A fall of 2.8% in the value of its assets was enough to wipe out Bear Stearns

2.8% The fall in asset values that erases the owners’ stake at a leverage of 35 to 1

Real U.S. house prices fell by more than a third after 2005, yet the mortgage losses were small next to the country’s wealth. So how did a housing bust bring the largest banks to the edge of collapse? And twelve years later, why did a far deeper slump fade so much faster?

Open with the number. It comes from the notebook’s leverage check: with a leverage ratio of 35, the fall in asset values that wipes out net worth is 1/(1 + 35) = 1/36, about 2.8 percent. Bear Stearns had about that leverage when it collapsed in March 2008.

Contrast it with a household that buys a house with 20 percent down: the same formula says a 20 percent fall in the house price wipes out its equity. An investment bank in 2007 had a cushion seven times thinner.

Two shocks, twelve years apart: the 2007–09 financial crisis and the 2020 pandemic. Not since the 1930s and the 1918 flu had the world seen shocks this large. The chapter has no new model; it measures the two recessions and adds the finance needed to understand the first one.

Ask the room: where were you, or your family, in 2008 and in April 2020? What changed first: jobs, prices, or the news?

Today: three questions

• What happened in 2007–09? A housing bubble, the Fed’s interest rate, fear between banks and a stock market crash.
• Why did a fall in house prices break the banks? Balance sheets, leverage, fire sales, bank runs and interest rate spreads.
• How did the two recessions compare, and what should policy do? Lives, output, jobs and inflation; then capital, rescues, jobs and lives.

These are the three acts of the deck and the three branches of the mind map on the title slide.

The headline result: leverage turned a fall in house prices into a banking crisis and a long, deep recession. The pandemic hit harder but more briefly, and inflation returned only a year later.

The chapter is a first look. Chapters 11 to 14 build the short-run model (the IS curve, monetary policy, aggregate demand and supply) that explains these booms, slumps and inflation. Here we collect the facts and the finance.

The crash of 2007–09

Act I

The housing bust

How unusual were the rise and fall of U.S. house prices in the 2000s?

Ask the room first. Most students know “house prices fell in 2008”. Ask: by how much? Was it the biggest fall since the 1950s, or one of several? Collect two or three guesses before moving on.

In real terms, U.S. house prices had their largest fall since 1953

Data: Robert J. Shiller, U.S. Home Price and Related Data, real home price index (from 1987 the S&P CoreLogic Case-Shiller U.S. National Index, deflated by the BLS CPI), shillerdata.com. Chart inspired by Fig. 10.1 in Jones (2024), Macroeconomics, 6th ed.

How to read it: Shiller’s index of the price of a typical U.S. house, divided by the consumer price index, with 1953 = 100. The vertical axis is a ratio (log) scale: equal distances are equal percent changes. The shaded band marks the bust chosen on the slider, and the two black dots are its peak and trough; the title computes the fall.

Do live:

1. Start on “2000s boom and bust”: a peak of 171 in December 2005 and a fall of 35.7 percent to February 2012.
2. Move to “1970s boom and bust”: a fall of 12.2 percent. Then “1980s”: 14.4 percent. The 2000s bust was more than twice as large.
3. Move to “2020s so far”: a fall of 6.9 percent from May 2022 to February 2023, and then a return close to the record.

Point at the flat line: for about 45 years real house prices hardly moved. The index was still near 100 in the late 1990s.

The 2000s bust was more than twice the size of any earlier one

• From the peak in December 2005 to February 2012, real prices fell 35.7%. The 1970s and 1980s busts were 12.2% and 14.4%.
• The boom was large, but not as large as often quoted: in this real index, prices rose 1.72 times from 1996 to 2006, about 5.6% a year. The textbook’s “nearly 3 times” must come from another index.
• On a ratio scale a fall by half and a doubling cover the same distance, so busts of different decades can be compared by eye.

Always ask which series a number comes from: nominal or real, national or the booming cities.

Numbers: the notebook’s Fig. 1 key numbers. Largest drawdown since 1953: −35.7 percent in February 2012; quarterly averages from 2006Q2 to 2012Q1 give −35.1 percent, the textbook’s “36 percent”. Annual averages: 1996 to 2006, prices times 1.72, or 5.6 percent a year by equation (2), g = 1.72^(1/10) − 1.

The textbook’s “factor of nearly 3” and “about 10 percent per year” do not match the real national index. The notebook guesses a nominal index, or a 20-city index with the booming coastal markets. This is a teaching moment: the same story told with two series gives very different numbers.

Beyond the U.S. (fig. 1b in the book chapter): in the BIS data, real house prices fell 56.7 percent in Ireland and 44.5 percent in Spain after 2007, more than in the U.S.; both had banking crises too.

The rule of 70: at 5.6 percent a year, prices double in about 12.5 years.

Cheap money and loose lending fed the boom

• A saving glut. After their own crises of the 1990s, many emerging economies saved more and lent abroad, largely to the U.S. Cheap funds chased assets.
• Low interest rates. The Fed kept its rate near 1% until mid-2004 (Fig. 2).
• Weaker standards. More mortgages went to borrowers with poor credit or high debt, on the belief that prices could only rise.
• Securitization. Mortgages were pooled and sliced into securities. Pooling spreads the risk of one house, but not the risk that all house prices fall together.

A nationwide fall in prices is an aggregate risk: no amount of pooling makes it go away.

This slide summarizes the causes the textbook lists, in our own words. They are not measured in the notebook, so the slide carries no numbers.

Saving glut: Bernanke’s 2005 Sandridge Lecture, which coined the phrase. Chapter 20 returns to the emerging-market crises of the 1990s.

Lending standards: Mian and Sufi (2009, Quarterly Journal of Economics) show that mortgage credit expanded most in ZIP codes with many subprime borrowers, where incomes were not rising, and that those areas later had the most defaults. Their reading is contested by other work that stresses middle-income borrowers and speculation; the point for us is that credit grew faster than the ability to repay.

Securitization: the originating bank sold the mortgage and no longer bore the loss if it went bad, which weakened the incentive to screen borrowers.

Ask: why would a sophisticated bank buy securities whose value depends on house prices always rising? (One answer: the bank’s managers are paid on profits during the boom, and the bank is highly leveraged; Act II.)

The Fed’s interest rate

What did the Federal Reserve do with its interest rate before the crash, and after it?

Ask the room first: did the Fed raise or lower rates in the years before 2007? And after the crash, how low did it go? Collect guesses.

The Fed raised its rate before the crash, then cut it to zero for seven years

Data: Board of Governors of the Federal Reserve System, Federal Funds Effective Rate (FEDFUNDS); NBER recession dates (USREC), via FRED. Chart inspired by Fig. 10.2 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the effective Fed funds rate, the monthly average of what banks paid one another for overnight loans, in percent. Shaded bands are NBER recessions. The buttons zoom in on four periods; the title reports the lowest and highest rate in the window.

Do live:

1. “2000–2026”: the whole story. The highest rate, 6.54 percent in July 2000; the lowest, 0.05 percent in April 2020.
2. “2004–06 hikes”: step by step from about 1 percent to above 5 percent.
3. “2007–08 cuts”: from August 2007 the Fed cut, and by December 2008 the rate was near zero.
4. “2022–23 hikes”: the fastest rise in the chart, to fight the inflation of Figs. 10 and 11.

Rates rose just as house prices peaked, then hit a floor at zero

• The effective rate was 1.00% in May 2004 and 5.24% in July 2006. Higher rates made mortgages dearer, above all adjustable ones whose low teaser rates were ending.
• From January 2009 to November 2015 the rate never rose above 0.22%. It cannot go much below zero: the zero lower bound of Chapters 11 and 14.
• In 2020 it went back to zero; in 2022–23 the Fed raised it to 5.33% by August 2023.

Once the rate was at zero, the Fed’s usual tool was used up for seven years.

