In 2009, the worst year of the Great Recession, U.S. productivity grew 3.5%

+3.5% Growth of U.S. total factor productivity in 2009, once changes in utilization are removed, in a year when business output fell 3.75%

The first modern business cycle models said that recessions are years of bad productivity. So what happened in 2009? And if productivity is not the whole story, what else moves the economy, and how would we know?

Open with the number. It is John Fernald’s utilization-adjusted TFP growth for the U.S. business sector in 2009: +3.53 percent, while business output fell 3.75 percent. The notebook’s “Key numbers: TFP in recessions” and the 2009 row of the annual table.

The puzzle: the real business cycle (RBC) models of the 1980s, the first DSGE models, explained booms and recessions as good and bad years for productivity. 2009 should then have been a terrible productivity year. In the cleaned-up data it was one of the best.

The chapter is about how macroeconomists model the business cycle today: dynamic, stochastic, general equilibrium (DSGE) models. We will measure the shocks, solve the most important block of these models by hand (the labor market), and see how sticky wages and prices change what each shock does.

Ask the room: when you hear “productivity fell”, what do you picture? Lost knowledge? Idle machines? Hold the answers; slide 10 sorts them out.

Today: three questions

• Do productivity shocks cause recessions? Real business cycle models, U.S. TFP since 1950, the Solow residual and 120 countries.
• How does the labor market of a DSGE model respond to a shock? Productivity, a sales tax and government purchases.
• What do sticky wages and prices change, and by how much? Monetary policy, good news, impulse responses and three policy debates.

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

The headline result: many shocks drive booms and recessions, and frictions decide how each one spreads. With flexible prices only TFP shocks look like a business cycle; sticky prices let money move jobs and wages together.

Where this sits in the book: Chapters 9 to 14 built the short-run model (IS, MP, Phillips curve, AS/AD). DSGE models are the research version of the same story, built from households and firms that optimize. Chapter 16 then studies the consumption decision that this chapter takes as given.

Real business cycles

Act I

A DSGE model follows optimizing people through time while shocks keep arriving

Letter	What it means	In the model
Dynamic	the economy is followed period after period	saving today is next quarter’s capital
Stochastic	random shocks keep hitting it	TFP, money, taxes, spending, energy, news
General Equilibrium	every market clears at once	wages, interest rates and prices are all endogenous

• Agents: households, firms, a government and a central bank.
• Frictions shape how a shock spreads: sticky prices, adjustment costs, missing markets.

A quarterly model of 50 years has 200 periods and 400 linked choices per household: it is solved on a computer.

The three building blocks of the Model overview: agents with objectives, shocks, and features of the economy (frictions).

Why the arithmetic: 50 years × 4 quarters = 200 periods, and in each one a household chooses consumption and hours, so 400 decisions, all linked, because saving becomes capital, capital changes tomorrow’s wage, and the wage changes tomorrow’s choices.

Shocks have three dimensions: size and persistence (temporary or permanent, fading fast or slowly); timing (today, or news today about the future); and uncertainty itself, which can delay hiring and investment because waiting has value.

The chapter’s two shortcuts: solve one block (the labor market) by hand, and read the numerical results of a large estimated model.

The first DSGE models were the Solow model with productivity shocks

\[Y_t = A_t K_t^{1/3} L_t^{2/3} \qquad\qquad \Delta K_{t+1} = \bar{s}\,Y_t - \bar{d}\,K_t\]

• TFP is no longer a constant \(\bar{A}\): it moves around it, hit by shocks. The shocks are real, not monetary: hence real business cycle (RBC) models.
• Impact: a 1% TFP shock raises output 1% with fixed labor, and about \(3/2\)% when people also choose their hours.
• Propagation: higher output means more investment and more capital, so a lasting shock raises GDP for years, and at its peak by more than the shock.

If TFP shocks drive the cycle, measured TFP growth should be low in recessions. That is a test.

The RBC models of Finn Kydland and Edward Prescott (1982) and of John Long and Charles Plosser (1983). Kydland and Prescott shared the 2004 Nobel Prize partly for this work.

The 3/2 comes from the labor-market block of Act II: TFP raises output directly, and again by drawing more people into work, 1 + (2/3)(3/4) = 3/2. Propagation is the transition dynamics of the Solow model (Chapter 5): capital is a stock, built up by investment.

A negative TFP shock need not mean forgotten knowledge. TFP measures how well inputs are used, so a new tax, a bad regulation or a disruption of trade can lower it. Act II makes this exact with a sales tax.

What survived from RBC models is the method: a growth model with optimizing agents, shocks and numerical answers. Adding other shocks and frictions turned them into today’s DSGE models, of which RBC models are a special case.

U.S. productivity

Does U.S. productivity growth fall when the economy is in a recession?

Ask the room first. Most students expect yes: recessions are bad times, so firms must be producing less per unit of input. Ask for a guess about 2009, the worst year of the Great Recession. Collect two or three guesses before moving on.

U.S. TFP growth swings widely, but not in step with recessions

Data: John Fernald, Federal Reserve Bank of San Francisco, utilization-adjusted TFP growth of the U.S. business sector (dtfp_util, 2026 vintage); NBER recession dates (USREC), via FRED. Chart inspired by Fig. 15.1 in Jones (2024), Macroeconomics, 6th ed.

How to read it: annual TFP growth of the U.S. business sector, in percent per year, from Fernald’s series, which removes changes in utilization (slide 10 explains why). Shaded bands are NBER recessions; the dashed line is zero. Hover to read a year.

Do live:

1. Point at the range in the title: from −2.4 percent (1993) to +4.0 percent (1953).
2. Hover over 2009: +3.5 percent, inside the Great Recession band.
3. Hover over 2001 and 2020: +2.3 and +1.8 percent, both recession years.
4. Hover over 1974 and 2008: two recession years in which adjusted TFP did fall.

Data caveat: Fernald revises the series as the BEA and BLS revise theirs, so recent years differ from the textbook’s vintage (2021 is 1.26 percent here, about 0.45 in the textbook’s figure).

TFP shocks are large, but they miss the recent recessions

• TFP growth ranges from −2.4% to +4.0% a year and is negative in 16 of the 73 years from 1950 to 2022. The shocks are real and big.
• But in 2001 and 2020 TFP grew 2.3% and 1.8%, and in 2009 3.5%, one of the largest values in the series.
• Business output fell in 11 years; adjusted TFP fell in only 2 of them (1974 and 2008).

TFP shocks cannot be the whole story: DSGE models had to add other shocks and frictions.

Numbers: the notebook’s Fig. 1 “What to notice” and “Key numbers: TFP in recessions”. Negative years 1950–2022: 1959, 1969, 1974, 1979, 1983, 1987, 1993, 1994, 2006, 2007, 2008, 2010, 2011, 2013, 2016, 2022. Of the 11 years in which business output fell, the raw Solow residual fell in 9, the utilization-adjusted series in 2.

The textbook makes the same point with its vintage: no negative TFP shock in 2001 or 2020, and a big positive one in 2009. Ask: what was the big positive shock of 2009? One answer: firms shed their least productive workers and activities, so measured productivity per hour rose. Another: the adjustment itself, which the next slides take apart.

