THE EVIDENCE
Did heavy investors really lag?
Key points
- 37 industries, 1947–2019: after years when an industry invested more than in most of its past 20, its shares lagged by 1.3% a year against the market over the next five years; after lean years they beat by 0.7% a year.
- The gap was clear in 1970–89 (2.8% a year) and 1990–2019 (2.1% a year). In 1947–69 it was small and could have been chance.
- It held for resource and heavy industries and, more strongly, for technology (chips, telecoms, software, IT services). It did not work for utilities, whose returns are set by regulators.
- Checked company by company for 2011–2020 (SEC filings), it held within industries: clear for gold miners, copper miners, oil and gas, power equipment makers, chipmakers with fabs, absent for utilities (the buyers), hyperscalers (data centres).
How we measured it
- Investment: each industry’s spending on equipment and structures (BEA), divided by depreciation of the same assets. R&D and software are left out; putting them in changes little (below).
- ‘More than usual’: that year’s ratio compared with the industry’s own previous 20 years. The top third counts as heavy, the bottom third as light. Different industries have different normal levels, so they are not compared with each other directly.
- Share returns: Kenneth French’s value-weighted US industry portfolios, relative to the whole market, over the 60 months from October of the following year (BEA publishes a year’s figures the following autumn, so nothing is used before it was known).
- 37 pairs of BEA and French industries, one-to-one, so the same investment figures are never counted twice.
Source: BEA / French Data Library
The result, period by period
| Period | After heavy (per year) | After light (per year) | Gap | Industries where heavy was worse | Years (industries compared with each other) | Random pairings doing as badly |
|---|---|---|---|---|---|---|
| 1947–2019 | -1.3%920 | +0.7%974 | -2.0% | 26 / 37 | 51 / 73 | 0.1% |
| 1947–1969 | -0.9%412 | +0.5%136 | -1.4% | 16 / 30 | 11 / 23 | 10.7% |
| 1970–1989 | -1.2%200 | +1.6%350 | -2.8% | 24 / 34 | 17 / 20 | 0.0% |
| 1990–2019 | -1.9%308 | +0.2%488 | -2.1% | 24 / 36 | 23 / 30 | 0.0% |
Small numbers under each figure: the number of industry-years. ‘Random pairings’: we shifted each industry’s investment history by five or more years against its share returns, 1,000 times, keeping each series’ own ups and downs. The share of those fake pairings with a gap at least as bad as the real one is how often chance alone would produce the result.
Change the definition, same answer?
The gap (heavy minus light, per year) under different choices. A negative number means heavy investment was followed by worse returns.
| 1947–2019 | 1947–1969 | 1970–1989 | 1990–2019 | |
|---|---|---|---|---|
| As above | -2.0% | -1.4% | -2.8% | -2.1% |
| Include R&D and software in investment | -2.0% | -1.6% | -2.5% | -2.0% |
| Compare with the past 10 years, not 20 | -1.6% | -0.5% | -2.6% | -1.4% |
| Next 3 years instead of 5 | -2.0% | +0.2% | -4.0% | -2.1% |
| Next 7 years instead of 5 | -1.5% | -1.6% | -1.7% | -1.5% |
Every definition gives a negative gap for the whole period and for 1970 onwards. The 1947–69 result flips sign with a 3-year window, another sign that the early period is not reliable.
Which kinds of industry?
| 1947–2019 | 1947–1969 | 1970–1989 | 1990–2019 | |
|---|---|---|---|---|
| Resources and heavy industry (15) | -2.2%10/15 | -0.8%7/14 | -0.4%8/14 | -4.7%12/15 |
| Technology (5) | -6.6%4/5 | -6.4%3/4 | -8.2%4/4 | -4.2%4/4 |
| Consumer, finance and services (17) | -0.7%12/17 | -0.7%6/12 | -3.8%12/16 | +1.4%8/17 |
Small numbers: industries where heavy investment was followed by worse returns / industries with both kinds of year. Technology shows the largest gaps: chips, telecoms and software build capacity fast, and prices fall fast when it arrives. Resource and heavy industries show it most clearly since 1990.