Numbers: the notebook’s Fig. 2 key numbers. The textbook says the Fed raised its rate “from 1.25 to 5.25 percent” between May 2004 and May 2006. Those are the Fed’s targets. The effective rate follows the target with a short delay: 1.00 percent in May 2004, 4.94 percent in May 2006, and 1.26 and 5.24 percent two months later. The latest value, September 2026: 3.75 percent.

Beyond the U.S. (fig. 2b in the chapter): Switzerland and Japan had negative policy rates for years (Switzerland −0.75 percent from 2015), so zero is not quite a floor.

A debate to flag, not settle: some economists argue the Fed kept rates too low in 2002–04 and so fed the bubble; others reply that house prices rose in many countries with different monetary policies. We return to it with the IS-MP model.

Fear between banks

Where in the data can we see that banks stopped trusting one another?

Ask: if you lend to another bank for three months instead of to the U.S. government, what extra interest would you ask for? And what happens to that extra interest if you start to fear that the bank might fail? Let students propose a measure before showing it.

Banks charged one another a fear premium that exploded in 2008

Data: ICE Benchmark Administration and Federal Reserve Board, TED Spread (TEDRATE); Federal Reserve Board, 90-day AA financial commercial paper (RIFSPPFAAD90NB) and 3-month T-bill (DTB3); NBER (USREC), via FRED. Chart inspired by Fig. 10.3 in Jones (2024), Macroeconomics, 6th ed.

How to read it: two spreads, in percentage points. The TED spread is the 3-month LIBOR rate (what banks charged one another) minus the 3-month Treasury bill rate (what the U.S. government pays). LIBOR was discontinued in 2022, so the second line, AA-rated financial commercial paper minus the T-bill rate, carries the story on. Shaded bands are NBER recessions. Click a spread in the legend to hide it.

Point at three moments: August 2007, when banks first doubted one another; the autumn of 2008, after Lehman Brothers failed in September; and March 2020, a short spike that the Fed’s lending quickly calmed.

The spread moved through three phases of the crisis

• Calm, 2002 to July 2007: on average 0.33 points, between 0.16 and 0.75.
• Doubt, August 2007 to August 2008: on average 1.31 points.
• Panic: a monthly peak of 3.35 points in October 2008. The textbook’s “more than 4.5” refers to single days; monthly averages smooth the spike.

A jump in a spread is the clearest sign of a financial crisis in the data. Act II explains what it measures.

Numbers: the notebook’s Fig. 3 key numbers. The two spreads correlate 0.97 over 2000–2022, so the commercial-paper spread is a good stand-in after 2022. Since February 2022 its highest value is 0.38 points (June 2022): no stress episode like 2007–08.

Flight to safety: lenders left interbank loans for T-bills, which pushed the T-bill rate down and the interbank rate up. The spread widened from both sides.

Beyond the U.S. (fig. 3b in the chapter): the euro area’s and the U.K.’s interbank spreads also jumped in August 2007 and peaked in October 2008; Canada’s hardly moved.

In two weeks of September 2008 the government took over Fannie Mae and Freddie Mac, Lehman failed, Bank of America bought Merrill Lynch, and the Fed lent to AIG. That is the moment of the spike.

The stock market

How far did stock prices fall, and why did an earlier crash of the same size not cause a crisis?

Ask: who remembers the dot-com crash of 2000? Were banks in trouble then? Collect answers; the next chart shows that the two crashes were of similar depth.

Real stock prices lost about half their value, twice in a decade

Data: Robert J. Shiller, U.S. Stock Market Data, S&P Composite price index and CPI (Irrational Exuberance, updated), shillerdata.com. Chart inspired by Fig. 10.4 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the S&P 500 index of 500 large U.S. companies, divided by the CPI, in September 2026 dollars (about 11 percent above the textbook’s 2023 dollars; percent changes are the same). Ratio scale: each tick doubles the one below it. The slider picks a crash; the shaded band, the dots and the title measure it.

Do live:

1. “Financial crisis, 2007–09”: −51.7 percent from October 2007 to March 2009.
2. “Dot-com bust, 2000–03”: −46.8 percent from August 2000 to February 2003. Almost as deep.
3. “Covid-19, 2020”: −19.1 percent from January to March 2020, and back above January’s level by August.
4. “Rate hikes, 2022”: −25.5 percent from November 2021 to October 2022.

A crash on borrowed money is far more dangerous than a crash on savings

• From October 2007 to March 2009 real stock prices fell 51.7%, below every month of 1997–99. They regained the 2007 peak only in November 2013.
• The dot-com bust of 2000–03 was almost as deep, 46.8%, but caused no financial crisis.
• Its losses fell on millions of shareholders who had not borrowed. In 2007–09 the losses fell on a few banks that held mortgage securities with high leverage.

Who bears the losses matters as much as how large they are.

Numbers: the notebook’s Fig. 4 key numbers. With the textbook’s dates, November 2007 to March 2009, the fall is 48.9 percent. The March 2009 level, 1,194, was below the lowest month of 1997–99, 1,600. In March 2009 the index was 58.6 percent below its record of August 2000.

Beyond the U.S. (fig. 4b): the crash was global. Germany −55.4, Japan −57.6, the U.K. −45.5 percent; Iceland, whose banks collapsed, −96 percent. After 2009 the paths split: real U.S. stock prices in 2026 are about 1.5 times their 2007 high, Greece’s about a third of theirs.

This slide is the bridge to Act II: leverage is what turns a loss into a bankruptcy.

Undoing a crash takes a larger rise than the fall

Stocks fell 50%, so a 50% rally brings them back to where they were.

A fall of a fraction \(f\) needs a rise of \(f/(1-f)\) to recover. After a fall of 51.7%, real stock prices had to rise 107%, which took until November 2013. A 10% fall needs 11.1%; a 35.7% fall needs 55.5%.

Equation (3) of the Model overview. The trough is (1 − f) times the peak, so getting back means multiplying by 1/(1 − f). The rise needed, in percent, is 100 × f/(1 − f).

The notebook’s checks: 100 → 90 → 100 needs +11.1 percent; 100 → 64.3 → 100 needs +55.5 percent (the house-price bust); 100 → 48.3 → 100 needs +107.0 percent (the stock crash).

The slope of f/(1 − f) is 1/(1 − f)², above one and rising with f: deep crashes take a long time to undo even with strong growth afterwards.

Log changes add up and percent changes do not: a fall of 50 percent then a rise of 50 percent leaves a series 25 percent below its start, because ln 0.5 + ln 1.5 = ln 0.75. This is why Figs. 1 and 4 use a ratio scale.

Why a housing bust broke banks

Act II

A balance sheet lists what a bank owns and what it owes

Assets	$ bn	Liabilities and net worth	$ bn
Loans	1,000	Deposits	1,000
Securities	900	Short-term borrowing	400
Cash, reserves	100	Bonds	400
Total assets \(A\)	2,000	Liabilities \(D\)	1,800
		Net worth \(E\)	200

\[A = D + E \qquad \ell \equiv \frac{D}{E} = \frac{1{,}800}{200} = 9\]

• Net worth (equity, capital) is what is left for the owners.
• Capital ratio \(E/A = 10\%\); reserves are 10% of deposits.

Every 10 dollars of this bank’s assets rest on 1 dollar of its owners’ money.

The hypothetical bank of the textbook, loosely modeled on a large commercial bank. Notation from the Model overview: A for assets (not productivity in this chapter), D for liabilities (debt), E for net worth (equity), ℓ for the leverage ratio.

Net worth is defined as A − D, so the two sides always balance. The bank satisfies an illustrative capital requirement of 6 percent of assets and a reserve requirement of 3 percent of deposits.

Step 1 of the Model overview: dividing A = D + E by E gives A/E = 1 + ℓ, so E/A = 1/(1 + ℓ). Leverage and the capital ratio carry the same information. A capital requirement of 6 percent is a ceiling on leverage: 1 + ℓ ≤ 1/0.06, or ℓ ≤ 15.7.

