Key takeaways
- Over the 100 years to July 2026 the eleven named US industries returned between 8.80% and 11.57% a year, a spread of 2.77 points that compounds into 12.35 times the ending wealth.
- Every one of the eleven industries has finished a decade in the top three and another in the bottom three. Business equipment returned 29.5% a year in the 1990s and -6.2% in the 2000s.
- Decade-to-decade rank correlation averaged -0.22 across the nine transitions since the 1930s, and it was negative in six of them.
- Drawdowns separate the industries far more than returns do. Since 1950 business equipment fell 79.59% at its worst and utilities fell 42.58%.
- Business equipment was 43.19% of US market value in July 2026, against 27.29% at the end of 2019 and 9.46% at the end of 1979.
A hundred years of sector returns fit inside a three point band
Which sector has paid best over the long run, and what did you have to sit through to collect it? Over the 100 years from July 1926 to July 2026, healthcare led the eleven named US industries in Kenneth French's data library, at 11.57% a year. Telecoms trailed at 8.80%. Everything else landed between the two. The US market as a whole returned 10.35% a year across the same months.
A 2.77 point spread sounds narrow. It isn't, because it compounds. One dollar put into healthcare in July 1926 was worth 57,465 by July 2026. The same dollar in telecoms was worth 4,653. That's 12.35 times the money out of a gap of under three points a year. These are gross figures, before fees, spreads and tax, and across a span no individual investor lives through.
The chart shows that hundred-year record for all eleven industries. It's the least interesting evidence in this piece, because it averages away the route. Sector returns by decade tell a different story, and that story is the rest of this piece.
The eleven industries here aren't the eleven sectors you can buy
These eleven come from the Fama-French 12 industry portfolios. Every NYSE, AMEX and Nasdaq stock is assigned to one bucket at the end of each June on its four-digit industry code, and the buckets are value-weighted. Monthly returns run from July 1926 to July 2026. Eleven of the twelve carry an economic name: consumer nondurables, consumer durables, manufacturing, energy, chemicals, business equipment, telecoms, utilities, shops, healthcare and finance. The twelfth is a residual labelled Other, holding mines, construction, transport, hotels, business services and entertainment. It's left out here, because "everything else" isn't a sector.
The eleven sectors you can actually buy are a different set. A prospectus filed with the Securities and Exchange Commission on 24 April 2026 describes one index as "one of eleven Select Sector Indexes", each holding the members of the S&P 500 that the Global Industry Classification Standard assigns to that sector. Those definitions were built for modern markets and get revised. They don't reach back to 1926. So the century-long record has to be read through the older industry portfolios, and the mapping between the two schemes is loose. Business equipment, defined in the source data as computers, software and electronic equipment, is not the same thing as the information technology sector.
Sector returns by decade, the 1930s to the 2020s
Here is the table. Each cell is the annualised value-weighted total return for that industry over that decade, in percent a year. The 2020s column covers January 2020 to July 2026, which is six and a half years rather than ten, and rows are ordered by hundred-year return.
| Industry | 1930s | 1940s | 1950s | 1960s | 1970s | 1980s | 1990s | 2000s | 2010s | 2020s |
|---|---|---|---|---|---|---|---|---|---|---|
| Healthcare | 2.7 | 9.0 | 22.3 | 12.5 | 3.8 | 19.1 | 18.6 | 2.7 | 14.6 | 9.8 |
| Business equipment | -2.7 | 8.2 | 25.7 | 12.2 | 2.9 | 8.2 | 29.5 | -6.2 | 16.5 | 22.4 |
| Shops and retail | 1.8 | 12.3 | 14.2 | 11.3 | 3.2 | 21.2 | 17.0 | 2.1 | 15.8 | 14.4 |
| Manufacturing | -0.8 | 9.1 | 22.1 | 6.7 | 5.1 | 14.6 | 16.5 | 5.0 | 13.5 | 15.2 |
| Consumer nondurables | 2.9 | 9.2 | 12.3 | 10.6 | 5.4 | 25.6 | 12.6 | 7.4 | 12.1 | 6.6 |
| Consumer durables | 6.1 | 10.4 | 21.6 | 7.9 | 4.6 | 16.2 | 12.8 | -1.8 | 10.6 | 21.1 |
| Energy | -1.1 | 13.2 | 18.8 | 9.2 | 13.7 | 15.0 | 11.6 | 11.1 | 2.9 | 15.8 |
| Chemicals | 5.7 | 7.8 | 18.3 | 3.7 | 6.9 | 17.1 | 15.4 | 4.5 | 11.7 | 3.7 |
| Finance | -5.7 | 13.1 | 15.7 | 10.5 | 5.2 | 17.3 | 18.6 | 0.1 | 13.6 | 12.4 |
| Utilities | -6.0 | 8.6 | 14.9 | 6.6 | 6.7 | 18.1 | 8.6 | 8.7 | 11.0 | 9.4 |
| Telecoms | 4.3 | 4.8 | 13.1 | 5.6 | 7.6 | 24.2 | 17.8 | -7.6 | 14.2 | 0.2 |
Read down any column and the range within a single decade is wide. Read across any row and the ordering keeps turning over. Those two facts together are the argument.
