Each month since  October, 2007 I have posted predictions for the U.S. stock market for the coming six months.  That’s 9 years of forecasts, so I think it’s about time for a reckoning. Are my forecasts for U.S. stock market gains worth anything? Or are my predictions just plain bogus?

The excellent news is that my break even models have performed quite well .  The models are just as valid now as when I started using them in 2007.  The models are far from perfect, but they generally foretell major stock market moves.

This blog post focuses on my forecasts of the probability that the U.S. stock market will at least break even over the next 6 months.  In another post I will review the separate models I use to estimate how large gains or losses will be.

Were Break Even Forecasts Basically Useful?

The chart below shows the growth of the Value Line Arithmetic Average (VALUA) from 1984 through 2016 Q1. This index tracks the roughly 1700 large U.S. companies that are followed by the highly respected Value Line Investment Survey.  Unlike measures such as the S&P 500 Index, or the NASDAQ Composite Index, VALUA gives equal weight to each stock. No stock is more important to the index that any other. The distinction is important — VALUA ends up being much less volatile and speculative than the other market indexes.That makes VALUA much more predictable than other indexes.

For the entire period  from 1984 through 2016 Q1 VALUA has risen  during 73% of all the 6-month periods, making 0.73 the average probability of VALUA rising in any half year period.

I have superimposed a series of green dots to show the dates when my forecasting model concluded optimistically that the probability of the market going up was 90% or better.  Likewise a series of red dots mark where my forecasts of the market at least breaking even were 50% or less. (The rest of the time the forecasts were middling — neither very optimistic or pessimistic.)

Overall, the pessimistic forecasts (red) do tend to come before significant market corrections, rather than coming after the market has already collapsed.  For the forecasts made in real time from 2007 on, all of the pessimistic forecasts were valid.  All but one were timed exceedingly well. The remaining recent pessimistic forecast did foretell a market break, but the break came months later than expected.

Optimistic forecasts (90% or better chance of a market rise)  were much more common in both  real time and historical data.  Generally, the optimistic forecast points do appear to lead favorable market periods. It is clear, however, that optimistic forecasts are frequently premature — even though the market future has improved, the stock market may still be in the process of crashing.

What About Break Even Forecasts That Were Terribly Wrong?

(Click on image to enlarge.)

Fortunately, there were relatively few points when the probability of gain forecasts were really bad. Eyeballing the chart above, it is reasonable to conclude that every few years, the model makes a really bad market call.  In most cases, even though the market rose after a very pessimistic forecast, it appears that the pessimistic forecast was just issued too early — the market did indeed tumble, but months after the model expected the market to break.

To my surprise there were no instances of really bad forecasts during my real-time testing from 2007 on.  I have no doubt that at some times in the future the model will again make some really bad market calls.  The accumulated experience, however, encourages me to think that really poor stock market calls will not happen more frequently than every few years.
basic

Has There Been Any Basic Change in the Factors that Move the Stock Market?

The general goal of modelling the stock market is to find consistent relationships for a small number of factors that reasonable might be expected to affect the market going forward.  The stock market responds almost instantly to countless sorts of news but no model can be reasonably be developed that will reflect all of those factors.  So, the real problem is finding just a small number of important factors that consistently affect the general direction of the market going forward.

In 2007 the stock market forecasting models I came up with tracked the past performance of the market pretty well.  The question now is whether the model needs to be changed to reflect new sorts of factors that may have developed over the past decade?  Maybe the basic nature of the stock market has changed enough so that an entirely new sort of model needs to be used going forward? Maybe some of the old factors that seemed to be relevant really aren’t?

Good news.  The same factors that seemed important based on data from 1984 to 2007 appear to be behaving in almost exactly the same way since then.  There really hasn’t been a large shift in the basic factors that affect stock prices.

The plot below compares the forecasts for the probability of stock market gains based on my old 2007 model and based on the same model that has been re-balanced to reflect all the new data since 2007.  The changes to the model that should be made are very minor:  the new forecasts for probability of breaking even match the old forecasts 99.8%. There is almost no change.