The other shocks the models added: monetary policy, government purchases, taxes, energy prices, financial frictions, markups, and news about the future.

Measured TFP is a residual, and idle capacity hides inside it

\[g_A = g_Y - \tfrac{1}{3}\,g_K - \tfrac{2}{3}\,g_L\]

With utilization \(u_t\), the inputs used are \(u_tK_t\) and \(u_tL_t\), so the raw residual is

\[\underset{\text{raw residual}}{g_Y - \tfrac{1}{3}g_K - \tfrac{2}{3}g_L} \;=\; \underset{\text{true TFP}}{g_A} \;+\; \underset{\text{utilization}}{g_u}\]

U.S. business sector, 2009	output	raw residual	utilization	adjusted TFP
growth, %	−3.75	−0.15	−3.68	+3.53

Growth accounting (Chapter 6), applied year by year: take logs of the production function and difference. TFP growth is what is left of output growth after capital (weight 1/3) and labor (weight 2/3). Worked example from the notebook: output +3%, capital +3%, hours +1.5% → g_A = 3 − 1 − 1 = 1 percent.

Step 5 of the RBC block: in a recession firms keep some workers and machines but use them less (shorter shifts, slower lines). With a utilization index u_t, the production function becomes A·u·K^(1/3)·L(2/3), so the raw residual equals true TFP growth plus utilization growth. When utilization falls, the raw residual falls even if technology did not change.

2009: −0.15 − (−3.68) = +3.53 percent. The notebook checks that the two sides of the equation agree to within 0.033 percentage points over 1950–2025.

Fernald estimates utilization from hours per worker and other indicators. The textbook says its line comes from the Solow accounting formula; it is in fact the adjusted series.

Raw versus adjusted TFP

Compare the raw Solow residual with the utilization-adjusted series. Which one rises and falls with GDP growth? Both?

Predict the sign and rough size of the correlation of each measure with GDP growth, 1950–2025.

This is the deck’s Predict card on a data figure. Ask students to write two numbers between −1 and 1. Many will predict that both measures move with GDP. Push them: if utilization is what moves with GDP, what is left once it is removed?

Raw TFP rises and falls with GDP; adjusted TFP does not

Data: John Fernald, Federal Reserve Bank of San Francisco, TFP data, U.S. business sector, annual: output growth (dY), raw Solow residual (dtfp), utilization (dutil), adjusted TFP (dtfp_util), 2026 vintage. Chart inspired by Fig. 15.1 in Jones (2024), Macroeconomics, 6th ed.

How to read it: each point is one year since 1950. Horizontal axis: growth of business output; vertical axis: TFP growth. Left panel, the raw Solow residual; right panel, the utilization-adjusted series. Green points are years in which output fell. The grey line is a least-squares fit in each panel; the title reports the two correlations.

Do live:

1. Hover over the green points in the left panel: almost all have negative TFP growth.
2. Hover over 2009 in both panels: −0.15 on the left, +3.53 on the right.
3. Click “GDP grew” in the legend to hide those years and look only at the falls.

Each panel’s grey line is a least-squares fit to that panel’s own points: slope 0.52 on the left, −0.01 on the right (correlations 0.82 and −0.03, as in the title).

Most of the co-movement of TFP with output is utilization

• The answer: the raw residual has a correlation of 0.82 with output growth; the adjusted series −0.03.
• In the 11 years of falling output, raw TFP fell in 9; adjusted TFP in 2.
• The arrow may run backwards: demand falls, firms use their inputs less, and measured TFP falls.

The original evidence for RBC models was largely a measurement effect.

Numbers: the Fig. 1b title and the notebook’s key numbers.

This does not prove that TFP shocks never matter: Fernald’s adjustment rests on its own assumptions about how utilization relates to hours per worker. But the strong correlation that made RBC models look so good in the 1980s weakens a lot once it is taken into account.

The debate it opened: Galí (1999, American Economic Review) found that hours worked fall after a technology improvement, the opposite of the RBC prediction, and Basu, Fernald and Kimball (2006, American Economic Review) found the same with utilization-adjusted data. Sticky prices explain it (Fig. 8 and the Smets–Wouters slide later). Others dispute the finding; it is still debated.

Around the world

Outside the United States, does measured TFP fall when GDP falls?

Ask: if the U.S. pattern is mostly utilization, what should we see in countries where TFP is measured without any adjustment? Collect guesses: almost always, sometimes, rarely?

In most countries the raw residual falls almost every time GDP falls

Data: Groningen Growth and Development Centre, Penn World Table 11.0 (doi:10.34894/FABVLR), TFP (rtfpna) and real GDP (rgdpna) at constant national prices; Feenstra, Inklaar and Timmer (2015). Chart inspired by Fig. 15.1 in Jones (2024), Macroeconomics, 6th ed.

How to read it: TFP growth in percent per year, for one country at a time, from the Penn World Table, which measures TFP with plain growth accounting (a raw Solow residual). Green diamonds mark the years in which real GDP fell. The menu at the top right picks one of 120 countries; the title counts in how many of the GDP-fall years TFP fell too. The axis widens only when a country needs it.

Do live:

1. Start on the United States: TFP fell in 4 of its 8 years of falling GDP, and grew in 2009.
2. Choose Japan, Italy or Mexico: TFP fell in every year in which GDP fell.
3. Choose Germany or the United Kingdom: falls of 3.8 to 4.5 percent in 2009.
4. Choose Iraq or Kuwait: −105 and −51 percent in 1991, the Gulf War (log changes can go below −100). These are collapses of output, not TFP shocks in the RBC sense.

Without a utilization adjustment, almost every slump looks like a TFP shock

• In the whole table TFP fell in 930 of 990 country-years with falling GDP: in Japan, Italy and Mexico every time.
• The United States is the exception (4 of 8), and its PWT series correlates 0.93 with Fernald’s raw residual, only 0.29 with the adjusted one.
• So the cross-country pattern is what idle capacity would produce, too.

Look at how a series is built before you read causes into it.

Numbers: the notebook’s “Beyond the United States” section. The PWT TFP is a raw Solow residual, like Fernald’s dtfp; only the United States has a utilization-adjusted series, because it needs detailed data on hours per worker and industry inputs that few countries publish.

Germany, Japan, Italy and the U.K. had TFP falls of 3.8 to 4.5 percent in 2009, with GDP falls of 4.7 to 5.9 percent: in plain growth accounting, almost all of the fall in output shows up as TFP.

Data lesson for the policy students: Iraq and Kuwait in 1991 (−105, −51) and Macao in 2020 (−80, casinos closed) are why the axis widens. A residual absorbs everything the inputs do not explain, including wars and shutdowns.

The labor market block

Act II

Firms hire up to the marginal product; households weigh the wage against consumption

\[\text{demand: } w = (1-\tau)\cdot\frac{2}{3}\cdot\frac{\bar{A}K^{1/3}}{L^{1/3}} \qquad\qquad \text{supply: } L^s = \bar{\ell}\cdot\frac{w}{c}\]

• Demand slopes down: diminishing returns. A sales tax \(\tau\) leaves firms the share \(1-\tau\).
• Supply rises with the wage relative to consumption, so hours stay flat as both grow.
• Consumption \(c\) is taken as given: temporary shocks barely move it.

