Labor Monopsony and the Limits of the Law

A. Models of Monopsonistic Competition

In this section we briefly outline the economic models of monopsony where firms have market power despite markets being “thick”—with large numbers of employers and workers. While these are reviewed in a number of other papers, we focus on their most recent variants and their implications for antitrust. These models formally capture the forces generating monopsony identified by Robinson and Reynolds and do not rely on concentration as an important determinant of monopsony power.²

The paradigm of antitrust is focused on market structures with few firms. Indeed, the primary diagnostic statistics are measures of concentration, for example, market share of the top four or five firms and Herfindahl measures of concentration (often motivated by underlying Cournot models of imperfect competition among a small number of firms). We first present a simple variant of this type of model here.

The Cournot model is the original workhorse model of industrial organization, where n firms take a demand curve as given and choose quantities produced. This model is simple to recast as an oligopsonistic model where employers choose employment facing an increasing, inverse labor supply curve w(.), as in Naidu, Nyarko, and Wang (2016), among many others. In a simple variant of this model with constant marginal productivity p_i, firms choose L_i to maximize:

Under a sufficiently convex w() function, this game will have an equilibrium solution which motivates the Herfindahl (, where is the share of employment at firm i) measure of concentration, and expresses the share-weighted markdown as the ratio between the Herfindahl-Hirschman index (HHI) and the aggregate labor supply elasticity to the labor market: η^agg = w/[w’(L)L], where

In this model, wages will fall as local labor market concentration increases, holding productivity and the aggregate labor supply elasticity constant, a result that has been found in a number of recent papers (Benmelech, Bergman, and Kim 2022; Azar, Marinescu, and Steinbaum 2022; Rinz 2022).

One difference between this literature and the previous literature on concentration is which “quantity” measure is appropriate. A more general form of the Cournot model, where employers are not choosing employment, but rather some other strategic variable, such as vacancies or recruitment intensity or retention efforts, could be warranted.³ Below we discuss measures of concentration that account for likely flows of workers between firms, which synthesize a number of sources of monopsony.

But even in these general models, market power falls as the number of firms, n, increases and approaches the perfectly competitive limit. The interpretation of the empirical relationship may be quite sensitive to how markets are defined, as it depends on which aggregate labor supply elasticity is used. More germane to our purposes here, the Cournot model also shows that market power cannot persist in the long run as more and more firms enter, attracted by positive profit. If there are no fixed costs of entry, for example, long-run markdowns are zero.

The Cournot model can help us compare product markets to labor markets. The literature on concentration in labor markets has pointed out that, for reasonable definitions of labor market boundaries, labor markets are much more concentrated than product markets (Marinescu and Hovenkamp 2018), with HHIs well exceeding the antitrust thresholds. Further, the scope for monopsony power in labor markets is larger, even holding the degree of concentration fixed: product market demand elasticities from the trade literature vary from 1.2 (footwear) to 17 (crude oil), with a median product elasticity of 4.5 (Broda and Weinstein 2006), while aggregate labor supply elasticities are generally between 0.1 and 1 (Chetty 2012). Within a Cournot model, these would imply that the same HHI would result in a larger markdown in the labor market than the corresponding markup in the product market.

The Cournot model has the virtue of being extendable and flexible. Naidu, Nyarko, and Wang (2016) use the average number of other firms workers transition to as a measure of the effective competition in a Cournot model. Krueger and Ashenfelter (2022) expand the Cournot model to accommodate anticompetitive practices. Franchise no-poaching agreements essentially reduce the number of competitors in the labor market, increasing “effective concentration.” We discuss more of these extensions below. Arnold (2020) extends the Cournot model to flexibly accommodate flows between markets and model the effects of mergers.⁴

While concentration is a clear empirical determinant of labor market monopsony, it has not been the traditional focus of labor economists interested in monopsony. Instead, labor economists have focused on models of imperfect competition that have many firms and many workers, yet still deliver upward-sloping labor supply curves to each firm. When market power is generated by forces emphasized in these thick-market models, traditional antitrust remedies may be inadequate.

There are two categories of these monopsonistically competitive models. The first— traditionally more emphasized in labor and macroeconomics—is based on search theory, where workers must actively look for employers. The second, while applied much more recently to labor market problems, is extensively developed in industrial organization and emphasizes product (in this case job) differentiation and heterogeneous tastes for jobs or costs of switching.

B. Search Frictions

As Robinson and Reynolds both pointed out, workers might not know about all the jobs available, and this simple idea has given rise to a vast literature on search in labor markets (see Rogerson, Shimer, and Wright 2005 for a review). Stigler (1961) developed the original model of job search, where workers sample from the distribution of jobs. This model generated a number of predictions, only some of which were borne out in the data. This idea was turned into a dynamic model of sequential search by McCall (1970), where workers would sample repeatedly from the distribution of jobs until a job was located. Rothschild (1973) criticized much of the search literature by pointing out that the distribution of offered wages was taken as exogenous and no underlying model of firm optimization delivered dispersion in the set of offered wages that motivated worker search.

Diamond (1971) allowed both firm optimization and identical searching workers and showed that when a model of costly worker search was integrated with a model of identical firms choosing wages, the unique equilibrium was complete monopsony, where all firms offered workers exactly their outside option, and no worker searched. In this equilibrium, no firm deviates by paying a higher wage because no worker is searching (so they get no additional workers), and no worker bothers searching because all firms are offering identical wages.

This model is somewhat extreme, and substantial progress was made by Burdett and Mortensen (1998) in extending the search model to include on-the-job search. In this variant of the search model, workers continue to search while working, which gives firms an incentive to try to poach workers from each other. Strikingly, in this model, the unique equilibrium, despite a large number of identical workers and firms, has wage dispersion, where each firm pays a different wage, trading off losing workers to firms paying more with making higher profits on each worker that stays.

The key parameter measuring the degree of labor market competition in this model is the ratio, κ, of the job-offer arrival rate to the (exogenous) job destruction rate. κ measures the rate at which searching workers encounter offers relative to the rate at which they lose jobs; when it is high, the labor market is tight, and indeed, as this ratio approaches infinity, the wage dispersion collapses and the model is identical to Bertrand competition, with all workers paid their productivity p. As κ approaches zero, the wage dispersion again collapses and the model collapses to the Diamond model described above, with all workers paid their reservation wage b.

Artifices that make it difficult to poach workers who are already employed naturally reduce κ. Naidu (2010) and Krueger and Ashenfelter (2022) model the effect of antienticement laws and no-poaching agreements, respectively, as falls in the job offer arrival rate.

The Burdett–Mortensen model has become a standard workhorse model in labor economics and macroeconomics. Quantitative and structural variants of this model have been fit to explain the job ladder over the business cycle, the fall in the labor share, wage inequality, minimum wage effects on employment and inequality, and empirical patterns of tenure and experience wage profiles.

