Employment protection and labour productivity growth in the EU: skill-specific effects during and after the Great Recession

Abstract

Does employment protection affect sectoral productivity growth differently during crises and recovery periods? This paper sheds light into this question by investigating the relationship between employment protection legislation (EPL hereafter) and sectoral labour productivity growth in the EU in the context of the Great Recession. We consider the crisis and recovery periods, evaluate the relevance of both levels and changes in EPL for productivity growth, and explore the conditioning role played by sectoral differences in terms of cumulativeness of knowledge as well as the skills of the labour force, captured by different levels of education. We find that stricter labour protection reduces labour productivity growth in sectors with a large share of workers with tertiary education, whereas this effect is negligible or positive in sectors where workers with secondary or only primary education are more prevalent (such as agriculture, mining and quarrying). We attribute this to a more intensive labour hoarding in the former, as EPL strengthens labour hoarding in sectors that rely on firm-specific knowledge accumulation and skilled human capital that are difficult to substitute with physical capital. Whereas it is simple to dismiss (and to find later) unskilled employees. They not only can be substituted more easily with capital, but also the costs of their firing are lower, they are overrepresented among workers holding temporary contracts, and they might be unequally informed and able to exercise their rights. This leads to low (if any) labour hoarding and little impact of EPL on labour productivity in such sectors. We also document that the negative effect is prominent only during the crisis, and an increase in the stringency of EPL over an extended period stimulates employers to substitute labour with investments in physical and knowledge capital.

1 Introduction

The Great Recession (2008–2009) brought with it a generalised consensus among economists and policymakers on the need to improve the flexibility of labour markets with a view to restoring unemployment to pre-crisis levels and combatting long-term unemployment (OECD 2013a). In fact, concerns surrounding the negative implications of too stringent labour market regulations were voiced even before the crisis, especially by key international institutions such as the Organisation for Economic Co-operation and Development (OECD), which started collecting indicators to capture the extent of employment protection in its member countries. When these indicators started being compiled, it was already believed that “the stringency of employment protection legislation (EPL) matters for innovation, and productivity too. Excessive job protection makes it costly for firms to restructure and also reduces the incentives for the employee to exert effort and to move to higher productivity jobs” (OECD 2008). In a very recent contribution, this view is reiterated: “when job protection is too high […], efficient job allocation and innovation are likely to suffer. Hence, overly strict dismissal regulation tends to reduce productivity growth and increase the duration of unemployment spells” (OECD 2020).

In the context of the Great Recession, and in the specific context of European Union (EU) Member States, the focus of policy and applied research has been on those countries that had been hardest hit by the crisis, mostly in the periphery. As an example, the OECD (2016) provides an overview of the lessons learned for the case of Greece in terms of macroeconomic outcomes after the plethora of structural reforms the country undertook since 2010. It is estimated that EPL reforms, and in particular the relaxation of some of the most stringent elements of job protection, have had a substantial impact on Greek Gross Domestic Product (GDP) through their impact on productivity growth. Similar studies are available for Italy and Spain. For the former, a similar analysis also found a positive and non-negligible effect of employment protection reforms on GDP, mainly through the employment generation channel in this case (OECD 2015a). For Spain, the reforms implemented in 2012, which reduced the stringency of job protection legislation, have been shown to have contributed to ‘wage moderation and increased hiring on permanent contracts’ (OECD 2013b),¹ although Villanueva and Cárdenas (2021) challenge the pro-flexibility results found in the literature. An evaluation of the effects of such reforms on other outcomes, such as productivity growth, that takes into account the resilience aspect during the crisis, analyses their impact conditional on the skill level of workers, and focuses on EU Member States—as we do in this study—is, however, absent in the literature.

This paper contributes to the extant literature in several respects. First, we study the effect of EPL on labour productivity’s resilience to the financial crisis—i.e., on changes in labour productivity during the crisis and the subsequent recovery—based on EU data. This is the first attempt, to our knowledge, at examining this issue in the context of the Great Recession and the EU. The choice of the Great Recession as the focus of our analysis is of special interest. The impact of recessions on productivity growth in EU member states could be different because of different stringency levels in EPL. Indeed, more stringent EPL might lead to greater labour hoarding, thus on the one hand hurting productivity growth due to a constant or lower level of output. On the other hand, however, more stringent EPL might induce rises in labour productivity during the recovery period, as retained labour facilitates the recovery, especially in sectors where high quality of human resources and knowledge capital plays crucial role in knowledge accumulation for innovations. Furthermore, costly dismissals during the crisis might provide incentives for employers to invest in physical and knowledge capital. Second, we provide an analysis of the impact of both the short and long-term changes in EPL, relating them to their impact on productivity growth at different time horizons. We find support for the view that, upon a negative shock, EPL is binding over short periods, preventing an optimal reallocation of factors of production, thus potentially leading to an inefficiently large amount of labour, and in turn, declining productivity. By contrast, over longer time horizons, firms facing relatively strong levels of EPL may find it optimal to increase their investment in capital in order to reduce their susceptibility to such shocks and avoid labour overhang. Third, we break down our analysis between high- and low-skill intensive industries based on several metrics. This distinction is important because not only can firms react under negative shocks differently with respect to qualified versus unqualified workers—due to unequal searching and hiring/firing costs—, but also because workers of differing skills and education might be unequally informed and able to exercise their rights (see, e.g., Meager et al. 2002; Denvir et al. 2013; OECD , 2019a; b). Fourth, relying on a proxy of cumulativeness of knowledge proposed by Peneder (2010), we show that part of the difference in sectoral effects might stem from the distinct dependence of firms (from these sectors) on internal/external knowledge sources that drive their innovation activities.² Fifth, we evaluate how the importance of the different channels with which EPL affects productivity growth differs between the crisis and recovery times.

Our analysis combines data from several sources. Data on EPL (regular and temporary contracts) come from the OECD. Labour productivity data is defined as output divided by the input of labour (hours) and stem from EU KLEMS dataset. Data on education was obtained from the EU Labour Force Survey (LFS). Most control variables are taken from the World Bank Development Indicators and Worldwide Governance Indicators databases. For different questions, we employ a number of tailored econometric techniques, including (robust) estimation of linear cross-sectional regressions, mixed effects models, Ridge regressions, as well as panel data models.

The paper is organized as follows. Section 2 provides a thorough overview of the most relevant literature. Section 3 presents the data used and discusses important methodological considerations in the empirical approach. Section 4 provides a motivation for the paper based on a number of empirical regularities observed in the data. Section 5 presents the econometric specifications in detail. Section 6 discusses the main empirical results. Section 7 provides robustness and further impact analyses. Finally, Sect. 8 contains concluding remarks.

2 Literature review

In this section we briefly review the main literature examining the impact of EPL on productivity growth. This set of contributions is not particularly large, since the issue that has garnered most attention so far is rather the impact on employment. The existing literature on the impact of EPL on outcomes other than employment emphasises that the idea that curtailing EPL invariably leads to better economic outcomes hinges on the assumption that it increases employment without reducing innovation and productivity growth (see a critical discussion in Vergeer and Kleinknecht 2010, 2014). From a theoretical standpoint, the net effect of EPL on productivity growth can take, in principle, a negative or positive sign. The main arguments supporting the first view are centred around the higher firing costs imposed by higher EPL and include: (i) EPL might reduce the productivity-enhancing reallocation of labour,³ (ii) EPL might also reduce incentives to invest in cost-saving innovation, (iii) EPL might provide incentives to invest in lower return, less risky projects, (iv) asymmetrically softer EPL for temporary contracts, as compared with regular ones, might encourage employers to rely more on workers with temporary contracts, who tend to be less productive, (v) EPL might hamper job turnover and lead to an increase in the mismatch between the qualifications required by firms, which change across time, and those possessed by their workers (OECD 2015a; b; Berton et al. 2017).⁴

By contrast, arguments pointing to a positive relationship between EPL and productivity growth relate mainly to incentives on the side of firms to invest in human capital and incentives on the side of workers to exert more effort: (i) EPL might induce firms to invest in training programs for its workers, partly to avoid large lay-offs, (ii) EPL can signal that workers are more unlikely to get laid off, which in turn can result in more commitment to the firm, (iii) EPL conditions power relations between employers and their employees, with stronger EPL potentially leading to a more equitable power dynamic and thus higher wages and likely reduced worker turnover, (iv) EPL reduces job turnover and hence increases familiarity between workers and thus might lead to more efficient teamwork, (v) EPL might lead to greater innovation through firm-specific knowledge accumulation and a selection effect in which only the firms which can bear the cost of more stringent worker rights regulations thrive.⁵ In addition, EPL may stimulate corporate innovation and new production methods that promote the substitution of capital for labour (Bena et al. 2022).

In response to these theoretical contributions, an extensive body of empirical literature has developed in parallel to investigate which of these forces governing the relationship between EPL and productivity growth may dominate. The two types of productivity measures generally analysed are total factor productivity (TFP) and labour productivity, chiefly depending on the level of granularity of the available data. In the next paragraphs we provide a succinct summary of the most important contributions to the topic analysed in this paper, in order of proximity to our analysis.

