Unconventional Monetary Policies and Inequality

Layer 1: Overview

This paper asks whether the Federal Reserve’s unconventional monetary policies (UMP) — specifically quantitative easing (QE) and forward guidance — exacerbated income and welfare inequality in the United States during the effective lower bound (ELB) episode following the Great Recession (2009–2015). The question is empirically and theoretically contested: QE raises profits and equity prices, benefiting wealthy households who hold most equity, while simultaneously reducing unemployment, which benefits poorer households who rely almost entirely on labor income. Resolving the net effect requires a unified framework that captures both channels simultaneously, with empirically realistic responses of profits, wages, and unemployment to monetary policy.

The paper builds a medium-scale Heterogeneous Agent New Keynesian (HANK) model that incorporates: (i) a two-asset structure (liquid deposits and illiquid equity) with portfolio adjustment costs; (ii) three working statuses — employed, unemployed, and business owner — with endogenous job-finding rates determined by a search-and-matching labor market; (iii) a banking sector modeled after Gertler and Karadi (2011), with a moral-hazard leverage constraint; (iv) a substantial fixed cost in production that, combined with wage rigidity, generates procyclical profit responses to monetary policy shocks — a feature absent from standard New Keynesian models and critical for capturing benefits to wealthy households; and (v) an occasionally binding ELB constraint with QE modeled as central bank asset purchases and forward guidance modeled as exogenous expected ELB durations following Jones (2017). The model is calibrated to match the 2007 Survey of Consumer Finances (SCF), targeting the top decile’s share of wealth (~70%), income composition across wealth groups, and standard labor market and financial sector moments. Remaining parameters are estimated using Bayesian methods on U.S. quarterly data from 1992 Q1 to 2018 Q4, using ten observables (output, consumption, investment, inflation, nominal interest rate, real wage, unemployment, lump-sum transfers, profits, and Federal Reserve assets), with the ELB regime handled via an inversion filter and the Kulish-Jones method for exogenous ELB durations.

At the posterior mode, the model attributes the Great Recession primarily to a series of large negative risk premium shocks around 2008–2009, causing investment to fall by more than 20% relative to the pre-crisis level. The central counterfactual compares the actual ELB episode (with UMP) against a scenario where the central bank held its balance sheet constant and allowed ELB durations to be determined endogenously by fundamentals. Between 2009 and 2015, UMP on average produced: a 3.3% increase in profits, a 0.9% increase in equity prices, a 1.5 percentage-point reduction in the unemployment rate, and only a 0.1% increase in real wages (reflecting high estimated wage rigidity). Output and investment were higher by approximately 1% and 3% respectively on average, with profits rising as much as 8% during the ELB episode.

These aggregate effects translated into non-linear distributional outcomes. For the Gini index, lower unemployment reduced the income Gini by up to 0.6 percentage points, but this was offset by about 80% by the increase in profits and equity prices — leaving only a marginal net Gini reduction of 0.04 percentage points on average. When computed for the bottom 90% alone, the Gini reduction was more pronounced because that group relies overwhelmingly on labor income. However, the income share of the top 10% rose by an average of 0.17 percentage points, driven mainly by higher profits and equity prices. Thus the answer to whether UMP raised inequality is measure-dependent: UMP reduced within-bottom-90% inequality while widening the top-decile income gap.

Welfare gains (consumption equivalents over the ELB episode) were U-shaped across the wealth distribution: the average gain was 0.27% of lifetime consumption, but households at both extremes gained more than the middle. The bottom 10% benefited from higher job-finding rates (gaining ~0.3%), the top 10% from profits and equity prices (also ~0.3%), and the top 1% gained ~0.33%. The middle 60% gained only ~0.26%. By working status, business owners gained the most (0.82%), followed by the unemployed (0.35%) and the employed (0.27%).

Decomposing UMP into QE and forward guidance, the paper finds that forward guidance accounted for approximately 55% of total UMP stimulus. Forward guidance amplified both the aggregate and distributional effects of asset purchases: QE alone raised the top 10% income share by about 0.1 percentage point, and forward guidance added a further 0.09 percentage point increase. Forward guidance lowered the overall Gini by about 0.05 percentage points more than QE alone around 2013, and reduced the bottom-90% Gini by an additional 0.2 percentage points during the same period. The interaction intensified what the paper calls a “hollowing out” of the middle class: forward guidance further reduced middle-60% income shares while leaving bottom-10% shares nearly unchanged, because the additional stimulus disproportionately raised profits and equity prices (by about 2% and 1%, respectively, between 2011 and 2014).

