New Classical vs. New Keynesian: can stabilization policy do anything?
Two schools, the same rational-expectations foundation, opposite answers — because they disagree about one assumption. Watch thirty years of data pick a winner.
The shared foundation
“That the predictions [of Keynesian models] were wildly incorrect, and that the doctrine on which they were based was fundamentally flawed, are now simple matters of fact… the task now facing contemporary students of the business cycle is that of sorting through the wreckage.”
— Robert Lucas & Thomas Sargent, “After Keynesian Macroeconomics,” Minneapolis Fed Quarterly Review, 1979
An obituary, written in 1979, for the macroeconomics that had run the postwar world. The striking part is what happened next: the school that became the modern mainstream did not fight this verdict. It accepted it — and then beat the people who wrote it.
To see why the obituary mattered — and why it did not kill what its authors thought it killed — you need the one idea both successor schools accepted: rational expectations. The claim is not that people are clairvoyant. It is that they do not make systematic forecasting errors, because they understand the model generating policy. If the central bank always prints money before an election, people learn the pattern and price it in. Expectations are part of the model, not a free parameter the modeler sets by hand.
From that single move comes the most consequential methodological result in modern macro: the Lucas critique. You cannot evaluate a change in policy using behavioral relationships estimated under the old policy. The reason is exactly rational expectations — those estimated relationships bundle in the expectations people held under the old regime, and the moment the regime changes, the expectations shift and the relationships move with them. A Phillips curve fitted to 1960s data could not tell you what would happen if the Fed tried to exploit it in the 1970s, because the act of exploiting it would destroy it. The old models, fitted and then used to simulate brand-new policies, were measuring something that would not survive being acted on.
This is the ground both schools stand on. It is not a New Classical possession and it is not a New Keynesian one — it is the shared inheritance, and that is the whole point of starting here. The duel that follows is not a fight about whether agents are smart. Both sides agree agents are smart. The fork is one assumption, and only one: do prices clear continuously, or are they sticky? Hold that question in your hand. Everything in the next three stages turns on it.
For the intellectual lineage — Friedman’s monetarist opening, Lucas’s rational-expectations revolution, the figure both schools descend from methodologically — the home is History of Economic Thought Ch.10 §10.2 (Lucas and the rational expectations revolution). This is the round after the one where Keynesians and monetarists fought over expectations themselves; that earlier battle — and how the natural-rate hypothesis settled it — is its own walkthrough.
“People even take offense if referred to as Keynesians. At research seminars, people don’t take Keynesian theorizing seriously anymore; the audience starts to whisper and giggle to one another.”
— Robert Lucas, paraphrasing the late-1970s mood the “After Keynesian Macroeconomics” manifesto announced
“Keynesian economics is dead.”
In 1979 the rational-expectations revolutionaries declared the whole Keynesian project a wreck. They were right about the old models — and wrong about what they were burying. The school that became the modern mainstream rose by accepting their verdict.
The agreement before the fork
It is worth dwelling on how complete the agreement is, because the size of the common ground is what makes the disagreement legible. Both schools build models from optimizing households and firms. Both impose rational expectations. Both accept the Lucas critique as binding — you must model the deep parameters (preferences, technology, the costs agents face) that do stay fixed across policy regimes, and derive behavior from them. Both treat technology shocks as real disturbances that genuinely move the economy. This is not a grudging overlap at the edges. It is nearly the entire method. The rational-expectations revolution did not split macroeconomics into two camps with different worldviews; it unified the field around one standard of rigor and then left exactly one question open.
The one open question is whether the price you see today is the price that clears the market today. Say yes — prices are flexible, markets clear continuously — and you get one school, with one answer about recessions and policy. Say no — some prices are sticky, set in advance and held for a while — and you get the other school, with the opposite answer. The two camps are not arguing past each other. They share a model and disagree about a single parameter in it. That is rare in any field, and it is what makes this duel resolvable by evidence rather than by taste.
