Our QQQQ signal (the Grail System signal) remains a ‘buy’ for the 20th day, and our mutual fund signal remains in the ‘buy’ st

FREE TRIALS: If you are interested in a 30-day Free Trial of our Strategist Newsletter, then please send an email to: Schulenberg & Associates, Inc., 30-day Free Trial. We are in the process of upgrading our two Tutorials, Part 1 and Part2. Part2 contains most of the explanatory material that has been published in our TimerTrac Broadcasts – and in a more concise and readable format. Additional information can be found on our web site: http://www.schulenberg.com

Schulenberg.com

Weekly Status (as of 9/14/07)

Stock Signal Status. Our QQQQ Voted Signal remains a SHORT for Monday, and we have been Short since the market Open on 9/13/07; admittedly, there is only a +0.16% gain at present (although our cumulative Long/Short gain since 8/22 is about 8%), and it is dangerous to go Short when a Fed meeting is imminent, but we are staying Short (with a 1% protective Stop) for at least one more day. Our indicators have shown considerable underlying market strength – 'except' for the tech stocks. Our MTI values this past week have been 1.14(M), 1.58(Tu), 1.20(W), 0.93(Th), and 0.50(F) (our ‘neutral’ value is 1.0 … at which level we expect the market to ‘tread water’). With the MTI value having just plummeted to a rather low value of 0.50, maintaining the Short on Monday seems quite reasonable. Our Voted Signal is becoming increasingly accurate at predicting ultra-short-term market trends, and the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report. In our expanded broadcast we include recent daily observations about this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm).

Mutual Fund Signal Status. Our Mutual Fund Timing Indicator (MFTI) switched back to a BUY on 8/22/07 (after having previously gone to a SELL on 7/26/07), went to a HOLD on 9/12/07, and the average 'score' for a set of 1100 mutual funds now has a value of 4.68 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is just below a HOLD state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 18 times.

Modeling (Findings/Insights).

Model Observations (Friday, 9/14/07): The market once again showed the inherent strength that is evident in our indicators (e.g., reflecting the MTI value of 0.9249), but we feel that a lot of this tenacity is due to the fact that the Fed meeting is imminent, and that the announcement will likely be 'market-friendly'. Admittedly, this is a dangerous time to be Short ... but the fact that our current QQQQ Short has been making money (albeit only +0.16% at this point) shows that models CAN make accurate predictions. The recent market strength has NOT been in the 'techs'. Now we have an MTI dipping to 0.4962 for Monday, and our Prediction Table suggests downward motion. So, we will carry the Short yet another day.
Model Observations (Thursday, 9/13/07): The past 3 weeks have demonstrated why our model is 'predictive' rather than 'reactive'; our QQQQ Voted Signal has accurately tracked virtually every market movement within this period. We went Short QQQQ this morning at the Open price of $49.30, and eked out an initial Shorting gain of +0.24%. This is a tricky time to be Short, especially with the Fed meeting next week, but we are becoming increasingly confident in the inherent accuracy of our method.: Our most recent trades (and their gains/losses) are:; Long: 9/7-9/12. Gain = 1.59%-1.96% (depending upon the use of Stops for the 2^nd figure).; Short: 9/5-9/7. Gain = 1.42%; Long: 8/22-9/5. Gain = 3.16%-4.56% (depending upon the use of Stops for the 2^nd figure).

This puts us up 6.41%-8.18% in a 3-week period. Let's hope this level of accuracy continues.

Model Observations (Wednesday, 9/12/07): A perfectly good market day was spoiled by rising oil prices and some less than stellar earnings projections. Our QQQQ Voted Signal model is switching to a Short for Thursday, but our indicators continue to look fairly decent; at this point we don't anticipate much of a pullback, and we still think that the market will soon resume its upward trend ... with the Dow once more topping 14000. But, since our methods are designed to look only one day into the future, we will simply suggest that Thursday is likely to be a 'down' day. The back-tested trading accuracy of our QQQQ Voted Signal continues to hold at about 79% (for about 400 trades over the past 5 years), and the Long/Short percentage is still approximately 60%/40% (see the voteqqqq.htm report).

Model Observations (Tuesday, 9/11/07): Our QQQQ Voted Signal model proved to be quite correct about Tuesday, and the issue was never seriously in doubt. For Wednesday we appear to have an even stronger situation, although we tend to become more apprehensive the more 'positive' our projections become. Our accuracy in the past few weeks has been excellent, and we have even profited from our 'misses'; our model is capable of learning from its mistakes, and our practice is to conduct 'post-mortem' analyses of our failures so that we can increase the accuracy of our model when confronted with similar situations in the future.

Model Observations (Monday, 9/10/07): We had been forced to make a judgment call on setting the signal as a BUY for Monday; there was simply NO back-tested (5-year history) guidance for setting signals based on the unusual confluence of market factors that we had to deal with. In other words, both a Long and a Short were 'compatible' with our existing model (although our neural networks suggested that “Long” was the appropriate signal). It turned out, however, that the optimal choice for Monday was a 1-day Short (Short at $48.62 at the Open and make 0.86% for the day), and our model 'now' (in the hindsight afforded by model optimization with the new data) would have gone that route. In any event, our current BUY took only a very minimal 'hit' on Monday, and we have a rather 'clean' BUY signal for Tuesday ... so it appears that either choice for Monday was 'reasonable' in some sense. So, we are going into Tuesday, 9/11, with an existing BUY for QQQQ – despite the cloud of foreboding that will probably always swirl about this particular date. Our back-testing suggests a 65+% probability for a profitable QQQQ Long on Tuesday.

Modeling (General Approach). Our method for calculating a Market Timing/Temperature Indicator (MTI) is unusual in that it is based on a ‘thermal’ model of the stock market. The Maxwell-Boltzmann Distribution governs the relationship between the absolute temperature of a gas and the distribution of its gas molecules in a spectrum of energy (speed) bands. Our Preprocessor views the stock market as a thermal system in which stocks and a spectrum of special portfolios correspond to gas molecules and energy bands. We believe that this method offers the best approach to actually ‘predicting’ the stock market direction, rather than simply ‘reacting’ to it.

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Weekly Status (as of 8/31/07)

Stock Signal Status. Our QQQQ Voted Signal remains a BUY for Monday. As measured by this indicator, the current BUY began on 8/20, although it wasn’t until 8/22 that this was confirmed by the switchover of our GR-L/GR-S QQQQ models to BUY states as well. And, of course, we had a 1-day Sell on 8/28. Our MTI values this past week have been 1.23(M), 1.07(Tu), 2.05(W), 1.20(Th), and 1.53(F) (our ‘neutral’ value is 1.0 … at which level we expect the market to ‘tread water’). We haven’t seen numbers consistently this good since the February/March time frame, and this leads us to suspect that the Dow may be set to power back above 14000 once again. Our Voted Signal is becoming increasingly accurate at predicting ultra-short-term market trends, and the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report. In our expanded broadcast we include recent daily observations about this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm).

Mutual Fund Signal Status. Our Mutual Fund Timing Indicator (MFTI) switched back to a BUY on 8/22/07 (after having previously gone to a SELL on 7/26/07), and the average 'score' for a set of 1100 mutual funds now has a current value of 6.51 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is in the Hold/BUY state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 18 times.

Modeling (Findings/Insights).

Model Observations (Friday, 8/31/07): We weren’t really optimistic that Friday would turn out as well as it did; although our Voted Signal model remained a BUY, our indicators were only mildly positive. However, the MTI value had spiked to 2.05 on 8/29, and this ‘may’ be viewed as a sign that the nature of this market has changed. Since May of 2007 we have had an unusual market … by ‘unusual’ we mean one that gives our thermodynamically-based timing model some difficulties … and this was the primary impetus for the rapid development of our Voted Signal model. These model changes were successful in coping with what we term a ‘cool’ uptrend (the soufflé effect), but now we seem to be back in a market regime that is ideally suited to our traditional methods.

Model Observations (Thursday, 8/30/07): Today the current strength (high MTI) of the market showed itself, keeping buying interest up (at least for most of the day) despite the ever-present concerns about subprime credit, housing, and consumer confidence. With an MTI value of 2.05 we were not seriously worried that the market was going to take a nosedive. We did see some initial tech weakness (QQQQ opening down), and this was in accordance with the negative indications from our Prediction Table (PREDTABL table II). Eventually, however, the very high MTI value was reflected in a steady climb in the techs (which helped bring the DOW up for a while as well). For Friday, however, our MTI has dipped sharply and we expect more of a tug-of-war kind of day.