C. Job Differentiation

Bertrand (1883), while reviewing Cournot, pointed out that firms are more likely to compete on prices than quantity. In Bertrand’s famous model, even a single competitor is enough to discipline a firm’s exercise of market power. The labor market analogue of Bertrand competition has firms maximizing:

In the classical Bertrand model with heterogeneous firms, firms can hire unlimited workers, and if firms have different productivities, all workers will be employed by the most productive firm, which will pay the productivity of the second most productive firm. Concentration in this model is uninformative, as there will be only one firm employing any workers, and yet that firm has no market power (if the firm lowered the wage even slightly it would lose all its workers immediately to a new entrant). The contestability of a market by new entrants is what leads many economists to dismiss measures of concentration as proxies for market power.

However, this model of workers choosing jobs considering only the highest wage is unrealistic. Firms are places to work and consist of a high-dimensional set of amenities, whose valuations will vary wildly in a given population of workers. A large literature in product competition has extended the Bertrand model to include models of product differentiation, where firms still compete on price, but products are imperfect substitutes for each other.

Differentiation can occur along a wide variety of characteristics, including spatial location, product characteristics, and product quality. Similarly, jobs are also highly differentiated, both spatially (for example, location of employer and hence commute times) and along bundles of amenities, both tangible (for example, benefits, job safety) and less tangible (relationships with managers and coworkers).

Jobs are not only differentiated along these various dimensions, but workers perceive and value these dimensions differently. Note that even jobs that seem quite standardized (for example, cashier) in terms of the productive tasks, so that there is little firm-specific human capital, can be valued idiosyncratically by workers due to particular tastes over commutes and workplace relationships. In models of random utility, workers have idiosyncratic utility over different jobs. Crucially, firms may not be able to observe this taste heterogeneity, and internal constraints on wage discrimination (for example, internal equity) may force firms to post only one wage per job. This restriction is what makes labor market power inefficient: if firms could perfectly tailor the wage to each worker’s taste for working at that firm—so-called perfect wage discrimination—there could still be market power, but it would not be inefficient, even as all surplus is extracted by the employer.

Firms thus know there are some workers who would work for the firm at a lower wage, but do not know which workers those are. Therefore, the profit-maximizing strategy is to pay below marginal product, accepting the loss of the workers who prefer working somewhere else in exchange for the profits made off those workers who stay.

These models have a long tradition on the product market side, being workhorse models in quantitative marketing research, as well as being the foundation for much of demand system estimation in industrial organization. Anderson, De Palma, and Thisse (1992) provide a survey of product differentiation models, while Card et al. (2018a) adapt a variant of multinomial choice to model monopsony in the labor market.

Gabaix et al. (2016) show that there is a subtle relationship between competition and the degree of price distortion in these random utility models. Depending on the shape of the distribution of the random utility terms, markups can stay positive even as the number of firms increases; indeed, if the distribution is fat-tailed enough, markdowns can grow without bound even as the number of firms approaches infinity! This result shows that simplistic measures of concentration or number of firms do not necessarily map into more or less wage (or price) distortion.

Finally, just as workers have distinct preferences over jobs, firms have different preferences over workers. This gives the market a two-sided matching characteristic, naturally implying thinner markets than when only one side has preferences over the other side. The set of workers one firm prefers may be small, and the set of workers that prefer that firm is even smaller. This “differentiation-squared” may exacerbate market power in labor markets relative to product markets.

Azevedo (2014) shows how market power can operate in a model of two-sided matching even when matches are optimal. In this model firms have some market power (even assuming wages are fixed) and reduce their hiring in order to shed marginal workers, who then get employed by another firm, who in turn sheds workers that may be supramarginal for the initial firm. These “rejection chains” give firms with labor market power an incentive to set quantities too low in order to obtain more productive workers, even if they cannot change the wage. Caldwell and Danieli (2018) propose a two-sided matching model of the labor market and use it to calculate an outside options index (OOI), which is the inverse of worker-specific Herfindahl index over likely other jobs. If a worker with characteristic X is likely to be employed by many firms, the OOI is high. Monopsony in this model could manifest as many workers with different OOI nonetheless being paid the same wage within a firm.

A related, but more sociological source of taste-based monopsony is endogenous worker loyalty. Bowles and Gintis (1976) emphasized the role of endogenous worker preferences in maintaining employer profits. A large literature in sociology, perhaps beginning with Michael Burawoy’s Manufacturing Consent: Changes in the Labor Process under Monopoly Capitalism (1979), argued that employers engineer work environments to maximize worker productivity and minimize conflict with management. Workers also develop attachments to coworkers (Fantasia 1995), employers, and in service-sector industries, even customers (Hochschild 1983). While this literature has not considered the effects of these types of attachments on quit behavior, job search, and labor market competition, it is plausible to believe that workers may be exploited due to their personal or emotional attachments to the job and to working. Employers may try to invest in increasing these attachments by fostering corporate loyalty or specializing the content of training programs.

When firms have product market power, advertising is quite profitable, as profits are positive on every additional unit sold. A similar logic is true for monopsonists, as monopsonists suffer losses of profits from every additional quit. But employers get eight hours a day of free advertising to their employees, and “corporate culture” together with “employee loyalty programs” may be inefficiently high as companies invest in these devices to cultivate loyal workers who are wage inelastic, rather than raising wages to prevent quits. There is very little in the way of compelling evidence on the causal effect of loyalty on wages; however, Masakure (2016) provide one such attempt, but the exclusion restriction, that religiosity is a valid instrument for loyalty, is dubious.

Workers’ social preferences for fairness may also militate against wage discrimination, even when employers have accurate information about worker’s tastes and outside options. A considerable amount of evidence (Card et al. 2018b; Fehr and Falk 1999; Breza, Kaur, and Krishnaswamy 2019; Dube, Giuliano, and Leonard 2019) has shown that workers react negatively to disequalizing pay schemes, and this may exacerbate the inefficiencies due to monopsony, even while preventing employers from fully exploiting their market power (via wage discrimination of workers).

D. Hybrid Models and Alternative Definitions of Concentration

The recent literature on monopsony has grown quickly, and there are a number of recent models that are hybrids of the three models above. Berger, Herkenhoff, and Mongey (2019) construct a nested logit model where workers first choose a labor market, then choose a firm within that labor market. The authors choose utility functions and random utility shocks that yield a nested constant elasticity of substitution formulation of the labor supply decision of a representative household. This model is a case of our reduced-form decomposition below, where the labor supply elasticity facing the firm can be decomposed into a firm-specific component (based on differentiation) and a market component (based on concentration). The model delivers a wage-share variant of the HHI that combines the monopsonistic competition dimension of monopsony with the concentration dimension.

Arnold (2020) uses the Longitudinal Employer Household Dynamics (LEHD) survey, a matched employer–employee data set, to look at the effect of mergers on wages. In his theoretical model, rather than defining concentration with respect to a fixed industry within a commuting zone, he uses job-to-job flows across industries to construct “shares” where a firm’s employment is divided by flow-weighted employment of all other firms in the commuting zone. This measure of concentration has the advantage of not requiring one to take a stand on which administrative boundary (for example, occupation or four-digit industry) is relevant for defining a labor market. Arnold shows that this measure is generally highly correlated with commuting zone × industry HHI indexes.

