5/1/08):
In the few days before and after a Fed announcement, extra caution
is always essential. This proved very true today when the market
inexorably marched upward and crashed through our rather tight
protective stop at $47.36 (QQQQ Open + 0.271%). The correct play was
to close out the Short at this point and go to cash, saving -2.84%
of additional shorting loss. An even better play, although one that
is not currently a part of our standard procedure, would be to
reverse at that point, and go Long up until the Close ... at which
point the Short would be re-instated. In any event the protective
stop today was much less optional than usual, and it was tight
enough (Open + 0.271%) to give a clear and early indication that
action needed to be taken.
-
Analysis
of Recent Model Behavior:
Depending on whether stops were taken or not, our recent shorting
drawdown ranges from -4.79% to -7.19%. These are hefty values, and
in such cases we always try to see if our model could have done
better. The sequence of events started, of course, with Google's
announcement after the Close on 4/17/08, causing a major gap-up at
the Open on 4/18/08. Despite continued analysis, there is simply no
way that our model could have taken any other action; there was no
way to know that Google's numbers were going to turn out to be
stellar, and furthermore, our Protocol doesn't currently provide for
actions to be taken 'after' the Close. Nonetheless, we are beginning
to consider preventative measures when Google's next earnings
announcement takes place...
-
Our
posted signals have stayed Short continuously since 4/16/08 (and we
notice that one of our competitors (kt-timing) has been solidly
Short since 4/14/08), but as our model continues to optimize itself
it has retroactively 'discovered' some Long periods embedded within
this larger interval; these occurred on 4/24 and 4/28-29. As usual,
the fact that the model belatedly sees different signals doesn't
affect our posted performance (via TimerTrac), but it is nonetheless
encouraging to see that the model continues to 'learn' new market
behaviors – and that there is enough power in our indicators
that they can adjust to new situations without degrading previous
back-tested performance. Moreover, we have been able to identify a
specific combination of indicators that 'would have' generated a BUY
for 5/1/08, based upon our neutral MTI reading (1.0), combined with
the fact that the preceding day had had a Fed announcement, AND
combined with the fact that our generalized candlestick (Table 2)
prediction for QQQQ (see the Prediction Table) had jumped up
considerably in Wednesday's numbers. So, we have added this new Rule
to our model, and it 'may' serve to protect us when similar
situations occur in the future. Despite all of this, however, we
don't see any way that a BUY can be generated for Friday ... and so
we are staying Short.
-
Model
Observations (Wednesday, 4/30/08):
Our
Short held up well on Wednesday, although it was a 'nail-biter'; our
(optional) protective Stop would have been set at $48.05 (Open +
0.727%), and QQQQ actually exceeded this bv $0.01 (reaching $48.06)
– showing the importance of either allowing some margin for
error, or monitoring the market in real-time when the Fed
announcement is issued. Our QQQQ Shorting drawdown decreased today
to -3.92%,
with about 2.70% of that resulting from the overnight
Google-engendered surge, and 1.2% of subsequent slow declines. The
Short continues for Thursday.
-
Model
Observations (Tuesday, 4/29/08):
Although
we got the tone right for the Dow (DIA) and the S&P (SPY), the
big 'techs' (QQQQ) continue to ratchet slowly upward. Apparently
there are some large-scale purchases going on, perhaps in
anticipation of favorable news from the Fed on Wednesday, or perhaps
in the belief that the worst is over and stocks will not get any
cheaper. In any event, our QQQQ Shorting drawdown increased today to
-4.78%, with about 2.70% of that resulting from the overnight
Google-engendered surge, and 2% of subsequent slow declines.
Wednesday will, of course, be a tricky day; although our Short
continues, we have a protective stop in place in the event that the
market turns against us in a major way once the Fed announcement is
made.
-
Model
Observations (Monday, 4/28/08):
The market continues to hold at elevated levels (in spite of
astronomical oil prices), and this is probably partly due to the
positive anticipation that leads up to a Fed announcement (which
takes place on Wednesday). Nonetheless, although it is not a lot of
fun to hold a protracted Short position when the market shows such
tenacity, we feel that the market is steadily weakening and is
overdue for some consolidation. Our Short continues for Tuesday.
-
Model
Observations (Friday, 4/25/08):
The market continues to show surprising tenacity, although we
perhaps saw some cracks in the edifice on Friday when QQQQ ended
lower. The drawdown on our current short is now at -3.79%,
although 2.74%
of
that occurred between the Close on the 17th
and
the Open on the 18th
following
Google's announcement. In other words, we are only down about 1%
more than we were on the morning after Google's announcement (an
event over which we had no control whatsoever), and so we don't seem
to be in danger of suffering significant additional losses. In any
event, the Short continues on Monday and we may see some continued
reduction in our drawdown. In the future, however, we 'may' take
note of those days in which Google (or Microsoft) is going to come
out with an earning's announcement, and set up a protective stop
'just in case...'. Such special measures, of course, go well beyond
the province of a market 'timing' algorithm.
-
Model
Observations (Thursday, 4/24/08):
The
market continues to show surprising (to us) strength, reflecting the
prevailing BLUE market color code and our neutral MTI value (1.0).
Nonetheless, our Short will continue on Friday, and once again the
after-hours action today suggests that the Short may work better
tomorrow. In any event, our current drawdown is now at -4.05%
(most
of which is due to the post-close Google rally). Our QQQQ Voted
Signal has now reached a record accuracy (back-tested) of 84.5%, but
it may be almost impossible to push this level higher without
increasing drawdowns and decreasing annualized gains. This number
does seem, however, to represent a near-optimal balance between
competing factors, and it has the advantage of reducing our trading
frequency by a significant amount.
-
Model
Observations (Wednesday, 4/23/08):
The
market continues to 'hang tough', reflecting our BLUE market color
code and our neutral MTI value (1.0). Nonetheless, our Short will
continue on Thursday, and the after-hours action today suggests that
the Short may work better tomorrow. In any event, our current
drawdown is now at -3.13%
(ALL
of which is due to the post-close Google rally). Our model
originally wanted to go Long for Thursday, but it seemed clear to us
that this was predicated on some very slim evidence; we found a
model enhancement, however, that preserved the Short while at the
same time boosting our overall back-testing gains ... and raising
our back-tested accuracy figure to a new high of 83.11%. Our goal
now is to develop an 85% model (combined Long/Short back-tested
trading accuracy of 85%). Whether or not we can achieve this is
still not clear, but 84% will likely be do-able and we will give it
a shot.
-
Model
Observations (Tuesday, 4/22/08):
Today was a pretty good day for our 'short', reducing our drawdown
to a more palatable -2.00%.
Our shorting signal continues for Wednesday, and perhaps we can
reduce our recent loss still further.
-
Model
Observations (4/21/08):
In hindsight our Voted QQQQ Model would have picked a BUY for
Monday, and this is 'ok' because we want our model to continually
refine itself, but we are glad that we did in fact have an official
'short' for Monday. Why? Well, it was overall a rather negative day
and thus we got the tone right, and second, we don't like it when
our model switches 'too' often (and especially when we are 'in the
hole') and third, because the market looks even weaker for Tuesday.
In any event, our Short continues ... and with the current drawdown
at a still not-too-bad -3.54%.
If it had not been for the Google-triggered after-hours surge we
would be quite pleased with the recent call; as it is, it's the hand
we were dealt.
