Competition policy and the labor share

Abstract

Recent years have seen intense debate about the causes of the observed decline in the labor share. We extend this inquiry by investigating whether the design and enforcement of competition law and policy are associated with changes in the labor share. Using a panel of 22 industries in 12 OECD economies, we find a positive statistical association between the effectiveness of competition policy and changes in the labor share over the period 1995–2005. This suggests a potential link between the design and effectiveness of competition policy and the labor share, and more broadly to distributional outcomes. Our results reinforce the importance of accounting for country-specific factors, including the design and enforcement of local laws, when examining dynamics in the labor share. The analysis implies that effective competition law and policy could mitigate the decline of the labor share, particularly in settings characterized by low levels of labor protection and limited labor bargaining power. (JEL: E21, E24, E64, J01, J21, K21, L40).

1. INTRODUCTION

The decline of the labor share in recent decades is well documented and widely acknowledged. However, the rate of change has varied across countries, a fact that has given rise to questions about the factors contributing to this decline (Gutiérrez and Piton 2020). A range of possible, and at times inconsistent, explanations for the decline in the labor share are presented in the literature (Grossman and Oberfield 2021). Broadly, some argue that the decline in the labor share is a global phenomenon reflecting structural change, technological advances (Acemoglu and Restrepo 2020), and changes in global trading patterns (Elsby et al. 2013). In contrast, others argue that the decline in the labor share cannot be generalized in this way and also reflects idiosyncratic local factors such as national and regional differences in labor regulation (Ciminelli et al. 2022), and product market competition (Gutiérrez and Philippon 2018). One prominent explanation is that the decline of the labor share is partly a consequence of an increase in market power over the same period (De Loecker et al. 2020). According to this view, lower levels of competition intensity have allowed firms to gain market power and set high consumer prices relative to costs (i.e., markups or monopoly rents). These rents, however, have not been equally shared between capital and labor, with a greater proportion of the rents being captured by capital vis-à-vis labor. This particular explanation has gained traction with policymakers and subsequently focused attention on the way in which changes in market power might account for declines in the labor share in specific countries.

While studies that show a connection between increases in market power and the decline in the labor share are compelling, they typically assume that market power is exogenously determined as a result of technological change or specific structural conditions. Accordingly, laws or other policies directed at constraining market power or increasing the bargaining position of labor are not factored into the analysis. In addition, these studies adopt different approaches to estimating market power, including different methodologies for estimating markups.

In this article, we investigate whether, and how, competition policy might interact with the labor share, given its ability to affect levels of concentration and market power. Introducing this aspect to the analysis is important for three reasons.

First, our inquiry introduces a new focus on how the effectiveness of competition policies in controlling market power may have contributed to changes in the labor share. Our analysis captures the variation within a country’s competition policy, allowing us to investigate how various aspects of the design and implementation of a country’s competition policy—for example, benchmarks for appraising merger transactions; use of threshold for “dominance” versus “monopolization”; choices about the enforcement of competition law with respect to mergers and abuse of dominance (e.g., killer acquisitions and approach to big tech); and decisions about the resources and independence of bodies charged with enforcing competition law—may have been more or less successful in controlling market power and thus contributed to changes in the labor share.

Second, it allows us to test whether changes in the labor share partly reflect domestic or local policy choices about the type and nature of competition law and labor protections, or whether the changes in the labor share are largely invariant to these policies and are driven by global factors. Specifically, we investigate two local legal instruments that could affect the level of market power and the labor share: competition law and labor protection laws (such as a minimum wage). These policies vary both between countries in our sample, and within countries, over time.

Third, by directly comparing how changes in competition policies relate to changes in the labor share, our main results are not contingent on how proxies for market power are estimated (such as markups) which can be difficult and contested. Although we do use estimates of “competitive margins” as part of the interaction analysis, this is not the focus of our investigation. Rather the competitive margins estimates are only used to examine whether different levels of labor protection interact with competition policy in shaping changes in the market outcomes.

To capture nuances in a country’s competition policy, and changes over time, we use the competition policy index (CPI; Buccirossi et al. 2011). We apply this index (as well as other variables) to a panel dataset of 22 industries nested in 12 OECD countries, to test whether changes in competition policy were associated with changes in the labor share over the period 1995–2005.

Overall, our findings provide support for the conclusions of previous studies on the decline in the labor share and its relationship with increases in market power by introducing another dimension to the debate. Our results imply that effective competition policy (as measured by the CPI) may have contributed to the rise of the labor share observed in some countries in our sample. In contrast, ineffective competition policy could have had the opposite effect (i.e., suggesting a decline in competition and the labor share). The results also highlight the importance of accounting for a range of local factors, particularly laws and institutions that shape and condition the levels of labor protection and labor bargaining power, when investigating changes in the labor share.

This article comprises eight sections. Section 2 summarizes the findings of relevant research on market power and the decline in the labor share, and on competition policy and market power. Section 3 describes our dataset and econometric specifications. Section 4 describes the main results from our analysis of competition policy and the labor share, while Section 5 discusses the results of models which account for differences in labor protections and institutions, focusing on their association with competitive margins. Section 6 summarizes the results of the sensitivity analysis. Section 7 considers the policy implications of our findings while Section 8 concludes. The  Appendix includes extensions such as an instrumental variable (IV) model, and sensitivity and robustness checks that use alternative model specifications and proxy measures for the level of competition.