Finally, a pair of papers relate concentration to wages via bargaining rather than monopsony. Schubert, Stansbury, and Taska (2021) use Burning Glass data to measure concentration but add an outside option that depends on occupational transition probabilities. In their model, wages are set by bargaining rather than wage posting. In another recent study, Jarosch, Nimczik, and Sorkin (2019) develop a model with search and bargaining where the worker’s outside option depends on the concentration of firms. Firms in this model commit to never renegotiating with a worker after the first match, and this means that large firms have a bargaining advantage because workers will be more likely to encounter offers from that large firm in the future; never being able to contract with a large firm again makes a worker quite willing to take whatever offer that firm is offering. Hence, highly concentrated labor markets have workers with low outside options and thus lower wages. While technically the assumption of efficient bargaining makes this not a model of monopsony, this model will also generate a negative relationship between concentration and wages.

As far as we are aware, there is no study that combines search, job differentiation, and concentration into a single model of monopsony. There is some structural work (Flinn and Mullins 2019) that endogenizes the choice between wage-posting and Nash bargaining, and integrating this into the empirical literature seems like a promising direction for future work.

Which measure of concentration is to be preferred? Bassier, Dube, and Naidu (2022) show correlations between a large number of HHI indexes (for example, flows adjusted, payroll vs. new recruits vs. employment shares, industry vs. geography) and firmspecific labor supply elasticities, with ambiguous results. While HHI indexes are all highly correlated with each other, they are only weakly correlated with the firm-specific residual elasticities. This suggests that more fine-grained measures of concentration ought to be constructed at the firm level, while allowing for variation in firm wages even in the limiting case of no strategic interaction. This perhaps could be done by synthesizing the Caldwell and Danieli (2018) measure (or a firm variant of Arnold 2020) with the Berger, Herkenhoff, and Mongey (2019) measure, combining flows across firms with shares of payroll rather than employment.

E. Evidence for Monopsonistic Competition

What is the evidence for monopsonistic competition in the labor market? Direct estimates of monopsony power that are obtained in thick labor markets are the most compelling evidence. For example, the minimum wage results that motivated the original monopsony model were obtained in thick labor markets. The most credible evidence is provided by the few randomized controlled experiments where wages are randomized for identical jobs in markets with many wage-setters and few barriers to mobility. Caldwell and Oehlsen (2018) randomize wages for Uber drivers, while also including those that also drive for Lyft. They examine the rate at which drivers switch, and find a surprisingly low elasticity of between 4 and 5, given that workers literally just have to switch apps on their phone. Dube, Jacobs, Naidu, and Suri (2020) experimentally vary wages for an identical task and find substantial monopsony power even on (putatively thick) Amazon Mechanical Turk, and Dube, Giuliano, and Leonard (2019) find a moderate quit elasticity (in response to wage changes induced by discontinuities in the national pay-setting scheme in a large low-wage retailer). Dube, Manning, and Naidu (2018) bound the labor supply elasticity facing the firm in low-wage labor markets using the extent of bunching at $10.00. Bassier, Dube, and Naidu (2022) use firm fixed effects to isolate the firm component of worker wages in matched worker–firm Oregon data and use this to estimate a labor supply elasticity between 4 and 5. They also find that this elasticity does not vary very much by employer concentration.

Other direct estimates include evidence on wage increases from small firms that exogenously receive shocks to the firm-specific marginal product of their labor. In the United States, these shocks include patents in small firms (as in Kline et al. 2019), fiscal stimulus shocks from American Recovery and Reinvestment Act (ARRA) contracts (Cho 2019), and wins in procurement auctions (Kroft et al. 2020). Goolsbee and Syverson (2019) use shocks to admissions at universities to show monopsony power in the tenure-track academic labor market, which is generally national in scope. By dividing the (log) employment response to the shock by the implied rent-sharing elasticity (the log wage effect of the shock), an estimate of the residual labor supply elasticity is obtained. Evidence in this vein against monopsonistic competition is provided by Matsudaira (2014), who finds that exogenous increases in nurse aide employment resulted in no increase in wages, suggesting perfectly elastic supply curves.

Arnold (2020) finds negative effects of concentration driven by mergers in thin labor markets but concludes that mergers and concentration are unlikely to be a big contributor to overall wage stagnation, with concentration resulting in wages only 5 percent below a fully competitive benchmark. This relatively small concentration markdown is difficult to reconcile with the rent-sharing elasticities or firm-specific labor supply elasticities unless there is a significant degree of non-concentration-based labor market power.

There is also evidence from structural papers that estimate both differentiation and concentration in the same model. While dependent on functional form assumptions, these papers show that there remains a significant amount of “pure” horizontal differentiation. Berger, Herkenhoff, and Mongey (2019) recover a within-market substitutability of 5.4, and Azar, Berry, and Marinescu (2019) estimate a structural labor supply model on vacancy data and also recover a pure differentiation elasticity of around 5.8.

A simple way to reconcile these estimates is to note that monopsony is not the only constraint firms face. Firms must also use wages to provide incentives, mitigate adverse selection, and induce loyalty and reciprocity. Rent-sharing elasticities reflect all of these other constraints on firm wage-setting, not simply the labor supply constraint. For example, in efficiency wage models combined with monopsony, the wage is potentially much higher than the pure monopsony wage (although employment is potentially much lower and would be characterized by job rationing). This is plausibly what is happening in the estimates of Matsudaira (2014), where wages are regressed on shocks to employment. It also cautions against taking the ratio of employment to wage responses to a shock as a clean estimate of the labor supply elasticity because it could, for example, reflect the shape of monitoring costs (Rebitzer and Taylor 1995). While the instrumental variables (IV) estimate does reveal a mean residual supply elasticity, one could imagine diverse rent-sharing mechanisms (for example, amenities) inducing heterogeneity in the estimates.

Where labor markets appear governed by “the law of one price,” it may be due to the effect of social norms and convention rather than competition. Dube, Giuliano, and Leonard (2019) show that the appearance of high quit rates in response to wage changes is driven more by comparisons across workers than sensitivity to one’s own wage. That is, when all workers are given the same wage increase, the quit rate falls by only a little bit, but when some workers are granted raises while others at the same store are not, the latter are much more likely to quit. Breza, Kaur, and Krishnaswamy (2019), whose paper is tellingly titled “Scabs...,” also suggest rural village markets exhibit substantial monopsony power. They randomize wages to workers in private and in public and find that workers are willing to take jobs at close to the same rate even at a 10 percent wage cut in private, but are unwilling to do so in public. The appearance of an extremely elastic labor supply facing the employer/experimenter is driven by social sanctions against accepting low wages in public. A possible explanation is that workers recognize that they are collectively harmed if they compete over wages and employ social sanctions to restrict competition, a practice that is of course formalized by unionization but can occur informally as well. As is well known, it is difficult to infer much about market structure from dispersion alone, as both competition and collusion can result in uniform wages; in labor markets the prevalence of social norms around pay-setting may exacerbate this difficulty.

A. Some Anti-Antitrust Arithmetic

In this section, we calibrate a model of imperfect competition that allows for both concentration and non-concentration-based sources of monopsony. Concentration works via Cournot competition in quantities, with each firm’s choice of quantities altering the wage that every other firm must pay. Non-concentration-based monopsony works by allowing the wage each firm must pay to be an upwards-sloping function of its own employment choice, irrespective of the employment choices of other firms. We then use existing estimates of concentration and aggregate and firm-specific labor supply elasticities to calculate how much monopsony power is “frictional” and would remain even if concentration were eliminated.