Comparing QE with a hypothetical conventional monetary policy (CMP) that would have allowed the nominal rate to drop to approximately -1%, the paper finds that CMP would have produced larger aggregate stimulus than QE but more adverse distributional effects. Under CMP, lower financing costs disproportionately boosted bank net worth, indirectly raising profits and benefiting wealthy households even more than QE did. Under QE, central bank asset purchases crowded out private bank investment by reducing expected equity returns even as they raised equity prices, partially dampening the profitability gains to the financial sector. Consequently, CMP would have delivered above-average welfare gains only to the bottom 1% (debtors benefiting from lower real rates) and the top 10% (through larger bank profit effects), while the broad middle class would have fared no better and in some dimensions worse.

The paper’s key methodological contribution is the first Bayesian estimation of a HANK model with an occasionally binding ELB constraint. Its key substantive finding is that standard NK models, which generate countercyclical profits, systematically understate the benefits that expansionary monetary policy delivers to wealthy households, producing a misleading or incomplete picture of the distributional effects of monetary policy.

Layer 2: Deep Dive

What is the model’s identification strategy and how is the ELB period handled in estimation?

The model is estimated with Bayesian methods using an inversion filter (following Guerrieri and Iacoviello 2017 and Cuba-Borda et al. 2019) on ten quarterly observables from 1992 Q1 to 2018 Q4. The key identification challenge is the occasionally binding ELB constraint. The paper follows Kulish et al. (2014) and Jones (2017), treating the ELB as a temporary alternative regime with exogenous expected durations. These expected durations are themselves estimated as latent variables, with priors informed by the New York Fed’s primary dealer survey. The Metropolis-Hastings algorithm is used for structural parameters (treating ELB durations as fixed in each draw), while ELB durations are drawn separately using a discrete uniform proposal density. To make estimation computationally feasible given the large idiosyncratic state space, the paper follows Bayer and Luetticke (2020) and updates only the subset of the model Jacobian corresponding to ‘aggregate’ and ‘summary’ equations during each iteration, leaving the ‘idiosyncratic’ blocks fixed across estimated parameters.

What are the main mechanisms by which UMP affects inequality and how does the model distinguish them empirically?

The paper identifies four main channels: (1) Profit and equity price channel — QE raises equity prices and reduces financing costs, increasing profits and the dividend rate on illiquid assets. Because the top decile holds ~70% of total wealth overwhelmingly in the form of equity, with capital and business income accounting for ~50% of their income, this channel benefits the wealthy disproportionately. (2) Unemployment channel — lower interest rates stimulate demand and raise the job-finding rate. Because households at the bottom of the wealth distribution are more likely to be unemployed at the onset of the ELB episode (8.75% of the bottom decile vs. 6.54% in the middle quintile in 2009 Q1), this channel is progressive. (3) Wage channel — nominal and real wage rigidity (only one-fifth of the real wage adjusts to labor productivity changes) means that the wage channel is very weak; average real wages rose by only 0.1% due to UMP. (4) Inflation/redistribution channel — forward guidance generates inflationary expectations that compress real rates, redistributing from savers to debtors. The empirical decomposition is performed by first isolating QE alone (endogenizing ELB durations) and then comparing to the full UMP scenario (exogenous ELB durations), attributing the residual effect to forward guidance.

What is the key modeling innovation regarding profits, and why does it matter for inequality?

Standard New Keynesian models generate countercyclical profit responses to monetary policy shocks: when demand rises, price rigidity keeps prices sticky while factor prices (wages) adjust upward, squeezing markups and reducing profits. This contradicts empirical evidence from structural VARs, which show procyclical profits. The paper introduces three interacting features that resolve this: (a) a substantial fixed cost of production calibrated to roughly 20% of steady-state output, so that average production cost falls even as marginal cost rises, boosting net profits; (b) wage rigidity with search-and-matching frictions, so that real wages respond very weakly to monetary shocks; and (c) a banking sector with a financial accelerator, so that rising equity prices boost banks’ net worth and their investment demand, further amplifying profits. Without procyclical profits, the model would understate the benefits wealthy households (whose income depends heavily on profits and equity returns) gain from expansionary monetary policy, producing an incomplete picture of distributional effects.

What heterogeneity in households’ balance sheets and income composition is documented, and how does it shape distributional results?

Using the 2007 SCF, the paper documents stark composition differences. The bottom 80% of the wealth distribution derives ~80% of income from labor, with transfer income making up most of the rest. The top 10% derives about 50% from labor and 50% from capital (equity and business income). For the top 0.1%, labor income is only 16% and capital/business income is about 83–85%. In the model, the top 10% hold about 70% of total wealth, overwhelmingly in illiquid equity. These composition differences mean that any policy raising profits and equity prices is strongly progressive at the top and neutral-to-mild at the bottom, while any policy reducing unemployment is strongly progressive at the bottom. The interplay of these two forces explains why UMP simultaneously reduces bottom-90% inequality (through the unemployment channel) and widens the top-vs.-rest gap (through the profit and equity channel), and why welfare gains are U-shaped rather than monotone.

