Microfoundations: An Honest Primer
The demand that macroeconomics be grounded in individual behavior transformed the discipline. Here's what microfoundations actually means, why it matters, and where the project has gone wrong.
The Revolution You Never Heard Of
Sometime in the 1970s, macroeconomics went through a revolution. It was not televised. It did not make the front pages. But it changed, fundamentally and perhaps irreversibly, how economists think about recessions, inflation, unemployment, and the role of government. The revolution was about microfoundations — the demand that macroeconomic models be derived from the behavior of individual agents making optimizing choices.
Before this revolution, macroeconomics and microeconomics were essentially separate disciplines. Macroeconomists studied aggregates — GDP, unemployment, inflation, the money supply — using models that described relationships between these aggregates directly. The Keynesian consumption function said that aggregate consumption depends on aggregate income. The Phillips curve said that inflation depends on unemployment. These relationships were estimated from data, and they worked well enough for policy purposes — until they didn’t.
After the revolution, macroeconomics was rebuilt from the ground up. Every macroeconomic relationship had to be derived from the choices of individual consumers, workers, and firms, each maximizing their own objectives subject to their own constraints. The aggregate patterns — consumption, investment, employment, output — had to emerge as consequences of these individual choices, not as independent assumptions.
This is what “microfoundations” means, and understanding it is essential for understanding why modern macroeconomics looks the way it does — why it is so mathematical, so reliant on a particular kind of model, and so often accused of being disconnected from reality.
The Lucas Critique: Why the Old Models Broke
The intellectual catalyst for the microfoundations revolution was a 1976 paper by Robert Lucas, “Econometric Policy Evaluation: A Critique.” The argument, known as the Lucas critique, is simple in principle and devastating in implication.
Lucas pointed out that the relationships estimated in traditional macroeconomic models — the consumption function, the Phillips curve, the investment equation — are not structural features of the economy. They are reduced-form relationships that reflect the behavior of individuals making choices under particular policy regimes. If the policy regime changes, the behavior changes, and the estimated relationships break down.
The Phillips curve is the classic example. Throughout the 1960s, economists observed a stable inverse relationship between inflation and unemployment: when unemployment was low, inflation was high, and vice versa. Policymakers treated this as a menu of options — you could choose low unemployment at the cost of higher inflation, or low inflation at the cost of higher unemployment. But when policymakers tried to exploit this trade-off in the late 1960s and 1970s — running expansionary policies to push unemployment below its “natural” rate — the relationship broke down. Inflation rose, but unemployment did not fall. The result was stagflation: high inflation and high unemployment simultaneously.
Lucas’s explanation was that the Phillips curve relationship depended on people’s expectations. When inflation was low and stable, workers and firms did not bother to predict it accurately, and an increase in aggregate demand could temporarily boost employment. But once policymakers systematically pursued expansionary policy, people caught on. They adjusted their expectations, demanded higher wages, and the Phillips curve shifted. The relationship that policymakers were trying to exploit was not a stable feature of the economy but a consequence of particular expectations, and those expectations changed when the policy changed.
The implication was radical: you cannot use historical relationships to predict the effects of new policies, because the relationships themselves depend on the policy regime. The only way to build a model that is immune to this critique — a model that remains valid when policy changes — is to build it from the ground up, starting with the preferences, technology, and constraints of individual agents, and solving for their optimal behavior under any given policy regime. This is the microfoundations program.
How a DSGE Model Actually Works
The standard vehicle for microfounded macroeconomics is the Dynamic Stochastic General Equilibrium (DSGE) model. The name is forbidding, but the basic structure is not as complicated as it sounds. Here is how a typical New Keynesian DSGE model works, in plain language.
The household: There is a representative household (or a continuum of identical households, which amounts to the same thing). The household lives forever and makes decisions in every period about how much to consume and how much to save. It prefers more consumption to less, and it prefers smooth consumption over time (a dollar of consumption today is worth about the same as a dollar tomorrow, discounted slightly for impatience). It also supplies labor, and it dislikes working — more hours of labor reduce its well-being. The household chooses consumption, saving, and labor supply to maximize its lifetime well-being (utility), subject to a budget constraint: in each period, its spending on consumption plus saving must equal its income from labor plus the return on its savings.
The firm: There is a representative firm (or a continuum of firms) that produces output using labor (and sometimes capital). The firm has a production technology that converts inputs into outputs. It chooses how much labor to hire (and how much to invest, in models with capital) to maximize profits. In a perfectly competitive model, the firm takes the price of its output and the wage as given. In a New Keynesian model, firms have some market power — they set prices rather than taking them as given — but they face a constraint on how frequently they can change prices (this is the “sticky prices” assumption, often modeled using Calvo pricing, where in each period, a firm has a random chance of being able to reset its price, and otherwise must keep it unchanged).
