After my last post on the relationship between 15-year normalized P/E ratios and long-term performance, I received several emails asking some interesting questions. So far, this one is my favorite:
I would be most interested in knowing if there is some predictive value in the normalized P/E ratio on the return in the third year of the election cycle.
What is the effects of P/E in years 2006, 2002, 1998, etc. on performance in 2007(?), 2003, 1999, etc.?
I am hoping that the third year effect is stronger than the P/E effect.
This is my favorite question, because I would never have considered it on my own. I'm usually focused on long-term performance. While I did include several short-term numbers in the data I used in this project, it never occurred to me to look at a question like this using the 15-year normalized price-to-earnings ratios I had calculated.
Anyway, it's an intriguing question, and it's one I can answer – or, at least investigate. I'm not sure there is an answer. But, there is relevant data, and I can share that with you here.
There are seventeen qualifying years in the set we want to study. The set includes every fourth year from 1938 to 2002.
Here, we are not looking at calendar year point changes in the Dow. For this project, I entered the Dow's average level for each year, because this was meant to be a study of long-term performance that individual investors could learn from.
I believe average prices are best suited for this task, because they better illustrate the results an investor could have achieved had he bought at some time during a particular year. So, please remember that the annual point changes in the Dow discussed here are based on comparisons of average levels in consecutive years – in fact, that's how all returns are measured in this study.
Now, on to the data…
Normalized P/E Effect
First, I needed to run a quick check to make sure there is still a measurable normalized P/E effect among this specific group of years – just as there had been in the full 1935-1990 set.
To look for this effect, I began by ordering the fourteen qualifying mid-term years by their 15-year normalized P/E ratios. This left me with two groups of seven years each, a low normalized P/E group and a high normalized P/E group.
Here is the average compound annual point growth in the Dow over the subsequent 15 years for each group:
Low Normalized P/E Group: 9.03%
High Normalized P/E Group: 6.35%
So, right away, we can see the election cycle does not completely obliterate the normalized P/E effect. It seems that years with low normalized price-to-earnings ratios tend to have better long-term performance than years with high normalized price-to-earnings ratios – even in the third year of the election cycle.
Of course, when people think about the third year of the election cycle, they are usually thinking about short-term performance rather than long-term performance. So, let's take a look at one-year point growth in the Dow (remember, these growth figures are based on yearly averages). This time, we have seventeen qualifying years, because we are only measuring performance one year out. Therefore, we can include every mid-term year from 1938 to 2002.
The one problem here is that seventeen is a prime number; so, I couldn't divide it into smaller groups consisting of an equal number of years. Therefore, I simply grouped the eight years with the highest normalized P/E ratios together and grouped the eight years with the lowest P/E ratios together – in other words, I ignored the median year. So, here is the average one-year point growth in the Dow for each group of eight years:
Low Normalized P/E Group: 17.41%
High Normalized P/E Group: 10.37%
I didn't expect that. As I'll explain in a later post, this whole project was based on simple logic. I started by imagining what normalization process should best predict long-term performance without actually looking at any data.
I chose a normalized P/E based on fifteen years of past earnings and measured compound point growth in the Dow over the subsequent fifteen years, because these numbers made sense to me. Basically, I thought if long-term performance was as closely tied to "earnings power" as I thought it ought to be, then fifteen years of past earnings data and fifteen years of future share price growth should be enough to detect this relationship.
I gave no thought to the possibility that any normalized P/E effect would be discernible over much shorter time periods. In this particular case, the one year performance numbers do seem to hint at some kind of normalized P/E effect even in the short-term.
Of course, I wasn't satisfied with merely comparing the average one-year performance posted by the two groups (17.41% vs. 10.37% in favor of the low normalized P/E group).
Remember, we're working with really small numbers here. Each group only had eight years in it. So, we need to look at the data from a few different angles and see if a clear and convincing pattern emerges.
To get a look from a slightly different angle, I decided that dividing the seventeen years into three groups (a lowest normalized P/E group, a highest normalized P/E group, and a middlemost normalized P/E group) was the best next step.
This might sound like a strange choice of action at first; however, I already had some familiarity with this data as part of the full study. I knew these normalized P/E numbers tended to "clump" a bit more than you might expect. When you consider how manic-depressive Mr. Market tends to be, this isn't really a surprising observation. But, it is still something we need to keep in mind.
Since there were seventeen years and I wanted exactly three groups, I had to discard the two years that separated our three groups of five years each. Again, this sounds strange, but when I go over the normalized P/E ranges for our three groups, you'll see why this approach makes a lot of sense.
