S T R A T E G I S T   F O R E C A S T

=====================================

=========================================

History of Software/Report Changes

=========================================

This document summarizes all major changes made to the Schulenberg Strategist System(TM) and its various reports since November 10, 2004.

*** New (4/4/08): We have made a small but important change to our QQQQ Voted Signal model; instead of a back-tested accuracy that tends to drift between 76.5% and 79.5% (and was recently just under 77%), we have redesigned the “penalty function” that we utilize during the optimization process to give more weight to the raw accuracy ... and not just the annualized gains and drawdowns. As a result, our back-tested accuracy for QQQQ trades has now jumped to just under 83% (82.8%), while the drawdowns and annualized gain figures remain virtually the same. This 82.8% Long/Short trading accuracy figure is based on almost 6 years of back-testing, and it is quite likely that performance similar to this will be obtained in future trades. The next step is to feed this improved signal into our 1000 Grail System models, hopefully boosting their bottom-line accuracy from the current 77% range to 80% or better. Finally, the PORTSTAT portfolios will be re-optimized to deal with these more accurate signals. This entire retuning process will take several more days.*** New (1/2/08): The “C”, “E”, and “U” portfolios (PORTSTAT report) have been made a bit more conservative. The “U” (ultra-conservative ETF portfolios) have done pretty well of late, as have the “C” portfolios, but the “E” (ETF-only) portfolios have been overly aggressive. Now all three of these portfolio classes are somewhat more selective as to which equities they buy during weak market conditions. We are currently looking at the allocation percentages for the intermediate-term portfolios (VARPORTA-G), and we will be making some adjustments to them in the near future.

*** New (8/29/07): The "H" PORTSTAT portfolios have now been changed so that they DO NOT short QQQQ. The "H" should now therefore be interpreted as "H"igh Performance rather than as "H"edging. Now that the Voted QQQQ Signal report (voteqqqq) is available, users are free to short QQQQ within ANY of our 50 portfolios if the QQQQ signal is a SELL/S(SHORT) ... and there is cash available.

*** New (8/29/07): The “N”, “C”, “E”, and “U” PORTSTAT portfolios have had their volatility levels dialed back somewhat. Their drawdown over the past 4 months has exceeded our target levels, and we have traded off some potential annualized gain in exchange for more moderate drawdown levels.

*** New (8/15/07): We have made our last planned enhancement to the Voted Signal Report, and no more changes are contemplated – other than minor refinements to the protective stop logic (and strategies for executing the various Stops). This new enhancement greatly simplifies all of the special case logic which has been added over the past 6-8 weeks, and now gives us a 'cleaner' (and more responsive) model. As a byproduct of this change, the Voted Signal model now recognizes that we have had a shorting situation for the past 6 trading days ... rather than a Long/Short/Long situation. From this point forwards, we intend to let the model 'do its thing', and only consider making additional model changes if a signal proves to be grossly off the mark. Back-testing of this model suggests extremely high accuracy, and as is usually the case with market timing models, the difficult part is following the signal ... and trying not to 'second-guess' it.

*** New (8/8/07): We have just completed a short (but intensive) analysis of our QQQQ Voted Signal model (see the voteqqqq.htm report) and made some changes to enable it to better cope with the volatility that usually accompanies 'Fed' announcements. This was a delicate operation, however, because the indications that the market 'might' rise on Tuesday/Wednesday were rather subtle; in fact, we had to make use of two Preprocessor indicators that have up until now been very little used (LSI and SSI). The revised model is now considerably more reluctant to short the market at such times, and we believe that this model has nearly reached the limits of achievable accuracy for predicting QQQQ on a short-term basis. It will now stay “long” about 60% of the time, and we expect its signal accuracy to approach 77-79% for future trades.

*** New (8/1/07): The slow and steady improvement of our QQQQ Voted Signal (voteqqq.htm report) continues; for 377 back-tested trades over a 5-year period, our average accuracy (Long and Short) is 77-78%. The back-tested annualized gains are 215-249%, with YTD 2007 gains of 67-74%. Overall, the model is Long 59.7% of the time, and Short 40.2% of the time. Obviously, it is very unlikely that such phenomenal performance will be realized in 'real' trading in the future ... but what we have achieved at this point is a finely crafted machine that allows new modeling components to be added in a controlled manner, permitting the model to 'learn' about additional market configurations. We have a lot of proprietary indicators (over a dozen or so) that we can utilize, and whenever the market goes 'against' our signal, we perform a 'post-mortem' analysis to see how the model could best be extended to account for the unanticipated behavior – so we are better prepared to deal with it the next time that it crops up.

*** New (7/30/07): We have begun restructuring our daily Forecast (this email) to accomplish the following objectives: to reduce the amount of redundancy (the same thing being said in several different sections); to reduce the overall size of the email document; and to present a clearer picture of current market conditions in a more compact manner. Section B and C have been totally redone (Section B is now largely computer-generated), and other Sections will be tackled bit-by-bit over the next few days. Also, the 'actual' performance of our 67 portfolios (formerly shown in Section E) has now been moved into its own separate report: portacts.htm. Finally, the information in Section D that is extracted from the Prediction Table (PREDTABL) has been removed; subscribers should refer to the PREDTABL report if they wish to view this information.

*** New (7/27/07): Although our protective stop (optional) on Friday would have gotten us out of our QQQQ Long position when it had fallen -0.60% from the Open, it is clear in retrospect that Friday should have been a Shorting day. We have added some new logic to our Voted QQQQ Signal model that would have greatly improved our accuracy over the past few days, and made the interesting discovery that these additions also substantially improved the back-tested performance for the last 4 months of 2002; in other words, we are currently seeing some behavior that was last seen almost 5 years ago. The fact that our back-tested accuracy and gains actually increase (and when there are enough cases to give some confidence that we actually have a 'reasonable' model change), suggests that our Voted Signal model is quite capable of recognizing a wide range of market conditions, and that it is gradually 'learning' more and more cues that predict imminent market swings. The $64,000 question, of course, is how soon the model will reach a state of maturity at which ALL of our available information has been fully incorporated ... as opposed to a never-ending series of 'enhancements' that provide better and better 'back-tested' results, without necessarily boosting our ACTUAL gains.

*** New (7/26/07): Following an extensive post-mortem analysis of the dip on 7/26/07, we have concluded that there was no way available to us to foresee a dip of that magnitude ... and certainly not enough information was at hand to permit switching to a shorting signal. We have, nonetheless, enhanced the protective stop logic so that a stop 'would have been' issued with the trigger point set to the Open price of QQQQ – 0.60% (and this change is thus in place when/if similar events occur in the future). Since QQQQ only fell -0.80%, however, this stop would not have been very effective – even it it had been available at the time. We did, however, have a cautious state going into Thursday: (1) a Market Color Code of YELLOW, (2) despite an aggressive Long allocation of 73%, the prudent allocation was only 20%, and (3) our “H” portfolios were only 37% Long – while the intermediate-term (I/T) portfolios were only 30% Long.

*** New (7/25/07): (voteqqqq.htm Report) After adding a few more refinements to the QQQQ Voted Signal model, we have arrived at the following (back-tested) figures: Average Trading Accuracy of just over 74% (for 357 Long/Short trades), Average Signal Duration of 3.47 days, and Long 61.3% of the time vs. Short 38.7% of the time. We first reached a Long percentage of 58% several days ago (whereas our original Voted Signal model was on the order of 50-51% Long), and now we have exceeded 61%. We believe that this is essentially the theoretical maximum; notice that the Long% divided by the Short% gives 1.58 (61.3/38.7 = 1.58). This is very close to the Fibonacci Ratio of 1.61803, and we believe (conjecture) that over a sufficiently large amount of time, that a 'perfect' model will have a Long/Short ratio of EXACTLY the Fibonacci Ratio – that in fact this is an innate characteristic of the stock market. We have encountered this ratio before when developing our Grail System models, and we intend (some day) to 'prove' that the Fibonacci Ratio does indeed play a role here. But at the very least, these figures suggest that the Voted Signal model will stay LONG to the maximum extent possible ... and when it Shorts, it's apt to be highly accurate.

*** New (7/24/07): In order to confirm the 'BUY' signal for 7/25/07 we re-examined our 'Fed' logic for Beige Book Days. When we made this logic a bit more sophisticated (and tuned it over our 5-year backtesting period), not only did it confirm the BUY for Wednesday, but it also boosted our back-tested annualized gains considerably. More importantly, our QQQQ Voted Signal model has now reached the 60% Long level (in back-testing it is Long 60% of the time and Short for the remaining 40%). This model really only signals Shorts when they are likely to succeed, and far prefers to be Long most of the time.

*** New (7/20/07): After a five month period (2/16/07-7/20/07), we have now concluded the primary developmental effort on the Voted QQQQ Signal report (voteqqqq.htm) with the finalization of our 'Last Ditch Long' logic.  This new logic examines every situation in which QQQQ has risen in value for at least one day (and up to 4 consecutive days), and utilizes every indicator available to us in order to decide if a Long signal is warranted.  This logic guarantees that we will quickly switch to a Long signal whenever it is justified, even if our underlying Grail System QQQQ signal (GR-L) remains in the SELL state.  As a result of this change, the Voted Signal is now Long 58% of the time (and Short 42% of the time) over our 5-year back-testing interval.  Our previous Voted Signal was only Long about 50% of the time.  On the positive side, this new algorithm produces a much higher annualized gain, with a lower maximum drawdown, over our back-testing interval.  On the negative side, the absolute trading accuracy has dropped from about 72-74% to the 70-71% range. This is still exceptionally good; we would be on the first plane to Las Vegas if we could win 7 Blackjack hands out of 10. There are other minor enhancements that will be made from time to time to this key Report, but we don't expect any quantum leaps in performance ... just incremental gains in trading accuracy, and in the clarity of the report format.

*** New (7/20/07). The PPF (Predicted Performance Factor) value has been reformulated so that true high-performanceogic stocks are easier to spot.  The average PPF for our Grail System equities now calculates out at about 25, and any equities in the BUY state with this PPF or greater are marked with an asterisk (*).  The PPF is a function of the G/d ratio (Annualized Gain divided by Average Maximum Drawdown), the Annualized Gain, and the model's back-tested trading accuracy.  Unlike the PPF values shown for our Intermediate-Term (I/T) equities and portfolios (see SECTION2.TXT and the VARPORT reports) which utilize some fundamental data (e.g., valuations), the PPF for the Grail System is strictly based on expected short-term performance (and at the higher trading frequency characteristic of the Grail System). Remember, however, that those equities with very high PPFs are going to be more volatile.

