Between 1945 and 1970 economics believed it had arrived. The discipline assembled what it took to be a finished, teachable, governing apparatus, married Keynes to the classics at the level of the diagram, and trained two generations inside the marriage. The working position here is that the synthesis was a real achievement at the level of the apparatus and a deferred problem at the level of foundations: it cohered because it taught and governed, and it fell because a coherence that had only been assumed was eventually asked to predict.
Named literature: Samuelson, Foundations of Economic Analysis (1947); Economics (1948). Hicks, "Mr. Keynes and the Classics" (1937). Solow, "A Contribution to the Theory of Economic Growth" (1956); "Technical Change and the Aggregate Production Function" (1957). Modigliani-Brumberg (life-cycle, 1954). Modigliani-Miller (1958). Markowitz, "Portfolio Selection" (1952); Portfolio Selection (1959). Sharpe (1964), Lintner (1965), Mossin (1966) on the CAPM. Mincer, "Investment in Human Capital" (1958); Schooling, Experience, and Earnings (1974). Haavelmo, "The Probability Approach in Econometrics" (1944); the Cowles Commission monographs (1944–55). Patinkin, Money, Interest, and Prices (1956). Arrow-Debreu, "Existence of an Equilibrium for a Competitive Economy" (1954); McKenzie (1954). Phillips (1958); Samuelson-Solow, "Analytical Aspects of Anti-Inflation Policy" (1960). Phelps (1967); Friedman, "The Role of Monetary Policy" (1968).
By the late 1940s a generation of economists had stopped arguing about whether Keynes was right and started building on the assumption that he was, but only halfway. The compromise that organized the next quarter-century has a name Paul Samuelson gave it: the neoclassical synthesis. Its content can be stated in a sentence the textbooks repeated until it became a reflex. The economy is Keynesian in the short run and neoclassical in the long run. In the short run, prices and wages are sticky, demand can fall short of what full employment requires, and a government that manages aggregate demand can close the gap. In the long run, prices adjust, markets clear, output returns to the level the economy's resources and technology permit, and the neoclassical apparatus of optimizing agents and clearing markets describes where the economy settles. Keynes and the classics were not rivals, on this reading. They were two halves of one machine, each describing a different horizon.
The machine had a diagram, and the diagram did more work than any prose. John Hicks had already, in 1937, reduced the General Theory to a pair of curves: the IS curve, where the goods market clears, and the LM curve, where the money market clears, that together fixed output and the interest rate. What Hicks lost in that reduction (Keynes's animal spirits, his treatment of fundamental uncertainty, the radical instability of investment demand) is the subject of the chapter before this one; here the diagram is picked up as an instrument rather than as a loss. IS-LM became the thing every economist was trained on and the thing every finance minister's advisers reasoned with. Over the 1950s it hardened into a still more compressed teaching form, AD-AS, an aggregate-demand curve crossing an aggregate-supply curve, in which the short-run/long-run division of labor became a property of which curve you drew. Draw an upward-sloping supply curve and demand moves output; draw a vertical one and demand moves only prices. The synthesis was, before it was anything else, an integration by diagram.
It is worth being precise about what the integration claimed, because the claim is what the rest of the prose collects on. The synthesis asserted that the short-run demand-driven model and the long-run market-clearing model describe the same economy, the same households deciding how much to consume, the same firms deciding how much to invest, the same workers deciding whether to take a job, behaving consistently as the horizon lengthens. That is a strong claim. It says the agents who are off their optimal paths in the short run, because prices have not adjusted, are the same agents who are on them in the long run, and that the transition from one to the other is something the model could in principle describe. The synthesis did not describe it. It asserted the consistency at the level of the apparatus and left the behavioral demonstration for later. The short-run model had no agents inside it at all; its relations were estimated regularities among aggregates, a consumption function, an investment function, a demand for money, while the long-run picture was populated by optimizers. The two were married at the level of the diagram and never reconciled at the level of behavior.