Arguably, the study closest to ours in scope is Van der Vorst (2019) who investigates the impact of EPL on both TFP and labour productivity growth. He also distinguishes between low-skilled and high-skilled sectors. However, there are a number of key differences, chiefly including a different classification of industries according to the skill level of the labour force, disparate data and time dimensions, and different econometric model specifications. In addition, there exist important qualitative differences worth highlighting. First, as opposed to us, he does not analyse the impact of changes in EPL over time and thus does not differentiate between long-term and short-term impacts. Second, this author does not account for potential endogeneity of the EPL indicator after the Great Recession, whereas we explicitly investigate this aspect. Third, he distinguishes between low-skilled and high-skilled sectors using a simple dummy variable,⁶ whereas we use a finer set of underlying variables to measure the skill intensity of industries, highlighting in addition the varying impact of EPL policies conditionally on three educational levels. Fourth, this author does not cover temporary workers, which might lead to biased inference due to omitted relevant variables. Fifth, he only analyses EU15 countries and the US, whereas we cover all those EU-27 Member States that are members of the OECD (plus the United Kingdom).Sixth, there is a key difference regarding the goal of both papers. Van der Vorst’s (2019) analysis is aimed at obtaining general conclusions on the relationship between EPL and productivity growth, our main goal is instead to investigate this link in the context of a crisis and the periods surrounding it, with a view to drawing lessons that can be applied to other economic recessions such as the COVID-19 crisis. We thus focus more on what EPL implies in terms of the resilience of labour productivity growth in different sectors to a recessionary shock. Nevertheless, despite these data and methodological differences, our main findings are broadly in line with those reported in Van der Vorst (2019), namely that stricter EPL policy is more binding in sectors dominated by high-skilled workers.

A seminal contribution in the literature on the relationship between productivity growth and EPL in the presence of business cycle shocks is Bassanini et al. (2009), who examine the impact of EPL on TFP by resorting to cross-country and time variation in both variables. Like us, they use data from OECD countries to classify industries according to their natural degree of net employment flows to find that more stringent dismissal regulations have a greater negative impact on those sectors where layoff restrictions are more likely to be binding. An instrumental variable approach is adopted in order to calculate the industry layoff propensity.⁷ The authors delve further into the effects of different types of employment regulation and observe that regulations concerning temporary contracts do not bear a significant effect on TFP growth.⁸ Although similar in nature, our main analysis focuses on labour productivity growth⁹ and breaks industries down according to the skill intensity of the labour force instead, distinguishing in addition between the recession and recovery periods. Contrary to Bassanini et al. (2009), we find a significant influence of EPL on productivity growth not only for regular but also for temporary contracts.

Many examinations focus solely on the impact of regulations related to temporary contracts. A paper belonging to this strand of literature, conceived along very similar lines as the one in Bassanini et al. (2009), is Lisi (2013). Focusing on data for the EU, this author also inspects the impact of labour market policies surrounding temporary employment protection, albeit with a different empirical strategy than previous cross-country studies. This approach consists of exploiting a particular difference-in-difference assumption. This author confirms, for the case of temporary contracts, the widely found result for regular contracts that EPL reduces labour productivity growth more in those industries requiring a greater employment reallocation. Moreover, he obtains the more general result that the use of temporary contracts has a negative, even if small in magnitude, effect on labour productivity, which is fully consistent with our findings.

In a follow-up paper, Lisi and Malo (2017) endeavour to analyse whether the impact of temporary employment differs across sectors according to the sectors’ skill intensity. Consistent with expectations, and our own results related to tertiary education, they find that temporary employment damages productivity across all sectors, and especially in relatively more skilled sectors. They show that this result is robust to different classifications of sectors according to skills and productivity measures. The main policy implication following from their analysis is that labour regulation should be designed to address the use of temporary employment as a flexible way to enter the market, rather than as a structurally cheaper form of labour. These findings are very close to ours, albeit we focus on overall EPL as opposed to temporary employment only.

Damiani et al. (2016) corroborate the negative correlation between the share of temporary employment and TFP growth for a panel of 15 EU countries, especially in sectors with higher propensity to hire transitory workers, further demonstrating that the deregulation of temporary employment deters on-the-job training and the acquisition of firm-specific competencies.

Another study utilizing an industry-level panel of EU countries to analyse the effect of temporary employment protection is Jacquier (2015). The data are drawn from the same sources as ours, namely the OECD’s EPL and EU KLEMS for the industry-level economic variables. Using a two-stage regression procedure, the author shows, first, that temporary employment has an unconditional, significantly negative impact on TFP and, second, that lowering employment protection on temporary jobs, conditional on employment protection on regular contracts being high, creates a surge in temporary employment that is detrimental to productivity performance.

From a more policy-oriented angle and with a wider scope, Rincon-Aznar and Siebert (2012) examine the impact of EPL in European countries on TFP, wage growth, and employment. They find that stricter EPL reduces both TFP and wage growth, while leaving employment broadly unchanged. In addition, contrary to the results of other contributions, they show that service sector industries are less affected by EPL than manufacturing, that industries dominated by large firms are affected to a greater extent than those dominated by smaller firms, and that hours worked per employee increase with EPL.¹⁰ Their results point to the existence of important policy trade-offs in terms of EPL, implying that reforms of the labour market need to progress with caution.

Beyond the EU context, Van Schaik and Van de Klundert (2010) analyse the role of labour market institutions as a driver of productivity growth in a panel of 21 OECD economies over the period 1960–2005. The relatively long time dimension allows the authors to find that the impact of employment protection is qualitatively different before and after 1980. The authors attribute this to technological change being mostly driven by imitation in the first sub-period, implying that incumbent workers were important in generating productivity growth and hence that the effect of EPL was more muted. By contrast, in the second sub-period, product and process innovation became the predominant factor, meaning that excessive protection of insider workers may hamper needed flexibility in terms of labour force adjustments, thereby reducing productivity growth.

Relevant studies based on data for individual countries include Autor et al. (2007), Vergeer et al. (2015) and Bjuggren (2018). The first authors exploit U.S.-state level data on wrongful-discharge protections over 30 years to examine their impact on employment flows and TFP. They find that the adoption of wrongful-discharge protections, a special type of employment protection, reduce employment flows and firm entry rates, while leading to a rise in capital deepening and a decline in total factor productivity. However, their analysis also shows evidence of strong employment growth following the adoption of dismissal protections, which runs counter to conventional wisdom. Moreover, analysis of plant-level data provides evidence of capital deepening and a decline in total factor productivity following the introduction of wrongful-discharge protections. In the second study, the relationship between labour flexibility and productivity is analysed through the lenses of the degree to which the former affects the commitment of workers and their training as well as the style of management. Using firm-level and survey data from the Netherlands, the authors find a negative relationship between flexible personnel and labour productivity growth, adding one further insight, namely that the relationship is moderated by the type of innovation regime.¹¹ The third study uses micro data on Swedish firms and a quasi-natural experiment on labour law change to study the effect of employment protection rules on labour productivity. In line with most results in the literature, it is found that increased labour market flexibility increases labour productivity, with these rises being driven mainly by the older and the smallest firms.

A recent in-depth survey of empirical works on the interplay between supply-side labour market reforms, innovation and productivity is offered in Kleinknecht (2020) and Hoxha and Kleinknecht (2020). After reviewing the most relevant evidence, the authors conclude that the negative impact of more flexible labour relations is significant in medium–high and high-tech sectors characterized by a high ‘cumulativeness’ of knowledge, while in low-tech sectors, where cumulativeness of knowledge is low, there is little or no effect. The empirical contributions collected in these papers differ in scope and working definitions from ours. In particular, the handful of papers reviewed that are closest to ours in nature focus on drivers other than strictly EPL. Buchele and Christiansen (1999) examine the role of a composite index, consisting of several indicators other than EPL, as the main explanatory variable.¹² Auer et al. (2005) analyse employment tenure, defined as the length of time that workers remain in their present jobs or remain self-employed. The authors argue that job tenure bears some relationship with EPL, but do not explicitly carry out an analysis of the impact that the latter may have on productivity through its impact on job tenure. Pieroni and Pompei (2008) examine the link between labour market flexibility and innovation, where the former is defined in terms of job turnover and the latter is proxied by patents. Their results do not thus pertain to the effect of EPL on productivity. Sanchez and Toharia (2000) build a model of efficiency wages and empirically test it using Spanish data. They find that the introduction of temporary contracts has an impact on wage formation, reducing the real wage cost. The most important departure with respect to our analysis is that they define labour market flexibility as the share of employees with temporary contracts, whereas only the effects on wages are analysed, and not the direct impact of this share on productivity. Besides these key differences, there are also the usual dissimilarities in terms of data sources and periods used in their econometric analyses.

