Check · Dividend as income

Is dividend safe income?

Published 26 August 2026 · Updated 2 September 2026

"Dividend is safe income" is one of the most-repeated sentences in investing, and a lot of people have built a calm, sensible portfolio around it. It is popular because parts of it are true. This piece splits the sentence into the four things it rests on, and lays each one next to a hundred years of data — so you can see which parts to lean on, and which parts to hold more loosely.

A hand seen in dark navy silhouette against a bright window, holding a single large coin up to the light between thumb and forefinger to inspect it. The rim of the coin catches a warm amber glow; everything else is calm blue and navy. No text or numbers. — **Figure 1.** Illustration. A coin, held up to the light — which is all this check does.

The idea we're checking

"Dividend is safe income."

A family of sentences: "passive income you can live on", "gets paid whether the market goes up or down", "cash on top of your capital". You will meet it in fund brochures, in courses, and from friends who have been happily collecting dividends for years. It is easy to believe for a good reason: the cheque really does arrive, on a schedule, whatever the share price did that week — and that feels like the one solid thing in an otherwise jumpy investment. This check asks which parts of that feeling the data supports.

The answer, in one sentence

For a broad, well-diversified basket of shares, the cheque really does keep coming: in nearly eighty post-war years the aggregate dividend never fell more than 21% and was back above its old level within five years — far steadier than the share prices it drops out of. That part holds, and this check says so. "Safe income" also rests on three more assumptions, and those do less well. That the cash comes on top of your capital: it doesn't — the share price drops by roughly the dividend on the day it is paid, and of the four this is the one we are most certain about. That a dividend basket cushions you when markets fall: a real effect on average, not a reliable one. And that more yield means more return: true against paying nothing, no longer true at the very highest yields. One assumption kept, two kept on average or in part, one not — that is the stack in Figure 2, with the bar for each coin under How we checked. What that means in practice is further down: a broad dividend basket has historically been a fine thing to own; "safe" is just the wrong word for what it gives you.

The Proof Stack for this check: four coin stacks under four labels. 'The aggregate dividend keeps coming' is a filled amber stack. 'The cash is extra, on top of the price' is a fully dotted outline. 'It cushions you when markets fall' and 'Higher yield means higher return' are each half-filled: the bottom two coins amber, the top two dotted. Under each stack a short line says what does survive. The count at the top reads: promised 4, proven 1, partly 2. — **Figure 2.** The Proof Stack — promised 4 · proven 1 · partly 2. Each coin names one thing that would have to be true for "dividend is safe income" to hold the way people mean it, and each has an explicit bar, stated under How we checked so you can judge our judgement. A filled coin clears its bar. A half-filled coin is a promise the evidence carries *on average or in part*, but not reliably enough for a whole coin — the line beneath it says which. A dotted coin is contradicted. **How sure of the finding: fairly sure** — that is about the finding itself, not about how big any return is.

For whom the claim is plausible: broad-basket investors with a long horizon who want a steady aggregate cheque, or who would rather not decide when to sell — and who can live with the share price going anywhere. For whom it isn't: anyone taking "safe" to mean the share price is safe, that the cash comes on top of your capital, or that the basket reliably holds up in a downturn; and anyone counting the cheque in today's money — from 1965 to 1989 it bought less than before. For whom it may cost money: yield-chasers buying the fattest yields (the top tenth earned less than the top third, and on average a cut is already in the price a year early); holders of a handful of single dividend stocks, where the basket's steadiness doesn't apply; and anyone who spends the cheque expecting the reinvested growth figures — those exist only if the cash goes back in.

First, what a dividend actually is

Skip this if you already know it; if you don't, everything below is easier with a clean picture. A dividend is a payment a company makes to its shareholders out of the profit it has earned — a small share of that profit, per share you hold. If you own a hundred shares of a company that declares a dividend of one euro, you get one hundred euros. The company's board decides how much to pay, if anything, and how often — quarterly is common in the United States, once or twice a year is common in Europe.

Where does that money come from? Mostly the business's own earnings. A company generates revenue, subtracts costs and taxes, and what's left is net profit. It can do four things with that profit. Pay it out as a dividend. Buy back its own shares (economically similar to a dividend, but the cash goes only to the shareholders who sell). Reinvest it in the business. Or keep it on the balance sheet as cash. The share of profit paid out as dividend is called the payout ratio, and it varies wildly by sector and by decade. Some companies choose to pay out most of what they earn; many pay out nothing.

Here is the part that usually gets left out, and it is the direct answer to "where does the money come from": a meaningful share of it is borrowed. The three authors of our main review followed American listed companies from 1989 to 2019 in a 2025 follow-up study. They found that 43% of the firms making a payout also raised capital in that same year, so that 31% of all money paid out was externally financed — "primarily with debt". A quarter of the total, 25%, could not have been paid without that simultaneous raise.⁸ For regular dividends specifically the share is higher still: 44.9% of dividend-paying firms raised capital in the same year.⁸ Worth knowing: the study leaves out utilities and financial firms, which pay some of the most reliable dividends there are. So "out of profit" is the normal case, not the only one — and the gap is filled by borrowing more often than the word "income" suggests.

A diagram with no numbers. A navy box on the left reads 'What's left after costs and tax'. Four arrows fan out to four boxes on the right: 'Paid out as a dividend' highlighted in amber and labelled 'the one you feel', then 'Used to buy back its own shares', 'Reinvested in the business', and 'Kept as cash on the balance sheet'. — **Figure 2b.** The four routes out of a profit. Only the first one lands in your account. Note that the first two both leave the company with less cash than it had — a buyback spends money just as a dividend does; the difference is who ends up holding it. No data here: this is an explanatory diagram, not a measurement.

Which companies do pay? Farre-Mensa, Michaely and Schmalz survey nearly a century of evidence in their 2014 review of payout policy and state one of the field's most robust patterns plainly: "large and profitable firms pay much more dividends than risky, growth firms."¹ Utilities, consumer staples, big financials and mature industrials are dividend-heavy; young technology and biotechnology firms typically pay nothing at all and reinvest instead. And even among payers, the money is concentrated: the same review notes that "roughly half of aggregate dividends [are] paid by only 25 firms."¹ A "world of dividends" is, in dollar terms, a fairly narrow cast.

