Is unemployment a labor problem or a money problem?
Same headline number, two completely different machines for reading it. One says the trouble is in the labor market. The other says the trouble is in the money. The argument you can’t see is which apparatus you reached for.
Als Debattengraph anzeigenThe labor market did it
“The American economy is changing fast, and too many workers are being left behind. We need to invest in the skills and training that let people compete for the jobs of the future.”
— the standard bipartisan framing, from a generation of Brookings briefs, State of the Union addresses, and op-eds on the “skills gap”
You have heard this sentence in a hundred costumes. It is so familiar it sounds like a description of reality rather than a theory about it. But it is a theory — a specific, confident claim about where unemployment lives. It says joblessness is a labor-market problem: a mismatch between the workers we have and the jobs that exist. Before we question that, let’s build it at full strength, because it is the framing almost every economist and almost every politician reaches for first — and for good reasons.
Start where the discipline starts. The labor market is a market. Firms demand workers, people supply their labor, and a wage clears the two. If everyone who wanted a job at the going wage could get one, there would be no unemployment to explain. So the labor-side question is sharp: why does this particular market fail to clear when so many others do?
Search and matching: friction is built in. The first answer is that the labor market never clears instantly, even when it is working perfectly. Diamond, Mortensen, and Pissarides — the 2010 Nobel — modeled hiring as a search process. Workers and jobs do not meet on a frictionless auction floor; they find each other slowly, through a matching technology that takes time. The result is a baseline rate of unemployment that exists not because anything is broken but because matching takes effort. Even a healthy economy has people between jobs at any moment.
The matching function makes this precise. The flow of new hires depends on the stock of searching workers $u$ and the stock of vacancies $v$:
$$M(u, v) = A\, u^{\alpha} v^{1-\alpha}$$More unemployed workers and more open jobs both produce more matches, but never enough to clear the market in an instant. Frictional unemployment is the geometry of search, not a failure of it.
Think of dating, not auctions. Even in a city full of compatible people and people looking, nobody is matched instantly — it takes time to meet, evaluate, and commit. A labor market always has some people mid-search. That floor of “between jobs right now” isn’t a malfunction; it’s what searching costs.
Sticky wages: the market can’t cut its way to clearing. In a recession, the textbook auction would simply lower wages until every worker found a buyer. It doesn’t happen. Wages are sticky downward — contracts, norms, morale, the refusal of firms to insult their existing staff with pay cuts. So when demand for labor falls, the adjustment comes through fewer jobs rather than lower wages. The Phillips relation captures the short-run trade this leaves behind: when unemployment runs below its baseline, inflation tends to rise; when it runs above, inflation tends to fall.
Inflation $\pi_t$ tracks expected inflation $\pi_t^e$ adjusted by the gap between actual unemployment $u_t$ and its natural rate $u_t^n$. The labor market and the price level are tied together at the seam of the sticky wage.
Monopsony: employers have power, and policy can use that. The simplest labor market assumes firms are price-takers. They aren’t always. Where employers have wage-setting power — a company town, a thin local market, high switching costs — they pay less than a worker’s contribution and hire fewer people than a competitive market would. This is the structure Card and Krueger’s credibility revolution used to explain why a higher minimum wage can sometimes raise employment rather than cut it: it pushes a monopsonist back toward the competitive outcome. The detail matters, because it is the labor-side apparatus generating a counterintuitive policy result on its own terms.
The natural rate is a property of the labor market. Pull these together and you get the keystone. Unemployment fluctuates, but it fluctuates around a level — the “natural rate” — set by the structure of the labor market itself: how fast workers and jobs match, how generous and long unemployment insurance runs, how protected jobs are, how tightly occupations are licensed, how well the skills people have line up with the jobs that exist, the age and education mix of the workforce. Friedman gave us the term in 1968, but the framing was older and the content is labor-side through and through. The natural rate is a statistic the labor market produces.