To end this section, it is important mentioning a stream of literature which underscores the firms’ knowledge management aspect and offers a number of insights relevant for the analysis of the links between EPL, employee relations, and labour productivity. Contributions in this domain suggest that such factors as strategic human resource practices, job satisfaction, benevolent atmosphere fostering collaboration, trust, loyalty, and commitment in the workplace may have a positive effect on knowledge sharing and accumulation, in turn increasing innovations and labour productivity either in general or, at least, in ‘Schumpeter-II’ industries with high internal cumulativeness of knowledge (Galunic and Rodan 1998; Chen and Huang 2009; Kuo et al. 2014; Vergeer et al. 2015; Kleinknecht et al. 2016; Kleinknecht 2020). As a result, these authors stress that lax hiring and firing conditions might hamper productivity by reducing workers’ commitment, increasing excessive management bureaucracies, and hindering firm-specific knowledge accumulation.

3 Data and methodology

We focus on labour productivity growth performance during and after the Great Recession by using data for the period 2008–2017.¹³ We further split the data into the recession episode (2008–2012) and the recovery period (2013–2017). We analyse all those EU27 Member States that are also members of the OECD, augmenting the list of countries with the UK.¹⁴Our analysis is performed at the country-sector level. We use the NACE Rev. 2 sector split by considering thirteen aggregate sectors specified in Appendix 1, Table 6.¹⁵

Our data are of annual frequency and come from various sources. Labour productivity is calculated as output (gross value added), expressed in real terms, per unit of labour input, expressed in hours. These data are taken from the EU KLEMS2019 release.¹⁶ This database is also used for retrieving data on capital at the sectoral level, which is used for the calculation of capital-labour ratios.

The OECD provides data on the strictness of EPL for regular and temporary contracts,¹⁷ with greater values of EPL corresponding to stricter regulation and a less flexible labour market. We explore various dimensions of EPL in our models by considering both jointly and separately the EPL strictness indicators for regular and temporary contracts. In a first step, we use the sum of the two indexes; although some studies find dissimilar effects on productivity between the two indexes (see Bassanini et al. 2009), in the main analysis we opt to aggregate the two indexes due to the limited number of observations and because the statistical hypothesis of equal coefficients cannot be rejected.¹⁸ Furthermore, including the two indexes separately with the related additional interaction terms into the regression specifications would greatly increase the number of parameters, rendering less precise estimates. Nevertheless, in a second step, we provide a robustness check for our aggregate results by including these two indexes separately in Table 5.¹⁹ In the robustness analysis reported in Table 12, we also explore if the share of temporary workers has any additional influence on our main results.

All data on the skills of workers are derived from the EU Labour Force Survey (LFS). In order to examine the conditioning effect of skills, we use education and occupation characteristics of workers as proxies of labour skills in a specific sector. For education, we separate between the share of workers with primary, secondary, and tertiary education, which are mutually exclusive categories. To account for the usual age restrictions found in the literature, we calculate the share of workers with secondary education from those who are older than 15, whereas we set the age threshold to 25 when considering tertiary education. Apart from a few robustness checks, we typically use primary education as the control group in our estimations. Occupation data from the EU LFS are used to derive the share of white- and blue-collar workers in different sectors.²⁰ Blue-collar workers are used again as the baseline in the estimations. As both types of proxies for workers’ skills yield qualitatively similar results and insights, and since the occupation-based proxy becomes insignificant when both are included jointly, we use hereafter the education-based shares in our main estimations.

The earlier discussed EPL data are at the country level, which prevents us from the inclusion of fixed country effects in our econometric models due to an identification problem, namely that EPL would be collinear with fixed country effects. However, whenever we are not interested in the impact of EPL as such but in the conditional influence of EPL identified through the usage of certain interactions of EPL with other sector-specific indicators—e.g. education levels in different sectors—, we also control for fixed country effects. In every case, fixed sector effects are always included.

Given that fixed country effects are absent in the main regression specifications, we aim at compensating that by controlling for a large number of potential additional country-specific variables. First, as an indicator of economic development, we use GDP per capita based on purchasing power parity (PPP) from the World Bank Development Indicators. Second, in order to account for the quality of institutions, we further control for corruption, rule of law and government effectiveness indicators, which are provided by the World Bank Worldwide Governance Indicators.²¹ Third, in extended model specifications, we also treat Eastern and Western EU countries separately.²² This is carried out by including a dummy variable, which equals 1 if the country is Eastern European, and zero otherwise. Fourth, in a few models, we control for the Economic Sentiment Index provided by the European Commission. Finally, we control for the size of the shadow economy by using data from Medina and Schneider (2019), which is expressed in terms of percentage of GDP. Summary statistics of all the variables used in the analysis are presented in Appendix 1, Table 7.

An inclusion of sector-fixed effects into the regressions controls for a number of important conditioning factors, which were studied in detail in previous works. For example, Bassanini et al. (2009) demonstrated the importance of layoffs as a mediating factor between EPL and productivity growth. Layoffs in that study are sector-specific. As we include sector fixed effects in our models, the role played by such sector-specific issues, are automatically controlled for.

The control variables we use to reflect the level of economic development, quality of institutions and the size of shadow economy are highly correlated. This leads to a large variance of the estimated parameters. Therefore, in most of our regression specifications, we substitute them with the two significant principal components that explain about 94% of the total variance of these variables.²³

To motivate the econometric specifications from which we initiate the analysis, we start from a simple illustration of relationships between productivity growth and EPL in the next section.

4 Motivating illustrations

In this section, we aim at motivating our research questions by presenting several figures using country-sector-level data. First, we provide a few scatterplots of economy-wide average yearly productivity growth in the whole post-crisis period against the EPL values observed in the year immediately before the crisis (2007). Three cases are considered using different EPL indicators, namely each EPL indicator for regular and temporary contracts separately, and the sum of these two. Second, using only the latter, we plot the same relationships separately for different economic sectors. Third, we hypothesise that the differences observed in the different sectors might be linked to the corresponding labour skills in these sectors, illustrating the relationship between EPL and labour productivity growth by conditioning on education levels.

Since economic policy changes generally gain momentum after crisis episodes (see, e.g., OECD 2019a, b), we fix the pre-crisis EPL level when considering its potential impact on productivity growth in the aftermath of the Great Recession; this aims at avoiding or reducing the potential endogeneity bias induced by simultaneity or reversed causality from productivity growth to changes in EPL policy. Figure 1 plots annual average labour productivity growth in 2008–2017 at country-sector level against the 2007 EPL levels. We distinguish the three EPL indicators discussed earlier in three panels and draw the estimated linear relationships in blue. It should be noted that, in the plots of this section, labour productivity growth rates are adjusted for sector-specific averages, whereas sector fixed effects (FE) will be used later in the econometric estimations to control for sector specificity.²⁴

Fig. 1

Productivity growth and EPL. Average annual labour productivity growth during 2008–2017 (vertical axis), adjusted for sector-specific means, and EPL in 2007 (horizontal axis)

A slightly negative relationship appears in all three cases that needs to be further analysed. As the slope of the linear equations in the left and middle panels of Fig. 1 is rather similar, in the main econometric specifications we use the sum of the EPL indexes for regular and temporary contracts (right-hand-side panel of Fig. 1). This summed index not only accounts for both indicators in a simple way, but also leads to slightly lower heteroscedasticity.²⁵

The figures above pool all sectors together, but different sectors might have diverse responses to EPL depending on their specific characteristics. Figure 2 plots the same relationships as Fig. 1, but for all sectors under consideration separately.²⁶

Fig. 2

Average labour productivity growth (2008–2017) and EPL (in 2007) by sectors. Sectors here are defined by the classification of economic activities as described in Table 6 of Appendix 1

Figure 2 shows that the link between EPL and labour productivity growth differs substantially across sectors. In most cases, we observe a negative relation between EPL and labour productivity growth. Nevertheless, in a handful of activities such as Agriculture (sector A) and Mining & quarrying (sector B), the relationship is positive. Furthermore, Accommodation and food services (sector I) and Wholesale and retail trade (sector G) activities exhibit very weak correlation, if any. A salient characteristic of these sectors is the relatively low skill level of their workforce. Motivated by this observation, we will augment our econometric models with several control variables to account for differences in the skill of workers across sectors.

In Fig. 3, we focus on the potential role of skills in the relationship between EPL and labour productivity growth, using education as the relevant proxy for skills. To illustrate this, we define simple thresholds where we distinguish between industries with a share of workers with at least secondary education above two-thirds, and industries with a share of workers with tertiary education above one-third (left and right panels of Fig. 3, respectively). The red and green lines, representing higher and lower education levels, respectively, reveal the conditional impact of EPL on labour productivity.

Fig. 3

The link between productivity growth and EPL and the role played by education. One dot corresponds to a country-sector pair

The left panel in Fig. 3 indicates that country-sector pairs that predominantly employ a relatively less qualified workforce are almost unaffected by the stricter EPL policy in terms of labour productivity growth, whereas EPL correlates negatively with productivity in sectors relying on more educated employees (in red colour ibidem). Similar observations are obtained in the right panel of Fig. 3, which compares tertiary education to lower education levels (primary and secondary taken together). We again observe that, in country-sector pairs characterised by a larger share of employees with tertiary education, the relationship between EPL and labour productivity growth is more negatively sloped. Both panels thus show that the link between productivity growth and EPL is negative and steeper for greater education levels.