A quick word on tax — the shortest we can make it. Tax rules on dividends differ from country to country, and inside a country they differ by account type and tax residence. As a rule of thumb: in a taxable account most jurisdictions tax dividends when they are paid, whereas an unrealised capital gain is only taxed when you sell (and sometimes not even then). So the tax treatment can make a small difference to what a euro of dividend is worth to you versus a euro of capital gain. At the level of the firm, the payout literature says it barely matters: differences in the taxation of dividends and capital gains have, in the review's own summary, "only a second-order impact on setting payout policy."¹

It is not nothing for you, though, and the exchange itself puts a number on it. Euronext publishes the AEX both gross of withholding tax and net of it: 10.88% a year since 1983 against 10.48%.⁶ Four-tenths of a percentage point sounds like a rounding error. On the €10,000 of Figure 7, held for those 43 years, it is the difference between €848,571 and €726,436 — about €122,000, or roughly a seventh of the final pot, handed over in withholding tax alone. That is Euronext's calculation for a non-resident investor without treaty relief; a Dutch retail investor gets relief and lands somewhere between the two lines. Ask a local adviser about your own situation; we're not going to fake being one.

Where the money actually comes from

Here is the sentence the whole "safe income" idea rests on: "the cash arrives on top of your capital." It's the reason a dividend feels like a paycheck and selling a share feels like eating your seed corn. Almost everyone feels it that way, and for good reason. But for the ordinary case of a listed company it is not what happens.

On the day a stock pays its dividend, the exchange marks its opening price down. That day is called the ex-dividend day: the buyer that morning is no longer entitled to that dividend, so a rational price is roughly the day-before price minus the dividend. The empirical average across markets, summarised in the same 2014 review, is that the price drops by about ninety percent of the dividend amount — "the exact value of the premium varies though time and averages around 0.9".¹ ("Though" is Farre-Mensa's own wording; they mean "through".) The small gap is attributed to tax preferences of different investor groups and to trading costs — it isn't a free ten percent of anything. Figure 3 puts it in one picture.

A schematic of what happens on an ex-dividend day. Day before ex-dividend: share price €50. The dividend is paid: €1 cash goes out to the holder while the share is still nominally €50. Opens on ex-dividend day: share price €49.10 (roughly the previous price minus the dividend). A red curved arrow marked 'the price falls by about the same amount' points from the cash to the new lower price. — **Figure 3.** The ex-dividend day, illustrated. A share worth €50 pays €1; the next morning it opens at roughly €49.10 — most of that dividend came out of the price. The 90% figure is the long-run empirical average of the "ex-dividend premium" across markets, reviewed in Farre-Mensa e.a. (2014).¹

Behind that empirical fact sits a piece of theory that has been standing since 1961. We take it from the 2014 payout-policy review that summarises it, since we have not read the 1961 paper itself: "Miller & Modigliani (1961) show that, in perfect and complete capital markets with no taxes, a firm's payout policy does not affect its value. … From the perspective of investors, payout policy is irrelevant, because any desired stream of payments can be replicated by appropriate purchases and sales of equity."¹ The reason, in gentler words: the net payout is just the residual — earnings, minus reinvestment, minus new share issuance — and any level of dividend can be replicated on the investor's side by buying or selling shares.

Real markets aren't perfect, so the theorem's assumptions get bent every day — taxes, informational asymmetries, transaction costs and the rest. That is exactly why payout policy can still matter in practice, and a large literature works out how. But the earlier Allen & Michaely handbook chapter, which we did read, states the cleanup line as sharply as anyone: "It could matter, not because dividends are 'safer' than capital gains, as was traditionally argued, but because one of the assumptions underlying the result is violated."² That is the sentence to keep. The dividend does not arrive as a gift on top of your capital. It arrives partly out of it — because the market price already adjusts to take it out. What it does do is hand you that cash without a sell decision — the one thing the theory assumes you would do for free, and most people don't.

A useful thing to keep straight. "Dividend yield" (the annual dividend divided by the share price) and "dividend growth" (the year-over-year rise in the dividend the company pays) are two different things and are easy to mix up. A stock can have a high yield because its dividend has grown, or because its price has fallen. Only the first is unambiguously good news. The second is often read as a warning that the dividend is at risk — though as with the payout ratio, we found no source establishing how reliable that reading is. Investors have a nickname for that second case — a yield trap — though that is market shorthand rather than a term from the research literature, and we have not found it used in the reviews we cite here.

Does the dividend keep coming through the bad years?

Now to the part where the claim actually earns its coin. If you buy a diversified basket of dividend-paying shares — the S&P 500 as a whole, or an index like the AEX — how reliable is the aggregate paycheck? Figure 4 shows the total dollars paid out by the S&P 500 basket per index point since 1927 — both as it was paid and corrected for inflation, because those two tell different stories.

A chart on a logarithmic scale of the dividend paid per S&P 500 index point each year from 1927 to 2025, drawn as two lines. The pale line is the cash as it was paid: it collapses in the early 1930s, then climbs almost without interruption to about $78 by 2025. The solid navy line is that same cash expressed in 2025 money: it also collapses in the 1930s, recovers into the 1950s, and then runs roughly flat from the mid-1960s to the late 1980s while the pale line keeps rising. That flat stretch, from 1965 to 1989, is shaded. Two labels mark the 1928-34 collapse and the 1965-1975 real decline of 24 percent. — **Figure 4.** The S&P 500's total dividend per index point, every year from 1927 to 2025. The scale is logarithmic so a percentage move looks the same size wherever on the chart it happens. The pale line is the cash as paid; the solid line is that same cash in 2025 money. The shaded stretch is the twenty-four years the cheque spent below its 1965 purchasing power — ten of them falling, fourteen climbing back — and the cash line shows none of it. Source: Aswath Damodaran, *Historical Returns on Stocks, Bonds and Bills*, "S&P 500 & Raw Data" and "Inflation Rate" sheets.³

The picture is genuinely striking. Measured in the dollars of the day, the aggregate S&P 500 dividend has fallen 10% or more only four times in ninety-eight years. The largest by far was the Great Depression: from a peak of $1.047 per index point in 1928 to $0.352 in 1934, a cumulative drop of about 66%, and it took until 1949 to get back to the 1928 level.³ The most recent was 2008-9: after the financial crisis the annual dividend fell from $28.39 to $22.41, about 21%, and was above the 2008 level again by 2012.³ The other two are smaller and mostly forgotten — 1957 and 1959, at about 19% and 13%.