-
Model
Observations (Friday, 4/18/08):
As it turned out, most of the damage had already been done by
Friday's Open. Google's positive announcement, and the market's
irrational response to it, had turned our 0.31% shorting gain (as of
Thursday's Close) into a -2.29%
shorting
loss (as of Friday's Open), that widened slightly by the Close to
give us a net -2.82%
shorting
loss. As it turned out, it 'would' have been productive to go Long
at Friday's Open, but our signal called for a continued Short and so
we were gratified to see that the additional damage was rather
slight. Our Short continues for Monday, and we are hopeful that the
damage will begin to reverse.
-
TimerTrac
Performance:
If
you look at our actual QQQQ performance over the past 6 months
(click the TimerTrac medallion on our website), you can see that we
are now up about 19% (since 10/16/07), versus a Buy-and-Hold
strategy being about -17% 'down'. This gives us a 36% differential,
and we believe that this 6-month spread will continue to widen as we
go forward. We will keep close track of this 6-month statistic
because only 'recent' performance counts for anything in this
business. Our model in effect changes every day (hopefully becoming
more accurate in the process), and by keeping tabs on this 6-month
'moving' statistic we will have an early warning sign if things
begin to go awry. Our number yesterday was 37%, and today's number
is 36%, and this is because in a 6-month period an old day drops off
every time a new day is added. So, this statistic will oscillate
around a bit from day to day. In any event, 36% is good ...
translating into a 72% lead over Buy and Hold during the course of a
year. This differential makes a good statistic because it shouldn't
be affected in a major way if the market is trending up, or down, or
just oscillating. If anything, the differential 'should' widen
considerably during choppy markets as the advantage of 'timing'
manifests itself.
-
Model
Observations (Thursday, 4/17/08):
The market showed quite a bit of resilience on Thursday, perhaps
reflecting the rather high MTI value that we had calculated.
Nonetheless, our Short 'worked' OK, giving us a net gain of 0.35%
for the day, following a gain of 0.91% from our closed-out Long. Our
Short continues for Friday, and it will be interesting to see how
well the fanatic buying that followed Google's after-the-close
announcement will carry over into Friday.
-
Model
Observations (Wednesday, 4/16/08):
There really wasn't much doubt about today's outcome (last night's
set of indicators were just 'too' good), and we had a massive win
that put our recent Long firmly back 'into the Black'.
-
Model
Observations (Tuesday, 4/15/08):
Today was a clean 'win' for our QQQQ signal, although it was sorely
tested. Even better is the fact that we are staying Long for
Wednesday, and have a good shot at breaking back into the Black
after our losses on Friday.
-
Model
Observations (Monday, 4/14/08):
Our Long was unable to gain traction on Monday (although there was
no dramatic weakness either), and as a result our drawdown deepened
slightly to -2.09%,
but the situation 'ought' to improve considerably on Tuesday –
and our Long continues.
-
Model
Observations (Friday, 4/11/08):
Right into the jaws of death ... our new Long got off to a terrible
start, although the “EXIT” (2.644% below the Open) did
not trigger, leaving us Long going into Monday. The previous short
wound up, however, with either a gain of 1.01% (if the position was
closed out at Friday's Open), or a loss of -0.31% (if Thursday's
EXIT had been taken). TimerTrac, of course, will show the more
favorable outcome in its graphs since we are NOT able to apprise it
of protective stops, EXITs, or 'profit targets'. Although our model
continues to do well, it is interesting that it HAS NOT changed its
mind about Friday; we remain convinced that if had not been for GE
that the market would in fact have risen on Friday.
-
Model
Observations (Thursday, 4/10/08):
Our new Short got killed on Thursday and triggered an EXIT to cash
at $45.62 (Open price of QQQQ + 1.516%). However, we are reverting
to a LONG on Friday, and our Voted QQQQ Model 'now' (too late, of
course) sees Thursday as a BUY as well. So, our Short ended up with
a net loss of -0.31%
(versus
a Buy-and-Hold that would have gained 0.31%), and we will start our
new Long on Friday morning at QQQQ's Open price. It is interesting
to note, however, that in 'retroactively' deciding that Thursday
should have been a Long (which would have been very profitable for
us), the bottom-line back-tested Long/Short accuracy has actually
risen to a full 82.8%. We don't like our model changing its mind
retroactively (although this is an inherent aspect of the modeling
process), but what would be worse is if it DID NOT change its mind.
-
Model
Observations (Wednesday, 4/9/08):
Our new Short worked well today, and we were able to close out our
previous Long with only slight additional damage. The final drawdown
for the Long was -0.63%
(or
-0.32%
if
the (optional) protective stop had been taken on Monday). We are
Short once again for Thursday. Notice that although our Profit
Trigger was 'hit' at $44.91, it effectively made no difference to
our 'score' since QQQQ closed at essentially the same price
($44.90).
-
Model
Observations (Tuesday, 4/8/08):
First, in retrospect our model now signals a Short for Tuesday. That
being said, what counts is that our 'official' (TimerTrac'ed) signal
for Tuesday had been a BUY. On that basis our current drawdown is
-0.59%,
or -0.28%
if
the (optional) protective stop had been taken on Monday, and 'IF'
QQQQ had been re-bought prior to the Close; if one had delayed
re-buying until Tuesday's Open, then the picture is even better
because there was some gain during the day. In any event, we are
going Short for Wednesday.
-
Model
Observations (Monday, 4/7/08):
Our (optional) protective stop was appropriate today because of the
gradual slide that followed the strong opening. For QQQQ this would
have 'saved' about 0.31% (Open price – 0.537%) if QQQQ had
been re-bought prior to the Close (our standard procedure), or xxx?
If QQQQ is re-bought at Tuesday's Open. Without the Stop our new
Long is -0.02%
in
the Red, but our Long continues for Tuesday.
-
Model
Observations (Friday, 4/4/08):
We
officially closed out our Short on Friday morning (at the open price
of $45.77), which actually gave us a small -0.15%
'official'
loss on the trade. However, for the purpose of tracking our trading
accuracy, we will nonetheless count this as a Win since we had
emailed out a QQQQ signal change to a BUY prior to the market Open
on Thursday ... recommending not a belated switch to oa BUY, but
rather that a protective stop be put in place to protect the opening
gains. In any event, it was easily possible to turn a slight profit
on the trade, and a Long wouldn't have been very productive either.
-
Our
new Long got off to a good start on Friday, began doing very well
indeed up until late in the day, and then ended up with a modest
0.20% profit. There's nothing wrong with that, and we remain Long
for Monday ... .so we shall see.Model
Observations (Thursday, 4/3/08):
Very early on Thursday morning (6am CST or so) our Voted QQQQ model
managed to tune itself into a BUY situation; the fact that this took
hours of cpu time to happen shows how finely balanced the day would
be on Thursday. Since we were 'officially' in a Short position, and
since we DO NOT change our signal once TimerTrac has received it
(late on Wednesday night), we decided to issue an email in the
pre-market period warning subscribers to set close stops so as to
protect some of the Shorting gains that had been made by the time of
the market Open. We ourselves covered our Short at approximately the
Wednesday close price ($45.47 to $45.49) shortly after the market
open on Thursday, and we stayed in cash the remainder of the day.
Our 'official' position, of course, was to hold the Short the entire
day, and on that basis our current shorting gain as of Thursday's
Close price was a meager 0.24% (and this is the way that TimerTrac
will score it). Since we are now going Long for Friday, we are
likely to see at least a slight shorting Loss as we cover and go
Long on Friday morning.