*** New (7/18/07): The Momentum Logic in our Grail System models has been further enhanced; this change causes quicker switchovers from BUY to SELL (or from SELL to BUY) when there have been N days (1 day, 2 days, etc.) of price decrease (or increase) AND the CNNI (composite neural network) value meets specific criteria. This change helps ensure that if the price of a stock is ratcheting up, then we will assign it a BUY signal if at all possible.

*** New (7/9/07): The 'E' Class portfolios (PORTSTAT) have been re-engineered to boost their annualized gains. Previously, there had been very little difference in performance between the 'E' and 'U' portfolios, whereas the 'E' portfolios were supposed to generate higher gains (with slightly higher drawdowns) than the more conservative 'U' portfolios. The 'E' class now performs more optimally, striking a balance between the conservative stock portfolios ('C'), and the conservative ETF portfolios ('U').

*** New (7/7/07): In our continuing effort to nudge the QQQQ Voted Signal (voteqqqq.htm) a bit further to the “Long” side, we have now reached a point where the signal is Long 52% of the time, and Short only 48% of the time (and essentially NEVER in cash). It may be possible to push the distribution a bit more, but this will have to be done over time – and on a case-by-case basis. Our back-tested 2002-2007 gains are now: 116%, 376%, 116%, 59%, 80%, and 42%. The 2002 gains were for the period 8/16/02-12/31/02, and the 2007 back-tested gains are year-to-date. The 2007 projection is now quite healthy, and we would love to make this much in actual trading over the next 6 months. We are looking more closely at 2005, however, to see why the back-tested gains were so much lower that year. The maximal gains, of course, are only attained if all of our protective stop alerts (and profit-grabbing signals) are followed; those gains would have been (again, back-tested): 161%, 489%, 126%, 64%, 91%, and 52%. The current Voted Signal would have made 283 trades over the past 5 years, and would have achieved an average trading accuracy of 73-74%. As always, however, the caveat applies that future performance levels cannot be guaranteed...

*** New (7/7/07): At this point over 40 Grail System models have been replaced by higher G/d (Annualized Gain divided by Drawdown) alternates. This is important because our Portfolio System (PORTSTAT) only utilizes those equities with reasonably high G/d ratios (hopefully 25.0 or better). The new models also include about a half-dozen ETFs, and this will continue to help our “E” and “U” portfolios come up with their requisite number of holdings.

*** New (7/6/07): We are enhancing the Voted Signal Report (voteqqqq.htm) by implementing several algorithms that are collectively referred to as 'Last Ditch Longs'.  The idea is to identify all days on which QQQQ has risen in price while we are currently in a SELL/S (SHORT) interval. We then divide these cases up based on whether QQQQ has risen for only 1 day (and was down the previous day), or whether it had risen for 2 or more days (actually, the 3+ case is already adequately handled by our current Reversal BUY logic). We then look at our full set of indicators to see if any one of them, or a combination of them, might be usable to generate a 1-day BUY signal (it might last longer, but one day of gains is quite acceptable).  As usual, any such rule must meet our standard criteria:
(1) It must be 'logical'; the indicators used must actually suggest a Long situation and not be just an arbitrary combination of values and limits.
(2) It must improve the bottom-line performance of our QQQQ signal over our entire back-testing range (currently 5 years), and ideally there should be multiple instances (not just one) in which the new Rule improved performance.
Thus, the central idea is to go the extra mile to try to go Long ... to 'err' on the Long side as it were.  There are three good reasons for this:
(1) The long-term historical trend of the market is UP.
(2) The Voted Signal itself is Long essentially 50% of the time, and thus pushing for another percent or two of Long situations is quite reasonable.  It 'seems' logical that the percentage of Long days should be GREATER than the percentage of Short days ... and not just equal.
(3) For an individual stock the Long interval is about 1.6 times its Short interval.  In other words, if you simply pick a day to go Long for a stock then your odds of winning (picking an Up day) are pretty good.  We believe that the exact ratio (Long interval length divided by Short interval length) is the Fibonacci ratio (1.61803), but we haven't yet figured out how to 'prove' this.

*** New (7/6/07): We have improved our 50 PORTSTAT portfolios by giving up on the idea of having 50 sets of portfolio parameters.  Up until now each portfolio has had its own set of 'tuning' parameters (20 variables), and had to be tuned separately.  As a result, it was impossible to tune all portfolios every night (we simply didn't have enough time and CPU power).  Tuning is highly desirable because our portfolios using our Grail System models.  These models are highly sensitive to the MTI value, and tend to fluctuate from day to day as our Preprocessor stage introduces small variations in the 'historical' values for the MTI.  The bottom line was that our portfolios were rarely 'optimal' (we were always able to tune them over a weekend), and showed strange variations from portfolio to portfolio within a given portfolio Class ("H", "N", "C", "E", and "U"); for example, the P6N and P8N portfolios might be rapidly filling with stocks while the P7N stayed mostly in cash.  The solution to this problem was to pick a 'master' portfolio within each Class (usually with about 6-8 stocks), and to heavily optimize this portfolio every night.  Then, these same portfolio parameters were used for all of the other portfolios (smaller and larger than the 'master' portfolio) within the same portfolio Class.  This approach is now being used and offers the following benefits:
(1) The PORTSTAT portfolios are now always fully tuned and optimized ... providing the highest possible annualized gains with the lowest possible drawdowns.
(2) Portfolio parameters now vary uniformly as one moves from the small portfolios toward the larger ones (there is a smooth gradation in annualized gains, drawdowns, and trading frequencies).
(3) All portfolios within a given Class tend to fill and empty in a similar manner, and at a similar rate.  Also, the holdings of each portfolio tend to closely reflect the holdings of the portfolios that are slightly smaller, or slightly greater.  For example, the P7N portfolio might have all 6 of the P6N equities, plus a 7th equity of its own.  This uniformity makes it far easier for a subscriber to move from one portfolio to another one that is slightly larger or smaller.

*** New (7/6/07): The majority of our 1000 Grail System models currently use the Voted QQQQ Signal as a subsididary input.  This recent change vastly boosts the G/d ratio (Annualized Gain divided by the Average Maximum Drawdown) for each model.  Although average trading accuracy consequently drops somewhat, and trading frequency increases substantially, the gains in performance are well worth it.  The Grail System models, of course, are the equities that are used within our PORTSTAT portfolio system (50 portfolios), and improvements to these models has an immediate positive effect on the portfolios.  We have made recent enhancements to our tuning process that speeds things up considerably; this is not only permitting us to optimize all models every night, but also to gradually substitute new models (with higher G/d ratios) for current models that are not performing as well.  The models with low G/d ratios are not utilized by our portfolios anyway, so phasing some of them out in exchange for high G/d ratio equities is a good strategy.  Our objective is to always maintain the Grail System as the 'best' set of 1000 equities that can be identified.

*** New (7/5/07): We have moved our historical record of QQQQ signal performance to a separate file. The link to that file is: http://www.schulenberg.com/download/history.htm

*** New (7/3/07): We have incorporated our Grail System signals for IWB/IWM/IWV into the Voted Signal Report. Those signals, coupled with some algorithmic improvements having to do with our 'new quarter' logic gives us a BUY signal for QQQQ. In fact, it would have given us a buy signal for Tuesday as well ... although the signal was very marginal for that day.

*** New (7/2/07):

(1) The follow-up of our Post-Mortem analysis of the pre-Fed market surge on 6/27/07 resulted in the addition of a new 'generalized' rule to our Voted QQQQ Signal algorithm. It turns out that the surge really had nothing to do with the fact that the following day was going to be a Fed announcement day – rather, it was because we had had 3 'down' days accompanied by a jump in our NN1 (1-day QQQQ price projection) neural network value. The implementation of this new Rule improved back-tested performance significantly over our 5-year optimization interval, and it made sense from several standpoints: (1) it is logical to expect a rebound ... especially after 3 down days, (2) when in doubt it is better to err on the side of going Long after a series of down days, rather than staying short, and (3) under such negative circumstances even 'slight' positive signs may be sufficient to augur a turnaround. Also most interesting is the fact that the magic number is 3 or more 'down' days ... if there have only been 2 'down' days then there is NO WAY to justify going Long based on any or all of our indicators (over the 5-year backtesting period).

(2) We have dropped a new 'engine' into our MTI calculation program; after 3 solid weeks of optimization on a dedicated computer, we have switched to an updated MTI algorithm that fully incorporates recent market action (on top of our 5-year set of data). This change improved our GR-L model (whose back-tested accuracy had dropped to only 94%), giving it a 97+% accuracy for Long trades, increased its trading frequency slightly (from 78 to 82), and finally, improved the back-tested accuracy of the Voted QQQQ Signal a bit as well. We hate like hell to incorporate new tuned parameters into our preprocessor because it temporarily clobbers all of our models; it took 2 days of work to re-establish a stable set of 1000 Grail System models following this enhancement. In general, we incorporate new Preprocessor parameters only about once per month because of the inordinate amount of work that this process creates.

(3) We have completed the retuning and rebalancing of all 50 PORTSTAT portfolios (“H”, “N”, “C”, “E”, and “U” classes). All portfolios now tend to fill more quickly (when the market strengthens) and also empty more quickly (when the market weakens), and in general should provide far greater gains during upturns. So, not only has the Voted QQQQ Signal been incorporated into most of our 1000 Grail System stock/ETF models, but the portfolio logic (PORTSTAT) makes use of this short-term signal as well in order to improve overall portfolio performance.

(4) We have eliminated the '100 Club' and the '95+ Club' because of the recent modification of our Grail System models to make use of the Voted QQQQ Signal. As a result, annualized gains have been boosted, and average maximum drawdowns have been lowered, but by the same token the effective back-tested Long trading accuracy has decreased (because we are more interested in making money than in being 'correct'). The decrease in the effective Long trading accuracy (now down to 75-85% on average) meant that the numbers of equities in the 100 Club and 95+ Club had dropped to the point that the concept no longer made much sense. Instead, we have added a very large QQQQ Club (consisting of all Grail System models that do in fact make use of the Voted QQQQ Signal), and a '80+ Club' (consisting of all models with back-tested accuracies of 80% or greater). The percentages of equities in the BUY state in each of these two 'Clubs' thus become new benchmark indicators of overall market strength ... with the QQQQ Club BUY% being a VERY sensitive estimator of market health.

*** New (6/28/07): The “C” (conservative) PORTSTAT portfolios are now fully operational, and preliminary tunings have been made for the “E” and “U” (ETF-only) portfolios. We will let all 50 portfolios optimize themselves over the weekend, and will thus start the new quarter with a revitalized set of portfolios that utilize the newly enhanced (incorporating the Voted QQQQ Signal) Grail System signals.

*** New (6/27/07): The “N” portfolios (see PORTSTAT) are now fully tuned, as were the “H” portfolios yesterday; the “C”, “E”, and “U” portfolios will be fully tuned by Monday.

