Labor Market Discrimination and the Racial Unemployment Gap: Can Monetary Policy Make a Difference?

Layer 1: Overview

This paper addresses two connected questions: why do Black workers face persistently higher and more volatile unemployment than white workers, and can the Federal Reserve’s August 2020 shift from a symmetric “Deviations” rule to a “Shortfalls” rule narrow the resulting racial unemployment gap? The authors build a New Keynesian search and matching model with endogenous separations (Mortensen-Pissarides) and add employer taste-based discrimination, calibrated to U.S. Current Population Survey microdata from January 1976 to December 2019.

The empirical motivation is stark. In CPS data, the Black unemployment rate averages 12.0 percent against 5.5 percent for whites — a gap of 6.5 percentage points that is largely unexplained by observable characteristics such as age, education, marital status, and state of residence (Cajner et al. 2017). The racial gap is also strongly countercyclical: its cyclical correlation with the aggregate unemployment rate is 0.77. A Shimer (2012)-style flow decomposition shows that the separation rate margin accounts for approximately two-thirds (67 percent) of the mean gap and 60 percent of its cyclical variance, with the job-finding rate contributing 20 percent of the mean and 27 percent of variance.

The model features two types of representative households that differ only in a non-productive attribute (race). Firms incur a per-period perceived cost κ₁ of employing a type-1 (Black) worker, following Becker (1971). This cost is time-invariant and not directly affected by monetary policy. Search is random (firms cannot direct search by race, consistent with anti-discrimination law). The model also incorporates Calvo price rigidities and an effective lower bound (ELB) on the nominal interest rate, solved via Dynare’s extended path method. Two aggregate shocks drive dynamics: a risk-premium (demand) shock and a productivity (supply) shock. The discriminatory parameter is calibrated to κ₁ = 0.0292 — equivalent to 3.6 percent of the steady-state average wage — to match the 6.4 percentage-point mean racial unemployment gap.

The baseline model (under the symmetric Deviations rule) generates four untargeted results that match the data: (1) higher mean separation rates and lower mean job-finding rates for Black workers, with the ratio of Black-to-white separation rates at 2.3 in the model (1.9 in data); (2) higher cyclical volatility of Black unemployment, driven by higher separation-rate volatility; (3) a strongly countercyclical racial gap (near-unit correlation with aggregate unemployment in the model); and (4) positively skewed unemployment distributions for both groups — skewness that arises endogenously from the ELB constraint, which is absent when the ELB is removed. The mechanism is geometric: because Black workers face a higher reservation productivity threshold (due to κ₁ > 0), more Black workers cluster near that threshold. A given aggregate shock therefore moves a larger mass of Black workers across the threshold, amplifying their unemployment response relative to whites.

Novel model-based discrimination measures — workers not hired or fired solely due to being Black — average 5.86 percent of the Black labor force under the Deviations rule and are strongly countercyclical (correlation with aggregate unemployment = 0.99 in the model vs. 0.64 in EEOC race-charge data). The welfare gap between white and Black households averages 2.4 percent in consumption-equivalent terms.

Shifting to the Shortfalls rule — which responds to unemployment shortfalls symmetrically but only tightens policy when unemployment is above its steady-state level — strengthens expansions by keeping interest rates lower. The aggregate unemployment rate falls by 0.7 percentage point, from 6.37 percent to 5.65 percent. Because Black workers are more cyclically sensitive, they benefit disproportionately: Black unemployment falls by 1.1 percentage points and white unemployment falls by 0.7 percentage points, narrowing the racial gap by 0.5 percentage point (from 6.50 to 6.03 percent). Model-based discrimination also declines (aggregate measure from 5.86 to 5.52 percent). The downside is a 0.5 percentage-point rise in average inflation, from 1.9 percent to 2.4 percent. The negative skewness in the racial unemployment rate gap is essentially eliminated under the Shortfalls rule, so the distribution shifts toward a lower mean with fewer episodes of extreme gaps.