Model Observations (Wednesday, 8/29/07): Our recent “long” is once again well 'in the Black' (and even more so if the protective Stop had been taken on 8/28/07). Despite the fact that our published signals have turned a profit, it would have been undeniably better if we had generated a Short signal for 8/28/07 (rather than a Long with a 0.50% protective stop). In fact, our current QQQQ Voted Signal model now does just that ... and we hope that future performance will reflect our steadily improving back-tested results. Our back-tested 5-year QQQQ trading accuracy has now topped 79%, although it is getting harder and harder to push accuracy any higher; we are probably close to the theoretical limits of what is actually possible in the realm of predictions, given the 'noisy' nature of a news-driven market.

Model Observations (Tuesday, 8/28/07): We were not altogether unprepared for the market dip on Tuesday; our Forecast had specified a color code of YELLOW, a prudent Long allocation of no more than 20%, and a max Long allocation of 60%. Furthermore, for
QQQQ we had a 0.50% Stop in place (based on a 0.50% drop from the Open price). However, a tighter stop would have been better, and a SELL(SHORT) signal would have been even better yet. There was a very tricky balance for Tuesday between a Long position (with a Protective Stop) and a Short stance, and today we ended up on the wrong side of the movement. Our QQQQ Voted Signal model will be less likely to be caught flat-footed like this in the future, however; we have made some minor algorithmic adjustments to our FOMC logic (the Fed released their Minutes on Tuesday), and this should help prevent our model from becoming quite so sanguine on 'Fed' days (and giving too little weight to the negative indicators).

Model Observations (Monday, 8/27/07): With the market situation negatively influenced by disappointing housing data, QQQQ fell substantially more on Monday than our worst-case projections would have suggested. Tuesday will be a tricky day – and especially because the Fed minutes will be released. Our QQQQ Voted Signal model is balanced on a knife edge between a continuation of the recent BUY, and a switch to a SELL. There is a very slight majority of data that has kept our BUY going, however, and this slim margin has resulted in the generation of an (optional) protective stop alert for Tuesday. Our MTI value remains decently high (at 1.2287), and the market could easily surge if the Fed minutes strike the right chord. As long as the MTI value stays well above 1.0 (as it is now) we aren't too worried about sharp dips – although the possibility, of course, is ever present.

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Weekly Status (as of 8/24/07)

Stock Signal Status. Our QQQQ Voted remains a BUY for Monday, and we are convinced that a firm bottom has now been established for this market. As measured by our QQQQ Voted Signal, the current BUY began on 8/20, although it wasn’t until 8/22 that this was confirmed by the switchover of our GR-L/GR-S QQQQ models to BUY states as well. Our MTI (Market Timing Indicator) value was very low up until Wednesday, when it abruptly spiked to 1.79 from its Tuesday value of 0.29; the Thursday/Friday values were 1.51 and 1.97, respectively (our ‘neutral’ value is 1.0 – at which level we expect the market to ‘tread water’). We haven’t seen numbers this good since the February/March time frame, and this leads us to suspect that the Dow may be set to power back above 14000 once again. Although our methods are geared toward predicting only a short time into the future, the fact that our mutual fund signal (which rarely changes) has just switched back to a BUY gives further corroboration that the market may have embarked on a new uptrend.

Our Voted Signal is proving adept at predicting ultra-short-term trends, and the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report. In today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm).

Mutual Fund Signal Status. Our Mutual Fund Timing Indicator (MFTI) switched back to a BUY on 8/22/07 (after having previously gone to a SELL on 7/26/07), and the average 'score' for a set of 1100 mutual funds now has a current value of 6.76 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is in the Hold/BUY state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 18 times.

Modeling (Findings/Insights).

Grail Signals and Portfolios (Friday, 8/24/07): Our portfolios are beginning to move back ‘into the Black’, based on their performance as measured since 5/4/07. In the PORTSTAT report (50 portfolios) it is clear that the best performing portfolios as a whole are our “U” portfolios (ultra-conservative investments in ETFs); these portfolios are up about 1.5-2% for the 5/4-8/24 period. On the other hand, among the “N” (stock) portfolios, the P1N portfolio is up 12.6% for this period, and the P2N is up 17.05%; these portfolios hold a single stock, or two stocks at a time, respectively. We are currently investigating why we have done so well with these very small portfolios, while their larger versions languish behind.

Model Observations (Friday, 8/24/07): Friday played out pretty much as we had anticipated; we now have a strong market that will most likely trend upwards for a while. Our QQQQ Voted Signal model is holding firm at a BUY, and we aren’t planning on making any more changes to this model for a while … other than addressing some long-standing questions about intra-day tactics: (1) if a protective Stop is ‘hit’, under which conditions should one re-establish the position prior to the Close, as opposed to waiting until the next Open?, (2) is there a way to set a protective stop to make things safer when the signal changes?, and (3) can both protective stops AND profit-taking measures be specified for the same day (to cover both possible extremes)?

The QQQQ Shorting Model (GR-S) has now converged to a new solution that no longer suggests that we will revisit our recent Lows; we thus no longer have that particular dark cloud hanging over us. Also, our Mutual Fund Timing Indicator (MFTI) has switched from a 'sell' (BEAR) state to a BUY state as well.

Model Observations (Thursday,8/23/07): Today was an unimpressive start for our new Long (a signal that is endowed with an extra measure of reliability due to the recent BUY signals coming from our GR-L/GR-S models), but we expect Friday to be a better day for us. We did note, however, that our NN1 neural network value (1-day QQQQ projection) had dipped into the negative range for Thursday, and this 'might' be something that we can use in the future to provide better entry timing for signal changes ... that is, we 'might' have been able to predict that the covering of the Short should have been deferred until after the market had had a chance to settle out a bit after the Open. Note, however, in the Voted Signal report that the current model (with our very recent algorithmic additions) would have gone Long on Tuesday morning (8/21/07), thus timing the recent market with high accuracy while at the same time maintaining excellent back-tested characteristics for the past 5 years. Such 'retroactive' model changes don't, of course, change the fact that our 'published' signals were less than perfect, but they DO help ensure that the model will perform much better when similar circumstances occur in the future.

Our QQQQ Voted Signal model (see voteqqqq report) continues to show superlative back-testing performance, and in particular a trading accuracy of about 78% for over 400 Long/Short trades over the past 5 years. While it may be possible to push this accuracy to 80% or thereabouts, we are probably very close to the limits of what can be accomplished given the unpredictable effects of after-the-Close (and before-the-Open) news announcements on the daily behavior of the market. We will settle for 78%, and hope that future actual accuracy is close to this...

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Schulenberg.com

Weekly Status (as of 8/17/07)

Modeling (Stock Signal Status). Our QQQQ Voted Signal is switching back to a SELL(Short) for Monday, and we fear that a firm bottom has not yet been established for this market. If this is the case, then QQQQ has a high likelihood of dipping back into the $44.00-$44.50 range (or lower). Our Voted Signal is proving adept at predicting ultra-short-term trends, and the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report. In today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm). This is not a time to be holding substantial stock portfolios; at MTI values this low (currently at 0.22 … and thus well below our neutral value of 1.0), only the nimble trading of a few ETFs like QQQQ is likely to be a profitable strategy.

Modeling (Mutual Fund Signal Status). Our Mutual Fund Timing Indicator (MFTI) switched back to a SELL on 7/26/07, and the average 'score' for a set of 1100 mutual funds now has a current value of 2.94 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is in a ‘SELL’ state at the present time (except, of course, for Bear Funds). A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 17 times.

Modeling (Findings/Insights). Model Observations (Fri.,8/17): For Monday it looks like the market will pull back somewhat, and we once again have a SELL/S(SHORT) signal for QQQQ. There is a good chance for some initial market strength (since our #L value is 32), and this would provide an opportunity to: (1) sell QQQQ, (2) set a close protective stop, or (3) go Short. Our indicators are so negative (MTI dropping, PREDTABL signal is SSSS, ...) that it is extremely unlikely that the market can eke out a win for the day. As always, however, shorting presents a lot more risk than just selling ... or setting a close stop.

What concerns us is that QQQQ never technically reached the low point that would be necessary if our GR-S (shorting) model is to retain its perfect (100%) back-tested accuracy. We might therefore dip once again down into the $44.00-$44.50 range for QQQQ.

Model Observations (Thur.,8/16): We 'may' have put in a bottom today; at least QQQQ dipped very close today to the entry price for our long-standing Short from our GR-S shorting model. Technically, we have not really had a Close price that would preserve the perfect 5-year (back-tested) trading accuracy of this shorting model ... but perhaps it was close enough to confirm that a bottom has been reached. Let's hope that we do not have to dip once again back into the $44.00 range for QQQQ in order to really lock in a bottom.