*** New (6/26/07): The “H” portfolios (see PORTSTAT) are now in a near-optimal state, and the remainder of the portfolio classes should be finalized within the next day or two; there is, of course, no real hurry – there is currently nothing to buy anyway.

*** New (6/25/07): We have completed an initial effort to get the PORTSTAT system (50 portfolios) back on its feet after the major design shift made to our Grail System over the weekend. Since the portfolios remain empty, we have another day or two (at least) to complete this effort by exhaustively optimizing all portfolio parameters.* New (6/25/07): The Voted QQQQ Signal report (voteqqqq.htm) has been enhanced slightly; our rebound detection logic now correctly identifies the 'bounce' on Thursday, while retaining the Short signal for the two dips on Wednesday and Friday. This retroactive change doesn't help us at the moment (we still missed out on a large upward/downward combination), but the program should recognize similar situations in the future when we 'will be able' to profit from them.

*** New (6/25/07): Our SIGNALS report clearly shows that most of our Grail System models have now become 3^rd Generation Models, that is, they incorporate the Voted QQQQ Signal as a short-term 'guide' to enhance their signal accuracy. It will take a few more days, however, before this change has been fully incorporated into our 50 standard portfolios; since all of our portfolios are currently empty, however, this is a good time to make the transition.

*** New (6/21/07): We are in the process of making a major enhancement to our 1,000 Grail System stock/ETF models (as summarized in the SIGNALS.TXT report). Just as we previously used one of our highly accurate ETF signals (IVV) to increase the overall modeling accuracy of other (less accurate) models (creating so-called “Second Generation” models), we are now beginning to merge the Voted QQQQ Signal back into our Grail System models. This is akin to a 'feedback' process; we use the Grail System signals for QQQQ (both the Long-optimized GR-L model, and its Short-optimized GR-S variation), SPY, DIA, the Fed Calendar, and a slew of other factors, to create the Voted Signal in the first place. Since the Voted Signal is a shorter-term signal than the standard Grail System signal (GR-L) (in fact it changes state about 3.3 times more frequently), it is just the thing to increase the performance (lower the drawdowns) of many of our other models. This new endeavor is proceeding in two Phases: Phase 1 encompasses the re-engineering of all 1,000 models to incorporate the Voted Signal – as appropriate (it isn't going to help some of the models). The results of Phase 1 will gradually appear in the SIGNALS report as the short-term performance of our models improves (mostly by decreasing drawdown and increasing trading frequency). Phase 1 will be complete by Monday (6/25/07). Phase 2 will take several more days; it involves retuning our Portfolio parameters so that the PORTSTAT portfolios will optimally select the new models (since trading accuracies will effectively be somewhat lower, and trading frequencies somewhat higher, the existing portfolios would at first glance reject many of the new models). This is a MAJOR enhancement to the bottom-line performance of our short-term (Grail System) portfolios, and it should be complete (both Phases) by the end of June.

We have been grateful for the forbearance of our QQQQ traders during the past few months while we have evolved the Voted Signal Report (voteqqqq.htm). We started with our Grail System QQQQ model (GR-L), which is a highly accurate 'swing trading' model, added the short-optimized GR-S model and everything else that we could think of, and we have now arrived at a highly accurate 'short term' QQQQ model. The underlying GR-L model remains quite accurate; as a 'swing trading' model it has a back-tested accuracy of 100% for QQQQ Longs over the past 5 years, and an accuracy of about 91% for QQQQ Shorts. However, since it is a 'swing trading' model, the GR-L model (and its companion GR-S) tends to stay Long (or Short) for appreciable periods (e.g., 12-13 days on average) ... usually 'winning' in the end, but incurring drawdown during these protracted intervals that often appears to be 'avoidable'. The answer to this kind of behavior was the development of the Voted Signal Report. This signal (when back-tested) changes state about 255 times during the past 5 years (compared with 76 times for GR-L), and currently has a back-tested YTD 2007 gain of over 20% (if no stops are used) and 35% (if all stops are taken). It remains to be seen, of course, if this level of performance can be maintained under actual trading conditions in the future, but one thing is certain: the current model is very responsive to short-term events, and doesn't get 'stuck' in either a Long or a Short state for more than 4-5 days, on average.

*** New (6/20/07): The FOMC (Fed) calendar information within our Voted Signal report (voteqqqq.htm) is now shown with different informational messages depending on which of the 3 basic types of 'Fed' events are about to take place: (1) A Fed rate change announcement, (2) the release of the Minutes from a prior Fed meeting, and (3) the release of the Beige Book. In all cases, appropriate Protective Stop alerts are issued. It is interesting, however, that such 'stops' are unnecessary (and in fact counterproductive) if we are LONG on the day of a Fed event; it is only during shorting periods that these extra precautions are necessary.

*** New (6/18/07): An important update has been made to the QQQQ Voted Signal Report (voteqqqq.htm) that substantially improves its short-term responsiveness. This increases the frequency of signal changes (thus increasing the number of Long/Short intervals), and boosts the bottom-line (back-tested) performance. In particular, the back-tested performance for the 2005-2007 timeframe has been greatly improved; the YTD performance (back-tested) for QQQQ is now over 17%, and with the maximum gain being over 30% (if all protective stops and profit-taking opportunities were exercised). Some of this improvement was due to the recent inclusion of the FOMC (Fed) calendar, but the bulk of this improvement was achieved by incorporating our new MRI indicator which detects ultra-short-term variations in QQQQ strength and 'tweaks' the voted signal appropriately. The bottom-line trading accuracy is actually lowered by increasing the number of trading intervals, but since we make much more on our 'wins' than we give up on our 'losses', the net effect is to make more money; the back-tested trading accuracy (over our 5-year test period) now varies between 68% and 73%, depending on whether stops are utilized.

Recent Development -- and Work in Progress: The record-breaking behavior of the market in 2007 has been a stress test of our 'thermodynamic' approach to market timing, and has highlighted 3 aspects of our System that have not performed up to our expectations; all 3 of these problems have now been addressed ... and solved.

(1) The Grail System QQQQ signal was shorting too aggressively at times when the market was showing obvious strength. Although the resultant drawdown was within historical 'specs', it was obvious that some of that drawdown was avoidable. The solution, implemented over the past few months, was the Voted Signal Report (voteqqqq.htm) that greatly increased the short-term responsiveness of the voted QQQQ signal ... and thereby reduced drawdown (as well as boosting overall gains).

(2) The Intermediate-Term (I/T) signals that are summarized in our SECTION2 report were predominately 'long', reflecting recent market strength, but our VARPORT portfolios (VARPORTA-G) were nonetheless staying largely in cash and missing out on profit opportunities. This was due to the fact that our Grail-System-based allocation vectors were being used to control the degree of 'long-ness' for our intermediate-term portfolios. This overly conservative behavior has now been improved by developing an allocation vector that is specifically tuned to the needs of our intermediate-term signals. As a result, our VARPORT portfolios are now maintaining a reasonable level of positions ... and are making money.

(3) Our MTI value (calculated by our 'thermodynamic' method) appears to have been too 'cool' over the past few months; although it has stayed at or above the neutral value (1.0) for much of 2007, its average value has been lower than we would have expected given the tenacious strength of the market. As a result, our Grail System models have given us a rather low Buy/Sell ratio, and this in turn has kept the Grail System (PORTSTAT) portfolios in a rather conservative state (an excessive number of cash positions). We have now implemented a 'feedback' process that utilizes the recent performance of Grail System stocks/ETFs to bias the MTI value slightly higher at times to reflect how well the Grail System equities are actually doing. This 'feedback' process is limited to small corrections, however, so that the predictive ability of the 'thermodynamically-generated' MTI value is not compromised (we want to ensure that QQQQ remains 'short' prior to the 2/27 meltdown, for example). So, the correction is fairly small (boosting the MTI by at most 0.25-0.35), and infrequent (when MTI is between 1.0 and 1.25 AND the recent Grail System performance suggests short-term strength).

*** New (6/14/07): The Voted QQQQ Signal Report (voteqqqq.htm) now includes the complete Fed (FOMC) calendar: (1) Fed interest rate announcements, (2) Release of the Meeting Minutes, and (3) Release of the Beige Book. This causes protective stop alerts to be generated for the day (and following day) of most of these three types of occurrences – and especially if we are shorting QQQQQ at such times. This effect (the increase in bottom-line gains) is relatively small, but every little bit helps. It is generally also true (on 'Fed' days) that when such a protective stop is triggered, then it is profitable to switch to the opposite position for the remainder of the day (Long>Short or Short>Long), and then re-establish the original position prior to the Close (or at the next Open). Such 'reverse plays', however, are not reflected in our performance numbers – and carry an obvious level of risk.

*** New (6/12/07): Reversal SHORT logic has been added to the Voted Signal Report (voteqqqq), but this appears to be just gilding on the lily; our Long signal was already extraordinarily accurate, and Reversal SHORTs provide only a small, incremental gain. The development of the Voted Signal Report is now essentially complete; there are still a few aspects that call out for further study and refinement, but we feel that this is the MOST ACCURATE QQQQ signal that has ever been devised.

*** New (6/8/07): The Rebound Detection Logic in the QQQQ Voted Signal Report has been enhanced to key off our #S (# of preprocessor Shorts) parameter as a possible bounce indicator. This new logic, after tuning and back-testing, would have given us a protective stop alert on Friday with a trigger point at the Open price of QQQQ + 0.282%. This would have been a nice savings. The ONLY indicator that we had that might have indicated a bounce possibility was the #S indicator, and to look for a bounce possibility when #S > 0 is definitely a CONTRARIAN move since this is a very NEGATIVE indication under normal circumstances.

*** New (6/5/07): A new Differential Signal ($NQ-AU) has been added to suggest the relative profitability of trading in QQQQ vs. gold (the GLD ETF). This signal has just switched to a '-/+' state, suggesting that a simultaneous Short of QQQQ and a Long on GLD might be profitable. This signal is currently only shown in the SIGNALS report, and in any event, is still 'experimental'. Since gold doesn't really provide an effective hedge (especially against a QQQQ Short), this signal should be viewed as primarily informational.