Try the bet on the interactive below before the prose argues it. Flip the regime and shift demand: the same disturbance produces output in the short-run regime and prices-only in the long-run regime, and the difference between the two outcomes is nothing but which curve you decided to draw vertical. That is the synthesis in one gesture, and the gesture is a choice about the diagram, not a result derived from how anyone behaves.
Toggle the regime and shift aggregate demand. In the short-run (Keynesian) regime the supply curve slopes up, so a demand shift raises output and prices a little; in the long-run (neoclassical) regime supply is vertical at potential output, so the same shift moves only prices.
Figure 9.1 (interactive). The synthesis as a diagram. The same demand shift behaves differently across regimes — output in the short run, prices-only in the long run — and which behavior you get is a choice of which supply curve is vertical. The split is asserted at the level of the diagram: the chapter's bet, made drivable. Flip the regime; drag the demand shift.
In the short run prices are sticky, so a demand shift moves output. In the long run prices have adjusted, so the same demand shift moves only the price level, and output is pinned at potential.
Aggregate demand slopes down in the price level: $Y = AD(P;\,\text{shift})$. Short-run supply slopes up because prices are sticky relative to expectations: $Y = Y^{*} + \gamma(P - P^{e})$. Long-run supply is vertical at potential output: $Y = Y^{*}$.
The regime toggle swaps which supply specification meets the same AD curve. Only in the short-run regime does a demand shift change $Y$; in the long-run regime it changes only $P$.
The full IS-LM/AD-AS apparatus, the curves derived, the multipliers computed, the policy experiments run, is the business of the intermediate-macro chapter in the economics book, and there is no point re-deriving it here. What this chapter owns is the integration: who built it, what it married, and what it assumed away. The boom that ran alongside it, the postwar golden age that made the apparatus feel vindicated, belongs to the economic-history book and is referenced, not re-narrated. The synthesis felt true partly because the economy it described was, for a quarter-century, behaving. The machine had to be taught before it could govern, and one man taught it.
Open the intermediate-macro chapter (IS-LM / AD-AS apparatus) ↗
Open the postwar golden age chapter ↗
You are reading this chapter as a stop on the employment walkthrough — the synthesis-side voice, before its falsifier.
The synthesis operationalized Keynes — IS-LM plus a stable Phillips trade-off — and sanded the radical edges off the General Theory into a fine-tuning menu, with Samuelson, Solow, and Modigliani the synthesis-side voices. For a generation the synthesis ran the show. Then one man stood up at the 1968 AEA meeting and said the menu was an illusion, and that the moment policy tried to exploit it, it would self-destruct.
On the government-spending walkthrough, this chapter supplies the synthesis-side lineage anchor.
The synthesis operationalized Keynes into a demand-management policy menu — the framework within which the fiscal multiplier was debated. The IS-LM apparatus is where "government spending raises output" became a diagram a finance ministry could reason with.
Paul Samuelson did two things to economics, a few years apart, and between them they fixed both how the discipline reasons and what every economist starts out believing. The first was a method. Foundations of Economic Analysis, published in 1947 from a thesis written years earlier, argued that the disparate problems of economic theory, the consumer choosing a basket, the firm choosing an output, the economy settling into equilibrium, were the same problem in different clothing: constrained optimization. An agent maximizes something subject to something, and the conditions that characterize the maximum, together with the assumption that the system is stable, generate testable predictions about how the agent responds when a constraint shifts. Samuelson called the link between stability and comparative statics the correspondence principle, and the move it licensed, read the comparative statics off the optimization and discipline them with the stability assumption, became the standard form of an economic argument.
This is more consequential than it sounds, because it fixed the discipline's method rather than any particular result. After Foundations, a piece of economics that did not reduce to an agent optimizing under constraints, with comparative statics derived from the optimization, was not quite economics in the way the profession had begun to mean the word. Samuelson's own contribution to consumer theory, revealed preference, is the purest expression of the program: instead of assuming a utility function and deducing choices, infer the structure of preferences from the choices an agent actually makes, so that the theory rests on observable behavior rather than on an unobservable mental quantity. The apparatus that does this work lives, as live machinery, in the advanced-microeconomics chapter of the economics book; what matters here is that one book fixed the shape of the discipline's arguments for the next half-century.
