Recession Mechanics
Recessions involve mutually reinforcing declines: firms cut output and hiring, households lose income and reduce spending, banks tighten credit, and the process deepens until some combination of policy response, price adjustment, and inventory clearing stabilizes the floor.
The depth and duration vary enormously. The 2001 recession lasted 8 months [1] and was mild by postwar standards. The 2007-2009 recession lasted 18 months [2], the longest since World War II, and destroyed trillions in household wealth.
Financial accelerator and balance-sheet recessions
The textbook IS-LM picture treats credit as a number on a chart. In a recession driven by deleveraging, credit is the chart. Bernanke, Gertler, and Gilchrist formalized the financial accelerator in 1999 [3]: a fall in asset prices weakens borrower balance sheets, which raises the external finance premium, which cuts investment, which lowers asset prices further. The mechanism is procyclical and non-linear, and it sits underneath every postwar recession that looked unusually deep relative to its trigger.
Mian and Sufi's work on the 2007-09 episode [4] put numbers on the household side of the same loop. Counties where household leverage rose most between 2002 and 2006 saw a sharp relative fall in spending on durable goods, starting in the third quarter of 2006, more than a year before the recession began. Aggregate consumption fell because deeply indebted households had to fix their balance sheets first. The crisis did not propagate evenly. It was concentrated in the counties that had borrowed the most.
The implication for forecasting is harder than it looks. A model that treats credit as exogenous will systematically understate the depth of recessions that follow asset-price corrections. The 1990 downturn, the 2001 dot-com mild recession, and the 2007-09 collapse all looked different on the way in than they did at the trough because the credit feedback was not in the projection. The 2020 episode was the exception: fiscal transfers replaced the missing income before the deleveraging loop had time to start.
Why COVID was a classification problem
The 2020 recession lasted two months on the NBER calendar [5], the shortest on record. By the textbook taxonomy it should have been a deep demand-side downturn. It was not. Service consumption collapsed and durable-goods spending boomed at the same time, which is impossible in a homogeneous-economy AD-AS picture. Total retail and food services sales fell 14.7 percent from March to April 2020 (seasonally adjusted), then rebounded 17.7 percent month-over-month in May, but May 2020 sales were still 6.1 percent below their May 2019 level and the three-month March-May period ran 10.5 percent below the prior year [6]. The restaurant industry collapsed harder: food services and drinking places (NAICS 722) fell 34.6 percent from March to April, leaving April sales 52.6 percent below their April 2019 level [6]. New auto sales recovered to pre-shock levels by mid-2020.
The right reading is that the pandemic was a sectoral reallocation shock layered on top of a demand contraction. Households could not consume haircuts, flights, or hotel nights, so they bought goods instead. The CARES Act delivered checks and an unemployment-insurance bonus that, for most UI-eligible workers, replaced more than 100 percent of lost wages [7]. The median statutory replacement rate was 145 percent during April through July 2020. Aggregate income held up. Composition rotated.
This combination broke the simple correlation between unemployment and inflation that had held since the 1990s. The 2021-22 inflation surge had three roughly independent components: goods scarcity (shipping, semiconductors), services normalisation (rents, insurance), and a wage Phillips curve in the still-recovering part of the labor market. Calling it 'demand' or 'supply' was the wrong framing. Different markets were on different curves at the same time.