*** New (6/1/07): The MTI calculation has now been enhanced so that it provides much more accurate estimates of market 'temperature' when the market is near a neutral state (MTI = 1.0). This is done by looking at the last 1-3 day's performance of all 1,000 Grail System stocks/ETFs, and appropriately boosting the MTI value if this recent performance suggests that the market is a bit 'hotter' than 1.0. If the MTI value lies within the range 1.0-1.25, and if the Grail System price gains for the recent period are supportive, then the MTI will be temporarily augmented (for 1 day only) by 0.25-0.35. This slight correction will kick in infrequently, but when it does the boost is sufficient to flip QQQQ from a SELL to a BUY state, and to cause the PORTSTAT portfolios to add to their current Long positions. Although this change causes some degradation of the back-tested performance for 2003-2004, the 2005-7 performance of the QQQQ signal retains its accuracy, and just as important ... the Grail System portfolios (PORTSTAT) should take better advantage of these short periods of market strength. The use of the Grail System portfolios (their actual performance) to modify the MTI value is appropriate; after all, the QQQQ signal is used to initially select the 1,000 equities that comprise the Grail System, and these equities thus tend to 'behave' a lot like QQQQ. To then use short-term actual performance to modify the MTI slightly simply reverses the process ... providing a 'feedback' loop in which the Grail System is able to influence the QQQQ signal itself and ensure that 'dangerous' SHORTs are terminated before they cause significant harm.

*** New (5/29/07): The intermediate-term (I/T) portfolios (see VARPORTA-G reports) have been modified so that their 'long' allocations are now substantially larger (and generally held for longer periods) than they were previously. This change was suggested by the results from a just-completed study into the optimal usage of our intermediate-term signals. This has also given us a new indicator -- the intermediate-term Long allocation – that now takes its place alongside our other allocation (hedging) vectors. Our I/T portfolios (with their much wider mix of equities of all types) now behave quite differently from their Grail System (PORTSTAT) equivalents that are highly sensitive to our QQQQ signals – and which tend to shed positions quite aggressively when market weakness is detected. The I/T portfolios can generally hold a larger number of holdings because they include a more diversified mix of equities, including a higher percentage of energy stocks.

*** New (5/22/07): The machinery that underlies our Voted Signal Report (voteqqqq) has been enhanced so that all empirical report parameters (37 constants at the current time) are maintained in a high state of optimality. This means that the voted QQQQ signal is always as accurate as we can make it every day .... despite the small fluctuations in the back-tested MTI value (due to the way that the MTI is calculated) that we have to contend with.

**** New (5/11/07): We have incorporated a new Rule in the Voted Signal Report that would have given us a BUY for Friday (5/11) based on our NN3 neural network (3-day forward projection). Using this same Rule, however, gives us a renewed Short for Monday. The Voted Signal Report (voteqqqq) will show Friday as a BUY based on this back-tested Rule, and thus doesn't count for anything officially (officially, we held our Short position on Friday, and with NO protective stop). In the future, however, this Rule may prove profitable when similar circumstances arise again – although this specific combination of market conditions occurs rarely. It is interesting to note that attempts to set a protective stop based upon those conditions would have resulted in degraded performance during the back-testing period; protective stops must be set ONLY when conditions seem to warrant them.

QQQQ Voted Signals Report: the Ultimate QQQQ Timing Model

The QQQQ signal shown in our Voted Signals Report (voteqqqq) is generated by a 3^rd generation timing model. Everyone has 1^st generation models; they rely on price and volume, and various indicators, oscillators, and neural networks. Our Grail System models (SIGNALS Report) are 2^nd generation timing models; they are dependent on the MTI (Market Timing Indicator) value PLUS all of the standard data utilized by traditional timing models (our Grail System models also take fundamental data into account). In the Voted Signals Report, however, we have a 3^rd generation timing model; the Voted QQQQ signal is a function of 9 different Grail System models: GR-L (Long-optimized QQQQ model), GR-S (Short-optimized QQQQ model), IXN (Global Technology ETF model), DIA, SPY, SMH, SWH, IGN, and IBB. In addition, sophisticated methods are used to minimize drawdown: protective 'stops', profit-taking exits, Reversal BUYs (during shorting periods), and Reversal SHORTs (during Long periods). This Voted Signal has evolved very rapidly in recent weeks, and this timing model will continue to be a major focus of development in the months to come. This signal is especially important because it is our ONLY shorting signal – and 'shorting' is intrinsically so hazardous that ALL possible protective measures should be utilized. As a result of these extra layers of logic, the 'voted' QQQQ signal changes state a bit more frequently than a typical Grail System model.

*** New (5/4/07): A Portfolio Performance Tracking system has been developed, and its results will henceforth be posted in the daily Forecast email. This tracking system permits direct comparisons between all of our 67 portfolios, and will maintain both day-by-day and cumulative performance figures.

*** New (5/4/07): The QQQQ Voted Signal (voteqqqq) has now been propagated throughout our portfolio system. The use of this new QQQQ signal has boosted portfolio performance somewhat (larger annualized gains, lowered drawdowns), but more importantly it has increased the overall Fill% (more holdings and less cash) at a given MTI level. It will take several more days before all 67 portfolios are optimally tuned, and a useful tool for accomplishing this is the new Portfolio Performance Tracking report; it will help ensure that all portfolios within a given family vary smoothly in the key parameters (annualized gain, maximum drawdown, trading frequency, and Fill%). Prior to the utilization of the voted QQQQ signal our Grail System portfolios had already been orecently improved by the use of the new timing models with MTI-dependent price momentum logic.

** New (5/2/07): The shorting accuracy of the QQQQ Voted Signal Report (voteqqqq) has been further improved by the incorporation of additional Grail System models into the 'voting' logic. These new models include SPY (S&P 500), DIA (the Dow), and IXN (iShares S&P Global Technology Sector), with IBB, IGN, SMH, and SWH as secondary contributors. The IXN signal is used as additional corroboration for our QQQQ shorting signal, and thus supplements the GR-L (Long-optimized QQQQ model) and GR-S (Short-optimized QQQQ model). IXN is useful because it is our ONLY model with a perfect (100%) backtested shorting accuracy over the past 5 years. SPY and DIA are important because if they are in the BUY state they can 'carry' QQQQ higher, or certainly make QQQQ 'shorts' rather dangerous. We are thus using a family of Grail System signals to produce our reference (voted) QQQQ signal. The Grail System models themselves, of course, were recently further boosted in accuracy by the addition of MTI-dependent price momentum logic.

*** New (4/30/07): The 1000 stock/ETF models in our Grail System (see the SIGNALS report) have been augmented with an MTI-dependent price momentum algorithm designed to improve their handling of short-term market trends (making our swing-trading signals work better for shorter time scales). This logic is similar to that which was recently added to our Voted Signal Report (voteqqqq), i.e., the Reversal BUY logic. There are now 56 empirical constants required by each Grail System model, giving us a total of 56,000 parameters to be optimized on a daily basis. The accuracy and drawdown figures for each Grail System model have been improved slightly as a result of this addition, and this increase in performance will pass through to our 50 PORTSTAT portfolios.

*** New (4/26/07): (Voted Signal Report – voteqqqq). An important extension has been made to the new 'short-term momentum' logic: the algorithm that tries to extend the duration of the Reversal BUY periods within a shorting interval. These Reversal BUYs are important because they reduce the drawdown (and increase the potential gain) that is inevitably associated with using swing-trading frequency timing models (i.e., the GR-L and GR-S QQQQ models) to 'short'. Despite the high back-tested accuracy of these two fundamental signals for QQQQ Longs and Shorts, these signals don't solve the problem of short-term countertrends that eat into our gains, as well as increasing drawdown levels. The 'momentum' logic has now been enhanced so that the BUY intervals are stretched out longer when either of the following conditions are met: (1) DIA has attained a new High (calculated during the past 60 trading days, (2) QQQQ has attained a new High (calculated during the past 60 trading days). Recognition of the powerful propulsive effect of approaching milestone index levels not only improves performance during March-April 2007 (as the DOW attained record levels), but it also improves bottom-line performance during our entire backtesting interval. The Reversal BUY signal itself is triggered, of course, when QQQQ's Close price is increasing AND the MTI (Market Timing Indicator) value exceeds a critical value.

*** New (4/25/07): We have added 'short-term momentum' logic to the Voted Signal Report (voteqqqq.htm). This algorithmic change extends the average duration of the Reversal BUY periods that occur within the shorting intervals of our basic swing-trading signals. In recent weeks there has been a staged development of methods for reducing the shorting drawdown that is inherent with any signal, including ours. First came the Type I and II Protective Stops, and this was followed by the Type III Profit-Grabbing 'stops'. Then came a Reversal BUY algorithm that detected upswings in the MTI value within a shorting interval, and capitalized on them by temporarily switching the signal to a BUY. Finally, the newest change extends the duration of the Reversal BUY so that more of the short term counter-move can be captured. Since these changes take place on the 'voted' signal (within the voteqqqq.htm report), they have no effect on the accuracy and signal duration of the two Grail System signals that we use for QQQQ: GR-L (our Long-optimized standard model) and GR-S (our Short-optimized QQQQ model). We have no wish to tamper with these fundamental signals because (as swing-trading signals) they are already highly accurate; GR-L has a back-tested accuracy for QQQQ Longs that is about 97% (over a 5-year back-testing interval), and GR-S has a similar accuracy for QQQQ Shorts. It is when these two signals are blended together within the Voted Signal Report that the above-mentioned additional measures are taken to reduce drawdown to the lowest possible levels.

*** New (4/15/07): We are now using a new QQQQ short-optimized model (GR-S) in our Voted Signal Report (voteqqqq.htm). This model backtests at 97+% accuracy for QQQQ 'shorts', and is thus our most accurate shorting model ... ever. Even before the inclusion of this model, however, our Voted Signal had switched to a Reversal BUY; both of our Long-optimized (GR-L) and Short-optimized (GR-S) models agree that QQQQ is still in a SELL state, but recent momentum (and the current MTI value) gives us sufficient impetus to trigger a 'reversal BUY'. This may be a very short-lived signal, but it is the one that the numbers (and our current algorithms) force us to accept. Note that the use of this new Reversal BUY signal will increase the frequency of QQQQ signal changes that we report to TimerTrac, although we still have one of the most stable QQQQ signals in existence (reflecting its derivation from swing-trading signal models).

*** New (4/10/07): A significant enhancement was made to the Voted Signal Report: the Reversal BUY logic. This is the natural culmination of our recent development work: (1) Type I and II Protective Stops, (2) Type III Profit Stops, (3) Early exits from successful Shorts, (4) Drawdown Alert messages, and now, (5) the Reversal BUY. A 'Reversal BUY' is a brief (usually 1 or 2 day) BUY that is interpolated in the midst of an ongoing SHORT (SELL/S), and which is intended to reduce drawdown by anticipating a significant short-term rise in QQQQ, and not only covering the Short, but also going “long” briefly before reimposing the Short. This amounts to a temporary 'override' of the Voted Signal, and it raises the QQQQ trading frequency by about 20-25%, but given the rather infrequent nature of our QQQQ signal this is not a significant problem. If it is not feasible to follow the Reversal BUY signal, then the prevailing SHORT (SELL/S) can be continued, although it will usually result in a higher drawdown than would be the case if the 'reversal' can be followed. The Voted Signal Report (voteqqqq) makes it clear when the 'BUY' signal is a very short-term override (Reversal BUY), rather than a MAJOR change of signal to a BUY state. The Reversal BUY signal allows us to enhance the short-term responsiveness of our swing-trading signal, and this new concept of a 'voted' signal gives us the freedom to modify the signal so as to achieve maximal performance ... without significantly impacting our overall approach (e.g., the Grail System models and portfolios).

