← Tools

Recession Explorer

· Methods

Explores U.S. recessions individually and in comparison with each other, and looks at where the economy stands today.

The National Bureau of Economic Research (NBER) defines a recession as “a significant decline in economic activity that is spread across the economy and lasts more than a few months.” By its dating there have been — U.S. recessions since 1948, lasting an average of — months and ranging from — months in 2020 to — months in 2007–09. The Recession Explorer lets you examine the recessions individually and in comparison with each other, and see where the economy stands today compared with past recessions. (The NBER calls a recession well after it begins, and sometimes after it has ended.)

By default, the changes in each recession are measured against a baseline: the trend of the previous ten years. A loss of 1 million jobs, say, is a bigger deal when jobs had been growing quickly in the years before a recession than when they had been growing slowly. You can instead take the baseline from the previous five years or from peak to peak, or use no baseline at all and compare the actual changes.

Data through —; the page refreshes on weekday mornings.

Recession
Compare with
The average and range
Look ahead
Link

Warning: these are the latest published figures, and many of them will be revised. Past recessions were sometimes clear only well after they began, in part because the first estimates were later revised down. For example, real GDP in the first quarter of 2008, as the 2007–09 recession began, was first reported as rising at a 0.6 percent annual rate; it now shows a fall at a — percent annual rate.

Where today stands

At a glance

Output

Jobs

Unemployment

What slowed

The pieces of GDP

Show

Inventories and net exports are missing here because they cross zero and cannot be indexed; they appear in “What slowed” as contributions to GDP.

Any series

Pick any series on the page. The first chart follows it from a year before the peak; the second ranks every recession on it.

Rank the recessions by

The Compare tab ranks the recessions on much more: their length, the labor market, each kind of spending, the recovery, stocks and the Fed. Rank the recessions →

The six monthly indicators

The National Bureau of Economic Research’s dating committee names six monthly indicators it watches when it dates a peak or a trough: real personal income less transfers, payroll jobs, household-survey employment, real consumer spending, real manufacturing and trade sales, and industrial production. Each panel has its own scale.

Employment-population ratios, by group

The share of each group with a job. The series for all groups by age and sex begin in 1948; White in 1954; Black in 1972; Hispanic in 1973; the four education groups (ages 25 and over) in 1992. A panel is blank where a group’s series starts too late for the recession picked, and where fewer than three other recessions have the series each of them is drawn as a thin line instead of a band. Each panel has its own scale.

Compare the recessions

Every recession on one baseline: the baseline picked above redraws every chart and table.

On this tab: Rank the recessions · Every recession on one chart · The scorecard · What slowed GDP (contributions, shares, the big pieces, each component over time, then and now, quarter by quarter, thirteen pieces, the numbers)

Rank the recessions

One chart, many rankings: how long and how deep, the labor market, which spending fell, how fast it recovered, stocks and the Fed. A recession with no reading is listed at the foot, with the reason.

Order
Depth at

Every recession on one chart

Each line is one recession, followed from a year before its peak and named at its end; the colors group the recessions by date. “Look ahead” in the control box sets how far the chart runs.

The scorecard

Every recession on one line, where a column heading sorts by that column; or head to head, a few recessions side by side on every criterion. Depth is read where the switch says: at each series’ low point (its worst reading under the baseline chosen within three years of the peak, or before the next recession began) or at the trough. “Back” is how long the series took, counting from the peak, to return to its level at the peak.

Show
Depth at

Best of the columnWorst of the column

What slowed GDP: the contributions in full

Every recession since 1953, broken down by who stopped spending: consumers, businesses investing in equipment and buildings, home building, inventories, government and trade. The default baseline compares each piece’s contribution to GDP during the recession with what it contributes on trend, so the pieces add up to how far the economy fell short of its usual pace. The baseline picked above redraws the four tiles and every chart and table here; the written findings, the Bottom line included, use the default baseline, the 10-year trend. The 1948–49 recession joins only under “Actual”.