Every one of the eleven has led a decade and lagged another
Rank the industries inside each decade and count the top-three and bottom-three finishes. Every single industry has done both at least once. Business equipment managed five of each: top three in the 1950s, 1960s, 1990s, 2010s and 2020s, bottom three in the 1930s, 1940s, 1970s, 1980s and 2000s. It returned 29.5% a year through the 1990s and -6.2% a year through the 2000s.
Energy shows the same shape on a different clock. It returned 11.1% a year through the 2000s, 2.9% through the 2010s, and 15.8% in the 2020s to date. Telecoms was the second-best industry of the 1980s at 24.2% a year and the worst of the 2000s at -7.6%. Utilities, the quietest name on the list, was the worst of the eleven in the 1930s at -6.0% a year and the second best of the 2000s at 8.7%.
A rank correlation puts a number on the reshuffling. Spearman's rho compares one decade's ordering with the next, running from +1 for an identical order to -1 for a perfectly reversed one. Across the nine transitions from the 1930s it averaged -0.22, and it was negative in six of them. The two deepest reversals were the 1990s into the 2000s at -0.75 and the 1960s into the 1970s at -0.71, with the 2000s into the 2010s at -0.60. Persistence turned up three times, at +0.31, +0.18 and, most recently and faintly, +0.02 from the 2010s into the 2020s.
Decade boundaries are arbitrary, so it's fair to ask whether the result is an artefact of starting at 1930. Shifting every window five years forward gives a mean of -0.21 across the eight transitions from 1935, which is the same answer. What the exercise cannot do is turn nine or ten observations into a large sample.
Sector drawdowns separate the industries far more than the returns do
Returns cluster. Falls don't. The table below gives the deepest peak-to-trough fall each industry took on month-end data since January 1950, beside its hundred-year annualised return. Both columns are percentages.
| Industry | Annualised return | Deepest fall | Peak to trough |
|---|---|---|---|
| Healthcare | 11.57 | -47.08 | Jul 1973 to Sep 1974 |
| Business equipment | 11.52 | -79.59 | Mar 2000 to Sep 2002 |
| Shops and retail | 10.96 | -57.54 | Dec 1972 to Sep 1974 |
| Manufacturing | 10.63 | -59.45 | Oct 2007 to Feb 2009 |
| Consumer nondurables | 10.54 | -51.93 | Dec 1972 to Sep 1974 |
| Consumer durables | 10.53 | -73.7 | Jun 2007 to Feb 2009 |
| Energy | 10.51 | -64.74 | Jun 2014 to Mar 2020 |
| Chemicals | 10.45 | -43.64 | May 2008 to Feb 2009 |
| Finance | 9.88 | -72.47 | May 2007 to Feb 2009 |
| Utilities | 9.22 | -42.58 | Nov 1972 to Sep 1974 |
| Telecoms | 8.8 | -72.04 | Mar 2000 to Sep 2002 |
Business equipment fell 79.59% from its March 2000 peak to September 2002. The worst utilities took since 1950 was 42.58%, from its November 1972 peak to September 1974. The market as a whole fell 50.31% between October 2007 and February 2009. So two industries 2.3 points apart on hundred-year return are 37 points apart on worst-case loss. That is the same distinction that makes volatility vs drawdown worth separating: a standard deviation and a worst year describe different experiences of the same asset.