Against that, 2020 barely registers: despite a pandemic that collapsed real economic activity, the aggregate dividend fell by only 2.6% (from $58.50 to $57.00) and was above 2019's level again by 2021.³ Set that against what the share prices did across those same episodes — they fell dramatically further and stayed down longer, as any long-run S&P 500 chart shows. That is what "the aggregate dividend is stickier than the price" means, and it is the part of the "safe income" idea that the data supports best.

Best, but not all the way — and the gap matters most for exactly the person the idea is meant for, the one planning to live on the cash. Everything above is measured in the dollars of the day, and a dividend that stays flat while prices rise buys less every year. Correct the same series for inflation and the picture changes twice over. There are six falls of 10% or more rather than four, and the longest of them is invisible in the cash line: between 1965 and 1975 the dividend lost 24% of its purchasing power, and it did not regain the 1965 level until 1989.³ Ten years falling, then fourteen climbing back: twenty-four years in which the cheque kept arriving, rose in dollars far more often than it fell, and still bought less at the end than it had at the start. (The Depression looks slightly milder in real terms — a 57% fall rather than 66%, because prices were falling too — but recovery took until 1950 rather than 1949.³)

To be fair to dividends, this is a limit on every income stream, not a special weakness of this one. A bond coupon is fixed in cash and never regains lost purchasing power at all; the dividend eventually did, because dividends grow with nominal earnings. That is the honest limit of the word "safe": the cheque kept coming, and for twenty-four years it bought less than it had in 1965. It's worth knowing, and it rarely gets mentioned.

That stickiness isn't an accident. It's engineered. In one of the review's most quoted findings, executives told researchers that they will do almost anything before touching the dividend: "selling assets, laying off employees, borrowing heavily, or bypassing positive NPV projects to avoid having to cut dividends."¹ Cutting a dividend is treated as a public admission of trouble, and management pays a real price for it. That is precisely why it happens as rarely as it does.

But there is a second reason the aggregate cheque looks as smooth as it does, and it is not in the earnings at all. Following American listed companies from 1989 to 2019, the same three researchers found that 44.9% of firms paying a regular dividend also raised capital in that same year, and that 41.6% of all the money paid out as regular dividends coincided with money raised — against 35.7% for buybacks. (That side-by-side is our own reading of the study's Table 1; the authors never put those two columns next to each other, and their own headline comparison, measured against the money raised rather than the money paid out, runs the other way.⁸)

Read that gently, because it is easy to over-read and we did so ourselves in an earlier version. Coinciding is not the same as replacing: a company with plenty of cash that happens to issue a bond in the same year is counted too. The stricter test asks, of that coinciding dividend money, how much could not have been funded internally at all: between 48% and 64% of it, depending on the measure — so roughly half of the overlap is genuine dependence and the rest is coincidence with cash to spare. For buybacks the dependent share is higher still, around 80%.⁸

What the authors conclude from all this is not that firms borrow to rescue the dividend in bad years. They test that idea and reject it: most financed payouts "cannot be explained by payout smoothing in response to volatile earnings or investment", and are instead "the result of firms persistently setting payouts above free cash flow."⁸ That is a plainer and heavier finding than the rescue story. It is not that the dividend gets propped up when times are hard — financed payouts actually fall in recessions, along with everything else. It is that across the ordinary run of years, a meaningful share of companies pay out more than the business generates and fund the difference elsewhere. The steadiness in Figure 4 is real, and part of what sits behind it is a payout level set above what comes in the door.

How real is that price? When companies do cut, the market treats it as bad news immediately. Averaged across a large sample of American cuts, the share price fell by about 3.7% more than the market as a whole moved that day; when a company skipped the dividend entirely (an "omission"), the gap was about 7%.¹ (Researchers measure it against what the rest of the market did, so that a generally good or bad day doesn't get mistaken for a reaction to the dividend.) The most striking part is how one-sided it is. Raising a dividend produced only a 1.34% move the other way on average, and starting one from scratch about 3.4%.¹ The market punishes cuts far harder than it rewards raises. That is exactly the pressure that keeps managers loyal to the dividend even in bad years.

Bar chart of four average share-price reactions to dividend announcements, measured against the rest of the market that day. Skipping the dividend entirely: minus 7.00 percent. Cutting it: minus 3.71 percent. Raising it: plus 1.34 percent. Starting one from scratch: plus 3.40 percent. The two negative bars are roughly twice the size of the two positive ones. — **Figure 4b.** The same four numbers as the paragraph above, side by side. The asymmetry is the point: a company that cuts loses about twice what a company that raises gains, which is why boards protect the dividend so hard — and why a cut, when it finally comes, tends to arrive late and hurt.¹

What the aggregate does not tell you. All the above is aggregate. What it tells you is that the S&P 500 as a whole keeps paying — because when one company cuts, another one raises. It does not tell you that any individual dividend stock will keep paying. Companies cut all the time; the aggregate is stable because the pool is diverse. Membership of even the most conservative dividend lists is not a guarantee either: an Aristocrat that cuts its dividend loses its place on the list, and it happens. If you own five dividend stocks instead of a broad basket, you're getting the concentration risk without the aggregate stickiness. The payout ratio mentioned earlier is the numerical version of this warning: a company paying out almost all of its profit has little room left if profits dip, and it is arithmetic rather than a forecast that the room runs out sooner the higher the ratio is. We looked for evidence that a high payout ratio is a reliable predictor of a coming cut and did not find it in the reviews we read, so we are not going to claim it. What the numbers do show is how much room there has typically been. Averaged across American listed companies, the dividend payout ratio bottomed at 7.2% of earnings in 2000 and had climbed back to 14.7% by 2012, while total payout including buybacks reached 28.7% — its highest since at least 1971, up from a low of 15.4% in 1996.¹ Two things to keep in mind with those: they are averages over companies, not shares of one big pot, and the underlying sample leaves out both loss-making firms and two of the most dividend-heavy sectors there are — utilities and financials. So they describe the typical firm with room to spare, not the ones where the room has run out. The number that matters for a company you actually hold is that company's own. And even at the aggregate level "sticky" is not "safe": in the Depression a diversified holder still saw their nominal income drop by 66% and did not see the 1928 level again for twenty-one years. A cheque that gives back a fifth every few generations is steadier than most things in investing — and still not a guarantee.