Notice that the posted low price for QQQQ today was $40.55. This is obviously some kind of 'glitch'; the actual low price (estimated by looking at a graph) must have been around $44.50-$44.55.

In any event, our Voted signal model returns to a QQQQ BUY for Friday -- after 6 successive days of SELL(SHORT) (as evidenced in our back-tested results, although not in our published signals, unfortunately).

Model Observations (Tue./Wed.,8/14-15): For the most part, our very low values for the Allocation Vectors have kept us from being overly Long in recent days, aside from the rather aggressive stance taken by our “H” portfolios (see PORTSTAT). Also, our Mutual Fund Timing Indicator has been in a SELL state since 7/26/07. What does tend to concern us when looking forward, however, is our QQQQ shorting model: GR-S. This model has a 100% back-tested shorting accuracy over the past 5 years, and it has been in a continuous Short state since 2/26/07. On 2/27/07 it would have initiated a Short at the open price of $44.36, and if this model DOES retain its perfect back-tested accuracy, then this means that QQQQ will eventually fall below $44.36 ... still another 3.4% or so. This is one of our few long-term gauges of market direction, and although it cannot be relied upon with a great degree of certainty (and it does obviously have a rather high back-tested drawdown), it is nonetheless a sinister backdrop against which the recent market trend may be viewed.

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Weekly Status (as of 8/10/07)

Modeling (Stock Signal Status). Our QQQQ Voted Signal is switching back to a SELL(Short) for Monday, despite our considerable (and ultimately futile) efforts to ‘coax’ it into a BUY state. We had several signal changes during this past week, and managed to avoid getting ‘whipsawed’ … actually ending up with a modest trading gain during this highly volatile period. Our Voted Signal is proving adept at predicting ultra-short-term trends; the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 68447% since 8/16/02), and in today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm). This is not a time to be holding substantial stock portfolios; at MTI values this low (currently at 0.54 … and thus well below our neutral value of 1.0), only the nimble trading of ETFs like QQQQ is likely to be a profitable strategy. When our Mutual Fund Timing Indicator switches back to a Buy or a Hold, then we will know that the current market malaise has ended.

Modeling (Mutual Fund Signal Status). Our Mutual Fund Timing Indicator (MFTI) switched back to a SELL on 7/26/07, and the average 'score' for a set of 1100 mutual funds now has a current value of 4.04 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is in a ‘SELL’ state at the present time (except, of course, for Bear Funds). A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 17 times.

Modeling (Findings/Insights). Since we have been expending a great deal of developmental effort on our QQQQ Voted Signals Report (voteqqqq), we have focused on the more important aspects of this QQQQ short-term trading signal in recent Broadcasts. We believe that we have now struck an optimal balance between the ‘predictive’ capability of this model (based on our ‘thermodynamic estimate of market strength), and the trend-following algorithms that ensure that we don’t ever lose sight of what the market is ‘actually’ doing. This balance is confirmed in back-testing, wherein our model continues to accurately track market movement in the highly volatile 2002-2003 time frame, and yet also locks in solidly to the strong uptrend that we have seen over the past 9 months or so. Another important characteristic of our Voted Signal model is that it doesn’t ‘whipsaw’; it is rather uncommon that the signal flips so often that we end up with back-to-back losses … e.g., losing on the “long” side, and then immediately losing on the “short” side.

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Weekly Status (as of 8/3/07)

Modeling (Stock Signal Status). Our QQQQ Voted Signal is switching back to a BUY for Monday, after going Short for Friday. We were Long (often with protective Stops) from 7/27-8/2, pinpointed the Short on Friday, and now anticipate a bounce on Monday. On 7/26/07 our MTI (Market Timing Indicator) plummeted to near-zero, causing our mutual fund indicator to switch to a SELL and our portfolios to drastically reduce their holdings. Our Voted Signal is proving adept at predicting short-term trends; the current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 49763% since 8/16/02), and in today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm). This is not a time to be holding substantial stock portfolios; at MTI values this low (near zero, and certainly well below our neutral value of 1.0), only the nimble trading of ETFs like QQQQ is likely to be a profitable strategy. When our Mutual Fund Timing Indicator switches back to a BUY or a HOLD, then we will know that the worst is over.

Modeling (Mutual Fund Signal Status). Our Mutual Fund Timing Indicator (MFTI) switched back to a SELL on 7/26/07, and the average 'score' for a set of 1100 mutual funds now has a current value of 4.25 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is in a ‘SELL’ state at the present time (except, of course, for Bear Funds). A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT . Although our primary stock signal (QQQQ) may change state every 3 or 4 days on average, during the past 5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 17 times.

Modeling (Findings/Insights). Since there is currently a lot of developmental effort being expended on our QQQQ Voted Signals Report (voteqqqq), we will discuss some of the more interesting aspects of this QQQQ swing-trading signal in the next few Broadcasts.

*** We have a signal change for Monday (8/6/07), and we are more comfortable with this one since we are returning to a Long signal after going Short on Friday.

Our current QQQQ Voted Signal model (see voteqqqq.htm report) has been housebroken of its tendency to Short over-aggressively. The current model stays long about 59-60% of the time, as determined by back-testing.

Obviously, our Short call for Friday was correct ... and it strengthens our confidence in our corrent daily Procedure: justifying every daily signal by trying to Force the model to take a contrary stance. This analysis step, which takes several hours, tries to find a 'logical' combination of our indicators that would not only reverse the newly generated signal -- but also INCREASE the performance of the model for the entire 5-year back-testing interval. If we can not find a 'logical' (defensible) function of indicators OR if the proposed change does not actually improve past performance, then we terminate the analysis and go with the proposed signal. If, however, we ARE able to come up with such a model modification, then the change becomes part of our standard model and we go with the new signal.

The following Table shows the actual (published) QQQQ signals for the past week (in column 2). Notice that (optional) protective Stops were generated for several days ... and that for this particular week at least, the Stops were rather more 'mandatory' than 'optional'. Column 3 shows what the signal SHOULD HAVE BEEN had we been omniscient. Finally, Column 4 shows what the current model would have generated (back-tested and optimized) were it able to do it all again. The current model has had several additions in recent days to sharpen its ability to detect Long situations (and thereby increase its accuracy in detecting Shorts).

Day/Date	Actual Signal	'Should' Have Been	Back-Tested Signal
Friday (7/27)	Long (w/ 0.60% Stop)	Short	Short
Monday (7/30)	Long (no Stop)	Long	Long
Tuesday (7/31)	Long (w/ 0.428% Stop)	Short	Short
Wednesday (8/1)	Long (w/ 0.428% Stop)	Long	Long
Thursday (8/2)	Long (w/ 0.428% Stop)	Long	Long
Friday (8/3)	Short	Short	Short
Monday (8/6)	Long	N/A	N/A

Remember that we are in an overall SELL state; our MTI value has been near zero since 7/26/07 ... forcing our mutual fund indicator to a SELL as well. This is a time when nimble trading of equities like QQQQ can be profitable (if suitably timed), and NOT a good time to be maintaining substantial portfolios. This is affirmed by the fact that our Allocation Vectors have been predominately weighted toward Cash rather than Long investments, and our various portfolios have been largely empty (except for our "H" Hedging portfolios that shorted QQQQ on Friday). In low MTI situations it is too difficult to maintain stock portfolios of any size; it is too hard to get in or out of them in a timely manner, and even more difficult to protect them with Stops.

******

Our Stock Strategist System(TM) has evolved greatly during the first 7 months of this year, and the major advances made during this productive period are discussed briefly in the following paragraphs.

I. Thermodynamic MTI (Market Timing Indicator)

Our fundamental idea continues to prove itself. After 5 years of real-time testing, there is no doubt but that our Preprocessor really is, in some sense, gauging the 'temperature' of the stock market. The MTI (Market Timing Indicator) value which is output by this program is our primary indicator, and it forms the basis for our 1000 customized stock and ETF timing models (our Grail System). As discussed at length elsewhere, the market 'temperature' is calculated by using an analogy with the Maxwell-Boltzmann distribution (from thermodynamics) that describes the relationship between the temperature of a gas and the resultant distribution of molecular speeds. In our case, instead of looking at the number of molecules that fall within certain speed ranges, we look at the number of stocks that are currently contained in a spectrum of specially-designed (synthetic) portfolios.