5 Econometric approach

Our econometric specification corresponds quite closely to the figures presented before, while extending the underlying relationships with additional controls. First, to allow for the possibility of a differing impact over time, we include, besides the EPL level in 2007, its short- and long-term changes over time. Second, we further augment the model with many additional country and country-sector-specific factors (or their principal components), mostly aiming to compensate for the absence of country fixed effects.²⁷ In order to avoid potential endogeneity of the post-crisis EPL level to the crisis event, the values for all explanatory variables are fixed at their pre-crisis level (i.e., their values in 2007).²⁸

Let EPL_c,t0 denote the level of the EPL indicator observed in country c in the pre-crisis period t₀ (equal to 2007). The two additional dynamic terms included in the model are the three year and seven-year differences, denoted by Δ_hEPL_c,t0 = EPL_c,t0—EPL_c,t0-h, h ∈ {3,7}, respectively. These particular values for h were selected relying on their statistical significance, with the seven-year horizon being approximately equal to the average business cycle length in the EU (see, e.g. Giannone et al. 2010). Additional country-sector specific controls and sector-specific fixed effects are denoted by x_s,c,t0 and α_s, respectively.

To capture the varying influence of EPL on labour productivity conditionally on education levels (primary, secondary, and tertiary), we augment the baseline specification with the interaction terms EPL_c,t0 x S2_s,c,t0 and EPL_c,t0 x S3_s,c,t0,. Here S2_s,c,t0 and S3_s,c,t0 stand for the shares of workers in sector s of country c in year t₀ with secondary and tertiary education, respectively. Note that, in the specifications containing such interactions, the unconditional EPL_c,t0 term is associated with the effects for the share of workers with primary education only, which thus serves as the baseline level by construction.²⁹

Average yearly labour productivity growth is our dependent variable. Similar empirical results are obtained using both the arithmetic and geometric averages, but the former is more widespread in the literature and thus is used also in our base estimations, leaving the results using the geometric average for robustness checks. We will distinguish between three periods, p: the whole 2008–2017 period (p = 0), the 2008–2012 recession period (p = 1), and the 2013–2017 recovery period (p = 2). Consequently, average yearly labour productivity growth during period p will be denoted by ${\overline{y} }_{s,c,p}$. Our main econometric specification is thus as follows:

$$\overline{y}_{s,c,p} = \alpha_{{\text{s}}} + \beta_{0} EPL_{c,t0} + \beta_{{1}} EPL_{c,t0} {\text{x}}S2_{s,c,t0} + \beta_{{2}} EPL_{c,t0} {\text{x}}S3_{s,c,t0} + \beta_{{3}} \Delta_{{3}} EPL_{c,t0} + \beta_{{4}} \Delta_{{7}} EPL_{c,t0} +\uptheta ^{\prime } x_{s,c,t0} + \varepsilon_{s,c,p} ,$$

(1)

where ε_s,c,p corresponds to the i.i.d. error term.³⁰

Here, we exploit only the cross-sectional variation, while analysing the effects of EPL on productivity growth over relatively long time periods—akin to a long-differences approach—rather than looking at year-by-year fluctuations. Since the values of all control variables are from the pre-crisis period (i.e., t₀ = 2007) while the dependent variable is the average growth rate for different post-crisis periods, we are effectively investigating the predictive performance of EPL.

It is worth noting that the unconditional shares of workers with secondary and tertiary education (S2 and S3) are included in x_s,c,t0. It should also be pointed out that Eq. (1) does not explicitly include the share of workers with primary education (S1), because the shares are perfectly collinear (add up to one). This implies that, although allowing for the inference about the direction of the EPL influence on labour productivity, Eq. (1) is a certain reduced representation of the underlying structure where all the education shares are present.³¹

In the next section, we show the main empirical results mostly using the ordinary least squares estimator and consider several alternative specifications to Eq. (1), including partial representations, alternative skill indicators, and various sets of control variables. In all the cases presented hereafter, sector fixed effects are always included and statistical inference is obtained relying on heteroscedasticity-robust standard errors. In particular, in the main estimations we apply the MacKinnon and White (1985) heteroscedasticity-consistent estimator (the so-called HC1) that adjusts for degrees of freedom and is the most commonly used robust standard error estimator (see Hausman and Palmer, 2012). Robustness to a number of alternative corrections will be explored in the robustness check section referring to the results provided in the Appendix. Apart from that, the ridge regression (shrinkage) estimator, the restricted maximum likelihood estimator of the mixed effects model, and a system GMM³² estimator of a panel data model will be employed later on in the respective robustness studies.

6 Results

Our empirical study is divided into four building blocks examining different aspects at an increasing level of complexity. First, using rudimentary specifications we analyse the central question of whether EPL and its changes over time significantly affected labour productivity growth during the entire 2008–2017 period. Second, we explore the role of skills in influencing the impact of EPL on productivity using education levels as a proxy of skills. Third, by splitting the entire sample period, we study the specific effect of EPL during the crisis (2008–2012) and recovery (2013–2017) episodes. Fourth, we evaluate if significant differences in the impact on productivity emerge when distinguishing between the EPL that applies to temporary versus regular contracts.

6.1 The significance of EPL

Table 1 presents the basic results, corresponding to the estimation of Eq. (1) without any interaction terms:

Table 1 Estimation results for the baseline specification

When considered separately in columns (1) and (2) of Table 1, both the EPL level and its changes over time are highly statistically significant. The EPL level in column (1) is significant, which is in line with the previously discussed literature. However, EPL in levels fails to be significant in column (3), where levels and changes are included jointly, potentially suggesting that the permanent impact of the level of EPL might be insignificant whenever the temporary effects (of EPL changes) are taken into account. Nevertheless, the signs of all coefficients are also retained in this case.

Importantly, the economic effects of EPL changes may not be immediate since they may only be realized after some time. For example, the related literature suggests that greater EPL may lead to capital deepening, thereby positively affecting labour productivity growth in the longer run (see, for instance, Cingano et al. 2016). However, capital accumulation is a relatively slow process, and this impact can thus only be observed with a certain lag. Capital deepening may also stimulate employees to acquire new skills and knowledge, which may require investments in human capital and a better knowledge management. This process also takes time. Furthermore, labour market effects may also not be immediate; both employers and employees may need some time to adjust to the new rules. This motivated our inclusion of shorter- and longer-term changes in EPL, aimed at capturing these distinct processes. Our empirical estimates indeed suggest that the short-run changes in EPL, captured by Δ₃EPL_c,2007 term, are associated with lower productivity growth, thus implying that variations in EPL are more binding in terms of productivity growth in the short term. However, longer-run effects, captured by the Δ₇EPL_c,2007 term, have a positive effect, thus partially softening the overall temporary impact. This might be potentially explained by the fact that stricter EPL stimulates the substitution of labour for capital and vice versa, which is in line with the findings in the literature.

In columns (4)–(6), we further evaluate the potential importance of simple non-linear EPL effects (a quadratic term added to column (4)), the share of the permanent workers (see column (5)), and the state of the business cycle, as captured by the economic sentiment indicator³³ in column (6). A straightforward augmentation of the basic model, given in column (3), with these additional variables does not result in their significant contribution neither when added one-by-one nor jointly (see column (7) for the latter).

The size of the estimated coefficient of unconditional EPL impact in Column (1) is similar as in previous studies. For instance, in Bassanini et al. (2009) it ranges from around − 0.17 to − 0.56 in their specifications using the difference-in-difference estimation framework with labour productivity growth rates expressed in percentage terms. Our dependent variable is the growth rate of labour productivity (without conversion to percentage points). Expressed in percentage terms, our estimate would be − 0.4 (= − 0.0040*100), i.e., around the middle of the previously reported range. The result in Column (1) would therefore indicate that a reduction of EPL index by a unit is associated with a larger productivity growth.by 0.4% point.

Regarding other conditional impact of EPL—including both the short and long-term separation considered in Table 1, and the EPL impact conditional on different education levels reported in Table 2 in the next section— we are unaware of studies with comparable conditioning. Even the closest research by Van der Vorst (2019) uses a dummy variable to classify the sectors which is incomparable with our specifications.

Table 2 Specifications with levels of education

6.2 Productivity growth and EPL: the role of education

The results presented so far reveal a negative overall dependence between EPL and labour productivity growth. However, they were obtained from a baseline specification of Eq. (1), which omits interaction terms and various controls. Next, we explore whether the link between EPL and productivity growth varies across sectors with different worker skills as captured by the share of workers with different education levels. To study this aspect, we augment Eq. (1) by including interaction terms of EPL with the share of employees with secondary (S2) and tertiary (S3) education. We also retain both the level and the short- and long-term changes of the EPL indicator, since, as we will see, they remain significant in almost all equations.

Table 2 presents the estimation results of model specifications of an increasing level of complexity and detail. Models corresponding to Columns (1) and (2) are similar to the specifications presented on the left and right panels of Fig. 3, respectively. Column (1) extends the previous model with a joint secondary and tertiary education indicator (S2 + S3) and its interaction with EPL, whereas Column (2) only adds the share of tertiary education, S3, (both its level and the respective interaction with EPL).³⁴ It should be noted that the control groups—associated with the unconditional EPL parameter (see Appendix 3 for details)—in this case consist of the remaining workers with, respectively: (i) less than secondary education in Column (1), and (ii) less than tertiary education in Column (2).