From a welfare perspective, however, the gains are quantitatively trivial. Both households experience slightly positive welfare gains under the Shortfalls rule — consumption rises by 0.62 percent for Black households and 0.64 percent for white households — but the differences are effectively indistinct from zero in consumption-equivalent terms. Crucially, the consumption-equivalent welfare wedge between the two groups actually widens slightly, because white wages rise more than Black wages under the Shortfalls rule (average productivity of Black employed workers falls more as the lower reservation threshold admits marginal workers). The authors note their welfare analysis is a lower bound, given within-group consumption insurance, the absence of liquidity constraints, and non-expiring unemployment benefits in the model.

Layer 2: Deep Dive

What is the identification strategy and what are the main threats to it?

The paper uses a structural calibration approach rather than quasi-experimental identification. The model is calibrated to match 10 aggregate moments (1976-2019 CPS data) with all parameters common across racial groups except κ₁. The racial unemployment gap in steady state is the sole targeted moment for racial differences; all other racial outcomes are untargeted predictions. Threats include: (1) the model attributes all cross-race labor market differences to discrimination, ruling out unobserved productivity heterogeneity; (2) the representative firm with taste-based discrimination abstracts from market-selection forces that, in Becker’s classic model, would erode discrimination in the long run (the authors cite Black 1995, Rosen 1997, Sasaki 1998 for equilibrium justifications); (3) the model is solved under perfect foresight (extended path), not fully stochastic, though Dynare’s method approximates stochastic dynamics; (4) the Shortfalls rule is a reduced-form approximation of the FOMC’s 2020 framework, not a structural representation.

What are the main mechanisms through which discrimination generates the observed racial unemployment patterns?

The core mechanism is that κ₁ > 0 raises the reservation productivity threshold for Black workers at both hiring (firms require higher expected productivity to justify the cost) and separations (existing matches must clear a higher bar to survive). Because idiosyncratic productivity is log-normally distributed, more Black workers cluster near their higher reservation threshold than white workers do near the lower white threshold. This concentration in the density means that any aggregate shock — moving both thresholds — shifts a proportionally larger mass of Black workers across the destruction margin, amplifying the volatility of Black unemployment and separations. The countercyclical racial gap arises because aggregate downturns raise both reservation thresholds, but since more Black workers are near their threshold, more are destroyed. The authors show that the separation-rate margin dominates: in the model it explains 92 percent of the mean gap and 81 percent of its cyclical variance, somewhat overstating the empirical 67 percent and 60 percent, because variation in the job-finding rate comes mostly from the common job-meeting probability.

How do the two types of discrimination in the model — hiring discrimination and separation discrimination — work quantitatively?

The hiring discrimination measure Df_t counts the fraction of Black job-seekers who are not hired because their idiosyncratic productivity draw falls above the white reservation threshold but below the (higher) Black threshold. The separation discrimination measure Dλ_t counts the fraction of employed Black workers who are endogenously separated for the same reason. Under the Deviations rule with ELB, the hiring margin averages 0.64 percent and the separation margin averages 5.22 percent of the Black labor force, for a total Dt of 5.86 percent. Both measures are strongly countercyclical (correlations with aggregate unemployment of 0.80 and 0.95 respectively). Under the Shortfalls rule, these fall to 0.56 and 4.95 percent (total 5.52 percent), and their skewness toward high discrimination levels is significantly reduced.

What are the aggregate macroeconomic effects of switching from the Deviations rule to the Shortfalls rule?

The Shortfalls rule keeps nominal interest rates lower during periods of below-target unemployment (its asymmetry means it does not tighten in expansions unless inflation rises). This raises average output and consumption. The aggregate unemployment rate falls by 0.7 percentage point (from 6.37 to 5.65 percent), driven by both a lower average separation rate (3.36 to 3.10 percent) and a higher average job-finding rate (50.14 to 56.99 percent). Average inflation rises by 0.5 percentage point (from 1.88 to 2.40 percent annually). The Shortfalls rule increases the volatility of all labor market variables (it has lower stabilization properties) but essentially eliminates the positive skewness in the aggregate unemployment rate. The probability of a binding ELB falls from 10.6 percent to 8.5 percent under the Shortfalls rule. The correlation between inflation and unemployment strengthens from -0.32 to -0.51.

How does the Shortfalls rule differentially affect Black and white workers?