Six months from now the U.S. stock market should wind up right where it is today, at least that is what my forecasting models expect. No net gain for he next six months and only about a 50/50 chance that stocks do better than break even. In between, the models have been expecting the market to stumble with a drop of perhaps 5%.

What’s up?  The market has been hitting new highs so there is relatively little likelihood that stocks will have any quick bounces up, rather a temporary drop becomes probable. Also, the market in September and October historically tends to under-perform.  Other than that, most economic indicators don’t point to any dramatic shifts coming.  The net result is that if past economic performance is any guide, the market shouldn’t be going anywhere in the next half year.

My six month stock market forecasting models expect zero gains over the next half year with about normal odds of at least breaking even.

The stock market rebound from the ‘Brexit’ dip has been spectacular. However, no two ways about it, my short term trading model was completely wrong, having turned negative last month. I missed the rebound entirely.

My 6 month forecast made at the start of February did better — it had projected very good 9% gains to come over the period. That is closer to what actually transpired.

Well, what can I say?  My models don’t know anything about politics or human emotions. They only consider basic economic variables that have been shown over the past half century to generally lead the stock market by several months.  What they say now is that some sort of short term market tumble of up to about 10% is looking pretty likely over the next several months. Doesn’t mean that will happen, just means that the models say it is likely.

My forecasting models expect some market weakness for the next few months, but project that the market will do somewhat better in the closing months of 2016.  Overall, the models forecast a gain of just 1.3 percent between now and the end of 2016.  The probability of at least breaking even over the coming half year is about 70%  — very close to the market’s long term average.

The short term market forecast remains negative.  The model turned negative at the start of June and remains mildly negative.  The model is not likely to issue a buy signal in August either.

Overall, the U.S. stock market has been behaving fairly close to the models’ expectations.  In January the model had projected a 7.3% market gain in te first half of 2016 and the actual market gain was exactly the same 7.3%.  In June the one-month trading model turned negative — ahead of the brief Brexit market tumble.

On the whole the market seems to be normal, reacting about the same as it usually does to the economic variables that typically foretell market direction. If that continues, we can expect a couple more weak months before prospects improve.

My near term stock market indicator has now turned red. My models do not predict a large stock market drop soon, but they do project that over the next few months the market is more likely to fall than to rise. This negative market outlook is likely to remain for at least the next several months.

This is the second month that my econometric models expect the U.S. stock market to drop over the next 6 months — by about 5%.  I calculate that there is about a 40% chance the market will break even in the next half year, versus a 74% chance on average.  For the past year or so the market has steadily performed worse than my forecasts so it wouldn’t surprise me if the actual results come in somewhat worse than my pessimistic forecast. Something outside the scope of my models is causing the market to perform poorly.

What’s behind this negative 6 month stock market projection? Typical seasonal market weakness is part of the reason: on average, but not always, the stock market performs worse during summer months than the rest of the year.  The bigger problem, however, is that as far as my model is concerned, there is little reason for the market to go up.  The market is near fair market pricing, leading economic indicators are weak, and interest rates appear more likely to go up a bit rather than falling.

As I noted above, the stock market has been performing somewhat worst than my models had expected.  Looking at the graph below, the market does appear to be following a normal seasonal pattern — the colder months of the year are generally more favorable to the stock market than warm months.  Overall, however, actual market changes over 6 month periods have been distinctly worse than the models had predicted.  Over the past 6 months, for example, the market rose about 2.5%, but my models had forecast a much stronger 9% rise.  As always, the market will do whatever it ‘wants’ to do, but my models are getting gloomy.

May 2016 through October 2016:  Watch out. The six month stock market forecast is bad and the near term forecast is flashing yellow.  Next month the short term forecast will almost certainly turn red. My forecasting models expect the U.S. stock market to fall significantly over the coming half year; and that doesn’t factor in the presidential election which is likely to create additional volatility.

The models predict that the stock market will decline about 6 percent from now through the end of October. The probability that the market will decline over the period is high.  The odds that the market will fall by at least 8% at some point over the next six months are high — about 1 in 4.

