As central banks across the globe have responded to the COVID-19 shock by rounds of extensive monetary loosening, concerns about their inequality impact have grown. But rising inequality has multiple causes and its relationship with monetary policy is complex. This paper highlights the channels through which monetary policy easing affect income and wealth distribution, and presents some quantitative findings about their importance. Key takeaways are: (i) central banks should remain focused on macro stability while continuing to improve public communications about distributional effects of monetary policy, and (ii) supportive fiscal policies and structural reforms can improve macroeconomic and distributional outcomes.
This is a preview of subscription content, log in via an institution to check access.
eBook EUR 117.69 Price includes VAT (France)
Softcover Book EUR 158.24 Price includes VAT (France)
Hardcover Book EUR 158.24 Price includes VAT (France)
Tax calculation will be finalised at checkout
Purchases are for personal use only
For example, heterogeneity in the sources of household income could imply that, everything else equal, wealthier households (who rely relatively more on business income) benefit more from monetary policy easing than poorer households (more reliant on labor income). However, poorer households are also more likely to be able to participate in labor markets or avoid unemployment when monetary policy eases in response to reduced economic activity.
The authors look at the evolution of inequality in the aftermath of SARS (2003), H1N1 (2009), MERS (2012), Ebola (2014), and Zika (2016).
Moreover, policymakers have often stressed that the increase in inequality and the decrease in interest rates are both long-term phenomena.
See Lenza and Slacalek (2018), for 17 euro area countries, using HFCS (2016) and Amberg and others, 2021 for Sweden.
For instance, in Italy, the youth unemployment rate among young workers was about 40 percent in 2014, above the national average and the unemployment rate among older workers (7 percent). As the economy recovered, the youth unemployment rate fell by 5 percentage points by 2016, while unemployment rate among older workers remained broadly unchanged over the same period (Marino & Nunziata, 2017).
While different maturity assumptions can lead to different results, our results are in line with those found in Auclert (2019) and Tzamourani (2021). By analyzing changes in net interest income, Dossche et al. (2019) find that due to the reduction in the interest rate an average euro area net borrower has gained close to EUR 2000 per year in lower interest payments during 2007–2017, whereas an average net saver has lost close to EUR 700 per year.
This simple simulation focuses on within cohort heterogeneity. However, it is also important to note that, even though higher house prices benefit households at the bottom of the distribution, rising house prices may have negative effects on young and middle-aged households that plan to purchase or increase their housing stock to accommodate a growing family. These intergenerational aspects, not explicitly discussed here, matter for wealth inequality (Bielecki et al., 2022).
See KMV, Table 8, column 4. This is the baseline specification in KMV. See Section IV.D for a detailed discussion. See KMV Online Appendix, Table E.1, column 4. See KMV Online Appendix, Table E.2, column 5.The effect on wages is analyzed separately for hourly wages and hours worked. We do not find evidence of an effect on hours worked. Hence, the latter results are not shown for sake of brevity.
See Colciago et al. (2019) for a more comprehensive stocktaking on the theoretical and empirical literature on monetary policy and inequality.
By contrast, Inui et al. (2017) find that monetary policy shocks do not significantly affect income and consumption inequality when using Japan’s data during the 2000s.
Amberg et al. (2021); also see Di Casola and Stockhammar (2021).Albrizio et al. (2021) investigate the distributional effects of U.S. monetary policy for the pre- and post GFC periods. In the latter period, monetary policy actions were mostly expansionary, took place in a period of weak economic activity, and were mostly unconventional measures. The results suggest that monetary policy expansion tends to increase employment and output and has small effects on income and consumption inequality and could even reduce inequality over the medium term. The effects are not statistically different between conventional and unconventional monetary policy, but tend to be larger for tightening than easing, confirming earlier results on asymmetric effects (Tenreyro & Thwaites, 2016).
See, for example, Aladangady (2014), Kaplan et al. (2014, 2018), Cloyne et al. (2019), Gelos et al. (2019), and Dossche et al. (2021).
More generally, even if monetary policy is unconstrained and can handle macroeconomic stabilization on its own, fiscal policy tools such as automatic stabilizers are useful as providers of social insurance.