In Column (1), all variables under consideration are highly significant. The coefficient of the interaction term of EPL with the joint indicator (S2 + S3) is significantly negative, revealing that a stricter EPL policy adversely influences labour productivity in sectors relying predominantly on a more highly educated labour force. On the contrary, the EPL level without an interaction (EPL_c,2007), which captures the impact of EPL in low-skill intensive industries, has a positive and significant coefficient.³⁵ In terms of the size of the coefficients, the result in Column (1) would indicate that a reduction of EPL index by a unit is associated with a larger productivity growth by around 1.3 in sectors where only workers with primary education are employed, whereas the productivity in sectors relying fully on workers with secondary and higher education would be smaller by around 0.85 of a percentage point (= 100*(0.0129—0.0214)). Table 8 in appendix 2 reports the calculated net effects linked with the estimates reported in Table 2.

Similar results follow in Columns (2) and (3) that include either tertiary only or secondary and tertiary education levels, respectively. The negative impact of the interaction with tertiary education remains always significant. In Column (4), we further account for the level of economic development, measured by GDP per capita in 2007 (in PPP), the size of shadow economy in 2007 (as a percentage of GDP), indicators for the control of corruption, government effectiveness, as well as the rule of law, and a dummy variable for Central and Eastern European (CEE) countries. The coefficients associated with our variables of interest remain highly significant. However, we find that the control variables capturing the quality of institutions are insignificant. In fact, the largest differentiation in terms of institutions during the analysed period in the EU was between the CEE and the rest of EU,³⁶ which is already captured by a highly significant CEE dummy variable.³⁷ It is also important to note that the number of controls is rather large and they are highly correlated with each other. The estimates from these models can therefore be susceptible to multicollinearity³⁸ Therefore, in Column (5), we replace the extensive list of macroeconomic and institutional control variables with only two significant principal components, keeping the remaining part of the specification the same as in Column (4). Column (5) indicates that the values of the coefficient estimates linked to the main variables of interest remain quite similar, with a slight increase in significance. At the same time, the explanatory power in terms of adjusted R² is highly similar in Columns (4) and (5). Hence, we conclude that these two principal components are good substitutes for the five country-level variables, and we use them in the more heavily parameterized specifications that follow hereafter.³⁹

Column (6) augments further the previous country-level list of controls with the growth rate of the capital-labour ratio in each country-sector pair during the post-crisis period. Contrary to the other explanatory variables, whose values are fixed at their 2007 level, the capital deepening variable is more likely to suffer from endogeneity problems due to its being measured for the post-crisis period.⁴⁰ Nevertheless, all coefficients of interest remain barely affected, except that of the long-run change impact (Δ₇EPL₂₀₀₇), which remains positive but becomes somewhat smaller and less significant. This is consistent with the interpretation that EPL changes over relatively long periods might be inducing, at least partially, a certain capital deepening effect. Since both the capital-labour ratio and the EPL long-run change variables account for the same process, the latter necessarily becomes less significant.

Next, in Column (7) we evaluate if the five country-level variables—or the respective principal components—used previously account well for country specificity in comparison with the country fixed effects. In this column, we include fixed country effects and thus do without all country-level variables, including the EPL indicators (without interactions). Only the effects of sector-level variables and EPL interactions with sector-specific variables remain identifiable, with their coefficients being quite similar to those obtained in the previous specifications. Hence, we conclude that the included country-level controls seem to account well for unobserved country-level heterogeneity.

Finally, in Column (8) we augment the key specification of column (5) with the same three additional variables as in column (7) of Table 1. As in Table 1, they remain insignificant also here.

These results yield several important economic insights. The significant negative and positive coefficients of short- and long-term changes⁴¹ in EPL (see the first two rows in Table 2) are consistent with our initial expectations regarding the role of labour hoarding and labour substitution with capital, correspondingly. The interpretation is straightforward. Greater EPL increases the employers’ costs of firing employees, which becomes especially relevant in the presence of negative shocks. This in turn distorts firms’ production choices (Cingano et al. 2016) and labour reallocation (Scarpetta 2014). Therefore, initially, over short-term periods, greater EPL negatively affects labour productivity growth by leading to inefficient labour choices by firms, whereas, over longer periods, firms aim at reducing their vulnerability to shocks by investing in more capital-intensive technology. This may also require investments in human resources and knowledge capital, because either new capital-intensive technologies may require other skills or the innovation process driving productivity heavily relies on internal knowledge accumulation. However, despite their differing window of impact, both these effects are temporary in our estimations and disappear altogether after about a single business cycle.

It should be underscored that there are two types of labour hoarding affecting productivity growth. Namely, the one determined by law enforcement (EPL) and the one that takes place because of firms’ rational response to its environment, irrespective of EPL. Even in the presence of a crisis, a firm could be interested in keeping knowledge carriers. This process should be more intensive in sectors relying on the internal knowledge accumulation with a relatively large share of highly qualified people. In the econometric estimations, the productivity growth linked with the labour hoarding part enforced by the law is identified through interactions with EPL.⁴² The negative sign of the significant EPL interactions with education in Table 2 shows that higher EPL levels reduced productivity in sectors with a large share of skilled (educated) employees. This suggests that softer EPL could have diminished the part of labour hoarding induced by the law enforcement, thus benefiting productivity growth during the analysed crisis period.

The long-term impact of EPL—linked to its level (and interactions) and not its changes over time—is strongly conditional on the level of education of the workforce in a given sector. First, note that, as expected, the coefficients capturing the unconditional effects of tertiary and secondary education (see S2 and S3 without interactions) are positive and tend to be higher and more significant for tertiary education, because more skilled workers tend to be more productive. However, the conditional impact, captured by the interaction of educational levels with EPL, is different. In particular, our estimates suggest that sectors relying more heavily on tertiary, and to a lesser extent, secondary, education experience a significant reduction in labour productivity growth due to stricter EPL, whereas sectors characterized by a lower average level of education of the workforce, such as agriculture, mining and quarrying, are found to even benefit from stricter EPL in terms of productivity growth.

There are several potential explanations for why higher shares of more skilled/educated workers are related to lower productivity growth under stricter EPL. First, from the workers’ perspective, low-educated employees may be unaware of changes in labour protection, or they may not be capable of taking advantage of these changes when firms initiate layoffs, whereas highly educated workers might know their rights better and/or be more inclined to exercise them (see, e.g., Meager et al. 2002, for such evidence with additional references in Denvir et al. 2013). Thus, under a negative economic shock, higher amounts of high-skilled labour hoarding emerge, leading to lower labour productivity growth in sectors with a large share of skilled workers. This is consistent with evidence presented in Egert and Gal (2016) that stricter EPL might even encourage hiring of high-skilled workers while impeding their firing. In principle, this problem could be mitigated by unionization reducing the imbalance of power in employment relationship (see, e.g., McCrystal 2008; Bachmann and Felder 2021). However, OECD data on unionization suggests that, in the analysed countries in 2019, the employee participation rate in trade unions exceeded 50% only in Scandinavian countries.⁴³

From the firms’ perspective, EPL may potentially imply higher costs associated with redundancy in higher-skilled sectors compared to lower-skill sectors. Once the financial crisis shock materialized, those firms employing a more skilled workforce likely found it more costly to dismiss their higher-skilled workers than firms with a relatively lower-skilled workforce,⁴⁴especially, in sectors relying more on human capital-based technology (in the next subsection we will examine one such case in more detail). This in turn might owe also to the fact that these workers are usually paid higher wages and thus tend to accumulate more rights in case of redundancies and/or be better organised (Heywood et al. 2018). This might lead to higher levels of labour hoarding of more skilled labour and, therefore, lower labour productivity growth. Stronger EPL amplifies all these effects, partly explaining why we find that EPL is more binding in industries with a higher skill composition of the workforce.

In summary, the negative effect of stricter EPL in sectors with highly educated workers during the crisis might be induced by EPL-linked additional costs of firing (and hiring) of highly qualified labour possessing firm-specific knowledge and requirements. Not only EPL prolongs workers’ stay within a firm increasing the accumulated rights and severance benefits payable during layoffs, but also highly educated employees better know and exercise their rights. Labour hoarding of such workers is therefore stronger, especially given that qualified labour with firm’s specific knowledge is much harder to find than unqualified labour with general knowledge. At the sector level, the negative EPL effect might also stem from a poorer allocation of resources: lower market exit rates of non-viable firms result in a weaker creative destruction and market selection processes (see, e.g., Adalet McGowan and Andrews 2016). Firms that under a weaker EPL would have exited the market releasing the resources found it preferable to continue their operations augmenting the share of low productivity ‘zombie’ companies⁴⁵ due to high costs and procedural complexity of liquidation induced by the EPL.⁴⁶

In country-sector pairs with lower levels of education of its workforce, the impact of EPL stringency on productivity is insignificant or even positive. It signals that there are only low levels of labour hoarding of unskilled employees, if at all. Workers in these sectors not only can be substituted more easily with capital, but also the costs of their firing are lower both in terms of direct monetary costs and the importance of workers in the production technology. The unskilled employees are also overrepresented among workers holding temporary contracts. When a crisis hits, employers may not prolong temporary contracts, hence avoiding labour hoarding irrespective of the level of EPL.⁴⁷ Furthermore, unskilled employees might be unequally informed and able to exercise their rights, which leads to an easier dismissal of such workers. All this might lead even to the ‘over-cleansing’ effect increasing the apparent labour productivity due to overly strong layoffs. In addition, greater EPL that induces firms to keep workers can be erroneously perceived by workers as a benevolent act of firms during the crisis, which rises their loyalty and commitment to the firms, thus resulting in a harder work and increased labour productivity.

