The Break-Even Line: 126 Years of US Inflation and What Standing Still Costs

One sentence in the first chapter of Burton Malkiel's A Random Walk Down Wall Street is the kind most readers nod at and pass straight over:

Remember, just to stay even, your investments have to produce a rate of return equal to inflation.

— Burton G. Malkiel, A Random Walk Down Wall Street, chapter 1, “Firm Foundations and Castles in the Air”

It arrives in a passage where Malkiel is explaining that the book is not a speculator's manual, and he follows it with several pages on what inflation does to purchasing power. Put as arithmetic: break-even means a return equal to inflation.

It reads like an accounting identity, and it is one. But an identity stated is not an identity felt. How high is that bar, actually? Was it the same bar in 1925 as in 1975? What does it cost, in real money, to hold the safest asset in the world for a working lifetime?

Those are measurable questions. This article answers them for the United States from 1900 through 2025 — 126 complete calendar years — using public data, with every figure produced by a script rather than recalled from memory. A companion piece takes the next step: what happens when the market falls apart, how long recovery has actually taken, and how the arithmetic of allocation follows from these numbers.

This is a history and methodology piece. It describes what happened. It is not a forecast, and it is not advice about any security.

1. Where the numbers come from

Three sources, chosen because all three can be republished without a licence:

• US Bureau of Labor Statistics, CPI-U (series CUUR0000SA0, not seasonally adjusted), monthly from January 1913. A US government work, in the public domain.
• Robert J. Shiller's Irrational Exuberance dataset, monthly from 1871: the S&P Composite price, trailing dividends, a consumer price index, and a long government bond yield. Publicly released for research.
• FRED series TB3MS, the 3-month Treasury bill rate, monthly from January 1934.

Two deliberate constraints shape everything below.

No index levels appear anywhere in this article. Index point values are the licensed product of the index provider. Everything here is expressed as a percentage return, or as the growth of 1.00 — a multiple, not a level. The underlying arithmetic is unaffected; the licensing exposure is.

Nothing is estimated to fill a gap. Where data does not exist, the article says so. One case is worth naming up front: the BLS never published a CPI figure for October 2025. Collection was suspended during that autumn's federal government shutdown, and the value is marked unavailable in the BLS file to this day. The annual figures below use the BLS's own published annual average for 2025, which the agency computed from the eleven months that do exist. Our monthly series interpolates that one month, and says so.

2. What US inflation actually looked like

Across the 126 calendar years from 1900 to 2025:

The mean of 3.15% is the number everyone quotes. It is also the least informative number in the table, because inflation did not arrive in even 3% slices. The standard deviation is larger than the mean. Ten separate years exceeded 10%. Seventeen years had prices falling.

The single most important pattern: there were two regimes, not one

Look again at the chart. The deflation years are not scattered across the century — they stop. Split the record at 1933, when the US left the domestic gold standard, and two different worlds appear:

The earlier era had lower average inflation and more than twice the volatility, with prices falling in one year out of three. The later era is higher and far steadier, and deflation has become close to extinct — six years in ninety-three, most of them mild.

This matters for reading history correctly. An investor in 1910 faced a genuinely two-sided price level: cash could gain purchasing power just by sitting still. An investor since 1933 has faced a one-sided one. Any intuition about "safe" assets carried over from the first regime does not transfer to the second, and a 126-year average silently blends the two.

By decade

At 1.77%, the 2010s had the mildest inflation of any decade since the US left the gold standard. The 1920s and 1930s came in lower still, but only because prices were falling — which is a different condition, not a calmer version of the same one. Anyone whose investing intuition formed between 2010 and 2020 therefore calibrated on the quietest inflation of the entire fiat era.

It is worth noting what preceded the two worst inflationary decades. The 1900s ran at 2.62% and the 1960s at 2.35%, both modestly below the long-run average of 3.05%. Neither was extraordinary, and that is the point: the decade before an inflation shock did not look like a warning.

3. The hurdle, in money

Compounded across the full period, the price level rose about 44-fold.

That number is abstract. Here is the same fact stated the other way: one dollar of 1900 purchasing power costs about $44 today, and a dollar held from 1900 to now would retain roughly 2.3 cents of what it could buy. Not lost 2.3 cents. Retained 2.3 cents.