Bottom line, against the 10-year trend (the default baseline)

  • Measured against trend, the average recession since 1953 took — points off GDP, against an average actual change in GDP of — percent. A recession costs more than the fall itself, because the economy would otherwise have kept growing.
  • Consumption was the largest single drag in — of the — and — percent of the combined shortfall, and its share has grown: — percent in the seven recessions from 1953 to 1982, — percent in the three from 1990 to 2009, and — percent in 2020.
  • Investment of all kinds (business, housing and inventories) was a bigger drag than consumption in — of the —; —. Consumption, business investment and inventories subtracted in every one.
  • Government was a large drag in two recessions, both through defense: 1953–54, which began in the month of the Korean War armistice and is the one recession where government was the largest drag of all, and 1969–70, as the United States drew down its forces in Vietnam. In the other nine it moved the total by less than a point either way.
  • Net exports cushioned the fall in — of the —, mostly because imports fell; —.
  • The main findings survive reasonable changes in method. Against the five-year trend, consumption is still the largest drag in — of the —; against the peak-to-peak trend, in —. The actual change, which subtracts nothing, tells a different story (consumption is the largest drag in only —), but for a bad reason (see the baselines).

Contributions to each recession

Each column is one recession. Pieces below zero pulled GDP down; pieces above zero held it up. The dot and number give the total.

Show

The deepest shortfalls before 2020, 1973–75 (— points) and 2007–09 (—), were broad: every private component fell well short of trend. 2020 was mostly consumption (— of — points), and most of that was services (—). 1981–82, when the federal funds rate stood near its record high (— percent in —), is the outlier: inventories (—), business investment (—) and a drop in exports (—) did the damage, while consumption barely slowed (—). In 1953–54 government did most of it (—).

Per year divides each total by the window’s length in years, which compares how fast the economy was slowing rather than how much ground it lost. Recessions ran from two quarters (1980, 1990–91, 2020) to six (2007–09).

Shares of each recession

Each bar scales one recession’s total to 100 percent. Segments to the right of zero added to the shortfall; segments to the left offset it, so whenever something cushioned the fall, the right-hand segments add up to more than 100 percent.

Consumption was the single largest drag in — of the —; the exceptions are —. Net exports offset part of the shortfall in — of the — (—); in 2007–09 they offset — percent of it.

The big pieces

The same numbers in three groups. The first view sets consumption beside investment of all kinds (business, housing and inventories) and government plus net exports. The second splits private final demand (consumption plus business and housing investment) from the inventory swing and from government plus net exports.

Group by

Investment of all kinds was a bigger drag than consumption in — of the —. Government plus net exports usually cushioned the fall; it added to the shortfall only in 1953–54 and 1969–70 (defense), and in 1957–58 and 1981–82 (a drop in exports).

Each component over time

One panel per component, on a common scale, with the recessions in order (’53 is the 1953–54 recession, ’20 is 2020, and so on). This shows how each piece’s role has changed.

Show

Consumption’s drag has grown, from — percent of the combined shortfall in 1953–82 to — percent in 1990–2009 and — percent in 2020. Inventories mattered most in the early recessions (— percent of the combined shortfall through 1982). Housing was a large drag only in 1973–75 and 2007–09, and government only in 1953–54 and 1969–70.

Then and now

Shares of the combined total, adding up each group of recessions so that bigger recessions count for more.

From 1953 to 1982, investment of all kinds accounted for — percent of the combined shortfall and consumption for — percent. From 1990 to 2009, consumption’s share rose to — percent and investment’s to — percent, while net exports offset — percent, because imports fell sharply. 2020 stands apart.

Quarter by quarter

One recession at a time, one column per quarter, from five quarters before the peak quarter to four after the trough, at an annual rate; the shaded quarters are the recession. This shows what a peak-to-trough window leaves out. Housing, for example, was already subtracting — a year relative to trend in the four quarters before the 2007 peak, on top of the — it subtracted during the recession.

Recession

Thirteen pieces

The six components split further: consumption into durable goods, nondurable goods and services; business investment into structures, equipment and intellectual property; government into defense, the rest of the federal government, and states and localities; net exports into exports and imports.

The detail, by recession

SubtractedAdded

Within consumption, durable goods (cars, furniture, appliances) were the biggest drag in — of the — and within a tenth of a point of the biggest in — more; in 2020 it was services. Equipment was the biggest part of the business-investment drag every time: in 2007–09 equipment subtracted — points and durable goods —. In the two recessions where government was a large drag, defense subtracted — points (1953–54) and — (1969–70). Net exports cushion recessions mainly because imports fall; falling imports added — points in 2007–09. In 2001 business investment (—) and exports (—) were the story, while durable goods added — more than trend.