In order to implement our calibration, we need a study that estimates the residual labor supply facing the firm, as well as concentration, to see if concentration can be a reasonable conduct parameter given an aggregate labor supply elasticity. Many of the papers that measure labor market concentration do so in markets where it is difficult to know the market-level labor supply elasticity. In Chetty (2012), a lower bound is given by the aggregate market labor supply with an extensive high end of the elasticities of 0.5, but this is likely much too low for the labor supply elasticity to a particular market. Another estimate is given by Monte, Redding, and Rossi-Hansberg (2018), who suggest a local employment elasticity of 1.52, although their model emphasizes heterogeneity, with elasticities between 0.5 and 2.5.

Alex Bartik (2018) provides a more appropriate estimate of the aggregate supply elasticity to a labor market that is closer to the labor market definition of many of the papers that measure concentration. He uses trade shocks with China and the fracking boom to estimate migration elasticities into labor markets, finding significant obstacles to mobility, with relocation costs around 20 percent of annual income (and between 4–5 percent for occupation and industry). These can be converted into labor supply elasticities using the ratio of the effect of each shock on employment to the effect on wages. Using estimates from Table 5 and Table 6 of his 2018 paper, we get labor supply (to the sector × commuting zone level) elasticity estimates of roughly 2.8 from both shocks.

We can then start to see why concentration cannot be the primary source of monopsony power from Webber (2015), who finds the average firm share of employment is only 9 percent in a county–industry, and yet he finds a firm-specific labor supply elasticity of 1.08. Taking the above values of the aggregate labor supply elasticity, this is simply too much monopsony power and too low an average firm share of employment for concentration alone to be the source of monopsony power, which suggests that the bulk of monopsony is coming from “nonconcentration” sources, such as search frictions and job differentiation.

Webber (2015), using the LEHD, finds little correlation between wages and employment concentration, but defines concentration as share of county X industry workforce employed at the firm. In contrast, more recent work, Azar, Marinescu, and Steinbaum (2022); Benmelech, Bergman, and Kim (2022); Rinz (2022); Lipsius (2018); Qiu and Sojourner (2019); and Hershbein, Macaluso, and Yeh (2019) find negative and significant effects of concentration (measured as HHI at the commuting zone–occupation level, or county × industry level, or county × “sector” level) on wages, although some of these papers only find effects in IV specifications. These differences suggest that more work on the definition of boundaries of the labor market is required, as well as credible identification of the effect of concentration.

The equation above expresses the firm-specific inverse labor supply function as the product of a firm-specific scaling factor α, times a w(Σ_j,L_j) term that represents the aggregate inverse labor supply function. If aggregate employment is low, then the wage is low. If the number of firms is small or a single firm has a large share of employment, then individual firms will be able to influence the wage they face by changing their employment. If α = 1 is constant, then this wage function is simply the Cournot model.

The α scaling only depends on firm i’s choice of employment. This term captures all the “competitive” labor supply frictions that would influence firm i’s choice of employment even if “concentration” did not matter, for example, search and differentiation. α scales the inverse aggregate labor supply curve by a factor that depends only on the employment choice of firm i, ignoring all the labor decisions of the other firms. Suppose the elasticity η^comp = α/[α’(L_i)L_i] is constant. If η^comp were large, then it would suggest that concentration was the primary source of monopsony. Writing the expression for the markdown we get: with s_i, = l_i/L denoting the share of employment at firm i. We can solve this for the “competitive” labor supply elasticity to get

Webber (2015) finds the average firm employs 9 percent (SD = 16) of the industry in its county and estimates firm-specific labor supply elasticities of 1.08 on average. This combined with the equation above gives a competitive labor supply elasticity of roughly 1.1 (at η^agg = 2.8). We can then plot the implied residual supply elasticity as the average firm employment share (or equivalent equal-firm HHI) changes. Figure 3 implies the bulk of the source of monopsony power is coming from competitive sources, and it implies that changing concentration would have a small effect on the labor supply elasticity. Note that the firm labor supply elasticity falls slowly as concentration increases, falling from close to 1.1 to around 0.81 as concentration goes from 0 to 1. This suggests there is still a considerable markdown (wages almost 50 percent of marginal product) even if concentration is extremely low.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Firm Residual Labor Supply Elasticities/Inverse Markdowns

Notes: This figure shows residual labor supply elasticities as a function of firm share of employment (or HHI with equal-sized firms). Residual supply elasticities are calculated as the harmonic mean of the competitive labor supply elasticity and the share-weighted aggregate supply elasticity. The competitive labor supply elasticity is estimated as described in the text using the residual labor supply elasticities reported by either Webber (2015) or Bassier, Dube, and Naidu (2022), along with firm employment shares for 1,000 firms drawn from a Pareto distribution with the mean and standard deviation reported in the respective paper.

We can also use comparable results from Bassier, Dube, and Naidu (2022), who argue the estimates recovered from the method used by Webber are too low and emphasize using firm-level variation in wages rather than worker-level variation. They find a residual supply elasticity of 4.5, with an average firm share of 8.2 percent (SD = 19). Figure 3 shows that even as firm shares approach 0, the residual supply stays below 5.

With heterogeneous firm employment shares, we can look at the share-weighted markdown, given by:

And summing across firms yields:

In a Cournot-type model, the firm-specific η = H/η^agg, the residual elasticity η^comp is infinite, and only concentration and the aggregate labor supply elasticity matters in determining markdown. Using the upper aggregate elasticity implied by Bartik (2018), the levels of HHI required to get a firm-level labor supply elasticity of less than 4 are all greater than 0.7, well outside the range of existing estimates. The mean level of HHI at the three-digit manufacturing industry × commuting zone level in Benmelech, Bergman, and Kim (2022) is 0.23. Azar, Berry, and Marinescu (2019) find a six-digit occupation × commuting zone HHI of 0.32. Hershbein, Macaluso, and Yeh (2019) find commuting zone HHIs between 0.2 and 0.1 for nonmanufacturing and around 0.35 in manufacturing. Rinz (2022) finds local HHI (at the industry × commuting zone level) has declined from 0.2 in 1976 to around 0.15. The divergence of labor market definitions across papers means that we do not yet have a clear picture of whether the Cournot-plus-concentration markdown can account for the low elasticity of the firm-specific labor supply curve. But these levels of concentration would imply residual supply elasticities three to six times aggregate supply elasticities, and with the aggregate supply elasticities 1.5 or higher, would be generally larger than recent estimates.

The takeaway from these exercises is that with concentration alone we might have a difficult time accounting for the low labor supply elasticities estimated in the literature.

Depending on the market definition used, concentration is too low, and the aggregate labor supply elasticities to local labor markets are too high for it to quantitatively matter— there is simply too much “competitive” market power.

