Monetary Policy and Heterogeneity: A Note (Inherently Wonkish)

Over the summer I began research for my MPhil thesis, which will seek to explore how introducing inequality in wealth and access to financial markets affects the monetary policy prescriptions of the standard workhorse model of modern macroeconomics used by policymakers in the central banks: the so called ``New Keynesian’’ model. My interest in this area stems from two strands of the modern macroeconomic literature, which I have come across at different points during my time as an Economics student to date.

The first such strand is the "heterogeneous agents’’ macroeconomic literature, which began with seminal contributions by Aiyagari (1994) and Hugget (1993). Before this literature, macroeconomic models had tended to feature a ‘’representative agent’’. Solving the optimising problem of the representative agent was effectively finding out the conditions of rational behaviour for everyone in the economy, which people concerned about maximising their standard of living subject to budget constraints would follow. This is simplifying assumption. It emphasises the similarities between people in the economy – that all people generally wish to consume ``stuff’’ (including leisure time) and generally don’t like having to work too much. This does not mean macroeconomists believe things like differences in wealth, income, preferences etc. between people are unimportant. Rather, it is a classic simplifying assumption for the purpose of simplifying the algebra involved in ``solving’’ the model, so that an algebraic solution to the mathematical problem actually exists, which is easily interpretable and can shed light on some key mechanisms when people make day-to-day decisions about consumption and saving in the aggregate.

As mathematical and computational techniques made advances, it became more feasible to model these ‘’differences’’ between people in the economy, and show that these can have important implications for the design of policy. I studied some simple examples of these in an Advanced Macro course I took at Durham led by David Chivers. Two papers stood out in this course, which seem to capture the benefits of capturing heterogeneity. One is a paper by Galor and Zeira (1993) in which people who can borrow but have differences in initial wealth lead to two equilibria: one where those with lower initial wealth cannot invest in ``human capital’’ (could be higher education or some other skills-based course) and so remain as low-skilled workers with a lower lifetime standard of living; and one in which those with higher initial wealth will choose to invest in themselves to become high-skilled workers and have a higher lifetime standard of living. The second was a paper by Alesina and Rodrik (1994) where people differ in the share of their income which comes from labour (working) and the share which comes from capital (investments, savings etc.): the authors show in this paper that the rate of tax on capital that will be chosen by people in this (democratic) economy is not the rate which maximises growth.  This model nests Barro’s (1990) result (where the optimal tax rate on capital maximises growth) as a special case, which instead in this model is that preferred by those whose entire income comes from capital. These are but two examples of an enormous literature on heterogeneous agents in macroeconomic models. In this light, hopefully the reader can see why economists find it irksome to read post-2008 criticisms which accuse macroeconomics of ignoring the differences between people and believing that the representative agent assumption is treated as a truism.

The second strand of literature is that on the New Keynesian model itself. The demand side of the New Keynesian model is the standard infinitely lived representative agent paradigm. The key features of the New Keynesian model (which set it aside from the older Real Business Cycle models) are two key assumptions: firstly, that firms produce different goods from each other (all of which are assumed to be demanded by households) and are thus monopolistically competitive so choose prices, unlike the perfect competition assumption of prior models where firms are price-takers; and secondly that prices cannot be instantly changed by firms, or rather that changing prices is costly which must be taken into account by firms when they solve the problem of maximising their profits. The optimal policy in the New Keynesian model is famously the one which “closes the gaps” – meaning the “inflation gap” (the difference between inflation and its steady state level, which is usually assumed to be zero) and the “output gap” (the different between the actual level of GDP and that level which would prevail in the absence of inflexible price-setting).

“Optimal” in this sense effectively means “efficient” in the classical sense. That is, there exist no price or market distortions which mean there is some trade which would have taken place between households and firms in the absence of such distortions, but didn’t. When learning about this in graduate macro (which is absolutely standard), it struck me that efficiency based on a representative agent may not really give truly “optimal” policies at all, as it has nothing to say about income/wealth distribution, or differing financial circumstances of people in the real world. As was the point in the Alesina and Rodrik (1994) paper, the “optimal” policy in terms of maximising growth is not actually the one that is best for the majority of people in the economy, as it ignores the distribution of the gains from growth. This was when I became interested in the merging of these two strands of literature in macroeconomics.

It turns out that this merger is a very recent phenomenon in macroeconomics. In a seminal contribution, Kaplan, Moll and Violante (2014) termed the introduction of heterogeneity into New Keynesian models “Heterogeneous Agent New Keynesian” Models (or “HANK” for short). In their paper, Kaplan et al. (2014) focus on the transmission mechanism of monetary policy through to the decisions of people, decomposing  the effect of a central bank’s decision to change the base interest rate into a direct effect (lowering the base rate of interest will generally lower the cost of present consumption for people) and an indirect effect (the change will affect overall demand for goods and therefore for labour, which may amplify or dampen the direct effect). One potential early advantage of HANK models is that they may help resolve some fundamental “puzzles” associated with the standard representative agent New Keynesian (or “RANK”) model. McKay, Nakamura and Steinsson (2015) claim that a HANK model can resolve the so called “forward guidance puzzle” raised by Del Negro, Giannoni and Patterson (2015) – the problem that in the New Keynesian model, the promise by the central bank to keep future interest rates lower for longer theoretically has a huge impact on current GDP today, though this has been completely absent in the events of the last 5 years or so. Bilbiie (2017) has more recently shown that HANK models do resolve the forward guidance problem, but at the cost of achieving amplification (the idea that an increase in consumption leads to a further increase in output, and so on) which is the key insight of “old” Keynesian macro – the standard undergraduate “Keynesian cross” diagram should spring to mind for anyone who has studied economics as an undergraduate. This is itself a puzzle, which I hope to look in to.

In terms of policy prescriptions from HANK models, there has not been enough research in this area as of yet to say definitively what the impact of introducing heterogeneity into the New Keynesian model has been in terms of its effect on policymaking prescriptions. Perhaps the most explicit exercise in deriving the optimal policy is that of Bilbiie and Ragot (2016), who conclude that policy prescriptions are quantitatively different in HANK models compared to the RANK model (Specifically, monetary policy is slightly more effective in a HANK world), but qualitatively this difference appears negligible. Indeed, the authors stress that this new framework doesn’t suggest that monetary policy needs to be radically different from where it is currently. Using a simpler two-agent setup (a so called “TANK” model, for obvious reasons…), Debortoli and Galí (2017) similarly find the optimal policy and conclude that central banks will care about preventing increases in inequality between people, but the weight they place on this is in the optimal solution to this model is perhaps surprisingly low.

This is by no means the end of the road for heterogeneity in the macroeconomic policy, and there are many more examples and applications which I neglect to mention here. There is much work still to be done, to see how these results change when the modelling assumptions and the economic environment are allowed to change. This is what I am hoping my thesis can achieve in some small part.