The numbers

The numbers

How it was done

  • Recessions. The National Bureau of Economic Research’s peaks and troughs, 1948–49 through 2020. Monthly series are lined up on its monthly peak, quarterly series on its quarterly peak. A recession’s length is the months from its monthly peak to its monthly trough.
  • One baseline for every chart. A baseline is a trend: a series’ own pace over a stretch of years that ends at the peak. Every number on the page is the gap between where a series stood and where that trend, continued from the peak, would have put it: in percent for a level such as GDP or jobs, in percentage points for a rate. The trend compounds, so a series that simply kept growing on trend reads zero however far out a chart runs. Under “Actual” there is no trend and the number is the plain change from the peak. Measured this way, real GDP’s reading at the trough is exactly the sum of the six contributions.
  • Which series have a trend. Levels use their own average growth over the baseline’s window. Employment-population and participation rates use their own average change, because they moved for decades as women entered work and the population aged: women’s employment-population ratio was rising by — to — point a year going into every recession from 1953 to 2001. The unemployment rate, jobless claims, the long-term share of the unemployed, the federal funds rate and stock prices get none. For the four rates, a ten-year change mostly says where the last cycle happened to stand, not a pace that would have continued: unemployment fell — point a year in the ten years before the 2020 peak, which was only the recovery from the previous recession. Stock prices do trend upward, but a market fall is always quoted as the plain change from a high, so they are shown that way. Those five read the same under every baseline: the plain change from the peak.
  • How much history a trend needs. Five years of data before the peak; with between five and ten, the 10-year trend uses what exists. Nothing before 1947 is used, since the quarterly national accounts begin that year. 1948–49 has no trend at all, because fewer than two years of quarterly accounts come before its peak, so for every series that takes a trend it appears under “Actual” only. The series that take no trend and reach back that far (the unemployment rate, the long-term share of the unemployed, stock prices) show 1948–49 under every baseline.
  • The trough and the low point. “At the trough” reads each series in the NBER’s trough month or quarter. A series’ “low point” is its worst reading under the baseline chosen within three years of the peak, or before the next recession began if that came sooner, as it did after the 1957–58 and 1980 recessions. For a rate that rises in a recession (the unemployment rate, the long-term share of the unemployed, jobless claims) the worst reading is the highest, so a ranking of one calls it the high point. Against a trend the low point can come later than the series’ own low, because the gap from trend can keep widening after the series turns up: payroll jobs touched bottom — months after the 2007 peak, but were furthest below their 10-year trend — months after it. Recovery is the time from the peak until the series is back at its level at the peak.
  • The average and range. The other recessions that have the series under the baseline chosen: their average, their full range and the middle half, month by month. 2020 is left out unless it is switched in, because jobs fell — percent in two months and it would stretch every band. A recession’s line stops where the next recession begins: the 1980 recession’s after 18 months. Ranks count every recession with a reading under the baseline chosen: the — since 1953 for readings against a trend, all — for the actual change, and for a series that takes no trend as many as it reaches.
  • Rankings. On the Compare tab, the line under the “Rank by” menu says what the ranks count first: the deepest fall, the largest rise in a rate that rises in a recession, the most negative contribution, the largest share, the longest, the slowest, the largest fall in stocks or the largest cut. The bars can be put in date order instead; the ranks stay the same. Everywhere on the page, two readings that print the same are a tie. Only the recessions with a reading are ranked; the others are listed below them with the reason. “Payroll jobs: months still falling after the trough” counts from the NBER’s trough to the month payroll jobs were lowest. The Fed’s cut is the fall in the federal funds rate from the peak month to its lowest within three years of the peak, or before the next recession began. The scorecard shades each column from its best reading (no shading) to its worst (the darkest). The four tiles above the contributions count a piece as subtracting, and net exports as a cushion, only where the contribution prints as more than a tenth of a point.
  • Today. The last 12 months, drawn as if the peak had been —, on the same charts and under the same baselines, on the page’s Today tab, which can start it 6, 18 or 24 months back instead. It is a comparison, not a claim that a recession has begun: the NBER names a peak only months afterward. There, each past recession’s contributions are measured over the same number of quarters from its own peak. The table at the top of that tab shows the unemployment rate and jobless claims as their fall, so that further right is better in every row; the charts below it keep the ordinary signs.
  • Contributions. BEA’s contributions to the percent change in real GDP, which measure what made up growth without being thrown off by the fact that chained-dollar components do not add up. Each quarter’s are converted to a quarterly rate and compounded over the recession, so the six components add up exactly to GDP’s cumulative change. Against a trend, each component is compared with its trend contribution: its share of GDP growth over the baseline’s window, times what trend growth would have added over the recession. The gaps are stated in percent of where trend growth would have put GDP at the trough, so the six of them add up exactly to real GDP’s shortfall from trend. Shares divide each component by the total; a total smaller than one point is not divided.
  • Details of the data. The household survey was not collected in October 2025. Where that month is the starting point of a comparison, the page starts from an estimate (the unemployment rate rebuilt from the September and November counts of unemployed people and the labor force, — percent; the other household-survey rates as the average of the two months), and every line breaks at that month rather than bridging it. BLS folds its yearly population adjustments into the January household-survey figures and does not revise earlier months, so those series can move in a step each January. The page marks the adjustments since — that moved a series by — point or more (— percent for employment), with the size of each step taken from BLS’s tables of their effects. The one with the largest effect on the ratio, in January 2026, lowered the employment-population ratio by — percentage point and household-survey employment by — million; the one in January 2025 raised employment by — million. None moved the unemployment rate by as much as 0.05 point. Where a line crosses a marked adjustment the page breaks it, and where a change is measured across one the page says how much of the change is the adjustment; nothing is adjusted. Smaller steps, and those before —, are in the series unmarked. Jobless claims are the monthly average of the weekly seasonally adjusted count, per 1,000 payroll jobs, so that six decades are comparable. Stock prices are the S&P 500: the path is the monthly average of daily closes, and the largest fall is the biggest drop in daily closes from a high reached between 18 months before the peak and the trough to a later low no more than a year after the trough and before the next recession began. Published tables of the index’s daily closes in the 1950s and in 1960 differ slightly from one another, so the falls shown for 1957–58 and 1960–61 can differ from other sources by a few tenths of a point. Real manufacturing and trade sales come in two tables on different bases (through 1996, and from 1997) that are never joined, so nothing is measured across December 1996.
  • What this is not. An accounting, not a model. It says how far each series fell and which kinds of spending fell short, not why.