Even if it were impossible to add more instruments to the central bank’s toolkit, it is fairly straightforward that assigning a loss function to the central bank that mimics the social loss function (i.e., one that could include a distributional term) would yield monetary policy that is optimal conditional on the toolkit and the state variables. However, the key point made by Davig and Gürkaynak is that “optimal monetary policy” may not be “optimal policy.”
The policies need to be designed carefully because some policies can rather have adverse distributional effects (Fabrizio et al., 2017). For example, making labor market flexible tends to increase inequality (e.g., Kahn, 2012; Ostry et al., 2021).
Slacalek et al. (2020) add the various income and balance sheet channels together and find that the effects on labor income benefit all households—particularly “hand-to-mouth households”—and quantitatively outweigh the effects on financial income.
Evidence from the United States suggests that, without quantitative easing in response to the GFC, unemployment would had been persistently higher through the end of 2018, hurting households at the lower end of the distribution the most (Eberly et al., 2019). Similarly, unconventional monetary policy evidently increased GDP through 2014–w18 in the euro area and potentially lowered the unemployment rate (Rostagno et al., 2019).
Demeaned earnings are defined as \( \logA possible consideration is that the empirical specification above does not allow for enough flexibility in the functional form, restricting the heterogeneous effect to be linear in the wage, while it could have a different, possibly more complex, relationship. As a robustness check, we interact the change in the FFR with a dummy for people with a wage larger than the (year-specific) average. This alternative specification does not alter our conclusions. Results are available upon request.
is a triennial cross-sectional survey of US families conducted by the Federal Reserve Board. It includes information on families’ balance sheets, pensions, income, and demographic characteristics. The 2016 SCF covers more than 6500 households.
Liabilities
Real Estate Prim = Real Estate Primary Residence
Secured by Primary Residence: Mortgages Secured by Primary Residence
Real Estate Sec. = Real Estate Secondary Residence
Stocks = directly held stock + stock mutual funds + other managed asset (50%) + Business
Secured by Secondary Residence: Mortgages Secured by Secondary Residence
Bonds = directly held bonds + bond mutual funds + savings bonds + other managed assets (50%)
Loans: Education loans + Vehicles loans + Other installments loans + Other loans
Liquid Assets = transaction accounts + CDs + Quasi-liquid retirement accounts (79%)
The Household Finance and Consumption Survey (HFCS) is a survey coordinated by the European Central Bank. It collects information on assets, liabilities, income, and consumption of households. The survey is based on 84,000 interviews in 18-euro area countries (Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Malta, Netherlands, Portugal, Slovak Republic, Slovenia, and Spain) as well as Poland and Hungary.
Liabilities
Real Estate Prim = Real Estate Primary Residence
Secured by Primary Residence: Mortgages Secured by Primary Residence
Real Estate Sec. = Real Estate Secondary Residence
Stocks = shares publicly traded + self-employment business + mutual funds (80%) + managed accounts (50%)
Secured by Secondary Residence: Mortgages Secured by Secondary Residence
Bonds = bonds + mutual funds (20%) + managed accounts (15%)
Loans: installment loans + credit line + credit card debt
Liquid Assets = deposits + money owed to households + voluntary/whole life insurance
To illustrate the decomposition of assets, we classify them into the following categories: real estate primary, real estate secondary, bonds, stocks, liquid, and others. Similarly, we classify liabilities into three categories: mortgages secured by primary residence, mortgages secured by other residence, and loans.
The SCF and the HFCS do not include the same variables. Hence, several simplifying assumptions are made in order to break down both assets and liabilities in the above broad categories. For example, for the US, we split the variable “other managed assets” equally into bonds and assets. Then we assume that 79% of the variable “quasi-liquid retirement accounts” is liquid assets. For the EU, we assume that 80% of the mutual funds are hold in stocks and the remaining 20% in bonds. Finally, for the variable “managed account,” 50% was allocated to assets while 15% to bonds.
Then we rank households according to their position in the net wealth distribution and we divide them into three groups: households in the bottom 20% of the distribution (the poor), households in the middle 70% of the distribution (the middle class), and households in the top 10% of the distribution (the rich).