The key assumption of everything that follows: liabilities are promises to repay fixed amounts; the value of assets moves with house prices, stock prices and defaults.

Balance sheets work the same way for households, firms and governments.

Leverage multiplies the owners’ gains and losses by \(1+\ell\)

\[\frac{\Delta E}{E} = \frac{A}{E}\cdot\frac{\Delta A}{A} = (1+\ell)\,\frac{\Delta A}{A}\]

Debt is fixed, so every dollar gained or lost on the assets lands on the owners: \(\Delta E = \Delta A\).

• The bank. Securities rise from 900 to 1,000: a return of 11.1% on them, but net worth goes from 200 to 300, +50% for the owners.
• A home buyer with 20% down has \(1+\ell = 5\). The house price rises 10%: equity goes from $100,000 to $150,000, +50%. It falls 10%: equity halves.

Leverage is a genie in good years and a trap in bad ones: the multiplier works both ways.

Equation (6). Change the value of the assets by ΔA; liabilities do not move; so ΔE = ΔA. Divide by E and multiply and divide by A.

For our bank 1 + ℓ = 10: a 1 percent gain on all its assets, 20 billion, raises its equity by 10 percent (the notebook’s check). When only one item changes, measure the change relative to total assets: 100 billion on the investments is 5 percent of all assets, so 50 percent for the owners.

The household: a 500,000 dollar house bought with 100,000 of the buyer’s money. Notebook: +10 percent → equity 150,000 (+50 percent); −10 percent → 50,000 (−50 percent); −20 percent → 0 (−100 percent).

Intuition: the owners hold a thin layer on top of a large pile of debt. A small change in the size of the pile is a large change in the thickness of the layer.

Ask: why do banks like leverage? (In good years the owners’ return is many times the return on the assets. Before 2007, banks that took leveraged bets that paid off nine years in ten showed long runs of spectacular profits.)

The thinner the equity, the smaller the fall that wipes it out

\[E' = (1-f)A - D = E - fA \le 0 \quad\Longleftrightarrow\quad f \ge f^* = \frac{E}{A} = \frac{1}{1+\ell}\]

• Our bank, \(\ell = 9\): \(f^* = 10\%\). Securities fall from 900 to 600: assets fall 15%, net worth goes from 200 to −100.
• Bear Stearns, \(\ell \approx 35\): \(f^* \approx\) 2.8%.

Higher leverage \(\ell\) means …	formula	at \(\ell = 9\)	at \(\ell = 35\)
a bigger multiplier on returns	\(1+\ell\)	10	36
a smaller fall that wipes out equity	\(1/(1+\ell)\)	10%	2.8%
more forced sales per $1 of losses	\(\ell\)	9	35

A bank whose assets are worth less than its debts is insolvent: selling everything would not repay what it owes.

Equation (7). Let the value of the assets fall by the fraction f while D stays put. Net worth becomes E − fA. It reaches zero when f = E/A = 1/(1 + ℓ).

The textbook’s example: investments fall from 900 to 600 billion, so total assets fall from 2,000 to 1,700, 15 percent, and net worth goes from +200 to −100.

The table is the Model overview’s comparative statics. In logs, ln f* = −ln(1 + ℓ): doubling 1 + ℓ halves the cushion.

Households: with an 80 percent mortgage, ℓ = 4 and f* = 20 percent. A larger fall leaves the house worth less than the mortgage (“under water”). Lost housing wealth is one reason consumption fell in the Great Recession (Act III).

Bear Stearns: roughly 3 dollars of equity and 97 of borrowing per 100 of assets. The third row of the table is the next slide.

Losses force sales, and sales cause new losses

\[\underset{\text{assets left}}{(A-x)} \;-\; \underset{\text{target}}{(1+\ell)(E-x)} \;=\; \ell\,x\]

• To bring leverage back to \(\ell\) without new equity, a bank that loses \(x\) must sell \(\ell x\) of assets and repay debt.
• Our bank: a loss of 10 forces sales of 90. At Bear Stearns’ leverage, 35.

Asset prices fall → losses \(x\) → forced sales \(\ell x\) → many banks sell the same assets → asset prices fall again

Through this loop, mortgage losses that were small next to U.S. wealth could shake the whole financial system.

Equation (8). After a loss of x, assets are A − x and equity E − x, so leverage jumps to D/(E − x). To restore the old leverage the assets must shrink to (1 + ℓ)(E − x). Expand and use A = (1 + ℓ)E: the sales needed are ℓx. The notebook checks it: a loss of 10 forces sales of 90 and leaves leverage at 9.0.

This is deleveraging. One bank selling is harmless. Many leveraged banks selling the same kind of assets at the same time push those prices down, which causes new losses everywhere, which equation (8) turns into new sales.

Systemic risk: banks lend to and insure one another every day. Paul O’Neill, a former Treasury secretary, compared it to ten bottles of water with one poisoned: if no one knows which, no one drinks. Banks stopped lending to one another (the spread of Act I) because no one knew which balance sheet held the bad mortgages.

The full derivation is in the book chapter’s Model overview.

A run can sink a bank that was solvent

\[\text{loss} = \frac{\delta\,(W-R)}{1-\delta}\]

• \(W\): withdrawals; \(R\): cash and reserves; \(\delta\): the fire-sale discount on loans and investments sold in a hurry.
• Full run on our bank: it fails if \(\delta \ge E/(E+W-R)\) = 18.2%.

Withdrawals \(W\)	loss at \(\delta = 25\%\)	net worth left
100	0	200
400	100	100
700	200	0
1,000 (all)	300	−100

With a 25% discount, a run on 70% of the deposits is enough to make this bank insolvent.

Equation (9). If withdrawals exceed cash, the bank must sell illiquid assets of book value S with (1 − δ)S = W − R. It gives up S and gets (1 − δ)S, so it loses δS = δ(W − R)/(1 − δ), and that comes out of net worth.

Two ways to solve the failure condition, loss ≥ E. For δ in a full run (W = all deposits, 1,000): δ* = E/(E + W − R) = 200/1,100 = 18.2 percent. For W at a given discount: W* = R + E(1 − δ)/δ; at δ = 25 percent, W* = 100 + 200 × 3 = 700, 70 percent of deposits. The table is the notebook’s run check.

Comparative statics: a run is harder to trigger (W* larger) with more cash R, more equity E or a smaller discount δ. Each policy works on one of them: reserve requirements raise R, capital requirements raise E, a lender of last resort removes the fire sale, and deposit insurance removes the reason to run.

In 2007–08 the runs were not on insured deposits but on banks’ short-term debt, which lenders could simply refuse to renew.

Illiquid is not the same as insolvent

A bank that cannot pay its depositors today must have made bad loans.

At normal prices our bank is solvent: net worth 200. It fails only because a run forces it to sell at a discount. If each depositor expects the others to run, running first is the best move, and the run proves itself right.

This is the logic of Diamond and Dybvig (1983, Journal of Political Economy): a bank turns illiquid loans into liquid deposits, and that service makes it vulnerable to self-fulfilling runs. Without a run the fire-sale discount never matters; with one, it turns a liquidity problem into a solvency problem.

Real cases mix the two. In 2008 many institutions had both: real losses on mortgage securities (solvency) and lenders who stopped renewing short-term funding (liquidity). In a panic it is hard, even for regulators, to tell which bank is which. That difficulty is at the heart of the policy debate in Act III.

Ask: why does deposit insurance stop runs even if it is almost never paid out? (If every depositor knows their money is safe, no one gains by running first, so the bad equilibrium disappears.)