One caveat is that we have chosen relatively large labor supply elasticity to the local labor market, based on Bartik (2018). If we had instead gone with a smaller elasticity (0.1 or 0.5) consistent with supply to any market (for example, Chetty 2012), we could have rationalized firm-level labor supply elasticities less than 3 with existing concentration estimates. A 0.2 to 0.4 HHI and a 0.5 to 1.5 aggregate supply elasticity can rationalize firm-specific labor supply elasticities within the plausible 1–4 range (Sokolova and Sorensen 2018). We show this in Figure 3 for completeness, but remind readers that the measures of concentration used in the literature are generally computed at levels much finer than the aggregate labor market, so the relevant supply elasticity is unlikely to be the aggregate one. While concentration may be important as a source of monopsony in some labor markets (for example, extremely concentrated labor markets), nonconcentration (that is, frictional) sources of monopsony also seem to be important.

We can compare the results of this exercise to the estimated parameters in Berger, Herkenhoff, and Mongey (2019), who specify a full nested labor supply system. They obtain an inverse residual labor supply function of the form where j subscripts distinct markets, summarizes strategic interactions across firms within a labor market j, and L is an index of aggregate employment.⁷ Using heterogeneous responses to corporate tax changes and the equilibrium in their model, they estimate η = 5.38 and ϕ = 0.66 and set φ = 0.5. Thus, even if all strategic interactions were eliminated, so that , the degree of labor market power would stay substantial, with markdowns of approximately 15–20 percent.

Finally, we can do a similar exercise with Amazon Mechanical Turk. Using the data from Dube et al. (2020), we calculate the HHI of MTurk jobs posted in a given hour–task category, which averages at 0.8 (consistent with other results showing high concentration on MTurk from Kingsley, Gray, and Suri 2013). But Dube et al. (2018) estimate a firm-specific labor supply elasticity of 0.14. For concentration to be the primary source of monopsony, the aggregate labor supply elasticity must be quite small, roughly 0.112. If we think of the aggregate labor supply elasticity on MTurk as akin to a Frisch hours labor supply elasticity, this is an order of magnitude below normal estimates of this parameter. For example, Angrist, Caldwell, and Hall (2017) estimate an elasticity of 1.2 using experimental Uber data, and Fehr, and Goette (2007) estimate between 1.12 and 1.25 on Swiss bicycle messengers.

A. Wage Regulation

A much-discussed response to the problem of wage suppression is the minimum wage or living wage law. A minimum wage prohibits the employer from paying workers a wage below a certain level. Mandatory minimums in this spirit can be, and often are, applied to other aspects of work. For example, maximum hours laws limit the number of hours that workers can be required to work or require extra pay for hours above that limit. Laws that require employers to meet minimum health and safety standards have a similar effect. They prevent an employer from underproviding what is effectively inkind compensation in the form of relatively safe or pleasant working conditions.

The standard criticism of minimum wage laws is that they will result in unemployment as employers fire workers to whom they must pay a wage greater than then the workers’ marginal revenue product. But this criticism assumes that labor markets are competitive. The more serious problem with minimum wage laws is that they can only help a small class of relatively poor people—workers who would otherwise be paid slightly less than the minimum wage—and not more deeply impoverished people or workers higher on the wage scale. When the monopsonistic wage level exceeds the minimum wage, minimum wage laws have no effect. If the wage is only a small part of the total compensation, or total compensation is fungible between wage and nonwage components, minimum wages alone will have a negligible effect. Moreover, the minimum wage must be carefully calibrated: if the wage level is set too high, then disemployment effects may be greater than the wage benefits. It may be difficult for governments to calibrate the minimum wage correctly. On the whole, uniform minimum wage laws can be only a small part of the response to wage suppression caused by monopsonistic competition.

A more thorough and flexible wage mandate response to pervasive monopsony would be wage boards, for example, as is prevalent in Australia and in some U.S. states for some industries (for example, New York and California). While firm-specific minimum wages are theoretically ideal, a more practical proposal might involve setting industry and region-specific minimum wages. Wage boards periodically set wage floors by industry, occupation, and location, using nonpartisan expert appointees (in the Australian case) or tripartite employer–worker–government commissions (as in the U.S. case) (Madland 2018). Dube (2019) calibrates effects of a system of wage boards on U. S. wage inequality.

B. Tax and Transfer Policies

It is well understood that the fiscal system solution to market power involves subsidizing the price paid by the firm. For example, the government could offer to pay a monopolist a subsidy for every unit sold, so that the monopolist voluntarily manufacturers and sells the number of units that it would produce in a competitive market. Transposed to labor markets, the government would subsidize the firm’s wage. Standing alone, such a scheme would have unattractive distributive effects, but if a corporate tax on pure profits was coupled with a precisely tuned (that is, equal to the optimal level of employment) subsidy on wages, the gains from alleviating the monopsony distortion via a subsidy could be redistributed.

Under this approach, the government should apply the subsidy only to employers with monopsony power, and the extent of the subsidy should be a function of the degree of monopsony power. But the existence, and especially degree, of labor market power is never self-evident. It is the domain of antitrust law in the first place to determine whether an employer has power in a labor market, and this fact-intensive inquiry seems to require lengthy hearings by courts. Further, firms will have an incentive to cherry-pick the best workers under the wage subsidy scheme. Taxes are not used to police product market power and are likely not a good instrument for labor market power.

A popular policy that has unanticipated consequences under monopsony is the Earned Income Tax Credit (EITC). The EITC subsidizes earnings of low-income households and is among the largest forms of redistribution in the United States. However, in monopsonized labor markets, the EITC can have perverse consequences. By increasing the effective (after-tax) wage earned by the worker, it encourages people to enter the labor market (shifting the labor supply curve to the right), enabling employers to pay a lower wage to the incumbent workers. While the EITC can undo the disemployment effects of monopsony, it will also transfer a share of those benefits to employers while reducing wages for incumbent workers.⁸

But a subsidy that leverages private information could be implemented in labor markets where firms do not have discretion over hiring. Imagine the following employment regime. Employers are required to make a public list of all the jobs that they offer, along with the qualifications and compensation, and are further required to hire the first qualified person who applies for it. Then monopsony power can be eliminated by subsidizing wages paid by employers. This subsidy has a similar economic motivation as the commonownership self-assessed tax proposed by Posner and Weyl (2018); a monopsonist employer has an incentive to quote too low a wage, and the subsidy blunts this incentive. The “take-all-comers” hiring policy is essential to make this work, but may not be such a stretch in the era of gig-work, where companies like Uber operate by offering a highly standardized form of work to workers who are hired based on their conformity to a rigid set of ex ante specified qualifications.

A recent literature on “robust monopoly regulation” (Guo and Shmaya 2019) may also be useful for adjudicating between wage mandates and subsidies. The idea in robust policy design is that policymakers do not even know the distributions of possible worker tastes, outside options, or firm valuations of workers. Can optimal (in some sense) policies even be formulated? The robust mechanism design literature suggests that not only can such policies be characterized, they are often simpler and much more transparent than policies that presuppose policymaker knowledge. Translating their product market results into the labor market case, Guo and Shmaya (2019) find that the “regret-minimizing” policy features either a minimum wage or a (capped) wage subsidy, or both, depending on how much the profits of business are valued relative to workers’ wages.