II. The Grail System (through 2006)

The 1000 Grail System models include 141 ETFs and 859 stocks. The stocks that are modeled are chosen based on the following criteria: (1) sufficient liquidity to permit reasonably large positions and low-spread trades, (2) a high G/d ratio (Annualized Gain divided by the average maximum drawdown), and (3)excellent (back-tested) Long trading accuracy (75% or better).

These 1000 models (each utilizing 69 tuned parameters) are primarily based on the MTI value, but also take into account the recent price and volume history of the equity, as well as some fundamental data. These models are essentially 'swing-trading' models, and their Long retention periods have always ranged between 10-13 days on average. These 1000 equities are used to construct the 50 portfolios of our PORTSTAT system. There are 5 'classes' of portfolios; three that utilize only stocks ("H", "N", and "C" classes), and two that have ETF-only portfolios ("E" and "U").

III. Initial Modifications (late 2006-early 2007)

The Grail System models provided (back-tested) trading accuracies of 75-100% (over our 5-year testing interval), and have provided excellent gains with rather moderate drawdowns. Our QQQQ model (GR-L) is the best known Grail System model, and its "long" accuracies have always tested at 90% or better. All of these 1000 models are Long models; they have two signal states, BUY and SELL, and DO NOT generate a shorting signal. The standard Grail System QQQQ model (GR-L), however, has always had one annoying characteristic: it sometimes tends to stay too long in a SELL state. This phenomenon is what we have called a 'cool rally'; during a cool rally the MTI value stays close to 1.0 (its neutral value), and does not fluctuate enough to kick the QQQQ model into a BUY state -- even when the market is slowly climbing. This problem has been most apparent since mid-2006, and despite the stellar performance of the model during volatile periods (2002-2004), its recent reluctance to go "long" has been an issue.

There were three possible ways to address the 'cool rally' problem: (1) modify the MTI value so that its value was artificially raised to reflect the 'actual' price trend of QQQQ, (2) modify the Grail System models themselves so that they paid more attention to price trends than the MTI value, and (3) develop a new kind of timing model. During the past year we have worked on all 3 approaches.

In late 2006 we modified the MTI calculation so that it took the actual market trend more fully into account ... boosting the MTI value slightly to reflect what the market was 'actually' doing. This modification was very limited, however, since any attempts to boost the MTI beyond rather modest amounts seriously degraded the back-tested performance of the model. At the present time the MTI calculation incorporates actual market trend data to the maximum possible extent.

The second step was to modify our Grail System models so that they included 'Momentum Logic'. This represented a further shift away from a purely 'thermodynamic' model (based strictly on MTI). Actually, our Grail System models were ALWAYS hybrid models; although the MTI is the dominant indicator (if the MTI is high then the models should switch to Long on the presumption that a 'rising tide lifts all boats'), our models have always included price and volume information. With the addition of our Momentum Logic, however, the price and volume data became appreciably more important -- moving us a bit farther from purely 'predictive' models (based on MTI), and a bit closer to traditional trend-following models. This change was more productive than was the MTI boost described in the preceding paragraph, although a delicate balance had to be maintained so that the models did not become excessively trend-following; after all, the objective of market timing is to be able to 'predict' what is about to happen next ... and not simply report on what has already occurred. The addition of the Momentum Logic did make our Grail System models more responsive, however, and this in turn helped fill our PORTSTAT portfolios more quickly (and more fully) when the market became stronger.

IV. The Voted QQQQ Signal (March-July 2007)

When it became clear that we had reached the limits of what was achievable by modifying the MTI, and the Grail System models themselves, we decided to develop an altogether new QQQQ timing model. This model is known as the Voted QQQQ Signal, and it now gives us for the first time a model that is accurate enough to generate profitable shorting signals.

You see, we actually have an overabundance of QQQQ signals; not only do we have the Grail System model for QQQQ (now known as GR-L), but we also have 10 'raw' QQQQ timing signals that are emitted by our Preprocessor. So, the idea arose of generating a new signal by 'voting' on these 11 available signals. We immediately took this one step further, however, by developing another Grail System model for QQQQ (GR-S) that has a higher 'shorting' accuracy than does the standard signal (GR-L). This was accomplished by changing our Penalty Function to put increased weight on shorting gains (and lower shorting drawdowns), and a bit less weight on "long" performance. Our current Grail System shorting model (GR-S) has a 100% backtested shorting accuracy over the past 5 years.

So now we had 12 signals to consider. In addition, we decided to throw everything else that we had into the mixture: 11 neural networks predicting QQQQ "n" days in advance (NN1, NN2, NN3, ..., NN10, and CNNI -- the 'combined' average value), (2) Table II of our Prediction Table (estimating QQQQ's performance for the day based on Generalized Candlesticks -- and a slew of neural networks as well), (3) our Grail System signals for SPY, DIA, IWM, IWB, and IWV, and (4) all available Preprocessor indicators (TOT, #L, #S, LSI, SSI, etc.). The basic idea was to do whatever it took to develop a single QQQQ timing signal that was as accurate as possible (for both Longs and Shorts), using every bit of information that we had available, and ENSURING that it would always look for all justifiable reasons for going "long" before resorting to a Short signal.

The current Voted Signal (see the voteqqqq report) has back-tested Long/Short trading accuracies of about 78% (for 378 trades over the past 5 years), annualized gains ranging from 215-248% (depending on the use of 'stops'), and YTD 2007 gains of 70-76% (depending on the use of 'stops'). Furthermore, this model prefers to be "long" rather than "short"; it is Long 59.1% of the time, and Short 40.8% of the time, and it has solved the two major drawbacks of our swing-trading Grail System signal (GR-L): (1) it will almost invariably detect profitable Long opportunities, and (2) it is a much more reactive, shorter-term signal (changing state every 3-4 days on average). Its back-tested performance for 2002-2007 is: 174%(8/16/02-12/31/02), 727%(2003), 188%(2004), 104%(2005), 108%(2006), and 76%(2007 YTD). These gain figures assume the use of all advertised 'stops', and are, of course, highly optimized numbers. It is interesting to note how productive a timing model can be when the market is extremely volatile (2002-2003), and how much less so when the market becomes steadier (2004-2007). 2007 has been a very difficult year for market timing algorithms because the slightest miscalculation puts us BEHIND a buy-and-hold strategy.

V. The Voted Signal and its Ripple Effects

Once the Voted QQQQ Signal was available, it was only natural to begin to utilize it throughout our System, and especially in our Grail System models, and in our PORTSTAT portfolio system.

Earlier work had increased the performance of our 1000 Grail System models by adding Momentum Logic. The marginally useful attempts to boost the MTI value (and thereby energize the Grail System models) now led to using the Voted Signal as an additional indicator within each Grail System model. In other words, if the Voted Signal is a BUY, then the model will try a bit harder to justify generating a BUY signal based on recent price/volume data; conversely, if the QQQQ Voted Signal has gone to a SHORT, then the model will become much more conservative ... and will switch to a SELL if the stock looks the slightest bit suspect. This has now made our 1000 stock/ETF models much more responsive to changes in market strength, and it helps our portfolios fill more quickly (because of more models in the BUY state).

A second use of the Voted Signal was within the PORTSTAT portfolio system itself. Within the "H" class portfolios, the Voted Signal is used as a final arbiter on whether to hedge (Short) or not; if the Voted Signal is a BUY then the "H" portfolios will refuse to short QQQQ ... opting for cash if no suitable Longs can be found.

VI. Signal Change Procedures

While there is undoubtedly more to do with our Voted QQQQ Signal, this will most likely be a gradual enhancement process from this point forward. We now have over 20 specialized algorithms that look to see if a Long signal is justified (at those times when our underlying QQQQ signals are actually 'short'), and we will undoubtedly be refining these algorithms... and adding a few more.

Our current procedure for dealing with the Voted Signal is as follows. If there is a signal change (BUY > SHORT, SHORT > BUY), then we will allot an hour or two each evening in attempting to suppress the signal change by adding more criteria to the Voted Signal model. We hate signal changes (and especially those which call for a SHORT), and so we look at all of our available indicators to see if there is ANY reasonable combination that would keep the signal unchanged. If we come up with an algorithmic modification that works -- and which improves our back-tested performance over our entire 5-year backtesting period -- then we will go with it, and the new criteria will become a permanent part of our model. If we can't come up with anything reasonable, however, or if the contemplated change would degrade our back-tested performance, then we reject the new logic and we ALLOW the signal to change state. Use of this procedure will permit our model to 'learn' new market setups (as our back-testing period steadily grows in length, and it will ensure that we never issue a signal change without an exhaustive analysis to be sure that it is as accurate as we can make it.