The second thing Samuelson did was write the textbook. Economics, first published in 1948, was not the first introductory text, but it became the one through which most of the postwar generation learned the subject, and through which the synthesis became not a research position but the water economists swam in. The book taught the income-expenditure model, the multiplier, IS-LM in its teaching form, and the short-run/long-run division of labor, and it taught them as settled apparatus rather than as a contested program. A student who learned economics from Samuelson learned the synthesis the way one learns a language, without the sense that it had been an argument someone won.
That is the mechanism by which the synthesis cohered, and it is worth naming carefully because it is easy to mistake for something deeper. The coherence the profession felt was, in large part, the coherence of a shared training. When every economist has reasoned through the same diagrams in the same order, the framework feels not merely agreed-upon but obvious, and an obvious framework is one whose foundations no one re-examines. The textbook did not prove the synthesis consistent; it made the synthesis familiar, and familiarity reads, from the inside, like depth. The cohesion was real and it was a teaching achievement, which is exactly why, when the foundations were finally tested, the discipline was surprised. The textbook had a confident short-run story and gestured at a long run it had not actually built. Robert Solow built it.
Where Samuelson sits on the intellectual-history timeline ↗
The Keynesians had a theory of the short run and, at the level of growth, an embarrassment. The dominant growth model before the mid-1950s, the Harrod-Domar model, made long-run full-employment growth depend on a knife-edge: the rate of growth warranted by saving and investment behavior had to equal exactly the rate the labor force and productivity made natural, and there was no mechanism that pushed the two together. Off the knife-edge, the model predicted either accelerating unemployment or accelerating labor shortage, with no return to balance. A theory whose central prediction was that balanced growth is a measure-zero accident was not a comfortable foundation for a discipline that had just declared the economy returns to equilibrium in the long run.
Robert Solow's 1956 model removed the knife-edge with a single assumption: capital is subject to diminishing returns. In the Harrod-Domar world, output was proportional to capital, so more saving meant permanently faster growth and the system never settled. Let the marginal product of capital fall as capital per worker rises, and the picture changes entirely. Saving adds capital, but each additional unit of capital adds less output, and meanwhile depreciation and a growing workforce are eroding capital per worker. At some level of capital per worker, the investment that saving finances exactly offsets what depreciation and population growth take away, and capital per worker stops changing. That level is the steady state, and the model's central result is that the economy converges to it from wherever it starts. Convergence, not the knife-edge, is the long-run prediction.
The steady state has a consequence the policy-minded found disappointing and the honest found clarifying. Because the long-run growth rate of output per worker at the steady state is set by the rate of technical progress and not by the saving rate, a country that saves more ends up with a higher level of income per worker but not a permanently higher growth rate. Saving buys you a richer steady state, not a faster one. This is the model's most counterintuitive claim and the one the interactive below is built to make visible: drag the saving rate up and watch the capital-per-worker path climb to a higher plateau without bending the long-run slope.
Drive the Solow model. Drag the saving rate, depreciation, and population growth; the capital-per-worker path converges to the steady state $k^{*}$ where investment just offsets depreciation and population growth. Raising the saving rate lifts the plateau but not the long-run slope. Switch to the growth-accounting view to see most measured growth fall into the residual, and toggle the Harrod-Domar overlay to see the knife-edge Solow replaced.
Figure 9.2 (interactive). Solow convergence and the residual. The path climbs to $k^{*}$ where $sf(k)=(n+\delta)k$; raising $s$ raises the plateau, not the long-run growth rate. The growth-accounting view shows most measured growth falling into the residual the model does not explain; the overlay shows the Harrod-Domar knife-edge diminishing returns replaced. Drag the sliders; switch the view.
Saving buys capital until depreciation and a growing workforce eat the gains; then capital per worker stops rising. A higher saving rate reaches a higher plateau, but the long-run growth rate of output per worker is set by technical progress, which the model does not explain.
Capital per worker stops changing where investment offsets depreciation and population growth: $sf(k) = (n+\delta)k$, with Cobb-Douglas $f(k)=k^{\alpha}$. The economy converges to this $k^{*}$ from any starting point.
