For decades, macroeconomic models at the core of central bank decision-making have rested on an elegant but deeply limiting assumption: that the behavior of an entire economy can be represented by a single, optimizing household. The representative agent framework simplified the mathematics beautifully. It also obscured something fundamental—that monetary policy doesn't land evenly across the population, and that this unevenness isn't a footnote. It is the transmission mechanism.

The emergence of Heterogeneous Agent New Keynesian models—HANK models, in the field's shorthand—has begun to dismantle this convenient fiction. By explicitly modeling households that differ in their wealth, income risk, and access to financial markets, these frameworks reveal transmission channels that representative agent models cannot see. The marginal propensity to consume out of a transitory income change isn't 0.05 for everyone. For a significant share of households, it's closer to one. That difference reshapes everything we thought we knew about how interest rate changes propagate through an economy.

This isn't merely an academic refinement. When central banks calibrate their policy responses using models that assume away distributional heterogeneity, they systematically misjudge the size, speed, and composition of the aggregate effects they're trying to produce. The policy prescriptions that emerge from HANK models differ—sometimes substantially—from those of their representative agent predecessors. Understanding why requires examining three dimensions: the failure of the representative agent assumption, the amplifying role of liquidity constraints, and the redistributive channel that operates quietly beneath every rate decision.

The representative agent in a standard New Keynesian model is a convenient analytical device. It collapses millions of heterogeneous decisions into a single Euler equation linking today's consumption to expected future consumption and the real interest rate. The implied marginal propensity to consume out of transitory income is small—often between 0.02 and 0.05—because the representative agent smooths consumption efficiently over a long horizon. This is the intertemporal substitution channel that dominates textbook monetary economics.

Empirical evidence has long told a different story. Studies using household-level data—from the work of Jonathan Parker and colleagues on tax rebates to more recent analyses of fiscal transfers—consistently find that the average MPC out of transitory income is somewhere between 0.2 and 0.5, with enormous variation across the wealth distribution. Households near the bottom of the liquid wealth distribution exhibit MPCs approaching unity. They spend additional income almost immediately, not because they're irrational, but because they face binding or near-binding constraints.

HANK models, pioneered in the influential work of Greg Kaplan, Benjamin Moll, and Giovanni Violante, build this heterogeneity into the structural framework. Households face idiosyncratic income risk, can hold both liquid and illiquid assets, and make consumption-saving decisions subject to borrowing constraints. The aggregate MPC that emerges is an equilibrium object shaped by the joint distribution of income, liquid wealth, and illiquid wealth—not a parameter assumed away by construction.

The consequences for monetary transmission are profound. In a representative agent model, a rate cut stimulates demand primarily through intertemporal substitution: lower rates make future consumption relatively more expensive, pulling spending forward. In a HANK model, the indirect income channel dominates. A rate cut stimulates aggregate demand, which raises labor income, which disproportionately affects high-MPC households, generating further demand increases. The feedback loop is substantially more powerful than the direct substitution effect alone.

This isn't a small quantitative correction. Kaplan, Moll, and Violante demonstrate that the direct intertemporal substitution channel accounts for a remarkably small share of the total consumption response to monetary policy shocks. The lion's share operates through general equilibrium income effects that only become visible once you allow households to differ. Representative agent models weren't just imprecise—they were identifying the wrong primary channel.

Takeaway

When the dominant transmission channel of monetary policy operates through income effects on constrained households rather than intertemporal substitution, the entire logic of how rate changes affect the economy shifts—from persuading a representative saver to reschedule consumption, to triggering a cascade of spending among those who spend what they earn.

The distinction between wealthy hand-to-mouth and poor hand-to-mouth households is one of the most consequential insights from the heterogeneous agent literature. Poor hand-to-mouth households hold neither liquid nor illiquid wealth—they are constrained in the traditional sense. But wealthy hand-to-mouth households hold substantial illiquid assets—housing, retirement accounts—while maintaining negligible liquid balances. They own wealth on paper but behave, at the margin, like constrained consumers. In U.S. data, this group constitutes roughly a third of all households.

This matters enormously for monetary policy amplification. When a central bank cuts rates, the initial demand stimulus raises output and labor income. For unconstrained households with low MPCs, much of this additional income is saved. The multiplier is modest. But when a large share of households are effectively constrained—spending most or all of incremental income—the general equilibrium feedback is dramatically amplified. Each round of spending generates income for other constrained households, who spend again.

Representative agent models, by construction, cannot capture this amplification mechanism. The single Euler equation governing consumption already assumes perfect access to financial markets and an interior optimum for saving. There is no room for borrowing limits to bind, no room for the distribution of liquidity to matter. The result is a systematic underestimation of the state-dependent nature of monetary policy effectiveness. The same rate cut has different aggregate effects depending on how many households are near their constraints—a feature invisible to the representative agent framework.