So far, we have discussed only the conditional marginal effects. We have shown that the total impact of EPL on aggregate labour productivity in a given sector hinges on the particular labour skill distribution of that sector. In extreme cases where the number of workers with no secondary education is prevailing, our results suggest that an increase in EPL can lead to a higher aggregate productivity growth in the whole sector. However, the number of country-sector pairs with a predominant share of workers who have lower than secondary education is small, mostly consisting of agriculture, mining, and quarrying. Therefore, at the sector level, we can expect the negative EPL impact to prevail.

6.3 Productivity growth and EPL: the role of cumulativeness of knowledge

In this subsection we evaluate a hypothesis that the described labour qualification-specific effects are driven by distinct types of innovators and their intensity of internal knowledge accumulation in different sectors. The cumulativeness of knowledge is one of key elements of the technological regimes determining the patterns of innovative activity (see, e.g., Malerba and Orsenigo 1993, 1997; Breschi et al. 2000). For creative firms with high internal cumulativeness of knowledge, firm’s ability to create new knowledge depends crucially on firm’s present stock of knowledge. The creation and transfer of knowledge over extended periods of time thus relies crucially on the continuity of (high quality) human resources, because the innovation process is often worker-embodied and tacit. A sudden loss (including firing) of labour disrupts the innovation process and technology which is especially detrimental for a firm in the Schumpeterian world of ‘creative destruction’.

In sectors where innovative firms with high internal cumulativeness of knowledge predominate, neither higher EPL nor temporary negative shocks outweigh the loss that would be incurred by a disruption of the innovation process due to the loss (or substitution) of the ‘carriers of knowledge’. Hence, labour hoarding naturally takes place in the presence of temporary negative demand shocks resulting in an instantaneous shrinkage of labour productivity (with an alternative for a firm to exit the market altogether). If this were the only source of labour productivity reduction, the productivity would recover over a longer period in such sectors (we will present some evidence consistent with such a process in Subsection Crisis vs recovery and Section Further robustness and impact analysis).

For other firms, including those that rely only on a low cumulative innovation regime, it is less crucial to retain the ‘carries of knowledge’. Hence, both higher EPL and negative shocks to a demand of products can induce less costly labour firing, potentially substituting it with capital. Consequently, in sectors where a low cumulative innovation regime is predominant, less labour hoarding in the interest of firms will be present, if at all.⁴⁸ Furthermore, the substitution of labour with capital can lead even to some labour productivity gains in such sectors.

The higher is EPL in a country, the larger differentiation of labour hoarding will appear between sectors that can and cannot adjust under a negative shock to production. Furthermore, firms that are inflexible in adjusting/substituting labour might be willing to cease their activity but are forced by the EPL to incur additional severance costs. At the sectoral aggregation level, this would look like an additional reduction in labour productivity. The intensity of such process is of greater extent in sectors where sunk costs are large while human capital substitutability with physical one is limited. At the same time, internal cumulativeness of knowledge relies on employees with high qualification and skills that have higher wages due to a skill premium. Workers in firms that need to rely on internal-knowledge-intensive innovation are likely to stay longer within a firm and accumulate larger severance benefits. All this will result in higher costs of severance payments.

Given that the accumulation and transfer of knowledge is based mostly on highly skilled/educated workers, the results obtained in the previous section are in line with the discussion presented above. The question remains if this is the main underlying source of the differentiation between sectors. To evaluate this aspect, we use an aggregate index of cumulativeness of knowledge at economic activity level that is based on the Peneder (2010) taxonomy of sectors in terms of their cumulativeness of knowledge.⁴⁹ Higher values of an indicator point to higher importance of cumulativeness of knowledge in a sector. Table 3 shows the estimation results.

Table 3 The role of cumulativeness of knowledge

Column (1) includes only the cumulativeness of knowledge indicator without distinguishing between different education levels. The cumulativeness of knowledge is significant in an interaction with EPL and the coefficient is negative. This is consistent with the hypothesis that the intensity of cumulativeness of knowledge in innovation activities matters, and results in stronger labour hoarding (lower labour productivity) in the aftermath of a crisis.

Column (2) includes both the cumulativeness of knowledge indicator and the share of more educated workers in a sector. The significance of the former becomes smaller by an order (as compared with Column (1)), suggesting that indeed a large part of the cumulativeness of knowledge effect seems to be captured by the share of more qualified labour.

Column (3) adds further some main controls from Table 2. In this case, the cumulativeness of knowledge becomes insignificant, whereas the share of better educated workers still conditions—and highly significantly—the impact of EPL on labour productivity.⁵⁰

In summary, the cumulativeness of knowledge seems to be related with labour productivity during the analysed period through the labour hoarding process as expected. However, a stronger significance of labour education could potentially suggest that some additional processes are also taking place.⁵¹

6.4 Crisis vs recovery

In previous sections, we focused on average labour productivity growth over the whole period under investigation (2008–2017). However, the impact of EPL policies on labour productivity might differ substantially during the crisis and recovery periods. A negative shock that reduces demand creates an immediate pressure on firms to act on the extensive margin by shrinking labour as the main variable of adjustment in the short term. Therefore, during the crisis period a stricter EPL might hurt labour productivity growth to a greater extent, due to forced labour hoarding, whereas it might induce an even more intensive labour-replacing process after the crisis than the one initiated already before the crisis. Both factors might thus give rise to higher productivity growth in the recovery period, once demand resumes, on the back of readily available labour and greater production capacity due to heightened capital investment.

Consequently, next we investigate if our relationship of interest differs between the crisis and recovery periods. In Table 4, we split the whole sample into two parts: (i) the period of the crisis (2008–2012), in Columns (1)–(3), and (ii) the recovery period (2013–2017), in Columns (4)–(6). The triplet of columns in each case again corresponds to the equation specifications with secondary and tertiary education levels together, with only tertiary education, and with secondary and tertiary education levels, respectively.

Table 4 The crisis and recovery periods

Two important differences emerge from inspection of Table 4 when comparing the estimates obtained for the crisis and recovery periods. First, the size and significance of the short-term change of EPL (Δ₃EPL) is much larger during the crisis episode than in the recovery period. This is consistent with our interpretation linking the short-term change in EPL to labour hoarding: upon a negative shock, stricter EPL policies imply falling labour productivity due to forced labour hoarding, whereas this channel becomes less relevant during the recovery period. On the contrary, the long-term change in EPL (Δ₇EPL), which we link to regulation-induced labour-saving investments, remains of a very similar magnitude.

Second, the impact of the level of EPL is only significant during the crisis, and its interactions with education levels are much larger in absolute terms during the crisis. During the recovery period, all EPL-linked coefficients become less significant and are accompanied by a substantial reduction in their absolute size. This is again consistent with the labour hoarding argument in the presence of a negative shock to output under stricter EPL: In the recovery period, the need for layoffs disappears and therefore EPL ceases to be binding, irrespective of the skill distribution of industries.

6.5 Regular vs temporary contracts

So far, we have explored the significance of an aggregate EPL indicator, obtained as the sum of the respective components for regular and temporary contracts. In this subsection, we examine the differences in results when these are used separately.

Table 5 presents the results when distinguishing between the EPL strictness indicators for regular and temporary contracts, again differentiating them according to the triplet specifications, namely with secondary and tertiary education levels together, only tertiary education, and secondary and tertiary education levels separately (Columns (1)–(3)).⁵²

Table 5 The regular and temporary contracts

The comparison of results for regular and temporary contracts reveals the following. First, the significance of the short-term change in EPL, Δ₃EPL_c,2007, is stronger for temporary contracts, although its size is about the same in both cases. In addition, its effect seems to be somewhat more muted for temporary contracts. Second, there is a qualitative difference in terms of the conditional impact of the interaction terms. For temporary contracts, the coefficient attached to the interaction term is significant only for tertiary education (Column (2)).⁵³ Whereas for regular contracts, the interaction of EPL with tertiary and secondary education together is strongly significant.^54,55 Thus, we conclude that regulatory provisions for temporary contracts only exert a negative effect on productivity growth in sectors relying more on the highest-skilled workers, while regulations referring to permanent contracts are binding also at lower skill levels. In Table 5, we also provide a statistical test to justify the use of the sum of EPL indicators for regular and temporary contracts. For that purpose, we test the hypothesis that the parameters corresponding to the EPL for regular and temporary contracts⁵⁶ are equal. We employ a standard Wald chi-square test and find that the hypothesis cannot be rejected at the usual significance levels (see p-values in the last row of Table 5). This provides evidence in favour of the suitability to use the sum of EPL indicators for temporary and regular contracts throughout our analyses.