Consumption is the one variable through which the future reaches today’s labor market.

Equations (5) and (6) of the Model overview, in Jones’s notation. The marginal product of labor is the derivative of Y = Ā K^(1/3) L^(2/3) with respect to L: (2/3) Ā K^(1/3) L^(−1/3). With no tax, τ = 0.

Where the supply curve comes from (the textbook’s appendix): a household maximizes ln c − (1/2)(1/ℓ̄) L² subject to c + S̄ = wL, with saving S̄ given. The first-order condition is w/c = L/ℓ̄: the gain from one more hour is the wage valued at the marginal utility of consumption, 1/c; the cost is the lost leisure. A richer household, with higher c, values extra income less and works less at any wage.

The shortcut: a full DSGE model chooses the whole path of consumption by looking ahead. Here c is a given number. Shocks that change lifetime income (lasting government purchases, news) act through c. Chapter 16 studies the consumption decision itself.

Setting the two wages equal gives employment and the wage

\[L^* = \left[(1-\tau)\cdot\frac{2}{3}\cdot\frac{\bar{A}K^{1/3}\,\bar{\ell}}{c}\right]^{3/4} \qquad\qquad w^* = \frac{c\,L^*}{\bar{\ell}}\]

Baseline	\(\bar{A}\)	\(K\)	\(\bar{\ell}\)	\(c\)	\(\tau\)	\(L^*\)	\(w^*\)	\(Y^*\)
illustrative values	1	1	0.9	0.6	0	1.00	0.67	1.00

Consumption is 60% of output and labor earns two thirds of it, as in the U.S. data.

Equation (7), the four steps of the Model overview: write supply as a wage, w = cL/ℓ̄; set it equal to the demand wage; multiply by L^(1/3) to get L^(4/3); raise to the power 3/4.

The baseline: the bracket is (2/3) × 0.9/0.6 = 1, so L* = 1, w* = 0.6/0.9 = 2/3, Y* = 1. The notebook checks that labor demand and supply are both 1.000 at w* and that the household’s best choice of hours is 1.0000.

The textbook draws its diagrams without numbers. Every live figure of this act starts from these values, so the numbers on the charts are illustrations of the mechanism, not estimates.

Labor’s share: along the demand curve with no tax, w = (2/3)Y/L, so wL/Y = 2/3 at every point, as in the Cobb–Douglas model of Chapter 4.

The baseline labor market

What does a rise in consumption do to the labor market, if nothing else changes?

Consumption rises from 0.6 to 0.7, with TFP and capital unchanged. Which curve moves, and which way? Do employment and the real wage rise or fall?

Ask students to commit before the reveal. Common answers: “more spending means more jobs” (demand thinking) or “nothing happens, consumption is not in the labor market”. The model has its own answer: look at where c appears.

Labor demand and supply cross at employment 1 and a real wage of 0.67

Model figure: \(w = \tfrac{2}{3}\bar{A}K^{1/3}/L^{1/3}\) and \(L^s = \bar{\ell}w/c\) with \(\bar{A} = K = 1\) and \(\bar{\ell} = 0.9\); the slider sets consumption \(c\). Adapted from Fig. 15.2 in Jones (2024), Macroeconomics, 6th ed.

How to read it: employment L across, the real wage w up. The blue curve falling from the left is labor demand; the teal rising curve is labor supply; the dot is the equilibrium, with its values in the title. The slider sets consumption c.

Do live:

1. Start at c = 0.6: L₀ = 1.00, w₀ = 0.67, Y = 1.00.
2. Move to c = 0.7: the supply curve moves up and to the left; demand stays put.
3. Move to c = 0.5: supply moves down and to the right.

Point at the shape of labor demand: curved and quite flat at the right.

Higher consumption lowers labor supply: employment falls, the wage rises

• The answer: \(c\) appears only in labor supply. At \(c = 0.7\) employment falls to 0.89 and the wage rises to 0.69.
• Richer households “buy” more leisure: they work less at every wage.
• \(\bar{A}\) and \(K\) appear only in demand. Demand is flat: a 1% lower wage raises the labor firms want by about 3%.

Ask of every shock: which curve does it enter?

The c = 0.7 point is the same equilibrium as Fig. 9 later (good news): equation (7) with c’ = 0.7 gives L = (0.6/0.7)^(3/4) ≈ 0.89 and w = 0.7 × 0.89/0.9 ≈ 0.69.

The demand elasticity: solve equation (5) for L, L^d = [(2/3)ĀK^(1/3)/w]³, so the elasticity is −3. The notebook’s check: a 1 percent lower wage raises labor demanded by 3.061 percent. This flat demand curve is why the wage moves little when labor supply shifts.

Also try ℓ̄ (in the book’s widget): it enters supply like 1/c, so a higher ℓ̄ has the same effect as lower consumption.

A productivity shock

Why did the first DSGE models put productivity shocks at the centre of the business cycle?

TFP falls temporarily, from 1 to 0.8. Which curve moves? Do employment and the real wage rise or fall, and by how much?

Ask for a direction for each, and then a rough size. Remind them: a temporary shock barely changes consumption, so supply stays put. Some students will expect the wage to fall but employment to stay (people still need jobs).

A fall in TFP shifts labor demand down along the same supply curve

Model figure: the baseline labor market; TFP after the shock \(\bar{A}'\) set by the slider, consumption unchanged. Adapted from Fig. 15.3 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the dashed grey curve is labor demand before the shock; the blue curve is demand after it. A is the old equilibrium, B the new one. The small arrows on the axes show how the real wage and employment move; the title gives the numbers.

Do live:

1. Start at Ā′ = 0.8: employment 1.00 → 0.85, real wage 0.67 → 0.56.
2. Move to 0.7: a bigger recession.
3. Move above 1, to 1.1 and 1.2: both rise, a boom.

Lower TFP cuts jobs and wages together, just as a recession does

• The answer: only demand moves. Employment falls from 1.00 to 0.85, the wage from 0.67 to 0.56: both about 15%.
• Why 15%: \(L^*\) rises with \(\bar{A}^{3/4}\), and \(0.8^{3/4} \approx 0.85\); along the unchanged supply curve the wage moves with \(L\).
• A rise in TFP raises both: a boom. Output responds with elasticity \(3/2\).

TFP shocks reproduce the signature of a business cycle: jobs and wages move together.

Numbers: the Fig. 3 title and the notebook’s “What to notice”. The elasticity table later collects these: a 1 percent rise in Ā raises L* and w* by 0.75 percent (the notebook’s check: +0.749 percent) and Y* by 1.5 percent (+1.504).

Why this made RBC models attractive: over actual business cycles employment and real wages tend to rise together in booms. A model in which the cycle is driven by TFP gets that right without any friction.

The textbook’s Exercise 2 asks for a temporary positive TFP shock, such as unusually good weather for farming: both rise.

A sales tax

Can a policy change do to the labor market what a bad productivity year does?

Instead of a TFP shock, firms must pay a temporary 20% tax on their sales. Compare with TFP falling to 0.8: a smaller effect, the same, or a larger one?

Ask for a comparison, not just a direction. Remind them of the demand curve: the firm keeps the share 1 − τ of what an extra worker produces.