-
Model
Observations (Wednesday, 4/2/08):
Our new Short got off to a wobbly start, but ended up producing a
0.46% gain for the day. Our Long closed out nicely as well, since
the market opened up a bit higher, giving us a gain of 4.70% from
the Long.
-
Model
Observations (Tuesday, 4/1/08):
There isn't much to say; our anticipation that the market would at
least hold on Tuesday, if not move up sharply, was exactly right ...
giving us a gain on this 2-day Long of 4.44% as of Tuesday's Close.
-
Model
Observations (Monday, 3/31/08):
Our
Short ended up decently (with a 2.37% gain), and our new Long got
off to an unexciting start ... with a meager gain of 0.16%. So far,
so good.
Model
Observations (Friday, 3/28/08):
Our Short once again did well, boosting our current shorting gain to
2.73%, and maintaining our recent 80% trading accuracy (for the last
10 trades). Notice that our MOS (Market Open Signature) analysis on
Friday morning (based upon the Prediction Table: PREDTABL) was borne
out during the day when our Short once again proved correct.
-
Signal
Change:
For
Monday, however, we are going back to Long. The justification is
thin (as far as we can see), but we have learned to trust our model
(especially when it decides to go Long); our current Protocol,
however, is to experiment with model 'extensions' that might keep
the model Long when in fact it has decided to go Short. Such model
extensions are only permitted, however, if the following conditions
are met: (1) the change to the model actually 'improves' performance
over the back-testing period (almost 6 years), and (2) the change is
'logical' (i.e., based upon our existing set of indicator values and
neural networks). We reject such model changes that attempt to force
a Long if there is NO back-testing support for such a change.
-
Trade
Triggers(TM):
We
are continuing our efforts to add as much protection as possible to
each trade by supplying (optional) protective stops (when the
position is re-established at the Close), profit targets (when
profits are taken and the position is re-established at the Close),
and trade 'exits' (when we go to cash and stay there until a new
signal is available). We have recently extended this logic so that
occasionally both a 'profit target' and a Stop (or Exit) are both
generated (covering both market extremes). Our goal is to increase
the bottom-line gains as well as to allow trades to be handled by
'Trade Triggers' (to use Ameritrade's nomenclature); for example, a
profit stop can be implemented by setting up a Trade Trigger to
occur when QQQQ reaches a certain critical value, at which time a
Trailing Stop can be automatically initiated that will trail QQQQ
upwards ... setting an increasing 'floor' that is $0.10 or $0.15
below the current price. In this way it is often possible to make
further gains by grabbing as much profit as possible as QQQQ
continues higher. At the same time that a Trade Trigger is in play
to grab profits (if QQQQ soars), a Protective Stop (or Exit) will
sometimes be specified that will protect profits in the case that
the market reverses. All of this is an ongoing research effort.Model
Observations (Thursday, 3/27/08):
Our Short did well today, boosting our gain to 2.24%, and locking in
our recent 80% trading accuracy (for the last 10 trades).
-
Model
Observations (Wednesday, 3/26/08):
Our new Short (initiated on Tuesday morning) has now moved into the
Black, even if only by $0.01. We will continue the Short on
Thursday, and hope that we obtain a more significant gain. Since our
'average' shorting gain is about 1.3% (based on back-testing), we
should 'expect' better results than our meager winnings thus far.
-
Model
Observations (Tuesday, 3/25/08):
Our
new Short got off to a weak start today, handing us an initial
drawdown of -0.27%.
Still, we are holding the Short into Wednesday, and hopefully we
will begin to make headway. Our previous Long finalized with a net
gain of 8.28%, which is a superb gain.
-
Model
Observations (Monday, 3/24/08):
One
more day of “Long” puts our recent gain up to 7.87%,
although we would have lost -1.38%
of
that if we had taken the 'profit target' at $43.97 today (and then
re-bought at the Close). We are going Short for Tuesday, however.
-
Model
Observations (Thursday, 3/20/08):
Our Long gain is now back up to 4.36%; while it would have been
ideal if we had called a Short for 3/19 (rather than continuing our
Long), we had nonetheless set a cautionary tone for the day ... and
in our continued model optimization process the QQQQ back-tested
signal now does indeed show a Short for 3/19/08.
-
Model
Observations (Wednesday, 3/19/08):
Our concerns about the viability of our Long on Wednesday were
justified. We didn't reach our 'profit target' (at $43.95), which
would have justified taking profits and going to cash, and since we
didn't have a 'protective stop' set for Wednesday we were
'officially' forced to hold the Long QQQQ position for the entire
day. However, we did warn subscribers to protect their profits if
the market showed weakness (which it did early on), and hopefully
most of you went to cash before QQQQ had dropped much below
Tuesday's Close price of $43.33. Our model was delicately balanced
about Wednesday's call, and in fact it now sees Wednesday as a
Short; this is fairly typical behavior for this type of model when
the indicators are exactly balanced (positive and negative forces
are roughly equal). It is frustrating when a model changes its mind
in hindsight, but it is also indicative of a 'good' model .. one
that is continually adapting to changing market conditions.
-
Model
Observations (Tuesday, 3/18/08):
Although our Long didn't look very solid for Tuesday, we felt better
knowing that it was a 'Fed' day, and as everyone knows it turned out
to be a very good day for a Long indeed. Our current gain on this
recent Long is 4.94% (and it gives us a combined gain of 11.29%
since 1/25/08).
-
Model
Observations (Monday, 3/17/08).
Our Short finished up in fine style, giving us a gain of 2.23% as of
Monday's Open, and our new Long got off to a good start as well,
producing an initial gain of 0.46%.
-
Model
Observations (Friday, 3/14/08).
Our Short finally dug itself out of its hole on Friday, leaving us
with a negligible drawdown of -0.05%.
There was a profit taking opportunity on Friday that would have
allowed us to exit with a good profit, but our model didn't set a
'profit target' and so we held the Short until the Close. For Monday
we are reverting to a Long signal, although our model is probably
figuring on a very short-term bounce in anticipation of the Fed
announcement ... and NOT because the indicators look particulary
good – which they DO NOT.
-
Model
Observations (Thursday, 3/13/08).
Our reluctance to go Short turned out to be well-founded; the market
opened significantly down, paring our recent Long gains to a mere
0.74%, and starting our Short on a day when the market tenaciously
fought its way upward later in the day ... giving us an initial
shorting drawdown of -1.67%.
Nonetheless, everything pointed to a lower Close (including our MOS
analysis in the Prediction Table – predtabl.txt), and we did
but follow the 'probabilities'.
-
Model
Observations (Wednesday, 3/12/08).
Our continued Long looked very good until late in the afternoon, but
as the market closed with the Nasdaq somewhat in the red our current
Long gain was whittled back to 1.88%. And now, for Thursday, we are
once again going Short. As always, we explored model enhancements
that might have justified staying Long, but based on our indicators
we were unable to come up with anything that would keep us Long ...
without seriously degrading our back-tested performance. So, Short
it is.
-
Note
that our current Long gain of 1.88% is well in line with our
back-tested figures for the Voted QQQQ Signal, where Long gains tend
to average between 1.65% and 1.83%. In other words, our algorithms
force us to be content with relatively small, but hopefully steady
gains.
-
Model
Observations (Tuesday, 3/11/08).