7 Further robustness and impact analysis

In this section, we provide twelve (augmented) variations of previous specifications to assess the robustness of our earlier findings and to obtain further insights. We cover several different definitions of the dependent variable (namely, TFP, quality-adjusted labour productivity, as well as different types of averages), one alternative skill indicator, and several additional control variables not covered previously (namely, post-crisis EPL policy, lagged productivity variable, unionization level, GDP growth rate, periphery indicator, etc.). Furthermore, we also carry out two additional estimations including country and sector random effects and using the RIDGE regression. Due to extensive robustness coverage, we look hereafter only at a single specification, corresponding to Column (5) in Table 2, corresponding to separate tertiary and secondary education variables, a dummy variable for Central and Eastern European (CEE) countries, and the two principal component controls. The findings with other specifications are very similar to the ones in Table 2.⁵⁷

First, we re-estimate the same specification of Column (5) in Table 2, using total factor productivity (TFP) as the dependent variable in Column (1) of Table 9. The impact of EPL conditional on education remains highly significant. However, all the other controls as well as the changes of EPL over time become statistically insignificant. This might partially be because the number of observations and degrees of freedom is smaller when using TFP compared with labour productivity.⁵⁸

Second, in Column (2) we provide evidence indicating that the use of the geometric average, which is more robust to outliers, instead of the arithmetic average of labour productivity, would only increase the significance of our earlier findings.⁵⁹ We retain the use the arithmetic average in our main estimations as it is more generally applied in the related literature.

Third, in Column (3) we augment the previous specification with an additional metric for workers’ skills. Specifically, we use the share of white-collar employees in each sector of each country, next to the education levels, as a proxy for the distribution of skills across industries. The coefficients associated with changes in EPL over time as well as EPL interactions with education retain the same signs as earlier and become even more significant. Although the unconditional white-collar variable is marginally significant, its interaction with EPL is not significant. This further substantiates our choice of education as the relevant proxy for skills in previous sections.

Fourth, the EPL policy might be responsive to the crisis event and, if modified as a consequence, it could influence labour productivity differently. To capture such a possibility, we introduce in Column (4) the change of EPL during the post-Great Recession 2008–2017 period and its interaction with the 2007 level of EPL.⁶⁰ We carry out this exercise for a simple robustness illustration since, as opposed to variables fixed at the pre-crisis level, this additional post-crisis EPL change might be endogenous. Nevertheless, neither post-crisis changes in EPL nor its interaction with pre-crisis EPL level are significant in Column (4), indicating that the relationship between labour productivity growth and EPL was not significantly altered after the crisis.⁶¹

Fifth, as different degrees of trade unionization in countries might lead to varying responses to EPL policies, in Column (5) we check the robustness of our results to the inclusion of the unionization level and its interaction with the EPL indicator. Both coefficients are not significant, while as in Column (4), the impact of EPL conditional on tertiary education is more stable compared with the secondary education indicator.

Sixth, in order to account for the dynamics of productivity (e.g., the capital accumulation process might be highly persistent), we include the average growth rate of productivity during the ten-year period prior to the crisis (1998–2007), as well as its interaction with the EPL level. Both variables appeared to be not significant. Nevertheless, the other explanatory variables remain significant, except for the secondary education and its interaction with EPL.⁶²

Seventh, in addition to accounting for the specificity of the CEE region, we check in Column (7) the significance of a dummy variable for ‘Southern periphery’ that takes value one for Greece, Italy, Spain, and Portugal, and zero otherwise. We do not detect any significant effect of belonging to any of these countries in conditioning the impact of EPL on productivity growth.

Eighth, in Columns (8) and (9), we explore the relevance of the share of temporary workers in two different ways. First, we include directly the share of temporary workers and its interaction with the EPL indicator in Column (8). This renders all the previously established results stronger in terms of statistical significance. Furthermore, a negative coefficient of the EPL interaction with the share of temporary workers indicates that a stricter EPL has a stronger negative impact in country-sector pairs with a larger share of temporary contracts.⁶³ Second, instead of using the sum of the EPL indicators for regular and temporary contracts, in Column (9) we employ a weighted EPL measure where the shares of regular and temporary workers are used to weigh the respective EPL indicators. All the main results are robust to this alternative way of defining the EPL indicator. Furthermore, we find a larger and more significant negative effect of secondary education in its interaction with the weighted EPL indicator.

Ninth, in Column (10) we perform an additional verification of the skill dependence of the impact of EPL on productivity established in previous sections. For this purpose, we calculate sector-specific productivity growth rates based on the Quality Adjusted Labour Input (QALI) skills indicator, instead of quality-unadjusted hours.⁶⁴ This aims at factoring out the influence of employees’ differences in qualification levels on labour productivity. With this metric, the previously established significance of the effect of EPL conditional on educational levels on labour productivity is expected not to be detected, since in principle it already controls for differences in skills. The results in Column (10) concord well with these expectations. Therefore, we conclude that qualification levels are indeed closely linked to the conditional impacts we found before. Nevertheless, the coefficient of the short-term change in EPL remains negative and highly significant, showing that this effect is robust to removing the heterogeneity stemming from differences in the skills of workers.

In Column (11) we report the estimation results obtained after controlling for sector specificity using random effects at both the sector and country levels. At the cost of potential bias due to existing covariance between the errors and the explanatory variables, this aims at checking whether the previously used combination of sector fixed effects with a set of country-specific variables is sufficient for explaining the underlying heterogeneity. Negative and highly significant coefficients attached to the short-term changes in EPL and the interaction term of EPL with tertiary education are found again, further reinforcing the robustness of our initial analyses.

In Column (12), the RIDGE regression estimation results are presented with the estimated coefficients of all education shares.⁶⁵ Despite the reduction in the value of the estimated parameters imposed by this approach, the results remain very similar in qualitative terms both for EPL changes over time and for the interactions of EPL with secondary and tertiary education shares. All coefficients related with the share of primary education are insignificant.

Despite these alterations and the limitations due to a sizeable variation in degrees of freedom, due to missing data for some additional variables, Table 9 reveals that the finding that EPL exerts a negative impact on labour productivity growth, and that the long-run effect is larger in industries with more qualified workers, still holds.⁶⁶

In Table 10, we extended the period of our analysis by including all years for which the data are available. In column 1, the conditioning year is 1998, which is the earliest period for which the education level differentiation is present in the LFS, although only for a much smaller sample of countries. For a number of countries from the Eastern and Central Europe, this period was linked with sever crises during 1997–1999 caused by the Asian and Russian crises with a subsequent period of recovery in the early 2000s. In column 1, we correspondingly look at the average growth rates of labour productivity over the longest feasible period (1999–2017). In column 2, we investigated the average labour productivity growth in 2001–2017, while fixing the explanatory variables at their values in 2000. This separation is motivated by the Dotcom Bubble deflation during 2000–2002: NASDAQ composite index tumbled by 76.81%. In most European countries, this didn’t cause an instantaneous real crisis, but many countries experienced a substantial decline in economic growth afterward. Despite prolonging the data samples, our estimates remain similar to the baseline models: The coefficient corresponding to the EPL indicator is positive, but small, and the coefficient corresponding to cross-term of EPL with secondary and tertiary educations remains negative.

In order to evaluate whether there is some bias due to the methodology applied until now, in column 3 of Table 10, we applied a dynamic panel model estimated using the system GMM (we used second and third lags of the dependent variable as instruments). The results remained similar. However, a problem with the consideration of the longer periods is that a sufficiently long and calm period of recovery that would be free of crises is absent. It is possible that, because of the peculiarities of these periods, we do not capture the positive impact of EPL attributed, e.g., to accumulation of knowledge. To address this issue in a simple framework, we introduce a variable accounting for the time elapsed since the latest crisis that could affect the nature of the impact. Column 4 in Table 10 includes the respective products of EPL and EPL cross-term with education with the number of years since the last crisis. The signs of the corresponding coefficients are the opposite of the coefficients without this product that retain the same features as established in the baseline case. This indicates that there is indeed a reversal of the previously discussed effects that are more pronounced during the periods of the crises and shortly after them, while in normal times, these effects vanish, giving place to other processes. However, our estimates should be treated with caution, because we do not account for the volumes of the crises, and possible nonlinearities.

All in all, we would like to underline that, out of all the effects, the short-term EPL change term (Δ₃EPL) exhibits the largest variation in terms of its estimated coefficient across the different models considered. In this sense, it is the least robust component in our estimations. This is directly connected with the fact that, in 2007, there were only three countries (Czech Republic, Ireland, and Spain) where Δ₃EPL changed. Furthermore, in a few countries, such as Greece and Portugal, there were employment protection legislation reforms during the period of analysis, which may cause reverse causality problems. To evaluate how influential they are individually, we perform an additional robustness check by dropping/keeping various combinations of these countries in the estimations (see Table 13 in the Appendix). Despite these variations, the short-term change remains significant, indicating that the considered change is informative and important even if derived from a handful of countries.