Here is the average one-year point growth in the Dow for each group:
Lowest Normalized P/E Group: 15.34%
Middlemost Normalized P/E Group: 23.59%
Highest Normalized P/E Group: 6.57%
This time I wasn't surprised. One thing you'll notice about data based on these normalized price-to-earnings ratios (and you may have already caught on to this) is that the "low" and "average" years tend to have a lot more in common than the "high" years. Returns among the highest normalized P/E group can sometimes be downright ugly.
In this case, the returns aren't bad at all; but, remember we're dealing with a group of seventeen years that had unusually excellent returns – the entire group saw an average one-year point gain in the Dow of 14.99%. So, the 6.57% average for the highest normalized P/E group is very poor relative to the entire set.
Furthermore, of the seventeen qualifying mid-term years (1938 – 2002), only two years saw negative one-year point growth in the Dow. Interestingly, these two years (1946 and 2002) had two of the three highest normalized P/E years in the entire set.
However, the year that had the highest normalized P/E in the entire set worked out just fine. Most investors remember the period from 1998-1999 quite fondly. Eventually, that extremely high normalized P/E ratio did catch up to the market; but, not during the years we're looking at in this little election cycle study. In fact, 1998's one-year point growth was over 20%, which although far from the best year in this group, is still an above-average performance within an above-average group.
I told you I would explain why dividing the group into years with the lowest normalized P/E, middlemost normalized P/E, and highest normalized P/E makes a lot of sense.
Basically, you need to look beyond the average (i.e., mean) and get a better look at what's driving that number. These are very small groups (each group consists of five years), so it's very easy to jump to conclusions based on data that might be purely coincidental.
Nonetheless, I would like to share some facts about these three groups with you. First, let's take a look at the range of fifteen-year normalized P/E ratios within each group.
Lowest Normalized P/E Group: 6.88 – 9.69
Middlemost Normalized P/E Group: 11.40 – 14.10
Highest Normalized P/E Group: 16.15 – 28.05
Normalized Earnings Yield
Right away, you can see that every year in the highest normalized P/E group has a normalized earnings yield that is unlikely to provide the kind of returns equity investors are accustomed to.
By inverting the normalized price-to-earnings ratios, we can calculate normalized earnings yields. For the highest normalized P/E group these yields range from 3.57% to 6.19%. This illustrates the logical basis for my original assumption that there is a relationship between normalized price-to-earnings ratios and long-term share price performance.
If the fifteen year normalized earnings numbers do tend to roughly approximate the Dow's "earnings power", then buying the Dow at a very high normalized price-to-earnings ratio is a lot like paying a large premium over par for a long-term bond – you're lowering the yield on your investment below that of the stated coupon. Obviously, this kind of thinking is nothing new. You've probably heard the term "earnings coupon" before. That's really all we're talking about here.
What's interesting in this case, though, is that while I fully expected to see unusually high "premiums" (or "discounts") disappear over time, I did not expect to find a normalized P/E effect in one-year returns. After all, that's why I started by looking at compound point growth in the Dow over 15-year periods – I thought that was the natural place to look, because there would be far too much noise in the market over shorter periods of time.
Here, we've seen that at least during mid-term years there is an apparent normalized P/E effect. But, I think the situation may be a bit more complicated than it first appears.
Variation and Consistency
To illustrate, let me show you one last set of data. Here is the range of one-year annual point growth in each of the three groups we've been discussing – with one change. This time I've included all seven years that are neither one of the five lowest normalized P/E years or one of the five highest normalized P/E years.
Lowest Normalized P/E Group: 2.95% - 34.55%
Middlemost Normalized P/E Group: 9.85% - 32.58%
Highest Normalized P/E Group: (7.31%) – 21.45%
Again, let me remind you that the only two negative numbers happen to show up in the highest normalized P/E group.
By far, we see the least variation in the middlemost P/E group. In fact, if we simply exclude the five highest and five lowest normalized P/E years, these are the one-year annual point growth numbers for the remaining seven years: 17.47%, 19.09%, 25.99%, 32.58%, 11.72%, 28.55%, 9.85%.
The middlemost P/E group doesn't just offer the best performance – it also offers the most consistently excellent returns.
The five highest normalized P/E years showed a 6.57% average one-year point gain in the Dow. That's much better than the average for high normalized P/E years at other times in the election cycle. But, it's still not truly spectacular.
The five lowest normalized P/E years showed a 15.34% average one-year point gain in the Dow. That's very good by any measure. In fact, it's quite a bit better than the average for low normalized P/E years at other times in the election cycle.