What is the welfare accounting methodology and what are the key welfare findings?

Welfare gains are measured as consumption equivalents — the fraction of lifetime consumption that a household in the counterfactual (no UMP) scenario would be willing to forgo to enjoy the UMP outcome. Households are sorted into wealth groups based on their 2009 Q1 wealth position (so group composition is not affected by UMP), and the same households are followed throughout the episode. Beyond the sample end (2018 Q4), no further shocks are assumed. The average welfare gain at the posterior mode is 0.27% of lifetime consumption. Bottom 10%: ~0.3% (driven by higher job-finding rates). Top 10%: ~0.3% (driven by profits and equity gains). Top 1%: ~0.33%. Middle 60%: ~0.26%. Business owners: 0.82%. The unemployed: 0.35%. The employed: 0.27%. Critically, the welfare gaps between extremes and middle are smaller than the income gaps, because anticipated tapering after the sample implies lower future profits and equity prices for wealthy households, narrowing their long-term advantage.

How do the contributions of QE and forward guidance compare in aggregate and distributional terms?

Forward guidance accounted for approximately 55% of the total UMP stimulus at the posterior mode. Exogenous expected ELB durations exceeded endogenous (fundamentals-based) durations by 1–2 quarters on average, and sometimes by up to 8 quarters, with the divergence widening from 2011 onward. In distributional terms, QE alone initially reduced the bottom-90% Gini and raised the top 10% income share by about 0.1 percentage point. Forward guidance amplified both effects: it lowered the overall Gini by an additional ~0.05 pp and the bottom-90% Gini by an additional 0.2 pp around 2013, but also added a further ~0.09 pp to the top 10% income share between 2011 and 2014. The amplification occurred because forward guidance raised profits and equity prices by about 2% and 1% respectively during that window, intensifying the income concentration at the top while also stimulating job creation at the bottom. The middle class saw its income share further compressed.

How does QE compare with conventional monetary policy in terms of aggregate and distributional effects?

In the counterfactual CMP scenario, the nominal policy rate drops to approximately -1% and remains negative for an extended period. CMP produces larger aggregate stimulus than QE: the stimulus effects of QE were partly crowded out by general equilibrium effects, specifically QE reduced banks’ expected return on equity even as it raised equity prices, discouraging private bank investment. Under CMP, lower nominal rates instead benefit banks through lower financing costs, boosting bank net worth via an accelerator mechanism more strongly than under QE. This difference has distributional consequences: CMP would have delivered higher welfare gains only to the bottom 1% (low-wealth debtors benefiting from lower real rates on their liabilities) and the top 10% (benefiting from larger bank profits). Households in the broad middle — already employed, holding limited equity, neither heavy borrowers nor large business income recipients — would have been no better off and in some dimensions worse off under CMP. The paper thus concludes that QE had less adverse distributional effects than CMP would have had, absent the ELB constraint.

What robustness checks and sensitivity analyses are conducted?

The paper checks results against: (a) the full 10th–90th percentile range of the posterior distribution for all key findings on aggregate effects, income inequality, welfare gains, and QE vs. CMP comparisons, showing that qualitative findings are robust to parameter uncertainty; (b) a comparison between rigid-wage and flexible-wage model variants (Table A1), showing that the flexible-wage version generates countercyclical profits, a weak unemployment response, and a strong real wage response — inconsistent with empirical SVAR evidence — validating the modeling choice of high wage rigidity; (c) a structural VAR analysis on U.S. data confirming procyclical profits, weak real wage responses, and significant unemployment responses to monetary policy shocks; (d) a comparison of the OccBin method (endogenous ELB durations, Guerrieri and Iacoviello 2015) vs. the Kulish-Jones method (exogenous durations) for solving the occasionally binding constraint; (e) a check that wages implied by the calibrated wage function always remain in the bargaining set, validating the equilibrium wage assumption.

What are the key differences between this paper and the closest prior work?

Kaplan, Moll, and Violante (2018) and Bayer et al. (2020) have two-asset HANK models but omit frictional labor markets, so they cannot capture how monetary policy affects employment and thus the progressive unemployment channel. Gornemann et al. (2016) include search-and-matching labor markets but only one asset, so they cannot capture the capital income benefits to wealthy households. Broer et al. (2019) and Auclert et al. (2023) identify the countercyclical profit problem but their solutions (wage rigidity alone) produce procyclical profits that are too weak quantitatively. This paper combines fixed costs, wage rigidity, and a banking sector to produce procyclical profits quantitatively consistent with SVAR evidence. On unconventional policy specifically, Lenza and Slacalek (2018) and Casiraghi et al. (2018) study ECB QE with partial equilibrium methods and find inequality-reducing effects; Bivens (2015) and Montecino and Epstein (2015) reach opposite conclusions for U.S. QE. This paper is the first to study both QE and forward guidance jointly in a Bayesian-estimated HANK model with an explicitly binding ELB, and is to the author’s knowledge the first to estimate a HANK model with an occasionally binding ELB constraint.