As it turned out, it made little difference whether we were Long or
Short on Monday – as long as we were Long on Tuesday; if we
'had' shorted on Monday, then we would have hated this morning's
Open, and it would have been difficult to switch to a BUY when the
market opened so high. So, all in all we did about as well as we
could have hoped. Also, our model (after another day of
optimization) now once again sees Monday as a “Long” –
our posted signal for Monday. Our most recent trade now has a gain
of 2.17%.
-
Model
Observations (Monday, 3/10/08).
As is often the case, our QQQQ Voted Signal model now retroactively
'sees' Monday as a Short. This is pretty typical behavior when we
have poorly sampled regions, that is, combinations of indicators
that have occurred very rarely during our back-testing interval; the
model 'learns' a bit more about that region, but often
after-the-fact. So, we now have a -1.53%
drawdown
on our recent Long, but since we are maintaining the Long position
into Tuesday we have an opportunity to knock this down a bit ...
hopefully.
-
Model
Observations (Friday, 3/7/08):
Our Long signal worked satisfactorily on Friday, racking up a gain
of 0.36%. Note that our “Exit” did not trigger (QQQQ
Open -2.2%) on Friday, although if we had been clairvoyant we would
have held our Short until QQQQ had dipped significantly prior to
rising near the Close – but there is no way to do this!
-
Model
Observations (Friday, 3/6/08):
Our Shorting signal was just the ticket today, racking up a gain of
2.14%. Since we are switching back to a Long for Friday, however,
this gain won't be finalized until we get the Open price. Going back
in time our recent track record is: (L) +0.23%, (S) -1.39%,
(L) -1.21%,
(S) 1.67%, (L) 1.36%. So, for the past 6 trades we made gains 4 out
of 6 times (a 67% trading accuracy), and our net gain is 2.80%. This
is 'OK' performance, but it would have been much better if we had
refrained from trying to 'improve' the model by making minor
adjustments whenever it went against our 'gut feel'. For the time
being we are keeping our hands off and letting it do its thing
without any second-guessing.
-
Model
Observations (Wednesday, 3/5/08):
Our previous Short lost some money, but so far our 1 day Long is up
0.30%, and hopefully we can gain a bit more if QQQQ gaps up at
Thursday's Open ... at which point we will once again go Short. We
expect a good-sized gap up based on our #L parameter, but we are
fairly confident that the market will then steadily lose ground.
-
Model
Observations (Tuesdsy, 3/4/08):
Switching to a Long (erroneously, according to the newly optimized
model) on Monday set us up for a bit of whipsawing, although that is
a fairly rare occurrence with our thermodynamically-based model. The
Short on Tuesday hurt us, of course, because we not only came in
'Long', but Cisco's announcement caused a late-day rally that
brought QQQQ back into positive territory. There isn't much we can
do about it now, but we have our fingers crossed as we switch back
to a Long for Wednesday.
-
Model
Observations (Monday, 3/3/08):
Fortunately, our new Long signal didn't do much damage; our previous
1-day Short closed out this morning with a 1.67% gain, and the new
Long left us with a -0.58% drawdown with or without our 'optional'
Stop. The 'stop', triggered at the Open price of QQQQ –
0.615%, had no effect since the exit price was the same as the Close
price.
-
However,
we are reverting to a Short for Tuesday – and the 1-day Long
for Monday turns out to have been 'spurious'; we knew that the call
for Monday was balanced on a razor's edge, and allowing our Voted
QQQQ Signal model to optimize for many more hours ended up with it
finally deciding that Monday should have been a continuation of the
Short. 'Should haves' don't count, of course, but it is nice to see
the model so finely adjusted to current market realities.
-
Observations
(Friday, 2/29/08):
Our
Short call proved to be correct, although we would have preferred
selling off prior to Thursday's Close. However, life is never that
easy. Our Shorting gain is nonetheless at 1.60%, and we are
switching back to a Long for Monday.
-
Model
Observations (Thursday, 2/28/08):
Our drawdown crept back up slightly today, and is likely to be even
greater when we exit from our Long position at the Open on Friday,
2/29/08. Our current drawdown for our extended (1+ month Long) is
-5.47%
(no
Stops) or -3.27%
(Stops).
-
Model
Observations (Wednesday, 2/27/08):
Our signal today worked well, and our Long drawdown was reduced a
bit more. Our 1+ month extended Long drawdown is now at -5.09%
(no
Stops) and -2.89%
(Stops).
-
Our
portfolios continue to do well; our P1N (see PORTSTAT) is up 55.6%
since May 2007, and P2N is up about 39.1%. Both figures, however,
are based on transactions at the Close price ... and thus only
approximate actual gains, with the approximation becoming more
accurate as the time span increases.
-
Model
Observations (Tuesday, 2/26/08):
Our QQQQ Voted signal did well today, reducing our current Long
drawdown a bit more. However, our 'optional' stop was triggered when
QQQQ fell to $43.37 ... one penny under the Stop setting of $44.38.
While we are tempted to ignore this as a situation in which the
average trader would have noticed that QQQQ only briefly fell below
the target figure and then rose, the fact is that a Stop is a Stop.
On this basis our Stop would have cheated us out of 1.59% of gain
... giving us current drawdown figures for the past 5 weeks of
-5.59%
(no
Stops) and -3.37%
(Stops).
Notice, however, that the Voted Signal report has now revised its
Stop criterion to 0.643% (Open price of QQQQ – 0.643%), and
this slightly wider tolerance would have avoided being stopped out
today. We have an identical Stop ('optional') in place for
Wednesday.
-
Model
Observations (Monday, 2/25/08): Our
QQQQ Voted Signal model proved rather accurate today, and the
drawdown on our protracted 1+ month Long was reduced a bit further.
-
Model
Observations (Friday, 2/22/08): Our 'optional' protective
stop was triggered on Friday, and for the first time in quite a
while produced a net loss for the day (-0.64%). For our recent 1+
month protracted Long period this now gives us a drawdown of -6.47%
(if NO stops had been used), or -1.98% (if our optional Stops had
been taken). Nonetheless, over this period our model has been
steadily improving, and we expect its performance to continue to
improve incrementally as time goes on.
-
If
we compare our QQQQ trading performance versus a 'Buy and Hold'
strategy (confirmed by the TimerTrac(TM) graphs available by
clicking the icon on our website), we see that the trading gains
resulting from use of our standard signal (with NO stops) are up 3%
for the past 6 months, versus a -7%
LOSS
for
a Buy and Hold strategy. For the past 3 months, the figures are a
-1%
Loss (our
signal) versus a -12%
Loss
(Buy
and Hold). Although we don't have independent verification of our
signal-with-stops like we do for our standard signal, the
history.htm
Report
is a very accurate way of estimating its performance. If you examine
the figures from 11/19/07-2/22/08 you can see that for this 3-month
period our standard gain is 0.94%(non-compounded) versus 10.66%
(noncompounded) for our signal-with-stops. Since our TimerTrac
statistic (-1.0% Loss) is time-lagged by 3 market days, this
correlates pretty closely with our estimated 0.94% gain (including
the extra 3 days), and helps confirm that our 10.66% figure is
pretty reliable as well. Besides, our subscribers help ensure that
these numbers are accurately maintained; if we should accidently err
on the side of making our results look better, then such errors are
quickly pointed out to us :)
-
PORTFOLIO
Performance (5/4/07-present):
Looking
at the portacts.htm
Report we see that our portfolio winners (our Grail System
stocks/ETFs) for the past 9+ months have been the P1H and P1N-P5N
portfolios. The “H” portfolios don't look very good
because they are still working off some major drawdown that they
incurred last summer when they shorted very aggressively (before our
Voted QQQQ Signal had become mature), but the P1H (holding a maximum
of 1 stock) has managed to move up 14.48% during this period
nonetheless. For the “N” (normal) portfolios the 9+
month gains are:
-
P1N 47.88%
-
P2N 32.68%
-
P3N 7.70%
-
P4N 18.45%
-
P5N 7.06%
-
Some of the
larger portfolios have done decently as well: the P8N is up 10.02%,
and the P10N is up 8.36%.