An interest rate spread is a thermometer of fear

Lend $1 to a bank that fails with probability \(p\), or buy a T-bill. A lender who cares about the average payoff is indifferent when

\[(1-p)(1+i) = 1+i^f\]

\[\Rightarrow\quad i - i^f = \frac{(1+i^f)\,p}{1-p} \approx p\]

• Calm years: a spread of 0.3 points ↔︎ a perceived default risk of about 0.3% a year.
• October 2008: 3.35 points ↔︎ about 3.2%.
• Fear of default rose about tenfold.

The spread rises almost one for one with the risk of default as lenders see it.

Equation (10). The approximation holds when p and the T-bill rate are small. The notebook’s check with a T-bill rate of 1 percent: a default probability of 0.3 percent gives a loan rate of 1.30 percent and a spread of 0.30 points; 3.2 percent gives 4.34 and 3.34 points. In both cases the expected payoff equals the T-bill’s, 1.0100.

The slope of the spread with respect to p is (1 + i^f)/(1 − p)², close to 1.

Caveats: if lenders recover part of the loan when a borrower fails, the spread is smaller for the same p; if they dislike risk, it is larger. Banks that worry about their own funding also hoard cash rather than lend it, which adds a liquidity premium. Either way the spread moves with fear.

This closes Act II. Back to the data: what did the crisis do to output, jobs and prices, and how did the pandemic compare?

Two recessions compared

Act III

Lives and output in the pandemic

Did the countries that lost more people to Covid-19 also lose more output?

Picture each country as a dot: deaths per million across, GDP lost up. Do the dots slope up, slope down, or show no pattern?

Ask students to commit. A common guess: more deaths → more lockdowns and fear → more lost GDP, so the dots slope up. Another: countries that locked down hard saved lives and lost output, so the dots slope down. Write both on the board.

More deaths did not mean more lost output

Data: Our World in Data, Covid-19 deaths per million (ourworldindata.org/covid-deaths), from the WHO; OECD, Quarterly National Accounts, real GDP volume index (DSD_NAMAIN1@DF_QNA), OECD Data Explorer. Chart inspired by Fig. 10.5 in Jones (2024), Macroeconomics, 6th ed.

How to read it: one dot per country. Across: cumulative confirmed Covid-19 deaths per million people. Up: the GDP loss, the sum of the quarterly shortfalls of real GDP below its 2019Q4 level from 2020Q1, in percent of one year’s GDP (equation 12). A value of 6 means the country lost as much output as 6 percent of one year’s GDP. The slider chooses the date of the death count; hover for the names.

Do live:

1. On “March 2023”, the textbook’s date, the title gives the correlation: 0.18.
2. Point at Spain (top: the most output lost), Bulgaria and Hungary (right: the most deaths per person), and China, Japan and South Korea (low on both).
3. Move to “August 2026”: the dots move right as deaths kept rising (by more than 500 per million in Sweden), but the ranking hardly changes.

Data hygiene: Chile’s cumulative count falls by 78 per million between the two dates, because Chile revised its earlier reports.

The cloud has almost no slope: the two costs were largely separate

• Answer: no clear pattern. The correlation is only 0.18.
• The U.S. lost 1 in 306 people, like Brazil (1 in 300) and the U.K., but only 3.3% of a year’s GDP. Spain lost 17.4%, the U.K. 13.5%.
• U.S. death rates were 3.4 times Norway’s and 5.5 times Japan’s.

Deaths depended on health systems, age and the virus; output losses on lockdowns, sectors and policy support.

Numbers: the notebook’s Fig. 5 key numbers. U.S. deaths, 3,264.6 per million by 22 March 2023, so one out of every 1,000,000/3,264.6 = 306 people, the textbook’s “1 out of every 300”. Brazil 1 in 300, the U.K. 1 in 305, Italy 1 in 316. U.S. death rates were 3.4 times Norway’s, 5.5 times Japan’s and 4.9 times South Korea’s, the textbook’s “lower by a factor of 3 or 5”.

The U.S. GDP loss is 3.3 percent of a year’s GDP with today’s data; the textbook’s 3.7 used GDP as published in early 2023. Data are revised long after a crisis.

A correlation across countries is not a causal trade-off. Countries differ in age structure, density, tourism, how they count deaths and how much they spent to support incomes.

GDP in the pandemic

Did every economy fall as far, and climb back as fast, in 2020?

Ask: which country do you expect to have fallen most in the spring of 2020: the U.S., Germany, Sweden (which avoided a strict lockdown), South Korea, Spain or the U.K.?

The fall of spring 2020 ranged from 4% to 22%

Data: OECD, Quarterly National Accounts, real GDP, chain-linked volumes, seasonally adjusted (DSD_NAMAIN1@DF_QNA), OECD Data Explorer. Chart inspired by Fig. 10.6 in Jones (2024), Macroeconomics, 6th ed.

How to read it: real GDP of each country as an index with 2019Q4 = 100. A value of 90 means GDP was 10 percent below its pre-pandemic level. The menu switches between the textbook’s six countries and regions of 49 economies; click a country in the legend to add or hide it.

Do live:

1. “Textbook’s view”: in 2020Q2, the U.K. and Spain about −22 percent; the U.S., Germany and Sweden about −9 to −11; South Korea only −4.
2. Point at the U.K.’s second dip in early 2021, during a new lockdown.
3. Add India: the deepest fall of all 49 economies, −23.8 percent, yet 46.5 percent above 2019 by 2026.
4. Data hygiene: add Ireland, 41 percent above 2019, much of it profits booked there by multinational firms.

Deep falls, fast climbs, then paths that split

Country	lowest index	back at 100	GDP loss to 2022Q3	index 2026Q2
U.S.	90.9	2021Q1	3.3	116.3
S. Korea	96.0	2021Q1	1.9	114.5
Sweden	91.2	2021Q1	3.5	109.3
Germany	89.4	2021Q4	5.0	101.9
U.K.	77.8	2021Q4	13.5	106.2
Spain	78.0	2022Q2	17.4	111.9

Most economies were back at their 2019 level within one to two years; the GDP loss is the area of the dip.

The notebook’s table for Fig. 6. GDP loss in percent of one year’s GDP, equation (12): add up, quarter by quarter, 100 minus the index when it is below 100, and divide by 4 because each quarter is a quarter of a year. The U.S. value, 3.3, is the GDP loss of Fig. 5.

After the textbook’s data end, the paths spread out: the U.S. kept growing fast, while German GDP has stayed close to its 2019 level for four years. That is a story for the long-run chapters (and for energy prices after 2022), not for the pandemic.

Should governments trade output for lives?

In 2020 governments closed schools, shops and borders. How should they weigh lives saved against output lost?

What the numbers say

• At $10 million per statistical life over 40 years, a year of life is worth about $250,000.
• Three lost years of U.S. life expectancy ≈ $750,000 per person.
• The GDP loss was about 3% of one year’s income.

What is debated

• Did lockdowns, or fear, empty the shops?
• Whose lives and whose jobs? Costs fell unevenly.
• One value for every life is a convention.

Numbers: the notebook’s “lives lost and their value”. A value of a statistical life V of 10 million dollars, the figure the U.S. Environmental Protection Agency uses to weigh safety rules, spread over T = 40 remaining years: v = V/T = 250,000 dollars a year. Discounted at 3 percent, a year of life is worth about 433,000 dollars, because later years count for less. Three years of life expectancy lost: 750,000 dollars per person, many times the lost output of equation (12).

The research. Hall, Jones and Klenow (2020, Federal Reserve Bank of Minneapolis Quarterly Review) value the deaths of the pandemic in consumption terms and find that society would accept a large fall in consumption to avoid them. Goolsbee and Syverson (2021, Journal of Public Economics) compare neighboring counties with and without stay-at-home orders and find that most of the fall in consumer visits came from people’s own fear of infection, not from legal orders: so lifting rules early would not have restored the economy. Both are contested; other studies find larger effects of orders.