*** New (4/9/07): A further significant enhancement was made to the accuracy of the voted QQQQ signal (see voteqqqq.htm): instead of going with a SELL/$ (CASH) signal when our Long-optimized model (GR-L) was a SELL, and the Short-optimized model (GR-S) was a BUY, it turns out to be quite advantageous to stick with a BUY signal if the MTI value is >= 1.0. We thus split the SELL(GR-L)/BUY(GR-S) into two cases depending upon the MTI value, and this means that the BUY state (which is, of course, the long-term direction of the market) is active even more of the time ... and the CASH (SELL/$) state correspondingly less.

*** New (4/8/07): Type III (Profit Stops) have been added for QQQQ LONGs, so that this feature is now supported for both Long and Short signals. The statistics at the end of the Voted Signal Report reflect the use of these new 'stops' in the columns denoted VOTE(2). The VOTE(1) column continues to reflect the performance of the straight voted signal, without the use of any stops of any kind. Also, a new Drawdown Alert message is issued when we have a current shorting signal (SELL/S), and our indicators suggest that QQQQ is likely to rise during the next market day. During a shorting period there will always be days in which we are forced to give back some of our profits, but the issuance of this message may be helpful to some of our subscribers who are not quite as 'risk tolerant' when it comes to shorting. When these messages begin to appear, then the more prudent investor might prefer to cover (or protect) the short, and instead engage in some 'long' plays as described in the header material of the Voted Signal Report (e.g., (1) follow one of the Preprocessor NEWPICKS portfolios, L/1-L/10, (2) pick up a few stocks/ETFs in the 'buy' state from our SIGNALS report, or (3) follow one of our intermediate-term portfolios, VARPORTA-G). In other words, if one continues the 'short' in the face of successive Drawdown Alert messages, then it is an acknowledgment that any resultant drawdown is acceptable, given the strong possibility of eventual gains as the 'short' continues.

*** New (4/4/07): We have added a Type 3 (“Profit-Taking”) stop to the QQQQ Voted Signals Report, although not all of the documentation has been updated. We still plan to add a corresponding 'profit taking' stop for BUY periods, but at the moment we are focused on the 'shorting' side of the problem for obvious reasons. The Type 3 stop occurs infrequently, but it specifies a trigger point based on the previous day's CLOSE price for QQQQ. The idea is that if, during the day, QQQQ should fall below the trigger point (even at the Open), then the QQQQ short can be profitably 'covered' at that point ... and QQQQ can be re-shorted later in the day. The profit-taking stop is signaled when two conditions are met: (1) we are currently 'short', and (2) we have indications that the market will show some strength during the following day. In such a case, it is best to grab profits when they are offered. We are in the process of evaluating put/call ratios (and other available data) to see if we can more optimally pinpoint the best time to cover a short position. This is an extremely difficult problem, and there are no guarantees that we will be able to improve upon the current signals... but we are working on it.

*** New (4/3/07): We have made a minor change to the QQQQ Voted Signals Report (voteqqqq.htm). Previously, whenever we have had a divergence between our Long-optimized QQQQ model (GR-L) and our Short-optimized model (GR-S), we have interpreted this as a CASH (SELL/$) situation. We have now verified that performance is slightly better if the BUY(GR-L)/SELL(GR-S) case is interpreted as a BUY (leaving the SELL(GR-L)/BUY(GR-S) as a CASH case). This makes sense for two reasons: (1) The GR-L model has a terrific Long accuracy (back-tested), and this should be respected, and (2) The market tends to favor 'longs' over 'shorts' because the market generally moves up when given half a chance (so, in questionable cases it turns out to be worthwhile to stay 'long').

*** New (4/1/07): We have completed our total retuning of our 10 primary Preprocessor portfolios (L/1-L/10), and the updated models have been incorporated into our MTI calculation, and into our 1000 Grail System models as well. Since this was a major perturbation to our fundamental algorithms, it will require a few more days before our entire system has reattained a steady-state equilibrium ... but in any event there have been no significant changes to our historical (or current) signals. We have now instigated a continuous tuning process for the Preprocessor; this will ensure that our MTI calculation is always as accurate as possible, and also that only very minor ripples will be created when newer tunings are incorporated each week. One of the reasons that we haven't retuned in a while is that this causes an enormous discontinuity to our MTI calculation algorithm, and requires days of re-optimization to ensure model stability; now, by tuning just one or two of the ten portfolios each week, we suffer only very minor impacts when a new tuning has been incorporated ... and yet our system will remain very nearly optimal at all times.

*** New (3/29/07): We have reduced the size of this document (forecast.htm) by moving more of the static material to separate linked files. Also, the ongoing trading experiment (L/1+QID) is now documented on a daily basis in the file exper.htm. This file is part of the daily report set.

*** New (3/23/07): In our new experiment to trade the Preprocessor's L/1 portfolio (see Section D), we have changed the procedure so that we utilize a QQQQ short (via the ETF QID) whenever the L/1 portfolio goes to 'cash' – BUT only if our QQQQ Voted Signal is a SELL/S (Short). In other words, if the L/1 portfolio has an equity then we buy it and hold it until the portfolio sells it. If we go to cash, and if the QQQQ Voted Signal is a SELL/S (Short), then we use the cash to buy QID (a 2x inverse QQQQ); otherwise we stay in cash. If we are holding QID (effectively a Short), then we will cover if one of two things happen: (1) L/1 buys another equity Long, or (2) the Voted Signal switches to BUY or SELL/$. Therefore, most of the time we will be either Long (in accordance with L/1) or Short (by using the QID inverse QQQQ ETF), and only occasionally in cash. So, the L/1 portfolio will control things, but rather than sit in cash we will utilize the Voted Signals Report so that we stay pretty much fully invested. If we are going to try to make as much money as possible with a small portfolio, then this approach is the most aggressive one that we can come up with. It should have an 'exciting' level of volatility :)

*** New (3/22/07): We have initiated a new experiment: following the L/1 Preprocessor portfolio (see the NEWPICKS report) with real money. Since this portfolio was just re-tuned several days ago, and since it has generated surreal back-tested annualized gains of 475% (a total 306,158% gain since 8/16/02), we just had to give it a try ... just for fun. We expect some interesting results, and with volatilities that can result in 20% gains in one stock, and 20% losses in the next. The average (back-tested) trading accuracy of this portfolio is only 58%, which is actually pretty good considering the kinds of stocks that it deals with. In Section D of this document we show the results after the 1^st day, and the results from succeeding days will be added as we go. The L/2 and L/3 portfolios are currently being re-tuned, and they will be carefully integrated into our system in about a week. This is a ticklish process since these portfolios are actually an important part of the MTI (Market Timing Indicator) calculation, and changes in portfolio performance can cause severe ripple effects in our Grail System models (1000 of them).

*** New (3/15/07): We have added more 'protective stop' performance information to the end of the QQQQ Voted Signals Report; in particular, a new Table shows the relative frequency of each Stop's occurrence (both the alert issuance and the triggering), and the average percentage gain for 'wins' ... and the average percentage loss for the 'losses'. We are considering incorporating the Table I information from our Prediction Table (PREDTABL), and this might well improve the accuracy of these 'optional' stops by looking also at the Market Open Signature (MOS), that is, whether QQQQ, SMH, and SPY open 'up' or 'down' from the previous Close. This should boost performance, if for no other reason than the fact that SMH is a major supporting player in whether QQQQ rises or falls –- and we already know that Table I is rather accurate (on a statistical basis).

*** New (3/14/07): We have finished with the refinement of our 'optional' Long and Short stops that are shown in the QQQQ Voted Signals Report (voteqqqq.htm). There are two kinds of Short 'protective stops' (Type I and Type II), and also two kinds of Long protective stops (again, Type I and Type II). Based upon back-testing, the data for these 4 stops can best be shown in tabular form:

The 'Stop Alert Frequencies' show how often the respective 'stop alert' is issued; for example, a Type I Short Stop alert is issued for 14.34% of market days (or about 1 market day out of every 7). The 'Stop Trigger Frequencies' show how often a 'stop alert' that has been 'issued' has actually been triggered during the course of a trading day; just over half (51.52%) of issued Type I Short Stop alerts will actually trigger. Finally, the 'Stop Accuracy' shows for what percentage of 'triggered stops' we actually saved money (by exiting a Short before the market began to move upward, or sold a Long before the market went down). Notice that there are NO CERTAINTIES in any of this data, and that is one reason that all such 'stops' are viewed as 'optional'. If we consider all 4 stops together then their frequency of occurrence will be about 21% (or one market day out of 5), but the Long Stops (Type I and Type II) and the Type II Short Stop occur rather infrequently (although they are almost 'sure things' when they do occur). What we can learn from this is that most extra gains can be made by utilizing the Type I Short Stop – despite its 56.47% success rate; of course, when these stops are successful they tend to save more money than the unsuccessful stops 'lose'. Another interesting point is that the accuracy of the 3 rarer 'stops' is sufficiently good that it might make tactical sense to hold a 'contrary position' between the trigger point (when the current Long/Short holding would be exited) and the market Close (when the original Long/Short holding would be re-established). This would be a much riskier procedure with the high frequency Type I Short Stops, although it should be quite profitable over a period of time. Note that the precise 'trigger point' (a percentage either above or below the Open price) now varies from day to day; the QQQQ Voted Signal Report (or this email) should be consulted to see the exact value. However, you need to use your own judgment if you employ these 'stops' to satisfy yourself that the 'stop' has truly been significantly 'broken' – and possibly add a 0.10%-0.15% of leeway before you take action in order to reduce the risk of being 'stopped out' (although this will cut into the possible gains, of course). In many accounts, of course, this type of day trading is not feasible (e.g., in retirement accounts), and the QQQQ Voted Signal would be followed 'without stops'. 

*** New (3/13/07): We devoted considerable time on Tuesday to exploring refinements to the new 'protective stop' logic within our QQQQ Voted Signals Report (voteqqqqq). Although there is much more to do, it turns out that the 'fixed' stop point of 0.50% (above/below the Open price for Short/Long stops) is indeed an accurate historical average, but that better results can be obtained if a variable stop point is used based on the following factors:

(1) A temporal factor (year-dependent) which sets the mean stop point higher during years of high volatity, and to more moderate values when the market shows more consistent trends.

