Strange things that happen in crises

The weirdness meter

How odd is odd? Each card is one oddity: the biggest move a series made in a crisis, at its own pace — a trading day, a week, a month or a quarter — set against the average move of the same series in all the years before that crisis. A card at ×1 would be an ordinary day at the office; each notch on the gauge is ten times more.

The champion is the Jobless claims card: ×249 the average week, or 170σ. Even the calmest card, Productivity, moved ×6.1 its usual pace. 9 of the 12 moves are so far out that, if economic data followed a bell curve, not one of them would be expected in the age of the universe. They do not follow bell curves — which is rather the point of this article.

Oil below zero

On April 20, 2020 the price of a barrel of West Texas Intermediate, the US benchmark crude, closed at −$36.98. The minus sign is not a typo: sellers paid buyers to take their oil. It is the only close below zero since the data begin in 1986. The trading day before, the barrel cost $18.31 — a fall of $55.29 in one session, the biggest one-day fall on record. The next day it was back at $8.91: $45.89 higher, the biggest one-day rise on record, too.

Why? The pandemic had emptied roads and skies, and the storage tanks in Cushing, Oklahoma, where WTI is delivered, were filling up. The spot price follows the nearest futures contract, and the May contract expired the next day: whoever still held one would have to take real barrels with nowhere to put them. Paying someone to take them was the cheaper option. On January 6, 2020 the same barrel had cost $63.27.

2008: $145.31 to $30.28 in 173 days

The other oil oddity ran the other way. On July 3, 2008 WTI closed at $145.31 a barrel. That is still the highest close on record. By December 23, 2008, after the failure of Lehman Brothers had turned a housing bust into a global panic, it was $30.28: −79% in 173 days. The world had not stopped needing oil; the price had simply gone from pricing a shortage to pricing a slump.

The week the seismograph broke

Every Thursday the Labor Department counts new claims for unemployment insurance — the fastest gauge of layoffs America has. In 2019 a week brought 218K on average. The worst week of any crisis before 2020 came in October 1982: 695,000. Then the week to March 28, 2020 came in 3.03M higher than the week before, and the week to April 4, 2020 brought 6,137,000 — 8.83 times the old record.

Claims stayed above the old record for 51 weeks, the last of them the week to March 13, 2021; in those weeks Americans filed 76.1M first claims. On a chart drawn to fit every earlier crisis, 2020 does not fit at all: the needle leaves the paper. The close-up shows those weeks on a scale of their own.

The unemployment rate made the same leap in monthly steps: from 3.5% in February 2020 to 14.8% in April 2020, 10.4 pp in a single month. The biggest one-month rise before that was 1.3 pp, in October 1949. The number of unemployed grew 4.01 times, from 5.75M to 23.1M. No month since the data begin in 1948 saw a higher rate.

The strangest twist is in how long people had been out of work. In June 2010 the typical unemployed American had been looking for 25.2 weeks, the longest since the series began in 1967. In April 2020, the month unemployment hit 14.8%, the typical wait was just 2.2 weeks, the shortest on record. Not because jobs were easy to find: almost everyone counted had lost a job in the previous few weeks.

Everyone saved at once

The personal saving rate is the part of income after taxes that households do not spend. Since 1959 the median month saw 8.2%, and the highest before 2020 was 17.3%, in May 1975. In April 2020 it was 31.8%: 3.88 times the median month. Nobody had suddenly turned thrifty. Shops, restaurants and travel closed while incomes held up, propped up by stimulus payments and extra unemployment benefits.

It happened twice: in March 2021, as another round of stimulus checks arrived, the rate jumped to 26.2%. The lowest month of the whole series, by contrast, came near the top of the housing boom: 1.4% in July 2005.

2020, month by month

The strange things came in a sequence. Read the rows from top to bottom on one shared time axis: claims peaked in the week to April 4, 2020, unemployment in April 2020, the saving rate in April 2020; stocks bottomed in March 2020 (−19% from February 2020, monthly average), and oil went below zero on April 20, 2020.

Then the money arrived and prices went strange. M2 peaked 41% above its Feb 2020 level in March 2022; in February 2022 used cars cost 54% more than in February 2020, after airfares had bottomed at −31% in May 2020. Eggs, hit by bird flu, peaked at $4.82 in January 2023.