The left column’s numbers are the textbook’s illustration. The right column’s points: the elderly carried most of the deaths; the young lost schooling and jobs; a statistical life is a policy convention, and some argue for valuing life-years instead.

Ask the policy students: would you use the same value of a life for a highway rule and a pandemic lockdown? Why or why not?

Jobs in two recessions

In 2020 the U.S. lost about two and a half times as many jobs as in 2008–10. Which recession took longer to win the jobs back?

Guess the number of months from the old peak until employment was back above it, for each recession.

Students often expect the bigger loss to take longer to undo. Collect a number of months for each recession, and write the class median on the board.

The pandemic destroyed jobs faster and brought them back faster

Data: BLS, total nonfarm payroll employment (PAYEMS); NBER (USREC), via FRED. Chart inspired by Fig. 10.7 in Jones (2024), Macroeconomics, 6th ed.

How to read it: total nonfarm payroll employment in millions of jobs, monthly, seasonally adjusted, from a survey of employers. Shaded bands are NBER recessions.

Do live: drag the range slider under the chart to zoom in on 2007–2015, then on 2020–2023, and compare how long each climb back took.

Point at the Great Recession: employment kept falling for months after the recession officially ended in June 2009. Then the cliff of 2020: more than half of the lost jobs were back by the end of 2020.

8.7 million jobs took 77 months to return; 21.9 million took 28

Recession	peak	trough	jobs lost	change	back at peak	months
Great Recession	Dec 2007	Feb 2010	8.7 m	−6.3%	May 2014	77
Covid-19	Feb 2020	Apr 2020	21.9 m	−14.4%	Jun 2022	28

• The answer to the prediction: the smaller loss took almost three times as long to undo.
• After 2008, banks and households repaired their balance sheets for years. In 2020 most jobs were paused, not destroyed.

The size of a slump and the length of the recovery are two different things.

Numbers: the notebook’s Fig. 7 key numbers. Great Recession: 8.69 million jobs, 6.28 percent of employment, from December 2007 to February 2010, and back at the old peak in May 2014, 77 months after it. Covid-19: 21.87 million jobs, 14.36 percent, from February to April 2020, back in June 2022, 28 months after the peak. Months are counted as 12 × (difference in years) + (difference in months).

The interpretation is our reading, not a measurement: a crisis rooted in balance sheets (Act II) leaves households and banks paying down debt for years, which holds back spending. In 2020 many workers were on temporary layoff and returned to the same employer.

Jobs across countries

Did every rich country lose a sixth of its jobs in the spring of 2020?

Ask: in April 2020 the U.S. unemployment rate reached 14.8 percent. Guess Germany’s. Then show the chart.

Some countries kept workers on the payroll; the U.S. let jobs go

Data: OECD, Infra-annual labour statistics, employment from labour force surveys (DSD_LFS@DF_IALFS_EMP_Q), OECD Data Explorer. Chart inspired by Fig. 10.7 in Jones (2024), Macroeconomics, 6th ed.

How to read it: employment from each country’s household survey (people aged 15 and over with a job), quarterly, as an index with 2019Q4 = 100, so countries of very different sizes can be compared. The menu shows the textbook’s view or a region; click a country in the legend to add it.

Do live:

1. The title: the largest Covid-19 fall, the United States, −13.3 percent from 2019Q4 to 2020Q2. Add Canada: −13.0 percent.
2. Add Germany, Japan and the United Kingdom: falls of only 1.7 to 2.4 percent.
3. Add Spain: before Covid-19 it had a much worse crisis, −17.5 percent from 2008 to 2013.

Data hygiene: Brazil and Indonesia show no data, because their series stop before the 2019Q4 base.

Germany, Japan and the U.K. lost hours, not jobs

• U.S. employment fell 13.3%, Canada’s 13.0%. Germany, Japan and the U.K. lost only 1.7 to 2.4%.
• Those countries paid firms to keep workers on shorter hours (Kurzarbeit, furlough). The slump showed up in hours worked, not in the number employed.
• The U.S. let jobs go and supported workers through unemployment insurance, and rehired fast.

Two ways to share the same shock: keep the job and cut hours, or cut the job and insure the worker.

Numbers: the notebook’s “Beyond the U.S.” notes for Fig. 7. In the household survey the U.S. fall was a little larger than in payrolls (11.8 percent in quarterly averages). Unemployment tells the same story (fig. 9b): Germany’s rate was at most 3.9 percent in 2020, while the U.S. rate jumped to 14.8 and Canada’s to 14.2.

This slide sets up the next policy question.

Should governments insure jobs or insure workers?

In a sudden stop, should the state pay firms to keep workers on short hours, or pay workers who lose their jobs?

Insure jobs (short-time work)

• Keeps the worker–firm match, so output can restart at once.
• Unemployment hardly moves.
• Cheapest when the shock is temporary.

Insure workers (unemployment insurance)

• Lets workers move to firms and sectors that grow.
• Short-time work can prop up jobs that would vanish anyway.
• Its money can go to firms that would have kept the workers.

The data are the book’s: employment fell 1.7 to 2.4 percent in Germany, Japan and the U.K., 13.3 percent in the U.S.; U.S. payroll jobs were back at their peak after 28 months.

The research. Giupponi, Landais and Lapeyre (2022, Journal of Economic Perspectives, “Should We Insure Workers or Jobs during Recessions?”) compare the two tools and argue that short-time work is most useful for large, temporary shocks, while unemployment insurance is better when reallocation is needed. Giupponi and Landais (2023, Review of Economic Studies) find that Italy’s short-time work saved jobs in the Great Recession but slowed the move of workers to more productive firms. Cahuc, Kramarz and Nevoux (2018, IZA working paper) find that in France it saved jobs in hard-hit firms but paid windfalls to others. On the U.S. side, Ganong, Noel and Vavra (2020, Journal of Public Economics) show that the CARES Act’s flat supplement gave many unemployed workers more than their previous wage. The size of these effects is contested.

The pandemic was the ideal case for short-time work: a shock everyone expected to end. The Great Recession, which reshaped construction and finance, was less so.

Ask the policy students: which tool would you choose for a shock that permanently shrinks one industry, such as coal or retail?

Short-run output

Which slump cost more output in total: about 5% below potential for years, or about 9% below potential for a year?

Recall Chapter 9: short-run output Ỹ is the percent gap between actual GDP and potential GDP. Ask students to estimate which slump was more costly before showing the chart.

After 2009 output stayed below potential for years; after 2020, for about a year

Data: BEA, real GDP (GDPC1); CBO, real potential GDP (GDPPOT), via FRED; 2023 vintage from ALFRED; NBER (USREC). Chart inspired by Fig. 10.8 in Jones (2024), Macroeconomics, 6th ed.

How to read it: short-run output Ỹ = (Y − Ȳ)/Ȳ, in percent, quarterly. Two vintages: today’s data, and the data as published in February 2023, when the textbook was written (it stops in 2022Q4). Since then the BEA has revised GDP and the CBO has re-estimated potential output. Click a vintage in the legend to hide it.

Point at the two shapes: a long, shallow trough after 2008 and a deep, narrow one in 2020. Then at the vintages: same shape, but today’s data put output higher, and above potential since 2021.

A slump’s cost is its depth times its length

\[\text{loss} \approx \frac{1}{4}\sum_q \left(-100\,\widetilde{Y}\!_q\right) \approx \tfrac{1}{2}\,d\,T\]

• A slump of depth \(d\) that closes in a straight line over \(T\) years is a triangle.
• To close a gap, output must grow faster than potential: \(g_Y \approx \bar{g} + \Delta\widetilde{Y}\).

	lowest \(\widetilde{Y}\)	back above zero	loss, % of a year’s GDP
Great Recession	−5.0%	2018Q1	21.3
Covid-19	−8.8%	2021Q2	3.4

The pandemic slump was almost twice as deep, but the Great Recession cost about six times as much output.