Growth accounting decomposes output growth: $\hat{Y} = \alpha\hat{K} + (1-\alpha)\hat{L} + \hat{A}$, where $\hat{A}$ is the residual (total factor productivity), the part of output growth left unexplained by measured capital and labor.
Then, in a 1957 companion paper, Solow did the thing that made the model immortal, and it was an act of measurement rather than theory. If output growth comes from growth in capital, growth in labor, and technical progress, you can use the model to take the measured contributions of capital and labor out of measured output growth and read off what is left. What is left is the part the model's own inputs cannot explain, the Solow residual, later called total factor productivity. Solow's finding, run on US data for the first half of the twentieth century, was that the residual was enormous: something on the order of seven-eighths of the growth in output per worker was not accounted for by more capital and more labor at all. It was the residual.
Sit with what that means. The era's most influential growth model, the one that organized how economists thought about the long run for thirty years, found that the overwhelming majority of growth came from a term the model itself did not explain, technical change, which the model treated as exogenous, as manna that falls on the economy from outside its own logic. Solow named the residual "a measure of our ignorance," and the candor is the point. A lesser model would have buried the gap; Solow measured it precisely and put the measurement at the center. That is why the model endured. It was not because it explained growth, it conspicuously did not explain most of growth, but because it located, with arithmetic, exactly what a theory of growth would have to explain, and admitted it had not.
Solow had grounded the long run in optimizing behavior, diminishing returns are what a profit-maximizing economy faces, and convergence is what optimizing saving and investment produce, and in doing so he had answered the question the Keynesians left open. The exogenous-technical-change limitation that the residual exposed is the opening the endogenous-growth revival of the 1980s drove through, making the growth rate itself depend on choices about research, human capital, and the returns to scale that ideas enjoy; that revival belongs to the Schumpeterian-tradition chapter and is referenced here, not re-narrated. What matters for the synthesis is the asymmetry the next section turns on. The growth side of the apparatus had agents inside it. The question was whether the demand side did.
Open the growth-theory chapter (the Solow model as live apparatus) ↗
On the growth walkthrough, this is the Solow stop you just drove the model for.
Diminishing returns deliver convergence to a steady state; growth accounting then finds that most measured growth is the unexplained residual, because technical change is exogenous to the model. Solow had found the engine of growth and then declared he couldn't open the hood, the gap the endogenous-growth revival later fills.
Through the 1950s and into the 1960s the synthesis pursued a quiet, powerful project: putting optimizing agents underneath the aggregate relations the macro model used. The consumption function, the demand for assets, the cost of capital, the determination of wages, the estimation of the whole system from data; for each of these the era produced a model in which the relation was derived from an agent solving a problem, rather than fitted as a regularity. The project succeeded, piece by piece, with a brilliance that founded entire subfields. And it left untouched the one relation the whole synthesis ran on. That asymmetry is this chapter's pivot, and it is worth building carefully.
Start with consumption, because the multiplier rests on it. Keynes had written the consumption function as a relation between current consumption and current income, and the multiplier's force depended on the slope. Franco Modigliani, with Richard Brumberg, asked the obvious optimizing question: why would a rational household tie its consumption to this year's income rather than to its resources over a whole expected lifetime? The life-cycle hypothesis answers that it would not. A household smooths consumption across the lifetime, borrowing when young against future earnings, saving in middle age, drawing down in retirement, so that consumption tracks lifetime resources, and current income matters only insofar as it changes the lifetime total. This is a microfoundation in the strict sense: the aggregate consumption relation is derived from a household solving an intertemporal optimization. It is a cousin of Milton Friedman's permanent-income hypothesis, which makes a similar move with a different formalization; that one belongs to the counter-revolution chapter and a different argument.
