Kartik B Athreya

Good evening. I’ll begin by thanking the VAE, Dr. Melanie Fennell and Dr. Evelyn Nunes for the honor to give this year’s Sandridge Lecture. I’m humbled to be among those who have given this lecture in years past. I’ll do my best to live up to that standard.¹

I’ll talk this evening about some research at the intersection of household finance and macroeconomics. But I also hope my remarks can inform teaching of economics to those at the very earliest entry points into our field, the latter being a stated goal of the VAE! As such, there will be a good deal of nonoriginal content in what follows. One of my goals today is to save you time later, so the printed version of these remarks will direct you to many papers, including some sterling overviews of the literature written by leaders in the field. Lastly, because these are my remarks, I’ll abuse the privilege slightly and cover the epsilon way in which I may have contributed to the field.

I’ll organize my remarks around work that studies three questions:

1. Why do we observe the consumption, income, and wealth disparities that we do?
2. (How) Do these disparities matter for outcomes, especially here in the U.S.?
3. (How) Do these disparities matter for policy, particularly consumer credit policy?

Why Do We See Disparities in Economic Outcomes?

Let me start with the first. Why do we see the disparities we do? My tentative and imprecise answers would be as follows.

For consumption disparities, I’d say: “Proximately, because individual consumption closely tracks individual income over longer horizons, it inherits the sizable dispersion in U.S. lifetime incomes. And because the evidence suggests imperfect income insurance, especially against long-lasting shocks, we get sizable disparities even within groups who have similar average lifetime incomes, such as the groups of all those with a high school education, or with a college education, and so on.’’

As to why consumption tracks income, I’d say: “The spending behavior of a large group can be understood as the outcome of a race between the discounting that consumers apply to the future, on the one hand, and precautionary motives and liquidity constraints on the other.” Discounting spurs borrowing, while precautionary motives and liquidity constraints not only restrain current spending but also inhibits routine usage of high debt, perhaps for fear of hitting a future constraint and losing access to credit in unusually bad times.^{2 3}

For the earnings disparities that loom large in generating consumption disparities, I’d say: “They seem attributable in roughly equal part to disparities in human capital investment early in life and shocks during working life.” Moreover, my sense is that, at the left tail of earnings, gender and racial biases operate in a vicious cycle to deter human capital investment, depress earnings, and in that way confirm biases, though I cannot say precisely to what extent.

And as for how to view wealth disparities, I’d say: “At the left tail, my current view is that limits to borrowing — and liquidity constraints more generally — are pervasive.” Note that this also means that absent constraints, we would see even more indebtedness! Conventional wisdom suggests that more indebtedness would be unwelcome, but the view from research suggests that it would instead likely reflect the easing of liquidity or credit constraints. As for wealth at the right tail, especially the far-right tail (e.g., top 0.5 percent), research suggests to me that the keys lie in extraordinary entrepreneurial success (think of the tech moguls), the presence of “superstar managers” (think high level executives for whom technology has allowed greater market reach, or firm size), and differential access to higher-than-mean returns on financial wealth for them (think hedge fund participants). The last point, especially, is an emerging view⁴, so I reserve the right to change my mind when future papers perhaps suggest I should.

How Do Disparities Matter for Outcomes?

Now let me turn to the question of how consumption, income and wealth disparities matter for outcomes. To set the table, let’s start with consumption. In the beginning, as always, apparently, there was Milton Friedman. In 1957, he put forward the model of consumption that I think we’d agree revolutionized how we think about consumption and saving. The biggest insight from his permanent income hypothesis (PIH) is that if credit markets work well and households are not too risk-averse, consumption is all about the annuity value of remaining lifetime income. This in turn meant that if an imperfectly insurable shock occurred, as Friedman allowed for, it would affect consumption via its effect on annuitized lifetime income. And that meant that the consumption response to shocks was fundamentally about whether a shock was temporary or permanent: The former didn’t much matter; the latter did. Twenty years later, Bob Hall came along and tested this and didn’t find it altogether crazy. The idea that the PIH was a reasonable model of consumption, including for macro consumption, took off.

But all was not well, as perhaps you know. While the PIH was brilliant in helping us see why short-run policies might not change consumption or savings, and why long-run ones might change them hugely, and why in the cross-section consumption rose less than 1-1 with income, it missed other facts. I noted a moment ago that individual consumption closely tracks income over the longer stretches of the life cycle (in the U.S. especially until perhaps age 40, but elsewhere for longer). This, as we all know, is in opposition to the simplest textbook PIH — where income paths don’t much matter given expected present values of lifetime income.   