2. MARKET POWER, THE LABOR SHARE, AND COMPETITION POLICY

Two strands of research are most relevant to our investigation. The first strand is the literature examining the causes of the decline in the labor share, and in particular the role of increasing market power as a potential explanation for that decline. The second strand of research explores whether the design and implementation of competition law and policy can affect market outcomes (e.g., in terms of market concentration, firms’ productivity, profits, and markups).

2.1 Rising markups and declining labor shares

Previous research on the decline of the labor share has identified various factors as potential causes for its global decline, including globalization and technological change (Acemoglu 2003; Azmat et al. 2012; Bassanini and Manfredi 2012; Karabarbounis and Neiman 2014; Piketty and Zucman 2014).

Within this work, the most relevant for our purposes are studies on rising markups (De Loecker and Eeckhout 2018) that find that the labor share is inversely proportional to markups and negatively correlated to profit rates and market concentration (The Herfindahl-Hirschman Index (HHI)) measures (Barkai 2016; De Loecker et al. 2020) in the United States and globally (De Loecker and Eeckhout 2018). Closely related is the work by Autor et al. (2020), who argue that the observed rise in measures of markups is a result of faster innovation of “superstar” firms like Google or Amazon reacting to changes in the product market environment, for example, globalization and technological changes. They suggest that monopolistic competition (“winner takes most”), and not necessarily market power, has contributed to the labor share decline. While they posit that the rise of superstar firms is a global phenomenon (contradicting the claim that it is relevant for US antitrust only), their analysis does not rule out a possible role of competition policy. Furthermore, they call for future work to analyze more precisely the economic and regulatory forces that lead to the emergence of superstar firms.

Some argue that these two macroeconomic trends, the rise of market power and the decline in the labor share, are not common to all developed countries. Specifically, issues of measurement of the labor share, like the ownership of housing (Rognlie 2016) and revenue of self-employed (Gollin 2002), might challenge the emerging consensus on the global decline in labor share. Similar issues of measurement error might challenge the observation of rising market power (Lipsius 2018; Traina 2018; Basu 2019; Cette et al. 2019).

A recent study by Gutiérrez and Piton (2020) directly contests the observation of global decline in labor shares and suggests that after adjusting for cross-country measurement issues, the US and Canadian economies are the only developed economies that exhibit a clear decreasing trend since the middle of the 1980s. This study puts greater focus on idiosyncratic country-specific explanations (i.e., local factors) for these macroeconomic trends. It suggests that reduced competition intensity in selected US industries might have led to rising profits, directly linking changes in antitrust laws and institutional drift, an observation supported by other research (Gutiérrez and Philippon 2018; Zac 2022).

Debates that associate the decline in labor share with market power in some, but not all, countries are of special interest to us. However, these studies often make particular assumptions about the market and institutional (legal) environment, and as such do not incorporate potentially relevant empirical factors such as the role of competition policy and other local factors which shape market dynamics.

2.2 Competition policy: empirical findings

Only a few studies have sought to empirically investigate the link between the design and implementation of competition law and policy and market outcomes (e.g., market concentration, firms’ productivity, profit, or markups). Gutiérrez and Philippon (2018, 2020) find that political and structural independence of competition institutions and other economic institutions like central banks can explain why EU markets have become more competitive than the US markets, in relative terms.¹  Affeldt et al. (2021) find evidence that stricter merger enforcement (proxied by the percentage of mergers blocked or remedied) is predominantly negatively related to concentration levels; implying that more active merger regulation enforcement is linked to higher levels of competition.²  Besley et al. (2021) study a large sample of countries over 10 years, using firm-level data and a proxy for the scope of competition law (“law of the books”). Their study compares firms operating in tradable markets (i.e., those sectors that are subject to international competition, namely agriculture, mining, and quarrying and manufacturing) with firms in non-tradable markets (i.e., all others). They find that profit margins of firms operating in non-tradable sectors are significantly lower in countries with wider competition policies (in terms of scope) compared with firms operating in tradable sectors.

Putting these two strands of research together one could reasonably argue that more effective competition law and policy should be associated with higher levels of competition which in turn could decrease markups and profits and increase the labor share. In this article, we test this hypothesis directly using data on industry level labor shares and country level implementation of competition policy.

3. DATA AND METHODS

This section describes the data used in the various model specifications. The country–industry-level dataset was compiled by combining data from different sources: the EU-KLEMS and the OECD database for 12 OECD countries—Canada, Czech Republic, France, Germany, Hungary, Italy, Japan, Netherlands, Spain, Sweden, United Kingdom, United States—in the period between 1995 and 2005.³  Table A1 in the  Appendix lists the 22 industries on which we perform our analysis.

The policy variables, which do not vary between industries, are recorded at country-year level. Importantly, they capture key changes in these countries’ competition policy during the period. For some countries, this includes fundamental changes such as the introduction of the Enterprise Act in the United Kingdom and the Dutch Competition Act (Mededingingswet). For other countries, changes captured reflect variations in the way competition policy is applied, changes in enforcement inputs (budget and staff) and outputs (enforcement activities), as well as institutional independence. Our sample includes countries from North America, Europe, Japan, and two ex-communist economies (Hungary and Czech Republic). The heterogeneity in the data is of value and allows us to capture differences in the goals and “philosophies” which underpin the national competition policy, as well as differences in the specific drafting of the laws. Overall, the variation between countries, and within countries over time, allows us to explore whether there is a link between the design and enforcement of a country’s competition policy and changes in the labor share.