Microsimulations are based on a welfare metric, “Unhedged Interest Rate Exposures” (UREs), capturing households’ exposures to changes in real interest rates, which has been proposed by Auclert (2019) and applied for the United States and Italy. Tzamourani (2021) follow the same approach to derive the URE metric for euro area households. Their findings suggests that a fall in interest rate can affect income distribution, redistributing away from households with large positive UREs (e.g., with large share of short-term fixed-income investments such as certificates of deposits and/or fixed-rate mortgages) towards households with large negative UREs (e.g., with large share of long-term bond investments and/or adjustable-rate mortgage liabilities). Auclert (2019) defines URE as:
where Y-T is net annual disposable income, C is a consumption measure that includes durable and non-durable goods as well as interest and principal payments. A and L are the remaining assets and liabilities maturing in a year, respectively.
Data. We use micro data surveys of consumer finances for the US and the EU, namely, the 2016 Survey of Consumer Finances (SCF) for the US and the 2016 Household Finance and Consumption Survey (HFCS) for the EU (see Appendix 1 for details on coverage and data processing).
Caveats. The simulations are based on several simplifying assumptions related to the type and maturity of various assets and liabilities, as in Auclert (2019). For the US, simulations also use the 2016 Survey of Consumer Expenditures to benchmark household consumption in the 2016 SCF. For the EU, simulations use broadly similar assumptions, except that household consumption data are directly observable in the HFCS while household disposable income is proxied by gross income due to data availability. Simulations also do not include the effect of a prolonged monetary policy easing on asset prices resulting from unrealized capital gains from higher assets prices, which typically benefit the wealthy but might not be immediately reflected in households’ income. Further, the simulations do not account for the effect of higher real estate prices on debt levels and home affordability, particularly at the bottom of the distribution. Also, given that our results are based on the 2016 survey data (latest available), potential changes in household income and finances since then can potentially alter the results. Finally, the results are subject to measurement error, a common caveat with survey data.
Given these caveats, the results are for illustration only, as different assumptions may lead to different results. Nevertheless, our broad findings are in line with Auclert’s results for the US and Tzamourani’s results for the EU.
Microsimulations are based on the 2016 Survey of Consumer Finances (SCF) for the US, and the 2016 Household Finance and Consumption Survey (HFCS) for the EU. We first compute household net wealth, defined as the difference between household asset and liabilities, using portfolio information available from the 2016 SCF for the US and the 2016 HFCS for the EU, and then scale household holdings of bond, real estate, and equities by their net wealth position. Further, we compute household capital gains on each asset by multiplying the relevant asset-to-net wealth ratio with a hypothetical 10 percent price increase.
For the EU, housing wealth includes households’ real estate and holdings of mutual funds that predominately invest in real estate. Bond holdings are defined as the sum of the direct bond holdings, mutual funds predominantly investing in bonds and 79 percent of private pension holdings. Equity holdings are the sum of stocks and business wealth, and equity mutual funds, and 21 percent of private pension holdings. For the US, real estate includes primary residence, other residential property, and 16 percent of other assets. Bond holdings are defined as the sum of savings bonds, bonds, 20 percent of pooled investment fund, 79 percent of retirement accounts, and 15 percent of other assets. Equity holdings are the sum of stocks, business wealth, 80 percent of pooled investment funds, and 21 percent of other assets.
Capital gains are scaled by the net wealth in line with the broadly agreed, but not uncontestable, principle of scale invariance, which requires the inequality measure to be invariant to equi-proportional changes in initial net wealth. For example, assuming the net wealth of the original distribution to be 1 unit for one bracket and 20 units for another and multiplying both of these by a factor of 100, the resulting distribution of 100 and 2000 would imply that inequality did not change. However, the same result would also mean that the top bracket would receive more than 95 percent of total new wealth of 2079, i.e., the top bracket wealth would increase by 1980 units, while the bottom bracket—by 99 units).
Main features. The Heterogeneous Agent New Keynesian (HANK) model of Kaplan et al. (2018) (KMV) combines a rich cross-sectional household heterogeneity under incomplete markets with standard New Keynesian features of price rigidities, monopolistic competition, and a Taylor rule. In the model, agents can save in two assets: a low-return short-dated asset and a high-return long-dated asset subject to transaction costs.
The model is calibrated to the US to match key micro data, including (i) net wealth inequality, with about 60 percent of households either in debt or holding close to zero cash; (ii) inequality in asset holdings, with the wealthiest decile holding 88 and 75 percent of all long- and short-dated assets respectively; and (iii) inequality in labor earnings, with 32 percent of labor earnings accruing to the wealthiest decile.