This tracking is key. It helps us see why wealth stays typically very small, for most people, for much of working life.⁵ And it led to the incorporation of two commonsensical features into the baseline models: 1) a precautionary motive that spurred savings merely because the future was uncertain — that is, abandonment of “certainty equivalence,” and 2) liquidity constraints that prevented severe indebtedness, and whose potential for binding later on further spurred precautionary savings now.

Nobel Laureate Angus Deaton’s book covers this earlier part of the story wonderfully if you’re interested.⁶ It’s probably fair to say that the work done on consumption and wealth at the left tail since then has really been about a couple of things: (1) building better models of the exact ways in which we think households might be exposed to risk and find themselves credit- and liquidity-constrained, and (2)  the macro implications of a world in which a positive fraction of people face such constraints.

Let’s do the macro part first. In the 70s, 80s, and 90s, Truman Bewley, and later Rao Aiyagari and Mark Huggett, among others, established key properties of models in which large groups faced potentially binding borrowing constraints. Taken together, their work showed that the economy’s interest rate was likely pushed downward by the presence of borrowing-constrained individuals. Notice how this dovetails with the most famous Sandridge Lecture of all — Ben Bernanke’s Global Savings Glut. The literature on those so-called “Bewley” models taught us that essentially any setting in which borrowing constraints were important would have a savings glut!⁷

As we learned quantitatively how tight borrowing constraints in the absence of externally provided safe assets could really depress market real interest rates,⁸ we learned that low real interest rates might reflect not just low productivity growth. That’s what would be true in the perfect market settings of the basic growth model after all. Now we saw that risk and borrowing limits and, correspondingly, the insufficient production of safe — and liquid — IOUs in which others could save, would matter too!⁹

If we broaden our question of how income and wealth disparities matter to ask how they matter for efficiency, or income levels, or growth, one view might be: “Well, disparities likely raise the average!” This line of thinking emerges from the commonsensical (and almost surely correct) idea that a highly equal society is also a poor and probably tyrannical one. It interprets disparities in an exceptionally rich society like the U.S. as reflecting a balance between insurance and incentives, allowed to play out over a century or more. The potential efficiency of disparities is formalized in the strikingly original paper of Atkeson and Lucas, in which ever-increasing inequality is optimal!¹⁰ Of course, one doesn’t need to accept that as the last word on real-world inequality, perhaps most of all because our generations are probably not as effectively long-lived as they are in that model, and it seems implausible that our earliest ancestors signed us up for a (very!) long-term insurance contract. Yet, it is thought-provoking. An optimal social insurance arrangement where the insurer lacks information on the needs of the insured offer a trade-off: a transfer now upon request in exchange for a commitment to a path of lower transfers in the future.

An alternative answer is, “Well, disparities likely hold us back.” This view suggests that a world with inequities, especially between large groups (e.g., men and women, Whites and Blacks, etc.), is unlikely to be tapping human potential anywhere near fully. American racial gaps especially are, under this view, not just symptoms of past or present failures to protect civil rights inflicting narrow damage, but are (as Atlanta Fed President Raphael Bostic put so well) a yoke on all of us. 

Moreover, when we see — as we plainly do — people “just like us” having to make wrenching lifestyle adjustments to events that in principle ought to be fully insurable, it is hard to accept disparities in wealth as simply part of efficient market functioning. The most vivid example to me is donation websites set up to finance critical surgeries for people. There is, to my mind, no meaningful “moral hazard” that could render (constrained-) efficient having to go out of pocket for several years of earnings simply because one’s kidneys fail.

At shorter, say, business cycle, frequencies, we have a newer, better answer in light of COVID and the massive policy response to it. Namely, prior to 2019, I would have said, “Well, the approximate aggregation result, due to Krusell and Smith, teaches us to maybe ignore heterogeneity in wealth across households when our goal is positive analysis of business cycles.”¹¹ In other words, like most other macroeconomists, I would have said that for the purpose at hand, the representative agent … represents! Seeing why this occurs also allows us to see the mechanics of baseline models of disparity more clearly.

The standard way to generate quantitatively-disciplined inequality in income and wealth within our models, especially during working life, is by using three ingredients: (1) empirically plausible permanent earnings differences (interpreted most naturally as arising from different human capital levels at the time of labor market entry); (2) more transitory earnings risk; and (3) imperfect tools — usually just a savings account or some credit access — to manage income and other (e.g., health) risks.¹²

These ingredients then make a funny kind of cake. The permanent parts of the differences between people in their earnings capacity naturally drive permanent wedges between their general living standards. Thoracic surgeons and schoolteachers live different lives on average, and aside from why one chose that versus another, it’s not some great mystery. If there is a policy implication, though, or a warning about waste in the economy, it surfaces in group-level gaps in career choices. For example, if I see that nearly all thoracic surgeons are male and non-Black, it stretches all credibility to think young Black people mysteriously dislike thoracic surgery more than everyone else. It is instead, a warning that we are throttling the potential, perhaps at a grand scale, of generations in a row.

