Monopsony in the U.S. Labor Market

The paper questions the assumption of perfect competition in the U.S. labor market. It argues that wedges between marginal revenue products of labor (MRPL) and wages indicate employer market power (monopsony). The study aims to quantify these wedges—'markdowns'—in U.S. manufacturing plants from 1976 to 2014. The authors highlight the importance of understanding employer market power for devising effective policy responses to labor market trends, especially concerning worker compensation, mobility, and regulations on large firms. The lack of readily available measures comparing MRPL and wages directly motivates this research. The paper estimates plant-level markdowns and studies their relationship with employer characteristics, along with the evolution of aggregate markdowns over time.

The paper situates itself within a growing body of research examining labor market monopsony in the U.S. This renewed interest is fueled by the secular decline in labor's share of income, often linked to increased industry sales concentration and the rise of 'superstar' firms. The authors contribute by providing direct estimates of the MRPL-wage wedge, documenting substantial dispersion in plant-level markdowns and their relationship with employer size, age, and productivity. They also propose a new theory-grounded method for aggregating markdowns to examine long-term trends. The paper contrasts its micro-founded markdown measure with commonly used concentration indices, highlighting the theoretical limitations of such simplified indices.

The authors employ a 'production approach' to estimate markdowns, combining insights from Hall (1988) and production function estimation techniques from industrial organization (IO) literature (Olley and Pakes, 1996; Levinsohn and Petrin, 2003; De Loecker and Warzynski, 2012; Ackerberg, Caves and Frazer, 2015). The key is to identify markups and markdowns separately, avoiding confounding effects of market power in both input and output markets. The method assumes at least one other input is flexible (meaning no adjustment costs and monopsony power) to identify the markdowns. Material inputs are assumed flexible. The authors use a translog production function—a second-order approximation to any arbitrary, differentiable production function—to estimate output elasticities and revenue shares. They use administrative data from the U.S. Census of Manufactures and the Annual Survey of Manufactures (1976-2014), with deflators from the NBER-CES Manufacturing Database. The authors' estimation procedure employs proxy variable methods (Olley and Pakes, 1996; Levinsohn and Petrin, 2003; De Loecker and Warzynski, 2012; Ackerberg, Caves and Frazer, 2015) to address unobserved productivity. They estimate production function parameters using a three-step GMM approach, creating instruments from lagged input values (except capital). For aggregate markdowns, a novel measure is proposed, incorporating two key requirements: consistency with aggregate wedges (gaps that a representative firm would face) and consideration for local labor markets. This leads to an aggregation rule based on sales-weighted harmonic averages adjusted for output elasticity heterogeneity. They aggregate across labor markets (3-digit NAICS-county pairs) using employment weights. The authors compare this aggregate markdown measure with employment concentration indices (Herfindahl-Hirschman Index or HHI).

The cross-sectional analysis reveals significant monopsony power in U.S. manufacturing. The average plant's MRPL is 53 percent higher than its wage, implying workers receive about 65 cents per marginal dollar. Substantial markdown dispersion exists within 3-digit NAICS industries (average interquartile range of 61.6 percent). Markdowns are positively associated with establishment size, whether measured by employment share, industrial scope, or geographic scope. The relationship with age is less robust, while the markdown-productivity relationship shows a U-shaped pattern, with higher markdowns for the most productive establishments. Plants in multi-unit firms and firms active in multiple sectors or locations have higher markdowns. Analysis distinguishing production and nonproduction labor showed similar markdowns for both categories, suggesting that monopsony is pervasive. The time series analysis of aggregate markdowns reveals a U-shaped pattern: a decline from the late 1970s to the early 2000s, followed by a sharp increase. This contrasts with trends in markups. The correlation between aggregate markdowns and employment concentration is weak at the local market level and over time, suggesting that concentration indices may not accurately reflect employer market power as measured by markdowns.

The findings challenge the assumption of perfect competition in U.S. manufacturing labor markets, demonstrating widespread monopsony power. The positive relationship between markdowns and size highlights the importance of considering firm size when assessing the welfare implications of market power. The U-shaped time trend in aggregate markdowns suggests that a reduction in monopsony power in the late 20th century has reversed since the early 2000s, which is interesting given observations about declining business dynamism. The weak correlation between markdowns and concentration indices raises questions about the adequacy of using concentration as a proxy for employer market power. The results suggest that more micro-founded measures like markdowns are needed to inform policy debates about labor market power.

The paper contributes direct estimates of employer market power in U.S. manufacturing, showing significant monopsony and substantial heterogeneity. It introduces a new theory-grounded measure of aggregate markdowns, highlighting a recent increase. The weak correlation with concentration indices underscores the need for more nuanced measures of employer market power. Future research could explore the impact of factors like unionization, non-compete agreements, and labor regulations on markdowns.

The study focuses on the U.S. manufacturing sector and might not generalize to other sectors. The reliance on material inputs as a flexible input and the potential presence of monopsony power in the material input market could lead to underestimation of labor markdowns. Unobserved heterogeneity in firms and the limitations of the proxy variable method, including the inability to directly observe physical output, may also impact the estimates. The time-series analysis is limited to census years.