Now finance, where the era founded a discipline. Harry Markowitz, in 1952, asked how an investor who cares about both return and risk should choose a portfolio, and answered with mean-variance optimization: the investor trades expected return against the variance of the portfolio, and because variance depends on how assets move together, diversification, holding assets whose returns are imperfectly correlated, reduces risk without sacrificing return. From Markowitz's foundation, Sharpe, Lintner, and Mossin built the capital asset pricing model in the mid-1960s: if every investor is solving the same mean-variance problem, then in equilibrium an asset's expected return rises linearly with its exposure to undiversifiable market risk, measured by beta, and exposure to risk that can be diversified away earns nothing. And Modigliani, again, this time with Merton Miller in 1958, proved that under stated conditions, no taxes, no bankruptcy costs, efficient markets, a firm's value is independent of how it finances itself; debt versus equity, the dividend it pays, none of it changes the value, because investors can replicate any capital structure on their own account. The Modigliani-Miller theorems founded modern corporate finance by establishing the irrelevance result that every real-world deviation from it has to be explained against. Each of these is an agent solving an optimization, and the asset-pricing and corporate-finance toolkits they founded are the descendants the economics book's finance material formalizes.
Labor, next. Jacob Mincer, in 1958 and definitively in 1974, modeled the worker's wage as the return on an investment. Schooling and on-the-job experience are costly to acquire, forgone earnings, tuition, effort, and they raise productivity, so a worker chooses how much to acquire by weighing the cost against the higher future earnings, exactly as a firm weighs an investment. The Mincer earnings function, relating log earnings to years of schooling and a quadratic in experience, became the workhorse specification of empirical labor economics and the foundation of human-capital theory, and it carries this chapter's thread into the credibility revolution in econometrics decades later.
And the method by which all of this was to be estimated. The Cowles Commission, gathering econometricians through the 1940s and early 1950s, formalized the program of structural estimation: write down a system of simultaneous equations that represents the economy's behavioral structure, confront the problem that you cannot in general recover the structural parameters from the data without prior restrictions (the identification problem, which Cowles stated with full rigor), impose theory-justified restrictions to achieve identification, and estimate. Trygve Haavelmo's 1944 probability approach gave the program its statistical foundation. The Cowles structural-equations program was the synthesis era's standard for what it meant to take an economic model to data, and it is the predecessor, engaged at full strength before any critique, that the econometric-methodology lineage measures the later time-series turn and credibility revolution against.
Now the asymmetry. Lay the era's micro-grounding achievements side by side and a pattern is impossible to miss. For consumption, the era built an optimizing household. For asset pricing, an optimizing investor. For the cost of capital, the arbitrage-driven indifference of value to financing. For wages, the human-capital investor. For estimation, a structural program with identification stated and solved. For each piece of the apparatus, the era produced an agent-level foundation. For the aggregate demand model itself, the IS-LM, AD-AS engine that the whole synthesis used to think about output, employment, and stabilization policy, it produced none. The demand side remained what Keynes had left it: a set of estimated relations among aggregates, a consumption function and an investment function and a demand for money, with no agents inside.
Micro-grounded pieces, agent-free whole
| Relation | Who micro-grounded it | The agent-level story | Year |
|---|---|---|---|
| Consumption | Modigliani-Brumberg (life-cycle) | Household smooths consumption over a planned lifetime; consumption tracks lifetime resources | 1954 |
| Asset pricing | Markowitz; Sharpe-Lintner-Mossin (CAPM) | Investor trades return against risk; in equilibrium return is linear in market risk | 1952 / 1964 |
| Cost of capital | Modigliani-Miller | Value independent of financing; investors replicate any capital structure | 1958 |
| Wages | Mincer (human capital) | Worker invests in schooling and experience; wage is the return on the investment | 1958 / 1974 |
| Estimation | Cowles Commission | Structural simultaneous-equations system; identification stated and solved | 1944–55 |
| Aggregate demand model | — | None. Estimated relations among aggregates; no agents inside | — |
They had grounded the pieces and left the whole a machine with no one inside it. The synthesis could tell you, agent by agent, why a household consumed as it did, why an asset priced as it did, why a worker earned as he did, and it could not tell you, from any agent's behavior, why the aggregate demand relations it used for policy held, or whether they would go on holding when the policy regime changed. The gap was not hidden; it simply did not seem urgent while the economy behaved and the relations held. And the most rigorous result the era produced was about the other half of the apparatus entirely.