A sales tax shifts labor demand down exactly as lower TFP does

Model figure: the baseline labor market with a sales tax \(\tau\) on firms, set by the slider; demand becomes \(w = (1-\tau)\tfrac{2}{3}\bar{A}K^{1/3}/L^{1/3}\). Adapted from Fig. 15.4 in Jones (2024), Macroeconomics, 6th ed.

How to read it: as in Fig. 3, the dashed grey curve is the old labor demand, the blue curve the new one, A the old and B the new equilibrium. The slider sets the sales tax τ.

Do live:

1. Start at τ = 0.2: employment 0.85, real wage 0.56. Go back one slide in your head: the same numbers as TFP at 0.8.
2. Move to 0.4: a deeper fall. Move to 0: back to A.

Only the product \((1-\tau)\bar{A}\) matters, so a tax can act like a TFP shock

• The answer: the same. A 20% tax and TFP at 0.8 both give employment 0.8459 and a wage of 0.5639.
• Both lower \((1-\tau)\bar{A}\) from 1 to 0.8, and only that product enters \(L^*\).
• So recession-like falls in jobs and wages need not come from technology at all.

A “negative TFP shock” can be a policy, a regulation or any distortion that works like a tax.

Numbers: the notebook’s “Key numbers: a tax and a TFP shock”, printed to four decimals to show that the two points are identical.

Exercise 5 of the textbook reverses the experiment: a temporary cut in a value-added tax from τ to zero shifts labor demand up and raises both employment and the wage. Act III’s first policy slide comes back to real VAT cuts.

Ask: which real-world events act like a fall in (1 − τ)Ā? Candidates: a new tax, a sudden regulation, a trade embargo, an oil price shock (energy is an input), a breakdown of supply chains.

A productivity shock does not mean forgotten technology

If TFP falls in a recession, people must have forgotten how to produce things.

TFP measures how well inputs are used, not what people know. A 20% sales tax moves the labor market to the same point as TFP at 0.8, and measured TFP also falls when firms leave workers and machines idle.

The textbook raises the same doubt: do we really believe recessions happen because people forget key ideas? Certainly not. Two answers from this chapter:

1. Taxes, regulations and other distortions enter like TFP (Fig. 4).
2. Measured TFP is a residual, and utilization hides inside it (Fig. 1b): the raw residual fell in 9 of the 11 years of falling output, the adjusted series in 2.

Real technology shocks do happen (the textbook’s example is the World Wide Web, a positive one), but most of the year-to-year swings in measured TFP have more mundane causes.

Government purchases

How can government spending that will be paid for later change jobs today?

The government raises its purchases for several years, paid for by future taxes. Households, poorer over their lifetimes, cut consumption from 0.6 to 0.5. Which curve moves? Employment and the real wage?

Government purchases appear in neither curve. Ask students to find the channel: the only forward-looking variable in the block is consumption. Many will say “more spending means more jobs and higher wages”; ask them to check the wage.

Lower consumption shifts labor supply down and to the right

Model figure: the baseline labor market; consumption after the rise in purchases \(c'\) set by the slider. Adapted from Fig. 15.5 in Jones (2024), Macroeconomics, 6th ed.

How to read it: now the dashed grey curve is the old labor supply and the teal curve the new one; labor demand does not move. A is the old equilibrium, B the new one.

Do live:

1. Start at c′ = 0.5 (a permanent rise in purchases of 0.1, 10 percent of output): employment 1.00 → 1.15, real wage 0.67 → 0.64.
2. Move to 0.55: a temporary rise cuts consumption by less, so the effect is smaller.
3. Move to 0.4: a very large rise in purchases.

More government raises employment, but the wage and consumption fall

• The answer: supply moves right. Employment rises from 1.00 to 1.15, the real wage falls from 0.67 to 0.64, and consumption falls too.
• People work more because they are poorer; with capital fixed, extra workers meet diminishing returns.
• The wage falls only a little because labor demand is flat.

Booms with falling wages and consumption do not look like real cycles: government purchases cannot be the main driver.

Numbers: the Fig. 5 title and “What to notice”. With c′ = 0.5, equation (7) gives L = (0.6/0.5)^(3/4) ≈ 1.15 and w = 0.5 × 1.15/0.9 ≈ 0.64.

Why consumption falls: the extra purchases must be paid with taxes at some point, so lifetime after-tax income falls by the present value of the purchases. A permanent rise of 0.1 cuts consumption by the full 0.1; a temporary one by much less (the book chapter’s exercise compares c′ = 0.5 with c′ = 0.58).

The Smets–Wouters model (later in the deck) finds the same pattern: GDP and hours rise, consumption falls, inflation barely moves.

The textbook’s Exercise 4 runs it in reverse: a temporary fall in purchases raises consumption, lowers labor supply, and so lowers employment and raises the wage.

More jobs is not always a boom

Employment is rising, so the economy must be booming.

In Fig. 5 employment rises 15% while the real wage and consumption fall: people work harder because they are poorer. In a typical boom, wages and consumption rise with employment.

A distinction worth practicing: the same movement in hours can come from a shift in labor demand (jobs and wages move together) or a shift in labor supply (they move in opposite directions). To tell them apart, look at the wage.

A historical example to discuss: wartime mobilization, when hours rose strongly while private consumption was squeezed. The textbook mentions a military buildup as its example of higher purchases.

Demand shifts move jobs and wages together; supply shifts split them

A 1% rise in …	\(L^*\)	\(w^*\)	which curve moves	example
TFP \(\bar{A}\)	+0.75%	+0.75%	demand up	TFP shock (Fig. 3)
after-tax share \(1-\tau\)	+0.75%	+0.75%	demand up	tax cut (Fig. 4 reversed)
capital \(K\)	+0.25%	+0.25%	demand up	fixed in the short run
consumption \(c\)	−0.75%	+0.25%	supply up, left	good news (Fig. 9)
labor-supply scale \(\bar{\ell}\)	+0.75%	−0.25%	supply down, right	more willing to work

A permanent TFP rise moves both curves, by \(\bar{A}\) and by \(c\): if both rise 10%, employment does not change.

The table is the notebook’s “What the model predicts”: take logs of equation (7) and read the coefficients as elasticities. The notebook’s numerical checks: +0.749 percent for Ā, −0.743 and +0.249 for c, +0.749 and −0.248 for ℓ̄ (the small differences from 3/4 and 1/4 are because a 1 percent change is not infinitesimal).

The footer is the textbook’s Exercise 3: a permanent TFP rise also raises permanent income and consumption. Ā and c enter equation (7) as the ratio Ā/c, so if both rise 10 percent, employment is unchanged and the wage w = cL/ℓ̄ rises the full 10 percent. That is the long run: wages grow, hours do not.

Frictions, money and numbers

Act III

Money can move jobs only if some nominal price is sticky

\[w = \frac{W}{P} \qquad\qquad W = \bar{W} \qquad\qquad L = \min\left\{L^d\!\left(\frac{\bar{W}}{P}\right),\ L^s\!\left(\frac{\bar{W}}{P}\right)\right\}\]

• With flexible wages and prices the classical dichotomy holds: money changes nominal variables only.
• Sticky wages: the nominal wage is stuck at \(\bar{W}\) while other prices move, so the real wage is set by \(P\).
• Trade is voluntary: the short side of the market decides how much labor is traded.

