Investors greatly overstate the short-term effect of earnings and ignore the long-term impact of dividends. The implications are momentous.
Overview
Mainstream financial media, as well as professional and retail investors, obsess about corporate earnings and their near-term prospects. The fixation is constant, but it’s most intense during “earnings season.” The media encourage investors, and investors abet the media: as a result of this feedback loop of mutual reinforcement, both groups apparently – and ardently – believe that the rate of growth of today’s profits, and expectations about tomorrow’s, propel stocks’ current and prospective returns.
On its face, this conjecture seems plausible. It’s hardly unreasonable to suppose that, all things considered, companies whose earnings are rising (or expected to rise) will be more valuable to investors – and thus generate higher returns – than those whose profits are static or falling.
Perhaps the self-evident nature of this tacit assumption and routine assertion to so many people, including prominent and influential people, explains why so few have bothered to investigate it.
In this article I ask: do actual and anticipated earnings really underpin stocks’ actual and prospective returns? I corroborate research which shows that this relationship is on average much weaker and far more erratic than investors and the media believe, and that the influence of dividends upon returns is much more significant than they realise (see also Dividends aren’t a bane – they’re a boon, 20 November 2023).
Specifically, as determinants of short-term returns short-term earnings typically don’t matter; but as determinants of long-term returns, long-term average earnings occasionally exert some influence. Even in the long term, however, earnings count much less than the crowd supposes, and far less than something that it ignores – long-term average dividends.
(Readers who’ll find tedious the descriptions of the analysis’ assumptions and results can skip the next several sections and proceed directly to the “Applying This Key Result” and “Implications” sections.)
Some Fundamental Preliminaries
“There is no way to predict the price of stocks and bonds over the next few days or weeks,” declared The Royal Swedish Academy of Sciences on 14 October 2013 in the press release that announced that year’s Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel (which is universally but erroneously known as “The Nobel Prize in Economics”).
This key point also applies over periods of at least several months and perhaps as long as 1-2 years: during these intervals, equities’ prices – and thus returns – fluctuate mostly randomly. “But,” the announcement continued, “it is quite possible to foresee the broad course of these prices over longer periods, such as the next three to five years.” These longer periods extend to 10 years. As The Wall Street Journal (“Can Stocks Surpass 2022 Highs? Yes, but the Math Looks Scarier from There,” 9 January) noted:
“How will stocks perform in 2024? The only honest answer to the obvious New Year’s question is that we don’t know, but we can have some idea how they might fare over the coming decade.”
“These findings,”the Academy’s press release elaborated, “which might seem both surprising and contradictory, were made and analyzed by this year’s Laureates, Eugene Fama, Lars Peter Hansen and Robert Shiller.”
Stocks’ prices and market indexes’ levels – and thus returns – “are nearly impossible to predict” over days, weeks, months and sometimes periods of more than one year, but they vary relatively systematically over periods ranging from several years to ten years. In the short term, noise overwhelms signal; but in the long term, the ratio of signal to noise rises substantially.
Using the longest, most valid and reliable series of monthly data – which Shiller compiled for his book Irrational Exuberance (Princeton University Press, 1st edition, 2001) and has updated thereafter, Figure 1, Figure 2 and Figure 3 apply the insight that the Academy’s press release summarised to the Standard & Poor’s 500 Index’s short-term (12-month), medium-term (five-year) and long-term (ten-year) changes of CPI-adjusted earnings, dividends and total (that is, its dividend and capital growth) returns.
Figure 1: CPI-Adjusted Earnings, Dividends and Total Returns, CAGRs over Three Time Intervals, January 1871-September 2023
The Index’s CPI-adjusted total return over rolling 12-month periods since January 1872 has averaged 8.0%. For rolling 60-month periods since January 1876, its total return, expressed as a compound annual growth rate (CAGR), has averaged 7.1%; and for rolling 120-month periods since 1881, it’s averaged 6.9%.
Figure 2: Standard Deviations of the Index’s CPI-Adjusted Earnings, Dividends and Returns over Three Time Intervals, January 1871-September 2023
Moreover,
- The same point applies to earnings and dividends: the longer is the interval, the lower is the CAGR.
- Figure 1 also shows, regardless of the interval, that the average total return greatly exceeds earnings’ average CAGR; and the rate of growth of earnings, on average, modestly exceeds the growth of dividends.
- Figure 2 shows that the longer is the interval and regardless of the variable, the lower is the standard deviation (that is, the dispersion of observations around their mean). In other words, long-term earnings, dividends and returns fluctuate much less than their short-term counterparts.
Finally, Figure 3 plots the co-efficient of variation (CV), i.e., the ratio of standard deviation to mean, of the Index’s short-term, medium-term and long-term earnings, dividends and total returns. For example, the mean return for rolling 12-month periods is 8.0% (Figure 1) and its standard deviation is 15.6% (Figure 2); hence its CV is 15.6% ÷ 8.0% = 1.95, and so on for the other intervals and variables.