The more transitory parts of income, and especially the risks to it along one’s life, then create disparities in lifestyle and wealth (and debts) between even those who are broadly similar in their average lifetime earnings prospects. The roots of these disparities are in the lack of availability of both credit and insurance. As a result, one’s financial position reflects not just one’s expected future income path, but also the history of shocks, good and bad, that one received. Those who arrive in old age without having been hit by misfortune will — if they have saved steadily — have a comfortable effective balance to preserve pre-retirement living standards. The opposite will be true for the unlucky. As households reach retirement, these differences can become significant, as time, fortune, and compound interest have done their work.

Returning to the business cycle consequences of such consumption and savings behavior, it turns out that motives to save to “self-insure” are powerful, too powerful, and lead anyone patient, and lucky enough to avoid visitation from ill luck, to build up sizable precautionary balances. As this occurs, the luckiest — and most patient — become the wealthiest and, critically, also become similar to each other in terms of their marginal responses to shocks. This means that there comes to be, in due time, a “representative” wealthy agent, who, moreover, is wealthy enough to own most of the economy’s wealth. Thus, the economy’s aggregate investment and capital accumulation dynamics largely play out as if there was a single representative agent. In other words, for positive business cycle analysis, models that use the heterogeneity as I’ve just described to generate the high level of wealth inequality we observe in the data will act — in the wages and prices and aggregates that it serves up — like models in which one “representative” agent who owns most of the wealth is making investment decisions!

Yet, the normative implications are not so rosy: Recessions are painful for the poor in these settings. It is an instance of a more general tension: What is required for positive analysis may be less than what it takes for a model to get the normative implications right. Indeed, in Krusell and Smith’s model itself, we see this: The dynamics of consumption (and hence utility) do depend on heterogeneity. To the extent that we use the simpler models in our calculations, we need to stay careful about the welfare economics: Here the representative agent may not, at all, represent appropriately in all the ways we need it to.

Now, in the wake of the past few of years of research especially, we can qualify this result even on strictly positive grounds. While I won’t cover it further today, I’ll note that this doctrine of “approximate aggregation” has been revised in light of a merger of the literatures on sticky-price macro — ubiquitous in monetary policy analysis — with the purely real heterogeneous agent models of the previous two decades. Work by McKay and Reis, Moll, Mitman, Kaplan, Violante and many others finds that disparities are important for how policy, including monetary policy, matters.^{13 14} That is, disparities matter for “Aggregate Demand” management.

At the heart of this is the recognition that a large group, maybe up to 40 percent of all households in the U.S., might be so liquidity-constrained that they are “hand to mouth.” Very surprisingly, the evidence suggests that maybe two-thirds of this group are hand-to-mouth despite being “wealthy!”  That is, despite owning homes, stocks, and other assets, this group lacks the ability to finance cheaply consumption above income for any sustained period of time. This work really affected my view of how to interpret the American consumer.¹⁵ Furthermore, the finding is key for another reason: It demands that we take the household’s whole portfolio seriously — net worth is simply too coarse. And this is why understanding deep disparities requires understanding balance sheet disparities.

Do Disparities Matter for Policy?

Here’s an example of how balance sheets can matter for policy. In general, the targets of stimulative transfers will be households that are the most credit constrained, and hence “high MPC” households. Yet we all know that income effects will push anyone to work less when their wealth or income rises. So, recipients may wish to work less, while the reverse is true for the relatively wealthy donors. We have a horserace. In a paper of mine with colleagues at the Richmond Fed, we show that when balance sheet and earnings disparities are taken empirically seriously, a wealth transfer from rich to poor may deliver a consumption boost but an output slump! ¹⁶ In other words, the distribution of marginal propensities to work across the wealth distribution, “MPWs” if you will, may be key. So, understanding balance sheet and earnings disparities may be important not just for long-run aggregates and interest rates, but for short-run policy too.

Let’s move now to one part of the population: those with at least part of their balance sheets in the left tail. They matter for normative reasons simply because they are closer to the edge in general, and they matter for positive reasons since policy recognizes this and effects transfers and sets consumer credit policy with that precarity in mind. Getting at this tail means getting at what determines the terms of credit, why people can’t or don’t borrow, and what they do instead. And here, not only can I tell you about real progress, I can engage in shameless self-promotion. What’s not to like?!