Output fell, productivity jumped

One more oddity hides in the quarterly numbers. In Q2 2020 real GDP fell 7.9%, the biggest quarterly change since the data begin in 1947. In the same quarter output per hour of work rose 4.9%, also the biggest since 1947. Both can be true only if hours worked fell even faster than output — and they did. The usual explanation: the jobs lost first were, on average, the lower-paid, less productive ones.

Used cars, airfares and eggs

In the twelve months to June 2021 used car prices rose 45.2%, the fastest since the index began in 1953; the median year since 1953 added 1.5%. From February 2020 to February 2022 they rose 54.3%, while consumer prices as a whole rose 9.7%. A used car is supposed to lose value every year; stories of people selling a car for more than they had paid were common. The usual explanation: a shortage of computer chips held back the production of new cars, and buyers turned to used ones.

Airfares did the opposite, then overshot: down 30.9% from February 2020 to May 2020, then up 74.0% from that low to June 2022 — 20.2% above where they had started. In the twelve months to October 2022 fares rose 43.0%, the most since the index began in 1989.

Eggs: two panics, one of them about hens

Eggs made the meter twice over. In April 2020, as shoppers stocked up for lockdown, the average price of a dozen jumped from $1.52 to $2.02: +32.4% in one month, the biggest monthly rise since the series began in 1980. The all-time record came later and from a different crisis — bird flu, which forced farms to cull their flocks: $6.23 a dozen in March 2025, 4.46 times the 2019 average of $1.40. Even after inflation, no month since 1980 was dearer.

BLS on its average prices: Average price paid in US cities, from the CPI survey; BLS publishes these to show price levels, but warns they are less precise than the CPI itself and some months are missing. Months without a published value are left out (2025-10).

The money stock: a jump, then a shrink

M2 counts the money Americans hold in cash, checking and savings accounts and similar deposits. In April 2020 alone it grew 6.4%, the most in any month since the data begin in 1959. In the twelve months to February 2021 it grew 26.8%; the fastest year before 2020 had been 13.8%, in the twelve months to February 1976. The jump is not an accounting trick: the Federal Reserve changed the definition of M1 in May 2020, but M2 already included the deposits that were moved.

Then came something the data had never shown: the money stock shrank. For 15 months in a row, from December 2022 to February 2024, M2 was smaller than a year earlier, by as much as 4.7% (Apr 2023), as the Fed raised rates and shrank its balance sheet.

Gold in panics: a safe haven, except when it is not

Since April 1968, when gold began to trade at a market price, the S&P 500's monthly average has fallen 15% or more from a high 8 times. Over the same months gold rose in 5 of those panics and fell in 3. The best showing was 1973–74: stocks −43%, gold +182%. The worst was 1980–82: stocks −19%, gold −46%.

And in the sharpest months of a panic, gold has often slipped too. In 2 of the 3 worst months for stocks since 1968 it fell: Oct 2008: stocks −20.4%, gold −2.8%; Mar 2020: stocks −19.1%, gold −0.3%; Nov 1987: stocks −12.6%, gold +0.5%. The usual explanation: when everything is being sold to raise cash, gold gets sold as well.

Homes fell everywhere at once

In the 53 years of monthly data before July 2006, US home prices as a whole barely ever fell: the deepest fall was −3.1% (1990–91). The common belief was that prices could drop in one city, never across the country. Then, from July 2006 to February 2012, Shiller's national index fell 27.4%. His yearly figures, which go back to 1890, show a deeper fall only once: −30.5% in 1925–1933, the Great Depression. Prices were back at their 2006 peak only in January 2017, 10.5 years later.

The rescue was strange too. From December 4, 2008 to December 16, 2015 the federal funds rate — what banks charge each other overnight — stayed at 0.25% or less: 7.0 years of nearly free money. It went back there from March 16, 2020 to March 16, 2022, another 2.0 years.

Stocks marked the panic in their own way: in October 2008 the S&P 500's monthly average fell 20.4%, the biggest fall since April 1932.

The famous ones we cannot measure

Some of the best-known crisis oddities have no reliable data series, so they get no numbers here — only a label saying what kind of claim each one is.