VII. Intermediate-Term (I/T) Signals and Portfolios

When the pace of development on our Voted QQQQ Signal, Grail System, and PORTSTAT portfolios tapers off, we plan to refocus our attention on our intermediate-term (I/T) signals (SECTION2 report) and portfolios (VARPORTA-G). These are rather 'traditional' signals and portfolios (although we believe they are several notches better than most that are available), and there is a lot more that can be done to improve them, e.g., creating more diversification in the portfolios, utilizing our Voted QQQQ Signal to fine-tune our allocation levels, etc. Unlike our Grail System (swing-trading) portfolios, our I/T portfolios incorporate a great deal of fundamental data (P/E ratios, earnings growth, dividend yields, market capitalization, Sales per Share, etc.). In recent months these portfolios have acquitted themselves well ... preserving their equity during dips by reducing their allocation levels.

VIII. Mutual Fund Signals

In the due course of time our Mutual Fund System will come back under our 'knife'. We have a growing list of ideas for improving the mutual fund reports (including adding more funds and fund families), and in the meantime we are encouraging our subscribers to contact us with their specific preferences.

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Schulenberg.com

Weekly Status (as of 3/23/07)

-----------------------------

Modeling (Stock Signal Status). Our QQQQ swing-trading signal remains in the SELL/SHORT state that was first signaled on 2/8/07. Since our cost basis was $44.56 (the Open price on 2/9/07), we currently have an ‘official’ shorting gain of 1.00%, giving us a net QQQQ trading gain of 4.24% for YTD 2007. If our occasional ‘protective stops’ had been utilized (which were very important on the day of the Fed announcement), then the current shorting gain would be 4.81%, yielding a YTD gain of 8.36%. Our current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 1491% since 8/16/02), and in today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm)

We characterize the market’s current status as “weakly negative”. For Monday (3/26/07) our Market Color Code is in the BLUE (weak “long”) zone, our Max Long% Allocation is at 60% (with our ‘prudent’ allocation at 40%), and only 3.00% of our 1000 stock/ETF models are in the ‘buy’ state. Also, among these short-term stock/ETF timing models we have 10 models (including 8 ETFs) with perfect (100%) back-tested Long accuracies over the past 5 years – and ALL of these models are in the SELL state. During this past week our MTI (Market Timing/Temperature Indicator) started at 33.6 deg F. (on Monday) and then moved steadily upward to a high value of 110 deg F. before falling back to 99.4 deg F on Friday (98.6 deg F, ‘body temperature’, represents an essentially ‘flat’ market). Since our QQQQ signal is a swing trading signal, it is not always reactive to very short-term events; it was ‘short’ for the big plunge of 2/27, but also stayed ‘short’ (with ‘stop’ alerts) when the market steamed back during the ‘Fed week’. Nonetheless, by following the ‘protective stop’ alerts given in our QQQQ Voted Signals Report, the shorting gain for the past 6 weeks would still be an impressive 4.81%.

Modeling (Mutual Fund Signal Status). In 2006 our Mutual Fund Timing Indicator (MFTI) was a ‘buy’ until 5/11/06, at which time it went to a ‘sell’, and as of 9/8/06 it returned to a ‘buy’ state. Because of the new QQQQ ‘sell’ on 2/8/07, however, this indicator went first to a ‘hold’, and then as of 2/27/07 – to a ‘sell’. During this past week our average ‘score’ for a set of 1100 mutual funds has increased significantly from 4.11 to its current value of 4.98 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is exactly at a ‘hold’ state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT

Although our primary stock signal (QQQQ) may change state once or twice per month on average, during the past 4.5 years our mutual fund timing signal has changed state (buy/hold>sell, sell>buy/hold) only 16 times.

Point #1: Our MTI (Market Timing Indicator) value is calculated based on the portfolio data emitted by our Preprocessor (10 tradeable Primary portfolios, and 100 highly artificial Secondary portfolios). It is tuned to QQQQ, however, rather than to the Dow (DIA), the S&P (SPY), the Russell 2000 (IWM), or the market as a whole (VTI). The point is that the MTI has to be tuned to ‘something’, and QQQQ was chosen for several reasons: (1) it is a highly liquid ETF with excellent modeling characteristics (high annualized gain, low drawdown, good shorting accuracy and excellent Long accuracy), (2) it correlates well with overall market action (if the ‘techs’ are strong then the overall market is likely to be as well), and (3) it is perhaps the most popular ETF for trading all by itself. The fact that our MTI value is tuned to QQQQ, however, was clearly evident during the past two days (3/22-23/07) when QQQQ declined at the same time that the rest of the market (DIA/SPY) gained. This is what one would expect if the MTI value was tracking the ‘tech stocks’, rather than the broader market.

Point #2: Our QQQQ Voted Signal (voteqqqq Report) is a swing trading signal just like our 1,000 Grail System models. All of these signals are based on our ‘thermodynamically’-calculated MTI (Market Timing Indicator) value, but they also reflect the ‘intrinsic’ natural frequency of the stock or ETF. We use a complex cpu-intensive Monte-Carlo procedure to determine the inherent drawdown and Long retention period of each of our 1000 modeled equities, and our model optimization process uses a Penalty Function to combine the MTI, this empirically determined ‘intrinsic’ data, plus the ever-lengthening price and volume history of the equity, in order to generate the final model. Also, these models are continuously optimized so that minor ‘ripple effects’ in the MTI calculation, as well as an additional day of price and volume data, are incorporated into the model. The Voted QQQQ signal is a blend of two separate Grail System models; the GR-L model is our standard Long-optimized model, while the GR-S sacrifices Long accuracy in favor of producing a more accurate shorting signal. In our QQQQ Voted Signals report these two models are combined, resulting in an aggregate model with 3 signal states: BUY, SELL/$ (CASH), and SELL/S (SHORT).

Point #3: Since our Voted QQQQ signal is inherently a swing trading signal, and with a mean Long period of about 13 trading days, the most effective way to improve its short-term accuracy is via the use of ‘protective stops’. The stop alerts generated by the Voted Signal Report are even more useful because they are not always triggered; they do, however, protect the current position from significant damage (e.g., being ‘short’ on a Fed Day). If instead of ‘stops’ we forced the overall signal to change state (e.g., SELL/SHORT to SELL/$, or SELL/SHORT to BUY), then we would effectively be setting MANDATORY stops and greatly increasing trading frequencies in the bargain. Instead, we prefer to maintain our swing trading (thermodynamically-based) signal, and to improve its performance (reducing drawdown) by issuing stop alerts when we ‘anticipate’ that the market is likely to turn against our current position.

Point #4: We have two good sources of data that can help us set ‘protective stops’ in an optimal fashion: (1) various parameters output from the Preprocessor, and (2) QQQQ day-trading signal data derived from the Generalized Candlesticks approach that underlies Table II of our Prediction Table (PREDTABL). If we issue protective stop alerts based on the Preprocessor data then we denote them as Type I Stops (either Long or Short); similarly, stop alerts issued on the basis of the PREDTABL signals (the RaDiSH Transform) are denoted as Type II Stops (Long or Short). The key parameter used as a guide for issuing a Type I stop alert is the #L value calculated in the Preprocessor. #L is the number of NEW Longs that was detected by the Preprocessor for the next trading day. If we are ‘short’ and #L is a large value, then this will cause a Type I Stop alert to be issued; or, if we are ‘long’ and #L is a very small value, then this will cause a Long Type I Stop alert to be issued. The Type II stop alerts (which tend to be more accurate than the Type I’s) are generated if we see very large (for a Short stop) or very negative (for a Long stop) values for the projected QQQQ gain for the next day. The back-tested results of QQQQ trading (year-by-year since 2002) are shown in summary form at the end of the Voted QQQQ Signals Report (voteqqqq).

Point #5: We are currently retuning all of our Preprocessor portfolios, and of the 10 Primary portfolios so far the retuning of L/1 has been accomplished. Each Portfolio requires about a solid week of cpu time on a fast PC, and we are now retuning two portfolios per week using dedicated machines. We anticipate two major results from this ongoing tuning activity: (1) the MTI calculation will be ‘sharpened’ slightly (made a bit more reactive), and (2) the trading performance of the 10 Primary Preprocessor Portfolios (L/1-L/10) will be substantially improved … making them even more suitable as candidates for serious trading. In addition to the Preprocessor retuning activity, we are looking at other ways to improve our ‘shorting’ performance: (1) using the Fed calendar to ensure that stop alerts are issued more liberally as Fed announcement days approach, and (2) using put/call ratios to try to maximize shorting gains by suggesting a ‘profit taking’ exit when the productivity of the ‘short’ appears to be diminishing.

