Monetary Policy and the Redistribution Channel

The paper rejects the “conventional view that redistribution is a side effect of monetary policy changes, separate from the issue of aggregate stabilization,” arguing instead that “redistribution is a channel through which monetary policy affects macroeconomic aggregates, because those who gain from accommodative monetary policy have higher marginal propensities to consume (MPCs) than those who lose” (Introduction, p. 1). The argument traces to Tobin (1982), quoted at length: “the population is not distributed between debtors and creditors randomly. Debtors have borrowed for good reasons, most of which indicate a high marginal propensity to spend from wealth or from current income” (p. 1). Most models of monetary transmission implicitly assume this channel away by using a representative agent, in which redistribution is definitionally impossible.

Monetary expansions “tend to increase real incomes, to raise inflation and to lower real interest rates,” and “not everyone is equally affected by these changes” (Introduction, p. 1). First, the earnings heterogeneity channel: “monetary expansions increase labor and profit earnings,” and “the distribution of these gains is unlikely to be equal.” Second, the Fisher channel: “unexpected inflation revalues nominal balance sheets, with nominal creditors losing and nominal debtors gaining.” Third, the interest rate exposure channel: real interest rate falls “increase financial asset prices,” but “it is incorrect to claim that asset holders generally benefit: instead, we have to consider whether their assets have longer durations than their liabilities” – since consumption plans count as liabilities and human capital as an asset, exposure depends on balance-sheet duration structure, not simply on net wealth.

Theorem 1 shows that a household’s consumption response to a shock decomposes into a substitution effect and a wealth effect equal to MPC times a net wealth revaluation dΩ = dy + ndw − NNP·(dP/P) + URE·(dR/R) (Sec. I.A, eq. 3-6). The theorem “makes no assumption on horizon or the form of u and v” and is shown (Theorem 2) to extend to incomplete markets, idiosyncratic income risk, and certain borrowing constraints, “generaliz[ing] previous findings by Kimball (1990) on the importance of MPCs in incomplete-markets consumption models” (Sec. I). The key implication is that a household’s MPC out of a windfall is exactly what determines how strongly it responds to inflation- or interest-rate-induced changes in its balance sheet, not just to income changes.

NNP is defined as the present value of a household’s nominal assets minus its nominal liabilities (Sec. I.B): “a nominal saver with NNP = $100k experiences a wealth effect of… the equivalent of $1000” from a 1% unexpected permanent price-level increase, while “a nominal borrower with NNP = −$100k gains the equivalent of $1000.” Citing Doepke and Schneider (2006), the paper notes NNPs “are large and heterogenous in the population: they are very positive for rich, old households and negative for the young middle class with mortgage debt” – so unexpected inflation mechanically redistributes from old, wealthy nominal creditors toward younger, indebted households, and the aggregate consumption effect of this depends on whether the latter group has systematically higher MPCs.

URE is defined as the difference between all currently maturing assets (including current income) and currently maturing liabilities (including planned consumption) – “the net saving requirement of the household at time 0, from the point of view of date −1” (Sec. I.B). The paper explains why simple net wealth is the wrong measure: a fall in the real rate raises the present value of both future assets and future liabilities (since planned consumption is itself a liability), so “consumers experience a net wealth gain only if their future assets exceed their future liabilities, which… can only happen if their currently-maturing liabilities exceed their currently-maturing assets, i.e. if URE < 0.” A household holding only short-term deposits (URE > 0) is hurt by falling rates even if it is a net asset holder, while a household with adjustable-rate mortgage debt or near-term spending plans (URE < 0) benefits.

Theorem 3 shows that the first-order response of aggregate consumption to simultaneous shocks to output, inflation, and the real interest rate is the sum of five terms: an aggregate income channel, a substitution channel (the only two present in representative-agent models), and three redistribution channels given by cross-sectional covariances – Cov(MPC, relative income exposure) for earnings heterogeneity, Cov(MPC, NNP) for the Fisher channel, and Cov(MPC, URE) for the interest-rate-exposure channel (Sec. II, eq. 19). These covariances are “sufficient statistics” in the sense used since Harberger (1964) in public finance: “equation (19) holds irrespective of the underlying model generating MPCs and exposures at the micro level… Most of the bracketed terms are cross-sectional moments that are measurable in household level micro-data,” so the aggregate consumption response can be evaluated without fully specifying the structural model generating household behavior.