The left tail of net worth, or even the left tail of gross indebtedness, are influenced by not just the income prospects and risks faced by consumers, but also by their ability to commit to debt repayment. This obvious fact is the jumping-off point for a now deep literature on defaultable consumer credit. In early work, Tao Zha and I, working independently, quantified a basic trade-off in allowing consumer debt default in settings where income insurance was not perfect. It was all just applied common sense: Default allows one to renegotiate — in effect — debt repayment if repaying as initially promised is painful. But creditors understand this and will price credit accordingly.

Whether we interpret observed default as healthy exercise of an embedded option, or as unhealthy constraint on credit foisted by the law, is a strictly quantitative matter, though. Hence, the details of risk, its direct insurability, the costs of default, and the availability of both collateral and informal debt delinquency all matter for one’s view of consumer debt policy. This is also why the subsequent explosion of micro-data rich macro models are so important.

Even absent such data, we can make progress, as shown in a paper of Livshits, MacGee and Tertilt in 2007.¹⁷ These authors start by pointing out that there are two kinds of reasons to borrow: to pull forward anticipated increases in income, and to deal with temporary bad luck or income disruption. As we just noted, in a world where comprehensive “income-interruption insurance” is not easily available, the ability to default on debt provides — at least in principle — an escape hatch if things go really wrong. But if creditors, anticipating default risk, make it costly to borrow in these instances, the ability of the young more generally to borrowing against a brighter future gets compromised. And the latter is all young people — not just the unlucky ones. So, we have a trade-off: Default might help smoothing of consumption across contingencies but will hinder smoothing of consumption over time. In their paper, “A Fresh Start,” (Chapter 7 Bankruptcy) can make sense for the kinds of risks U.S. households face — particularly ones that resemble the out-of-pocket kidney surgery I gave earlier, but absent catastrophes or if earnings shocks were mainly transitory, not so much.

These models also help us interpret the indebtedness we observe. Is it high or low? We are tempted to say “high” — as evidenced by the endless policy discussion about debt relief for all stripes of borrowers, students, homeowners, credit card borrowers, etc. But high relative to what? Our models tell us that seeing lots of people getting into trouble with creditors may well indicate that we are too lax on default, and that we are borrowing too little! One concludes this by studying an alternative in which default is simply prohibited — and hence one in which people could borrow up the ratio of their worst-case income and the net interest rate. Of course, if the worst-case income level were zero, that would preclude all default-free borrowing. In the U.S., with substantial if imperfect safety nets, the worst-case income level is nowhere near zero for most. Hence, we can conclude that default risk is meaningfully choking off borrowing capacity among U.S. households, a point I fleshed out in a paper in 2008.¹⁸

A corollary to this is that details matter — maybe too much. If we can’t sharply measure the worst-case scenarios and the details of the safety net (formal and informal) people have access to, we can’t get a tight answer to the question of whether the debt we observe is too high or low in any clear manner. This is where the explosion of empirical work using rich administrative datasets has and will really help us.¹⁹ And this is a highly active area of research today.

A more general lesson is that we want to think about policies toward default and debt forgiveness in concert with the general safety net and redistributive posture we strike. Economists like us need little reminding, but using price-modification and institutional tweaks instead of direct transfers to achieve equity goals is of course generally a bad idea. Nowhere is this exhortation ignored more often than in credit market commentary, with its Manichean language of rapacious creditors and victimized borrowers. Yet, the defaultable-debt literature does clarify how, at times, the victims of seemingly debtor-friendly policy will be mainly those who lack the wealth to post the collateral needed to liquefy their earnings and wealth. The well-off will more easily bypass the unsecured credit market, and due to their higher average living standards, perhaps barely notice.

None of this means extreme penalties for default are a great idea either, though. People do at times get duped, and people do at times find themselves in terrible binds. Hence, coordination between social insurance and consumer credit policy is surely valuable, with the general idea being that harsher sanctions on defaulters — even as those assist with borrowing capacity — may be best when accompanied by better safety nets, to make borrowing not so scary in the first place. Though even here, some of the hurdles from a poor safety net may be overcome by making default costlier — to allow lending to temporarily unfortunate consumers work better, with fewer spikes in credit costs.²⁰