Point #6: The ‘protective stop’ alerts are optional, but it is clearly advantageous to heed them whenever possible. Since this would involve day trading, however, it is not feasible for all traders to do this – nor is it possible in certain types of accounts. When it is feasible, however, our Protocol calls for using the ‘trigger’ price to force an exit from the position, and then staying in cash until near the Close, at which time the original position (Long or Short) would be re-established. The use of such ‘stops’ is even more important for ‘shorts’ than it is for Longs, and this is because shorting signals are inherently less accurate than Long signals to begin with, and the maximum drawdowns are also considerably greater. The ‘trigger point’ is set at the OPEN price of QQQQ plus or minus a percentage value that is specified in the Voted Signals report. For ‘short’ protective stops the trigger point may typically be 0.4-0.5% ABOVE the Open price. In general, it is better to ‘monitor’ QQQQ as the trigger point is approached in order to ensure that it is truly exceeded in a convincing fashion (rather than just lightly ‘tapped’) before exiting from the position. Similarly, since the trade should be re-established prior to the Close it is usually possible to get a more favorable price by observing QQQQ as the Close approaches.

Point #7: If protective stops ‘work’, of course, then it is natural to consider taking a contrary position between the trigger point and the Close … in order to put the Cash to work in the opposite direction. We call this a ‘reverse play’, and it goes beyond our Protocol because it adds quite a bit more risk to the QQQQ trading process. Nonetheless, this is often a very profitable thing to do … especially if the current position is a ‘short’. As can be seen in the Voted Signals Report, the accuracies of the Type II Short Stops, and both the Type I and II Long Stops, appears to be high enough (from a back-testing perspective) to make a ‘reverse play’ attractive. The Type I Short Stops are not as accurate, however, and it probably is best to restrict making ‘reverse plays’ on these Stops UNLESS it is a Fed announcement day.

In future segments of this weekly Broadcast we will discuss more concepts and procedures behind our short-term Grail System models and portfolios:

(1) More on Model Optimization: dealing with a myriad of empirical constants

(2) The ‘Voted’ QQQQ Signal: the most accurate QQQQ timing model in existence

(3) A ‘tie-in’ between Grail System models and Fibonacci analysis (and the foundations of Elliott Wave Theory)

(4) The Perfect Trader™ program: building ‘natural’ models that reflect the inherent rhythm of a stock

(5) The Prospector™ program: selecting the best 1,000 stocks/ETFs for short-term modeling

(6) Optimal portfolio design using Grail System signals: the need for different ‘classes’ of portfolios

(7) The “Stagecoach” Principle: rotating the tired ‘horses’ out of a portfolio

(8) Hedging, and other asset protection techniques

We will also be addressing a variety of other topics: Differential Signals (QQQQ vs. IWM, SPY, DIA) for futures trading, Estimating the relative strength of QQQQ, SPY, and DIA by a ‘component’ signal analysis, Intermediate-term models and portfolios (the STAR System™), the Mutual Fund Timing Indicator (MFTI), the ETF ‘Decomposition’ of Mutual Funds (determining the current ‘holdings’), the Prediction Table (knowing when to hold ‘em, and when to fold ‘em), the unification of MOS/MCS analysis and the QQQQ day-trading signal, proprietary indicators (‘Nibble’ indicator, waxless candlesticks, etc.)…

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Schulenberg.com

Weekly Status (as of 3/16/07)

-----------------------------

Modeling (Stock Signal Status). Our QQQQ swing-trading signal had gone to a ‘sell/short’ based on the market Close data of 2/8/07, and it remains firmly in this state. Since our cost basis was $44.56 (the Open price on 2/9/07), we currently have a shorting gain of 3.88%, giving us a net gain of 7.12% for YTD 2007 (our peak drawdown during the ‘short’ was –2.04%). Our current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 1871% since 8/16/02), and in today’s expanded broadcast we discuss some additional aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm)

We currently characterize the market’s status as “moderately negative”. For Monday (3/19/07) our Market Color Code remains in the RED (shorting) zone, our Max Long% Allocation is at 4% (with our ‘prudent’ allocation at 0%), and only 0.90% of our 1000 stock/ETF models are in the ‘buy’ state. Also, among these short-term stock/ETF timing models we have 22 models (including 19 ETFs) with perfect (100%) back-tested Long accuracies over the past 5 years – and ALL of these models are in the SELL state. During this past week our MTI (Market Timing/Temperature Indicator) started at 23.6 deg F. (on Monday) and for the past four days has languished in the very low range of 5.2-5.7 deg F. (0.0524-0.0571) (98.6 deg F, ‘body temperature’, represents an essentially ‘flat’ market). There are as yet NO signs that this dip is over, and it may yet turn out to be a full-fledged correction; we always take it one day at a time, however.

Modeling (Mutual Fund Signal Status). In 2006 our Mutual Fund Timing Indicator (MFTI) was a ‘buy’ until 5/11/06, at which time it went to a ‘sell’, and as of 9/8/06 it returned to a ‘buy’ state. Because of the new QQQQ ‘sell’ on 2/8/07, however, this indicator went first to a ‘hold’, and then as of 2/27/07 – to a ‘sell’. During this past week our average ‘score’ for a set of 1100 mutual funds has decreased slightly from 4.23 to its current value of 4.11 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is well below a ‘hold’ state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT

The following Table (extracted from the end of the Voted Signals Report) shows the back-tested gains, calendar year by calendar year, since our signal’s inception date of 8/16/02. The figures for 2002 thus represent only a 3.5 month period. The VOTE(1) column shows the back-tested gain for a particular year, assuming that QQQQ was held Long during a BUY signal, shorted during a SELL/S signal, and otherwise equity was kept in cash during a SELL/$ signal. This is the signal that TimerTrac is now monitoring, although we didn’t start with TimerTrac until 9/28/06. Despite the fact that we only have TimerTrac verification for the last 3 months of 2006, the annual figure of 29.45% fairly closely reflects our actual QQQQ gains during the year. The match isn’t exact, however, since our new ‘voted’ QQQQ signal has only been available since January 2007; before that we were using our standard Grail System signal (GR-L) – with its substantially lower shorting accuracy.

QQQQ Performance by Calendar Year
------------------------------------
VOTE(1): Performance without "stops"
VOTE(2): Performance with "stops"
------------------------------------------
YEAR   VOTE(1)   VOTE(2)  BUY&HOLD  #Intvls
------------------------------------------
2002   135.70%   153.96%    -0.20%     13    (From 8/16/02 - 12/31/02)
2003   134.80%   194.98%    48.30%     21
2004    69.21%    72.63%     9.30%     16
2005    52.60%    56.49%     1.45%     17
2006    29.45%    48.29%     6.91%      8
2007     6.56%     9.29%    -1.45%      2    (YTD: Year-to-Date)

The VOTE(2) column will be discussed in depth next week; it reflects the back-tested gains that ‘would have been made’ if protective stops had been used to minimize drawdowns. Since this requires the ‘day trading’ of QQQQ (at a frequency of about one day out of five), this procedure is not feasible for all investors; nonetheless, the improvements in annualized gains are impressive if these 4 stops (Type I Short Stop, Type II Short Stop, Type I Long Stop, Type II Long Stop) are utilized.

The BUY&HOLD column, of course, shows exactly what its name implies: the gain in QQQQ if it had simply been held Long all year.

A salient fact about this Table is that the gains in 2002-2003 were enormous, while the gains in 2004-7 appear to be steadily tapering off. Why is this? Is our model slowly degrading? No, it has to do with ‘volatility’. Notice that the last column (#Intvls) counts the number of distinct BUY and SHORT (SELL/S) periods that occurred during the year (according to our QQQQ voted signal model). 2002 had 13 such intervals – in only a 3.5 month period. We don’t know how this would have extrapolated for the entire year (if our signal had been available before 8/16/02), but 2002 was obviously a year characterized by extremely frequent signal changes. 2003 was also a busy year for signal changes with 21 distinct Long and Short periods. Remember that we can only make money when we are Long or Short (our occasional ‘cash’ periods (SELL/$) merely hold equity constant). In 2004 and beyond we have seen a trend toward longer Buy/Short intervals, and consequently have had far fewer changes of signal direction.

The number of Long/Short intervals (#Intvls) determines the maximum potential gain from a market timing strategy. For example, if the market goes up like a ruler during the course of a year, making 15% in the process, then the maximum gain that a market timing signal could possibly make would also be 15%. If the market is absolutely smooth (flat, increasing, or decreasing), then the best that one can do is to stay Long while it is moving upward, and Short while it is moving downward. Timing methods only provide an advantage if the market is changing, going from Long to Short and Short to Long, and it is when there are a lot of changes that big profits can be made (staying Long during the uptrends and shorting the dips).