As shown in the graph below, the stock market has performed below the expectations of the model for more than a year. If that continues, actual market results in the next six months could be significantly worse than the bleak forecast now given by the model.

The immediate prospects for the stock market, however, are much less clear and are not as bad as the longer term forecast.  As shown below the short term forecasting model is now flashing a warning.  Based on the current negative 6 month forecast next month the short term forecasting model will probably be flashing a sell signal.

April 2016 through September 2016: My forecasting models expect positive, but below average market gains over the next 6 months. (roughly 3% gain, 70% chance for break even with 73% being the long term break even probability.)

Spring flowers are just starting to come up where I live, but my stock market forecasting models are already anticipating that stock market gains over summer months will be weak following a normal seasonal pattern. The model says that 3% projected gains through the end of September will be below long term stock market price appreciation.  My models know nothing about politics, but personally I will bet on some sort of market stumble as the Republican and Democratic presidential conventions get underway.

That said, it is still springtime so April, and probably May will likely produce better than average results. The second graph below shows that a time lagged forecast using my models remains bullish.

(Click on image to enlarge.)

(Click on image to enlarge.)

March 2016 through August 2016: Above average gains (roughly 8%, 85% chance for break even)

The stock market continues to lag the models’ forecasts, but the gap appears to be closing. The model is saying that once again the market has predicted a recession that didn’t happen. Current forecasts from the model remain optimistic — nothing beats a crazy stock downturn for making the market future look brighter.

February 2016:  U.S. stock market returns well above average
March 2016: U.S. stock market returns well above average
February 2016 through July 2016: Above average gains (roughly 9%, 85% chance for break even)


Last month the short term forecast from my forecasting model proved to be dead wrong. Looks like I have some explaining to do.

The prediction was strongly positive, but the U.S. stock market suffered it’s worst January performance EVER.  So, what does that mean?  Is the model junk? Or is something else going on?

My models focus on a small number of key measures of the economy that have proven over the past several decades to be leading indicators for the broad U.S. stock market.  Together these factors forecast about one-third of the actual variation in the market over 6 month periods. However, that also means that about 2/3 of 6-month market variations are not predicted by my models. The goal here is not to be accurate, but rather to foretell the most likely general direction of the broad stock market

If the model is to be believed, the U.S. stock market should be heading upward fairly soon.

The chart below shows a multi-year plot of the S&P 500 index with buy/hold recommendations from the model in green. The current forecast is positive.

The next chart displays actual 6-month changes in the stock market (blue) and predicted 6-month changes (red).  In the most recently completed periods, the market has performed significantly worse than expected. Note that in previous downturns the market has routinely performed worse than expected.

January 2016:  U.S. stock market returns well above average
February 2016: U.S. stock market returns well above average
January 2016 through June 2016: Above average gains (roughly 7%, 80% chance for break even)

I have more-than-normal confidence in this month’s forecast. My short term (1 to 2 month) stock market forecasting models have been anticipating the timing of market shifts pretty well for the past couple of years. The most recent call for a change in market direction (Buy) was made at the end of August.  The forecast was matched by a market turn around starting in September that now has produced a gain of roughly 7%.  

Looking forward, the short term models remain quite bullish for the first months of 2016. While my longer term models are already starting to worry a bit about market slow downs in May and June, the short term models are optimistic.

The buy/hold calls of the short term model (green) and the sell/avoid calls(red) for the past two years are shown in the chart below.  The model  turned green at the start of September and the market has risen about 7% since then. For January the model remains bullish. It suggests “Buy/Hold”.

Let’s shift to the longer 6 month forecast.  For the past year or so the stock market generally has been following the same course that my 6 month macro economic models had predicted six months months in advance.  On the whole, however, the stock market has come in somewhat below my econometric models’ expectations. Last July, for example, the model expected below average gains for the second half of 2015, but the actual result in the 6 months since then was significantly worse, a loss of about 7%.  I am not too concerned about this difference — as you can see in the chart below, it is more the rule rather than the exception that actual market performance is more extreme than my models had predicted.

Happy New Year!