What are the main policy implications and their scope conditions?

First, UMP’s inequality effects are measure-dependent: policies that simultaneously stimulate employment and profits can reduce within-bottom-90% inequality while widening the top-vs.-rest gap. Policymakers who cite Gini reductions and those who cite rising top-income shares are both correct, pointing to different parts of the distribution. Second, forward guidance amplifies inequality effects as much as it amplifies aggregate effects, so its use carries a distributional cost concentrated at the top of the distribution. Third, QE had less adverse distributional effects than conventional monetary policy would have had, suggesting that concerns about QE’s inequality effects should be placed in context of the ELB constraint — the relevant comparison is not QE vs. no policy but QE vs. CMP with the ELB absent. Fourth, models that generate countercyclical profits will systematically understate benefits to the wealthy and potentially reach qualitatively different conclusions about whether monetary policy raises or reduces inequality. These findings are scoped to the U.S. Great Recession ELB episode, estimated with the specific HANK model structure and Bayesian posterior; findings may differ for different financial structures, more generous unemployment insurance, or different asset price dynamics.

What drives the Great Recession in the model and how is UMP modeled mechanically?

At the posterior mode, the Great Recession is primarily attributed to a series of large negative risk premium shocks (shocks to banks’ discount factor) around 2008–2009, which caused banks to sharply contract their investment, leading to the investment collapse (>20% below pre-crisis). QE is modeled following Gertler and Karadi (2011): the central bank issues bonds (sold to the private sector) and uses proceeds to purchase equity directly, converting non-productive asset demand into productive capital demand and raising equity prices and investment. Forward guidance is modeled as setting exogenous expected ELB durations longer than would be implied endogenously by the Taylor rule fundamentals, effectively mimicking future negative interest rate shocks and inducing inflationary pressure via intertemporal substitution. The expected ELB durations at the posterior mode range from 6 to 8 quarters through 2013, falling sharply to 1–2 quarters by late 2014–2015.

Key Concepts

Heterogeneous Agent New Keynesian (HANK) model: As used in this paper, a DSGE model where households differ ex-post in idiosyncratic productivity, asset holdings (liquid deposits and illiquid equity), and employment status; combined with search-and-matching labor markets, a banking sector with leverage constraints, and a zero lower bound on the policy rate. The heterogeneity in wealth composition and income sources determines how aggregate policy shocks translate into distributional outcomes.

Procyclical profits: The property, established empirically via SVAR and reproduced in the model, that firm profits rise in response to expansionary monetary policy shocks. Standard New Keynesian models generate the opposite (countercyclical profits) because price rigidity compresses markups when demand rises. In this paper, the combination of large fixed costs in production, wage rigidity, and a banking sector financial accelerator is required to generate quantitatively realistic procyclical profit responses.

Effective lower bound (ELB) episode: The period from 2009 Q1 to 2015 Q4 during which the Federal Reserve’s policy rate was constrained at zero. In the model, this is treated as a temporary alternative regime with exogenous expected durations; when the policy rate hits the ELB, the central bank can only affect the economy through asset purchases (QE) and forward guidance.

Forward guidance (as exogenous expected ELB durations): In this paper’s framework, forward guidance is operationalized as the central bank committing to maintain the policy rate at zero for a longer period than the endogenous (fundamentals-based) Taylor rule would prescribe. This is parameterized as an exogenous expected ELB duration that exceeds the endogenous one, creating anticipations of future negative interest rate shocks and thus stimulating activity through intertemporal substitution.

Consumption equivalent welfare gain: The fraction of lifetime consumption that a household in the counterfactual scenario (no UMP) would be willing to forgo in order to instead experience the outcomes under UMP. Used to compare welfare across heterogeneous households in a cardinal, utility-based metric rather than income alone.

Business owner working status: A third working status (alongside employed and unemployed), following Bayer et al. (2019), in which households receive a fixed fraction of aggregate profits as income without supplying labor. Business owners transition into and out of this status exogenously and are the highest-income group in the model, calibrated to match the top-decile’s share of liquid assets and the income composition data showing that capital and business income dominate the very top of the wealth distribution.

Inversion filter: The likelihood evaluation method used in this paper for Bayesian estimation, following Guerrieri and Iacoviello (2017). Rather than running a Kalman filter, structural shocks are backed out directly by inverting the linear solution of the model given the observed data and a given set of expected ELB durations. This avoids continuously updating the large state-transition matrix and makes estimation computationally feasible.