New Keynesian models add a nominal rigidity; this act adds two, one at a time.

Now we need nominal variables: W, the nominal wage in dollars per hour, and P, the price level; the real wage is W/P. Equation (8) of the Model overview: when the wage does not clear the market, the quantity traded is the smaller of demand and supply.

Why wages might be sticky: contracts, norms against nominal pay cuts, the cost of renegotiating. The idea goes back at least to Keynes. Downward stickiness is the most common.

Sticky wages first (Figs. 6 and 7), then sticky prices (Fig. 8): we will see that they give different answers for the real wage.

Sticky wages

What happens in the labor market when the nominal wage cannot fall?

The nominal wage is stuck at 0.75 with prices at 1, above the market-clearing wage of 0.67. Who decides employment: firms or workers? Is anyone unemployed?

Ask: at a wage above the market-clearing level, do firms want more or fewer workers than at 0.67? Do people want to work more or less? Which side gets its way?

A wage stuck too high leaves willing workers without jobs

Model figure: the baseline labor market with the nominal wage stuck at \(\bar{W}\) (set by the slider) and \(P_0 = 1\); employment is the short side, \(\min\{L^d, L^s\}\). Adapted from Fig. 15.6 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the vertical axis is now the real wage W/P. The green horizontal line is the sticky real wage W̄/P₀. Point A is where firms hire, on the demand curve. The bracket marks unemployment: the gap between what people want to supply and what firms demand at that wage. The unlabelled dot is where the market would clear.

Do live:

1. Start at W̄ = 0.75: employment 0.70, unemployment 0.42.
2. Move to 0.9: a much larger gap.
3. Move to 0.7: a small gap. (Below 0.67, in the book’s widget, labor supply becomes the short side and there is no unemployment.)

A sticky wage creates involuntary unemployment, and a small excess costs many jobs

• The answer: firms are the short side. They hire 0.70, while people would like to supply 1.125: unemployment is 0.42.
• These people want to work at the going wage and cannot find a job: involuntary unemployment.
• Labor demand is elastic (\(-3\)), so a wage 12.5% too high cuts hiring by about 30%.

Sticky wages are one of the ways DSGE models produce unemployment.

Numbers: the notebook’s sticky-wage check. Labor demand at the real wage 0.75 is ((2/3)/0.75)³ = (8/9)³ ≈ 0.702; supply 0.9 × 0.75/0.6 = 1.125; unemployment 0.423.

The 12.5 percent: 0.75 is 12.5 percent above 2/3, and employment is 0.70 instead of 1.00. (8/9)³ ≈ 0.70.

Other ways DSGE models produce unemployment: search and matching frictions in the labor market (Chapter 7’s bathtub), efficiency wages. The textbook mentions search as one of the features added to modern models.

Monetary policy with sticky wages

Can a monetary expansion put these people back to work?

The Fed cuts its rate, spending rises, and prices go up 5%, from 1 to 1.05. The nominal wage stays at 0.75. What happens to the real wage, to employment and to unemployment?

Ask for three directions. Then ask the harder one: in this story, is the real wage higher or lower in the boom?

Higher prices lower the real wage, so firms hire more

Model figure: the sticky-wage labor market with \(\bar{W} = 0.75\); period 0: \(P_0 = 1\); period 1: \(P' = 1.05\); period 2: the nominal wage has adjusted. Adapted from Fig. 15.7 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the slider steps through three periods of one event. The dashed grey line is the old real wage W̄/P₀ with point A; the green line is the current real wage with the current point.

Do live:

1. Period 0: the situation of Fig. 6, unemployment 0.42.
2. Period 1: prices rise 5 percent, the real wage falls to 0.71, employment rises to 0.81 at B, unemployment shrinks to 0.26.
3. Period 2: the nominal wage catches up, W′ = w*P′, the real wage is back at 0.67 and the market clears at C.

With sticky wages, money raises jobs only by cutting the real wage

• The answer: the real wage falls from 0.75 to 0.71, employment rises from 0.70 to 0.81, unemployment falls from 0.42 to 0.26.
• Employment rises by \(1.05^3 \approx 1.16\): its elasticity with respect to the price level is \(+3\).
• But real wages and jobs move in opposite directions; in actual booms they rise together.

Sticky wages let money matter, but they predict a real wage that falls in booms.

Numbers: the notebook’s check. P = 1.05: real wage 0.714, employment 0.813, unemployment 0.258; the ratio of employment, 1.1576, equals 1.05³.

Period 2 is not in the textbook’s figures: once the nominal wage adjusts, the real wage is back at w* and the market clears. In the long run money is neutral again (the classical dichotomy).

The countercyclical real wage was a classic objection to Keynes’s sticky-wage story, raised in the 1930s by Dunlop and Tarshis: in the data, real wages did not fall in recoveries. It is one reason modern models give a big role to sticky prices.

With sticky prices, firms hire the labor that demand requires

\[Y = \bar{A}K^{1/3}L^{2/3} \quad\Longrightarrow\quad L = \left(\frac{Y}{\bar{A}K^{1/3}}\right)^{3/2} \qquad\qquad \frac{W}{P} = \frac{c\,L}{\bar{\ell}}\]

• Each firm keeps its price and sells whatever its customers want: spending decides output \(Y\).
• The production function then decides the labor needed. The wage does not appear: labor demand is a vertical line.
• Wages are flexible, so the real wage lies on the supply curve.

A monetary expansion now works by raising the output demanded.

Equation (9) of the Model overview: invert the production function, L^(2/3) = Y/(ĀK^(1/3)), and raise both sides to the power 3/2.

Elasticities: employment rises 3/2 percent for each 1 percent of extra output demanded, and so does the real wage (with c fixed). A rise in TFP with the output demanded unchanged lowers employment, with elasticity −3/2: firms need fewer workers to make the same goods.

Chapter 12 introduced sticky prices in the short-run model; this is the labor-market view of the same assumption.

Monetary policy with sticky prices

If prices, not wages, are sticky, does a monetary expansion still lower the real wage?

Prices are sticky, wages are flexible. A monetary expansion raises the output demanded by 15%, from 1 to 1.15. What happens to employment and to the real wage?

Ask students to compare with Fig. 7 before answering: will the real wage move the same way as there?

Higher demand shifts the vertical labor demand curve to the right

Model figure: the baseline labor market with sticky prices; labor demand is vertical at \(L = (Y/\bar{A}K^{1/3})^{3/2}\), and the slider sets the output demanded \(Y'\). Adapted from Fig. 15.8 in Jones (2024), Macroeconomics, 6th ed.

How to read it: the vertical blue line is labor demand at the new output; the dashed grey vertical line is the old one. The teal curve is labor supply. A is the start (the flexible-price baseline, Y₀ = 1), B the new point.

Do live:

1. Start at Y′ = 1.15: employment 1.00 → 1.23, real wage 0.67 → 0.82.
2. Move to 1.3: a bigger boom, and a bigger rise in the real wage.
3. Move to 0.9: a monetary contraction, with employment and the wage both falling.