There is no way to tell whether 2007 will be a year of many signal changes, or few; so far we are in our second interval. Although the first quarter is not yet over, our projected QQQQ gain for the entire year currently looks like it may fall in the 28-32% range (like 2006). If one has a highly accurate signal, of course, then additional gains can be make with relatively low risk by using ‘leveraged’ QQQQ instruments like QID (double Inverse QQQQ) and QLD (double Long QQQQ); these two ProShares ETFs will approximately double the gains (or losses) when trading QQQQ.

In future segments of this weekly Broadcast we will discuss more concepts and procedures behind our short-term Grail System models and portfolios:

(9) More on Model Optimization: dealing with a myriad of empirical constants

(10) The ‘Voted’ QQQQ Signal: the most accurate QQQQ timing model in existence

(11) A ‘tie-in’ between Grail System models and Fibonacci analysis (and the foundations of Elliott Wave Theory)

(12) The Perfect Trader™ program: building ‘natural’ models that reflect the inherent rhythm of a stock

(13) The Prospector™ program: selecting the best 1,000 stocks/ETFs for short-term modeling

(14) Optimal portfolio design using Grail System signals: the need for different ‘classes’ of portfolios

(15) The “Stagecoach” Principle: rotating the tired ‘horses’ out of a portfolio

(16) Hedging, and other asset protection techniques

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Schulenberg.com

Weekly Status (as of 3/9/07)

-----------------------------

Modeling (Stock Signal Status). Our QQQQ swing-trading signal had gone to a ‘sell/short’ based on the market Close data of 2/8/07, and it remains firmly in this state. Since our cost basis was $44.56 (the Open price on 2/9/07), we currently have a shorting gain of 3.66%, giving us a net gain of 6.90% for YTD 2007 (our peak drawdown during the ‘short’ was –2.04%). Our current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 1847% since 8/16/02), and in today’s expanded broadcast we discuss additional trading aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm)

We currently characterize the market’s status as “weakly negative”. For Monday (3/12/07) our Market Color Code remains in the RED (shorting) zone, our Max Long% Allocation is at 46% (with our ‘prudent’ allocation at 0%), and only 1.0% of our 1000 stock/ETF models are in the ‘buy’ state. Also, among these short-term stock/ETF timing models we have 22 models (including 18 ETFs) with perfect (100%) back-tested Long accuracies over the past 5 years – and ALL of these models are in the SELL state. During this past week our MTI (Market Timing/Temperature Indicator) started at 0 deg F. (on Monday) and then moved as high as 51 deg. F (0.5073), before ending up on Friday at 26 deg F. (0.2586) (bear in mind that ‘body temperature’ represents an essentially ‘flat’ market). There are as yet NO signs that this dip is over, and it may yet turn out to be a full-fledged correction; we always take it one day at a time, however.

Modeling (Mutual Fund Signal Status). In 2006 our Mutual Fund Timing Indicator (MFTI) was a ‘buy’ until 5/11/06, at which time it went to a ‘sell’, and as of 9/8/06 it returned to a ‘buy’ state. Because of the new QQQQ ‘sell’ on 2/8/07, however, this indicator went first to a ‘hold’, and then as of 2/27/07 – to a ‘sell’. During this past week our average ‘score’ for a set of 1100 mutual funds has increased slightly from 4.12 to its current value of 4.23 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is well below a ‘hold’ state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT

Modeling (Findings/Insights). Our ‘voted’ QQQQ signal report can be seen at http://www.schulenberg.com/download/voteqqqq.htm. In order to produce this accurate signal we utilize two different Grail System models for QQQQ: our standard (long-optimized) model (GR-L), and our new (short-optimized) model (GR-S). Our GR-L model has a back-tested accuracy of 94% for QQQQ Longs, but only an 81% accuracy for QQQQ Shorts. The GR-S model, however, has a lower back-tested accuracy of 90% for QQQQ Longs, but an excellent 94% back-tested accuracy for QQQQ Shorts. The Voted QQQQ Signals Report combines these two models by calling for a ‘BUY’ when both models agree on a Buy, and a ‘SHORT’ (SELL/S) when both models agree on a SELL. When the two signals disagree (about 5% of the time), we generate a CASH signal (SELL/$).

Although this combined model (GR-L/GR-S) shown in the Voted QQQQ Signals Report has an extraordinarily accurate back-tested history, this doesn’t mean that there is no room for improvement. There are always drawdowns associated with any timing signal (even the most accurate ones), and this is a big source of concern to traders who are holding large QQQQ positions – and especially if the current holding is a SHORT while the market is going up. In order to address this issue we have developed two different types of ‘protective stops’; although these stops are truly ‘optional’ (and since they require day trading they are not feasible for many types of trading accounts), their use can substantially increase bottom-line gains while at the same time minimizing drawdown.

We do not generate ‘stop’ alerts every day; any attempt to do this invariably reduces profits. Instead, we utilize two of our accurate predictors of incipient market strength or weakness: (1) the #L (# of new Longs) parameter generated by our Preprocessor, and (2) the projected QQQQ percentage gain for the day (from Table II of our Prediction Table). If our #L is very large, and we are currently in a QQQQ Short (SELL/S), then we generate a Type 1 protective stop alert … and a Type 2 stop alert is generated if the day-trading signal for QQQQ is a “long”. Similarly, if we are in a QQQQ Long period (BUY), then we generate a Type 1 alert if the #L parameter is unusually small, and we issue a Type 2 alert if the QQQQ day-trading signal is a “short”. With all of these ‘alerts’, a trigger point is set at 0.50% above or below the Open price of QQQQ (for ‘short’ and ‘long’ stop alerts, respectively). If the stop is triggered, then our standard protocol calls for re-establishing the Short or Long position near the market Close – either before or after (in the after-hours extended session), depending on when the entry looks best. Our tracking of the gains resulting from these protective stops assumes that the position is re-established exactly at the Close price.

None of the statistics in the Voted QQQQ Signals report reflect any gains or losses due to these ‘protective stops’, although when a stop ‘would have been’ triggered the Report does show the percentage gain (or loss) produced by that particular ‘stop’. It is then easy for the reader to add these up over a certain period of time in order to gauge their overall effectiveness. We plan to begin maintaining two sets of ‘books’ within the Report: the first one would show performance WITHOUT the impact of protective stops, and the second one would show the true performance (including drawdowns) WITH the usage of all triggered protective stops.

So, these ‘stops’ are optional, but their use can significantly increase the bottom-line profit, reduce drawdown, and greatly reduce the anxiety level associated with seeming to be on the wrong side of the ‘fence’ (e.g., “short” when the market is showing strength). These stops ‘have to be’ optional because many types of accounts will not support day trading (e.g., retirement accounts or other cash accounts). For accounts that WILL support day trading, however, there is one further step that might be considered, and that is to take an opposing position at the time that the ‘stop’ is triggered UNTIL the market Close – at which time the original position should be re-established. We cannot ‘recommend’ this procedure, but obviously if going to cash SAVES money (when the protective ‘stop’ is triggered), then taking a contrary position should INCREASE profits. For example, if our Voted QQQQ signal is a SELL/S (Short signal), and we have a protective stop alert for the day (with the trigger point set 0.50% ABOVE the Open price of QQQQ), then if the stop is triggered we have lost 0.50% (plus any loss that might have been incurred if the market ‘gapped up’ from the preceding Close). At that point a Long position might be considered, and such a Long position would be maintained up until near the Close, at which time the intra-day Long position would be closed out and QQQQ would be re-shorted.

In addition to trading QQQQ there are substantial benefits to be reaped by utilizing the new ProShares QLD and QID ETFs. QLD and QID both provide a leveraging factor of 2 (as though 100% margin were being utilized), and while QLD is thus a double-Long QQQQ equivalent, QID behaves as though QQQQ were shorted with 100% margin. These two ETFs allow leveraging (and shorting) within non-margin-enabled accounts (e.g., retirement accounts), and clearly this raises the level of risk as well as the potential for gain. We NEVER advise the use of leveraging techniques within retirement accounts, however. This is a strictly personal decision, and it definitely increases the volatility of accounts that SHOULD BE invested in a conservative fashion.