On the finance walkthrough, this is the postwar-founding stop.
Markowitz (1952) made portfolio choice a risk-return optimization; if everyone solves the same optimization, what happens to prices? Sharpe-Lintner-Mossin answered with the CAPM (1964), and Modigliani-Miller (1958) made firm value independent of financing. Modern finance theory starts here, on its way to Fama's efficient markets in the next chapter.
On the econometric-methodology walkthrough, this is the structural-ambition stop, at full strength before any critique.
The Cowles Commission formalized structural simultaneous-equations estimation and stated the identification problem with rigor, with Frisch and Haavelmo behind it. It set the synthesis era's standard for taking a model to data, the predecessor the later time-series turn and credibility revolution measure themselves against.
Eighty years before, Léon Walras had written down a system of equations purporting to describe an entire economy in general equilibrium, every market clearing simultaneously, every price determined by the requirement that supply equal demand everywhere at once, and had been unable to prove that the system had a solution. Counting equations and unknowns, as Walras did, shows the system might have a solution; it does not show one exists. For eighty years the foundational claim of the whole equilibrium tradition, that a competitive economy can be in equilibrium, rested on a hope dressed as arithmetic.
In 1954 Kenneth Arrow and Gérard Debreu, and independently Lionel McKenzie, proved it. Using fixed-point theorems, the mathematical machinery that guarantees, under the right conditions, that a continuous mapping has a point it leaves unmoved, they showed that under stated assumptions a competitive general equilibrium exists: there is a vector of prices at which every market clears simultaneously. The Arrow-Debreu existence theorem is the rigorous completion of the Walrasian program, the summit toward which the marginalist general-equilibrium lineage had been climbing since the 1870s, and as a piece of mathematical economics it is among the genuine monuments of the century.
It is also, read carefully, an ambivalent monument, and the ambivalence is what makes it matter for this chapter rather than for the micro chapter that owns the apparatus. Consider first what it proved and what it left open. The proof establishes that an equilibrium exists; it says almost nothing about whether the economy would reach it (stability) or whether there is only one to reach (uniqueness). Those silences were later made damning: the Sonnenschein-Mantel-Debreu results of the early 1970s showed that the assumptions which guarantee existence place essentially no restrictions on the structure of aggregate excess demand, so that stability and uniqueness cannot be had in general at all. The theorem proved equilibrium possible and left the discipline unable to show it would be found.
What Arrow-Debreu proved, and what it left silent
| Proved | Silent on |
|---|---|
| A competitive general equilibrium exists, a price vector clearing all markets at once | Whether the economy would reach it (stability) |
| Existence holds under convexity and completeness conditions, rigorously stated | Whether the equilibrium is unique (SMD later: not in general) |
| The Walrasian program is mathematically complete after eighty years | Any connection to the Keynesian macro the synthesis actually ran on |
But the silence this chapter cares about most is the third one. The conditions the proof requires are extraordinarily restrictive, a complete set of markets, including markets for every good in every future state of the world; convexity of preferences and technology; no externalities; price-taking behavior everywhere. These are not approximations of a real economy; they are the assumptions under which the existence question becomes tractable, and Arrow and Debreu were entirely candid that the theorem buys its rigor by accepting them. And the equilibrium it establishes is the equilibrium of the neoclassical long run, the market-clearing world the synthesis assigned to the long horizon. It connects not at all to the short-run, sticky-price, demand-determined Keynesian model that the synthesis used to think about unemployment and stabilization, the model that did the actual policy work. The most rigorous achievement of the era was a proof about the half of the synthesis that was not running the economy.
Set the achievement and the silence next to each other and the chapter's frame sharpens into a single uncomfortable observation. The synthesis had, on its long-run neoclassical side, a result of the highest rigor, and that result was silent on stability, silent on uniqueness, and disconnected from the macro the synthesis governed with. On its short-run Keynesian side, the side that ran policy, it had no agent-level foundation at all. The rigor was about the wrong half. When the bill came due, in 1968, it came due on the half that had been running on borrowed foundations.