-
The “U”
(ultra-conservative ETF portfolios) are up only 2-2.5%, while our
problem portfolios (the “C” and “E”
portfolios) are still rather negative, although rapidly digging
themselves out of their holes as a result of algorithmic changes
made several months ago.
-
Remember that
these performance figures are NOT quite kosher, however; they are
based on a simplified calculation that assumes that the stocks are
bought or sold at the current Close price rather than the next day's
Open price. This assumption is reasonable when longer periods of
time are considered since a stock is about as likely to 'gap down'
as it is to 'gap up'. By using this method, however, our algorithms
are much simpler, and stock splits are easily handled.
-
-
Model
Observations (Thursday, 2/21/08): Our
'optional' protective stop hit today, and would have saved about
1.29% (the Open was at $44.36, our 0.615% stop point was at $44.09,
and the Close was at $43.52). Our standard protocol if these stops
are taken is to close the position out at the trigger price (in this
case $44.09) and re-buy just prior to the Close (in this case
$43.52). In our recent 1-month “Long” interval, our
drawdown is now -6.84% (if no stops had been used) ... or -2.41% (if
Stops had been employed). Going into Friday we remain “Long”,
and once again with an 'optional' protective stop set at the Open
price of QQQQ – 0.615%. Our estimated probability for success
on Friday is 75%.
-
Our
Voted QQQQ Signal model continues to slowly evolve as we
occasionally add special case logic to 'force' the signal in the
direction that we 'think' it should go – but always based on
our complete set of indicators – and only when so doing
improves (or leaves invariant) our complete 5.5 year back-testing
history. We have been more prone to make model extensions when the
signal wants to go Short ... and especially when we have a Long
drawdown that we are trying to work our way out of. Currently,
however, our model stays long over 62% of the time, and this is
probably as high as it is ever going to get; remember that we
'believe' that the optimal number is 61.8% (the Fibonacci Ratio).
Furthermore, our model back-tests at a trading accuracy of about
79+%, and once again it appears that this is as good an accuracy as
it will ever be possible to achieve. The bottom line is that we are
about ready to cease making ANY tweaks to the model, or if we do, we
will try to set a close Stop in the event that we have nudged the
signal in the contrary direction. There is no way to avoid making
continual changes to a model (even if this just consists of daily
re-optimization) because the market itself continually changes. Only
a model that can adapt to changing conditions can ever hope to have
real predictive power.
-
Model
Observations (Wednesday, 2/20/08):
Wednesday was a reasonably 'normal' day; our signal proved correct,
and our drawdown was reduced slightly.
-
Model
Observations (Tuesday, 2/19/08): Our signal remains a BUY
for Wednesday, although our model has now retroactively decided that
Tuesday should have been a SELL(Short). Considering the wildness of
the day it is not surprising that the model has refined its
predictions and consequently changed its mind, and the fact that the
model now back-tests at almost an 80% trading accuracy lends
credibility to these occasional fine adjustments. The fact that QQQQ
ended up lower on Tuesday, while SPY and DIA ended up higher, shows
how difficult it was for the model to decisively 'call' the outcome.
-
Model
Observations (Friday, 2/15/08): Friday
was not good for us, but not so bad either; our drawdown increased
only slightly, and no 'tweaks' to our model are warranted.
-
Model
Observations (Thursday, 2/14/08): Today
Bernanke took the wind out of the sails of a pretty nice rally; we
don't believe that our QQQQ Voted Signal is to be faulted, and
although our drawdown has once again increased somewhat ... it
remains in an 'acceptable' range.
-
Model
Observations (Wednesday, 2/13/08): Our signal (QQQQ BUY)
proved to be quite correct on Wednesday, and this has now reduced
our combined drawdown for our recent 1-month Long (1/14/08-2/14/08)
to -3.97% ... OR -0.82%
if all posted 'stops' had been executed. Given the extreme variation
in the market during this period we are quite satisfied with this
level of performance. Notice that our model possesses many ideal
attributes: (1) it doesn't trade too often, (2) it doesn't whipsaw,
(3) it sets reasonable 'stops' that actually WORK, and (4) it
doesn't overreact to scary market drops, nor does it become euphoric
when the market seems irresistibly strong; it keeps a steady hand on
the tiller at all times.
-
On
Tuesday we made a significant improvement to the tuning algorithm
for our Voted QQQQ Signal: (1) the penalty function optimizes not
only the overall performance for the past 5.5 months (since
8/16/02), but also the recent performance (for the past 12 months),
and (2) the penalty function tries to keep the Long% figure (the
percentage of time that we have a BUY signal) at 61.8% or above –
relecting the Fibonacci Ratio. These two changes should help
stabilize our model, and ensure that it doesn't deviate much from
these 'ideal' characteristics during the tuning process.
-
Model
Observations (Tuesday, 2/12/08):
Today
our signal was less than perfect ... getting the sense of the market
correct (DOW and S&P up), but giving back some of our gains on
QQQQ. Nonetheless, our protective stop would have saved us 0.54%,
getting us out at $44.06, and back in at the closing price of
$43.82. Our 'optional' Stops continue to give us a higher level of
performance... saving us a full 3% during the course of our recent
“long” interval.
-
Market
Observations (Monday, 2/11/08): Our
signal was proven correct today ... including the prediction that
the 'techs' should outperform the rest of the market.
-
Model
Observations (Friday, 2/8/08): Our QQQQ signal proved
quite correct on Friday, giving us a solid gain despite weakness in
other sectors.
-
We decided to
recalibrate our QQQQ Voted Signal model over the weekend because it
was obvious that it was gradually tuning itself towards more
frequent signal changes ... with an actual decrease in trading
accuracy. This is the first time that we have seen this behavior,
but we now have an explicit procedure for keeping the model 'on
track'. You see, we have determined empirically that the percentage
of “Long” days should be approximately 61.8% (the
Fibonacci Ratio), and that the maximal trading accuracy for QQQQ
trades (Long and Short) should be close to 78%-79%. When our model
begins to deviate from these statistics, then it is time to
're-tune' by using a previous 'baseline' model and re-starting the
optimization process from that point. Note that although a trading
accuracy of 78%-79% doesn't seem that great at first sight, it is in
fact a superb figure. In fact, if the number were much higher then
we would suspect it ... figuring that there must have been some
error in our back-testing procedure. We doubt that accuracies above
80% are possible – with any timing model!
-
The current QQQQ
model has an average trading accuracy (back-tested over 5.5 years)
of about 77%, produces (back-tested) annualized gains ranging from
225%-244% (depending upon the use of our optional Stops), a 2008 YTD
QQQQ gain ranging from 27%-35% (theoretical ... based on
back-testing), an average Long period of 3.55 days, and an average
Short period of 2.21 days. Notice that the Ratio of Long/Short
periods is 1.61 – almost exactly the Fibonacci Ratio. These
facts give us confidence that our model is, in a sense, 'perfect'.