3.1 Description of the variables

3.1.1 The competition policy index

The CPI was compiled by Buccirossi et al. (2011) as a measure of the quality of competition policy.⁴ The index includes the data for 12 developed countries over the time span 1995–2005 as presented in Figure 1. The CPI is an aggregated index of a sub-set of indices which are calculated covering two main areas of study: competition policy infringements (including hard-core cartels, anti-competitive agreements, and abuse of dominance) and merger control policy. The aggregate CPI index is constructed using a “pyramidal” approach from “low-level” indicators, through “medium-level” sub-indices, to “high-level” indices. This is a key advantage of the CPI compared with other classic “scope” or “law of the books” measures (Bradford and Chilton 2018). At the lowest level, data were gathered on each area of policy under five categories: independence of enforcement; separation of powers; quality of law; powers of investigation; sanctions policy and damages (the level of the overall loss that can be imposed on firms and their employees); resources (budget, staff) and in the case of hard-core cartel and mergers enforcement activity the number of cases opened or reviewed. Hence, the CPI reflects the nuances of competition policy, allowing for a comparison of the relative effectiveness of such policies.⁵

As can be seen in Figure 1, some countries show large variation in the CPI over time, notably: the Czech Republic, Hungary, Netherlands, and United Kingdom. In contrast, other countries—such as Italy or Germany—exhibit a flat trend, while Sweden shows a negative trend in the CPI. These last countries are still relevant to our investigation because they allow us to test whether no changes in the levels of the labor share (or a decline of the labor share) are associated with a flat or negative trend in the CPI.

3.1.2 The labor share

The labor share (Labor share) is defined as the ratio between the gross total labor compensation (including wages and the additional costs of employing labor) and the gross total value added (both components are in nominal terms). There are two potential limitations of the available data on labor share which could potentially limit cross-country comparisons and should be acknowledged.

First, labor compensation for the self-employed is not recorded in the national accounts. Yet, our data on self-employed labor compensation and value added are retrieved from the EU-KLEMS database, which applies a standard imputation by assuming that the hourly compensation of the self-employed equals the hourly compensation of employees,⁶ and therefore allows for cross-country comparisons of the labor share.⁷ A second issue that can affect cross-country comparisons is the inclusion or omission of housing data in some national accounts. To mitigate this issue, we have excluded the housing sector from our sample of 22 industries in the 12 countries we examine, but we still retain data for other non-manufacturing sectors.⁸

3.1.3 Other variables

The first group of variables we use are non-legal, economic factors, at country–industry level. They include some factors that were found to be associated with changes in the labor share in previous studies. These include: trade liberalization and globalization (tradelib) which is a ratio of industry imports over value added; high skilled workers (HHS) and medium skilled workers (HMS) to capture the degree of human capital accumulation/education and the changing skill composition of the labor; labor compensation (wagegrowth), as reductions in the salaries of workers across industries tend to decrease the labor share (Arpaia et al. 2009; Ennis et al. 2019); business cycles (industry_trend), which allows us to account for different demand fluctuations; and R&D intensity as a proxy for technological progress (R&D), measured as the ratio between R&D expenditure and the value added.

The second group of variables are legal factors at the country level. These variables are used to control for other product market regulations, competition policy metrics, and labor institutions that could also impact the labor share. These include: a product market regulation indicator (prod_mkt_reg); a scope of competition law index (CLI), and text similarity proxies which examine the extent to which various national laws replicate the language used in the US and EU competition laws using human coding (US_text or EU_text)⁹; and a variable (empl.protec) which captures the strictness of employment protection and data on unionization.

In Table 1, we present the descriptive statistics of our main variables, based on the sample size used in the main analysis (Table 2). In the  Appendix, we include a full list of variables (Table A2) and sources.

3.2 Econometric specifications

Overall, we expect a positive relationship between competition policy (CPI) and the labor share (Labor share) with the coefficient E(⁠$βCPI)$ > 0. In Equation (1), “i” refers to country, “j” to industry, and “t” to year. As described in Section 3.1, we include a set of country–industry controls $Xi,j,t-1$ related to the main dependent variable (such as human capital and import penetration), while $Zi,t-1$ captures country-specific controls (such as product market regulation and scope of law). Following Buccirossi et al. (2013), we model the unobserved heterogeneity through the error term’s structure: $εj,i,t=ψi,j+φt+ui,j,t$⁠. Specifically, we control for country–industry fixed effects $ψi,j$ to capture a variety of specific time-invariant unobservable factors, and year fixed effects $φt$⁠, to consider macroeconomic shocks that might affect the labor share (Labor share) in all countries at the same time. Lastly, we include the residual standard component of the error term $ui,j,t$⁠. We cluster the standard errors at country level (as our main independent variable CPI), to allow for correlation across industries within the same country (and for the presence of possible spillovers).¹⁰

To limit the potential bias from reverse causality, we use the standard approach of lagging the control variables by one period with respect to the labor share indicator (Griffith et al. 2004; Buccirossi et al. 2013), based on the assumption that lagged values are uncorrelated with the error terms of the equation. While we do not suspect the presence of two-way causality bias in our analysis, we prefer to adopt a conservative approach.¹¹

The main source of endogeneity which might affect our analysis is omitted variable bias. While we control for the time-invariant unobserved heterogeneity, by using the wide set of fixed effects in our estimation, there might be other variables and policies that may affect the CPI and the labor share and are not captured in our baseline model. Specifically, an index like the CPI, might capture the effect of the more general quality of economic institutions in a country such as central banks (Gutiérrez and Philippon 2018). To address this issue, we implement an IV estimation strategy to make the analysis more robust and reliable. The results are included in the Table A4.

