Where 155 Years of Stock Returns Actually Came From

A share of stock pays you in three ways. It hands you cash. The earnings behind it grow. And the price other people will pay for a dollar of those earnings drifts up or down. Those three, and nothing else, add up to what you earned.

That framework is standard. What is rarely done is running it on actual data. So we did, over Robert Shiller’s S&P Composite series from 1871 to 2026.¹ Real annualized, over 155 years:

Three things in that table are worth more than the framework itself. Dividends supplied 61% of the return. Real earnings growth ran about 2% a year, which is far below what “stocks grow with the economy” suggests. And the multiple, which went from 11 times earnings to 28, contributed 0.60% a year across the whole span.

Those figures use trailing twelve-month earnings, which is what Bogle and most of the older literature used. The calculator below defaults to a more conservative setup: ten-year average earnings, starting in 1881 because the smoothing needs a decade of history first. That version reports 6.69% real, made up of 4.16% income, 1.91% earnings growth and 0.51% multiple change. Same story, and the fact that it barely moves when you change both the earnings measure and the start date is the first sign the result is not an artifact of either choice.

How the split is calculated

The price leg is an identity, not an approximation. If M is the price-to-earnings multiple, then by definition:

Price return therefore splits into earnings growth and multiple change exactly, with no residual. Income is then the wedge between the realized total return and the price return, so the three parts always reconstruct what actually happened rather than approximating it. Total return comes from a monthly dividend-reinvested index built from the same series. Everything is deflated by CPI unless labelled nominal.

This is deliberately not the Campbell-Shiller log-linearization, which is the usual tool for this job. Ben-David and Chinco catalogued its failure modes in 2026, including the one that matters most here:

“The formula registers buybacks and new issuance as phantom cash-flow shocks.”²

Their paper is an NBER working paper and has not been refereed, and the authors are explicit that it is an inventory of qualifications rather than an attack. The elementary ratio above sidesteps the issue entirely.

The method reproduces published work

A computation is only as good as its checks. Ours reproduces four independent published decompositions, which is the reason to believe the numbers above:

The extraction script refuses to write its output if any of those drift, so the checks run every time the data is refreshed.

Two percent is the number that should surprise you

Real earnings per share grew about 2% a year over 155 years. Real US GDP grew roughly 3% to 3.5% over comparable spans. The gap is not a measurement error, and closing it explains most of what confuses people about stocks and the economy.

Per-share earnings track real GDP per capita, which has run closer to 1.8% to 2.0%, rather than aggregate GDP. The difference is population growth plus the capital that flows into companies that did not exist when you bought your shares. Jay Ritter put the mechanism plainly:

“Historically, much of economic growth has come from the infusion of capital into new firms, which does not result in a higher growth rate of dividends per share for existing firms. Technological change has tended to benefit consumers and labor, not the owners of capital.”⁶

Ritter also found a cross-country correlation of −0.37 between real equity returns and growth in real per capita GDP across 16 countries from 1900 to 2002. Worth stating carefully: that result is not statistically significant, at p = 0.16, and Ritter says so. The claim the data supports is that faster economic growth has not reliably produced better shareholder returns, which is weaker and more interesting than a negative relationship.⁶ Our guide on GDP and stock returns works through that separately.

One measurement warning, because it changes the answer by more than a percentage point. Growing 2.06% is an endpoint-to-endpoint figure. Fit a trend line through the whole series instead and you get 1.68%. Endpoint measures are hostage to whether the first and last years happened to be good ones, and the effect is largest post-1945, where the endpoint method reads 1.16 points hotter than the trend because 1945 itself was a depressed-earnings year. When someone quotes a long-run earnings growth rate, the method matters as much as the number.

The multiple contributes nothing over a century and everything over a decade

Across 155 years the P/E went from 11 to 28, and that entire re-rating is worth 0.60% a year. Spread thinly enough, even a near-tripling of the multiple barely registers.

Nobody invests for 155 years. Over the windows people do hold, multiple change is frequently the largest single term:

A sixteen-point spread between the best and worst decade, on a term that averages out to almost nothing. The historical average conceals which valuation regime you happened to start and finish in, and that is not something a saver chooses.

Where the calculation falls apart, and why that is useful

Set the tool to 1900 through 2009 with trailing earnings and it reports a P/E of 70.9 and real earnings growth of −0.05% a year. Both are nonsense. January 2009 trailing earnings had briefly collapsed to almost nothing, so the multiple exploded and took an offsetting bite out of the growth term.

