Economic paradoxes

01John Maynard Keynes, 1936 · after Bernard Mandeville, 1714Holds up

The paradox of thrift

If every household tries to save more at the same time, spending falls, incomes fall — and the country can end up poorer without saving any more in total.

The idea is old — Bernard Mandeville's Fable of the Bees (1714) praised the spendthrift — but John Maynard Keynes made it famous in 1936. One family that spends less builds a cushion. A whole country that spends less takes away someone else's income: my spending is your paycheck. Recessions are the test, because that is when worried households pull back.

And they do. In 6 of the 9 recessions since 1960, the personal saving rate was higher during the recession than in the 12 months before it, while real GDP fell below its pre-recession peak. In 2008–09 the saving rate averaged 4.8% against 2.5% before (+2.3 points) and output shrank by 3.8%. In 1973–75 and 1990–91 the saving rate barely moved (less than 0.25 points either way). In 2001 real GDP never dipped below its level at the peak — a see-saw with nobody on the other end.

The steepest see-saw was 2020: the saving rate jumped 15.0 points, to 31.8% in April 2020, as output shrank by 9.1% — less thrift than lockdown, with shops shut and relief checks in the mail. Whether saving caused any of these slumps is a different question: the data show the see-saw, not who pushed.

Measured with: the personal saving rate, monthly since 1959 (BEA) · real GDP, quarterly (BEA) · recession dates (NBER).

02Robert Solow, 1987Holds up

The productivity paradox

Computers made every office faster, yet productivity growth slowed down just as they spread.

In 1987 the economist Robert Solow quipped: “You can see the computer age everywhere but in the productivity statistics.” He had a point. Output per hour in US business grew 2.7% a year from 1947 to 1973; from 1973 to 1995 — the years of mainframes, PCs and spreadsheets — only 1.5%.

Then the computers showed up in the numbers after all: 3.1% a year from 1995 to 2004, 2.1 times the pace before, as the web and new ways of working spread. And then the paradox came back: 1.6% a year in 2004–2019, the smartphone years. Since 2019: 2.2% a year to Q3 2026 — faster than the 1.6% of the years before; whether AI deserves the credit is too early to say.

Measured with: output per hour in the nonfarm business sector, quarterly since 1947 (BLS).

03William Stanley Jevons, 1865Holds in part

Jevons paradox

When engines get more efficient, fuel gets cheaper to use — and people may end up burning more of it, not less.

William Stanley Jevons spotted it in 1865, in The Coal Question: better steam engines had not saved Britain's coal, they had multiplied its uses. America's cars are the classic modern test. Since 1973, the year the fleet's fuel economy bottomed out, the average US motor vehicle went from 11.9 to 19.2 miles per gallon (+61%). Yet the country burned 102 billion gallons of motor gasoline in 1973 and 138 billion in 2024 (+35%).

So Jevons wins? Only in total. Per American, gasoline use fell 16%, from 483 gallons in 1973 to 405 in 2024, and it peaked back in 1978 at 510. Per vehicle, fuel used went from 850 to 577 gallons a year (−32%) while miles driven went from 10,099 to 11,071 (+10%). The efficiency did save fuel; more people, more cars and more miles spent the savings. Most studies of driving find the rebound real but much smaller than the saving — a Jevons effect, not a full Jevons paradox.

The dashed curves show what efficiency alone would have done: had Americans driven no more miles in total than in 1973, today's fleet would burn about 63 billion gallons a year — not 138. (Right panel: no more miles per person.)

Measured with: fuel economy and miles of all US motor vehicles, and motor gasoline used, annual since 1949 (EIA, compiled from the Federal Highway Administration) · US population (BEA).

04Edward Simpson, 1951Holds up

Simpson's paradox

A total can move one way while every one of its parts moves the other — when the mix of the parts shifts.

The statistician Edward Simpson described it in 1951 (Karl Pearson and Udny Yule had noticed it decades earlier). American paychecks offer a crisp example. Take the median weekly earnings of full-time workers aged 25 and over, after inflation, in four education groups: no high-school diploma, high school, some college, and a bachelor's degree or more.