Where this leaves us
The rational-expectations revolution won, and both schools that fought the next thirty years were its children. They did not disagree about whether people are smart — they agreed people are smart. They disagreed about whether prices are flexible. Everything that follows is one assumption, pushed to its conclusion in two directions. Keep this in mind whenever someone tells you this was a replay of the old expectations fight. It was not. That fight was over by 1979. This is the next one.
Take rational expectations seriously and add the one extra assumption — that markets clear continuously, that prices are fully flexible — and you arrive at a startling conclusion. The business cycle is not a failure to be fixed. It is the economy working perfectly. And if that is true, the case for stabilization policy collapses before it starts.
New Classical at full strength
“Costly efforts at stabilization are likely to be counterproductive. Economic fluctuations are optimal responses to uncertainty in the rate of technological change.”
— Edward Prescott, “Theory Ahead of Business Cycle Measurement,” Minneapolis Fed Quarterly Review, 1986
Read that twice. A recession is not a malfunction. It is the optimal response of rational agents to a real shock — and trying to smooth it makes people worse off. It is the most counterintuitive claim in modern macroeconomics, and it won a Nobel Prize.
Two results carry the New Classical case, and both follow directly from the shared foundation once you add flexible prices.
First, the policy-ineffectiveness proposition. Sargent and Wallace showed in 1975 that if agents have rational expectations and prices are flexible, then systematic monetary policy — any rule the central bank follows that people can learn — has no effect on real output. Only the part of money agents could not anticipate moves anything real, and a predictable rule is, by definition, anticipated. The central bank cannot fool people on average. A policy of leaning against the wind is leaned-against right back by the expectations it provokes.
Second, the real-business-cycle model. Kydland and Prescott built a stochastic-growth model — rational households, optimizing firms, continuously clearing markets, and one source of disturbance: shocks to technology. No money, no nominal frictions, no policy. They calibrated it (set its parameters from long-run growth facts and microeconomic studies rather than fitting it to the cycle) and asked whether technology shocks alone could reproduce the statistical signature of actual business cycles — the size of output swings, how consumption and investment co-move, how employment tracks output. The answer was: to a surprising degree, yes.
The policy-ineffectiveness result is sharpest as a decomposition. Let money growth be a systematic, rule-based component the public can forecast plus a pure surprise:
$$m_t = \underbrace{\mathbb{E}_{t-1}[m_t]}_{\text{systematic (anticipated)}} + \underbrace{\varepsilon_t}_{\text{surprise}}$$With rational expectations and flexible prices, output deviates from its natural level only through the surprise term:
$$y_t - y_t^{*} = \beta\,\varepsilon_t + (\text{real shocks})$$The systematic part drops out entirely — people already priced it in. A central bank that follows any learnable rule contributes nothing to the real economy through that rule. In the pure RBC limit the monetary term vanishes too, and the cycle is driven entirely by real (technology) shocks $z_t$ propagating through optimal saving and labor-supply decisions.
Imagine a farming economy hit by a bad harvest — a genuine drop in what its technology can produce this year. Rational families respond by working a little differently and consuming a little less until conditions improve. Output falls; that fall is not a mistake, it is the best available response to a real shortage. Now ask the central bank to “fix” the recession by printing money. If everyone sees it coming, prices simply rise to match; nothing real changes. The recession was the economy adjusting correctly to a real event, and the stabilizer is pushing against a door that is already where it should be.
Put the two together and you get the disciplinary payoff that makes the Prescott quote coherent rather than absurd: if the cycle is the efficient response of optimizing agents to real shocks, there is no welfare case for stabilization. Smoothing the cycle would mean overriding choices that were already optimal — making people worse off, not better. The right policy posture is credible rules and a stable environment, not activist demand management. Recessions are not failures of the market; they are the market doing its job under bad conditions.
The policy-ineffectiveness result also lives in the monetary-theory apparatus — see Ch 16 §16.2 (Time Inconsistency and the Inflation Bias), which is the formal home for why rules beat discretion. The intellectual lineage of the whole program — Lucas, Sargent, Prescott, Kydland — lives in History of Economic Thought Ch.10 §10.3 (Kydland-Prescott and real business cycles).