Equation (12) with −100Ỹ in place of 100 − I: add the quarterly gaps below potential and divide by 4. The notebook computes the exact sums: 21.3 percent of a year’s GDP for the Great Recession and 3.4 for Covid-19. The triangle rule gives about ½ × 5 × 9 ≈ 22 and about 4.

Numbers: the notebook’s Fig. 8 key numbers. Today’s data: lowest −4.98 percent in 2008–09 and −8.83 in 2020; Ỹ averaged −1.48 percent from 2010 to 2019; first quarter above zero after 2009, 2018Q1; after 2020, 2021Q2. The 2023 vintage: −5.48 and −10.42, the textbook’s “more than 5 percent” and “−10 percent”; back above zero only in 2021Q4.

Growth and the gap (Chapter 9): closing a gap of 5 percent in five years takes growth one point a year above potential, every year. The U.S. recovery after 2009 grew only a little faster than potential, which is why the gap closed so slowly.

Potential output is an estimate; the OECD’s U.S. gap for 2020 is −4.2 percent against −3.3 for the CBO-based annual average.

Unemployment

How high did unemployment go in each recession, and how fast did it come down?

Ask for two numbers per recession: the peak rate and the number of years until it was back near 5 percent.

Unemployment rose 5.6 points in the Great Recession, 11.3 in two months of 2020

Data: U.S. Bureau of Labor Statistics, Unemployment Rate (UNRATE); NBER (USREC), via FRED. Chart inspired by Fig. 10.9 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the U.S. unemployment rate, monthly, seasonally adjusted. The slider picks a recession; the title gives the lowest rate before the peak and the highest rate in the window.

Do live:

1. “Great Recession”: from 4.4 percent in March 2007 to 10.0 percent in October 2009, the textbook’s numbers. Then six years to get back to 5 percent.
2. “Covid-19 recession”: from 3.5 percent in September 2019 to 14.8 percent in April 2020; below 4 percent again by the end of 2021.
3. “2001 recession”: from 3.8 to 6.3 percent, a rise of 2.5 points whose peak came in mid-2003, long after the recession ended.

Data hygiene: October 2025 is missing; the BLS could not run its household survey during the federal government shutdown.

Across countries

Was 10% unemployment high by international standards?

Ask: name a rich country where unemployment went above 20 percent in the last two decades.

The euro crisis pushed unemployment far above the U.S. peak

Data: OECD, harmonised unemployment rates, monthly (DSD_LFS@DF_IALFS_UNE_M), OECD Data Explorer; NBER (USREC). Chart inspired by Fig. 10.9 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the OECD’s harmonised unemployment rates, each country’s labour force survey adjusted to one international definition. For the U.S. it is exactly the BLS rate of Fig. 9. Shaded bands are U.S. recessions. The menu shows the textbook’s view or a region; click a country in the legend to add it.

Do live:

1. Add Spain and Greece: above 26 and 28 percent in 2013, during the euro crisis. The euro area as a whole peaked at 12.3 percent in 2013.
2. Add Germany: its rate fell through the Great Recession, from 11.2 percent in 2005 to below 4 percent, and in 2020 it was at most 3.9 percent.
3. Add Canada: 14.2 percent in 2020, close to the U.S.

Both recessions were far deeper than the average one since 1953

Change from NBER peak to trough	average of previous 9	Great Recession	Covid-19
Real GDP	−1.6%	−3.8%	−9.1%
Nonfarm employment	−1.8%	−5.0%	−11.8%
Unemployment rate	+2.3 pp	+4.5 pp	+9.4 pp
Consumption	+0.4%	−2.3%	−10.2%
Investment	−12.9%	−29.3%	−16.1%
Government purchases	+1.2%	+5.3%	+3.3%
Exports	−1.1%	−9.2%	−23.9%
Imports	−4.2%	−17.9%	−19.9%

In a typical recession consumption hardly moves; in both recent ones it fell.

The notebook’s Table 10.1, with today’s data: the change between the NBER peak and trough quarters, for the nine recessions from 1953 to 2001, the Great Recession (peak 2007Q4) and Covid-19 (peak 2019Q4).

Our numbers differ from the textbook’s for two reasons. Data revisions: today’s data make the Great Recession milder (GDP −3.8 instead of −4.7, investment −29.3 instead of −34.0); the chapter’s data notes show that data as published in 2012 give the textbook’s values. Dates: the textbook’s employment fall of 6.3 percent is the monthly peak-to-trough fall of Fig. 7, not the change between NBER quarters (5.0).

Every pattern of the textbook’s table holds. Investment, exports and imports fall sharply in every recession; government purchases rise, in part because of stimulus.

Consumption: households usually smooth consumption on their long-run income (Chapter 16). In 2008–09 the falls in house and stock prices cut household wealth, and consumption fell. In 2020 people stayed away from shops, restaurants and travel.

Investment pulled GDP down in 2008–09; consumption in 2020

\[\frac{\Delta Y}{Y} = \sum_X \frac{X}{Y}\cdot\frac{\Delta X}{X}\]

• Each term is a contribution: the component’s share times its own change, in points.
• Imports enter with a minus sign.

Contribution, points	Great Recession	Covid-19
Consumption	−1.51	−6.86
Investment	−4.79	−2.89
Government	+1.05	+0.57
Exports	−0.99	−2.82
Imports (fall adds)	+2.50	+2.88
Sum (GDP)	−3.74 (−3.82)	−9.13 (−9.09)

A small component that swings a lot can matter more than a large one that barely moves.

Equation (13): divide the change in Y = C + I + G + EX − IM by Y and write each term as a share times a percent change.

The notebook’s contributions. Great Recession: investment, 16.3 percent of GDP, fell 29.3 percent, a contribution of −4.79 points, more than the whole fall in GDP; consumption, 67.0 percent of GDP, fell 2.3 percent, −1.51 points. Covid-19: consumption, 67.3 percent of GDP, fell 10.2 percent, −6.86 points.

Imports fell 17.9 and 19.9 percent, which added 2.50 and 2.88 points: part of what households and firms stopped buying had been made abroad.

Real GDP is in chained dollars, which do not add up exactly, so the sum misses GDP growth by a small residual (in parentheses).

Recessions are mostly investment slumps

Recessions happen when households stop spending.

In the average recession since 1953 consumption rose 0.4%, while investment fell 12.9%. In 2008–09 investment alone subtracted 4.79 points from GDP, three times as much as consumption. Only 2020, when shops and travel shut, was led by consumption.

Numbers: the notebook’s Table 10.1 and the contributions of equation (13).

Why investment swings so much: firms can postpone new machines and buildings, and households can postpone new houses (residential investment is part of investment). Postponing consumption is harder, and households smooth it on long-run income.

The exception proves the point: in 2020 consumption fell because the goods and services themselves were not available, and because people feared infection.

Inflation after the pandemic

When did inflation return after the pandemic began, and how high did it go?

Ask: did prices start rising in 2020, when the pandemic began, or later? Which economy had the highest inflation: the U.S., the U.K. or the euro area?

Inflation came back a year into the pandemic, highest in Europe

Data: BLS, CPI-U (CPIAUCNS), via FRED; ONS, CPIH, via the OECD prices database (DSD_PRICES@DF_PRICES_ALL); Eurostat, HICP, euro area (prc_hicp_minr). Chart inspired by Fig. 10.10 in Jones (2024), Macroeconomics, 6th ed.

How to read it: 12-month inflation in the U.S., the U.K. and the euro area, monthly. The slider switches the measure: the log change 100 × ln(P_t/P_t−12), which the textbook’s figure uses, or the usual percent change.

Do live:

1. On “log change”: the 2022 peaks, U.S. 8.7, U.K. 9.2, euro area 10.1 percent.
2. Switch to “percent change”: the peaks rise by about half a point, to 9.1, 9.6 and 10.6. Equation (4): π = 100(e^(π_log/100) − 1); the gap grows with the square of inflation, about 0.02 points at 2 percent and about 0.5 at 10.
3. Point at 2015–2019, between about −0.5 and 3 percent, and at 2020, when euro-area inflation went below zero.