-
Without our
Stops (but 'with' our non-optional market EXIT at $44.84 on
1/25/08), our drawdown for the recent Long is at -6.65%.
“With” our optional Stops, however, our drawdown would
be only -4.04%. This is still bigger
than we would like, but it is an 'acceptable' figure. Bear in mind
that the market remains in a dangerous state that makes even the
best of timing algorithms look crappy. Our Mutual Fund Timing
Indicator has been in a SELL state since 1/2/08; our Market Color
Code has been almost solidly RED since 12/27/07 ... with only a few
days of ORANGE (another 'dangerous' color); our MTI value has been
less than 1.0 since 12/27/07, and has gone to zero (and negative)
for much of this time. Despite these obstacles, our objective is to
try to generate an accurate QQQQ signal 'every' day. It can be
frustrating trying to trade such signals during weak market periods,
however, and there are obvious ways to minimize risks and reduce the
stress of daily trading:
-
(1) The safest
way to proceed is to trade QQQQ only Long, and only when the MTI
value is >= 1.0 (and the Market Color Code is usually GREEN or
BLUE).
-
(2) If you are
going to trade when the MTI is less than 1.0, then you should
minimize your use of margin (especially when Shorting), and you
should consider following our 'optional' protective Stops and Profit
Targets.
-
(3) Also, even
when trading QQQQ Long, it is not a bad idea to keep an eye on the
Allocation Vectors and avoid being 'too' Long.
-
Bear in mind
that the use of 'Stops' is often the better solution to a declining
market than going Short. In other words, by setting Stops your daily
losses on a QQQQ Long position are kept to a minimum, while at the
same time you are able to stay fully invested ... and poised to take
advantage of a rebound. Going Short may 'seem' like a good idea at
such times, but it is much more hazardous. Our QQQQ Voted Signal
(voteqqqq.htm) is pretty astute; it 'knows' when to play the Long
side (with Stops) ... and when to go Short. We are currently
investigating the merits of a possible 'tweak' to our Protective
Stop logic, however, and this would be to re-buy at the next Open
when a Protective Stop is hit, rather than the Closing price –
WHEN the MTI value is < 1.0. This strategy has not yet been
confirmed, but it would tend to avoid losses when the market 'gaps
down' (as it is likely to do under these low-MTI conditions), and
would leave you ideally positioned to go Short in case the signal
changes direction for the next day.
-
-
Market
Observations (Thursday, 2/7/08): Our
drawdown was somewhat reduced today, but the market action also
helped vindicate some of our 'optional' Stops and Profit Targets.
Remember that our intra-day Stops and Exits are NOT accounted for by
TimerTrac, and in fact we do pretty well without them as the graphs
will show ... especially if you limit the graphs to the past 6
months or so when our model became rather mature (we look 'really'
good if you set the Graphing interval to the last 3 months).
-
Model
Observations (Wednesday, 2/6/08): The
market was doing decently today until an unwelcome comment by a
member of the Fed rekindled fears that interest rate cuts were no
longer likely. This has increased our drawdown for our current
“Long”, which is now back to a painful level. This shows
that no timing method is perfect when it comes to eliminating
drawdown, but it should be borne in mind that our Mutual Fund Timing
Indicator has been in a SELL state since 1/2/08, and that our market
color code has been almost solidly RED for this entire period. With
MTI values periodically dipping into the negative range, it is
almost impossible to make headway at the present time.
-
Note
that according to the TimerTrac service (see the TimerTrac medallion
on our web site), our net gain for the past 3 months is about -1%.
This is using our 'straight' signal, with all trades made at the
Open price, and with absolutely NO 'stops', 'exits', or 'profit
targets'. This compares well to a Buy and Hold approach –
which would be down about -17% for the same 3 months.
-
Market
Observations (Tuesday, 2/5/08): An
adverse manufacturing report sent the market down sharply today. Our
drawdown on our recent 'combined' Long is now -6.5%,
or -5.32% if our optional Stops had been used (see the history.htm
Report). Our optional Stop on Monday would have had us re-buy at the
Close on Monday – and that would have been disadvantageous
since the market gapped down so much at today's Open. We are
considering a strategy change for these optional Stops that would
set the re-buy point at the Close (if the MTI is >= 1.0) ... or
at the next Open (if the MTI is < 1.0). We plan to run some
analyses to see how the strategy might best be adjusted based upon
the prevailing MTI value.
-
Model
Observations (Monday, 2/4/08):
The market (and QQQQ) slumped today, obviously disregarding our Long
signal :) Nonetheless, the drawdown on our 'combined' Long is only
at -3.59%, and we would have saved an additional 0.77% by taking the
optional Stop on Monday (Open price of QQQQ – 0.615%) and
rebuying at the Close ... giving us a drawdown in this case of only
-2.82%.
-
Model
Observations (Friday, 2/1/08):
Once again the market behavior reflected our signal rather
accurately... and further reduced our recent drawdown: our drawdown
had reached -5.70% when we made an “EXIT” to cash. Our
subsequent re-buy (Long continuation) has now made a gain of 3.57%,
giving us an aggregate drawdown of -2.13% for the entire trade.
-
Model
Observations (Wednesday, 1/30/08): The
performance of our Voted QQQQ Signal was excellent on Thursday. Yes,
the market did retreat after showing strong initial gains in
response to the Fed rate cut, but QQQQ did quite well nonetheless
... dropping by only $0.04, and validating our statement that we
thought that the tech stocks would do well. Given that it was a
'Fed' day, the volatility doesn't surprise us. Our “EXIT”
was not hit, and thus we remain Long going into Thursday.
-
Model
Observations (Tuesday, 1/29/08):
Our signal seemed quite appropriate on
Tuesday, and therefore there is no need for further commentary on
the performance of our QQQQ Voted Signal. Although we are not
satisfied with the recent performance of the model (e.g., because of
the -7% drawdown that was reached at one point), it is not yet clear
what, if anything, could have been done to avoid it ... other than
simply staying in cash when the MTI value is zero (or negative).
-
Market
Note: One of the hallmarks of an 'ideal' timing model is that
it doesn't whipsaw, and that is one of the outstanding
characteristics of our QQQQ Voted Signal model; it does necessarily
allow some drawdown to occur, of course, but the model tries very
hard to avoid switching to a 'Short' just when a rebound is likely
to occur that will reduce the drawdown. Our recently added “EXIT”
(to Cash) feature is now a key part of this kind of strategy; it
tries to limit Long drawdown by ducking out on the worst of the
dips, while at the same time staying in a position to catch a
rebound. As a result, our net 2008 gains to date vary from -7.16% to
-3.28%, depending on whether our optional stops were NOT used ... or
WERE used. This is in sharp contrast to the -14.0% loss that would
have resulted from a simple Buy and Hold during this same period
(12/28/07-present).
-
-
Model
Observations (Monday, 1/28/08): So
far our “EXIT” on Friday (at the Open price of QQQQ –
2.2%), and our reestablishment of the Long at the Open on Monday,
appears to have been a prudent move. Our recent Long drawdown has
now been lowered to -5.0%. This gives us a net 2008 gain/loss
(12/28/07-present) that varies between -7.16% and -3.28%, depending
upon whether our various optional stops were NOT taken ... or were
taken. This is in contrast to a -14.0% loss that a straight Buy and
Hold of QQQQ would have produced during this same period. So, we are
running a net loss for the year ... but the drawdown is a heck of a
lot lower than that of a Buy and Hold. Moreover, our MTI has been
zero or negative for most of this time, the Market Color Code has
been RED, and our Mutual Fund Timing Indicator has been a SELL
(since 1/2/08). Thus it is not surprising that it has so far been
impossible to make a net gain. During such subzero temperature (MTI)
periods, it is a whole lot easier to just stay in cash...