Black workers benefit disproportionately because their unemployment is more cyclically sensitive. The unemployment rate falls by 1.1 percentage points for Black workers (from 11.89 to 10.78 percent) versus 0.7 percentage points for white workers (from 5.39 to 4.74 percent). The racial gap narrows by 0.5 percentage point (from 6.50 to 6.03 percent). Separation rates fall more for Black workers (6.53 to 6.29 vs. 2.90 to 2.65 for whites). Average wages for Black workers increase by 0.43 percent and for white workers by 0.48 percent. The slight relative wage disadvantage under the Shortfalls rule arises because the lower reservation threshold for Black workers admits workers with lower average productivity, pulling down average Black wages relative to whites.

What are the welfare implications of the policy change, and why are they small?

Both households gain welfare under the Shortfalls rule, but the gains are quantitatively very small in consumption-equivalent terms (effectively indistinct from zero). The aggregate benefit — lower average unemployment — is partially offset by the cost of higher average inflation (price dispersion loss in the Calvo framework). Consumption rises by about 0.62 percent for Black households and 0.64 percent for white households. The consumption-equivalent welfare wedge between Black and white households (2.4 percent under the Deviations rule) actually widens slightly under the Shortfalls rule, because white wages increase more than Black wages. The authors emphasize several reasons their welfare analysis understates true racial inequality: (1) within-group consumption insurance prevents individual unemployment spells from being welfare-costly; (2) no liquidity constraints; (3) unemployment benefits do not expire; (4) the model abstracts from labor force participation margins and involuntary part-time employment. These features, if relaxed, would likely reveal larger welfare differences between the two groups.

What role does the effective lower bound (ELB) on nominal interest rates play?

The ELB is essential to generating positively skewed unemployment distributions in the model. Without the ELB, the model produces essentially symmetric (near-zero skewness) distributions for both aggregate and racial unemployment outcomes. With the ELB, the baseline model matches the observed positive skewness of the unemployment rate (1.25 aggregate; 1.23 for Black workers, 1.26 for whites). The ELB also raises the mean unemployment rate by about 0.25 percentage point and slightly amplifies labor market volatilities. It introduces a deflationary bias (inflation averages 1.88 percent vs. the 2.0 percent steady-state target). Critically, the main results — the 0.5 pp narrowing of the racial gap and 0.7 pp fall in aggregate unemployment under the Shortfalls rule — are robust to removing the ELB constraint (Appendix B.2.2), confirming they are not artifacts of the nonlinearity introduced by the ELB.

What robustness checks are conducted?

Key robustness exercises include: (1) removing the ELB constraint, which confirms the main results hold (aggregate unemployment falls 0.7 pp, racial gap narrows 0.5 pp, inflation rises 0.5 pp without the ELB; Table A.8-A.9); (2) extending the unemployment flow decomposition to a three-state system (employed, unemployed, out of labor force), which confirms that the employment-to-unemployment (EU) transition is the primary driver of the racial gap even accounting for labor force participation transitions (Appendix A.2); (3) verifying that employer-to-employer transition rates are similar across racial groups (2.20 percent for Blacks vs. 1.96 percent for whites, 2004-2019), supporting the assumption of equal exogenous separation rates; (4) confirming that inflation experiences are similar between Black and white households using the Chicago Fed IBEX data (2.80 percent for Blacks vs. 2.87 percent for whites, 1983-2013), supporting the equal-inflation assumption; (5) presenting impulse response functions under both a productivity shock and a demand shock, in models with and without monetary policy inertia.

How does this paper relate to and differ from closely related prior work?

The paper contributes to four literatures. First, versus Cajner et al. (2017) on empirical racial labor market gaps, it provides a structural explanation rather than documenting gaps. Second, versus search-and-matching discrimination models (Bartel 1995, Bowlus-Eckstein 2002, Rosen 2003, Flabbi 2010, Borowczyk-Martins et al. 2017), the key contributions are: (a) endogenous separations (prior models used exogenous exit), which the authors view as essential since separation rates dominate the gap’s dynamics; and (b) incorporating nominal rigidities and an ELB, enabling analysis of monetary policy. Third, versus Ravenna-Walsh (2012) and Bergman et al. (2022), who embed worker heterogeneity in New Keynesian search models, this paper differs by modelling heterogeneity as discrimination rather than productivity differences, and by studying the Deviations-to-Shortfalls rule change specifically. Fourth, versus Bundick-Petrosky-Nadeau (2021) who study the same Deviations/Shortfalls comparison for the aggregate economy, this paper adds the racial dimension. Versus Lee et al. (2022), Nakajima (2023), and Ait Lahcen et al. (2023) — all of which also study monetary policy and racial inequality — the contribution is generating racial disparities endogenously from discrimination rather than taking them as given, and including endogenous separations.