The central bank: The central bank sets the nominal interest rate according to a rule — typically a Taylor rule, which says: raise the interest rate when inflation is above target, and lower it when output is below potential. The central bank’s rule is the main source of monetary policy in the model.
Shocks: The model includes random shocks — unexpected changes in technology (a good harvest, a new invention), in preferences (consumers suddenly want to save more), or in policy (the central bank raises rates unexpectedly). These shocks are the source of the fluctuations that the model is designed to explain.
Equilibrium: In each period, the household makes its optimal choice, the firm makes its optimal choice, and the market clears — the quantity of labor supplied by the household equals the quantity demanded by the firm, and the quantity of goods produced equals the quantity consumed plus the quantity saved. The household and the firm have rational expectations: they understand the structure of the model and use all available information to forecast future prices, wages, and interest rates. The equilibrium is the sequence of prices, quantities, and interest rates in which every agent is optimizing and all markets clear in every period.
Solving the model: The model is expressed as a system of equations (the household’s optimality conditions, the firm’s optimality conditions, the market-clearing conditions, and the central bank’s rule). This system is typically solved by linearizing around a steady state — finding the long-run equilibrium and then studying small deviations from it. The solution describes how the economy responds to shocks: how much output falls after a negative technology shock, how quickly inflation rises after a monetary expansion, how employment adjusts over time.
What DSGE Models Are Good At
To their credit, DSGE models have several genuine virtues.
Internal consistency: Every relationship in the model is derived from the optimizing behavior of agents, so there are no ad hoc assumptions about how aggregates relate to each other. The consumption function is not assumed; it is derived from the household’s optimization problem. The Phillips curve is not assumed; it emerges from firms’ pricing decisions under Calvo frictions. This makes the model immune to the Lucas critique: if you change the policy rule, the agents re-optimize, and the model’s predictions adjust accordingly.
Policy counterfactuals: Because the model is structural — because it specifies the deep parameters (preferences, technology, institutional rules) that are presumed to be invariant to policy — it can be used to simulate the effects of policies that have never been tried. What would happen if the central bank adopted a different inflation target? What would happen if the government implemented a particular fiscal stimulus? The model can answer these questions (subject to its assumptions) because its structure does not depend on the current policy regime.
Discipline: The requirement of microfoundations imposes a discipline on model-building that prevents certain kinds of mistakes. You cannot simply assume a relationship between aggregates; you have to show that it can emerge from individual behavior. This forces you to think carefully about incentives, constraints, and the mechanisms through which policy operates.
Common language: DSGE models have become the lingua franca of macroeconomics. Central banks, finance ministries, and international institutions use them for forecasting and policy analysis. Academic macroeconomists use them to formulate and test new ideas. Having a common framework facilitates communication and cumulative progress.
What DSGE Models Are Bad At
The list of what DSGE models struggle with is, unfortunately, at least as long as the list of their virtues.
Financial crises: The most spectacular failure of DSGE models was their inability to predict, explain, or provide useful guidance during the 2008 financial crisis. Most DSGE models in use before the crisis did not include a financial sector at all — there were no banks, no leverage, no asset bubbles, no contagion. The models simply could not generate the kind of systemic collapse that occurred. Post-crisis DSGE models have added financial frictions, but the question of whether these additions are sufficient to capture the dynamics of financial crises remains contested.
Heterogeneity: The representative-agent assumption — the assumption that the macroeconomy can be modeled as if there were a single, typical household and a single, typical firm — rules out by construction one of the most important features of real economies: the fact that people and firms are different. They have different incomes, different wealth, different risks, different access to credit, and different responses to policy. A tax cut has very different effects on a cash-constrained household living paycheck to paycheck and a wealthy household that is saving most of its income. The representative-agent model cannot capture this heterogeneity. Recent work on Heterogeneous Agent New Keynesian (HANK) models has begun to address this limitation, but the field is still in its early stages.
Unemployment: In most DSGE models, the labor market clears — everyone who wants to work at the prevailing wage can find a job. Unemployment, when it appears at all, is voluntary (people choose not to work because the wage is not high enough) or frictional (people are temporarily between jobs). The involuntary, persistent unemployment that characterizes real recessions — people who want to work, are willing to work at the prevailing wage, and cannot find a job — is difficult to generate within the standard framework. Search-and-matching models (Mortensen, Pissarides) have been incorporated into DSGE models to address this, but the resulting models are complex and their quantitative performance is uneven.