Industry by industry
Show the numbers
| Industry | Gap (%/yr) |
|---|---|
| Software | -24.8 |
| Textiles | -10.1 |
| Healthcare services | -7.8 |
| Aircraft, ships & rail | -6.0 |
| Fabricated metal | -5.4 |
| Chips & electronics | -5.1 |
| Food | -4.7 |
| Telecoms | -4.7 |
| Apparel | -4.6 |
| Paper | -4.6 |
| IT & business services | -4.2 |
| Oil & gas | -3.7 |
| Mining | -3.2 |
| Chemicals | -2.8 |
| Beverages & tobacco | -2.4 |
| Real estate | -2.4 |
| Securities & asset management | -2.2 |
| Entertainment | -2.1 |
| Restaurants | -2.1 |
| Banks | -1.8 |
| Transport | -1.8 |
| Retail | -1.7 |
| Insurance | -1.6 |
| Rubber & plastics | -1.4 |
| Electrical equipment | -0.9 |
| Printing | -0.1 |
| Household & other | 0.0 |
| Medical equipment | 0.0 |
| Steel & metals | 0.2 |
| Utilities | 0.2 |
| Autos | 0.5 |
| Machinery | 0.6 |
| Wholesale | 1.7 |
| Building materials | 1.8 |
| Agriculture | 2.1 |
| Personal services | 6.5 |
| Construction | 7.1 |
26 of 37 industries point the same way. The exceptions: household & other, medical equipment, steel & metals, utilities, autos, machinery, wholesale, building materials, agriculture, personal services, construction.
A second check: company by company, 2011–20
We repeated the test with the companies on this site: each company’s capex over depreciation from its SEC filings, compared with its own earlier years, and its shares over the next five years against the S&P 500 (2011–2020, 297 company-years).
| Industry | After heavy (per year) | After light (per year) | Gap |
|---|---|---|---|
| Gold miners | -5.2%7 | -1.4%20 | -3.8% |
| Copper miners | +1.3%6 | +8.1%3 | -6.8% |
| Oil and gas | -11.9%11 | -6.6%51 | -5.3% |
| Power equipment makers | -2.3%6 | +9.8%11 | -12.1% |
| Utilities (the buyers) | -3.3%27 | -4.5%23 | +1.2% |
| Chipmakers with fabs | +2.6%30 | +6.3%11 | -3.6% |
| Hyperscalers (data centres) | +5.3%12 | +5.4%10 | -0.1% |
| All | -1.2%99 | -1.7%129 | +0.4% |
Within an industry, it shows up clearly (2% a year or more) for gold miners, copper miners, oil and gas, power equipment makers, chipmakers with fabs, and not for utilities (the buyers), hyperscalers (data centres). Pooled across all industries the gap is +0.4% a year, because industries with naturally different returns get mixed together (the industry test avoids this by comparing each industry with its own past). Three reasons to trust this check less than the industry test: only ten years and one cycle; individual companies move for many reasons besides their own spending; and the list is of companies that exist today, so the ones that went bust after over-investing are missing. The capital cycle is also a statement about an industry’s total supply, not about one firm.
Source: SEC EDGAR / Yahoo Finance
What this does and does not show
The five-year windows overlap, so 75 years contain only about 15 independent five-year stretches per industry. That is why we lean on the random-pairing test rather than on the raw count of observations. What the test supports is a tendency, a couple of percent a year on average, not a timing rule: heavy investment can go on for years while shares keep rising, as in 1998–2000. So this site uses the ratio as a warning light next to each industry, and checks it against what the demand-side sites expect.
A related, well-documented finding is that individual companies whose total assets grow fastest tend to have lower returns afterwards (Cooper, Gulen and Schill, 2008). The Capital Returns book itself admits that its editor left out letters whose calls turned out wrong, which is one reason to test the idea again rather than take the anecdotes on trust.
Source: Cooper et al. (2008) / Capital Returns
This page organises information; it does not recommend buying or selling anything. Data belongs to each publisher and may contain errors or delays. Make your own investment decisions.
Updated 30 Sep 2026 01:15 JST · company filings to Aug 2026 · industry history to 2024