What does the paper find about the countercyclicality of racial discrimination?

Both the model and the data exhibit strongly countercyclical discrimination. In the data, EEOC race-based discrimination charges (normalized per non-white labor force member) have a contemporaneous correlation of 0.65 with the cyclical component of the aggregate unemployment rate from 1997 to 2019. In the model, the aggregate discrimination measure Dt has a correlation of 0.99 with aggregate unemployment. The countercyclical pattern arises mechanically from the higher density of Black workers near the reservation productivity threshold: during recessions, both thresholds rise, destroying proportionally more Black matches and blocking more Black hires. The model-based discrimination measure also shows positive skewness (1.13 aggregate skewness under the Deviations rule with ELB), consistent with the asymmetric incidence of recessions.

What are the quantitative scope conditions and limitations the authors themselves identify?

The authors identify several scope conditions and limitations: (1) the model abstracts from labor force participation, so it misses the racial gap in participation rates and involuntary part-time employment; (2) within-group consumption insurance and no liquidity constraints imply welfare estimates are a lower bound on true racial inequality — the consumption-equivalent wedge of 2.4 percent would be larger with incomplete insurance or borrowing constraints; (3) the welfare analysis assumes equal inflation rates across racial groups, which is empirically supported but abstracts from possible differences in consumption baskets; (4) the discriminatory parameter κ₁ is time-invariant and unresponsive to monetary policy, so all channels are indirect (through business cycle dynamics); (5) the model assumes a representative firm with taste-based discrimination, abstracting from firm heterogeneity in discrimination and from customer or statistical discrimination; (6) the Shortfalls rule is a reduced-form approximation of the FOMC’s 2020 framework and may not capture all aspects of the actual policy change.

Key Concepts

Shortfalls rule: A Taylor-type monetary policy rule that responds symmetrically to inflation deviations from target but responds to unemployment deviations from steady state only when unemployment is above its steady-state level — not when it is below. This captures, in reduced form, the FOMC’s August 2020 revision from ‘deviations’ to ‘shortfalls’ of employment from maximum.

Deviations rule: A symmetric Taylor-type interest rate rule that responds to deviations of both inflation and unemployment from their respective steady-state values, regardless of the direction of the unemployment deviation. The baseline monetary policy in the model before the 2020 FOMC framework change.

Taste-based discrimination (κ₁): A per-period perceived cost κ₁ borne by employers for each period they employ a Black worker, following Becker (1971). In this model, κ₁ = 0.0292 (≈3.6 percent of the steady-state wage), is time-invariant, and is not directly altered by monetary policy — only indirectly through business cycle conditions.

Reservation productivity threshold (zRi): The minimum idiosyncratic productivity level at which it is profitable for a firm to either hire or retain a worker of type i. Because of κ₁, the Black reservation threshold exceeds the white threshold, generating higher endogenous separation rates and lower job-finding rates for Black workers.

Model-based discrimination measures (Df_t, Dλ_t): Novel measures of the fraction of the Black labor force that is not hired (Df_t, hiring margin) or is fired (Dλ_t, separation margin) solely due to discrimination — i.e., workers whose idiosyncratic productivity exceeds the white reservation threshold but falls below the Black threshold. These are expressed as fractions of the Black labor force and compared to EEOC race-based charge data.

Consumption-equivalent welfare wedge (Ψ_t): The percentage increase in per-period consumption that must be given to Black households every period to equalize their welfare with that of white households, given the same stochastic future. Under the Deviations rule, this averages 2.4 percent. The change under the Shortfalls rule is effectively zero in quantitative terms.

Endogenous separation: A separation that occurs because a matched worker-firm pair draws an idiosyncratic productivity below the reservation threshold — as distinct from exogenous separations (random layoffs unrelated to productivity). The dominance of the separation margin in explaining the racial unemployment gap motivates the use of endogenous separations as a key model ingredient; prior search-and-discrimination models assumed exogenous exit.