These features allow the model to create a large endogenous aggregate marginal propensity to consume (MPC), making consumption much more responsive to monetary policy shocks through general equilibrium effects, or what KMV call “indirect effects.” A monetary policy loosening incentivizes an immediate increase in consumption, particularly for agents facing a borrowing constraint. Increased demand for goods leads to larger labor demand and wages, which reinforce the effects on consumption.
Caveats. In this model, the portfolio channel is very strong and typically swamps the direct and other channels through which monetary policy affects income inequality. That is because wages are assumed to be flexible but not prices, and thus profits contract after a monetary expansion which lowers the price of equity and thus raises the subsequent returns to equity. If wages were also rigid, households would not see a significant rise in their labor income following a monetary policy shock, which would suppress the income and consumption of the poorest the most. At the same time, profits could become procyclical under sufficient wage rigidity, and thus asset returns would not rise as much as under flexible wages. Thus, moving to wage rigidity would support the wealth of the rich but suppress their income, while their consumption would be roughly unaffected due to stable permanent income. The KMV model puts some channels together but like any model it does not capture all relevant channels. Gornemann et al. (2016) build a HANK model featuring the crucial employed-unemployed margin, not modeled in KMV. They find that looser monetary policy reduces not only consumption but also income inequality, the latter being the opposite of the response in KMV. Note that Gornemann et al. (2016) abstract from the liquid/illiquid assets dichotomy that KMV argue is crucial to understand monetary policy transmission.
The empirical analysis of monetary policy easing on unemployment is based on Albrizio et al. (2021), who use the local projection method of Jordà (2005), augmented with the smooth transition regression approach of Granger and Teräsvirta (1993) to allow state-dependent responses. The analysis uses the US macro and household data at quarterly frequency over 1980–2016, which allows looking at monetary policy actions in the pre- and post-2008 periods to account for potential changes in distributional effects.
To quantify the effects of monetary policy shocks on wages, we estimate the following equation for labor market outcomes:
where w it is demeaned earnings for individual i in quarter t and FFR t is the Federal Fund Rate. Footnote 25 The vector X it consists of the following controls: a second-order polynomial in age, state fixed effects, occupation fixed effects, industry fixed effects, education fixed effects (10 categories), seasonal fixed effects (12 categories), gender, race, individual relationship to the household’s head (4 categories). The parameter γ q are our parameters of interest. If large and significant, it suggests a heterogeneous impact of monetary policy shocks on labor market outcomes.
Monetary policy shocks (i.e., changes in the FFR) are identified by exploiting high-frequency variations in interest rate futures within a narrow time window around Federal Open Market Committee (FOMC) announcements, following Gertler and Karadi’s (2015) framework. The response of futures to FOMC announcements is a proxy of the investors’ surprise around monetary policy announcements. If the announcement was in line with investors’ expectations, then future prices would not shift around the announcement. By contrast, surprisingly tighter or surprisingly looser monetary policy will result in shifts in future prices proportional to the size of the surprise. In the first stage, we regress changes in the FFR on surprises in futures in response to the FOMC announcements.
We use individual-level data on earnings and wages from the Current Population Survey (CPS). The CPS data is a well-established source of data for labor market indicators. However, it is important to note that data are top-coded. Therefore, we are not able to observe the dynamics at the upper end of the distribution.
Table 2 Second stage estimates of changes in earnings in response to changes in the FFRThe results of second stage regressions in response to changes in the FFR instrumented by surprises in futures are reported in Table 2. Both columns include the full set of control variables. Footnote 26 Column (1) reports the dynamic effect of an exogenous quarterly change in the FFR on the year-on-year (yoy) change in wages. In particular, a 100 bps increase in the FFR decreases yoy average wages by 0.6 percent in the first quarter after the monetary policy shock, compared to 1.1 percent in the fourth quarter. Column (2) reports the same regression introducing interaction terms for the initial level of wages (1 year ago). The results confirm significant heterogeneity in labor market outcomes in response to a monetary policy shock. Given that individual-level wages in the controls are demeaned using quarter-specific wage averages, the point estimate for coefficients on the change in the FFR is unchanged by construction. Footnote 27