-
Model
Observations (Friday, 1/25/08): On
Friday our new “EXIT” logic called for going to Cash at
the trigger point of the Open price of QQQQ – 2.2%. This exit
at $44.84 avoided the subsequent continued drop, and left us with a
drawdown of -5.70%. The Long will 'resume' on Monday when we
re-establish the position at the Open price of QQQQ.
-
Model
Observations (Thursday, 1/24/08): Our prediction of an
'up' day for Thursday was borne out; our model is staying Long, and
slowly digging itself out of its rather substantial drawdown.
-
***
New (1/23/08).
We
have recently added a new kind of 'Stop' to the QQQQ Voted Signal
report (voteqqqq.htm): the “EXIT” Stop. This type of
“Stop” may occur during Long (Buy) periods when the MTI
value is unusually low. The 'trigger point' for this kind of Stop is
usually about 2% below QQQQ's Open price, and the idea is to go to
cash ... and stay in cash ... if one of these kinds of Stops is
'hit'. Our normal (and optional) protective Stops usually result in
a reacquisition of the position just prior to the Close, whereas in
these new “Exit” Stops the position will not be
reestablished the same day, but 'may' get reestablished at the next
Open – assuming that the signal hasn't changed.
-
Model
Observations (Wednesday, 1/23/08): We
got the major bounce that we had anticipated based on Tuesday's
data, although it didn't do much for QQQQ (which ended up down just
slightly for the day). The after-hours action, however, gives us
hope that QQQQ will lift further on Thursday, and begin to reduce
our drawdown somewhat.
-
Model
Observations (Tuesday, 1/22/08): Our
signal has not been perfect in the last few trading days except in
one important sense: it doesn't 'whipsaw' ... changing direction a
day late and thus incurring losses on both Longs and Shorts.
Generally this model will 'hang tough', permitting compensatory
market forces to help reduce our drawdown. Moreover, our Voted QQQQ
Signal model continuously 'learns'; the current model would have
gone short a bit more during the past few weeks, thus eliminating a
substantial part of our current Long drawdown. Such retroactive
tuning is of little help at the moment (with a drawdown of almost
-7%), but it shows that the model is able to switch direction in a
nimble fashion .... and potentially do an even better job in the
future.
-
Model
Observations (Friday, 1/18/08): A wild market day left
QQQQ almost unchanged (down just slightly); it wasn't the kind of a
day that we had expected, but what we got was sufficient to provide
some validation of our signal. We thus see no reasons to begin
second-guessing our model.
-
Model
Observations (Thursday, 1/17/08): Evidently
we haven't yet found the bottom, but nonetheless our Voted QQQQ
Signal remains a BUY ... in expectation of a strong rebound. Our
drawdown is just over -4%, but this remains well within historical
tolerances. As we mentioned previously, our model has tuned itself
to the point that it now considers Tuesday's market action a
'Short', and so our 'theoretical' drawdown would be lower than what
we currently have. So, we haven't spotted any model behavior that we
can fault, and so we are continuing to trust it. The tendency of the
model to do some altering of the past (resulting from back-testing)
is normal, unavoidable, and actually a positive thing since it shows
that the model is able to rapidly adapt to changing market
conditions.
-
Model
Observations (Wednesday, 1/16/08):
Our Voted Signal model (voteqqqq Report) guessed the overall 'long'
aspect of today's market correctly; unfortunately, late news
resulted in a moderate loss instead of what appeared to be a likely
market gain. There is therefore no reason to delve deeper, and we
believe that the market will show considerable strength on Thursday
... confirming the upward movement that we had expected on
Wednesday.
-
Model
Observations (Tuesday, 1/15/08):
Today was obviously not the day to cover our Short and go Long.
Admittedly, we did achieve a very slight profit with our recent
Short (+0.11%), but we would have greatly preferred staying Short
for at least one more day. Interestingly, the Voted Signal was
obviously much more finely balanced for Tuesday than we had thought
... and the model currently (after a brief back-testing exercise)
shows that Tuesday was indeed still a 'short'. This is obviously no
consolation now, but at least it suggests that there is no need to
rip apart our model and make major changes. Nonetheless, we are
going to analyze our model to see if there is any way that
protective stops can 'always' be set when the MTI is below various
critical values. In any event, although we were wrong about Tuesday,
we were not totally unprepared for a sell-off: (1) our MTI was only
0.2135 (obviously much, much below the neutral value of 1.0), (2)
our market color code was Red, (3) our maximum recommended Long
allocation was only 34% (for portfolios), and (4) our mutual fund
timing indicator has been in the Sell state since 1/2/08.
-
The
safest way to trade QQQQ is to take only Long positions (and never
Short), and to be even safer ... to enter a trade only when the MTI
is >= 1.0.
-
-
Model
Observations (Monday, 1/14/08):
We were right about the market regaining some lost ground on Monday,
and in retrospect it wasn't such a bad move to keep the Short going
throughout the day. If we 'had' switched to a Long at the Open, then
we would only have gained $0.19 (QQQQ Close – QQQQ Open),
amounting to 0.40%; admittedly, that would have been 0.40% saved off
the Short, plus 0.40% more gained from a Long ... but, of course,
our Voted QQQQ Signal did not call for such an insightful play.
Based on over 5 years of back-testing, the signal that we published
was most likely the 'best' one in view of the existing market
conditions.
-
Model
Observations (Friday, 1/11/08):
Our model pretty clearly predicted that Friday would be a 'down'
day, and in fact, QQQQ dropped enough in price to put our current
Short into the Black.
-
Technical
Note:
It
is interesting to look at our LSI and SSI indicators (shown each day
in our MKTSTATS report). The LSI is our Long Strength Indicator
(higher values suggest 'Long' conditions), and the SSI is our Short
Strength Indicator (higher values suggest 'Short' conditions). In
the last week or so of December our LSI value was running in the 2-3
range, while the SSI was about 0.25. This suggests a 'Long'
situation. Abruptly, on 1/2/08 things reversed; since 1/2/08 our LSI
has been < 1.0 while the SSI has been >= 2.0 ... and the SSI
has now spiked to 3.57 while the LSI has gone negative. This all
suggests that the market decisively changed direction on January
2nd.
-
Model
Observations (Thursday, 1/10/08):
Our 'profit target' of $47.93 was well chosen for Wednesday, and
anyone who exited to cash just before the Close (at the $47.93 mark)
would have reduced their net loss on our recent Long trade to -0.97%
-- instead of the -2.00% that would have resulted when closing out
the Long at Thursday morning's Open price. Even so, our new Short
has taken a -1.18% 'hit' as the market tenaciously rose during the
day. It is too early, of course, to know just how 'wise' this new
Short will turn out to be. Note that we were also right about the
fact that shorting conditions for Thursday were likely to be 'poor',
and that holding cash was probably a better play. Note also that the
'profit target' of $47.72 for Thursday's Short never came into play
because QQQQ opened 'below' that level; while this 'might' have been
interpreted as a sign that the Short had 'already' run its course
before the Open, we have never performed an analysis along these
lines and so we cannot answer this question at the present time ...
but it's something we intend to analyze when we get a chance.
