Today, I'll be detailing how I calculated the normalized P/E numbers I referenced in two previous posts: "On 15-Year Normalized P/E Ratios for the Dow" and "On Normalized P/E Ratios and the Election Cycle."
That explanation will come in a later post. First, I'd like to revisit two topics about which I've received quite a bit of email over the last twenty-four hours. The two topics readers seem most interested in are the election cycle and the relationship between 15-year normalized price-to-earnings ratios and one-year point growth in the Dow.
First, let's tackle the election cycle. When writing about this (normalized P/E) project, I run a lot of numbers I never report to you. For the most part, I only share interesting or unexpected findings. However, I still routinely check to make sure I'm not missing something obvious. Despite these checks, I encourage (and ultimately depend on) your attempts to keep me honest by pointing out the possible holes in my logic.
So, let's poke a bit at the findings from the last post and see if we can find a hole.
One obvious explanation for the election cycle effect is that mid-term years might tend be abnormally cheap years. Is this hypothesis supported by the data?
Technically, mid-term years do have below-average 15-year normalized P/E ratios. But, I wouldn't say these years have abnormal 15-year normalized P/E ratios, because other randomly selected groups from within this same set of years (1935-2005) would also have normalized P/E ratios that fall a bit below the average for the entire set.
The "full set" (1935-2005) had an average (mean) 15-year normalized P/E of 14.08, a median of 13.59, and a range of 6.88 – 30.84. Just under 44% of the years in this set had a normalized P/E of less than 12.50.
The "election cycle set" (1938, 1942, 1946…) had an average 15-year normalized P/E of 13.46, a median of 13.00, and a range of 6.88 – 28.05. Just over 47% of the years in this set had a normalized P/E of less than 12.50.
The 12.50 Rule
The importance of this last check (percentage of years with normalized P/E of less than 12.50) is based purely on logic. Before beginning this study, I felt that when the Dow has a 15-year normalized earnings yield of 8% or more (i.e., a normalized P/E of 12.50 or less) there is a very good chance it is an attractive purchase for long-term investors, because other assets don't tend to offer long-term returns superior to those expected from an asset priced at 12.5 times its "earnings power", and sometimes present greater risks (including a loss of purchasing power) than a diversified group of large businesses like the Dow normally does.
Obviously, the fact that, since 1935, the Dow has been (what I would call) "undeniably cheap" nearly 44% of the time helps explain why it has done so much for long-term investors. Stocks are not inherently attractive; they have often been attractive, because they have often been cheap.
Possible Measurement Mistakes
The election cycle set strongly outperforms over a one-year period despite not having a much lower normalized P/E than the full set (1935-2005). Maybe my year-to-year, average-to-average point growth measurement is the problem.
I'll take this question up in the next post, because it involves going over the one-year point growth numbers for the entire set. That particular investigation will lead us to an interesting and unexpected finding that deserves its own post.
The one-year point growth outperformance in the election cycle set (i.e., the difference in the Dow's yearly average for year 3 of the cycle over year 2) can not be explained by mid-term years having abnormally low normalized P/E ratios.
Between the election cycle set and the full set, the difference in performance is both large and convincing; the difference in normalized P/E ratios is small and unconvincing.
In other words, insofar as one-year point growth in the Dow is concerned, there is both an apparent "normalized P/E effect" and an apparent "election cycle effect". These effects do not tend to coincide or "mirror" one another. They are separate and distinct.
In fact, both tendencies seem to be operative at once. However, within the election cycle set, the normalized P/E effect is somewhat obscured, because the election cycle effect tends to be of a much greater magnitude than the P/E effect over a one-year time period.
Nonetheless, the normalized P/E effect can be seen within the election cycle set – especially in high normalized P/E years. In those years, the election cycle effect is substantially muted and the frequency of negative and below-average returns is much greater.
Simply put, past experience shows that although betting on the Dow in the third year of the election has resulted in good overall results, that bet is considerably less attractive in high normalized P/E years.
Based on this logic (which I don't have much confidence in), betting with the election cycle effect during 2007 would be a much less attractive wager than betting with the election cycle effect in most years, because 2006 has the sixth highest normalized P/E on record – and the second highest normalized P/E among mid-term years.