Monopsony in the U.S. Labor Market

The paper asks whether the U.S. labor market is perfectly competitive and examines wedges between the marginal revenue product of labor (MRPL) and wages as direct evidence of employer monopsony power. Using plant-level data from U.S. manufacturing (1976–2014), the authors estimate “markdowns” (MRPL-to-wage ratios). They document that markdowns are substantially above one, indicating widespread monopsony power, and study how these wedges have evolved over time. Understanding employer market power is crucial for evaluating policies affecting wages and mobility (e.g., minimum wage, merger policy). The study fills a gap by providing direct, minimally parametric estimates of MRPL–wage wedges at the plant level and a theory-consistent aggregation to study long-run trends.

The study contributes to renewed interest in labor market monopsony and its evolution in the U.S., related to declines in labor’s share and the rise of “superstar firms.” Prior work has proxied employer power via labor market concentration (e.g., Azar et al.; Benmelech et al.; Rinz; Rossi-Hansberg et al.) or structural models (Posner et al.; Azar, Berry, and Marinescu; Jarosch et al.; Berger, Herkenhoff, and Mongey). The paper builds on the production function approach (Hall; De Loecker and Warzynski; Ackerberg, Caves, and Frazer) to estimate wedges with minimal structure while separating product markups from labor markdowns. It relates to evidence on rising product markups (De Loecker, Eeckhout, and Unger; Eggertsson et al.) and to studies on rent sharing and productivity-wage links (Abowd, Kramarz, and Margolis; Van Reenen; Card, Devicienti, and Maida). It also engages with critiques of production-approach identification and deflated-revenue biases (Bond et al.) and with literature on the locality of labor markets (Manning and Petrongolo; Marinescu and Rathelot).

Measure of monopsony: Define a plant’s markdown ν as MRPL divided by the wage; under profit maximization with a differentiable, upward-sloping firm-specific labor supply, ν = 1 + 1/ε_s (ε_s is the perceived labor supply elasticity). Production-approach identification: Using cost-minimization first-order conditions, the wedge between an input’s output elasticity and revenue share reflects market power. If there exists at least one flexible input (assumed to be materials) with no adjustment costs and no monopsony, then product markups μ are identified as the ratio of the materials elasticity to its revenue share. Labor markdowns are then identified by comparing labor’s elasticity-share wedge to the markup, enabling separation of μ and ν at the plant level. Production function estimation: Estimate output elasticities via proxy-variable methods (Olley-Pakes; Levinsohn-Petrin; De Loecker-Warzynski; Ackerberg-Caves-Frazer). The specification uses a translog gross-output production function in capital, labor, materials, and energy, allowing elasticities to vary with input levels. Estimation proceeds in three steps: (1) nonparametrically approximate output net of measurement error; (2) recover productivity as a function of observables and model its law of motion to obtain innovations; (3) use GMM-IV with lagged inputs (except capital) as instruments to estimate production parameters. Data: U.S. Census Bureau’s Census of Manufactures (quinquennial) and Annual Survey of Manufactures (annual rotating sample), 1976–2014, with standard deflators (NBER-CES Manufacturing Database). Aggregation: To study secular trends, propose a local-market (3-digit NAICS × county) aggregate markdown V_jlt consistent with aggregate wedges. Under flexible materials, aggregation implies sales-weighted harmonic means (with corrections for heterogeneity in elasticities), and an economy-wide aggregate markdown V_t is an employment-weighted average over local markets. Robustness: Test constant-returns-to-scale restrictions, labor adjustment costs (convex and nonconvex via biennial estimation), inclusion of benefits in compensation (post-2002), and labor heterogeneity (production vs. nonproduction workers). They address identification critiques (deflated revenues, materials as flexible input, scalar-unobservable) theoretically, via simulations, and via alternative specifications.