We employed an interaction analysis to allow for greater variation in our CPI measure. Specifically, we examine the marginal effect of competition policy on the labor share, by accounting for: (i) the type of industry (services versus manufacturing); (ii) the scope of competition policy (i.e., the “size of the net” capturing anti-competitive behavior); and (iii) the legal background or model of law (i.e., countries that adopt the US antitrust laws’ text compared with the EU competition law text). Our last interaction model offers an additional contribution to the debate on whether movements in the labor share are global or partly reflect local factors and could help explain the different results observed in the literature for North America (United States and Canada) compared with other developed countries.

Lastly, we also explore the effects of competition policy against the background of labor protection levels in a separate reduced model, as we believe that both factors and their interplay can influence the level of competition intensity and ultimately the labor share.

4. RESULTS

4.1 Main results

Table 2 presents the results for the baseline model, with different sets of fixed effects. The positive sign and the statistical significance of the CPI are consistent across all specifications. We started our analysis with the univariate model and gradually included all the controls in the full specification (see Column 5). The coefficient of the (lagged) CPI is always positive and statistically significant, suggesting that more effective competition policy is positively linked to the labor share. As expected, given that we account for several fixed effects in our model (allowing us to control for time-invariant unobserved heterogeneity at country–industry level), the (adjusted) R-squared takes high values in Columns 3–5. This means that specific country–industry characteristics already explain much of the variance of our outcome variable. Table A5 reports the standardized betas to aid the comparison in terms of magnitude of the effect across the controls. One SD change in the CPI is associated with a statistically significant increase of 0.087 SD in the labor share. Although the CPI does not have the strongest effect on the labor share (high skilled human capital [HHS] is the highest), it is also not the weakest (the effect of the CPI on the labor share is higher than the effect of trade liberalization in our sample).

With respect to control variables in the baseline model (Table A3), the negative coefficient for the ind_trend variable suggests that economic fluctuations and business cycle effects (deviations from the predicted levels of production) tend to be negatively associated to the country–industry labor share (see Growiec at al. (2018) for the counter-cyclicality of the labor share in the short-run).¹² As expected, the coefficient for wage growth on the labor share is positive and highly statistically significant (ranging from 0.103 to 0.116). The rest of the controls are statistically insignificant as we elaborate on in the  Appendix, where we present different specifications by including each control at a time. Given the relevance and possible policy implications, we further investigate different underlying settings of labor markets in Section 5.

Overall, the findings of the OLS model suggest that effective competition policy is positively linked to the labor share. As set out in the  Appendix, this result is confirmed using an IV approach and is robust to a range of sensitivity tests.

4.2 Interaction analysis

We make use of the heterogeneity in our data and investigate the interaction between the CPI and other potentially endogenous factors using an interaction analysis. This analysis allows us to explore how the scope of competition policy, and the degree of industry-specific regulation, might affect the relationship between the CPI and the labor share. Table 3 presents the results for three different interaction specifications.

4.2.1 Type of industry

The first interaction specification tests whether the relationship between the CPI and the labor share differs according to the type of industry (Column 1), namely manufacturing versus services. The results presented in Column 1 of Table 3 suggest a stronger and more statistically significant relationship between competition policy and the labor share in capital-intensive industries (e.g., manufacturing), compared with service industries.¹³ These results contrast with the findings of Besley et al. (2021) who differentiate between tradable sectors (i.e., manufacturing) and less-tradable sectors (i.e., services) and find that the “scope” of competition law only affects profitability in the latter case. However, a direct comparison of the results of the two studies and others (e.g., Affeldt et al. 2021) is limited by the fact that the studies use: different proxies for competition law and competition intensity; are conducted at different levels of aggregation; and examine different countries and time frames.¹⁴ Additional work using consistent measures and datasets is needed to shed light on these findings.

4.2.2 “Scope” of competition policy

Column 2 of Table 3 presents the results of the second interaction analysis which focuses on the relationship between the “scope” of competition policy and the labor share. We test the hypothesis that the greater the scope of competition policy, the greater the impact of the CPI (which captures competition policy effectiveness) on the labor share. As a proxy for the scope of competition policy, we use the CLI, retrieved from the dataset of Bradford and Chilton (2018). As the CPI already reflects an element of the scope of competition policy, we expect the CLI and the CPI to correlate to a certain extent, but the correlation is small and negative.¹⁵ The CLI is limited to scope alone (presence of competition policy) and reflects the letter of that law. By contrast to the CPI, this more formal index does not take account of the actual level of enforcement or other factors in the policy. This limitation of the CLI should be noted, as some jurisdictions may have enacted the laws but engage in limited effective enforcement. Other jurisdictions may have gone through a significant policy change which affected the intensity of enforcement, without changing their laws. Accordingly, measures of scope could fail to capture meaningful differences between developed countries that already have a more harmonized scope of law including articles for mergers, cartels, and abuse of dominance.