Worth considering as part of the policy mix for combating monopsony is unemployment insurance (UI). While a large literature considers the optimal level of unemployment insurance in light of the insurance–incentives trade-off, little of this literature has accounted for the role of UI in blunting employer market power. For example, legitimate reasons to quit (for example, harassment on the job or poor health and safety standards) can allow a worker to exit employment into the UI system, putting some brakes on employers’ ability to exert monopsony power. However, almost all UI systems do not permit otherwise voluntary quits to take UI, so the role of UI in increasing the outside option of workers is limited, and this may account for why recent evidence has shown that within-firm increases in UI generosity do not lead to appreciably higher wages (Jäger et al. 2018). The enormous expansion of the UI system during the COVID-19 pandemic may alter the terms under which workers are eligible for UI, increasing its role as a possible counterweight to monopsony.⁹

C. Mandatory Benefits

Workers are protected by a range of laws that require employers to offer certain benefits to them. Federal mandates include workers’ compensation, safety and health requirements, family and medical leave requirements, and special treatments for veterans.¹⁰ States also impose mandates. Illinois, for example, requires employers to give workers time for a meal if they continuously work 7.5 hours or more and prohibits employers from penalizing employees who miss work in order to vote or serve on a jury.¹¹ Mandates can be loosely defined as legally required in-kind transfers from the employer to the workers, where the workers attach or may attach an intrinsic value to the benefit. We abstract away from certain legal requirements that are designed to increase workers’ bargaining power, for example, union organization rights.

These policies have often puzzled economists because they seem to substitute the government’s judgment about the conditions of employment for the employee’s own judgment as to what may be best for them. Consider, for example, a mandate that employers grant unpaid leave to workers who experience a family medical emergency. It would seem that if workers value unpaid leave of this type a sufficient amount, employers would grant it to them even in the absence of the mandate. The unpaid leave is simply an in-kind benefit—effectively, a kind of weak employer-supplied insurance policy. Suppose, for example, that a worker would be willing to pay $100 for such a policy because it gives them peace of mind, while the cost to the employer is only, say, $50 in lost productivity. By incorporating unpaid leave into the employment contract, the employer should be able to reduce the wage by between $50 and $100. Summers (1989) observes that mandates might be justified where externalities are present, or for paternalistic reasons, but otherwise they are a puzzle.

The logic is the same if the employer is a labor monopsonist. Indeed, it is possible that the labor monopsonist has stronger incentives than a nonmonopsonist to offer certain packages of benefits because the monoposonist will obtain a larger share of the marginal surplus from retained workers. To see why, following the Spence (1975) model of monopoly, imagine that the marginal worker, the one indifferent between working at this firm and the outside option, values amenities more than all the other inframarginal workers A monopsonist may have an incentive to offer a package of low wages and amenities that appeals to the marginal worker but is also transferring surplus from the inframarginal workers to the employer.¹² In such circumstances, labor market mandates, just like quality standards in product markets, may have a role to play in curbing the distortion in job design induced by monopsony.¹³

As Summers also notes (1989, p. 179), the story is more complex if, as will usually be the case, the monopsonist has limited information about employees and potential hires. To avoid adverse selection problems, employers may use packages of wages and benefits that are, from the social standpoint, inefficient. But a policy of mandating benefits in such circumstances does not have straightforward efficiency effects.

Further, to the extent that the cost of benefits is larger than the value workers have for those benefits, mandates will act as a tax and thus magnify the monopsony distortion, resulting in even lower employment and wages than the competitive case. We suspect that mandates will not generally help address labor monopsony power except in the limited case where the minimum wage is binding, so the addition of a mandate has the effect of increasing the effective compensation of a low-income worker. Even here, however, raising the minimum wage would be the better remedy to the problem of labor monopsony, unless the wage is a small share of total compensation. Mandates do not address wage suppression caused by monopsony power.

D. Job Protection

In the United States, most jobs are at-will, meaning that the employer can fire the worker for any reason not specifically forbidden by law (such as racial discrimination). In one state, Minnesota, the law provides that employers may fire workers only “for cause.” Under the for-cause standard, employers may fire workers only if they can prove that the workers are unable or unwilling to perform the job up to standards. In other countries, some workers have even more secure forms of tenure. Laws that put limits on termination of workers also typically prevent the employer from taking lesser forms of actions against workers, like reducing wages or even failing to make cost-of-living adjustments.

A crucial observation with respect to monopsonists is that they are labor constrained: they always want more labor at the given wage, and so it is unclear why monopsonists would fire workers without cause (other models, like efficiency wages, may be needed to rationalize these protections).

In the simplest variant of the Burdett–Mortensen model of search, however, job protections could be understood as lowering the (exogenous) separations rate, and thus increases the tightness of the labor market (ratio of recruitment to separations rate), moving the labor market closer to efficiency. Because fewer workers are fired, fewer are available to hire, and employers must compete more vigorously to hire those who remain. But if employers lose profits because they cannot fire underperforming workers and instead choose vacancies and recruitment effort, then the employer reduction in recruitment effort may outweigh the reduction in the separation rate.

Job protection rules may reduce the bargaining power of employers by depriving them of the ability to fire a worker who refuses to accept a low wage or insists on a higher wage, but they do not help workers in concentrated labor markets—the initial wage will be set at the monopsony rate. At most, they help workers who obtain work at the market wage, or a relatively high wage, and then lose bargaining power as the labor market consolidates or the workers’ outside options diminish for other reasons. These workers will be unable to obtain raises that they would receive in a competitive labor market.

Job protection also has negative consequences. Many economists worry that the job tenure laws in some countries damage the macroeconomy by decreasing labor mobility and reducing employers’ incentives to hire in the first place. Labor rigidity may also make it more difficult for economies to recover from recessions.

A weaker form of job protection comes in the form of notice requirements. The Worker Adjustment and Retraining Notification Act requires employers to give workers notice before laying them off. Notice benefits workers by enabling them to start their job searches while they are still being paid. Notice requirements may therefore enhance workers’ bargaining power by reducing search costs, although perhaps only modestly.

E. Occupational Licensing

Many types of employment are subject to occupational licensing statutes. These statutes require people to undergo training and certification before offering services to the public. Traditional examples include lawyers and doctors, but in the last few decades the list of occupations that are subject to these rules has lengthened considerably, and now includes (depending on the state) hair dressers, auto mechanics, financial advisers, civil engineers, electricians, and funeral directors, among many others. A survey reported by Kleiner and Krueger (2009) found that 35 percent of workers were licensed or certified.

The traditional justification for occupational licensing is quality control. If the government can screen out incompetent service providers, consumers will benefit. Many economists are skeptical of this justification and have argued that the main effect of occupational licensing has been to erect entry barriers that raise prices for services, reduce supply, and benefit incumbents. The crucial observation here is that occupational licensing lowers the supply of labor to a given market and thus raises wages of the licensed, lowers profits of firms (and raises prices to consumers), and lowers the wages of the unlicensed.

Unfortunately, occupational licensing also imposes a cost on people who want to enter the workforce in the first place—since they must pay for training that may otherwise be unnecessary, as well as the fees for certifications. For this reason, occupational licensing may not on balance be a useful way to counter employer monopsony power.

It is possible that occupational licensing could help workers counter labor monopsony power of employers if licensing makes the supply to the firm more elastic as well as lower, which may or may not be the case, and has not been a consideration in the empirical literature on licensing. Monopsony also implies that the deadweight loss from binding occupational licensing will be smaller than if the no-licensing equilibrium were competitive, so the social costs of licensing are smaller.