And here is the payoff: a real policy menu. If unemployment is a labor-market property, the levers are labor-market levers. Raise the minimum wage where monopsony bites. Fund training and re-skilling — Workforce Investment, Trade Adjustment Assistance, Job Corps — for workers whose skills no longer match available jobs. Attach work requirements to benefits. Tune immigration to the labor supply. Strip out occupational licensing that walls people out of trades. Loosen or tighten employment protection. This is not a fringe agenda. It is the menu both American parties have actually reached for, decade after decade, when joblessness rises. The framing has a coherent apparatus behind it and a working set of tools in front of it. That marriage is exactly why it feels like common sense. The wage-flexibility default it rests on — markets clear if you let prices move — goes back to the marginalists, and that lineage lives in History of Economic Thought Ch.5 §5.3 (Marginalist revolution).
“We need to invest in the skills and training that let people compete for the jobs of the future.”
— the bipartisan “retrain the workforce” consensus, ~1990–present
“The answer to unemployment is retraining workers”
It’s the safest sentence in American economic policy: when jobs disappear, train people for new ones. Nobody loses an election saying it. But how much of the unemployment we actually observe is a skills problem the training can fix?
Inside the labor-side conversation
“The labor market is not a single market but a vast collection of local markets, occupations, and skill groups. Who gets hired, at what wage, and how fast depends on the matching technology and the institutions that govern it — not on some aggregate that floats free of all that structure.”
— the labor-economist’s creed, in the spirit of Autor, Goldin, and Manning
This is the framing argued by people who measure labor markets for a living, and it is powerful precisely because it is granular. It explains things aggregates can’t: why Black unemployment runs roughly twice white unemployment in every phase of the cycle, why teen unemployment runs triple the adult rate, why Spain can sit above 15% for a decade while the US sits near 5% under broadly similar monetary conditions. None of that is about the money supply. It is about institutions, frictions, and the structure of who-matches-with-what. The labor economist says: look at the structure, and the structure answers.
“We keep prescribing training because it is the policy that lets us feel responsive without admitting how little of the problem we actually understand. The evidence that adult retraining moves earnings much is, to put it gently, not strong.”
— the internal labor-side critic, in the spirit of Heckman on human-capital timing and Cowen on program efficacy
Notice this critic is still inside the labor-side conversation — not flipping to monetary policy, just turning the apparatus on its own favorite remedy. The point is that “unemployment is a labor problem” does not automatically imply “therefore the labor-policy menu fixes it.” The matching frictions are real; the natural rate is real; but the toolbox we reach for to move them is weaker than its political popularity suggests. That is a serious objection. It is not, however, the objection that breaks the frame open. The objection that does that comes from somewhere else entirely.
Where this leaves us
Within the labor-side framing, unemployment is a labor problem and the policy menu is the labor menu. The apparatus is real: search and matching, sticky wages, monopsony, the natural rate. The empirics are real: structural mismatch, the cross-country spread, the demographic gaps that persist through every cycle. The policy results are real, even where they disappoint. None of that is what is about to be questioned. What is about to be questioned is whether the labor-side apparatus is the only apparatus that opens on the same data — or whether there is a second machine, reading the same headline number, that diagnoses something the first one cannot see. For the deeper labor-side policy fight at the heart of Stage 1 — monopsony, Card-Krueger, the minimum wage — see Do minimum wages cause unemployment?
There is a strange assumption hidden inside every labor-side story we just told. Each one quietly takes the amount of spending in the economy as given — fixed by something offstage — and treats the labor market as merely trying to clear whatever demand happens to exist. But what if the amount of spending is not given? What if it is the thing that decides whether five million Americans spend next year looking for work?
The flip — what if the money did it?
“The Fed thought monetary policy was easy in 2008 because interest rates were low. But low rates are usually a sign that money has been tight. The mass unemployment of 2009–2014 was not a labor-market problem the Fed was powerless to fix. It was a monetary disequilibrium the Fed mistook for a labor-market problem.”