The results shown in Column 2 of Table 3 are in line with our hypothesis that the greater the scope of competition policy the greater the impact of the CPI on labor share. The overall effect of both indexes of competition law is positive and so is their interaction effect, captured by the coefficient of the term CPI_t−1* CLI_t−1. Figure 2 shows the predicted marginal effects of the CPI on the labor share (on the y-axis), for fixed values of the CLI (on the x-axis). As shown, the marginal effect of the CPI on the dependent variable becomes gradually bigger as the CLI increases.

4.2.3 Similarity in legal text (“model of law”)

Column 3 in Table 3 shows that the interaction between EU countries (or countries whose competition laws closely resemble the EU legal text) and the CPI is positive and statistically significant. Theoretically, under the pure global argument, we should not expect to see heterogeneous effects for competition law text and no significant cross-country differences should be observable. In other words, if the claim that market power is on the rise merely due to global factors is robust (Diez et al. 2020), then competition laws have likely played a passive part in these two common trends, that is, decline of the labor share and the reduction in the level of competition (i.e., rise of markups). However, if market power is also affected by local factors, then the rise in concentration levels observed in previous studies could be explained by a combination of innovation and efficiencies (the superstar firms hypothesis or other global forces), as well as ineffective competition policy (Gutiérrez and Philippon 2018; Covarrubias et al. 2019).¹⁶ In this case, it is possible that some local competition law enforcers did slightly “better” than others in controlling market power and anti-competitive practices. In our small sample, the countries that experienced a positive change in the CPI (United Kingdom, Netherlands, Hungary, and the Czech Republic) were all in Europe, while the “US similar” countries (like the United States, Canada, and Japan) experienced a stagnant CPI in the period examined (see Figure 1). Our results are, therefore, more in-line with the local (or mixed) explanations for changes in the labor share and can partly explain why some countries seem to experience stronger declines in the labor share than others. As we only observe a decade (1995–2005), we cannot contribute to the debate over the long-term trends of markups and the labor share.

5. COMPETITION LAW AND LABOR PROTECTION AND INSTITUTIONS

An important interaction we set out to explore concerns the link between a country’s competition law and labor protection laws and institutions. To explore how these two laws jointly associate with market outcomes, we empirically investigate the link between both laws and a proxy for the level of competition. Our analysis in this section is also consistent with a markups (i.e., market power) mechanism, showing that the CPI is negatively correlated with our proxy of competition.

Note that the capital costs measure is not a residual of the labor share but rather is estimated by multiplying the capital stock for the user cost of capital. This approach takes into account the real interest rate and the extent of capital depreciation (Griffith et al. 2007; Buccirossi et al. 2013).¹⁹ The “competitive margins” measure we use tends to closely approximate markups (or the price cost margin such as the Lerner Index) whenever average costs closely match marginal costs. Indeed, the major limitation of this approach lies in the assumption of constant returns to scale. Whenever this assumption does not hold, the proxy will be downward (upward) biased with increasing (decreasing) returns to scale. However, by controlling for year and country fixed effects, we mitigate this possible concern (Griffith et al. 2007). While this approach has limitations (including that marginal costs are not observed), it has the advantage of being easily implementable and data driven.²⁰ As our data are aggregated at the industry level, we need not go into difficulties of estimating production functions (Traina 2018).

As a robustness check, we compare this measure to other publicly available datasets including the recent estimates of the OECD for aggregated country markups. This can be seen in Figure 3, where we plot our estimates of the “competitive margins” (defined in Equation (2)) against recent work on country-level markups from Schreyer and Zinni (2020).²¹

In our sample, the average value of the competitive margins proxy is 1.23 and is consistent with the findings of Schreyer and Zinni (2020), who find an average markup of 1.25. This is also in line with Christopoulou and Vermeulen (2012) (who find an average of 1.37 for European countries and 1.32 for the United States) and Loecker and Warzynski (2012), using firm-level data of Slovenian manufacturing firms, obtaining markups in the range of 1.17–1.28. De Loecker et al. (2020) find average markups in the United States increased from 1.31 in 1986 to 1.45 in 2006, which is relatively close to our time-series range of values. While the competitive margins proxy we use is broadly consistent with other estimated markups, it should nevertheless be interpreted with care, given the discussed limitations.²²

We followed a similar econometric specification to Equation (1) using the competitive margins proxy. The only difference is the inclusion of several economic factors identified in previous studies as being co-founding factors next to competition policy.²³

Besides these standard economic controls, we also include the CPI and three additional policy proxies to capture local variations in laws and institutions. These are product market regulations (prod_mkt_reg) and labor market policies and institutions in the form of employment protection, trade union density (t. union), and number of union members (union m.). We expect the CPI to be negatively correlated with competitive margins, as it should reduce distortions in the market. As for the product market regulation, increasing levels of prod_mkt_reg indicate a higher economic burden in the form of high barriers to entry, which are expected to be positively associated to profits (Autor et al. 2017) and market distortions.²⁴ The relationship between labor market regulation and institutions and competition is more difficult to predict. On the one hand, stricter labor market regulation and institutions could bolster the ability for labor to negotiate which could raise costs and reduce markups and profits. On the other hand, in some industries this might induce firms to substitute labor with capital, hence widening the price–cost margins (assuming that attributable capital costs are lower than labor costs).