With sticky prices, money moves jobs and real wages in the same direction

• The answer: employment rises from 1.00 to 1.23 (\(1.15^{3/2}\)) and the real wage from 0.67 to 0.82.
• Jobs and wages rise together, as in actual booms: the signature of new Keynesian models.
• Flip side: if TFP rises 10% and demand stays at 1, firms need only 0.867 of labor. Hours fall.

Modern DSGE models combine both flavors: TFP shocks and sticky prices with monetary shocks.

Numbers: the notebook’s sticky-price checks: Y′ = 1.15 gives employment 1.233 (= 1.15^1.5) and a real wage of 0.822; TFP +10 percent with Y = 1 gives employment 0.867.

Employment rises more than output (by the power 3/2) because each extra worker adds less output than the last.

The flip side is the textbook’s Exercise 2(b): with sticky prices, a TFP rise can lower hours at first, because demand rises only slowly. That is exactly what the estimated Smets–Wouters model reports, and what Galí (1999) and Basu, Fernald and Kimball (2006) found in the data. Real business cycle models, with flexible prices, predict the opposite.

Which shocks look like a business cycle?

Shock	prices and wages	what moves	employment	real wage	like a cycle?
TFP falls	flexible	demand down	↓	↓	yes
sales tax rises	flexible	demand down	↓	↓	yes
purchases rise	flexible	supply right (\(c\) falls)	↑	↓	no
monetary expansion	sticky wages	along demand (\(\bar{W}/P\) falls)	↑	↓	no
monetary expansion	sticky prices	vertical demand right	↑	↑	yes
TFP rises, demand fixed	sticky prices	vertical demand left	↓	↓	no

The question “how sticky are prices?” decides which shocks can drive the cycle.

The table is the notebook’s summary of the four blocks. “Like a cycle” means that jobs and real wages rise and fall together, as they tend to do over U.S. business cycles.

Two flavors of DSGE models: TFP shocks with flexible prices (the RBC flavor) and monetary shocks with sticky prices (the new Keynesian flavor). Today’s models include both, plus shocks to government purchases, news, financial frictions and markups. The conventional view: fluctuations come from many shocks.

Ask: which row would you use to describe 2009? (Hint: adjusted TFP rose and hours fell: the last row, with demand falling as well.)

Good news with flexible prices

Can good news about the future make fewer people work today?

People learn that a discovery will raise productivity in five years. Expecting to be richer, they raise consumption today from 0.6 to 0.7. With flexible prices, what happens to employment and the real wage today?

This is the textbook’s Exercise 1. Ask first: does labor demand change today? (No: the technology is not usable yet, and capital is unchanged.) Then: what does higher consumption do to labor supply? Students who remember Fig. 2 should get it.

Good news raises consumption and shifts labor supply up

Model figure: the baseline labor market with flexible prices; consumption after the news \(c'\) set by the slider. Adapted from Fig. 15.14 in Jones (2024), Macroeconomics, 6th ed., the worked solution to its Exercise 1.

How to read it: the dashed grey curve is the old labor supply, the teal curve the new one after the news; labor demand does not move. A is the old point, B the new one.

Do live:

1. Start at c′ = 0.7: employment 1.00 → 0.89, real wage 0.67 → 0.69.
2. Move to 0.8: bigger news, a bigger fall in hours.
3. Move to 0.5: bad news about the future raises hours today.

With flexible prices, good news lowers employment today

• The answer: employment falls from 1.00 to 0.89; the real wage rises a little, to 0.69.
• Nothing about production has changed yet: the whole effect comes through expectations and consumption.
• People work less because they feel richer, so they buy more leisure.

In a DSGE model, the future moves the present through consumption.

Numbers: the Fig. 9 title. Equation (7) with c′ = 0.7: L = (0.6/0.7)^(3/4) ≈ 0.89, w = 0.7 × 0.89/0.9 ≈ 0.69.

A famous puzzle sits here: in the data, good news (for example, booming stock markets on expected technology) tends to come with more hiring, not less. Getting news shocks to produce booms needs extra frictions; the next figure shows the simplest one.

Good news with sticky prices

Do sticky prices change the effect of the same good news?

Same news, same rise in consumption to 0.7, but prices are sticky: firms must produce the extra consumption, so the output demanded rises from 1 to 1.1. Does employment still fall?

Ask: with sticky prices, what decides labor demand? (The output demanded.) Then: what happens to labor supply? (Still up and to the left.) Which effect wins for employment? For the wage?

With sticky prices, the same good news raises employment

Model figure: good news raises consumption to \(c' = 0.7\); the slider switches between flexible prices (Fig. 9) and sticky prices, where output demanded rises to \(Y' = 1.1\). Adapted from Fig. 15.15 in Jones (2024), Macroeconomics, 6th ed., the worked solution to its Exercise 1(d).

How to read it: on “sticky”, labor demand is the vertical blue line, shifted right from the dashed grey one; labor supply has moved up and to the left (dashed grey is the old supply). On “flexible”, it is Fig. 9 again.

Do live:

1. Start on “sticky”: employment 1.00 → 1.15, real wage 0.67 → 0.90.
2. Switch to “flexible”: employment 0.89, real wage 0.69.
3. Switch back and forth: the same news, opposite effects on jobs.

One friction flips the sign of the response

• The answer: with sticky prices employment rises to 1.15 (0.89 with flexible prices) and the real wage jumps to 0.90.
• Firms must produce more, so labor demand moves right; people want to work less, so firms pay much more.
• The same shock, opposite effects on jobs: the answer depends on how sticky prices are.

This is why macroeconomists argue so much about the size of nominal rigidities.

Numbers: the Fig. 10 title and “What to notice”. Employment 1.1^(3/2) ≈ 1.15; real wage 0.7 × 1.15/0.9 ≈ 0.90.

The textbook’s lesson from comparing its Figs. 15.14 and 15.15: the presence of sticky prices can change the effect of a shock on employment completely.

How sticky are prices in the data? Bils and Klenow (2004, Journal of Political Economy) found that half of U.S. consumer prices change within a few months; Nakamura and Steinsson (2008, Quarterly Journal of Economics) showed that, leaving out temporary sales, prices change much less often. The answer matters for every row of the table two slides back.

Estimated models answer in numbers: impulse responses

\[x_{t+h} = \rho^h\,x_t \qquad\qquad h_{1/2} = \frac{\ln 2}{-\ln\rho}\]

• An impulse response function traces one variable after one shock.
• Smets–Wouters TFP shock: half gone after 16 quarters, so \(\rho = 0.5^{1/16} \approx\) 0.958 and 84% is left after a year.
• Adjustment costs delay the peak: closing 30% of the gap a quarter, it comes 3 quarters late.

Not just which way a variable moves: how much, and for how long.

Equation (1) of the Model overview: if a shock keeps the fraction ρ of its size each quarter, after h quarters it is ρ^h of its original size. Set ρ^h = 1/2 and take logs for the half-life.

The notebook’s checks: ρ = 0.9576, half-life 16.0 quarters; after 1 year 84 percent left, after 4 years 50 percent, after 8 years 25 percent; a simulation matches the formula (quarter 4: 0.8409).