“Economic fluctuations are optimal responses to uncertainty in the rate of technological change… costly efforts at stabilization are likely to be counterproductive.”
— Edward Prescott, 1986
“The recession is the economy working perfectly.”
RBC says business cycles are efficient adjustments to real shocks, so smoothing them destroys welfare. It sounds like a provocation. Given its assumptions, it is a theorem — and that is exactly why it is dangerous to dismiss.
Where this leaves us
Given its assumptions, New Classical is airtight. Rational expectations plus flexible prices really does imply that systematic policy is neutral and the cycle is efficient. There is no logical gap to attack. The entire question reduces to one word: flexible. Are prices, in fact, flexible? If they are, Prescott is right and stabilization policy is a category error. If they are not — even a little, even some of them — the conclusions unravel. Everything now rides on that single empirical question.
There is one stubborn fact the RBC model had to wave away. Prices are not flexible. The menu at your local restaurant does not change every time the Fed moves. And a generation of economists who agreed with every word Lucas said about expectations decided that this one fact — small, mundane, well-documented — changes absolutely everything.
New Keynesian at full strength
“Small costs of changing prices can cause large fluctuations in output. The menu cost is privately small but socially large.”
— N. Gregory Mankiw, “Small Menu Costs and Large Business Cycles,” Quarterly Journal of Economics, 1985
A few dollars to reprint a menu. That is the entire friction. The New Keynesians accepted every methodological demand the New Classicals made — rational expectations, microfoundations, all of it — and then added this one trivial cost. Watch what it does.
The New Keynesian move is a chain, and every link is built on the shared foundation. The end of the chain is the opposite of Stage 2’s conclusion.
Menu costs make prices sticky — from optimization, not assumption. A firm pays a small fixed cost to change a posted price. Facing a small shock, a rational firm leaves its price where it is, because the private gain from adjusting is smaller than the cost. Individually trivial; collectively decisive. When many firms hold their prices, the aggregate price level fails to move, a fall in nominal demand becomes a fall in real demand, and output contracts. This is not the old ad hoc rigidity the Lucas critique discredited — it is stickiness derived from optimizing firms, which is why it survives on New Classical terms. Calvo pricing makes it tractable: each period a firm resets its price with some fixed probability, so prices adjust sluggishly and at staggered times.
Sticky prices make money non-neutral. If prices do not adjust instantly, a change in nominal demand falls partly on quantities. Money moves real output — the exact result RBC was built to deny — recovered without giving up rational expectations. And because the resulting recession comes from prices being stuck rather than from any real shock, output sits below the level it would reach if prices were free. That gap, the output gap, is a genuine welfare loss: idle resources, not efficient adjustment. This is the hinge. In RBC the cycle is optimal; here it is a coordination failure that policy can correct.
Calvo pricing yields the New Keynesian Phillips curve — inflation is forward-looking, driven by expected future inflation and the current output gap $\tilde{y}_t$:
$$\pi_t = \beta\,\mathbb{E}_t[\pi_{t+1}] + \kappa\,\tilde{y}_t$$Close the model with a forward-looking IS curve (the output gap depends on the expected real interest rate) and a Taylor rule for the central bank’s policy rate:
$$\tilde{y}_t = \mathbb{E}_t[\tilde{y}_{t+1}] - \tfrac{1}{\sigma}\left(i_t - \mathbb{E}_t[\pi_{t+1}] - r_t^{n}\right)$$ $$i_t = r_t^{n} + \phi_\pi\,\pi_t + \phi_y\,\tilde{y}_t, \qquad \phi_\pi > 1$$These three equations — NK Phillips curve, IS, policy rule — are the 3-equation NK model, the analytical core every inflation-targeting central bank runs. The condition $\phi_\pi > 1$ (the Taylor principle: raise the real rate when inflation rises) is what makes the system stable. Stabilization policy is no longer a vague aspiration; it is a reaction function.