Gaps: no U.S. CPI for October 2025 (government shutdown); the U.K. figure for September 2026 was not yet published.

Prices did not jump when the pandemic began

The pandemic sent prices up right away.

In 2020 inflation fell, and in the euro area prices fell outright. Inflation first passed 3% in April 2021 in the U.S., and in September and October 2021 in the euro area and the U.K. It peaked at 8.7% (U.S., June 2022) and 10.1% (euro area, October 2022).

Numbers: the notebook’s Fig. 10 key numbers (log changes). The textbook’s “more than 8 percent in the United States and more than 10 percent in the euro area”.

Since the pandemic did not raise inflation at once, something else must explain 2021–22: the textbook lists the candidates and returns to them in Chapter 12 with the short-run model (large fiscal transfers, supply bottlenecks, the shift of demand from services to goods, and the energy shock after Russia invaded Ukraine). Economists still debate their weights.

By 2026 inflation was back to about 3–4 percent: U.S. 3.67 (September 2026), U.K. 3.19 (August), euro area 3.69 (September), still above the 2 percent the central banks aim for.

Beyond the three (fig. 10b): Germany 10.9 percent in October 2022, Brazil 11.5 in April 2022, Estonia and Hungary above 22 percent; China, Japan and Switzerland below 4.5 percent all through 2021–23.

Core inflation

Take food and energy out of the price index. Does Europe’s inflation in 2022 still exceed America’s?

Headline inflation peaked higher in the euro area (10.1%) than in the U.S. (8.7%). Predict the ranking of core inflation in 2022.

Core inflation leaves out food and energy, whose prices jump with harvests and oil and gas markets. Ask students to rank the three economies by core inflation in 2022, and to give a reason.

Without food and energy, U.S. inflation was the highest of the three

Data: BLS, CPI less food and energy (CPILFENS), via FRED; ONS, CPIH excluding food and energy, via the OECD; Eurostat, HICP excluding energy, food, alcohol and tobacco (prc_hicp_minr); other economies from Eurostat, the IMF CPI dataset and the OECD. Chart inspired by Fig. 10.11 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the menu shows the core rates of the three economies, as in the textbook (“core, all three”), or the headline and core rates of one economy (46 in all). Twelve-month changes, in percent.

Do live:

1. “core, all three”: the U.S. core rate peaks at 6.4 percent in September 2022, above the U.K. and the euro area in 2022.
2. Choose “Euro area”: the gap between headline and core is largest there, up to 5.2 points in October 2022, because energy prices soared after Russia invaded Ukraine.
3. Choose “Estonia”: food and energy added up to 11.3 points in 2022. Choose “Türkiye”: core inflation peaked at 55.9 percent in April 2024, and the axis widens.
4. Data hygiene: choose “Japan”: no data for 2022, because the OECD’s monthly Japanese core index stops in June 2021.

Europe’s inflation was mostly energy; America’s was broader

• The answer: the ranking flips. In 2022 U.S. core inflation reached 6.42%, the U.K.’s 5.66% and the euro area’s 5.05%.
• Food and energy added up to 5.2 points to euro-area inflation in 2022, against 2.9 in the U.S.
• Core inflation in the U.K. and the euro area kept rising into 2023 (6.29% and 5.51%), after the textbook was written.

Core inflation separates a shock to a few prices from a rise in prices across the board.

Numbers: the notebook’s Fig. 11 key numbers. U.S. core peak 6.42 percent in September 2022, the textbook’s “just over 6 percent”. Largest 2022 gaps between headline and core: U.S. 2.92 points (June 2022), U.K. 3.52 (October), euro area 5.19 (October). U.K. core peaked at 6.29 percent in May 2023, euro-area core at 5.51 percent in March 2023.

Why it matters for policy (Chapter 12): an energy shock raises headline inflation and may fade on its own; broad inflation in services and wages points to demand pressure and is what central banks fear will persist. Core is a guide, not a rule: high energy prices feed into the cost of everything else with a lag, which is one reason European core inflation peaked later.

The Great Recession was a typical financial crisis in markets, milder in output

	U.S. 2007–09	average financial crisis
Real house prices	−35.7%	−35.0%
Real stock prices	−51.7%	−56.0%
Unemployment rate	+5.6 pp	+7.0 pp
Real GDP	−3.8%	−9.3%

Asset prices fell as much as in the typical crisis; output and jobs fell less. Was that luck, or policy?

The notebook’s comparison with the textbook’s Table 10.2, which reports the averages of twentieth-century financial crises from Reinhart and Rogoff (“The Aftermath of Financial Crises”, 2009, American Economic Review Papers and Proceedings): the Great Depression, Japan in the 1990s, Sweden in 1991, the Asian crisis of 1997 and others. We do not have their data, so the right column is their averages, and the left column is ours: Figs. 1 and 4 for the asset prices, Fig. 9 for unemployment and Table 10.1 for GDP.

The headline: financial crises are long and costly. In the average one, unemployment rises by 7 points over almost five years and real GDP falls by close to 10 percent. The U.S. recession was deep, but milder than that.

One reason may be the policy response: the Fed’s lending and rate cuts, the bank rescues and fiscal stimulus. That is the next debate.

How much of their own money should banks put in?

Regulators set a floor on equity as a share of assets. How high should it be?

What the framework says

• The capital ratio is the cushion: \(f^* = E/A\).
• At 10% equity a bank survives a 10% fall in its assets; at Bear Stearns’ leverage, 2.8%.
• More equity also makes runs harder to start.

What is debated

• Banks: equity is costly, so lending shrinks.
• Critics: most of that cost is private, not social.
• Higher floors push risk to less-regulated lenders.

The left column is Act II: equation (7) says the fall that wipes out the owners equals the capital ratio, so a capital requirement is a ceiling on leverage (6 percent ↔︎ ℓ ≤ 15.7) and a floor on the cushion. The bank-run model adds that more equity E raises the run needed to sink a bank.

After 2008 the Basel III accords raised the minimum equity banks must hold and added a simple leverage-ratio floor alongside the risk-weighted rules.

The research. Admati and Hellwig (2013, The Bankers’ New Clothes, Princeton University Press) argue that equity is expensive for bankers mainly because debt is subsidized (tax deductions and implicit guarantees), and that much higher equity, a multiple of the pre-crisis levels, would cost society little. Miles, Yang and Marcheggiano (2013, Economic Journal) estimate the costs and benefits and find an optimal capital ratio far above the pre-crisis requirements. Others warn that higher requirements raise loan rates and move lending to “shadow banks” that the rules do not cover. The level is contested; the direction since 2008 has been up.

Ask the policy students: if you were the regulator, would you rather set one simple leverage limit, or many risk-weighted ones that banks can game?

Should the central bank and the government rescue banks in a panic?

In 2008 the Fed lent to banks and AIG, and Congress created the TARP rescue fund. Lehman Brothers was allowed to fail. Was rescuing right?

The case for rescues

• A run can sink a solvent bank (Act II).
• Lending against illiquid assets removes the fire sale.
• The spread peaked the month after Lehman failed.

The case against

• Rescues reward risk-taking.
• In a panic, illiquid and insolvent look alike.
• Uninsured lenders still run: SVB in 2023.

The left column is the bank-run model: with δ* = 18.2 percent a full run sinks our solvent bank, and a lender of last resort that lends against illiquid assets spares it the fire sale. Deposit insurance (FDIC, since 1933) removes the reason to run. Fig. 3: the TED spread peaked at 3.35 points in October 2008, the month after Lehman failed.

Bagehot’s rule (Lombard Street, 1873): in a panic, lend freely, at a penalty rate, against good collateral, to solvent institutions. The hard part is the last word.