What a dividend-focused portfolio actually returns

So much for the aggregate. The second thing "safe income" usually implies is that going out of your way to hold dividend-heavy stocks earns you a competitive return — that you're not paying for the cheque with a poorer total return. The longest clean dataset on this is Ken French's, at Dartmouth, which sorts every American stock by its dividend yield each year and calculates what each yield bucket would have returned. We took the value-weighted annual returns for 1928-2025 and recalculated the long-run annualised return in each bucket ourselves — the library doesn't publish that as a headline number.⁴ Figure 5 puts them all together.

A bar chart of six annualised returns for the period 1928-2025: stocks with no dividend at 9.33%, the bottom-30% yield bucket at 9.90%, the middle 40% at 10.43%, the top-30% at 11.07%, the top-20% at 10.84%, and the top-10% at 10.49%. The tallest bar is the top-30%; the top-10% bucket is shorter than both the top-20% and the top-30%. An amber callout points to the top-10% bar with the words 'Yield-chasing stops paying off'. — **Figure 5.** Annualised total return of American value-weighted stock portfolios sorted by dividend yield, 1928-2025. "Top 30%" means the highest yielders in the market; "no dividend" is the group paying nothing at all. Recalculated for this piece from the Ken French Data Library's D/P series.⁴

Two things sit inside that picture. First: yes, dividend-paying stocks as a group did modestly better than non-payers — the top 30% by yield returned an annualised 11.07% over 1928-2025, against 9.33% for the non-payers.⁴ Over a lifetime that ~1.7-percentage-point gap compounds to real money. In plain terms, one dollar left to grow at 11.07% for 98 years turns into about $29,400; the same dollar at 9.33% into about $6,260 — roughly four-and-a-half times as much. That is a real edge, and it lines up with what factor-model researchers have long attributed to value: the same stocks that show up in high-yield buckets tend to score high on value too.

Second, and this surprised us: more yield does not keep buying more return. Once you go past the top 30% and start selecting for the top 20% and then the top 10%, the numbers drop: 10.84% for the top 20% and 10.49% for the top 10%, versus 11.07% for the top 30%.⁴ That is the yield trap showing up in the data. The very highest-yielding stocks disproportionately have high yields because their prices have fallen for a reason — and part of that "reason" turns into a dividend cut later. "More yield" is not "more return": a portfolio built only from the fattest yields did worse, historically, than one that took the whole high-yield third — though still better than the market's 10.02% a year. The drop is inside the top third, not below it.³

The best-known dividend-focused product family is the S&P 500 Dividend Aristocrats, and its own factsheet lets us look at a specific ten-year window. The Aristocrats index takes S&P 500 companies that have increased their dividend every year for at least 25 consecutive years, weighted equally, with a 30% cap on any one sector and a minimum of 40 constituents. A detail worth knowing: that last condition outranks the first. If fewer than 40 companies qualify, the methodology drops the bar to more than 20 years of dividend growth, filling by yield — and if that still is not enough, it tops up from the S&P 500 by yield with no growth requirement at all. So "25 consecutive years" describes the intention, not a guarantee about every member.⁵

As of the factsheet we have — dated May 2021 — over the previous decade the Aristocrats returned 14.53% a year and the S&P 500 returned 14.38% a year: a wafer-thin 0.15-percentage-point win.⁵ They did it with slightly less bounce along the way. Standard deviation of monthly returns, which is a way of saying how much a return wobbled around its average from month to month, came in at 12.83% for the Aristocrats against the S&P 500's 13.61%. On the factsheet's own return-per-unit-of-risk measure — which is return divided by that same wobble, not the classical Sharpe ratio (Sharpe subtracts the risk-free rate first, this one does not) — the Aristocrats came in at 1.13 versus the S&P 500's 1.06 over ten years.⁵ A small edge, in the right direction, over a specific window. Zoom out to shorter or different windows and the edge flickers: over the five years ending May 2021 the Aristocrats returned 14.56% a year and the S&P 500 17.16% — the dividend basket trailed by about 2.6 percentage points a year.⁵

Three things worth naming before we go on. First, the factsheet we hold is dated 28 May 2021 — every Aristocrats number above is as of that date, not "current". None of the four calendar years since (2021 through 2024) are in these figures, and the specific ten-year window used here ends in May 2021. If you are weighing an Aristocrats-linked product today, look at the current factsheet rather than these numbers. Expect a different and probably less flattering picture: the broad S&P 500 has since been driven by a small number of mega-cap technology names, which is not the sector mix a dividend-focused index leans into.⁵ Second, the Aristocrats index only went live on 2 May 2005. Every published number for a period ending before that is hypothetical back-tested data, calculated with today's constituent list and today's rules, and S&P's own disclaimer notes the survivorship and hindsight caveats that come with that.⁵ That covers a large part of any Aristocrats "history" you'll see quoted, and the further back it goes the heavier the caveat. Third, S&P Dow Jones Indices licenses this index to product providers — the best-known exchange-traded fund tracking it is ProShares NOBL — and it also publishes the S&P 500 the index is compared with. That is normal in this industry; we mention it so you can weigh the factsheet accordingly.⁵

Stricter and looser cousins of the same rule. The 25-year threshold is one line in a family. Tighter: the Dividend Kings require 50 straight years of increases (the club is small — a few dozen names). Looser: the S&P High Yield Dividend Aristocrats require 20 years of increases within the broader Composite 1500. All three are constructed the same way — a rule on the past, applied to today's list — and all three carry the same catch: passing the past-rule tells you a company hasn't cut yet, not that it won't.

Is a dividend basket "dip-proof"?

The last piece of "safe" is downside protection: the idea that when a normal stock portfolio drops 30%, a dividend basket drops less. This is the part of the claim where the evidence is genuinely mixed, so it is worth going slowly. We have three windows on it: the Aristocrats' own calendar years, S&P's analysis of the 2008 crisis, and a century of American stocks sorted by yield.