Explore the intellectual-history timeline ↗
On the value walkthrough, this is the summit the lineage was climbing toward.
Walras wrote down a whole-economy equilibrium in 1874 but nobody could prove such a system has a solution. In 1954, two mathematicians did: the Arrow-Debreu existence theorem is value as the equilibrium price vector, rigorously shown to exist, and silent on stability, uniqueness, and any connection to the macro the synthesis ran on.
On the market-structure walkthrough, Arrow-Debreu 1954 is the general-equilibrium summit before the lineage turns toward mechanism design.
Arrow-Debreu 1954 is the proof that a competitive price system can clear all markets at once, the general-equilibrium summit, before the market-structure lineage turns toward the economics of information and mechanism design.
The synthesis had a teaching triumph in Samuelson and a long-run model in Solow, but what turned it into a governing apparatus, what let it walk into a finance ministry and offer a choice rather than a forecast, was an empirical regularity discovered almost by accident. In 1958 A. W. Phillips, examining nearly a century of British data, found a stable inverse relationship between the rate of wage inflation and the rate of unemployment: when unemployment was low, wages rose fast; when unemployment was high, they rose slowly or fell. Two years later Paul Samuelson and Robert Solow, examining American data, restated the relation in terms of price inflation and named its implication for policy in as many words. The Phillips curve was, they wrote, a menu, a set of inflation-unemployment combinations a government could choose among. Want lower unemployment? It is available, at the price of higher inflation. Want lower inflation? Also available, at the price of higher unemployment. Pick your point.
This was the synthesis at its most confident, and it is worth feeling how reasonable the confidence was. The apparatus described the economy; the economy was booming; the Phillips curve turned the abstract promise of demand management into a concrete dial a government could turn. By the mid-1960s the position had a slogan, "we are all Keynesians now," uttered, tellingly, by people who were not Keynesians by temperament, because the apparatus had become the common language regardless of where one started. The synthesis cohered because it taught (Samuelson) and because it governed (the Phillips menu), and for a quarter-century both halves of that sentence were true. A reader who feels the pull of the position is feeling exactly what the discipline felt. The fault line was real, but it was not visible from inside the boom.
Here the historiography genuinely divides, and both readings have force. On one reading, the synthesis was a coherent achievement: it integrated demand-side and long-run economics into a single working apparatus, trained the profession, governed the longest expansion in modern history, and described that expansion well enough to be repeatedly vindicated. On the competing reading, it was a pragmatic compromise: the marriage of Keynes and the classics was a diagram, not a theory; the consistency of the two halves was assumed rather than demonstrated; and the apparatus worked, while it worked, for reasons it could not itself articulate. The honest verdict is that each reading captures a real layer and neither captures the other's. The synthesis was a durable achievement at the level of the apparatus and a deferred problem at the level of foundations. It cohered because it taught and governed; it would fall because the coherence had been assumed, not proved, and an assumed coherence holds only until something asks it to predict.
What, precisely, did the synthesis lack? Here is the analytic payoff, and it is worth stating carefully. The synthesis macro lacked microfoundations, derivations of its aggregate relations from the optimizing behavior of individual agents. Recall the asymmetry of the previous section: the synthesis had ground the pieces (consumption, asset pricing, the cost of capital, wages) but not the aggregate demand model itself, whose relations remained estimated regularities among aggregates. The Phillips curve was the canonical such regularity, and it concentrated the whole problem into one relation. The synthesis treated the estimated Phillips relation as if it were structural, as if it reflected a stable feature of how the economy works, invariant to policy, so that a government could move along it at will. But an estimated relation is only a fitted regularity, and a fitted regularity holds only as long as the behavior that generated it holds. The distinction between a structural relation (invariant to the policy regime, because it reflects underlying behavior) and an estimated one (a regularity that may shift when the regime shifts) is the exact distinction the synthesis conflated, and the Phillips menu is where the conflation was about to be exposed.