To account for possible heterogeneity in the relationship between the CPI and the competitive margins in different labor market settings, we split our sample according to whether a particular labor market regulation or institution is below or above the median, indicating, respectively, less and more regulated labor markets. This method has advantages in dealing with possible outliers or multiple distributions (Bailey and Goodman-Bacon 2015). The central hypothesis is that the CPI might impact competition differently, depending on the strength of labor regulation in place.²⁵  Table 4 reports the results. The coefficient of the (lagged) CPI is negative and statistically significant at 1% level across all specifications.

With the exception of Columns 1 and 2 (that model the variation in the level of employment protection), the coefficient for the CPI drops dramatically in the above median sub-samples (Columns 4 and 6), which account for the relative strength of labor bargaining power (e.g., low versus high level of trade unions and low versus high level of union membership). As discussed further in Section 7, this could suggest that competition policy may be more important in environments characterized by limited labor protection and labor bargaining power.

6. SENSITIVITY ANALYSIS

The results of testing the baseline model using an IV approach are contained in  Appendix A.1, next to several sensitivity tests. We summarize here the main tests and their results: IV, CPI sensitivity, sample sensitivity, clustering, and modelling the error term’s structure.

First, we applied an IV model to control for endogeneity as a direct consequence of potential omitted variable bias, for the main results in Tables 3 and 4. We followed a similar approach to that of Buccirossi et al. (2013) and used as an instrument the average of the CPI in neighboring countries. The logic is straightforward: while it is likely that competition policy in each country may be influenced by the policy of neighboring countries over time, there is no reason to believe that the latter should substantially affect the labor share dynamics of a neighboring country (especially in the studied period, where the global convergence of competition law and markets was only beginning). Results are confirmed and provide us with further reassurance about the validity of the main analysis (Table A4).

Second, the aggregate CPI is constructed according to the discretionary weights chosen by Buccirossi et al. (2011) which might affect our overall estimation in a systematic manner. We therefore re-ran the baseline model using one of the alternative weighting schemes for the CPI, which implements a factor analysis. Our results are confirmed (Table A5).

Third, to address the risk that potential outliers might be skewing our results, we ran sample sensitivity tests by removing four pairs of country observations (keeping 10 countries in our data). For historical-development reasons, we removed Hungary and the Czech Republic on the basis that they were in the process of transitioning toward market-based economy in the period we examine (1995–2005) and thus might bias our estimation upward. Similarly, we removed the top CPI values in our sample (treating them as outliers), that are associated with Sweden and the United States, both highly developed economies at the time. The results are consistent with the main findings and presented in the  Appendix  (Tables A8 and A9).

As described above, to test whether our results hold using other estimates of the labor share we ran our baseline model (in Table 2) excluding the United States and Canada (Table A10) and again without Japan and the Netherlands (Table A11). Our results are confirmed when we exclude these countries. The exclusion of the United States and Canada from our sample shows that our findings are not affected by the cross-country labor share estimation issues identified in Gutiérrez and Piton (2020). The exclusion of the Netherlands and Japan serves to test the possible difference between our measures of competitive margins and the markups extracted by Schreyer and Zinni (2020), which as shown in Figure A1 (in the  Appendix) are mostly similar except for these two countries.

Finally, we presented our results based on specific assumptions about fixed effects and the error term structure based on previous work and standard practice. However, to test the robustness of our findings, we checked whether they are sensitive to the types of individual effects, the type of clustering, and an alternative structure of the error term (see Table A13). Therefore, we re-run the model by clustering at different levels of aggregation and by controlling for different fixed effects. Moreover, as the choice of the error term structure might still appear subjective, we re-run the OLS model by shaping the error term structure as an AR(1) model. Results are confirmed, both in size and statistical significance. This additional set of results further validates our analysis.

7. POLICY IMPLICATIONS

Our results suggest that the effectiveness of competition policy, as measured by the CPI, is positively associated with the labor share. These results hold over several specifications, using fixed effects, IV, and interaction models. Several points can be made about these results.

First, our findings support and expand on the conclusions of previous studies (Affeldt et al. 2021), by introducing another factor that can help explain changes in the labor share, and its association with changes in market power. Our results imply that the effectiveness competition policy (as measured by the CPI) may have contributed to the rise of the labor share observed in some countries in our sample, and therefore the ineffectiveness of competition policy could have had the opposite effects (i.e., suggesting a decline in competition and the labor share).

Second, our results suggest that the specific characteristics of a country’s competition policy (i.e., the scope of competition law [as measured by the CLI] and the text of competition law in a jurisdiction) matter, and reinforce the positive link between the competition policy and the labor share. In addition, our results suggest that choices about whether to model a country’s competition law on the EU or US text can have potential implications. These findings connect to the comparative literature on US antitrust, institutional differences, and appetite for intervention in the US, relative to EU competition law model (Zac 2022).

Third, the results highlight the importance of accounting for a range of local factors, particularly laws and institutions which shape and condition the levels of labor protection and labor bargaining power, when investigating changes in the labor share. The interaction between competition policy (which focuses on the product market prices), labor policies, and institutions (labor bargaining power and wage setting) illustrates how these forces jointly contribute to the competitive environment, labor market outcomes, and macroeconomic trends. When strong labor institutions are in place, they help compensate for limitations in competition policy, as employees can use their bargaining power in labor markets to demand higher wages. However, where the bargaining power of labor is weak, or there are inadequate employment protections, this provides scope for firms with market power to increase profits and markups by lowering wages, and the task falls to competition policy alone to constrain the implications of market power. In the context of inequality, this suggests that competition policy will have a greater role to play in reducing economic inequality in jurisdictions with weak labor laws or limited labor bargaining power.