(2) Upward and downward bias factors based upon the Preprocessor parameters #L (# of new Longs), #S (# of new Shorts), and the MTI.

(3) Upward and downward bias factors based upon Table II of the Prediction Table (QQQQ day-trading signal).

When the requisite tuning has been performed, the trigger point settings vary from about 0.20% to 0.72% (with an average of 0.50%). The program logic then 'knows' when the trigger point needs to be set more 'loosely', and when it can be tightened up so as to minimize losses. Of course, no 'stop' strategy can be perfect; we have worked on this problem periodically for over 6 years, and this is the first time that we have come very 'close' to an ideal solution. Still, there is nothing more frustrating than 'stopping out' by a penny or two and then missing a big move. For this reason the protective 'stop' trigger points should be used as guidelines, supplemented by your own assessment of breaking market news ... and risk tolerance. Stops are 'optional', and this means that there is no need to set them so 'tight' that the probability of 'stopping out' becomes unacceptably high. It is generally better to cut the stop some 'slack' (as much as 0.20% or so) ... or simply set the stop at about 0.72% -- which is currently the maximum trigger point observed over the past 4.5 years.

*** New (3/9/07): A new Report (#13) has been added to our daily set of reports. The PORTSTAT Portfolio Performance Report (portperf.htm) shows the back-tested performance statistics for all 50 portfolios, quarter-by-quarter, from 8/16/02 to the current day.

*** New (3/8/07): Both Type 1 and Type 2 'protective stops' were added to the Voted Signals Report for LONG trades. These work in a manner identical to the stops previously added for Short trades.

*** New (3/7/07): A second kind of shorting 'protective stop' has been added to the Voted Signals Report. The original 'stop' is now known as a Type 1 'stop', and it is triggered when our Preprocessor detects a high number of new Long positions (#L) for the next day. A large value of #L almost always signifies that the market will show considerable strength. The new Type 2 'stop' takes advantage of our QQQQ day-trading signal that is contained in Table II of the Prediction Table (PREDTABL). If the QQQQ day-trading signal is sufficiently strong, then a protective 'stop' is important to protect against a 'runaway rally'. Whether the 'stop' is a Type 1 or a Type 2, the treatment is the same: if it is desired to set these 'optional' stops then the trigger price will be the Open price plus 0.50%. When the stop is triggered (covering the QQQQ 'short'), then the short should be reestablished near the market Close (just before the Close – or in the after-hours session) so that the 'short' is held overnight. It is important to remember 2 things: (1) these stops are 'optional' (although highly useful), and (2) use of these 'stops' is NOT reflected in the QQQQ signal performance statistics shown in the Voted Signals Report (voteqqqq.htm).

*** New (3/7/07): Two new subsections have been added to the top of Section 2 in the SIGNALS report: (1) a listing of the current members of the 100% Club, and (2) a listing of the current members of the 95+% Club. Remember that equities may shift into or out of these lists as a result of the daily model optimization process.

*** New (3/5/07): Logic has been added to the voteqqqq.htm report so that protective 'stops' will occasionally be signaled during SELL/S (shorting) periods. The 'stop' warning will be generated if the Preprocessor has detected 21 or more new Long positions (#L) for the next market day, and the stop will be set at the Open price + 0.50%. If the 'stop' is triggered, causing the 'short' to be covered, then QQQQ should be re-shorted prior to the Close. None of the statistics in the voteqqqq.htm report will reflect the usage of this 'stop'; the purpose is simply to help the trader reduce drawdown on days when the market might be expected to surge. Since the 'short' will be re-established by the Close, then the 'trade' will technically still be in force ... and thus the trading statistics don't need to reflect these extra (optional) day trades.

*** New (3/2/07): Table I of the Prediction Table Report (PREDTABL) has now been renamed to Table I-A. A new Table (I-B), has been added which uses the MTI value, the MCS (Market Close Signature), and information about whether the ETF has risen (or dropped) during the day, in order to provide the probability that the ETF will Open higher the next day than the current day's Close. Table I-B provides probability estimates for QQQQ, SMH, and SPY. The objective of this Table is to provide guidance as to whether a trade should be held overnight, or whether it should be sold just prior to the Close (when the MCS pattern is known) ... or in the after-market hours.

*** New (2/27/07): We have added some additional statistics to the voteqqqq.htm report: (1) Max (mid-trade) Loss for Longs and Shorts (the maximum amount of starting equity lost 'while' a Trade was still in progress, and (2) Max (end-trade) Loss (the maximum amount of starting equity lost on an unprofitable trade). Remember, however, that the voteqqqq.htm report represents 'back-tested' data, although the information for 2006-7 is extremely close to our actual data (and performance since 9/06 has been TimerTracked).

*** New (2/26/07): We have modified the voteqqqq.htm report slightly: (1) A Total Equity column has been added to show the accumulated equity day-by-day since 8/16/02 (by back-testing), (2) the CASH signal has been changed to read SELL/$, and the SHORT signal has been changed to read SELL/S (both signals are, after all, 'sells'), and (3) horizontal demarcation lines are used to indicate all changes of signal.

*** New (2/26/07): We have revamped our Daily Forecast (forecast.htm and this 'email') to make our overall market assessment more concise; there is now a Table at the beginning of our Tactical Plans section (the TACTICAL SUMMARY) that consolidates ALL of our commentary in one place. We are open to suggestions for further improvements. Please let us know what you think.

*** New (2/22/07): Another major change was made to the new VOTEQQQQ report; this was the addition of a 'simulator' that shows the cumulative gain, and the gain for each individual Long/Short period, resulting from back-tested trades over the past 4.5 years. This simulator provided the mechanism for finalizing the 'conflict strategy', that is, how best to resolve those cases in which the Long-optimized QQQQ signal (GR-L) disagreed with the new Short-optimized version (GR-S). It turned out that the original concept worked quite well: when the signals disagree, set the signal to CASH. We still plan on looking at the possibility of setting some protective 'stops' so as to minimize the risk of being 'short', for example, on a Fed Day. The simulator will make it easy to evaluate possible strategies, and to check their performance over the entire back-testing period.

*** New (2/21/07): We have made some additional improvements to the new voteqqqq.htm report, e.g., adding the $-OPN field which shows the Open price of QQQQ for the subsequent market day. This is the 'action price' for the transaction since we operate on the principle that all signals are acted upon at the market Open. We are still finalizing our strategy for the best way to 'vote' when the GR-L and GR-S signals conflict, but since that is not currently an issue we have some time for a more in-depth study of these situations. This is the first time that we have had such an accurate 'shorting' model, and it will take some time to fully integrate it into our System ... and our thinking.

*** New (2/20/07): The voteqqqq.htm Report has been enhanced with two new signal columns: GR-S and VOTE. We have been working on this enhancement for quite some time, and the idea is as follows:

(1) The traditional Grail System signal for QQQQ (henceforth also known as GR-L) has always been optimized for QQQQ Longs, with a lesser focus on the 'shorting' accuracy of the signal (SELL intervals). As a result, the back-tested accuracy of the QQQQ BUY signal has always been relatively high (ranging from 91% to 97%, and currently at 97%). The 'shorting' accuracy, however, has been lower ... about 81% of late (for a 4.5 year back-testing interval).

(2) We have added a second Grail System signal for QQQQ (henceforth known as GR-S), and which is optimized for 'shorting' accuracy rather than Long accuracy. This new signal has back-tested Long accuracies of a more modest 86%, but has shorting accuracies of about 94%. This signal will only show up in this email (Daily Forecast), and in the voteqqqq.htm file; its sole purpose is to give us more accuracy when QQQQ is very weak.

(3) The voteqqqq.htm report now shows the following signals:

(a) 10 'Raw' QQQQ signals (one for each of the 10 smallest Preprocessor portfolios)

(b) A Consensus signal (CONS) that represents the 'weighted' average of these 10 fundamental signals.

(c) The official Grail System signal for QQQQ (see GR-L in (1)), that is, our normal swing-trading QQQQ signal that is shown in our SIGNALS report.

(d) The new shorting-optimized Grail System signal for QQQQ (see GR-S in (2)), that is, a 'variation' on our normal model that is tuned to provide higher accuracies for QQQQ SELL intervals.

(e) A 'voted' signal (VOTE) that uses the GR-L and GR-S information to produce a new triple-valued signal (which we will henceforth use as our standard TimerTrac signal). By comparing the GR-L signal (our traditional Grail System signal for QQQQ) with the new GR-S signal we can define 3 different states for QQQQ: BUY, CASH, and SHORT. If both GR-L and GR-S are 'buys' then the voted signal will be a BUY. If both GR-L and GR-S are 'sells', then the voted signal will be a SHORT. In the 'mixed' cases (e.g., GR-L = BUY and GR-S = SELL) we will call for a CASH holding. We will probably do some additional fine-tuning of the strategy for these mixed cases, but for now the voteqqqq.htm report presents our 'best' assessment of the current strength of the Nasdaq 100.

*** New (2/16/07). We have added a new Table (B1-2) to Section B1 (Primary Indicators). This Table provides detailed descriptions of all of the specialized indicators that are shown in Table B1-1, and provides references to the various Reports that show historical values, or supply additional detail.

*** New (2/16/07). We have updated Part 1 of our Tutorial, and we have created Part 2 by incorporating most of the explanatory material previously seen in our weekly TimerTrac broadcasts. Links to the Tutorials are contained in Section A.2. We will now be upgrading both Tutorials on a weekly basis.

*** New (2/16/07). We have just completed a thorough retuning and rebalancing of all of our PORTSTAT portfolios. Gains have now been maximized, and drawdowns minimized, for all 50 portfolios. There is still more to do, however, in optimizing the performance of the 'hedging' strategy used in our “H” portfolios (QQQQ shorting).

*** New (2/16/07). We have begun a process of slowly adding new Closed End Funds (foreign equities) to our System. 62 new funds are now being utilized to improve the accuracy of the ETF Decompositions used to generate Scores and Recommendations in our Mutual Fund Report (MUTFUNDS), and some of the better funds will be added day-by-day to our Grail System – and will thus show up in our SIGNALS report, and in our “E” and “U” (PORTSTAT) portfolios. This process must be done slowly so that we do not destabilize our MTI calculation. The TTF (Thai Fund) and LDF (Latin American Discovery Fund) are included in this new batch of equities.

*** New (2/17/07). We have added a new report, voteqqqq.htm, which shows the current and historical values for our 10 'raw' QQQQ signals. This report shows the values of two of our primary indicators: L/1 and the # of 'raw' QQQQ signals in the Buy ('L') state.