From 2000 to 2012, the median of every one of the four groups fell — from −1.9% (bachelor's or more) to −5.7% (some college). The median of all of them together rose 0.4%. No arithmetic trick: the share of full-time workers with a degree grew from 31% to 38%, and graduates are the best-paid group, so the mix lifted the total while every group slipped. The reversal shows for 5 end years in all, from 2007 to 2014.

Over the whole span to 2025, the overall median changed +11.4% after inflation — more than 3 of the 4 groups, because the mix kept moving toward graduates: from 31% of full-time workers to 47%.

Measured with: median usual weekly earnings and the number of full-time wage and salary workers 25 and over, by education, annual since 2000 (BLS, Current Population Survey) · consumer prices, CPI-U (BLS).

05Rajnish Mehra & Edward Prescott, 1985Holds up

The equity premium puzzle

Stocks are riskier than government bonds, so they should pay more — but they paid far more than any reasonable fear of risk can explain.

In 1985 Rajnish Mehra and Edward Prescott asked how much more stocks should earn than safe government debt if investors dislike risk the way standard economic models assume. Their answer: at most 0.35 percentage points a year. History disagrees. In an average calendar year from 1872 to 2025, stocks with dividends beat 10-year Treasury bonds by 6.2 points — about 18 times the model's limit. Stocks came out ahead in 100 of those 154 years.

Compounded, the gap is enormous: from January 1871 to October 2026 stocks returned 9.4% a year and bonds 4.5%. A dollar became $1.19M in stocks and $1,009 in bonds; after inflation, 7.1% against 2.4% a year. Over the years Mehra and Prescott studied, 1889–1978, stocks beat these bonds by 4.9 points a year compounded; they measured against short-term Treasury bills, which paid less, so their gap was wider still.

Decades of explanations — rare disasters, sticky habits, loss aversion, and the luck of measuring the one market that became the world's largest — have narrowed the puzzle without closing it. The scoreboard of 20-year periods shows how lopsided the race was: stocks came out ahead in 98% of all 20-year stretches since 1871. Bonds won 30 of 1630, all starting in the 1870s, 1910s, 1920s, 1980s and 2000s; the worst for stocks began in June 1912: 2.4% a year against 4.4% for bonds.

Measured with: the S&P Composite with dividends reinvested and a 10-year Treasury bond with coupons reinvested, monthly since 1871 (Robert J. Shiller's data).

06named by Keynes, 1930Faded

Gibson's paradox

Long-term interest rates should follow inflation. For a century under the gold standard, they followed the level of prices instead.

Keynes named it in 1930 after Alfred Herbert Gibson, who had charted it in British data, and counted it among the best-established facts in economics — and the least explained. Irving Fisher's theory says lenders ask for more when prices are rising, so rates should track inflation. Under gold they tracked the price level.

US data show it plainly. From 1872 to 1913, the long-term rate moved with the price level — a correlation of +0.82 — and not with inflation (−0.19). Prices fell 48% from 1871 to 1896 and the rate fell with them, from 5.3% to 3.5%; then prices rose 53% to 1913 and the rate climbed back to 4.3%. Robert Barsky and Lawrence Summers (1988) traced it to gold itself: when real interest rates were high, holding gold was less attractive, its value sank and prices measured in gold rose.

07Alan Greenspan, 2005Holds up

Greenspan's conundrum

When the Fed raises short-term rates, long-term rates usually rise too. In 2004–2006 they barely moved.

From June 2004 to June 2006 the Federal Reserve raised its target rate at every meeting. The effective federal funds rate went from 1.03% in June 2004 to 5.24% in July 2006 (+4.21 points). The 10-year Treasury yield went from 4.73% to 5.09% (+0.35): the short rate moved about 12 times as much as the long one.

In February 2005, with the funds rate at 2.50% and the 10-year yield lower than when the hikes began (4.17%), Fed chairman Alan Greenspan told Congress the bond market's behavior was “a conundrum”. By June 2005 the 10-year yield was down to 4.00% while the funds rate had reached 3.04%. Suspects named since: foreign central banks and oil exporters buying Treasuries (Ben Bernanke's “global saving glut”), tame inflation expectations and a shrinking premium for lending long. Cheap long-term money is often blamed for feeding the housing boom that ended in 2008.

Measured with: the effective federal funds rate (Federal Reserve, H.15) and the 10-year Treasury yield (US Treasury), monthly averages of daily rates.