While our results provide new insights into the factors that may have contributed to changes in the labor share in some jurisdictions, there are three main caveats to our analysis which should be borne in mind. First, the results only cover a single decade (1995–2005). While they are corroborated with other studies in that time horizon (Affeldt et al. 2021), we cannot exclude the possibility that relationship between competition policy and the labor share we have investigated changed after 2005. Second, while we consider the CPI to provide the most suitable available metric to measure competition policy (Ezrachi et al. 2022) we recognize that competition policy is, in fact, not merely a corrective tool for markets but, at the same time, it may shape or correlate with the quality of economic institutions in general (e.g. central banks). Such factors could act as omitted confounders if they are also connected with economic performance and labor market outcomes (Hartmann et al. 2017; Gutiérrez and Philippon 2020). Third, our analysis is based on aggregated industry data which do not allow us to account for labor shares among firms within the same industry. Recent studies by De Loecker et al. (2020) and Song et al. (2019) suggest that re-allocation between firms of markups and profits is key to understanding aggregated trends, suggesting more studies are needed to inform policy in this area.

8. CONCLUSIONS

We find that more effective competition policies, as measured by the CPI, were statistically associated with higher labor shares at country–industry level in 12 OECD sample of economies during the period 1995–2005. By focusing on how competition policy interacts with the labor share, our findings supplement existing work on the relationship between markups and the labor share. This positive relationship between competition policy and the labor share is robust to several econometric specifications, using IVs methods and sensitivity tests, while controlling for the standard confounders identified in previous studies.

Competition policy appears to be more strongly positively associated with the labor share in sectors not subject to heavy or sector-specific regulatory oversight (manufacturing sector versus services sector). In addition, the link between competition policy and the labor share is stronger in countries with a broader scope of the competition law, and countries that use an EU text of competition law. Finally, we find evidence that competition policy is negatively correlated with competitive margins and that the effects of competition law are reinforced when labor protection is weaker. Our findings contribute to studies that seek to better understand the observed changes in labor share, by identifying the effectiveness of competition policy as a relevant “local” factor that could explain some of the differences between developed countries, and some of the apparent inconsistencies in debates about the causes of the decline of the labor share.

Footnotes

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of Law, Economics, & Organization online.

Conflict of interest statement. None declared.

ACKNOWLEDGMENTS

This article forms part of a research project funded by a grant from the Leverhulme Trust (Award RPG-2019-270, ‘The Effects of Competition Law on Wealth Distribution’).

APPENDIX

This appendix comprise two sections. Section A includes the main extensions to our analysis such as the instrumental variable (IV) model and the standardizing betas. Section B includes sensitivity testing. Tables A1–A3 are elaborations of the tables in the main text.

APPENDIX A: EXTENSIONS

A.1 Endogeneity and omitted variable bias: labor share and CPI

We apply an IV model to control for endogeneity as a direct consequence of potential omitted variable bias. An IV is valid when it: (i) is uncorrelated with the error term (exogeneity); (ii) is correlated with the regressor of interest, namely the CPI in our case (relevance); and (iii) does not exert a direct effect on the dependent variable or it is excluded in the structural form (exclusion).

We follow a similar approach to that of Buccirossi et al. (2013) and use as an instrument the average of the CPI in neighboring countries, CPI_G. The logic is straightforward: while it is likely that competition policy in each country may be influenced by the policy of neighboring countries over time, there is no reason to believe that the latter should substantially affect the labor share dynamics of a neighboring country (especially in the studied period, where the global convergence of competition law and markets was only beginning). Results are confirmed and provide us with further reassurance about the validity of the main analysis.

While for historical reasons and similar institutional development, we might expect a positive correlation between the CPI in a given country and the average policy in similar/neighboring countries, this intuition is not confirmed in our model (see the first-stage regression). Indeed, it is not a priori clear whether these policies should be expected to positively correlate for the entire sample period.²⁶  Table A4 reports the first and the second stage of the IV and the coefficient of the CPI is similar to one obtained in the main results (Table 2).

A.2 Model with standardized coefficients

To better compare the magnitude of the impact among variables in a “unitless” way, we ran the baseline model (see specification 7 in Table A3) by standardizing the betas, as shown in Table A5. βStdX refers to the standardization of the independent variables only, while βStdXY presents the betas if the dependent variable is also standardized. One SD change in the CPI is associated with an increase of 0.087 SD in the labor share. In both types of the model (partially or fully standardized) the magnitude of the effect of the CPI is lower than other statistically significant controls, such as the high skilled human capital (HHS) or R&D, even though the last one is not statistically significant. However, the effect of the CPI is higher than the effect of trade liberalization, and broadly similar to that of medium skilled workers (HMS) and R&D.