8 Final remarks

In the aftermath of the Great Recession, the effect of EPL on labour productivity varied substantially across economic activities in the EU countries studied in this paper. Part of this dissimilar effect can be attributed to a proxy of cumulativeness of knowledge,⁶⁷ which is one of a few key components in determining the technological regime of innovation activities. We find that in sectors where firms relied heavily on the internal accumulation of knowledge in their innovation process, an instantaneous reduction of labour productivity appeared after a negative shock to production due to labour hoarding. These firms could not fire their workforce or substitute with capital their ‘knowledge carriers’ because this would disrupt their innovation process and would be detrimental for their survival. In general, the less crucial labour is in the production technology, and the more easily substitutable with capital, the lower the degree of labour hoarding and the negative impact on labour productivity.

The sectoral differences found in terms of the conditional impact of EPL are even more pronounced when examining the educational differences of employees. Partially, this is consistent with the cumulativeness of knowledge hypothesis, because the most important ‘knowledge carriers’ are, typically, highly skilled and educated workers. However, the conditioning effect linked with education is much stronger in terms of its statistical significance and robustness to model perturbations. This potentially points to the presence of other factors at work, including the fact that higher wages that are paid for more qualified labour due to wage skill premia in regular times result in larger costs of firing of such employees under more binding labour legislation.

The EPL impact on productivity related to highly skilled/educated workers was significant and negative in connection with the underlying stronger labour hoarding during the crisis. It is harder to find highly qualified workers, especially possessing the company-specific knowledge and requirements. Labour hoarding is therefore expected to be stronger (and labour productivity lower) under a temporary negative shock to production in sectors employing highly skilled workers, especially knowledge carriers. This can take place by a firm’s choice, but EPL magnifies this by inducing additional accumulation of workers’ benefits and increasing firing costs. Furthermore, stricter EPL could result also in a weaker creative destruction and market selection processes because of lower market exit rates of non-viable firms during the crisis. This might take place, because large (and instantaneous) EPL-induced liquidation costs related to market exit might outweigh the costs of continuing the operations with some probability to recover to the break-even production level sufficient to cover firms’ sunk costs. All this is consistent with our result that EPL had a negative and significant effect on labour productivity growth during the Great Recession. However, the lower market-exit-rate argument seems to be less apt to explain the qualification-dependent differentiation of the EPL effect, unless a larger mismatch in the high-skill segment of the labour market is pre-assumed.

The impact of EPL stringency on productivity was insignificant or even positive in sectors with low levels of education of its workforce, which signals that there were no or low levels of labour hoarding of unskilled employees during the analysed crisis period. Unskilled workers could have been substituted more easily with physical capital and their firing costs were lower in terms of both direct monetary expense and the disruption of the innovation process. The unskilled employees were also overrepresented among workers holding temporary contracts, which might have not been prolonged by employers during the crisis. Unskilled workers might be also insufficiently informed and able to exercise their rights, which lead to an easier dismissal of such employees. All this might have resulted also in the ‘over-cleansing’ of low-skilled labour, increasing the apparent labour productivity due to overly strong layoffs. On the other hand, greater EPL could have made workers feel safer during the crisis, which increased their loyalty and commitment to the ‘benevolent’ firms, thus resulting in a harder work and productivity growth.

Looking from a broader perspective, temporary reduction in labour productivity due to a temporarily higher labour hoarding should not be regarded as a problem per sé and might be a firm’s rational choice (see, e.g., Pissarides 1993). However, our estimations show that during the Great Recession, the EPL-linked regulation significantly affected labour productivity growth. The simplest specification suggests that, during the Great Recession, a reduction of the EPL index by a unit could have increased the average labour productivity growth by 0.4% points. Additional decomposition reveals the negative and positive effects on productivity during about three- and seven-year periods of a change in EPL connected, correspondingly, with labour hoarding and investments in physical and knowledge capital. More complex specifications highlight further the importance of sectoral specificities, especially in terms of the varying distribution of skills across industries. Overly stringent EPL was reducing productivity growth in sectors employing a relatively more skilled workforce (those with tertiary education) in connection with an increased labour hoarding under a negative shock to the production. At the same time, in sectors where the majority of employees have attained primary education—such as agriculture, mining and quarrying—, greater values of EPL did not hamper or even increased labour productivity growth in connection with low or no labour hoarding in these sectors. From the social perspective, the latter fact might be worrying and signal either insufficient legislative regulation for less educated workers or their inability to properly understand and exploit their rights provided by EPL,⁶⁸ which weakens the effectiveness of labour market policies towards more vulnerable groups.⁶⁹

Another important finding is that the estimated relationships were stronger during the initial period of the crisis, when output decreased strongly. It became much less binding during the recovery phase. The fact that the EPL effects related to both skilled and unskilled workers became weaker or completely disappeared in the recovery phase might indicate that the results discussed above are more specific to the (severe) crisis period. In the international context, this is consistent with the picture seen in Kleinknecht (2015, Slide 10, and 2020, Fig. 1), where EU12 and EU15⁷⁰ largely underperformed in terms of labour productivity growth during the Great Recession. Nevertheless, moderate EPL levels might even enhance productivity in the longer run: EPL might not only stimulate the substitution of physical labour for capital (including knowledge capital), but also facilitate the firm-specific internal knowledge accumulation which is crucial for firms with ‘routinized’ (Schumpeter II) model of innovations (see, e.g., Hoxha and Kleinknecht 2020; Kleinknecht 2020). The previously estimated temporary (present only during the crisis period) 0.4-percentage point “reduction” in labour productivity induced by the actual EPL might be a small price to pay for the cumulative permanent dynamic gains in knowledge accumulation fostering innovations. At the same time, the discussed estimated gain does not take into account the potential social burden of unemployment increase if other market frictions would prevent the lax labour regulation from clearing the labour markets under a substantial demand shock. This is consistent with studies showing that labour market reforms implemented during the crisis, while demand was low, had limited success (see, e.g., Gehrke and Weber 2018). All these aspects are outside the scope of this paper and would require further exploration and evaluation.

Our conclusions are subject to various limitations. One important caveat is that EPL data from the OECD database are at the country level and hence do not capture the specificities of sectors, although formal employment protection and informal rules of conduct can vary between the sectors. Furthermore, the aggregate EPL indicator masks the detailed regulatory policies within it, which might also exert a particular influence. Moreover, we focused on aggregate economic activities using country-level data. An analysis at a more granular disaggregation of both sectors and regions is left for future research. Although we explicitly conditioned on the pre-crisis situation and used internal instruments in estimating the panel data model, the potential for endogeneity remains an important caveat. In addition, nonlinearities besides the explored simple interactions are highly likely to be present. Accounting for these methodological issues and investigating more fine-grained regulatory aspects as well as additional sources of heterogeneity in responses would require further analysis, necessitating access to richer datasets and the use of different econometric models.

Appendices

Appendix 1

Tables 6 and 7.

Table 6 Sector codes

Table 7 Descriptive statistics

Appendix 2

Tables 8, 9, 10, 11 , 12 and 13.

Table 8 Net effect of EPL increase for sectors with workers in secondary and tertiary education (derived from estimates reported in Table 2)

Table 9 Robustness checks

Table 10 Further robustness checks with longer data periods

Table 11 Further robustness checks

Table 12 Robustness to different estimators of heteroscedasticity-consistent (HC) standard errors

Table 13 Robustness to dropping various combinations of potentially influential countries: CZ (Czech Republic), ES (Spain), IE (Ireland)

Appendix 3

Let us consider a generic underlying regression equation with the three interactions present while omitting unaffected terms and indexes:

$$y = \lambda_{0} EPL + \lambda_{1} EPL \times S1 + \lambda_{2} EPL \times S2 + \lambda_{3} EPL \times S3 + \varepsilon .$$

Since the shares satisfy S1 = 1−(S2 + S3), it holds

$$\begin{aligned} y & = \lambda_{0} EPL + \lambda_{1} EPL \times [1 - \left( {S2 + S3} \right)] + \lambda_{2} EPL \times S2 + \lambda_{3} EPL \times S3 + \varepsilon \\ & = (\lambda_{0} + \lambda_{1} )EPL + (\lambda_{2} - \lambda_{1} )EPL \times S2 + (\lambda_{3} - \lambda_{1} )EPL \times S3 + \varepsilon \\ \end{aligned}$$

Hence, the conditional expectation

$${\text{E}}({\text{y}}|EPL,S2,S3) = \beta_{0} EPL + \beta_{1} EPL \times S2 + \beta_{2} EPL \times S3,$$

where the link between the observed parameters β₀, β₁, and β₃ and the underlying ones is given by

$$\beta_{0} = \lambda_{0} + \lambda_{1} ,\beta_{1} = \lambda_{2} - \lambda_{1} ,{\text{and}}\,\beta_{2} = \lambda_{3} - \lambda_{1} .$$

Hence, when the reduced form is under estimation, one cannot prescribe the change in variable to only the impact of a particular share. However, the conditional impact can be properly predicted from the reduced-form equation.