— Scott Sumner, the market-monetarist case, The Money Illusion (blog) and The Money Mirage (2021)
Read that sentence again with Stage 1 fresh in your mind. Everything we just built — the matching frictions, the structural mismatch, the natural rate as a labor-market property — Sumner is calling a misdiagnosis. Not wrong about the labor market in general, but wrong about this: the specific mountain of unemployment after 2008, which the labor-side framing read as friction and skills, and which he reads as the Fed having let spending collapse. Same statistic. A different machine. Let’s build the second machine at full strength — because it has its own apparatus, its own lineage, and its own hard evidence.
The flip has a founding move, and it is Keynes. Involuntary unemployment — the General Theory, 1936 — is the claim that workers willing to work at the going wage can be unable to find jobs because total spending in the economy is simply too low at the prevailing price level. The constraint does not sit inside the labor market at all. It comes from outside it: aggregate demand. Cut wages and you don’t fix it — you might make it worse, because lower wages mean lower incomes mean lower spending. This was the move that made “unemployment is a money problem” a coherent sentence rather than a confusion. The founding statement of the demand-side reading lives in History of Economic Thought Ch.8 §8.1 (The Keynesian revolution).
Friedman 1968 fixed the boundary. This is the single most important move in the whole argument, so it gets its own paragraph. In his 1968 presidential address to the American Economic Association, Friedman accepted that the labor market has a natural rate set by its own structure — conceding the labor side its territory — but then drew the line that has organized macroeconomics ever since. The Phillips curve, he argued, is only short-run. Monetary policy can push unemployment below the natural rate for a while, but only by fooling people with inflation they didn’t expect; once they catch on, unemployment snaps back to natural and the inflation stays. The deep consequence is the one that matters here: the deviation of unemployment from its natural rate is a monetary phenomenon. The level is labor-side; the gap is money-side. Friedman did not collapse the two framings into one — he drew the border between them, and we have argued over exactly where it runs ever since. That boundary-fixing move is the spine of History of Economic Thought Ch.10 §10.1 (The monetarist counter-revolution).
The New Keynesian synthesis bakes it into the workhorse. Modern macro runs on a three-equation model — an Euler equation for spending, a Phillips curve for inflation, a Taylor rule for the central bank. In it, unemployment moves because monetary policy moves: a demand shock or a policy mistake opens an output gap, the gap shows up as unemployment above the natural rate, and the central bank’s response decides how fast it closes. The labor-market structure is held fixed in the background; the cyclical action is monetary. This is not a heterodox view. It is the standard model taught to every graduate student.
The canonical three equations — output, inflation, policy:
$$y_t = E_t y_{t+1} - \sigma^{-1}\,(i_t - E_t \pi_{t+1})$$ $$\pi_t = \beta\, E_t \pi_{t+1} + \kappa\, y_t$$ $$i_t = \phi_\pi\, \pi_t + \phi_y\, y_t$$Nowhere in these equations does a worker re-skill. The cyclical unemployment that maps to the output gap $y_t$ is determined by spending and the central bank’s reaction to it.
There is an even simpler way to see the money-side claim. Total spending in the economy is just the quantity of money times how fast it changes hands: $M \cdot V = P \cdot Y$, nominal GDP. If that nominal flow suddenly drops — people hoard money, the central bank lets it shrink — then either prices fall or output and jobs fall. Prices are sticky, so jobs take the hit. On this reading, a recession is what a shortfall of nominal spending looks like from the unemployment office.