-
Model
Observations (Wednesday, 1/9/08):
On Wednesday our model proved accurate, and there is nothing further
to say. Our ;profit target' of $47.93 was attained, however, and so
anyone following that (optional) recommendation would now be in cash
... and prepared to go Short at Thursday's Open.
-
Model
Observations (Tuesday, 1/8/08):
Our QQQQ Voted Signal looked very solid until late in the afternoon,
but then several news developments killed our upward momentum ...
putting the market once more into a sell-off. There is NO way to
take account of such things in a mathematical model, and there is
thus no reason to try to 'fix' the model to predict the inherently
unpredictable.
-
Model
Observations (Monday, 1/4/08):
Our Voted Signal model did fine today; although our new Long is off
to a weak start, with a drawdown of -0.48%, this is a quite
acceptable number ... and we hope for better performance on Tuesday.
-
Model
Observations (Friday, 1/4/08):
The QQQQ Voted Signal model obviously worked perfectly for Friday.
The thing worth noting, however, is that it is often better to trade
QQQQ (when we have a signal change) in the pre-market period
(800-930am EST), rather than waiting for the Open – and
especially when there is going to be a market-moving news
announcement like Friday's Jobs Report. Even if you don't generally
trade during this early period, it is highly advisable to be
monitoring the market if you have to make a signal change!
-
Model
Observations (Thursday, 1/3/08):
Our new Long trade reduced its drawdown somewhat today, but still
leaves us with a net loss of -1.84%.
A bit of consolation, however, is provided by the fact that our
Voted Signal model (voteqqqq.htm) now shows a SELL/S(SHORT) for the
period from 12/27-12/31. Since we had done some creative 'nudging'
to produce a BUY signal during the 12/28-1/2/08 period, it is good
to see that the model has now tuned out our meddling and is doing
its own thing. We are really at a point now where it is best to
leave the model totally alone ... despite our occasional temptations
to get it to flip in the direction that we 'feel' that it should go.
The fact is ... that it is smarter than we are!
-
Model
Observations (Wednesday, 1/2/08):
Our observation that we had no way of estimating a success
probability for Wednesday's Long trade (because similar
circumstances had not hitherto come up in our back-testing history)
turned out to be prescient. In retrospect, our Voted Signal
algorithm has now decided that a Long call was inappropriate for
Wednesday, and thus it has 'learned' ... and will thus hopefully
avoid similar situations in the future.
-
Model
Observations (Monday, 12/31/07):
Obviously the Open was not the ideal time to cover our QQQQ Short.
Nonetheless, the new “Long” trade is only down -0.68%
...
and the trade continues for Wednesday.
-
Model
Observations (Friday, 12/28/07):
Our new Short acquitted itself well, although it made only a small
amount of money (0.37% on QQQQ). Now, we have a short-term reversal,
and we are going Long for Monday.
-
Model
Observations (Thursday, 12/27/07):
The market took a minor 'hit' today, apparently confirming the
negativity implied by our L/1 indicator switching from 1 to 0.
Moreover, the market situation changed enough today (for whatever
reasons) to throw our Voted QQQQ Signal back to a SELL/S(SHORT) –
and our Market Color Code to a RED condition.
-
Model
Observations (Wednesday, 12/26/07):
Overall it was a pretty good day for us; our QQQQ signal once again
proved accurate, and this has pushed our recent Long gains to over
5%.
-
Model
Observations (Monday, 12/24/07):
Once again our Voted Signal model has generated a winning signal,
and there is really nothing to add.
-
Model
Observations (Friday, 12/21/07):
Once again our Voted Signal model has generated a winning signal,
and there is really nothing to add.
-
Model
Observations (Friday, 12/20/07):
Our new Long signal was amply confirmed today, and there is little
to add to that. So far, so good.
-
Model
Observations (Wednesday, 12/19/07):
Today we had a near-perfect changeover from a Short to a Long, with
the Open price on Wednesday equalling Tuesday's close price.
Although we lost $0.01 on the day, the after-market action looks
positive and we expect a good day for the market on Thursday.
-
Model
Observations (Tuesday, 12/18/07):
The market on Tuesday reflected pretty much what we had forecast.
Our (optional) profit-taking target of $49.53 turned out to be a
reasonable exit point today, gaining us 0.70% over holding our Short
the entire day. In retrospect, now that we have a BUY signal for
QQQQ, it would have been better to have avoided re-shorting prior to
the Close ... and much, much better, of course, if we had had the
foresight to go Long when QQQQ appeared to have bottomed out. But,
obviously, such precision is NEVER possible when dealing with the
stock market. Our tracking statistics (see history.htm file)
generally show both the gains that would have been achieved without
the use of any stops or profit targets, and also the delta gains
that would have been possible if the stops had been followed.
-
Model
Observations (Monday, 12/17/07):
Our shorting signal was clearly confirmed today, giving us a current
gain of 2.66%. Our previous Long recently finished up with a 2.28%
gain (or 3.43% if stops had been used), and the Short prior to that
yielded a 2.82% gain. So, we continue to do well with our QQQQ
signal.
-
Model
Observations (Friday, 12/14/07):
It took a while before our shorting signal began to look reasonable,
but our signal proved to be correct by the end of the day. There was
certainly no 'hurry' on closing out the Long, however; by waiting a
while it would have been possible to sell QQQQ for better than the
previous day's Close of $51.50 (we actually sold at $51.31).
However, our tracking statistics (history.htm file) are based on
transactions at the Open price, and on this basis our new Short has
a gain of 0.23%, and our previous Long finished up with a 2.28% gain
(or 3.43% if stops had been used), and the Short prior to that
yielded a 2.82% gain. So, we have been doing pretty well of late.
-
Model
Observations (Thursday, 12/13/07):
Our current BUY signal seemed reasonable based on Thursday's market
action, but we now have a signal change to a SELL/S (Short) state.
The justifications for this change are still a bit weak, but after
running with a BUY signal for almost a month it is certainly time
for a change. In any event, this is certainly a warning to be
cautious.
-
Model
Observations (Wednesday, 11/12/07):
Give our signal credit for staying Long for Wednesday, although our
huge opening gains dissipated considerably as the day wore on.
-
Model
Observations (Tuesday, 12/11/07):
A protective stop would have been extremely handy today, but our
back-testing studies didn't suggest that one was desirable under
these conditions. However, our current Long (QQQQ) is still well in
the black, and we anticipate that we will soon regain some of the
ground lost today. It figures, of course, that it would be a 'Fed'
day that would trip our signal up.
-
Model
Observations (Monday, 12/10/07):
Our Forecast was reasonably accurate on Monday, including the
suggestion that the S&P and DOW would likely outperform the
'techs' (QQQQ). For Tuesday our MTI value is higher, and we 'expect'
that the market should fare well.
-
Model
Observations (Friday, 12/7/07):
The market's behavior on Friday reflected rather accurately our
weakened MTI value. Since our MTI value for Monday is even lower,
however, we are hoping that the upcoming Fed announcement (perhaps
reflected in our short-term neural network projections) will
counterbalance any tendency for the market to slump.
-
Model
Observations (Thursday, 12/6/07):
Our
signal proved correct once again, but despite a continued BUY signal
for Friday, our suspicion is that we will have to give up at least a
portion of our current gains on Friday.
-
Model
Observations (Wednesday, 12/5/07):
Our signal proved correct for Wednesday and the Voted QQQQ Signal
model (VOTEQQQQ) continues to look very sound.
-