Switch to ten-year average earnings and the same window reports a P/E of 16.0 and real growth of 1.67%. Antti Ilmanen’s published figures for exactly that window are 21.9 and 1.3%.⁷ The smoothed version is in the neighborhood; the trailing version is not in the same postcode.

This is the entire reason Shiller built a cyclically adjusted P/E. A single year of earnings is a noisy denominator, and the noise is worst precisely when markets are most interesting. The tool defaults to the smoothed measure and warns you when an endpoint is distorted.

The largest component has shrunk by two thirds

Income supplied 4.30% a year over the full period. By decade it ran around 5% to 6% through the middle of the twentieth century and roughly 1.8% to 2.0% since 2000. The biggest contributor to the historical record is structurally about three points smaller than it was.

The necessary caveat, immediately. That figure counts dividends only, and American companies moved much of their payout to share repurchases. A buyer today receives a meaningful net buyback yield on top of the dividend, so the honest comparison is total shareholder yield rather than the dividend line, and the gap narrows. It does not close. Our guide on shareholder yield covers the full net payout measure. Nothing here contradicts dividend irrelevance: this finding is about the size of the payout, while that argument is about its form.

Two further reasons not to extrapolate the recent record. Corporate profits sit near the top of their historical range as a share of the economy, so the strong real earnings growth of the 2020s reflects margin expansion and a shrinking share count rather than a new trend rate. And Michael Smolyansky at the Federal Reserve estimated that falling interest expense and falling corporate tax rates together accounted for more than 40% of real profit growth among S&P 500 non-financial firms between 1989 and 2019, neither of which can repeat from current levels.⁸ That is an unrefereed working paper, and worth weighing as one estimate.

Reconciling with our own dividend attribution guide

Summitward has a guide arguing that dividends account for about 31% to 33% of S&P 500 returns, and this page puts income at 61%. Both are right, and the difference is entirely definitional:

• Real versus nominal. Inflation is a large slice of nominal return and none of real return. Income is 61% of the real figure and about 37% of the nominal one over 1926–2026.
• Period. That guide runs from 1926; this one from 1871. The pre-1926 decades had the highest dividend yields on record.
• Method. A geometric decomposition of annualized return is not the same calculation as an arithmetic attribution of monthly returns.

Any figure between roughly 30% and 60% is defensible depending on which three choices you make, which is itself the lesson. Ask which basis, which period, and which method before accepting a dividend-contribution statistic from anyone.

What this changes about a planning number

The 7.09% real figure is not a forecast, and using it as one repeats a mistake Fama and French identified. Estimating the equity premium from fundamentals over 1951–2000, they found 2.55% from dividend growth and 4.32% from earnings growth against a realized 7.43%, and concluded:

“Our evidence suggests that the high average return for 1951 to 2000 is due to a decline in discount rates that produces a large unexpected capital gain. Our main conclusion is that the average stock return of the last half-century is a lot higher than expected.”⁹

The decomposition shows the same thing mechanically. A forward assumption starts from today’s payout yield and a defensible real growth rate, and treats expected multiple change as zero unless you can argue otherwise. That lands well below the historical realized number, which is why our guide on planning assumptions argues for roughly 5% real rather than 7%. This page is the arithmetic underneath that recommendation.

What this evidence cannot tell you

• It is one country, and the winning one. The US was the strongest large equity market of the twentieth century. Survivorship runs through every figure here. See the global record and what long histories can and cannot support.
• January is arbitrary. Every window here runs January to January. Different months move the decade figures.
• Accounting earnings changed. A dollar of reported 1890 earnings and a dollar of 2026 earnings were produced under different rules.
• It says nothing about the future. The decomposition is an accounting of what happened. Each term still has to be forecast separately, and the multiple term is not forecastable at all.

Frequently asked questions

Does this mean I should buy dividend stocks?

No. The finding is that the market’s payout yield was historically large, not that high-yielding companies beat low-yielding ones. Paying out cash and generating returns are different things, and a company can raise its dividend by cutting investment. The comparison here is the whole market against itself over time.

Why is real earnings growth so much lower than GDP growth?

Per-share earnings track GDP per capita rather than aggregate GDP, and even then leak value to new firms that existing shareholders do not own. Population growth and net new share issuance account for most of the gap.

Should I use the endpoint figure or the trend figure?

Trend, for anything forward-looking. It is far less sensitive to whether your first and last years were unusual. Report which one you used, since the two differ by roughly 0.4 points over the full sample and more than a point post-1945.

Why does the tool default to ten-year average earnings?