F. Government Subsidies, Including Training and Employment

Numerous government programs offer various skills training for people. The U.S. government subsidizes student loans and offers tuition grants. States and local governments provide subsidized schooling, vocational training, and university training. Many programs help workers who have lost jobs. For example, the Department of Labor runs the Employment and Training Administration, which offers retraining programs to dislocated workers, among others. The Workforce and Innovation Opportunity Act, passed in 2014, provided additional resources for supporting and retraining people who have lost their jobs.¹⁴ States and local governments also offer numerous services to unemployed workers, including training and matching.¹⁵

These programs offer benefits to ordinary people, but most of them do not address the problem of labor market power. Consider, for example, federal grants and loan subsidies for students who seek to attend college. In the absence of such benefits, people will either borrow in the private market or refrain from going to college. In the first case, the benefit is equal to the difference between the cost of borrowing in the private market and cost of subsidized borrowing along with any grants. In the second case, the benefit is equal to the difference between future income that is obtained as a result of the college education (net of costs) and future income otherwise obtained. In both cases, the benefit is a transfer from taxpayers to the generally lower-income people who qualify for these programs. Employers may benefit from the larger pool of qualified labor. Monopsonistic employers remain free to use their market power to suppress the wages of the people they hire. It is even possible that as the pool of trained workers increases, the workers lose bargaining power, which further enhances the bargaining power of monopsonistic employers, who thus obtain a larger share of the surplus generated by the government programs.

Insofar as worker and firm underinvestment in training is a symptom of excessive labor market power, policies to encourage training can mitigate the distortion. However, without reducing the degree of monopsony power, the bulk of the returns from training will likely be captured by employers. Indeed, some of the hunger for government training programs emanating from the private sector may be due to labor market power: monopsonists are always labor-constrained and demand more and more skilled labor without wanting to raise the wage.

On the other hand, some educational programs may help counter labor market power. We have in mind job-retraining programs, particularly those that give relatively general skills that facilitate occupational mobility. To see why, imagine that a single meatprocessing plant dominates the local labor market for meat-processing workers. Because the workers have few outside options if they are fired, the employer can suppress wages. Now imagine that the government offers job retraining for anyone who has been fired from a job. The program improves the value of the workers’ outside option by enabling them to earn a higher income once they undergo the program after they have been fired. This should increase their bargaining power vis-à-vis the employer, who in turn should refrain from suppressing wages as much as it otherwise would. Note that this pathway for countering labor market power works by reducing search frictions for workers rather than by reducing market concentration or directly regulating the terms of employment. A recent paper by Manoli and Patel (2019) compares experimental results on job search assistance in Nevada with Job Corps and Workforce Investment Act Training programs and finds that only the former delivers persistent earnings gains.

Retraining programs and other programs that help laid-off workers find new, wellpaying jobs could thus be a useful way to counter labor market power. But these programs also have many limitations. They are costly and will only be justified when the benefits for workers exceed those costs. It may also be difficult for the government to offer appropriate retraining programs. The government needs to be able to forecast the demand for the jobs for which training is needed, as well as the willingness of workers to take those jobs and undergo training for them. This type of forecast may be challenging.

G. Job Standardization

None of the proposals we have discussed address the problem of job differentiation— where labor market power arises because apparently similar jobs are actually quite different for workers because of variation in amenities across workplaces. The simplest amenity to think about is commute time: employers that are dispersed across residential locations have more market power than employers that all occupy a dense central business district. But jobs being far away from each other could be true in a much higherdimensional space of job characteristics than simply distance. This problem seems intractable because the variation of amenities may reflect the different preferences of workers, and employers would normally be justified in catering to different preferences. The result is that employers can underpay workers who cannot find valued amenities in other workplaces.

At least as a theoretical matter, however, workers (and the economy) could benefit if labor market differentiation was deterred at the margin.¹⁶ Unions have sometimes performed this function by standardizing jobs across firms within industries (Freeman and Medoff 1984). Nonwage characteristics of unionized jobs are very important to workers’ preferences for unionization (Farber and Saks 1980). The law also plays a role in standardizing work. Minimum wage and maximum hours laws push employers to offer standard eight-hour workdays. This puts a limit on the duration of shifts, which in turn should reduce the variation across employers for this dimension of work. Government-mandated health and safety regulations should also reduce job differentiation by putting a floor under the health and safety conditions of any workplace. However, as far as we know, no study documents the job-differentiation effects of union practices and legal regulations on employer market power, likely because many of these regulations also come along with mandated changes in wages, limiting the value of the exercise.

Some markets for services seem to be naturally standardized. Consider, for example, the services offered to households by plumbers, electricians, drycleaners, locksmiths, carpenters, and other skilled workers. Because these workers typically sell services to numerous households, they are not subject to typical labor monopsony pressures. The law recognizes this distinction by classifying such workers as independent contractors and denying them the protections of employment and labor law—including, for example, minimum wage laws and the right to unionize. The law apparently assumes that market competition protects these workers, so legal protections are unnecessary or less necessary. In contrast, if a (say) plumber goes to work for a single firm like the owner of an office building or hotel, the plumber is likely to be classified as an employee because the buyer of the services has more “control” (or labor market power) over them (Posner 2020b).

In recent years, some employers have evaded the work restrictions imposed by employment regulations by classifying employees as independent contractors. Consider, for example, the rise of ride-sharing companies, which compete with taxi and limo companies. When taxi and limo companies organize as employers, their drivers are treated similarly, and this means that a driver will not see much difference between working for employer A and for employer B. In contrast, an independent contractor could be given insurance by company A and not by company B. This means that the independent contractor, while legally treated as independent of the ride-sharing companies, may actually be more constrained in their ability to move from one to another.

Thus, companies might be able to gain market power over workers if the independent contractor rules are not enforced with sufficient strictness, but at the same time, relaxation of the independent contractor rules might also give workers flexibility that they value. How these factors balance out is a complex empirical question. Nonetheless, recent efforts to restrict abuse of the independent contract rule, in California and elsewhere, seem appropriate.

H. Support for Unions

Workers have historically turned to union organization to counter the labor market power of employers. Unionization deprives the employer of its main source of market power: the ability to set wages and working conditions. If the employer does so, the union can strike, and the threat of the strike should deter wage suppression in the first place. However, unions are fragile organizations. They must maintain discipline among members, and employers can bust unions by countering those disciplinary efforts.

Governments can counter wage suppression by providing legal protections for and subsidies to unions. This strategy has been pursued in many countries. In the United States, the law prohibits employers from engaging in various types of union-busting activities, including bribery of workers, intimidation, the creation of company unions, and much else. The law also regulates union elections, collective bargaining, and work stoppages. These regulations limit fraud and coercion, enhance transparency, and encourage peaceful negotiation and collective action.

Unions operating in monopsonistic labor markets also generate spillovers to other, nonunion, workers, without any threat effects. This is because union density raises wages for unionized employers, and nonunion employers must raise their wages to compete for workers.