The four baselines

10-year trend (the default)
From the peak, how far each series stood from where its own average pace over the previous 10 years would have taken it. For every recession but 2001 and 2020, those ten years hold at least part of an earlier recession, so the trend includes the occasional downturn; the ten years before the 2001 and 2020 peaks were unbroken expansions. Where it misleads: ten years before the 2020 peak is early 2010, the bottom of the job market after the previous recession, so trend job growth comes out at — percent a year, against — percent between the 2007 and 2020 peaks. Three years after the 2020 peak, jobs were — percent below this trend but only — percent below the peak-to-peak trend. For 1953–54 only about six earlier years of accounts exist.
5-year trend
The same, with the trend taken from the five years before the peak. It tracks the pace just before each recession more closely, but five years are easily unusual, and then it misleads. The five years before 1960 include the 1957–58 recession, so trend growth looks slow (— percent a year, against — over ten years) and the shortfall small (— points, against —). The five years before 1957 were years of falling defense spending, so government’s trend contribution to growth was close to zero (— point a year, against — over ten years) and it shows up as a cushion in 1957–58 (—, against —). The 1980 recession falls inside the five years before 1981, and consumption’s drag in 1981–82 nearly disappears (—, against —). For today, the five years hold the rebound from 2020, so trend job growth is — percent a year, against — over ten.
Peak to peak
From the peak, against each series’ average pace since the last business-cycle peak at least five years earlier (for 1953–54 the 1948 peak, a little under five years earlier, is the only one there is). Peaks are similar points in the business cycle, so the pace between two of them is less distorted by where a fixed ten-year window happens to start: for 2020 it gives trend GDP growth of — percent a year, against —. It misleads when the two peaks were not alike. Unemployment was — percent at the 1953 peak and — percent at the 1960 peak, so trend growth for 1960–61 looks slow (— percent a year, against — over ten years).
Actual
No baseline: the plain change from the peak, with nothing subtracted (the index from 100). It is the number most often quoted, and as a gauge of what a recession cost it is misleading. A component that merely grew more slowly than usual still shows up as holding GDP up: consumption looks like a cushion in 1969–70 (—), 1981–82 (—) and 2001 (—), and government’s ordinary growth looks like an offset in — of the —. And in —, real GDP barely changed, so shares of the change are not meaningful; the charts leave them blank.

The series

Every series on the page

“Recessions” counts the 12 since 1948 the series covers; “with a trend” counts those with at least five earlier years of data, for the series that take a trend.

Sources: Bureau of Economic Analysis (national accounts, personal income, manufacturing and trade sales); Bureau of Labor Statistics (household survey, payroll survey, hours worked); Federal Reserve Board (industrial production, the federal funds rate); Department of Labor (jobless claims); S&P Dow Jones Indices (the S&P 500); National Bureau of Economic Research (recession dates and the six monthly indicators). Calculations by Econvitals.