• U.S. manufacturing plants exhibit sizable monopsony power: the mean plant-level markdown is 1.53 (median 1.364), implying the average worker receives about 65 cents per marginal dollar produced. The implied average labor supply elasticity is about 1.88 (from ν = 1 + 1/ε_s).
• Dispersion: Markdown dispersion is large. Whole-sample interquartile range is 0.618; within-industry average interquartile range is about 61.6% (SD ≈ 60.4%).
• Determinants: Variance decomposition shows markdown variation primarily reflects variation in labor output elasticities and labor revenue shares and their covariance; product markups contribute little to markdown variance.
• Size gradient: Markdown increases monotonically with establishment size (employment share in local market). Plants in the top size bins have ≈20% higher markdowns than the smallest plants, controlling for age, industry, state, and year.
• Age: A positive age gradient appears without size controls but largely vanishes after controlling for size.
• Productivity: Relation between markdowns and TFPR is nonmonotonic (qualitatively U-shaped), with wide but imprecisely estimated variation across productivity percentiles.
• Organizational scope: Plants in multi-unit firms or firms spanning multiple industries or counties have ≈0.25 higher markdowns on average than stand-alone or single-scope firms (effects persist controlling for size). High-tech plants have weakly lower markdowns on average.
• Labor heterogeneity: Both production and nonproduction workers face markdowns above 1; no systematic dominance of one group. Results are consistent with most plant labor used in production.
• Aggregate trends: A novel, wedge-consistent aggregate markdown for local labor markets declines from the late 1970s to early 2000s (minimum around 2002) and then rises sharply thereafter (U-shaped pattern through 2012). This differs from a steadily declining local-concentration-based aggregation and only loosely matches simple employment-weighted averages.
• Concentration vs. markdowns: Cross-sectional correlations between local employment HHI and local markdowns are near zero (sometimes negative). Over time, local concentration declined in late 20th century; it does not capture the sharp post-2000 rise in aggregate markdowns.
• Robustness: Findings are robust to assuming constant returns to scale, to accounting for convex and nonconvex labor adjustment costs (estimated corrections small, ~0.03 on average), and to including benefits in compensation (markdowns modestly lower but still >1). Using energy as the flexible input produces noisier estimates and often lower markdowns (consistent with monopsony in energy markets), supporting the materials-as-flexible-input choice.

The findings directly document substantial and pervasive wedges between MRPL and wages in U.S. manufacturing, indicating widespread employer monopsony power rather than perfectly competitive labor markets. The positive size-markdown relationship suggests larger employers capture greater rents from labor, aligning with concerns about the distributional and efficiency implications of firm size in labor markets. The U-shaped trend in aggregate markdowns indicates that monopsony power decreased through the early 2000s but subsequently intensified, implying that rising monopsony is unlikely to explain earlier declines in labor’s share but may contribute to more recent wage stagnation and reduced dynamism. Weak correlations between concentration indices and markdowns caution against using concentration as a sufficient statistic for monopsony power in policy and antitrust contexts. The approach, which separates output markups from input markdowns and accounts for local labor market structure, strengthens the interpretation of wedges as employer market power and informs debates on minimum wages, merger review, and mobility frictions.

The paper provides direct, plant-level estimates of labor markdowns in U.S. manufacturing and a new aggregation consistent with aggregate wedges and local labor markets. It shows substantial monopsony power (mean markdown 1.53), considerable dispersion even within industries, a robust positive association with employer size and scope, and a U-shaped secular pattern with increasing monopsony since the early 2000s. Concentration measures are weakly related to markdowns cross-sectionally and miss recent increases in employer power. Future research could investigate institutional and regulatory determinants (e.g., unionization, noncompete clauses, right-to-work laws), extend analysis beyond manufacturing, incorporate factor-biased technological change and bargaining models, and refine measurement with richer input price/quality data and occupation-level local market definitions.

• Identification relies on materials as a flexible input without monopsony or adjustment costs; violations would bias markdowns toward zero (labor power understated). Energy as a flexible input performs poorly and appears subject to monopsony.
• Production function estimation uses proxy-variable methods with a translog form and deflated revenues. While deflated-revenue biases cancel in markdown ratios, markups in isolation may be biased; scalar-unobservable assumptions may be violated under market power, though simulations suggest robustness.
• Assumes upward-sloping firm-specific labor supply; results do not map to models where monopsony arises without such supply (e.g., pure search-bargaining à la Diamond–Mortensen–Pissarides without upward-sloping supply), though alternative interpretations via bargaining are possible.
• Potential dynamic considerations beyond quadratic adjustment costs (e.g., customer capital) are not modeled explicitly.
• Benefits are only observed post-2002; inclusion lowers markdowns modestly but data limitations remain.
• Analysis is limited to manufacturing establishments; generalizability to other sectors is untested. Unionization and some local institutional variables are not observed at the plant level.