Because a single year of earnings is a noisy denominator and the noise peaks in recessions. Measured on trailing earnings, the window ending in January 2009 reports a P/E of 70.9 and negative real earnings growth, neither of which describes anything real.

Does the 0.60% multiple contribution mean valuations do not matter?

The opposite. It means they matter enormously over the horizons people invest over and wash out only across spans nobody lives. See what CAPE does and does not predict.

Key takeaways

• Real US stocks returned 7.09% a year over 155 years: 4.30% income, 2.06% real earnings growth, 0.60% multiple change. Cash paid out was the largest component by a wide margin.
• Real earnings per share grew about 2% a year, tracking GDP per capita rather than GDP. New firms and population growth absorb the difference.
• Multiple change averages ~0 across the full span and swings from −7.3% to +9.0% by decade. The average hides the thing that determines your outcome.
• The income leg has fallen from roughly 5-6% to under 2%. Buybacks close part of that gap and not all of it, so the historical total is not a forward expectation.
• Method changes the answer. Endpoint versus trend growth differs by more than a point post-1945; trailing versus smoothed earnings can invert the split entirely.

Related guides

• What Real Return Should You Assume for Stocks?: the planning number this arithmetic sits underneath.
• Shareholder Yield: the full net payout measure, which is the right modern version of the income leg.
• Do Most S&P 500 Returns Really Come From Dividends?: the attribution question, and why the answer moves with the method.
• Do Stock Valuations Still Matter?: the multiple term, forward-looking.
• The Stock Market Is Not the Economy: the GDP-to-returns chain in full.
• Growth Stocks Do Not Mean Higher Expected Returns: the same decomposition applied to a single company.

Sources and method

1. Robert J. Shiller, long-run US stock market data (shillerdata.com), the series behind Irrational Exuberance. Monthly S&P Composite price, dividends, trailing earnings, CPI and CAPE from 1871.01, read here through 2026.08. Note the widely cited Yale mirror is frozen at 2023.09.
2. Itzhak Ben-David and Alex Chinco, “Crimes Against Campbell-Shiller”, NBER Working Paper 35189, May 2026. Unrefereed.
3. John C. Bogle, The Little Book of Common Sense Investing, Wiley, 2007, exhibits 2.2 to 2.4. Nominal, S&P 500. Bogle’s “investment return” is dividend yield plus earnings growth and his “speculative return” is the multiple change.
4. Research Affiliates, on long-run real earnings growth measured as a log-linear trend fit rather than an endpoint ratio: 1.5% for 1871–2017, 0.8% pre-war and 1.8% post-war.
5. Elroy Dimson, Paul Marsh and Mike Staunton, “The Worldwide Equity Premium: A Smaller Puzzle,” in Rajnish Mehra (ed.), Handbook of the Equity Risk Premium, Elsevier, 2008, table 4. Their decomposition uses dividends and price-to-dividends rather than earnings and price-to-earnings, so exact agreement is not expected.
6. Jay R. Ritter, “Economic growth and equity returns”, Pacific-Basin Finance Journal 13, no. 5 (2005): 489–503. The quoted passage is from page 502. The −0.37 correlation covers 16 countries, 1900–2002, at p = 0.16.
7. Antti Ilmanen, Expected Returns on Major Asset Classes, CFA Institute Research Foundation, 2012, exhibit 2.5.
8. Michael Smolyansky, “End of an era: The coming long-run slowdown in corporate profit growth and stock returns,” Finance and Economics Discussion Series 2023-041, Board of Governors of the Federal Reserve System, June 2023. Covers S&P 500 non-financial firms. Unrefereed.
9. Eugene F. Fama and Kenneth R. French, “The Equity Premium,” The Journal of Finance 57, no. 2 (April 2002): 637–659. The 2.55% estimate comes from the dividend growth model and 4.32% from the earnings growth model.

Method. Every figure on this page is computed by scripts/extract_shiller.py and the calculator above, from January observations of Shiller’s series. Total return comes from a monthly dividend-reinvested index where each month’s cash is the annualized dividend divided by twelve. Price return is split by the exact identity in the method section. Income is the residual between total and price return, so the parts always sum to the realized figure. Real figures are deflated by CPI. Headline numbers use trailing twelve-month earnings for comparability with Bogle and the older literature; the calculator defaults to ten-year average earnings, which holds up better and which we recommend for any window ending near a recession. The extraction script validates against four published decompositions on every run and refuses to write output if they drift.

Editor’s note

Educational content, not investment advice. Historical decompositions describe what happened and do not forecast what will. Figures computed August 8, 2026 against Shiller data running through August 2026; citations verified against primary sources on the same date.