The market monetarists push the flip to its edge. Scott Sumner and the market-monetarist school take the money-side reading further than the mainstream is comfortable with. Their claim: if the central bank had committed to keeping nominal GDP growing on a stable path after 2008 — not just cutting rates, but credibly promising to make total spending recover — the great overhang of unemployment would never have happened. What the labor-side framing diagnosed as “structural” was, to Sumner, monetary disequilibrium wearing a labor-market costume. The relevant counterfactual was never “the Fed has done a lot.” It was “the Fed could have targeted the path of spending and didn’t.” The apparatus that grounds this — the nominal anchor, the central bank’s authority over the spending path — lives in Ch 16 §16.2.
And from the opposite edge, MMT. Modern Monetary Theory flips from the other direction. A currency-issuing government, it argues, never faces a financing constraint the way a household does — it faces a real-resource constraint. So persistent involuntary unemployment is not a fact of nature or a skills mismatch; it is direct evidence that the state has chosen to supply too little nominal spending. The structurally correct response, on this reading, is a job guarantee: the government stands ready to hire anyone who wants work at a fixed wage, putting a floor under employment that the labor-side framing never imagined. Sumner and MMT disagree about almost everything, but they share the move that defines Stage 2: the binding constraint on jobs is nominal, not in the matching technology.
The lineage doesn’t stop here. Where the labor-side framing held a fixed nominal background, the money-side framing’s modern formalism — how a demand shock propagates through unemployment, how the zero lower bound traps the central bank, how the natural rate of interest couples to all of it — is the dedicated business of the New Keynesian chapter. The formal demand-shock-to-unemployment channel and the zero-lower-bound trap are worked out in History of Economic Thought Ch.12 §12.4 (New Keynesian economics).
Two episodes the flip reads better than the labor side alone. Theory is cheap; the money-side machine earns its keep on the evidence. Take 2008–2014. The labor-side framing predicted a fairly quick return to the natural rate once the financial shock cleared — instead, unemployment sat above 7% for five years. The money-side framing predicted exactly that overhang: with the central bank pinned at the zero lower bound and nominal spending well below its old path, demand-short unemployment was supposed to persist, and it did. The 1970s tell the mirror story: stagflation — high unemployment and high inflation at once — was the thing the simple labor-side Phillips relation said couldn’t happen, and it broke the labor-side framing’s monopoly. That episode is the spine of Economic History Ch.16 §16.3 (Stagflation and the neoliberal turn); the 2008–2014 overhang, and the very different 2020 story, run through Economic History Ch.19 §19.4 (The Global Financial Crisis and after).
Then 2020 closed the case with the cleanest natural experiment in living memory. Unemployment hit 14.7% in April. The labor-side instinct — hysteresis, skills atrophy, sectoral reallocation friction — forecast a long, grinding, 1930s-style recovery. Instead, an unusually aggressive monetary-plus-fiscal response pushed nominal spending right back, and unemployment was near 3.5% by early 2022. The recovery’s speed tracked the size of the demand response, not the depth of the labor-market scarring. If you only had the labor-side machine, 2020 made no sense. With the money-side machine, it was almost a prediction.
“Anyone who wants a job and can’t find one is proof that the government has not spent enough. Involuntary unemployment is a policy choice, and the job guarantee is the way to stop choosing it.”
— the MMT case, in the spirit of Stephanie Kelton, The Deficit Myth (2020)
“Unemployment is a money problem, not a labor problem”
Two camps that agree on almost nothing — market monetarists and MMT — agree on this: when people can’t find work, look at the nominal spending in the economy before you look at their skills. How far does that take you?
The flip and its limit
“Never reason from a price change — and never reason from an unemployment rate without asking what nominal spending was doing. The 2008 recession was caused by tight money, full stop, and the recovery was slow because money stayed tight for years while everyone debated retraining.”
— the market-monetarist voice, in the spirit of Scott Sumner, The Money Mirage, 2021
Argued in its own voice, the flip is bracing. Sumner’s charge is that the entire 2009–2014 conversation about skills, mismatch, and structural unemployment was a category error — economists staring at a labor-market thermometer to diagnose a monetary fever. The natural rate didn’t suddenly jump three points in 2008 because American workers forgot how to work. Spending fell, the central bank let it, and the labor market showed the wound. On this reading, the labor-side framing wasn’t just incomplete in 2008. It was pointing the policy response at the wrong organ.
