May 17, 2004...Tinkering with the market data in our daily market summaries going back to July of 1993, we've formulated a few rules for predicting tomorrow's market...

• If the worst performing group in our "25-slice" daily tables loses less than .8%, tomorrow will likely be positive.

• If the last market session gained strongly (up 1% or more), the following day will likely be positive.

• If the market 3 or 4 sessions ago was particularly strong (up 1.8% or more), the next session will likely be strong.

• A strongly losing Friday or Monday is often followed by a weak Tuesday.

• If yesterday's market was weak (losing more than .5%) and the most recent session showed a large spread (5% or more) between the gains in our best and worst groups, tomorrow will likely be strong.

The differential between best and worst groups is an interesting indicator, we think.  It indicates whether stocks are rising and falling in lockstep, or if there are pockets of strength and weakness that seem to be disregarding the general market.

********************

We ran similar tests to what you see above, but this time we glommed all our half-monthly data tables into a single text file, and looked for trends with our software.  

Nothing of great significance emerged.  The general message seemed to be one of momentum, as opposed to bottom-fishing....if the previous half-month period gained, the odds increase (slightly) that the next half-month period will do the same.  In fact, the same logic applies when you look back two, as opposed to one, half-month periods...if stocks lost in the first half of month X, they tend to lose in the first half of month X+1.

Again, though, these results are not particularly significant...nothing breached p<.05.

***************************

Finally, we ran tests that are similar to what you see above, but on full-month scales.  The problem here is, we've only got about 130 rows of data...one for each month since 1993.  It's a bit difficult to extract anything of significance from small sets of data like this, especially when you're dealing with something as chaotic as the stock market.

Again, the levels of significance were not particularly impressive, but the general themes were:

• a high "Group Stdev" (above 30) is a good thing in our "25 slice" tables.  Similarly, a large differential between high and low groups is a positive sign.

• months with particularly large losses in the weakest group (more than 13%) tend to be followed by gains, but be careful...months with more moderate losses in the weakest group tend to be followed by more weakness.

• months with large gains in the general market (more than 8%) tend to be followed by more gains.

Our own thinking on the logic behind the tendency toward future gains when the difference between high and low groups is large is this:  markets where stocks move in lockstep are driven by fear.  Nobody seems willing to go out on a limb and bid a particular group up.  You might also say that particular economic or politic themes...interest rates, inflation, oil supply, terrorism...play an excessive role in investment decisions in these environments.

On the other hand, whether the general market is gaining or losing, when various groups don't seem to be paying attention to the more general trends, optimism is reigning.

Currently (5/2004), stocks are moving very much in lockstep...a negative sign.

***************************

The tables below involve predicting the performance of the tomorrow's Nasdaq National Market composite index.  We took several indices and used their one-day historical performances as a basis for prediction.  The indices were the FVX (a treasury bill index), the XCI (computers), XAU (gold), XII (institutional buying), VIX (volatility), IXI (industrial), and VLIU (utilities).  The Nasdaq National Composite itself goes by "IXQ".  The test goes back to October of 1990...3300 lines of data.

Perhaps we shouldn't be surprised that nothing amazingly significant emerged...it's in these domains that the "efficient market hypothesis" would function most strongly.  Nevertheless, there were a number of trends that beat throwing the dice by a decent margin.

A gain in the XCI index was the strongest indicator of a gain in the IXQ.  Breaking the data up into deciles of XCI gains caused a loss in significance.  There was about a 10% chance that the result was pure luck.  We're not hugely surprised that this indicator would serve as a decent indicator of tomorrow's performance...we've noted quite a few times that gains in volatile stocks tends to be a short-term bullish signal.

To put the results in a somewhat more digestible form, the Nasdaq Composite rose 59.7% of the time following an XCI gain.

If the IXI falls into the lowest 1/3 of % gains (historically, a loss between 10% and .03%), the market tends to fall the next day.

The VIX, a popular index for traders, did have a slight tendency to forecast up days when its daily change was negative.  Combining this negative reading with a gain in the XCI improved the gains further, without losing significance.  This dual strategy, however, was slightly outperformed by that of combining an XCI gain with an FVX loss.

A loss in the VIX predicted a Nasdaq gain 56.8% of the time.

To repeat:  nothing amazing emerged.  The most interesting results occurred with large "slices" of data (e.g. when the VIX was simply divided into two performance groups)...don't assume you could milk the market for greater gains by, say, buying stocks when the VIX falls into the lowest 10% decile group.

After making the tables below, we thought we'd throw some "streaks" into our data...using knowledge of how many days in a row the VIX, XCI, or IXQ itself have gained or lost to make predictions.  In the end, nothing of great interest emerged...the Nasdaq does have a very slight tendency to gain after winning sessions in its own index or in the XCI, but these sorts of results could be more the result of chance than anything else.

In fact, regardless of whether the XCI is coming off a 1,2,3,4,5, or 6 day winning streak, the odds for a gain in the Nasdaq are all about 60%.  Bear in mind that over the course of the test, the Nasdaq went up 54.6% (1836 out of 3362 days) of the time anyway.

Another test that we didn't bother to summarize in table form was to look at data since the year 2000, as opposed to 1990.  If anything, the results are murkier than the longer term-results.  The main observation:  the Nasdaq has had a decent tendency to bounce back following relatively long losing streaks (4 to 7 days) in its own index or in the XCI.

A final test (not shown) was to throw the Nasdaq high/close differential into the data mix.  We were a tad surprised to find that the Nasdaq has a small tendency to move upward following a session where it closed near its high...this is the opposite of what we've noticed with individual stocks.  The market-wide tendency reverses, however, if you combine a large high/close differential with a decent volume increase.

We threw out five tables with entirely boring results.  In statistics, this is known as "cheating", but it doesn't make much sense to post these tables just as "proof" that we're not data-snooping.  The "NUM" indicator in several of the tables is essentially meaningless...ignore it.  The gains and losses (red text), are in percentages.

Copyright © 2010 MarketSynopsis.com. All rights reserved.