This is the break-even line. It is not a benchmark someone chose; it is the definition of standing still. And the crucial property of the line is that there is no way to decline it. An investor who buys nothing has not opted out of the comparison — they have taken the position that loses to it by the full amount.

Which raises the obvious question about the asset people actually hold when they want to be safe.

Treasury bills: the safest asset, and a 92-year real loss

From January 1934 — when the 3-month bill series begins — through July 2026:

• Nominal: +3.47% a year, turning 1.00 into 23.51.
• Real: −0.08% a year, turning 1.00 into 0.93.

Ninety-two and a half years of rolling the safest instrument the US government issues, reinvesting every payment, never missing a single one, and ending with less purchasing power than you started with. Not dramatically less — but the direction is what matters, because this is the asset that carries no credit risk, no duration risk, and no volatility worth mentioning.

The loss was not evenly distributed. Bills lost about 4.6% a year in real terms during the 1940s, when rates were held down by policy while wartime and post-war inflation ran, and lost ground again through the 1970s. They earned a positive real return in the 1960s, 1980s, 1990s, and — barely, at +0.20% a year — the 2000s. On net, across nine decades, the ledger is slightly negative.

The word "risk" usually points at volatility. This is the other kind: an asset that almost never moves and quietly fails to clear the break-even line for a lifetime.

4. What the assets actually returned

Over the full span for which we have all inputs — December 1899 to July 2026, about 126.6 years:

A note on the two inflation figures: §2 reports 3.05% and §4 reports 3.00%. Same data, different basis — §2 compares calendar-year averages, which is the BLS's official annual convention, while the table above uses month-end points from December 1899 to July 2026. Neither is wrong; mixing them would be.

Three observations.

The equity premium is large and it is real, not nominal. 6.87% a year in purchasing power over more than twelve decades is the headline result, and it survives every crisis in the record — two world wars, the Depression, the 1970s, 2008, 2020.

Bonds cleared the bar, but by little. 1.52% real a year turns 1.00 into 6.71 over 126 years. Equities' real multiple is roughly 670 times larger. Both beat inflation; only one of them compounds into a different life.

These figures triangulate against the standard reference. The 2026 edition of the Dimson–Marsh–Staunton Global Investment Returns Yearbook puts US real returns since 1900 at 6.6% a year for equities, 1.6% for bonds and 0.5% for bills, with inflation averaging 2.9%. We compute 6.87% and 1.52% for equities and bonds — agreement to within about a quarter of a percentage point, from independent construction on independent data. That is the point of checking.

Two differences are worth naming rather than glossing over. Our inflation figure runs about 0.15 points hotter than theirs, and across 126 years that compounds into something visible: they state that a 1900 dollar has the purchasing power of roughly $38 today, where our CPI-U series gives $44. And our bill figure is not comparable to theirs at all. Theirs starts in 1900 and so captures the deflationary pre-1933 decades, when bills earned a genuine real return; ours starts in 1934 and captures only the era after that. Same instrument, different centuries — which is exactly the regime split described in §2.

The finding most people's mental model gets wrong

Look at the top two rows of the table again.

The gap is 4.16 percentage points a year. Compounded over 126 years, it is a factor of about 152.

The number quoted in every news bulletin — "the market is up 12% this year" — is the price line. It is the lower one. Dividends, reinvested, account for well over half of the real long-run return, and the entire visual difference in the chart above.

There is a second, subtler point buried in the same two rows. Price-only real return was 2.71% a year — barely above the 1.52% of a government bond, and derived from an era in which dividend yields were far higher than they are now. An investor who owned equities purely for price appreciation, spending the income, historically earned something closer to a bond-like real return while carrying full equity risk.

Where each decade actually landed

Annualized real returns, by decade:

Three of the twelve complete decades produced a negative real return in equities: the 1910s, the 1970s and the 2000s. A decade is not a short holding period. It is most of the time a person spends saving for something specific.

Two of those three — the 1910s and 1970s — were the two highest-inflation decades in the record. The third, the 2000s, was not an inflation event at all; it was two crashes with a low starting yield. Which is a useful correction to a tidy story: high inflation is one way to lose a decade, but it is not the only way.

Note also the 1930s row. Equities returned +2.11% real across the decade containing the worst crash in the record — because the crash happened in 1929, prices fell 2% a year through the decade, and the 1932–1937 recovery was violent. Decade boundaries are arbitrary, and they can hide as much as they reveal. The companion article measures the damage properly, peak to trough to recovery.