In future segments of this weekly Broadcast we will discuss more concepts and procedures behind our short-term Grail System models and portfolios:

(17) More on Model Optimization: dealing with a myriad of empirical constants

(18) The ‘Voted’ QQQQ Signal: the most accurate QQQQ timing model in existence

(19) A ‘tie-in’ between Grail System models and Fibonacci analysis (and the foundations of Elliott Wave Theory)

(20) The Perfect Trader™ program: building ‘natural’ models that reflect the inherent rhythm of a stock

(21) The Prospector™ program: selecting the best 1,000 stocks/ETFs for short-term modeling

(22) Optimal portfolio design using Grail System signals: the need for different ‘classes’ of portfolios

(23) The “Stagecoach” Principle: rotating the tired ‘horses’ out of a portfolio

(24) Hedging, and other asset protection techniques

Craig W. Schulenberg

http://www.schulenberg.com/ Schulenberg & Associates, Inc.

http://www.schulenberg.com/download/broadcast.htm Previous (Expanded) Broadcasts

Schulenberg.com

Weekly Status (as of 3/2/07)

-----------------------------

Modeling (Stock Signal Status). Last week we had said: “Based on model parameters there is a 94% probability that QQQQ will dip by at least 1.6% over the next few days, and likely starting on Monday… Despite the optimism implied by the Market Color Code (GREEN) and the Max Long% Allocation, the remainder of our indicators (and especially our neural networks) suggest that a correction may be imminent”.

Our QQQQ swing-trading signal had gone to a ‘sell/short’ based on the market Close data of 2/8/07, and it remains firmly in this state. Since our cost basis was $44.56 (the Open price on 2/9/07), we currently have a shorting gain of 4.67%, giving us a net gain of 7.91% for YTD 2007 (our peak drawdown during the ‘short’ was –2.04%). Our current and historical (back-tested) values for our ‘voted’ QQQQ signal are contained in the voteqqqq report (showing a compounded gain of 1968% since 8/16/02), and in today’s expanded broadcast we discuss important trading aspects of this QQQQ signal (see http://www.schulenberg.com/download/broadcast.htm)

The ‘tenacity in the techs’ that we have discussed for some time has indeed finally dissipated. For Monday (3/5/07) our Market Color Code remains in the RED (shorting) zone, our Max Long% Allocation is at 0%, and only 0.6% of our 1000 stock/ETF models are in the ‘buy’ state. Also, among these short-term stock/ETF timing models we have 19 models (including 16 ETFs) with perfect 100%) back-tested Long accuracies over the past 5 years – and ALL of these models are in the SELL state. During this past week our MTI (Market Timing/Temperature Indicator) started at 101 deg F (on Monday), and then dropped to 0 deg F (0.0) for the remainder of the week(bear in mind that ‘body temperature’ represents an essentially ‘flat’ market). There are as yet NO signs that this dip is over, and it may yet turn out to be a full-fledged correction.

Modeling (Mutual Fund Signal Status). In 2006 our Mutual Fund Timing Indicator (MFTI) was a ‘buy’ until 5/11/06, at which time it went to a ‘sell’, and as of 9/8/06 it returned to a ‘buy’ state. Because of the new QQQQ ‘sell’ on 2/8/07, however, this indicator went to a ‘hold’, and as of 2/27/07 – to a ‘sell’. During this past week our average ‘score’ for a set of 1100 mutual funds has plunged from 4.66 to its current value of 4.12 (a 0 is a hard ‘sell’, a 5 is a ‘hold’, and a 10 is a strong ‘buy’); in other words, the average mutual fund is well below a ‘hold’ state at the present time. A recent mutual fund report can be seen at: http://www.schulenberg.com/download/MUTFUNDS.TXT

Modeling (Findings/Insights). The following commentary was contained in an email sent out at midday on 2/27/07:

Although the day is not yet over, the following will probably prove true:
(1) The market would have fallen even without the China selloff -- although there is, of course, no way to prove this. Our Forecast for Monday said: 'a dip is expected', and last evening's Forecast for Tuesday said: 'Dip will likely deepen on Tuesday'. Last evening our neural networks had gone sharply negative, and that was 'before' China.
(2) Our current QQQQ Short (initiated at $44.56 at the Open of 2/9/07) will likely once again be 'in the black' ... putting us ahead of a QQQQ buy-and-hold strategy for YTD 2007.
(3) Our advice that 'prudent' mutual fund investors should be in money market funds looks quite perceptive, and those warnings have been coming for several days now.

We have recently added some more interesting statistics to our Voted Signals Report, and the new Report can be seen at: http://www.schulenberg.com/download/voteqqqq.htm

The QQQQ performance statistics, extracted from the end of this Report, are as follows:

Long/Short Acc%:   91.30%  #Trans.:   92
Total Gain:      1868.73%  (since 8/16/02)
Annualized Gain:   92.90%

Avg. Long  Gain per Trade:     5.22%
Maximum Long  Drawdown:       -5.28%
Max. (mid-trade) Long  Loss:  -5.01%
Max. (end-trade) Long  Loss:  -0.92%

Avg. Short Gain per Trade:     3.55%
Maximum Short Drawdown:       -7.39%
Max. (mid-trade) Short Loss:  -5.18%
Max. (end-trade) Short Loss:  -2.10%

QQQQ PERFORMANCE BY CALENDAR YEAR
---------------------------------
YEAR    VOTE     BUY&HOLD
---------------------------------
2002   135.70%    -0.20%    (From 8/16/02 - 12/31/02)
2003   134.80%    48.30%
2004    68.03%     9.30%
2005    52.33%     1.45%
2006    29.45%     6.91%
2007     7.36%    -2.25%    (YTD: Year-to-Date)

Now, our new 'voted' signal isn't all that much more accurate than our standard (Long-optimized) QQQQ signal, but every little bit helps when dealing with the market. The 'voted' signal, however, appears to be substantially more accurate when it comes to QQQQ 'shorts'.

The following observations can be made about the data shown above:
(1) The figures for 2006 correspond very closely to our posted (actual) results, including the performance reported by TimerTrac since September 2006, and the figures for 2007 are ‘actual’ data.
(2) Notice that the phenomenal gains in QQQQ trading came in 2002 (although we only track performance from 8/16/02 to year-end) and 2003. 2004 and 2005 were also stellar years. 2006 didn't offer nearly as much profit, but 29.45% is certainly much better than the 6.91% from a buy-and-hold strategy. In 2007 Tuesday’s (2/27) Close put our 'voted' gains well above a QQQQ buy-and-hold strategy.
(3) Note that the average gain for Long trades (in QQQQ) is only 5.22%, and that the Max Loss% is -5.01%. This means that at several times in the past 4.5 years it was necessary to 'temporarily' lose 5% of real equity in order to make a gain of 5.22%. The Max Drawdown figure is slightly higher, but since drawdown measures the dip from Peak to Trough, it was generally the case that some initial gains had been made in a trade before the drawdown began. The bottom-line is: with QQQQ Longs you have to be prepared to lose (at least temporarily) as much money as you hope to finally gain. Our worst bottom-line ‘loss’ was –0.92%.
(4) The Short situation is even more hazardous; with average Short gains of only 3.55%, the Maximum (hopefully temporary) Loss is even greater (-5.18%), and the Max Drawdown is more than double. Still, if one has the fortitude to 'stick it out', one can make money on QQQQ short trades. For ‘shorts’ our worst bottom-line ‘loss’ was –2.25%.

All of the foregoing (and the voteqqqq Report) are based on back-tested results (although the 2006-7 performance is essentially 'actual' data), and therefore there is no guarantee that such performance can be expected in the future; after all, 4.5 years of back-testing is not really all that long of a period. Nonetheless, there are clear lessons to be learned from the above data: (1) you have to learn to deal with Drawdown if you are going to make money, (2) you have to be prepared to lose (at least temporarily) at least as much as you hope to gain, (3) it isn't possible to make money every single day, (4) some years are just better than others even with almost perfect trade signals, and (5) Buy-and-Hold sucks.

In future segments of this weekly Broadcast we will discuss more concepts and procedures behind our short-term Grail System models and portfolios:

(25) More on Model Optimization: dealing with a myriad of empirical constants

(26) A ‘tie-in’ between Grail System models and Fibonacci analysis (and the foundations of Elliott Wave Theory)

(27) The Perfect Trader™ program: building ‘natural’ models that reflect the inherent rhythm of a stock

(28) The Prospector™ program: selecting the best 1,000 stocks/ETFs for short-term modeling

(29) Optimal portfolio design using Grail System signals: the need for different ‘classes’ of portfolios

(30) The “Stagecoach” Principle: rotating the tired ‘horses’ out of a portfolio

(31) Hedging, and other asset protection techniques

Modeli