Because the menus are sticky, when the Fed cuts rates the price tags do not all jump at once. For a while the extra demand lands on real activity — more orders, more hours, more output — before prices catch up. That window is exactly what the RBC world denied could exist. The Taylor rule is the central bank’s playbook for using it: when inflation climbs, push the real interest rate up enough to cool demand; when the economy slumps, push it down to lift demand. Sticky prices give the central bank a lever, and the Taylor rule is the hand on it.
Output gaps as genuine welfare losses is not a slogan — Woodford grounded it in a welfare-theoretic loss function derived from household utility, so “stabilize the gap” is a statement about making people better off, not a planner’s preference. The Taylor rule as the operational form of all this lives in the monetary apparatus at Ch 16 §16.2 (rules, discretion, and the inflation bias).
The lineage — Mankiw and David Romer (co-editors of the 1991 collection New Keynesian Economics and author of the standard graduate text; not Paul Romer of growth theory), Woodford, Galí, Blanchard — lives in History of Economic Thought Ch.12 §12.1 (The Synthesis on the Victor’s Foundations).
“Small costs of changing prices can cause large fluctuations in output.”
— N. Gregory Mankiw, 1985
“One friction changes everything.”
New Keynesians conceded the entire New Classical method and added a single well-documented fact — that prices are sticky — and recovered money non-neutrality, real recessions, and a job for stabilization policy. They beat RBC on RBC’s own turf.
Where this leaves us
One friction flips the answer. Money is non-neutral, recessions are real welfare losses, and there is a job for stabilization policy — and the Taylor rule is what that job looks like when a central bank actually does it. We now have two complete, internally airtight frameworks built on the same foundation, separated by a single assumption, reaching opposite verdicts on the question this walkthrough opened with. You cannot settle that with more theory; both sides are valid given their premises. You settle it with evidence about which premise the world obeys. For the policy specifics this framework opens onto — the fiscal multiplier, the zero-lower-bound case for spending — see Does government spending help the economy?; for the central bank’s power to steer the cycle in practice, see Can central banks control the economy?
So which assumption was right? You do not settle that with a chalkboard — both frameworks are airtight on their own premises. You settle it with thirty years of data. And the data arrived in three waves: a long, strange calm; a catastrophe nobody’s model called; and an aftermath economists are still arguing about.
What the data said, and the layered resolution
“Macroeconomics in this original sense has succeeded: its central problem of depression-prevention has been solved, for all practical purposes, and has in fact been solved for many decades.”
— Robert Lucas, AEA Presidential Address, 2003
The high-water mark of pre-crisis confidence, spoken five years before the worst financial crisis since 1929. It is evaluable precisely because of what happened in 2008. The data on which school was right came in three waves — and the first wave looked exactly like a New Keynesian victory lap. See the great-moderation volatility record (History Ch.18 §18.7) for the calm itself.
Wave one: the Great Moderation (1984–2007). The volatility of output and inflation roughly halved across the advanced economies. Recessions grew shallow and rare. Inflation-targeting central banks, running New Keynesian models and Taylor-rule stabilization, appeared to have tamed the cycle — the era of the “divine coincidence,” where stabilizing inflation seemed to stabilize output for free. It looked like decisive vindication of the New Keynesian program.
But the Moderation is not clean proof, and honesty requires saying so. The calm had non-policy causes too: better inventory management, structural shifts in the economy, and — the awkward part — a genuinely quieter run of shocks. The Stock-Watson and Bernanke “good luck versus good policy” debate established that a real share of the moderation was smaller disturbances, not better steering. The interactive volatility record in History Ch.18 §18.7 makes the point physical: the same data that shows volatility falling after 1984 shows the underlying shocks shrinking too. The Moderation was apparent vindication. The crisis would test whether it was real.
Wave two: 2008. The stress test arrived, and it exposed gaps in both frameworks — this is the part a New Keynesian victory lap would skip. RBC had no financial sector and no account of a demand-driven collapse; a recession as deep as 2008, relabeled as technological regress, was not credible. But the canonical New Keynesian model was also caught short: it too had no financial frictions, no banking sector, and a weak answer at the zero lower bound, where the Taylor rule calls for a negative interest rate that cannot be set. Neither pre-crisis workhorse called the crisis. The side-by-side impulse responses below are exactly where the two frameworks’ predictions diverge — and where the adjudication actually happens.