*** New (2/13/07): (PORTSTAT) The shorting algorithm for the "H" (Hedging) Portfolios has been streamlined and made more consistent across the 12 portfolios. There is a bit more to do in order to fine-tune the historical performance of these portfolios, but at least their 'hedging' behavior is now more explicable.

*** New (2/8/07). The overall accuracy of the "Decompositions" used to estimate the current holdings of a mutual fund have been increased by a simple expedient: If the R**2 (correlation coefficient) for the decomposition into a weighted linear sum of 5 ETFs is too low, then the calculation is redone using all of the Grail System signals (1,000 stocks and ETFs) in addition to the original set of ETFs. This means that 5 'basis' equities are used that are drawn from a set of over 1,000 equities ... rather than the 175 standard 'basis' ETFs. Now, this doesn't necessarily mean that the decomposition is more 'correct', even though its R**2 (correlation coefficient) is substantially greater, but at least it gives us a reasonable basis for assigning a current SCORE (ranking) for the mutual fund. Since we utilize a combination of SCORE plus 3-day percentage gain/loss when we perform mutual fund portfolio rebalancing, it doesn't matter as much if we are absolutely correct or not ... that is, since we are always choosing a set of funds from amongst the recent top performers, whether or not the SCORE is really correct or not becomes less significant. In any event, the uncertainty in the SCORE value will generally be small enough that the bottom-line Recommendation (Buy/Sell/Hold) is unlikely to shift.

*** New (2/7/07). The SCORE values (and consequently the REC=Recommendation Field) are now based not solely on our intermediate-term signals (Star System), but also on our Grail System signals -- provided that the component stock/ETF in the decomposition is actually modeled within our Grail (short-term) System. This change makes our SCORE values more accurate in the short-term, and thus more reactive to current market conditions.

*** New (2/6/07). In Section 1 for each fund family we now show the Inception Date (calculated by looking at the data) for each fund IF the fund was started after 8/16/02. The Inception Date (IN:mm/dd/yy) is not shown if it was earlier than 8/16/02. More importantly, the percentage gains for all funds are now normalized by calculating an "annualized" percentage gain. This means that funds that have been recently created are directly comparable with the older funds.

*** New (2/6/07). In the 3-day performance section (Section 2) for each fund family, those funds that are currently in the SELL state (S or S?) are prefixed with '***' to emphasize that the fund should probably be swapped out for a better fund (one in the HOLD (H) or BUY (B) state). Remember that if you cannot find funds in the HOLD (H) or BUY (B) state, then you should probably be in cash (money market fund).

*** New (2/6/07). A new fund 'family', MISC01, has been added to our report so that specific fund needs of our subscribers can be satisfied. We are planning on developing and distributing a new program, however, that users will be able to run on their own PCs, and that will allow them to generate 'subsets' of the MUTFUNDS report that consist exclusively of the funds that they are personally interested in.

*** New (2/5/07): We have completed an exhaustive analysis of the unprofitable 'sell' period which our QQQQ model had previously flagged for 9/21/06, 9/22/06, and 9/25/06. This was the only 'sell' interval signaled during the entire period from 9/11/06 until the present, and since this was (in retrospect) an undesirable signal, we followed our standard procedure of determining what (if anything) our model might have done differently (based upon the information available to it) so that the signal would have stayed 'Long' for those 3 days. We have now identified some criteria that would have averted this 'sell' signal, and we have modified our algorithm accordingly. This is not, of course, historical 'revisionism'; rather, the objective is to try to identify similar combinations of signals in the future that might allow us to generate a more accurate signal. We use the following criteria when modifying our model:

(1) Any prospective changes must be functions solely of the signals and indicators that were available at the time (e.g., MTI, TOT, LSI, SSI, #L, #S, NN1, NN3, NN8, CNNI, TSI, etc.).

(2) The changes must be “logical” and reasonable, that is, capable of being extended to future situations that have a similar setup.

(3) The changes must improve the back-tested accuracy (or drawdown) over our entire back-testing period (currently 4.5 years).

(4) The changes should not excessively perturb the back-tested signal history (minor signal changes are acceptable, but the overall pattern of Longs and Shorts must be maintained).

The end result of this change can be seen in the MKTSTATS report; the entire interval from 9/11/06 to 2/5/07 now shows a continuous QQQQ 'buy' ... and the 3-day 'sell' is now absent. The MKTSTATS report is a 'back-tested' record –- rather than a historical record -- but it does effectively show how smooth our QQQQ timing indicator can be in actual practice; our QQQQ signal is a true swing-trading signal that accurately tunes in to the underlying dynamics of the market, and does not flip-flop back and forth in response to superficial variations. For our 4.5 year back-testing period our QQQQ model has a Long trading accuracy of 96.97% for 34 Long intervals, an annualized Long gain of 50.22%, and an Average Maximum Drawdown of -1.73%. In view of the recent extended Long period (with its moderate ups and downs) the natural question arises as to whether our model should have done better ... that is, detected more of these 2-3% dips. The answer is NO. Our thermodynamic model has functioned optimally, and has thus stayed Long for almost all of the recent market uptrend; only our day-trading model (see PREDTABL) has the potential to detect these very short dips, and even this method boasts an accuracy level of only about 70% for typical market days. As far as we can tell, our models continue to show exceptional accuracy.

*** New (1/29/07). A new Report, the Differential Signals (Back-Tested History) Report, has been added to our daily Report file. This report shows the back-tested (and current) differential signals for $NQ-RU, $NQ-SP, and $NQ-DW (showing the relative strength of the 4 indices), and this may be useful as guidance for trading in options and futures, or in setting up pairs of hedged trades using ETFs (e.g., a QQQQ Long with a DIA Short), or as further corroboration of our overall market status assessment. This report comes in 3 formats: the TXT version (DIFFHIST.TXT) is in our traditional report format, while the html version (diffhist.htm) has more eye appeal, and finally, the CSV format (DIFFHIST.CSV) is suitable for direct loading into a spreadsheet program (e.g., Excel) for further analysis. We are especially interested in feedback as to how the format of the CSV file could be made more useful.

*** New (1/25/07). The format for our 3 'Differential' Signals has now reached its final configuration. It now consists of a pair of positive or negative signs, e.g., (+/+), (-/-), (+/-), (-/+), depending upon whether the 1^st equity (always QQQQ, representing the Nasdaq 100) should be held Long (+), or Shorted (-), and whether the 2^nd equity (IWM [Russell 2000], SPY [S&P 500], DIA [the DOW]) should be held Long (+), or Shorted (-). The objective is to try to identify those time periods in which 'hedged' trades (one Long and one Short) are likely to be profitable, and the remaining times when it is best to go fully Long (+/+) or fully Short (-/-). Although this is the final format for the Differential Signals, this change has not yet percolated throughout all of the impacted reports (e.g., MKTSTATS and SIGNALS).

*** New (1/23/07). The signals for our 3 'Differential' Signals have been reformatted so that they now consist of two sections separated by a colon(:). The first section shows the short-term Grail System 'raw' signal, e.g., a 'BUY' is interpreted as the combination L/S, while a 'SELL' is represented by 'S/L'. The second section takes account of the PPF value and it either 'confirms' the L/S with a 'l/s' notation, or expresses a lack of information with a '?/?' notation. The same holds true for the 'S/L' signal, which would be expressed as S/L:s/l or as S/L:?/? depending on the much more sensitive PPF value:

L/S:?/? means that no hedging combination appears feasible at the moment although the 'trend' is for a QQQQ Long with the second equity shorted.

S/L:?/? means that no hedging combination appears feasible at the moment although the 'trend' is for a QQQQ Short with the second equity held Long.

L/S:l/s means that QQQQ would be Long and the second equity Short

S/L:s/l means that QQQQ would be Short and the second equity Long

*** New (1/22/07). The PORTCORR report program was modified so that it avoids processing any of the ETF-only portfolios. The feeling here is that ETFs are sufficiently robust to hold up reasonably well in a 'dip', and in any event it is not feasible to try to find 4 equivalent 'stocks' that can represent a family of ETFs. Also, portfolios with fewer than 6 active holdings are not processed since it isn't reasonable to try to 'hedge', say, a 5-stock portfolio by shorting 4 other stocks.

*** New (1/19/07). 3 new 'Differential Signals' have been added to this report: $NQ-DW, $NQ-RU, and $NQ-SP.  These are 'hedging' signals that try to identify market periods in which there are divergences between the Nasdaq 100 (QQQQ), the Dow (DIA), the S&P 500 (SPY), and the Russell 2000 (IWM).  The signal 'yes' is used to indicate that QQQQ might be held "long", while DIA, IWM, or SPY are simultaneously 'shorted'.  In other words, if the symbol $NQ-RU is a 'yes', then it might be profitable to hold QQQQ Long while shorting IWM as a hedge. These signals are still EXPERIMENTAL, and we are still looking at the even more difficult issue of the inverse situations, e.g., QQQQ Short/IWM Long.

*** New (1/17/07): We have made still more formatting changes to the daily FORECAST report (this email). This includes adding shortcut links at the top of the document which bypass the introductory (background) material and jump directly to the Signals, Indicators, Tactical Plan, and Prediction Table information.

*** New (1/16/07): We have extensively modified the format of the daily FORECAST report (this email) in order to improve its readability. This is an ongoing process, and there are a few more changes in the works over the next few days.

*** New (1/16/07): We have completed our portfolio optimization procedure, and the 50 PORTSTAT portfolios now vary with portfolio size in a more uniform manner. Also, we have made an important enhancement to our Prospector(TM) tool that will further help us in maintaining the best 1000 equities (i.e., the best 'models') in our Grail System.

*** New (1/11/07): We are in the process of optimizing the portfolio generation algorithms used to build the 50 PORTSTAT portfolios. When this process is completed there should be a more uniform variation in portfolio parameters as one moves upward (to larger portfolios) or downward (to smaller portfolios) within each of the 5 portfolio classes. In addition, the tendency of the “H” portfolios to do an excessive amount of trading will be substantially curbed. These “H” portfolios will 'always' trade more than the “N”, “C”, “E”, or “U” portfolios, but we feel that the current level is a bit excessive.

*** New (1/10/07): An enhancement has been made to the PORTSTAT "C" portfolios (Stocks - Conservative) and "E" portfolios (ETFs) to force more careful trading when the MTI level dips below approximately 1.45. Instead of selecting equities strictly on the basis of the Grail System signal, at these low MTI values the portfolio logic now also requires that any newly added equities also be current Long holdings of the Preprocessor. In other words, when it is a bit dicey to add any more "long" positions, the logic demands that a new equity have shown a respectable long-term trend as evidenced by the current Preprocessor holdings.