Collective bargaining by unions may allow contracting to overcome a lack of competition. In an extreme example, where the marginal product of labor is constant, transferring monopoly power to workers can be efficient. Rather than wages being distorted downwards by monopsony, resulting in too few workers, the union will set the wage equal to marginal product (having no reason to set it higher than that, as then the firm exits). In a more realistic case with diminishing marginal product of labor, the choice between laissez-faire monopsony and union monopoly will depend on the elasticity of labor demand versus labor supply. Crucially, however, whether or not efficiency is enhanced by firm-level unions depends on whether employment increases under the union. Existing evidence is mixed; Frandsen (2013) suggests employment increases for low-skill workers and decreases for high-skill workers.

When the source of monopsony is job differentiation, either due to specific skills or idiosyncratic tastes, unions may have a further role in reducing monopsony power by facilitating contracting, overcoming asymmetric information and commitment failures. A union may use nonpecuniary rewards to facilitate first-degree price discrimination and then use the wage to split the resulting (first-best) surplus.

Despite the legal protections they have been given, unions have lost ground in the United States over the last 50 years. There are many reasons for this, including technological change and globalization. Employers have developed more sophisticated union-busting strategies (Schmitt and Zipperer 2009). Workers have become increasingly isolated from each other as a result of broad economic trends, and this isolation interferes with organization. Right-to-work laws at the state level have further weakened union discipline by allowing workers to free-ride on the collective bargaining efforts of the union leadership. Many employers have taken advantage of legal forms that allow them to classify workers as independent contractors (as discussed above) or fragment their workforce by operating through independent franchisees. General economic changes have also apparently created more highly differentiated jobs, which further interferes with organization, and supplies employers with an independent source of market power.

I. Shareholder Activism/Codetermination

Monopsony implies that a component of firm profit is rents from underpriced labor. This profit then accrues to shareholders. But what if at least some of these shareholders are workers themselves?

Suppose workers have a portion a of the firm’s shares, and managers weight shareholders’ interests according to shares held. The objective function of the firm then becomes:

Thus, the monopsony incentive to distort the wage downwards is mitigated to the extent that the firm’s manager internalizes the wage bill. Of course, the equation above does not effectively capture the true objective of the (partially) labor-managed firm: if the labor supply is the extensive margin (number of workers) the question of how many workers a worker-owned firm would want needs to be answered, and that depends on how profits are shared and the extent of diminishing returns, as in the Ward (1958) model of the worker-owned firm under market socialism (see also Dow 2003).

A prominent example of employee-ownership is employee-stock ownership plans (ESOPs), which are the subject of an extensive economics literature, much of it using the NBER Shared Capitalism database. Roughly 20 percent of private-sector U.S. workers own some of the stock their company. Kruse (2016) surveys this literature and finds that employee ownership is linked to better company performance. Part of the mechanism (besides higher compensation, effort and less conflict) is lower turnover and absenteeism, which suggests that employee-owned firms are moving up the labor supply curve facing the firm. However, cleanly identified causal effects of employee ownership are still missing from the literature.

Union pension funds have been used successfully to alter corporate governance practices, particularly around labor relations. By organizing shareholders around worker interests and mobilizing proxy votes, union pension funds are able to influence a variety of firm decisions.

But most pension funds, wanting diversification of risk, would likely invest only a small share of their savings in the firms that employ their members. And workers’ holdings may be small relative to holdings of other investors, so a is small. Even if a were large, firm managers might be required, by fiduciary duty, to exploit monopsony power even against some of their owners.

By asking its managers to raise wages in monopsony, the pension fund would (i) lose some value in profits, but (ii) increase contributions and members. Depending on the degree of monopsony, exposure to the firm, and the extent of contributions of workers, the value to the pension fund of the latter could offset the costs from the loss of value in profits.

One interesting case is public-sector union pension funds, for example, CalPERS. If we take literally the idea that these funds should maximize the returns to their members, then it may be sometimes appropriate for these funds to demand that monopsonistic firms raise wages. Higher wages benefit the workers more than their lost capital gains. Tax revenues should also increase because the tax rate on wages is higher than the tax rate on capital gains—although the problem is complicated because the public goods funded by these tax gains will benefit people other than the fund’s members, and the taxes paid on capital gains will mostly be paid by nonmembers as well. But to the extent that the tax bill increases, and to the extent that public-sector union members get higher wages from additional tax revenue, public-sector pension funds may have a pecuniary interest in requiring their holdings to raise wages.

Another way to induce a positive a is via worker codetermination, which would not require ownership of firm shares. Instead workers get votes on firm policies, including wage-setting policies. To the extent that workers’ votes count, this will influence firm wage-setting and mitigate the exercise of monopsony power.

J. Macroeconomic Considerations

A considerable amount of theoretical and quantitative research has gone into the cyclical implications of job search models with bargaining, but comparatively much less has gone into variants of search models that feature monopsonistic wage-setting.

The job search model implies that the residual labor elasticity will fall during economic downturns because workers have more trouble finding new jobs. Depew and Sorensen (2013) and Webber (2015) both find evidence for this hypothesis. Tight labor markets are also more competitive. Thus, countercyclical macroeconomic policy that successfully minimizes job loss during downturns will have broadly positive effects on labor market efficiency and wage levels.

While more research is necessary, it is intriguing to consider monopsony as the proximate mechanism behind the famous “wage curve” widely used in macro, where wages and unemployment exhibit a negative relationship. Models like the Mortensen–Pissarides model or the Shapiro–Stiglitz model deliver wage curves because of how unemployment lowers the outside options of workers. A monopsonistic variant would suggest that a low residual elasticity (perhaps proxied by a low voluntary quit rate) is the mechanism that transmits high unemployment into low wages, not just outside options.

Given the recent staggering increase in unemployment due to COVID-19, even as labor demand for essential workers has increased, the relationship between macroeconomic slack and labor market competition is potentially even more salient. Lack of labor market competition may be one reason compensating wage differentials for essential workers facing increased hazards have not materialized (Cajner et al. 2020).

K. Who Enforces?

All of these proposals raise a cross-cutting question of agency costs. Many employment regulations are enforced by federal agencies, including the Occupational Safety and Health Administration (OSHA) and the Equal Opportunity Employment Commission. Others, like minimum wage laws, are enforced by state and local governments, as well as by the federal government. In all these cases, the government serves as an agent of the workers. Union leadership, too, serves as an agent for the members of the union. And in our pension example, one might think of the pension as serving as an agent for workers, though indirectly. Many employment laws are enforced by private litigation, and because of the high cost of litigation, usually in the form of class actions, where private lawyers serve as agents for classes of workers. These class actions can be thought of as contingent unions that spring into existence to enforce the law. All of these approaches raise questions about whether the agent actually has the interest of the workers at heart.

Worries about agency costs have led in many cases to a round of metaregulation. The law requires unions to act in the interest of workers, and the same is true for class action lawyers. But we might also wonder whether the government agents charged with enforcing the law will act in the interest of workers, who often have little political power. The old union movement was based in part on suspicion about government responsiveness to the interests of workers, and such concerns continue to be aired today.

Pervasive monopsony, together with employment at will, also means that whistleblower mechanisms for enforcement (like OSHA) will be weaker in labor markets than in other markets. Under monopsony, inframarginal workers are getting rents from their job and thus would be afraid of termination or forgone pay increases. Employer retaliation is thus costly for whistleblowers and difficult to prevent.