A grouped bar chart with two pairs of bars. Left pair, 2018 (a broad down year): S&P 500 Dividend Aristocrats at -2.73%, S&P 500 at -4.38%. Right pair, 2020 (pandemic dip and rebound): S&P 500 Dividend Aristocrats at +8.68%, S&P 500 at +18.40%. — **Figure 6.** Two years, two very different outcomes. In 2018, when the broad market ended down, the Aristocrats lost about a percentage point less. In 2020, when the market dipped hard in March and rebounded, the Aristocrats trailed the S&P 500 by nearly ten percentage points on the calendar year — because a dividend basket that skews away from technology and growth had less of what led the recovery. Source: S&P 500 Dividend Aristocrats factsheet as of 28 May 2021.⁵

Start with the years that support the idea, because there are several. In 2018, a broad down year, the Aristocrats lost 2.73% against the S&P 500's 4.38% — a cushion of about a point and a half. In 2011 they returned 8.33% against 2.11%.⁵ And over the three years to November 2008, the window that contains the financial crisis, S&P's own analysis puts the Aristocrats at −3.32% a year against −8.67% for the S&P 500, with a lower spread of outcomes along the way.⁷ That is a real pattern, from the index provider's own numbers, in exactly the kind of period the idea is about. And the most recent bad year is a clear case: in 2022 the top fifth of American stocks by yield returned +11.54%, while the companies paying no dividend lost 37.35% and the lowest-yielding fifth lost 25.15%.⁴ Whoever held that top-yield fifth through the year felt exactly the cushion the idea describes — at least against the companies paying nothing; our set has no Aristocrats figure for 2022. The same buckets show the other side two years earlier: in 2020 the top-yield fifth lost 5.35% while the non-payers gained 50.17%.⁴ Same basket, same tilt, opposite year.

Now the years that don't. 2020 is often produced as evidence and it should not be: the market ended that year up 18.40%, so it does not measure who was cushioned in a fall. It measures who kept up in a rebound — and there the Aristocrats managed only 8.68%, nearly ten points behind, because the sector mix that pays reliable dividends is not the mix that leads a technology-heavy recovery.⁵ That is a genuine cost of a dividend tilt, but it belongs under "what you give up", not under "does it protect you".

The wider test is the one that settles the word consistently. Sorting all American stocks by dividend yield back to 1928 gives 26 years in which the S&P 500 lost money. In those 26 years the highest-yielding third beat the market in 15 of them, and beat the companies paying no dividend at all in 20. On average across those falling years it lost 9.6% where the market lost 13.5% and the non-payers lost 21.9%. (Our own calculation: the yield buckets are Ken French's,⁴ the S&P 500 return each year is derived from Damodaran,³ and the count of falling years is ours.) So the cushion is real on average — and it shows up barely more often than a coin flip. The same data complicates the 2018 story we opened with: measured by yield buckets rather than by the Aristocrats list, the high-yield third actually lost more than the market that year (−5.79% against −4.23%). Two different S&P 500 figures for 2018 appear on this page, −4.38% and −4.23%. The first is the benchmark line printed on the Aristocrats factsheet; the second is our own calculation from Damodaran's series. We quote each next to the data it belongs to rather than silently picking one.

Two reasonable ways of asking "did dividend stocks protect me in 2018?" give opposite answers — and the reason is worth knowing, because it decides which one applies to you. The Aristocrats list is a quality filter: to be in it a company must have raised its dividend every year for decades, which selects for stable, established businesses. A yield bucket is a price filter: a share lands in the top third because its dividend is large relative to its price, which happens both when the dividend is generous and when the price has fallen. Those two select very different companies, and only the first has the defensive character the idea assumes. If you own a broad high-yield fund, you own the second.

So: sometimes a dividend basket protects you and sometimes it doesn't, and the years it doesn't tend to be the years growth stocks run. "Safe" is the wrong word for that. "Often a bit softer in a fall, and sometimes not" is less catchy, but it is what the data supports — and that is not an argument against owning one, as long as you know which of the two you have.

The part of the story that really does compound

None of the above should obscure the one thing about dividends that is genuinely unambiguous, and genuinely good news. If you own a broad stock index and reinvest your dividends — buying more shares with each payment instead of spending the cash — the difference to your final pot is enormous over a working life. Figure 7 shows it for one index a Dutch reader might know: the AEX, over the last forty-three years.

Two lines from March 1983 to March 2026 showing what €10,000 invested in the AEX became: with dividends reinvested (AEX Gross Return at 10.88% a year) it reaches €848,571; price only (AEX price index at 7.34% a year) reaches €210,256. Both curves grow gently at first and diverge dramatically after year twenty. — **Figure 7.** €10,000 invested in the AEX in March 1983, held to March 2026. The two annualised return figures — 7.34% for the price index alone and 10.88% with dividends reinvested — are the ones Euronext itself publishes on the AEX factsheet.⁶

Both lines are the same index; the only difference is whether the dividends were put back to work. At 7.34% a year for 43 years, €10,000 becomes €210,256. At 10.88% a year over the same period, it becomes €848,571.⁶ The gap — €638,315, or three quarters of the larger pot — is what dividends did once they were fed back into the machine. Nothing about that argues the dividend is "safe income". It argues that reinvested dividends are one of the biggest slow-cooking forces in long-run stock returns, which is a different sentence.

The same math shows up in American data with the same shape. In Damodaran's long-run series, $100 invested in the S&P 500 at the start of 1928 with all dividends reinvested became about $1,157,000 by the end of 2025 — a 98-year annualised return of 10.02%.³ The same $100 held only for its price change — no dividends — became roughly $38,763, about a thirtieth as much.³ The "with dividends" pot is about thirty times bigger, and the reason is not that dividends are extra: it's that the cash the company paid out was continuously used to buy more shares in the same company. If instead you had taken the cheque every year and spent it, most of the top line goes away — because most of the top line was the reinvestment.

How often the money arrives, in one paragraph. American-listed companies typically pay quarterly, most European ones pay once or twice a year, and a smallish set of specialty vehicles pay monthly. For a working-age investor who is reinvesting, the frequency is a rounding difference — same total, slightly earlier compounding. For a retiree who is spending the dividend as income, the frequency becomes a household-cashflow question, not a return question, and can usually be smoothed with a small buffer in a bank account. Monthly-payout funds sound appealing for exactly this reason; worth knowing before you buy one is that some "monthly income" products get to that headline by returning part of your capital — that is a separate check, and it is in our queue.

What this means for you

Start with what holds. If you own a broad basket of shares, the aggregate cheque is far steadier than the price it comes out of. Since the war its worst year was a 21% fall, from 2008 to 2009; in 2020 it slipped just 2.6%.³ That solves a real problem for anyone living off a portfolio: choosing when to sell, which is hardest exactly when prices are lowest. A dividend hands that choice to boards that go a long way to avoid a cut.