A.3 Alternative CPI index and disaggregated components of the CPI

This section presents the set of findings by using an alternative weighting scheme for the competition policy index (CPI_FA). The CPI index we used in the main analysis was created by Buccirossi et al. (2013) who constructed it by assigning arbitrary (equal) weights to the CPI’s features and sub-indices. We want to test whether our results remain consistent if we use an alternative CPI_FA index (created by Buccirossi et al. (2013) as well) which implements the factor analysis and overcomes the problem of exerting too much discretion in choosing the weights to assign while constructing the index. Table A6 reports the results, which appear to be similar, in both size and statistical significance (0.165***), to the main estimates presented in Table 2.

We also disaggregate the CPI into its subcomponents and run the baseline model to see whether heterogeneous effects emerge. Table A7 reports the set of findings. Interestingly, the CPI_institutions and the CPI_mergers are the subcomponents that seem to be more strongly correlated with the labor share (both in size and statistical significance). However, the CPI_enforcement is not statistically significant. This suggests that the main drivers of the effect of competition policy on the labor share are the institutional component and merger policy. These findings are consistent with Gutiérrez and Philippon (2018) and Affeldt et al. (2021).

APPENDIX B: SENSITIVITY

B.1 Sample composition and outliers

To check the sensitivity of our results to the sample selection, we run the model by excluding countries that, for different reasons, might be considered as outliers.²⁷ In Table A8, we exclude two former Communist economies Hungary and Czech Republic, that experienced different economic and market development process. This might have influenced the labor share dynamics and the quality of the competition law enforcement. However, even excluding these countries our baseline results still hold and are not dissimilar from the main analysis reported in Table 2. We also run the analysis by excluding the United States and Sweden, to check whether the overall results are being affected by the high CPI values. Table A9 reports the findings and shows that the coefficient associated to the CPI is even greater in size (0.21***, Column 6).

In Table A10, we report the baseline results by excluding United States and Canada. The decision to exclude these observations is based on the findings of Gutiérrez and Piton (2020), who, after adjusting for the share of self-employed and for housing residential assets, found a downward trend in the labor share only in the United States and Canada. To test whether our results are only driven by the inclusion of these two countries we run the analysis excluding the United States and Canada. Our results show that the coefficients remain very similar in size and statistical significance, albeit at a lower level of significance (10% level).

As discussed in Section 5, we use a proxy for competition margins to study the interaction between competition policy and labor protection. As already explained, in doing so we follow the approach of Griffith et al. (2007) and Buccirossi et al. (2013). While we are aware of other proxies and empirical methods for estimating market power, we believe this choice is appropriate for the purposes of our analysis which is to investigate whether there is a link between competition policy and the labor share.²⁸

As seen in Figure 3 of the main text, for most of the countries (e.g., United Kingdom and France), the markup proxy and our proxy are very similar. To confirm that they are comparable measures we inspected any divergences over time, especially for Japan and Netherlands (see Figure A1). The sample correlation for our full sample is 0.40***, but when we remove these two countries, it jumps to 0.70***. We therefore run the reduced model (CPI and competitive margins) analysis by removing Japan and Netherlands (for which we have the highest divergence between the proxies) and report the results in Table A11. We found coefficients that are bigger in size and statistically significant for the reduced model. However, we treat these estimates with caution as the size of the subsamples below and above the median is very unequal in most cases. Nevertheless, we still find similar qualitative heterogeneous effects, by accounting for different labor market regulation settings (Table A11).

B.2 Profit share, CPI, and labor market regulation

Table A12 replicates the analysis reported in Table 4, in the article, but using a measure of “profit share” instead of the competitive margins as dependent variable.²⁹ Specifically, profit share is defined as:

The magnitude, the size, and the pattern across the subsamples (below and above the labor market regulation) in the main analysis are confirmed, strengthening our confidence in our results and in the role of competition policy in limiting market distortions in the presence of weaker labor market regulation.

B.3 Alternative model assumptions

Our analysis relies on some sets of assumptions with respect to (i) the type of errors’ clustering (at country level in the baseline model), (ii) the types of individual effects (country-industry FE and time FE), and (iii) the structure of the error terms (see Equation (1) in the article). However, some of these assumptions might be subjective and based on a certain degree of discretion. Hence, we want to test the robustness and consistency of our findings, by relaxing some of these assumptions and be further reassured on the consistency of our analysis. Results are reported in Table A13.

We first re-run the benchmark model with OLS, by clustering this time at industry level (rather than at country level) allowing for spillovers effects within sectors across countries (Column 1), and then clustering at country-industry level, controlling for different and separate sets of fixed effects (year, country, and industry FE), as reported in Column 2.

We also include in Column 3 both country-time FE and industry-time FE, to make sure to identify plausibly exogenous shifts in the CPI and not merely picking-up heterogeneous time trends. In doing so, we can better account for country-specific shocks, such as the economic transition and institutional change that some economies might have undergone over the time sample considered in our analysis (such as the Czech Republic and Hungary) and some sector-specific shocks (e.g., the boom of the financial service sector or the oil/gas industry).³⁰

Lastly, as reported in Column 4 of Table A13, we decided to focus on the correlation structure of the error term and shape it as an AR(1) component, to account for a possible presence of serial correlation, by also including year, country, and industry fixed effects. Similarly, in Column 5 we impose an AR(1) structure on the errors and control for industry-within-country effects and year fixed effects.

The overall estimates, especially with respect to the CPI, are similar in size and in statistical significance to the OLS estimates of the benchmark analysis reported in the main body of the article. The coefficient of the CPI ranges between 0.126 and 0.40 and is statistically significant at the 1% level.

REFERENCES

Author notes