Figure 3: Coefficients of Variation, S&P 500 Index’s CPI-Adjusted Earnings, Dividends and Total Returns, Three Intervals, January 1871-September 2023
In plain English, the higher is the CV, the greater is the ratio of noise to signal. For each variable, the ratio is highest in the short-term and lowest in the long-term. Accordingly,
- If we wish better to understand the S&P 500 Index’s CPI-adjusted total returns, it makes sense to analyse its long-term (rolling ten-year) total returns. That’s because its ratio of noise to signal is lowest.
- As factors explaining short-term total returns, short-term changes of earnings and dividends are useless – they contain much more noise than signal.
- As a candidate explaining long-term total returns, dividends make at least as much sense as earnings.
Clarifying Some Key Assumptions
Do companies’ earnings drive the S&P 500 Index’s returns? It’s a simple question, yet some complex assumptions underlie it, and it’s vital to clarify and justify them. Let’s start with what’s indisputable and obvious: every month, as the newest observation of the Index’s earnings, dividend and total return enter the rolling 12-month, 60-month and 120-month series and the oldest one leaves it, these series’ means, CAGRs, etc., update. Table 1, which enumerates the Index’s CPI-adjusted earnings during two randomly-selected rolling 12-month intervals, provides a simple example – and clarifies an important distinction.
Table 1: the S&P 500 Index’s CPI-Adjusted Earnings and Varying Summary Statistics, Two Consecutive 12-Month Intervals
The two intervals are November 1952-October 1953 and December 1952-November 1953. The “Actual” column enumerates the Index’s actual CPI-adjusted earnings during each month of the relevant period; the entries in “12-month mean” compute average earnings for the year ending in a given month. For example, earnings averaged $28.29 in November 1952-October 1953; that quantity appears both as the mean of the “Actual” column and as observation #12 (bold font) in the “12-month mean” column. The mean for October 1952-September 1953, $28.16, appears as observation #11 in the “12-month mean column, and so on for that column’s other observations, and so on for the interval from December 1952 to November 1953.
Which earnings drive the Index’s returns: the actual monthly earnings or the rolling 12-month means? The later remove (some of the) the random month-to-month variation of earnings; accordingly, their standard deviations are one-third to one-half as large as those of the former.
The rolling 12-month means provide a slower-moving target – one that investors, to the very limited extent that they can discern current and forecast prospective earnings, will hit more easily than the actual observations. That’s why these data – specifically the CAGRs, which allow direct comparisons of 12-month, 60-month and 120-month series – underpin my subsequent analysis.
From one 12-month interval to the next, the oldest observation from the previous interval (#1) exits the series and the newest one (#12) enters it. Consequently, if the correlation of a given series of earnings with another corresponding series (say, total CPI-adjusted returns from November 1952 to October 1953) is perfect, during each of an interval’s 12 (or 60 or 120 as appropriate) months the change from one month to the next (i.e., from observation 1 to 2, etc.) of the earnings series begets an equivalent magnitude of change of the corresponding observations in the return series.
Under these conditions, by buying or selling the Index as appropriate, and thus adjusting the total return series’ CAGR to change in lock-step to the earnings series’ CAGR, investors respond instantly and perfectly accurately to each month’s earnings.
More generally, the more quickly and accurately investors react – by buying or selling the Index and thereby adjusting its total return – to new information about earnings throughout a given (12, 60 or 120-month) interval, the higher will be the two series’ correlation during that interval; conversely, the less quickly and accurately they respond, the lower will be their correlation.
Do Earnings Drive Returns?
Adjusted for CPI and expressed as CAGRs, and over one-year, five-year and 10-year rolling intervals, the Index’s growth of earnings and total returns are generally positive (I very much doubt that’s controversial, so for the sake of brevity I’ll skip several graphs that substantiate this point) but vary greatly (recall Figure 1 and Figure 2). But are they correlated?
If, from one month to the next throughout a given interval, the earnings and total return series’CAGRs changed by an identical amount and in the same direction (but not necessary by the same magnitude each month), then the two series would be perfectly positively correlated and their correlation coefficient (r) would attain its maximum value of 1.0.
Conversely, if during each month one series increased by some amount and the other fell by an equivalent magnitude, then the two series would be perfectly negatively correlated and the coefficient would attain its minimum of -1.0. Finally, if the change of one series from one month to the next bears to relation to the change in the other, then their correlation is 0.0 and the PE ratio fluctuates randomly.
Figure 4: Correlations of S&P 500’s CPI-Adjusted Earnings and Total Return, 12-Month Rolling Intervals, January 1872-September 2023