The Great Recession had a profound effect on business cycle theory. Public opinion turned against the profession, and even other economists seemed to disparage the state of the field, with Nobel laureates like Joseph Stiglitz and Paul Romer rubbishing the last 30 years of business cycle research. Bob Solow even testified to the United States Congress that “popular DSGE models [don’t] pass the smell test”. While tremendous criticism was directed towards rationality, perfect financial markets, and a lack of empiricism, the feature of business cycle models that perhaps provoked the most ire was the use of a “representative agent”.

Essentially, in most modern business cycle models, there is only one person. It is usually perfectly rational, and (given its informational constraints) makes an optimal forecast of all economic variables, then behaving to maximise its utility function. But, using modern computing and innovative modelling, macroeconomists have begun to move towards agents that are heterogenous in income, wealth, access to liquidity, and even rationality. Combine this with the price stickiness found in standard New Keynesian models and you get Heterogenous Agent New Keynesian models, or HANK for short.
So how revolutionary really is this? It’s easy to satirise the idea of a representative agent – have economists just discovered that people aren’t identical? But the question is, if a central banker asks what the effect of raising interest rates will be, or a politician asks whether expansionary fiscal policy works, does HANK actually give qualitatively different answers? For all their apparent absurdity, there is a justification for the representative agent assumption. If markets are “complete” (basically meaning it is possible to insure yourself against any eventuality) then we have theorems that tell us we can always construct a representative agent to model the dynamics of aggregate variables like GDP and inflation arbitrarily well, no matter the degree of heterogeneity. Now markets are obviously never complete, but economics is a social science and all our assumptions are wrong. What matters is whether it is wrong enough that HANK meaningfully changes our understanding of basic questions concerning the causes of business cycles, or the efficacy of monetary and fiscal policies.

The most obvious use-case for HANK is it changing how we understand the behaviour of “aggregate” variables – that is, economy-wide variables like GDP, inflation, and total consumption – in response to monetary and fiscal policies. To understand why HANK might differ, first consider a representative agent’s consumption behaviour. Behaving rationally, if it randomly receives £1000 from the government, it will want to smooth the increased consumption over its lifetime. So rather than spending most of the £1000 now, it will save most of it so it can enjoy higher consumption later. In technical terms, the representative agent has a very low marginal propensity to consume (MPC). Thus fiscal policy based on stimulus checks is ineffective, and the “indirect channel” of monetary policy – whereby monetary policy works by temporarily changing incomes – is of little importance.

But empirical assessments pretty soundly falsify this hypothesis. As a stylised fact, it seems there are two types of consumers. Some behave like the aforementioned representative agent when given transitory increases in income, but others are “hand to mouth” and consume the entirety of changes in income immediately. In other words, there is substantial heterogeneity in consumption behaviour. Indeed, the Campbell-Mankiw model (an early precursor to more sophisticated modern HANKs) has two agents, one that consumes all its income in every period, and one that behaves optimally.
So in HANK, we can specify richer consumption behaviour. The canonical way to do this is assume all consumers are rational but some are “liquidity constrained”¹ – they would like to borrow money so they can consume more in this period, but cannot get the credit. So if they are given a one-off lump sum transfer, they may consume most, if not all of it. Thus MPCs are dramatically higher in HANK than representative agent models.
This allows a richer discussion of monetary and fiscal policies. Kaplan, Moll and Violante (2018) – the paper that coined the term HANK – find that while their estimates of how aggregates change with the real interest rate do not differ substantially from representative agent models, around half of the mechanism by which monetary policy actually works is “indirect channels” only possible in HANK. Furthermore, conventional New Keynesian models usually imply monetary policy dominates fiscal policy unless we are at the effective lower bound on nominal interest rates. But in HANK, fiscal policy can be very effective due to higher MPCs, and indeed Wolf (2023) finds an equivalence result between government stimulus checks and rate cuts in certain models with heterogeneity. So it seems HANK does change our understanding of monetary and fiscal policies, and thus how we advise policymakers to manage business cycles.
HANK also has another advantage over representative agent models. Even if the behaviour of the economy on aggregate is described perfectly well by conventional models without heterogeneity, they cannot tell us anything about the dynamics of outcomes like inequality. For example, consider the oft-made statement that inflation is equivalent to a regressive tax. Representative agent models cannot study the regressivity of inflation because there is only one agent, rather than a continuum of agents with different incomes. So we need HANK to study the distributional effects of monetary and fiscal policies.

This is important because the study of optimal policy is normative economics; to say a policy is “optimal” requires explicit value judgements. While monetary policy is not a tool well-suited to addressing distributional concerns, it is critical economists and policymakers understand the policies they promote as, even if they are beneficial on aggregate, they may disproportionately harm some groups according to wealth and income. To illustrate, HANKs usually imply that more hawkish monetary policy benefits creditors and harms debtors, who see the real value of their liabilities rise.² So monetary policy has distributional consequences; in the words of Kaplan, Moll and Violante (2023), HANK shows the notion “a rising tide raises all ships” is “a fiction”, and so-called optimal policy could affect some households negatively.
Ultimately, this field, while exciting, is still in its infancy, and it is impossible to know what its lasting impact will be. In macroeconomics there have been many ideas that, while received as gospel at first, have had scant impact on the subject in the long-run, from 1960’s class struggle models to new dynamic public finance more recently. And I have even neglected to mention the associated mathematical and computational difficulties – when economists are co-authoring with Fields medalists and begging mathematicians for help,³ the onus is on the model to justify all the effort.
But HANK has already produced results that change our understanding of how macroeconomic policy works – and thus how we advise policymakers. And once you start thinking about how policies designed to target aggregate variables impact individuals based on income, age, and wealth, it becomes hard to stop; indeed if you care at all about inequality, it changes how you view macroeconomic policy in general. So it appears that HANK is here to stay, and I for one am excited to see how it ends up changing macroeconomics.

Footnotes

1. A lot of HANKs also have consumers who are hand to mouth due to short-sighted optimisation, rational inattention, or just plain irrationality
2. This particular insight is of course older than HANK; Sargent (2023) is a good article relating old theories to HANK
3. See https://benjaminmoll.com/wp-content/uploads/2020/10/MFG_macro.pdf

References