Recent work by Alisdair McKay, Emi Nakamura, and Jón Steinsson formalizes this by showing that forward guidance—promises about future rate paths—is far less powerful in HANK models than in representative agent models. In the standard New Keynesian setup, forward guidance has implausibly large effects, the so-called forward guidance puzzle. Introducing heterogeneous agents with borrowing constraints naturally resolves this puzzle. Constrained households cannot respond to promises about future rates because they cannot borrow against future income. The aggregate response to forward guidance falls to empirically plausible magnitudes.

The policy implication is immediate and practical. Central banks relying on representative agent models may overestimate the potency of forward guidance while underestimating the power of policies that put income directly into the hands of constrained households. The balance between conventional rate policy and unconventional tools looks different through the HANK lens—and the optimal policy mix depends on the current state of the wealth and liquidity distribution, not just on the output gap and inflation.

Takeaway

The share of households operating near their liquidity constraints acts as a hidden multiplier for monetary policy—and since that share varies over the business cycle, the effectiveness of any given rate change is fundamentally state-dependent in ways that standard models cannot capture.

Every monetary policy action redistributes. A rate increase transfers real resources from debtors to creditors through the Fisher effect on nominal contracts. It alters the relative returns on different asset classes, benefiting holders of short-duration bonds while depressing equity and real estate valuations. It shifts labor income risk across sectors and skill levels. In a representative agent world, these redistributions net to zero in aggregate because there is only one agent—they wash out by assumption. In a heterogeneous agent world, they are a primary channel through which policy affects aggregate demand.

The work of Adrien Auclert provides perhaps the most systematic decomposition of this redistributive transmission mechanism. Auclert identifies several distinct channels: an earnings heterogeneity channel (rate changes affect labor income unevenly across households), an interest rate exposure channel (households with net nominal liabilities lose from rate increases while net savers gain), and a Fisher channel (unexpected inflation redistributes from creditors to debtors). When these redistributive flows move resources between households with different MPCs, they generate aggregate demand effects that are absent from representative agent frameworks.

Consider a concrete example. A 100-basis-point rate increase transfers income from mortgage holders—many of whom are high-MPC, wealthy hand-to-mouth households—to depositors, who tend to be wealthier and have lower MPCs. The net effect on aggregate consumption is negative not just because of the standard intertemporal substitution channel, but because purchasing power has shifted toward agents less inclined to spend. Auclert's estimates suggest this redistributive channel accounts for a quantitatively significant share of the total consumption response—on the order of the direct substitution effect itself.

This reframing has uncomfortable implications for central bank communication and mandate design. If monetary policy inherently redistributes, then distributional outcomes are not side effects to be noted and set aside—they are integral to the transmission mechanism. A central bank that ignores distributional consequences isn't being apolitical; it's using an incomplete model of how its own tools work. The aggregate outcomes it cares about—output gaps, inflation—are themselves shaped by the distributional shifts it sets in motion.

For optimal policy design, the implications are significant. The welfare-maximizing interest rate path in a HANK model generally differs from the one prescribed by a representative agent model, because the social planner in the HANK world internalizes the redistributive costs and benefits of rate changes. In some calibrations, this implies a more cautious approach to rate increases during periods when household balance sheets are fragile and the share of constrained borrowers is elevated. In others, it suggests that conventional monetary policy should be paired with fiscal instruments that offset undesirable distributional consequences—a coordination that representative agent models never motivated.

Takeaway

Monetary policy is redistributive by nature, not by accident. Once we recognize that shifting resources between high-MPC and low-MPC households is a core transmission channel, the boundary between monetary and distributional policy dissolves—and optimal policy design must account for both.

The transition from representative agent to heterogeneous agent frameworks is not a marginal improvement in macroeconomic modeling. It is a paradigm shift in how we understand the mechanisms through which monetary policy operates. The channels that matter most—income effects on constrained households, state-dependent amplification through liquidity constraints, redistribution between debtors and creditors—are precisely the channels that the representative agent assumption eliminates by construction.

For central banks, the practical stakes are high. Policy calibrated to representative agent intuitions may misjudge both the magnitude and the composition of its own effects. Forward guidance may be less potent than expected. Rate changes may be more potent—and more unevenly felt—than standard models predict. The optimal policy response becomes contingent on the distribution of household balance sheets, not just on aggregate gaps.

The HANK revolution is still unfolding. But the direction is clear: macroeconomic policy analysis that ignores heterogeneity is not just incomplete—it is systematically biased. The models we build shape the policies we design, and better models are beginning to demand better policy.