The paper measures its sufficient statistics using the 2010 Italian Survey of Household Income and Wealth (self-reported MPC via a hypothetical windfall question, following Jappelli and Pistaferri 2014), the 1999-2013 U.S. Panel Study of Income Dynamics (a semi-structural approach backing out MPCs from consumption-income co-movement, following Blundell, Pistaferri and Preston 2008), and the 2001-2002 U.S. Consumer Expenditure Survey (MPCs identified from the randomized timing of 2001 tax rebate receipt, following Johnson, Parker and Souleles 2006) (Sec. III.A). Using three methods across two countries and different time periods is deliberate: “given that sufficient statistics are likely to vary over time and across countries, this exercise gives a sense of robustness to the fundamental setting as well as the estimation method” (p. 25) – the paper is explicit that each method “has its own limitations” and that the exercise is “tentative,” intended to “give a sense of magnitudes” rather than definitive point estimates.

Across all three surveys, “the covariance between MPCs and UREs is also negative,” implying “the interest rate exposure channel acts in the same direction as the substitution channel, and with comparable magnitude provided that σ is between 0.1 and 0.4” – squarely within the range macroeconomists typically use (0.1 to 0.5, per Hall 1988 and Havránek 2015) even though financial economists often prefer values around 2 (Bansal et al. 2016) (Sec. III, p. 3-4, 31). The estimated redistribution elasticity Ê_R ranges roughly from −0.06 to −0.20 across duration-assumption scenarios in the SHIW, and around −0.11 to −0.14 in the PSID (Table 5), so representative-agent analyses “may fail to capture an important reason why real interest rates affect consumption, especially if σ is small.”

The Fisher channel is correctly signed but small: “across datasets, the covariance between MPCs and NNPs is negative on average… However, when cast in terms of elasticities, the magnitude is small: an unexpected 1% permanent increase in the price level raises consumption today by no more than 0.1%,” so “while changes in monetary policy can entail significant nominal redistribution, the aggregate effect of this redistribution on consumption is likely to be modest” (Introduction, p. 3-4). For the earnings heterogeneity channel, “the covariance between MPCs and gross incomes” is negative in all three surveys, “confirming previous findings in the literature,” and “if, in addition, low-income agents disproportionately benefit from increases in aggregate income – as suggested, for example, by Coibion et al. (2017) – the earnings heterogeneity channel also amplifies the effects of monetary policy” (Introduction, p. 4; Sec. III, p. 31).

The conclusion stresses that “capital gains and losses, both nominal and real, matter for understanding monetary policy transmission,” with practical consequences: “a change in the inflation target can create large redistribution in favor of high MPC agents and be expansionary over and beyond its effect on real interest rates,” while “with long asset maturities, lower real interest rates can benefit asset holders with lower MPCs and make interest rate cuts less effective at increasing aggregate demand than they would otherwise be” (Sec. IV, p. 35). The paper concludes that “monetary policy becomes intertwined with fiscal policy, but also with government debt maturity management and mortgage design policies” – i.e., the size of the redistribution channels depends on institutional features (debt maturity structure, mortgage contract design) that are themselves policy choices, not fixed technological parameters.

The paper is explicit about scope limits (Introduction, fn. 1): it “abstract[s] away from aggregate risk, so cannot handle changes in risk premia,” does “not model limited [asset market] participation,” and, “since I assume that all assets are remunerated at the risk-free rate,” does not address “the unequal incidence of inflation due to larger cash holdings by the poor.” Relative to full structural HANK models such as Gornemann, Kuester and Nakajima (2016), McKay, Nakamura and Steinsson (2016), and Kaplan, Moll and Violante (2018), the paper argues its sufficient-statistics approach makes “two contributions”: it offers “a decomposition of the monetary policy transmission mechanism into its various sources of effects on consumption that is useful to shed light on the underlying mechanisms in any such model,” and it proposes that “sufficient statistics can discipline the construction of these models,” since “making sure that the model’s sufficient statistics match the data” provides a model-robust check on a structural model’s predictions (Introduction, p. 4-5).