The research. Diamond and Dybvig (1983, Journal of Political Economy) give the case for deposit insurance. Gorton and Metrick (2012, Journal of Financial Economics) show that the 2007–08 panic was a run on short-term “repo” funding of banks, outside deposit insurance. Bernanke (2018, Brookings Papers on Economic Activity) argues that the panic in funding markets, more than the housing losses themselves, explains why the recession was so deep. In March 2023 Silicon Valley Bank lost uninsured deposits in a day; the authorities then guaranteed all its deposits (Board of Governors of the Federal Reserve System, 2023, review of the supervision of SVB). Each rescue reopens the moral-hazard debate on the next slide.

Ask the policy students: should deposit insurance cover all deposits, or only small ones?

“Bailouts reward the reckless”

A skeptic says: “Banks earned huge profits from leverage before 2007, and taxpayers covered their losses after. If every big bank knows it will be rescued, the next one will take even bigger bets.”

Reply: the worry is real, but letting a leveraged bank fail in a panic punishes more than its owners. Each dollar it loses forces \(\ell\) dollars of fire sales that hit healthy banks too. The better answer is to make owners bear losses before a crisis, with more equity, and to rescue the system, not the shareholders.

Take the objection seriously: it is the moral-hazard argument, and the evidence supports it. Kelly, Lustig and Van Nieuwerburgh (2016, American Economic Review) find that options on the financial sector in 2007–09 were priced as if the government would cap losses for the sector as a whole, an implicit guarantee investors could see. Large banks borrow more cheaply than small ones, which is consistent with a too-big-to-fail subsidy.

The reply uses Act II. Equation (8): every dollar of losses forces ℓ dollars of sales; with many banks selling the same assets, prices fall for everyone. The TED spread’s peak came after Lehman failed, not before. Veronesi and Zingales (2010, Journal of Financial Economics) estimate that the October 2008 capital injections raised the value of banks’ debt and equity by more than they cost taxpayers, because they stopped a run, while warning that the gains went largely to bondholders.

Rescues can be designed to hurt owners: in 2008, shareholders of Bear Stearns, Fannie Mae, Freddie Mac and AIG lost most of their stake. Capital requirements (previous policy slide) are the ex ante answer.

Ask: what is the difference between rescuing a bank and rescuing its shareholders?

What to remember from this chapter

• A housing bust (−35.7% in real prices) started the crisis; spreads exploded and stocks lost half their value.
• Leverage multiplies owners’ returns by \(1+\ell\); a fall of \(1/(1+\ell)\) wipes them out; each $1 lost forces \(\ell\) dollars of sales. Runs can sink a solvent bank.
• The Great Recession was long (77 months to regain the jobs); the pandemic slump was deeper but short.
• Inflation returned a year into the pandemic; in Europe it was mostly food and energy.
• Measure carefully: percent vs points, log vs percent changes, which series, which vintage.

Ask each student to write one sentence answering the opening question: how did a fall in house prices break the banks, and how was the pandemic slump different? A good answer: banks held mortgage securities with high leverage, so a modest fall in their value wiped out equity, forced fire sales and froze lending; the pandemic was a sudden stop with no broken balance sheets, so the economy came back fast.

For the policy students, three habits: ask who holds the losses and with how much leverage; separate a liquidity problem from a solvency problem; and judge a slump by its depth times its length.

Walk the three branches of the mind map in about a minute each: the crash of 2007–09 (cyan), why a housing bust broke banks (lavender), two recessions compared (green). Click the image to zoom in.

Leverage turned a housing bust into a long slump; the pandemic hit harder but healed faster

Next: Chapter 11, The IS Curve

This chapter collected the facts of two recessions and the finance behind the first. Chapters 11 to 14 build the short-run model that explains them: the IS curve (how interest rates and shocks move output), monetary policy and the Phillips curve, and aggregate demand and supply. Chapter 14 returns to the financial crisis with the full model.

Sources: data and figures

• Figs. 1 and 4, house and stock prices. Robert J. Shiller, U.S. Home Price and Related Data and U.S. Stock Market Data, shillerdata.com (S&P CoreLogic Case-Shiller index; S&P Composite; BLS CPI). Inspired by Figs. 10.1 and 10.4 in Jones (2024).
• Figs. 2 and 3, rates and spreads. Federal Reserve Board, FEDFUNDS, RIFSPPFAAD90NB, DTB3; ICE Benchmark Administration, TEDRATE; NBER (USREC), via FRED. Inspired by Figs. 10.2 and 10.3 in Jones (2024).
• Figs. 5 and 6, the pandemic. Our World in Data (WHO Covid-19 deaths); OECD, Quarterly National Accounts (DSD_NAMAIN1@DF_QNA), OECD Data Explorer. Inspired by Figs. 10.5 and 10.6 in Jones (2024).
• Figs. 7 to 9, jobs, the gap and unemployment. BLS, PAYEMS, UNRATE; BEA, GDPC1; CBO, GDPPOT, via FRED and ALFRED; OECD labour force statistics (DSD_LFS@DF_IALFS_EMP_Q, DSD_LFS@DF_IALFS_UNE_M). Inspired by Figs. 10.7 to 10.9 in Jones (2024).
• Figs. 10 and 11, inflation. BLS, CPIAUCNS, CPILFENS; ONS CPIH via the OECD (DSD_PRICES@DF_PRICES_ALL); Eurostat, HICP (prc_hicp_minr); IMF CPI dataset. Inspired by Figs. 10.10 and 10.11 in Jones (2024).
• Tables and finance. Table 10.1 from BEA/BLS quarterly data; Table 10.2 averages from Reinhart and Rogoff (2009); the hypothetical bank, leverage and bank runs follow Section 10.5 of Jones (2024); the run, spread and loss derivations are the book chapter’s own.
• Textbook. Jones, C. I. (2024). Macroeconomics, 6th ed. W. W. Norton. Book chapter, with the data, code and every figure: Data Science and Macroeconomic Models, Chapter 10.

Sources: research cited

• Causes of the boom. Bernanke, B. (2005), Sandridge Lecture on a “global saving glut” and the U.S. current account, Federal Reserve Board; Mian, A. and A. Sufi (2009), Quarterly Journal of Economics 124(4).
• Financial crises, runs and rescues. Reinhart, C. and K. Rogoff (2009), “The Aftermath of Financial Crises”, American Economic Review 99(2); Diamond, D. and P. Dybvig (1983), Journal of Political Economy 91(3); Bagehot, W. (1873), Lombard Street; Gorton, G. and A. Metrick (2012), Journal of Financial Economics 104(3); Bernanke, B. (2018), Brookings Papers on Economic Activity, Fall; Board of Governors of the Federal Reserve System (2023), Review of the Federal Reserve’s Supervision and Regulation of Silicon Valley Bank.
• Capital and moral hazard. Admati, A. and M. Hellwig (2013), The Bankers’ New Clothes, Princeton University Press; Miles, D., J. Yang and G. Marcheggiano (2013), Economic Journal 123(567); Kelly, B., H. Lustig and S. Van Nieuwerburgh (2016), American Economic Review 106(6); Veronesi, P. and L. Zingales (2010), Journal of Financial Economics 97(3).
• Jobs in 2020. Giupponi, G., C. Landais and A. Lapeyre (2022), Journal of Economic Perspectives 36(2); Giupponi, G. and C. Landais (2023), Review of Economic Studies 90(4); Cahuc, P., F. Kramarz and S. Nevoux (2018), IZA Discussion Paper; Ganong, P., P. Noel and J. Vavra (2020), Journal of Public Economics 191.
• Lives and lockdowns. Hall, R., C. Jones and P. Klenow (2020), “Trading Off Consumption and COVID-19 Deaths”, Federal Reserve Bank of Minneapolis Quarterly Review 42(1); Goolsbee, A. and C. Syverson (2021), Journal of Public Economics 193.