And how do these kinds of models inform high-frequency macroeconomics? I’ll mention one ongoing paper, by Kyle Herkenhoff and co-authors, and a less recent but important one by Kurt Mitman.²¹ The first addresses a natural question: “How can consumer credit help people navigate job loss?” Kyle and co-authors show that unemployed individuals maintain significant access to credit. Following job loss, the unconstrained borrow, while the constrained default and delever. They find that both defaulters and borrowers are using credit to smooth consumption. One interesting feature of their work is that they connect the level of information held by creditors to the ability of consumers (workers in their setting) to maintain credit access, with credit registries playing a role. And to the point about coordinating social insurance with consumer credit, they find that the optimal provision of public insurance is unambiguously lower with greater credit access — in their setting this is a reduction in the generosity of overall “replacement rate” of benefits provided by the safety net. Returning to the “wealthy-hand-to-mouth,” notice that consumer credit terms surely matter for their willingness to tolerate consumption drops — they are non-wealth-poor after all. So their effective illiquidity may come from a combination of transactionally-costly secured borrowing, and unsecured borrowing costs afflicted by any prevailing level of consequences for nonrepayment.

Turning to Kurt’s paper²², he studies how the 2006 reform of the rules governing U.S. consumer bankruptcy — which made Chapter 7 bankruptcy harder to access — significantly reduced bankruptcy rates. But his work is important because it connects unsecured credit to other forms of credit that households have — mortgage debt especially. He shows that when one can “force” consumer debt repayment, but one may need to accept that other debts get deprioritized: He finds that foreclosure rates rose following house price declines. This again suggests that good macroeconomic policy needs to consider the whole balance sheets and institutional settings in which liquidity-constrained households, especially, operate.

So, what do we know about the group in the left tail? Are they permanent residents, or just passing through? In recent work with co-authors, we use data on consumers’ financial distress (as defined by being delinquent on debts)²³ and document that it is highly persistent. Namely, the same people year after year end up in trouble, with the rest almost never in it.

As for why, we show that this is “explained” by what appear to be effectively different discount factors. Indeed, there is a growing view that such permanent differences in time preference may really be “a thing,” and if so, it will offer a nontautological explanation for the left tail of U.S. wealth inequality.  But I’ll be slow to take this literally. For me, deeper investigation is critical into what such “revealed impatience” means for the circumstances under which at least a portion of the U.S. population is operating. Think about two groups, one with reliable cars and flexible salaried work, and other with clunkers and inflexible shift work. The first group, in the rare instance of car trouble for their nice cars, would get a tow and work from home. The second one will not only more often need to fix the car, the failure to do so might cost them their job. So they’ll overdraft, or borrow at high rates, etc., to get through the day. Yet, they’ll look impatient in terms of balance sheet evolution simply because we’re not seeing the double whammies they often have to navigate. Data that can’t peer this far into lives will lead us to assign observed differences to things like consumers’ patience and risk-aversion.²⁴ Caution is therefore warranted. Nonetheless, it is a finding that I think is still informative about drivers of left-tail balance sheet health, and it only buttresses the need for richer microdata.

Concluding Thoughts

Let me conclude. The quantitative macroeconomics of disparities is in an exciting place. The workhorses of this literature are now one human generation old, and their children are healthier and richer. The original models for business cycles have now been extended hugely. They now help us think about monetary and fiscal policy, and indicate that distributions matter for the positive economics.  Importantly, from a normative view, these models have helped us better understand the distributional consequences of both. Best of all, we are getting more clear-headed all the time about what drives the left tail of wealth holdings.

The path ahead will also demand discipline from economists to realize that the ability to produce inefficient disparities is a siren. It will lure partisans seeking to reverse engineer economics to yield to distributional priorities they may hold. This is perhaps fine so long as the papers are transparently written so we can see where the bodies are buried. But publicly facing communications will rarely be closely vetted, and the trust in which we are collectively held by the public will be tested given these newfound powers.

I’ll end with a PSA: Let’s, please, find ways to bake these learnings into early economics coursework.  The “standard-incomplete-markets” model is perhaps advanced material, but the ideas are, if anything, easier to grasp than single-agent macroeconomic models.²⁵ Job loss, illness, insurance glitches, expensive unsecured credit, trouble with bill collectors, lack of collateral, and aging are things so many navigate. So surely models that feature them explicitly will resonate? We have a golden opportunity to inspire and reassure a new generation of would-be economists who are motivated to think about disparities in our economy and the implications for high- and low-frequency policy. We can show them that we get it, and that we have made progress too.

So, our undergraduate texts and lectures ought, in my view, to show them just how rich our engagement with the topic is — even if only as a “teaser.” Maybe we just say: “Hey kids, we’ll do the classical stuff now, but remember, in more advanced classes, you’ll learn how to build models of rich and relevant inequality, and use theory and empirics in amazing ways, because that is what is already happening.” I leave the “how” to experts among you. Thank you for your time.