But, in both groups, you had a few really great years and a few mediocre to quite poor years. With the middlemost group you had seven years of very good to truly great performances – without a single blemish on the record. You also had an average one-year point gain of more than 20%. That's quite a combination.
What does all this mean? I have no idea. I've already looked into other factors that might explain the outperformance during years in which the normalized P/E ratio was between 11.40 – 14.10 (e.g., the difference between normalized earnings and actual earnings, the one-year earnings growth rate, and simply when the years fell).
So far, none of those factors offered any promising clues. In fact, most of these years have very little in common except for the fact that they all managed to squeeze into a very tight normalized P/E band.
Obviously, you can go either of two ways with this data. You can either say, "It's only seven years among a group of seventeen years; it's clearly just a random fluke of absolutely no significance." Or, you could try to come up with a logical explanation for the outperformance.
For instance, you could hypothesize that in years where the Dow is within a tight normalized P/E range that happens to be centered around a very normal sounding "earnings power" multiple of about 12-15, there's a very good chance that investors themselves feel quite "normal" and are somehow more receptive to whatever good things come in the third year of the election cycle.
Do I believe any of this myself? Well, I certainly wouldn't be comfortable making a bet on the future direction of the Dow using this information. But, then again, I wouldn't be entirely comfortable making a bet it's all a random fluke either.
Anyway, I should probably try answering the two questions this post started with.
1. Is there some predictive value in the normalized P/E ratio on the return in the third year of an election cycle?
2. Is the third year effect stronger than the P/E effect?
I don't know about "predictive value", because we're working with a very small set of historical data and a very complex system (the market). But, yes, there is a discernible pattern in the data.
Simply put, the more "normal" the fifteen-year normalized P/E is, the better the one-year point growth in the Dow tends to be. Also, when the fifteen-year normalized P/E is abnormally high, the election cycle effect tends to be muted.
As for the second question, there are a few ways you can assess whether the third year effect is stronger than the P/E effect. First of all, the one-year point growth in these mid-term years does tend to be, on average, better than even very low normalized P/E years that occur at other times in the election cycle. However, among the mid-term years we looked at, the normalized P/E was the only variable that appeared to be related to the one-year returns for these years.
In several different ways, we saw that the normalized P/E ratio still seems to matter – even during the third year of the election cycle. Clearly, betting on a big one-year move up in the Dow during high normalized P/E years would have exposed you to greater risk and poorer returns than making that same bet during the years that didn't have especially high normalized P/E ratios.
Remember, two of the three years with the highest normalized P/E ratios actually posted negative point growth in the Dow. The very worst third years of the election cycle both followed high normalized P/E years.
Finally, during years in which the normalized P/E ratio happened to be close to a very "normal" sounding range of 12 – 15, returns were consistently excellent.
I can't find any other variable that does a better job of selecting good years (and weeding out bad years) from this set of mid-term years than the normalized P/E ratio. Simply put, a "normal" normalized P/E ratio seems to be a good sign going into the third year of the election cycle.
Considering how small this set of data is, I wouldn't consider it much more than a sign. I have about as much faith in this third year normalized P/E combination as I do in augury – but, I can disprove neither.
So, I have to plead agnostic on this one. Not only do I not know, I think I know it's not knowable.
The Year Ahead
If you really believe in this third year normalized P/E combination, I hope you believe in a higher power as well, because a quick prayer before New Year's might not be a bad idea.
Based on the Dow's most recent close, the fifteen-year normalized P/E ratio for 2006 is 21.39. The good news is that the only mid-term year with a higher normalized P/E ratio was 1998 and that worked out great. The bad news is that 2006 will have a normalized P/E ratio that's higher than both of the mid-term years that posted negative point growth in the Dow.
If you really want a prediction based on past third years in the election cycle with similar normalized P/E ratios, I would say the Dow's expected return (ex-dividends) would be in the 4% - 6% range and there's at least a 50/50 chance the Dow will end the year below where it starts the year.
Personally, I think it would be unbelievably foolish for someone to actually bet on next year's performance based on the information presented in this post.
I don't think the fifteen year normalized P/E ratio can tell you much about how the market is likely to perform in 2007.
I do, however, think it can tell you that the Dow will have a hard time matching its historical point growth pace over the next 15 years – but, that's a post for another day.
Note: I've been so busy responding to questions about my last post on this subject, that I haven't even told you how I calculated the fifteen year normalized price-to-earnings ratios. Don't worry, I'll do that in my next post.