Wave three: the resolution. The model the profession rebuilt after 2008 is not RBC, and it is not the 2007-vintage New Keynesian model either. It is New Keynesian extended with financial frictions — the workhorse central banks run today couples nominal rigidities to a banking and balance-sheet sector. On the question this walkthrough opened with — can and should stabilization policy do anything? — the answer the discipline and the central banks adopted is New Keynesian’s yes. The empirical detail of the great-moderation calm and the 2008 break sits in History Ch.19 (The 2008 crisis and after); this walkthrough names the outcome rather than re-telling the crisis.
New Keynesian won the stabilization question. New Classical won the methodology both now share.
The popular story — “New Keynesian beat New Classical” — is only half right, and the missing half is the interesting one. The school that “lost” the policy argument supplied the rules of the game the winner plays by.
The verdict, in two layers
The duel resolved with a split decision — and the split is itself the consensus. Two layers, and they must not be collapsed into one:
Layer one — New Keynesian won the substantive stabilization question. The modern macro mainstream is New Keynesian DSGE, now extended with financial frictions. Every inflation-targeting central bank runs Taylor-rule stabilization on a New Keynesian workhorse. Output gaps as real welfare losses, and the warrant for smoothing them, are textbook. The pure-RBC position — cycles efficient, money doesn’t matter, no stabilization role — is a minority view. On “can and should stabilization policy do anything?”, the answer the discipline adopted is New Keynesian’s yes, and it won by answering RBC on RBC’s own microfoundational turf.
Layer two — New Classical won the methodology both now share. Rational expectations as the default, the Lucas critique as a permanent constraint on policy evaluation, microfoundations as the standard, and the real-shocks backbone RBC contributed — these are universal, not contested. The New Keynesian workhorse is an RBC real-side core with New Keynesian nominal frictions added. New Classical lost the policy argument but won the rules of the game.
Neither won wholesale. The one-line verdict: New Keynesian won the stabilization question; New Classical won the methodology both now share; the synthesis is built from both. The disagreement that started all this reduced, in the end, to a single assumption about price flexibility — and the data picked the sticky-price side for policy while keeping the flexible-price program’s entire method. For what happened to that synthesis after 2008 — whether the discipline was broken or merely extended — that is a different question with its own walkthrough.
Where this leaves us
We started with a 1979 obituary for Keynesian economics and an apparently simple question: can stabilization policy do anything? Four stages took us from agreement to a split verdict. The shared foundation showed that both schools accept rational expectations and the Lucas critique, so the duel is not the old expectations fight — it is one assumption, price flexibility, pushed two ways. New Classical at full strength gave the most rigorous and most counterintuitive answer macroeconomics ever produced: flexible prices imply systematic policy is neutral and the cycle is efficient, so there is no welfare case for stabilization — airtight given its premise. New Keynesian at full strength conceded that entire method and added one well-documented friction, sticky prices, and recovered money non-neutrality, real output gaps, and a stabilization role, with the Taylor rule as the operational payoff. The data’s split verdict ran through the Great Moderation (apparent vindication, complicated by good luck), 2008 (a stress test that caught both pre-crisis models short), and the rebuilt mainstream (New Keynesian with financial frictions, on an RBC backbone).
The honest verdict lives in two layers, and the temptation to flatten them is exactly the error to resist. New Keynesian won the stabilization question — decisively, and this is mainstream, not contested: central banks stabilize, output gaps are real, pure RBC is a minority view. New Classical won the methodology both schools now share — rational expectations, the Lucas critique, microfoundations, and the real-shocks backbone are universal, and the model every central bank runs is an RBC real-side core with New Keynesian frictions bolted on. The popular “New Keynesian beat New Classical” is half the story; the missing half is that the school which lost the policy argument supplied the rules of the game the winner plays by. The next time someone tells you who won the great macro debate of the last forty years, you have the tools to ask: won which question? — because the answer is genuinely different on the two axes.