Adjustment costs: the simple illustration y_t = y_{t−1} + λ(x_t − y_{t−1}) with ρ = 0.8. With λ = 1 the response peaks at once, in quarter 0; with λ = 0.3 it peaks in quarter 3, at 0.509. This is a toy, not the Smets–Wouters model, but it shows how Friedman’s “long and variable lags” arise.

“Estimated”: parameters such as ρ, the weight on leisure or the Taylor-rule coefficients are chosen so that the model matches U.S. data as closely as possible (Smets and Wouters use Bayesian methods).

In the Smets–Wouters model, a rate rise cuts GDP most after about a year

Shock	GDP	consumption	hours	inflation
fed funds rate +1 point	−0.2% at once; −0.33% after 3–4 quarters	falls nearly as much as GDP	falls	just under −0.2 points for 2 years
TFP +1%, decaying slowly	peak above +1%	rises much less	falls at first	falls
government purchases +1%	about +1%: a multiplier near 1	falls	rises	barely moves
financial friction +1 point	like a rate rise	largest effect	falls	falls

The labor-market block gets the signs; the estimated model adds timing and size.

These are the responses the textbook reports from its Figs. 15.9 to 15.13, as summarized in the book chapter. The book chapter does not redraw them: they are the output of a large estimated model with sticky prices and wages, investment adjustment costs and many shocks, not data, and reading numbers off printed curves would invent precision.

Read with the deck:

• Monetary shock: like Fig. 8 run in reverse (sticky prices). Most of the effect is gone after 5 years (20 quarters). Consumption falls by more than the permanent-income hypothesis suggests because of investment adjustment costs.
• TFP shock: GDP rises above the shock because capital accumulates (Solow); hours fall at first because prices are sticky (the 0.867 of Fig. 8); inflation falls because potential output rises faster than actual output.
• Government purchases: the Fig. 5 pattern: hours and GDP up, consumption down.
• Financial friction: a wedge between the rate households pay and the fed funds rate. The model was built before the 2008 crisis, so it has no bank balance sheets.

Smets and Wouters (2007, American Economic Review) build on Christiano, Eichenbaum and Evans (2005, Journal of Political Economy). Versions are used by the ECB, the Fed and other central banks.

Averages hide who loses: heterogeneous-agent models

• A representative agent stands for everyone. But job losses in a recession fall on some people, not on all a little.
• With incomplete markets people cannot insure their jobs, and many have little savings: their consumption follows their income.
• HANK models (heterogeneous agent, new Keynesian) keep these differences. A recession is costlier when it hits those with the least.

In HANK models, monetary policy works mainly by changing jobs and incomes, not by changing the reward to saving.

The textbook’s case study on heterogeneous agents. The Great Recession made heterogeneity obvious: falling house prices hurt households differently depending on their balance sheets.

Kaplan, Moll and Violante (2018, American Economic Review, “Monetary Policy According to HANK”) show that in a model with many hand-to-mouth households, the direct effect of a rate cut (people save less because saving pays less) is small; most of the effect comes indirectly, through higher labor income as demand rises. In representative-agent models it is the other way round.

Policy relevance: who bears the cost of a recession, and whether central banks should care about inequality, are now questions that DSGE models can address. Also heterogeneity in reservation wages: even if each person works 40 hours or not at all, the aggregate labor supply elasticity depends on the spread of reservation wages.

DSGE models have room for money and for unemployment

DSGE models assume that money does not matter and that everyone who wants a job has one.

That was the early RBC flavor. With sticky wages the model has involuntary unemployment (0.42 in Fig. 6); with sticky prices a monetary expansion raises employment 23% (Fig. 8). The models central banks use have both frictions.

RBC models are now a special case of DSGE models. Prescott’s 1986 essay already called for adding public finance, a foreign sector and monetary factors, and the profession did.

Unemployment in modern DSGE models comes from sticky wages, from search and matching frictions, or from efficiency wages. Smets–Wouters has both sticky prices and sticky wages.

The 23 percent: employment 1.00 → 1.23 with output demanded up 15 percent.

Temporary VAT cuts: a short-run boost from the tax side

The U.K. cut its VAT from 17.5% to 15% in 2008–09, Germany from 19% to 16% in the second half of 2020. Can a temporary tax cut fight a recession?

What the model predicts

• A cut is Fig. 4 reversed: labor demand shifts up.
• Jobs and wages rise together.
• Announced as temporary, it pulls purchases forward.

What it leaves out

• Whether shops pass the cut into prices.
• A dip in demand when the cut ends.
• Effects at the zero lower bound.

The left column: Exercise 5 of the textbook. Our sales tax is equivalent to a value-added tax when there are no intermediate inputs. Cutting τ to zero raises (1 − τ)Ā by 25 percent if τ was 20 percent, and employment and the wage rise by the same proportion (elasticity 3/4 each).

The pull-forward effect is outside the one-period block: a cut that people know will end makes goods cheap now relative to later, so they buy now (intertemporal substitution), much as a lower interest rate would.

The research. Crossley, Low and Wakefield (2009, Fiscal Studies) analyze the U.K. cut and argue its main effect works by bringing spending forward. Bachmann, Born, Goldfayn-Frank, Kocharkov, Luetticke and Weber (2021, NBER Working Paper 29442) find that the German cut raised household spending, above all on durable goods. Correia, Farhi, Nicolini and Teles (2013, American Economic Review) show that a path of consumption-tax changes can do what monetary policy cannot at the zero lower bound. How much of each cut reached consumer prices, and how much spending was only moved in time, is debated.

Ask the policy students: if you had to choose between a temporary VAT cut and a cheque to every household, which would you pick, and for which households?

How much does a dollar of government purchases raise GDP?

A government expands its purchases in a slump. Will GDP rise by more than the spending, or by less?

What the models say

• Block: hours rise, people are poorer.
• Smets–Wouters: a multiplier near 1.
• Consumption falls in both.

What is debated

• Estimates vary with data and method.
• Maybe larger with rates at zero.
• Who pays, and when, matters.

The left column is Fig. 5 and the Smets–Wouters slide: a temporary 1 percent rise in purchases raises GDP by about 1 percent in the short run; consumption falls because future taxes make households poorer.

The research. Ramey (2011, Journal of Economic Literature, “Can Government Purchases Stimulate the Economy?”) surveys the evidence and finds that most estimates of the aggregate multiplier lie around one, with a wide range. Ramey and Zubairy (2018, Journal of Political Economy) find multipliers below or near one with a century of U.S. data, even in slack times. Nakamura and Steinsson (2014, American Economic Review) use regional military spending and find larger local multipliers, which are not the same as national ones. Christiano, Eichenbaum and Rebelo (2011, Journal of Political Economy) show that in a new Keynesian model the multiplier can be much larger when the central bank’s rate is at the zero lower bound. Present this as contested: the size of the multiplier is one of the most debated numbers in macroeconomics.

Ask: why would the multiplier be larger when the central bank does not raise rates in response?

Central banks forecast with estimated DSGE models, and 2008 tested them

The ECB, the Fed and other central banks use versions of the Smets–Wouters model. How much should a policy decision rest on it?

What the models offer

• A consistent story behind each forecast.
• Answers to “what if we raise rates 1 point?”
• Separate roles for each shock.