“Grant the gap is monetary. You still have to explain the level around which it moves — and that level is built from institutions, frictions, and demographics the money supply never touches. Worse, a long enough demand shortfall can scar the labor market and raise the natural rate itself, which means the two framings are entangled, not separable.”
— the labor-side defender, in the spirit of Goldin, Manning, and the hysteresis literature
This is not a strawman dismissal — it is the strongest labor-side rebuttal, and it concedes a lot. It grants that the cyclical gap is the money-side framing’s home turf. What it refuses is the takeover. The decomposition the money-side framing relies on — clean natural rate here, clean cyclical gap there — is not as clean as it looks: hysteresis means a deep enough recession can drag the natural rate up behind it, as long-term unemployment scars skills and detaches workers from the market. So the labor side doesn’t just own the baseline; it bleeds into the cycle the money side claims. That entanglement is exactly the problem Stage 3 has to adjudicate.
Where this leaves us
Pure labor-side framing is incomplete. The monetary regime decides whether the labor market gets a chance to clear at all; when it doesn’t, what looks like a labor-market problem can be a monetary disequilibrium misdiagnosed — and 2008 and 2020 are the receipts. But the reverse is also true. The money-side framing cannot say why some labor markets clear at 4% and others at 15% under similar monetary regimes, and it cannot wish away the entanglement the hysteresis literature insists on. So the question is no longer “which framing is right.” It is “which framing applies to which part of the variation” — and that is a question with an answer. For the deeper monetary-policy-authority debate behind Stage 2, see Can central banks control the economy?; for the fiscal side of the same regime question, see Does government spending help the economy?
We are left with two machines that both diagnose the same statistic, and each is right about something the other can’t see. The mainstream’s honest answer is not to pick a winner — it is to name which machine reads which part of the number. Let’s see what that calibration actually looks like when you do it carefully.
The verdict — the level and the gap
“My colleagues and I may have misjudged the strength of the labor market, the degree to which longer-run inflation expectations are consistent with our inflation objective, or even the fundamental forces driving inflation. … Our understanding of the forces driving inflation is imperfect.”
— Janet Yellen, “Inflation, Uncertainty, and Monetary Policy,” September 2017
This is the chair of the Federal Reserve, in public, admitting she is not sure where the natural rate is or how the Phillips curve runs. That is not a confession of incompetence — it is the calibration problem stated honestly by the person who has to act on it. She has to separate the part of unemployment that is the labor market’s baseline from the part that is the cycle her policy can move, and she is telling you the line between them is blurry and shifting. The mainstream’s real answer to our question lives right here, in that difficulty.
The whole walkthrough collapses into one piece of arithmetic. Observed unemployment splits into a slowly-moving baseline and a cyclical gap around it:
The labor-side framing owns the first term — the natural rate $u^n_t$, built from institutions, frictions, demographics, and the match between skills and jobs. The money-side framing owns the second — the gap $(u_t - u^n_t)$, which opens and closes with aggregate demand and the central bank’s response to it. That is the calibration, in one line. Neither framing is wrong; each is reading a different term of the same sum.
Think of the tide and the waves. The natural rate is the tide — the slow level the water sits at, set by the shape of the coastline (the labor market’s structure). The cyclical gap is the waves — the fast ups and downs driven by the weather (aggregate demand). Ask a question about why the average water level differs between two beaches, and you’re asking a tide question: labor-side. Ask why the water jumped two feet this afternoon, and you’re asking a wave question: money-side.