*** New (1/9/07): We have added a new indicator to our system: “The 95+ Club”. To compute this indicator we look at all of the 1000 Grail System short-term (S/T) timing models, select those with back-tested accuracies of 95.0% or greater (over a 4.5 year period), and then calculate the percentage of those that are in the 'buy' state. The idea is to use the most accurate models as a finer measurement of overall market strength than simply using the total percentage of models that are in the 'buy' state.

*** New (1/8/07):  We have now added 5 new ultra-conservative ETF-only portfolios to our PORTSTAT report.  This now gives us 50 portfolios that are subdivided into 5 different classes that progressively become 'safer': "H", "N", "C", "E" and "U".  The new portfolios (P6U-P15U) will hold from 6 to 15 ETFs, and they are restricted to those ETFs with significant volume (> $ 5 million in value traded per day), and with back-tested trading accuracies of 96% or greater, and back-tested drawdowns of -3.00% or less (over 4.5 years).  Even these 'play it safe' ETF-only portfolios can result in significant gains, however, despite the fact that they spend protracted periods in cash.  Take the P10U (a maximum of 10 ETFs). for example.  This is arguably one of the safest, and least greedy, of our portfolios. The back-tested results are as follows -- assuming that $100,000 was invested on 8/16/02.

On 12/31/02, value = $ 125,511.  Gain from 8/16/02-12/31/02 = 25.5%
On 12/31/03, value = $ 218,074.  Gain for 2003 =  73.7%
On 12/31/04, value = $ 295,082.  Gain for 2004 = 35.3%
On 12/31/05, value = $ 349,023,  Gain for 2005 = 18.2%
On 12/29/06, value = $ 403,096.  Gain for 2006 = 15.5%

*** Remember that these are 'back-tested' gains, and may not be indicative of future performance.  The maximum back-tested drawdown was just under -4.00%.

*** New (12/27/06): We have added one more ETF-only portfolio to our PORTSTAT report: P15E will hold a maximum of 15 ETFs, and can be expected to average about 1 buy or sell per day.

*** New (12/26/06): We are continuing to increase the accuracy of our 1,000 Grail System stock/ETF models, and thereby the accuracy of our 44 PORTSTAT portfolios. We will discuss this activity in more detail in a few days, and we will address the following topics:

1. Speeding up our software: Following data downloads (stocks and mutual funds data), stock renames, and split adjustments, it has been taking almost 6 hours to run our complete software suite. We have devoted considerable effort to making our software run faster – not only to permit preparation of our reports in a more timely manner, but also to allow us to more fully re-tune our models and portfolios after the inclusion of yet one more day's worth of market data.

2. Optimizing our choice of stocks and ETFs to model: Our Grail System consists of about 100 ETF models and 900 stock models, and these are now chosen in a 2-step process. First, our Preprocessor stage winnows our initial universe of about 9,000 stocks and ETFs into a 'selected' set of 1,600 equities. This selected set of equities is used by the Preprocessor to derive its portfolios (see NEWPICKS), generate the MTI (Market Timing/Temperature Indicator) value), and ultimately to produce intermediate-term (I/T) timing models. In the second stage, we utilize our new Prospector(TM) software to extract the best 1,000 equities for our Grail System short-term (S/T) modeling process. The objective is to maintain about 900 stock models (and 100 ETF models), and to periodically make minor additions and deletions so as to retain only stocks with back-tested modeling accuracies of 80% or greater (and with acceptable annualized gains and Average Maximum Drawdowns). Since we have plenty of stocks to choose from, there is no point in modeling any stocks that do not meet a set of high performance and accuracy standards.

*** New (12/18/06): About 60 ETFs have been added to our system, bringing the total to well over 100 in our Grail System. In order to take advantage of this, four new ETF-only portfolios have been added to PORTSTAT (P9E-P12E), now giving us a total of 44 portfolios based on our Grail System signals.

*** New (12/11/06). Modeling Accuracy. We have increased the fidelity of some of our Grail System signals by capitalizing on the extremely high back-tested accuracy of the IJS model. Although our principal model is the one for QQQQ, and this forms the basis for the preponderance of our Grail System stock/ETF models, it turns out that there are a number of stocks which can benefit from utilizing the trade signals generated by the IJS model (S&P 600 smallcap value) – as 'adjustments' to the underlying QQQQ signals. The average back-tested Long accuracy of our 1000 Grail System stock/ETF models is now 82.51% (over the past 4.5 years). During this period QQQQ has a Long accuracy (back-tested) of 97.14%, while IJS is our only model which boasts a (back-tested) 100% track record for 41 Long trades.

*** New (12/11/06). Bond ETFs -- we have added four bond ETFs to our Grail System, and their buy/sell signals can be seen in the SIGNALS report. These ETFs are TLT, LQD, IEF, and SHY. Note that these buy/sell signals are for informational purposes only; such ETFs may be fine for diversifying long-term portfolios, but offer insufficient annualized gain potential to be useful in our style of trading. We will probably add a few more of these in the near future. These ETFs were added to satisfy subscriber requests.

*** New (12/8/06). We have added a suite of 8 ETF-only portfolios to the PORTSTAT Report. These portfolios, named P1E-P8E, will handle a maximum of 1 ETF (P1E) up to a maximum of 8 ETFs (P8E). Since these portfolios follow our short-term (S/T) Grail System signals, they represent a logical step toward diversification over simply holding QQQQ Long. These portfolios do not 'short' – rather, they reduce their risk by dropping positions as the market weakens ... and refusing to add new positions unless the MTI level is > 1.45. The trading frequency of these new portfolios is, naturally, very low.

*** New (12/8/06). We have renamed the 'non-H' (normal) portfolios of PORTSTAT so that they use the suffix “N”. The portfolio names are now P1N, P2N, etc. This gives us the following 4 classes of portfolios, and in increasing order of safety and conservativeness (and steadily diminishing trading frequency) these are: “H”, “N”, “C”, and “E”.

*** New (12/8/06). We have modified the 'weighting factors' that we employ with the QQQQ day-trading 'Poker strategy'. Since the preponderance of our signals are relatively weak (?+,?-,*,S), we need to boost the allocations somewhat so that we are using 100% of available trading equity more frequently. We will thus use a 50% allocation for the '?+' and '?-' signals, a 100% allocation for the '*' and 'S' signals, and allocations in excess of 100% (i.e., using 'margin') for the much stronger **,***,****,SS,SSS,SSSS signals.

*** New (12/7/06). We have now switched from Microsoft Word (for our html editing) to the Star Office Suite by Sun Microsystems. This cures the html formatting problems that we encountered when we switched from Microsoft Outlook to Mozilla's Thunderbird. We are thus now using Firefox 2.0, Thunderbird 1.5, and Star Office as our primary tool suite.

*** New (12/5/06). Two new “C” (conservative) portfolios have been added to PORTSTAT. P20C is designed for a maximum of 20 stocks, while the P25C is designed for a maximum of 25 stocks.

*** New (11/14/06).  We have re-added two Reports to our daily report set.  The first, SECTION2.TXT, is a report that summarizes all of our intermediate-term (I/T) signals.  These signals have hitherto only been found in Section 2 of the Market Status Report (MKTSTATS).  Also, the reports VARPORTA.TXT, VARPORTB.TXT, VARPORTC.TXT, VARPORTD.TXT, VARPORTE.TXT, VARPORTF.TXT and VARPORTG.TXT contain intermediate-term portfolios constructed from these I/T signals.  The maximal sizes of these portfolios ranges from 6 stocks (VARPORTG) to 20 stocks (VARPORTA).  It is recommended that anyone interested in these portfolios paper-trade them for a while to gain familiarity with their performance, and their drawdowns.

*** New (11/14/06).  Two more VARPORT portfolios were added, bringing the total up to 7.  These intermediate-term portfolios range from 6 to 20 stocks in size.

Recent Changes to the MUTFUNDS (mutual fund) Report:

*** New (11/10/06). Many new funds have been added for AllianceBernstein and AIM Funds.

*** New (11/10/06). The Recommendation (REC) field for Short/Bear/Ursa/Inverse funds is now forced to a 'B' (Buy) when the QQQQ Grail System signal is a 'SELL',and to an 'S' (Sell) when the QQQQ Grail System signal is a 'buy'. This is currently our 'best guess' as to the appropriate time to purchase shorting funds.  Note that this is PROBABLY not the best signal to employ for non-tech Bear funds, e.g., funds that short GOLD, OIL, REAL ESTATE, or Japan. This is currently an area that we are trying to improve.

*** New (11/13/06). Shorting funds are now tracked according to the status of the QQQQ Grail System timing signal.  In other words, the performance (%gain) shown in Section 1 for each Fund Family assumes that the shorting (BEAR) fund was bought when the QQQQ signal was a 'sell', and sold when the QQQQ signal went back to a 'buy'.  Remember, however, that many shorting funds have rather recent inception dates, and so their Section 1 performance will not be indicative of their long-term performance.  All Shorting (BEAR) funds are now clearly marked with the tag '**BEAR**'.

*** New (11/13/06).  We have now reached a total of almost 1,100 funds.

*** New (11/3/06).  The current value of the Mutual Fund Timing Indicator (MFTI) is now shown at the beginning of the Report.  This indicator has the states: 'buy', 'hold', and 'SELL', and for each state the report gives a recommendation about actions that should be taken for mutual fund portfolios.

*** New (11/6/06).  SunAmerica Funds have been added.

*** New (11/7/06).  Many new funds have been added for GE Funds, Delaware, Vanguard, Fidelity, Oppenheimer,  BlackRock, Rydex and Profunds.

*** New (11/7/06).  The REC field is now shown for all "long" funds, but a trailing question mark (?) is applied if the correlation coefficient (R**2) is less than 0.9800.

*** New (11/8/06).  More Putnam Funds have been added.

*** New (11/9/06).  Many new funds have been added for American Century, Morgan Stanley, Janus, Dreyfus, T. Rowe Price, USAA, and Van Kampen. We have now reached a total of nearly 1000 mutual funds.

*** New (11/9/06).  Special error-checking logic has been added to detect mutual fund data errors and dividends.  Dividends are now fully accounted for in the historical data used to perform ETF 'decompositions'.

Recent Changes to the MUTFUNDS (mutual fund) Report:

*** New (11/2/06).  American Funds have been added.

*** New (11/3/06).  ING Funds have been added.

*** New (11/6/06).  SunAmerica Funds have been added.  Also, many more DELAWARE and GE Funds have been added.

*** New (11/3/06).  Section 3 has been eliminated.  Mutual fund recommendations are now contained strictly within Section 2, which shows the performance of each mutual fund (%gain) over the past 3 trading days.  The REC field has been added next to the SCORE field, and it will contain a S, H, B, or ? code depending on whether the  mutual fun