5. What this establishes, and what it does not

Established, on this data: the US price level rose about 3% a year for 126 years and about 3.45% a year since 1933; short-term government bills failed to preserve purchasing power over 92 years; long bonds preserved it and little more; equities compounded at roughly 6.9% a year in real terms, and more than half of that came from reinvested dividends.

Not established: anything about the future. This is one country over one span — the country that happened to produce the best-documented and among the most successful equity markets of the twentieth century. Reading a 126-year US average as an expectation ignores exactly that selection. The Dimson–Marsh–Staunton work exists partly to make this point across twenty-plus markets, several of which went to zero at some stage.

Also not established: anything about what any particular reader ought to hold. These are historical measurements published for general readership, identical for everyone reading them.

6. Methodology and known limits

Every figure above comes from scripts/build_inflation_study.py in this site's repository, which writes data/inflation-study.json. Nothing is typed in by hand. The known limits, in order of how much they matter:

1. Shiller's monthly price is an average of daily closes, not a month-end close. This smooths the series. Drawdown depths computed from it are understated — materially so for one-month collapses like October 1929 and October 1987. Annualized returns over long spans are barely affected; the companion article, which is largely about drawdowns, returns to this.
2. No official CPI exists before 1913. For 1900–1912 we use the Warren–Pearson-derived index carried in Shiller's dataset, rescaled to splice cleanly onto BLS CPI-U at January 1913 (the required scaling factor came out to 1.000000, which is a good sign for the splice). That early series is interpolated to monthly from annual observations, so month-to-month movement in 1900–1912 is smoother than reality.
3. October 2025 CPI does not exist. Discussed in §1. Annual figures use the BLS's own 11-month average; the monthly series interpolates.
4. Bond returns are an approximation. We price a 10-year par bond each month off the long yield and reprice it a month later — a constant-maturity approximation, not an actual bond index. Before 1953 the yield itself is a spliced long-government series rather than the modern 10-year constant maturity.
5. Monthly dividends are the trailing-twelve-month figure divided by twelve, the standard convention for this dataset. It smooths dividend timing within a year.
6. Everything is pre-tax and pre-cost. No fees, no commissions, no taxes, no bid-ask spread. Real-world outcomes were lower — for taxable investors in the high-inflation decades, considerably lower, because tax is levied on nominal gains.

References

Data sources

• BLS Consumer Price Index (CPI-U, CUUR0000SA0) — monthly from 1913; bulk time-series file
• BLS CPI home and release schedule — methodology and publication calendar
• Robert J. Shiller, online data — the ie_data.xls workbook, monthly from 1871
• FRED — 3-Month Treasury Bill Secondary Market Rate (TB3MS)
• FRED — Consumer Price Index for All Urban Consumers (CPIAUCNS) — used to cross-check the BLS file

Cross-checks and further reading

• UBS Global Investment Returns Yearbook — Dimson, Marsh and Staunton's long-run returns across 35 markets; the standard reference against which the figures here were validated. The US figures quoted above are from the 2026 public summary edition
• Burton G. Malkiel, A Random Walk Down Wall Street: The Best Investment Guide That Money Can Buy, 50th anniversary edition (W. W. Norton, 2023) — chapter 1 is the source of the break-even framing this article set out to measure. The Chinese edition of this article quotes the Commonwealth Publishing translation of the same passage
• Robert J. Shiller, Irrational Exuberance — the dataset's origin and documentation

Disclaimer

This article is published by Elnath Finance Academy for general informational and educational purposes. It is written for all readers and is identical for everyone. It is not investment advice, not a personalized recommendation, and not an offer or solicitation to buy or sell any security. Elnath Finance Academy is not a registered investment adviser.

All figures are historical measurements computed from the public sources named above. Historical returns describe the past. They are not a forecast and not a guarantee of any future result. The single most important limitation of any long-run study of this kind is that it examines one market that survived; markets that did not survive are absent from the average by construction.

All returns are gross of fees, costs and taxes. Index and company names are the property of their respective owners and are used here for identification and educational reference only. Elnath Finance Academy is not affiliated with, sponsored by, or endorsed by any index provider.

Investing involves risk, including possible loss of principal. Consult a qualified professional where appropriate.