Run 2008 through the decomposition. The pure labor-side reading failed because the skills-and-mismatch story could not explain why the gap stayed open for five years with the central bank stuck at the zero lower bound. The pure money-side reading failed too, just less: the natural rate itself drifted upward after 2008 — the reassessment Yellen is gesturing at — so some of the persistence really was structural. The calibrated answer is not a shrug. It is specific: most of the post-2008 overhang was a monetary-side gap problem, and a smaller slice was a labor-side natural-rate drift. The components have different sizes, and naming the sizes is the work.
Run 2020 through it and the calibration flips emphasis. The labor-side framing forecast a slow, scarring recovery from 14.7%; the money-side framing forecast a fast one driven by the size of the demand response, and the money-side framing won that round cleanly. But then the tight post-2022 labor market — low unemployment, real wage pressure — reopened the natural-rate question the labor side owns. So 2020 is not a victory for one machine. It is the calibration in motion: the money side called the cyclical recovery, and the labor side still set the baseline the recovery settled onto. Both terms of the sum, doing their jobs.
The split is a stable disposition, and it locates the edges. Long-run averages and cross-country variation: labor-side dominates. Cyclical fluctuations and disinflations: money-side dominates. The strong heterodox positions slot in honestly against this map. Sumner’s market monetarism is more right than wrong about the gap — “the Fed has done enough” was too often wrong because the right counterfactual was a committed spending path — but overreaches when it claims the baseline. MMT is right that involuntary unemployment is a real category and that currency-issuers have more spending authority than the conventional framing admits, but its job guarantee collapses the labor/money distinction in a way that obscures the very calibration we just drew. The two edges pull in opposite directions from the center, which is exactly why naming them honestly is useful rather than tidy.
“Our understanding of the forces driving inflation is imperfect.”
— Janet Yellen, September 2017
The reflex worth keeping: “which component do you mean?”
The real take-away isn’t a verdict on labor vs. money. It’s a habit: when someone says “unemployment is a labor problem” or “a money problem,” the right first question is which part of the number they’re talking about.
The verdict
Unemployment is a labor problem and a money problem, and naming which component each framing dominates is the discipline, not a dodge. The labor-side apparatus owns the level unemployment fluctuates around — the natural rate, set by institutions, frictions, and the match between people and jobs — which is why it answers the cross-country and demographic questions the money side cannot. The money-side apparatus owns the deviations from that level — the recessions, recoveries, and disinflations — which is why it called 2008 and 2020 when the labor side did not. The cases where they tangle, where a long slump scars the baseline itself, are real and worth respecting, but they don’t dissolve the split; they just remind you the calibration is empirical, not a slogan. This is not a refusal to answer — it is the answer, and it names which framing dominates which regime, and by roughly how much. The cyclical-causation half of the picture is the dedicated subject of What causes recessions?, the labor-policy half of Do minimum wages cause unemployment?, and the monetary authority behind the gap of Can central banks control the economy?
Where this leaves us
We held one number still — the headline unemployment rate — and changed the machine reading it three times. First we built the labor-side machine at full strength: search and matching, sticky wages, monopsony, the natural rate, and a real policy menu of training, minimum wages, and work requirements. It is the framing every textbook and every politician reaches for, and it earns that place. Then we flipped to the money-side machine: Keynes’s involuntary unemployment, Friedman’s boundary-fixing natural-rate address, the New Keynesian workhorse, and the two strong edges — Sumner’s market monetarism and MMT — that read the post-2008 overhang and the 2020 collapse-and-recovery as nominal events the labor side misdiagnosed. Then we calibrated: the labor side owns the level, the money side owns the gap, and the cases where they tangle are real but don’t erase the split.
The reframe was the point. The component-calibrated verdict matters, but the durable thing you should walk away with is the reflex underneath it: the apparatus you bring shapes what you see. The same five million unemployed Americans are a skills crisis to one machine and a spending shortfall to the other, and both machines are partly right. So the next time you hear “unemployment is a labor problem” or “unemployment is a money problem” stated as the whole truth, you have the only question that cuts through it: which component do you mean?