The exposure came from inside, in 1967 and 1968. Edmund Phelps and, more publicly, Milton Friedman in his 1968 presidential address to the American Economic Association made the same argument: the Phillips relation is not agent-invariant, because it depends on what workers and firms expect inflation to be. The downward-sloping menu describes an economy whose participants have not yet adjusted their expectations. Let a government try to exploit the menu, push unemployment below its natural rate by accepting higher inflation, and at first it works, because expectations lag. But workers and firms learn. They build the higher inflation into wage demands and price-setting, the short-run curve shifts up, and the economy returns to the same unemployment rate at the new, higher inflation. There is no long-run trade-off. The menu was an illusion that lasted exactly as long as it took expectations to catch up, and the moment policy tried to buy a permanent reduction in unemployment, the relation it was buying along moved. The interactive below walks the move: pick a point on the menu, exploit it, and watch the curve climb out from under you until the long-run relation stands vertical at the natural rate.
Step through the falsification. Step 1: the stable menu, pick a point. Step 2: exploit it, and expectations adjust, shifting the short-run curve up. Step 3: repeated exploitation traces the vertical long-run curve at the natural rate. Step 4: the gap named.
The synthesis Phillips relation: a stable, downward-sloping menu. 1960s policy picks a point, lower unemployment, accept higher inflation.
Figure 9.5 (interactive). The synthesis Phillips menu falsifying itself. Advancing the steps shifts the short-run curve up as expectations adjust, until the long-run relation stands vertical at the natural rate. A static figure can show the final vertical curve; only the sequence shows the move from stable menu to illusion. Advance the steps.
The menu looks like a free lunch; the moment you order from it, the price changes. Lower unemployment buys higher inflation only until people expect the higher inflation, then you are back where you started, paying more.
The synthesis relation is a stable menu: $\pi = f(u)$. The expectations-augmented relation adds the term the synthesis omitted: $\pi = \pi^{e} - \beta(u - u^{*})$.
With expected inflation $\pi^{e}$ adjusting to actual inflation $\pi$, the long-run curve is vertical at the natural rate $u^{*}$: no permanent trade-off. The expectations-augmented form belongs to the counter-revolution chapter; here $\pi^{e}$ is named as the missing term.
Notice what kind of failure this was. It was not that the 1970s happened and the curve stopped fitting, though that is how it is sometimes told, and the stagflation record belongs to the next chapter. It was that the relation broke for a stateable reason: it had no agent-level foundation, so there was no reason for it to be invariant to policy, and the moment policy leaned on it the behavior that had generated it changed. Friedman and Phelps named the gap the synthesis had carried all along. A few years later Robert Lucas would generalize the point into a critique that cut deeper still, that any estimated macroeconomic relationship is built from the decisions of agents who respond to the policy regime, so that the relationship's parameters will shift whenever the regime shifts, and using estimated relations to evaluate policy changes is therefore unsound in principle. The Lucas critique is named here, not derived; deriving it, and the rational-expectations program it launched, is the work of the counter-revolution chapter, which drives through exactly the opening the previous section located and this one has now stated.
So the close arrives where the opening pointed. The synthesis was a real achievement and a real failure, and the two were the same thing seen from different sides. It cohered because it was taught until it felt obvious and governed until it felt vindicated; it fell because the coherence had been assumed at the level of the diagram and never built at the level of behavior, and an assumed coherence survives only until it is asked to do the one thing assumptions cannot do, which is predict out of sample. The gap it left did not end the project, it set the next two. The counter-revolution would drive through the gap, insisting that macro relations must be derived from agents who anticipate policy. The New Keynesian program would, a decade further on, close the gap on the synthesis's own side, rebuilding the sticky-price short run on explicit microfoundations and answering the Lucas critique on its own terms. The synthesis was not refuted into oblivion. It was diagnosed, and the diagnosis became the agenda.
On this walkthrough, the synthesis is the Keynesian baseline the 1970s tested.
The Samuelson-Solow Phillips menu is the synthesis-side voice, the Keynesian baseline, before the monetarist challenge that Friedman's natural-rate argument opens. The menu's self-destruction under exploitation is exactly where the duel begins.