Now the parts to hold more loosely, starting with costs: nearly every return on this page is gross, with no fund fees, trading costs or tax. Gross, the Dividend Aristocrats index beat the S&P 500 by a whisker over ten years, 14.53% against 14.38% a year. On the factsheet's net total return line — dividends after a withholding-tax deduction at a rate the factsheet does not state — it was 13.65% against 13.68%: the net ten-year line tips it the other way.⁵ Inflation bites too: the cheque lost 24% of its buying power from 1965 to 1975 and did not get back above its 1965 level until 1989.³ That is 24 years below the peak — a long wait for anyone living on it.

A dividend basket cushions a fall on average, not reliably. In the 26 losing years since 1928 the highest-yield third lost 9.6% on average, against 13.5% for the market and 21.9% for non-payers. 2022 was the good version: the top-yield fifth gained 11.5% while non-payers lost 37.4%. 2018 was the other kind: the highest-yield third lost 5.8% while the market lost 4.2%.⁴³ And the very highest yields earned less than the wider high-yield third. The price often knows first: in the year before a cut, a share has on average already fallen about 28% behind comparable shares.¹

Two useful things survive all this: a cash line that takes the sell decision off your plate, and a discipline you can automate. $100 in the S&P 500 in 1928 grew to about $1,157,000 by the end of 2025 with dividends reinvested, against $38,763 on price alone — roughly thirty times more.³ On the AEX the dividend added 3.54 percentage points a year for 43 years.⁶ Those numbers exist only if the cash goes back in. Plenty of the best businesses in the world pay dividends, and a diversified stake in them has historically been a fine thing to own. So the honest label is not "safe income" but "a steady cheque you decide what to do with".

What to take away

A broad basket's aggregate dividend is genuinely sticky — in cash. Four falls of 10% or more in ninety-eight years, and management works hard to keep it that way. Corrected for inflation there are six, and one of them kept the cheque below its 1965 purchasing power until 1989. And the reluctance to cut is an Anglo-American habit, not a law of nature: more than 80% of loss-making German companies with five good years behind them scrap the dividend in the loss year, which our own main review contrasts explicitly with the American pattern.¹ Worth knowing — and the stickiness is a property of the basket, not of a handful of individual dividend stocks.
The dividend isn't extra money. On the day it's paid the share price drops by about the same amount, and the theoretical case for "payout policy is irrelevant" has been standing since 1961. It is a cash movement inside your portfolio, not a paycheque from your employer.
Dividend-focused portfolios are not automatically safer in a downturn. Sometimes they hold up better (2018); sometimes they miss most of the rebound (2020). "Dip-proof" is not the right word.
High yield helps, but only up to a point. The top third by yield beat the non-payers by nearly two percentage points a year — but the top tenth did worse than the top third (though still ahead of the market), because the fattest yields are often the market's warning that a cut is coming.
Reinvested dividends are the real multiplier. Same AEX index, same 43 years: €210,256 versus €848,571. That is the piece of the story that survives every test in this piece — but only if you reinvest.

Sources

Farre-Mensa, Joan, Roni Michaely and Martin Schmalz, "Payout Policy", Annual Review of Financial Economics vol. 6, 2014, pp. 75-134. DOI: 10.1146/annurev-financial-110613-034259. The peer-reviewed survey of a hundred years of dividend research, and the source we quote for the Miller-Modigliani payout-irrelevance theorem (§3), the ~0.9 ex-dividend premium (§4.1), the market reactions to dividend increases and cuts (§5.4.1), the management-behaviour findings around dividend cuts (Brav et al. 2005, §2.3), the "second-order" tax finding (abstract), the average 28% abnormal return in the year before a dividend cut (Benartzi, Michaely & Thaler 1997, §5.4), the "large and profitable firms" stylised fact, and the concentration of aggregate dividends in a small number of firms (§1). We hold the PDF locally; we have not read Miller & Modigliani (1961) themselves — every MM passage above is quoted through this review.
Allen, Franklin and Roni Michaely, "Payout Policy", chapter for the Handbook of the Economics of Finance, North-Holland, 2002. SSRN: ssrn.com/abstract=309589. The earlier of the two payout-policy surveys we read. Source of the sharpened line quoted in §"Where the money actually comes from" — that if dividends matter, it is because one of the Miller-Modigliani assumptions is violated, not because dividends are "safer" than capital gains.
Damodaran, Aswath, "Historical Returns on Stocks, Bonds and Bills — United States", updated series through 2025 (dataset). Author's page: pages.stern.nyu.edu/~adamodar/. Source of the annual S&P 500 dividend per index point (sheet "S&P 500 & Raw Data", column C) used in Figure 4 and in the three crisis-cut figures cited in the text, and of the $100-invested-in-1928 comparison in the compounding section (sheet "Returns by year", column "Value of $100 invested at start of 1928 in S&P 500 (includes dividends)"). Local file: refs/damodaran-histretSP.xls.
French, Kenneth R., Data Library — Portfolios Formed on D/P (dividend yield), downloaded June 2026, CRSP-based, value-weighted annual returns 1928-2025. Library: mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. We used the value-weighted annual return series for the "<= 0" bucket (non-payers), the deciles Lo 30 through Hi 30, and the finer Hi 20 and Hi 10, and calculated the 98-year annualised geometric mean ourselves — the library publishes the annual data but not the long-run summary numbers in Figure 5. The 2020 and 2022 yearly bucket figures in the "dip-proof" section and in "What this means for you" come from the same file.
S&P Dow Jones Indices, S&P 500 Dividend Aristocrats — Index Factsheet, as of 28 May 2021. We hold a third-party archived copy of this PDF (from suredividend.com); the publisher blocks automated access to the original at spglobal.com. Source of the ten-year total return, the ten-year monthly wobble figure (which is the standard-deviation number quoted above), the ten-year return-per-unit-of-risk figure, the ten-year net total return pair (13.65% against 13.68%; the factsheet does not state the withholding-tax rate it applies), the five-year total return comparison, the 2018 and 2020 calendar-year returns in Figure 6, and the launch-date and back-tested-history disclaimer quoted in the sidebar. Methodology — the 25-year rule, the equal weighting, the 30% sector cap, the minimum of 40 constituents — comes from S&P Dow Jones Indices, S&P 500 Dividend Aristocrats — Methodology, November 2015 (third-party copy via miraeasset). The commercial line about index licensing (relevant here because ProShares NOBL is licensed off this same index) is the boilerplate on the factsheet's own back page.
Euronext, AEX-Index Factsheet, as of 31 March 2026, published by Euronext N.V. Source of the annualised return figures used in Figure 7: AEX (price only) 7.34% a year since 01/03/1983, AEX Gross Return 10.88% a year over the same period, and AEX Net Return 10.48%. The dividend yield of 2.48% and the top-10 concentration of 75.42% quoted in this note come from the same factsheet — the AEX is, for three-quarters of its weight, ten companies. Local file: refs/euronext-aex-factsheet-2026-03-31.pdf.
Soe, Aye M., S&P 500 Dividend Aristocrats, Standard & Poor's, December 2008. S&P's own analytical paper on the index, data as of 30 November 2008. Source of the three-year crisis window in the "dip-proof" section: Exhibit 12 gives three-year returns of −3.32% a year for the Dividend Aristocrats against −8.67% for the S&P 500 and −11.44% for the S&P 500 Equal Weight, with standard deviations of 12.44%, 15.26% and 17.97% respectively — that last set is what "a lower spread of outcomes" refers to. ⚠️ The paper is from 2008 and carries no current performance figures; we use it only for that crisis window and for the index's construction. Local file: refs/soe2008.pdf.
Farre-Mensa, Joan, Roni Michaely and Martin Schmalz, "Financing Payouts", Journal of Financial and Quantitative Analysis 60(4), June 2025, pp. 1586–1624. DOI 10.1017/S0022109024000231, published online 1 April 2024, open access under CC-BY. The follow-up to source 1 by the same three authors, and the source for the financing figures in this piece: 43% of paying firms raise capital in the same year, 31% of aggregate payout externally financed — "primarily with debt" — 25% that could not have been paid otherwise, 44.9% of regular-dividend payers raising capital alongside, the 41.6%/35.7% coincidence shares from its Table 1 (our own column comparison), and the 48–64% and ~80% dependence shares from its Table 4. Sample: Compustat-CRSP, 1989–2019, 11,557 firms, 106,407 firm-years. ⚠️ Like source 1 it excludes financial firms and utilities — two of the most dividend-heavy sectors there are. Local file: refs/farre-mensa2024-financing-payouts.pdf.