Go further back and the numbers stop being useful for planning, though they set the outer bound. Finance fell 92.12% and business equipment 92.07% from their August 1929 peaks to May 1932, against 83.65% for the whole market to June 1932.
Single events show the dispersion more clearly than any average. Between April 2000 and September 2002, consumer nondurables gained 18.51% while business equipment lost 79.59%. From November 2007 to February 2009, finance lost 69.73% and healthcare lost 30.78%. In February and March 2020 together, energy fell 44.58% and healthcare fell 10.19%. In calendar 2022, energy gained 64.64% while consumer durables lost 55.67%. Each of those pairs sat inside one index return.
What a cap-weighted index does with an eleven-way split
The same file that carries the returns carries the number of firms in each industry and their average size. Multiply the two, divide by the total, and you get each industry's share of US market value. In July 2026 business equipment was 43.19% of it, spread across 574 firms. At the end of 2019 the same industry was 27.29%, and at the end of 1979 it was 9.46%. The three largest industries together were 63.56% in July 2026.
A cap-weighted fund takes that arithmetic as given. The semi-annual report of the SPDR S&P 500 ETF Trust, for 31 March 2026, breaks the fund down by industry: semiconductors and semiconductor equipment 14.5% of net assets, software 8.2%, interactive media and services 7.6%, and technology hardware, storage and peripherals 7.4%. Those four lines are 37.7% of the fund between them. The classification and the universe both differ from the French data, so the two numbers aren't the same measurement, but they point the same way. The related question of how far a rules-based cap helps is covered in sector concentration.
None of this is intrinsic to cap weighting. It's a description of one market at one moment. The FTSE Russell factsheet for the FTSE 100 at 31 August 2026 puts banks at 19.28% of the index and technology at 2.53%, with health care at 11.65% and energy at 10.43%. The five largest constituents are 33.37% of the index. A UK investor holding the home index carries a concentration of a completely different shape, and the top 10 index weight is where that usually shows up first.
The case against reading any of this as a sector rotation rule
Sector returns by decade are a record of what happened, not a rule about what happens next. The obvious reading is that reversal across decades is tradeable: hold whichever industry just had a bad ten years. There are three objections to that, and none of them is a technicality.
The first is the sample. Nine transitions is a thin basis for a mean of -0.22, and the same nine contain +0.31 and +0.18. The second is that a return series is not a strategy. These figures carry no dealing cost, no spread, no tax and no timing error, and anyone acting on them has to decide when a decade ends while it is still running. The third is that an industry is a code, not a business. The label persists while the companies inside it turn over completely, which is as true of the modern sector indices as it is of the 1926 data. That is also the pattern critics of the published factor premia point to, and the value premium is the best-documented case of a historical spread that shrank once people started trading it.
The strongest counter-argument runs in the other direction, and it deserves stating properly. The last two decades did not revert. Business equipment led the 2010s at 16.5% a year and leads the 2020s at 22.4%. If a 43.19% share of market value reflects durable economics rather than a repricing, then reversal is simply the wrong frame for the present, and a century of decade tables is a record of a world that no longer applies. The data can't settle that. It only measures what already happened, in one country, in a currency that isn't the pound, before any of the costs a real holder pays.
What would change the conclusion
If sector returns by decade started repeating their order rather than reversing it, the reading here would need replacing. If decade-to-decade rank correlation turned reliably positive over the next two or three decades, the reversal in this record would look like a period effect rather than a structural one, and ten draws from 1930 onward would read as a coincidence. If business equipment's share of US market value fell back toward its 2019 level of 27.29% without a matching fall in its earnings, the concentration would have been a valuation story. If it is still near 43.19% after a full recession, it wasn't.
The measure worth watching is narrower than any of those, and it updates every month: the gap between the best and worst industry inside a single decade. That gap was 13.6 points a year in the 2010s and stands at 22.2 points in the 2020s so far, against 8.8 points in the 1960s. Dispersion that wide is what makes the industry you happen to own matter more than a hundred-year average of eleven industries suggests it should.