Distribution: DSGE models are designed to study fluctuations — how the economy moves around its long-run trend — not the distribution of income and wealth. They can tell you how aggregate output responds to a shock, but they have little to say about who gains and who loses. This is a serious limitation in an era when inequality is one of the central economic concerns.
Expectations: The assumption of rational expectations — that agents understand the model and use all available information to forecast the future — is both the strength and the weakness of DSGE models. It is a strength because it prevents the modeler from assuming that agents are systematically fooled by policy (the whole point of the Lucas critique). It is a weakness because real people do not have rational expectations. They use rules of thumb, they are influenced by narratives and emotions, they disagree with each other about the future, and they are often surprised by events that a rational-expectations agent would have anticipated. Behavioral macroeconomics is a growing field, but integrating behavioral assumptions into DSGE models without losing the discipline of the framework is an ongoing challenge.
The Critics
The critics of microfoundations and DSGE models are numerous and distinguished.
Robert Solow, one of the founders of modern growth theory, has been consistently skeptical. In a 2010 congressional testimony, he argued that DSGE models are built on assumptions that are “wildly at variance with what is known about actual economic behavior” and that their conclusions are “built on sand.” He characterized the representative-agent assumption as “silly” and argued that macroeconomics had taken a wrong turn in the 1970s by abandoning empirically grounded models in favor of theoretically elegant but empirically empty ones.
Paul Romer, in his provocative 2016 paper “The Trouble with Macroeconomics,” coined the term “mathiness” to describe the practice of using mathematical formalism to disguise weak or unsupported arguments. Romer argued that macroeconomics had become a field in which theoretical models were valued for their mathematical sophistication rather than their empirical accuracy, and that the profession’s standards for evaluating models had deteriorated to the point where models that were clearly inconsistent with the data were nevertheless taken seriously because they were mathematically elegant.
Agent-based modelers — Leigh Tesfatsion, Robert Axtell, Doyne Farmer, and others — argue that the entire microfoundations program is misconceived. The goal should not be to derive macro behavior from the optimization of a representative agent, but to simulate the interactions of many heterogeneous agents following simple behavioral rules and observe the macroeconomic patterns that emerge. Agent-based models can generate phenomena — bubbles, crashes, fat tails, persistent unemployment — that DSGE models struggle to produce, precisely because they do not assume equilibrium, rational expectations, or representative agents.
Post-Keynesian economists argue that the Lucas critique, while technically correct, does not justify the microfoundations program. The fact that reduced-form relationships change when policy changes does not mean that the only alternative is to model every relationship as the outcome of individual optimization. It might mean, instead, that macroeconomic relationships are historically contingent — that they depend on institutional structures, power relations, and conventions that are not captured by any model, whether micro-founded or not.
Where the Field Is Heading
Macroeconomics is in a period of ferment. The 2008 crisis shattered the complacency of the pre-crisis consensus, and the discipline is still trying to figure out what to put in its place.
Several trends are evident. First, heterogeneity is being taken more seriously. HANK models, which allow for differences in income, wealth, and constraints across households, are becoming increasingly influential. These models can capture phenomena — such as the high marginal propensity to consume among low-income households — that are crucial for understanding the effects of fiscal policy.
Second, financial frictions are being integrated into macroeconomic models. The work of Brunnermeier, Gertler, Kiyotaki, and others has produced models that can generate financial crises, bank runs, and credit crunches — phenomena that were absent from pre-crisis DSGE models.
Third, there is a growing interest in bounded rationality and behavioral macroeconomics. Xavier Gabaix’s work on “behavioral inattention” and the broader program of behavioral New Keynesian economics are exploring how departures from rational expectations affect macroeconomic dynamics.
Fourth, agent-based models are gaining credibility. The Bank of England, the European Central Bank, and other institutions have begun to explore ABMs as complements to DSGE models, particularly for analyzing financial stability and systemic risk.
None of these trends represents a clean break from the microfoundations program. HANK models are still microfounded. Financial friction models still use rational expectations (mostly). Even behavioral macroeconomics often works within the DSGE framework, modifying the assumptions about agent behavior without abandoning the overall structure. The revolution, if there is one, is evolutionary rather than radical.
The honest assessment is that microfoundations, as a methodological principle, is not going away. The Lucas critique is too powerful to ignore: if you want to evaluate policy, you need a model whose structure does not depend on the current policy regime. But the specific form that microfoundations has taken — the representative agent, rational expectations, continuous market clearing — is being challenged from many directions. The question is not whether macroeconomics needs microfoundations, but what kind of microfoundations it needs. And that question is still very much open.