Corrections. This check was revised on 27 August 2026, the day after publication, after two adversarial reviews of our own, and again on 2 September 2026 for tone. What changed, so that anyone who read an earlier version can see exactly what moved:

2 September: the same findings, a friendlier voice. A reader told us the first version read as if we were against dividends. We weren't, but the writing kept addressing a "pitch" and its "selling" rather than the reader, and carried eleven warning signs in the running text. No number, no bar and no coin changed. The opening block now says why the idea is easy to believe before it is tested; the answer opens with what holds; "the pitch", "simply false" and the like are gone; the warning signs are ordinary sentences; and the interest of product providers has moved from the opening to this fine print, where it belongs. Two things were added on the way: the strongest fair case for the idea, written out separately and used in the closing section, and a reader's-eye pass by our completeness reviewer. Four figures came in from the same verified set, none of them new to our files: the 2022 yield-bucket returns, the factsheet's net-of-tax ten-year pair, the average 28% fall in the year before a dividend cut, and the AEX dividend contribution of 3.54 percentage points a year.
Three recovery dates were wrong. The 2008 level was regained in 2012, not 2013; the 2019 level in 2021, not 2022. The Depression date was corrected too, and then corrected again in the anchor-year item below once we found that the peak year itself was wrong.
The compounding comparison was overstated by about a third. We had put Damodaran's "$100 from the start of 1928 with dividends" next to a price-only figure that started a year later, after a 38% up-year. On the same start date the price-only pot is $38,763, not $28,000, and the multiple is about thirty rather than forty. ⚠️ Both of our blind extractions made the same choice of starting year, so the reconciliation reported no discrepancy on a number that was wrong. Double extraction catches misreadings; it does not catch a shared assumption.
Inflation was missing entirely. The word did not appear anywhere in the first version, while the only coin we had marked as carried was measured in the dollars of the day. Figure 4 now shows the series in real terms as well, and the 1965–1989 stretch it reveals is the longest decline on the page.
The dip-proof section was one-sided in both directions. It leaned on 2020 as evidence against, though the market ended that year up, and it left out both 2011 and S&P's own three-year crisis window, which support the idea. It now carries all three, plus the century-wide test.
Smaller: announcement figures are described as what they are (moves relative to the rest of the market, not raw price falls); the claim that the market reaction figures came from a "34% intra-year drop" is removed as unsourced; "yield trap" is now flagged as market shorthand rather than a term from the literature; a made-up reconciliation of two lists of Miller-Modigliani assumptions is gone; and S&P's 2008 paper is now listed as a source rather than used without being credited.
A second pass, the same day, after three more reviews. The first round of fixes changed the running text but left the captions, alt texts and takeaways behind, so several of them still described the older version — including a screen-reader description of a chart that no longer existed. Those now match what is actually on the page. Two numbers moved as well. The Depression is measured from 1928, not 1929. 1929 was not the peak: the dividend per index point was $1.047 in 1928 against $0.879 in 1929, so the fall is 66% rather than 60% and the recovery year is 1949 rather than 1948. Our own reconciliation had overruled the one extraction that got this right. ⚠️ Note what this means: the correction makes the check harder on the claim, not softer, and it is the second anchor-year error in two days — the same mistake in a different place. And the 1965–1989 stretch is described more precisely: the dividend's purchasing power fell for ten years and then took fourteen to recover, so it was below its 1965 level for twenty-four years rather than falling for twenty-four.
A source we said we could not open, and could. The first version of this revision carried a paragraph naming a follow-up study by our main authors, saying every route to it was blocked and that no figure on the page relied on it. The first of those was wrong. The paper was reachable — an unauthenticated request to the Oxford repository returns it, and we had simply not tried that one — and "no figure relies on it" is not a reason to publish around a gap in the first place; that judgement is not ours to make alone. The study is now source 8, it went through the same double blind extraction as everything else, and it changed the answer to the question this piece opens with: paying out and raising capital in the same year is normal practice, not an edge case. ⚠️ We also had its citation wrong: it is 2025, not 2024.