What 2008 exposed

• No bank balance sheets before the crisis.
• Strong assumptions on expectations.
• Most work fitted to calm times.

The left column is the chapter’s quantitative theory: an estimated model gives impulse responses for every shock, so a central bank can compare policy paths in numbers, and decompose a recession into shocks.

The research. Smets and Wouters (2007, American Economic Review) and Christiano, Eichenbaum and Evans (2005, Journal of Political Economy) are the workhorse models. Del Negro, Giannoni and Schorfheide (2015, American Economic Journal: Macroeconomics) show that a version with financial frictions, run at the New York Fed, would have forecast the depth of the Great Recession and the small fall in inflation that puzzled many. Blanchard (2018, Oxford Review of Economic Policy, “On the Future of Macroeconomic Models”) argues that DSGE models should be one of several classes of models in a policy toolkit, not the only one; Lindé (2018, same journal) defends their use for policy analysis while listing their gaps.

Ask the policy students: when a staff model and your own judgment disagree, what would make you trust the model?

“DSGE models failed in 2008, so they are useless”

A skeptic says: “The frontier models of 2007 had no banks and did not see the financial crisis coming. Why should we trust them to guide policy now?”

Reply: the criticism of 2007 is fair, but no model forecast the timing of the crisis, and the test of a model is whether it answers “what if” questions well. The framework absorbed the lesson: financial frictions, bank balance sheets and HANK households were added. The alternative to a stated model is not no model, but an unstated one.

Take the objection seriously. Before 2008 most estimated DSGE models had no financial sector; Smets–Wouters has only a wedge between the rate households pay and the fed funds rate. The textbook’s conclusion admits that the precise quantitative predictions often clash with the real world and that the past decade was humbling.

The reply. After the crisis, models added financial intermediaries and their balance sheets (for example Gertler and Kiyotaki, 2010, Handbook of Monetary Economics), and the version of Del Negro, Giannoni and Schorfheide (2015) with financial frictions fits the Great Recession much better. HANK models (Kaplan, Moll and Violante, 2018) add who loses.

The deeper point from the deck: a DSGE model is a disciplined way of asking how a shock spreads, with stated assumptions that can be argued about (how sticky are prices? how elastic is labor supply?). Every forecast uses some model; the virtue of a DSGE model is that its assumptions are on the table.

Ask: what would you need to see to be convinced that a model is useful for policy, even if it cannot predict crises?

What to remember from this chapter

• DSGE models are dynamic, stochastic and general equilibrium; they grew out of RBC models, the Solow model with TFP shocks.
• U.S. TFP growth swings a lot, but once utilization is removed it rose in most recessions (+3.5% in 2009).
• In the labor-market block, demand shifts (TFP, taxes) move jobs and wages together; supply shifts (purchases, news, through \(c\)) split them.
• Sticky wages make money matter but lower the real wage; sticky prices move jobs and wages together, and can flip the sign of a shock.
• Estimated models answer in impulse responses: how much, and for how long.

Ask each student to write one sentence answering the opening question: do productivity shocks cause recessions, and how do sticky wages and prices change the answer? A good answer: measured TFP falls in recessions partly because inputs sit idle, so productivity shocks are at most part of the story; with sticky prices, demand shocks such as monetary policy move jobs and wages together like a real cycle, and a TFP rise can even lower hours.

For the policy students, three habits: ask which curve a policy enters; check the wage, not only employment, to tell supply from demand; and ask how sticky prices are before trusting a model’s sign.

Walk the three branches of the mind map in about a minute each: real business cycles (cyan), the labor market block (lavender), frictions, money and numbers (green). Click the image to zoom in.

Many shocks drive the cycle; sticky prices decide which ones move jobs and wages together

Next: Chapter 16, Consumption

This chapter took consumption as given and let it carry the future into today’s labor market. Chapter 16 opens that box: the permanent-income and life-cycle hypotheses, and why consumption reacts so little to temporary shocks.

Sources: data and figures

• Fig. 1, U.S. TFP growth, raw and adjusted. John Fernald, Federal Reserve Bank of San Francisco, A Quarterly, Utilization-Adjusted Series on Total Factor Productivity, U.S. business sector, annual (dY, dtfp, dutil, dtfp_util, 2026 vintage); NBER recession dates (USREC), via FRED. Inspired by Fig. 15.1 in Jones (2024).
• Fig. 1, 120 countries. Groningen Growth and Development Centre, Penn World Table 11.0 (doi:10.34894/FABVLR), rtfpna and rgdpna; Feenstra, R., R. Inklaar and M. Timmer (2015), American Economic Review 105(10); World Bank country classifications. Inspired by Fig. 15.1 in Jones (2024).
• Figs. 2 to 8, the labor market with shocks, sticky wages and sticky prices. Our own model figures with illustrative values; adapted from Figs. 15.2 to 15.8 in Jones (2024).
• Figs. 9 and 10, good news. Adapted from Figs. 15.14 and 15.15 in Jones (2024), the worked solution to its Exercise 1.
• Impulse responses. Smets, F. and R. Wouters (2007), a Bayesian estimated DSGE model of the U.S. economy, American Economic Review 97(3), as reported in Figs. 15.9 to 15.13 of Jones (2024).
• 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 15.

Sources: research cited

• Real business cycles and TFP. Kydland, F. and E. Prescott (1982), Econometrica 50(6); Long, J. and C. Plosser (1983), Journal of Political Economy 91(1); Prescott, E. (1986), Federal Reserve Bank of Minneapolis Quarterly Review 10(4); Galí, J. (1999), American Economic Review 89(1); Basu, S., J. Fernald and M. Kimball (2006), American Economic Review 96(5); Dunlop, J. (1938), Economic Journal 48(191); Tarshis, L. (1939), Economic Journal 49(193).
• New Keynesian and estimated models. Christiano, L., M. Eichenbaum and C. Evans (2005), Journal of Political Economy 113(1); Smets, F. and R. Wouters (2007), American Economic Review 97(3); Bils, M. and P. Klenow (2004), Journal of Political Economy 112(5); Nakamura, E. and J. Steinsson (2008), Quarterly Journal of Economics 123(4).
• Heterogeneity and finance. Kaplan, G., B. Moll and G. Violante (2018), “Monetary Policy According to HANK”, American Economic Review 108(3); Gertler, M. and N. Kiyotaki (2010), Handbook of Monetary Economics 3; Del Negro, M., M. Giannoni and F. Schorfheide (2015), American Economic Journal: Macroeconomics 7(1); Blanchard, O. (2018) and Lindé, J. (2018), Oxford Review of Economic Policy 34(1–2).
• Tax cuts and multipliers. Crossley, T., H. Low and M. Wakefield (2009), Fiscal Studies 30(1); Bachmann, R. et al. (2021), NBER Working Paper 29442; Correia, I., E. Farhi, J. P. Nicolini and P. Teles (2013), American Economic Review 103(4); Ramey, V. (2011), Journal of Economic Literature 49(3); Ramey, V. and S. Zubairy (2018), Journal of Political Economy 126(2); Nakamura, E. and J. Steinsson (2014), American Economic Review 104(3); Christiano, L., M. Eichenbaum and S. Rebelo (2011), Journal of Political Economy 119(1).