The four coins, one by one. Figure 2 splits the claim into four things that would all have to be true. Each has an explicit bar, stated below with what we measured against it and how sure we are — so you can judge our judgement rather than take it. 1. The aggregate dividend keeps coming — filled. Bar: on a modern investor's horizon (the post-war era, nearly eighty years), no cash fall of more than 30%, and recovery within five years. How sure: fairly sure. The worst post-war fall was 21% (2008–09, from $28.39 to $22.41 per index point), back above its old level by 2012; 2020 managed only −2.6% and was recovered by 2021. That clears the bar with room. ⚠️ Two honest caveats travel with this coin. The Great Depression sits outside the window we chose, and it was far worse: −66% from 1928 to 1934, not regained until 1949 — if your test is "survives anything history can throw at it", no income stream passes. And measured in purchasing power the cheque spent 1965 to 1989 below its 1965 level (Figure 4). ⚠️ That second caveat cuts less against dividends than it first seems: every income stream loses to inflation — a bond coupon is fixed and never catches up at all — and this one did catch up, because dividends grow with nominal earnings. It is a real limit on the word "safe"; it is not a reason to prefer the alternatives the same buyer is offered. 2. The cash is extra, on top of the price — dotted. Bar: on the day it is paid, the share price does not drop by roughly the amount of the dividend. How sure: very sure. The measurement runs straight against the promise. On the ex-dividend date the share price is reduced by the dividend just paid, and across markets the observed fall averages roughly 90% of the dividend. The payment moves value out of the company and into your account; it does not add any. Of the four assumptions this is the one that does not hold, and it is the one the whole "income" feeling rests on. 3. It cushions you when markets fall — half-filled. Bar: reliably better in falling markets. How sure: not very sure. The half that holds: across the 26 years since 1928 in which the S&P 500 lost money, the highest-yielding third lost 9.6% on average where the market lost 13.5% and the non-payers lost 21.9%; it beat the non-payers in 20 of those 26 years, and S&P's own three-year crisis window to November 2008 shows the Aristocrats at −3.32% a year against −8.67%. The half that does not: it beat the market in only 15 of the 26 — barely better than a coin flip — and 2018 flips sign depending on how you measure it. A real cushion on average, not a reliable one: exactly what a half-filled coin means. 4. Higher yield means higher return — half-filled. Bar: the income does not cost you return, and more yield brings more of it. How sure: not very sure. The half that holds: over 1928–2025 dividend payers returned 11.07% a year (top 30% by yield) against 9.33% for the non-payers and 10.02% for the market — leaning toward dividends has not cost the long-run investor return. The half that does not: more yield is not more return. The top 20% returned 10.84% and the top 10% just 10.49% — the line stops rising before you reach the highest yielders, which is the opposite of what yield-chasing assumes. ⚠️ Our figures here are gross and not adjusted for risk; a full coin would need more than that. How the three states work. A filled coin clears its bar with at least "fairly sure" certainty. A half-filled coin is a promise the evidence carries on average or in part, but not reliably or certainly enough for a whole coin — the stack never says "half" without the line under the coin and the paragraph here saying which half. A dotted coin is contradicted by the measurement. The count — one kept, two half, one not — is also the honest summary of this page: nothing here says dividends are bad; it says "safe income" assumes more than the evidence delivers, and precisely where. Who has an interest. Providers of dividend-focused products — from broad Aristocrats-linked funds to covered-call and high-yield funds — benefit when the sentence is believed as a whole. That is not disqualifying, and the arithmetic on this page does not depend on it; we state it so you can weigh their materials accordingly.

How we checked. Verified 26–27 August 2026 against the eight sources above, with the four figures added on 2 September 2026 verified against the same reconciled set that day; all opened and read by us — with one important exception: we have not read Miller & Modigliani (1961) in the original, because the paper is behind a paywall we could not clear in the time available. Every Miller-Modigliani passage on this page is quoted through Farre-Mensa e.a. (2014), which restates the result and its assumptions in the section we cite. That is stricter than paraphrasing the abstract of a paper we cannot open. On study selection. This piece adopted the literature already screened by Farre-Mensa e.a. (2014) and Allen & Michaely (2002) — the two payout-policy reviews that between them cover a hundred years of the dividend literature. We did not run a separate literature search of our own; the four sub-promises are all addressable with data from the anchors plus Damodaran's long-run series, the Aristocrats factsheet, the Ken French library and the Euronext AEX factsheet listed above. The Aristocrats caveats matter and are worth repeating together. The factsheet we hold is dated May 2021, so every Aristocrats performance figure in this piece is as of that date and not "current". The Aristocrats index itself only went live on 2 May 2005, so anything about its behaviour before then is hypothetical back-tested history calculated with today's rules — S&P's own disclaimer notes this may reflect survivor and look-ahead bias. And the same organisation that publishes the index licenses it to product providers, so the commercial interest is real and worth stating. The Ken French geometric means in Figure 5 are our own calculation from the library's yearly value-weighted returns; the library publishes the annual series but not that summary number. The formula is a straightforward product-of-(1+return)-to-the-1/98; the reconciliation between our two independent readers matches to two decimal places, and can be reproduced by anyone with the CSV. Three source-internal disagreements are worth stating. Farre-Mensa lists six assumptions behind the MM theorem while Allen & Michaely list five. ⚠️ We have not read Miller & Modigliani in the original and cannot say from the source how the two lists map onto each other, so we are not going to invent a reconciliation: they are two summaries of the same theorem that carve up its assumptions differently. The AEX Gross Return (10.88% a year since 1983) sits half a percentage point above the Net Return (10.48%): the difference is dividend withholding tax as Euronext calculates it for a non-resident investor without treaty relief. A Dutch retail investor gets treaty relief and lands somewhere in between, so the "with dividends" line in Figure 7 is the correct order of magnitude for either. And S&P's own 2008 analytical paper and its May 2021 factsheet paint a shifting picture of the Aristocrats' edge — the 2008 paper (leaning heavily on pre-2005 back-tested data) shows a consistent crisis-era outperformance, while the 2021 factsheet shows a marginal 10-year win alongside a clear 5-year loss. The framing of "consistently beats" therefore depends on which window a broker picks. Not investment advice — a check of a widely repeated claim. Think we got something wrong? Tell us at info@proofofreturns.com — we correct in public.