Labour market reforms and the wage-setting mechanism

Prior to the outbreak of the economic crisis, the Spanish economy was registering increasing external deficits, along with a large accumulation of private debt. Nevertheless, the restrictive macroeconomic policies that were passed to boost the economy and recover from the crisis and the sovereign bond crisis generated new imbalances such as high levels of public deficit and debt, which caused the post-2010 crisis (Reference Cárdenas, Villanueva and ÁlvarezCárdenas et al., 2020). After the outbreak of the economic crisis, the diagnosis of domestic and international authorities pointed to the inadequacy of Spanish labour market institutions, as real wages did not fall sufficiently, or fast enough, to prevent the unemployment rate from rising and to restore price-competitiveness (EC, 2012). Following this argument, larger wage flexibility would have prevented the increase in the unemployment rate in case of a severe economic shock (Reference Doménech, García and UlloaDoménech et al., 2018).

In line with this narrative, the adjustment mechanism during the recession was external, that is, individual or collective dismissals, given the institutional rigidities of the labour market (particularly strong employment protection of open-ended contracts and downwards negotiated wage rigidity). In addition, the adjustment was uneven due to the lack of wage flexibility, as dismissals and wage cuts concentrated on ‘outsiders’ (Reference Bentolila, Dolado and JimenoBentolila et al., 2012). Consequently, relatively unprotected workers, with temporary and other non-standard contracts, were affected to a higher extent by the wage adjustment.

The policy conclusion was that higher flexibility reduces labour market dualism and decreases the unemployment rate. Following this prescription, several flexibilisation reforms of the labour market and the industrial relations system were introduced before the crisis hit, particularly between 2008 and 2013. In Spain, the two most important reforms were those of 2010 and 2012, which implied profound changes in labour regulations. Their goal was to reverse the structural weaknesses of the labour market; reduce wage-setting rigidity and labour dualism; and resolve the situation of high unemployment.

The content of these legal changes is well known (Reference Garcia-Serrano and MaloGarcia-Serrano and Malo, 2013; Reference Muñoz de Bustillo, Esteve, Piasna and MyantMuñoz de Bustillo and Esteve, 2017; Reference Sola, Alonso, Fernández, Koch and FritzSola et al., 2014). In a nutshell, the 2010 and 2012 reforms facilitated the procedures and lowered the costs of dismissal by reducing employment protection legislation (Reference Cruces, Álvarez, Trillo, Gyes and SchultenCruces et al., 2015). Precisely, the 2012 reform deepened the changes introduced by the 2010 reform and introduced new ones (Reference Horwitz and MyantHorwitz and Myant, 2015).

These reforms deeply facilitated unilateral decisions by employers in collective bargaining, dismissals, work organisation and working conditions (Reference Köhler, Calleja, Lehndorff, Dribbusch and SchultenKöhler and Calleja, 2017). As a result, employers’ discretionary power increased considerably in terms of external, internal and wage flexibility decisions (Reference López-AndreuLópez-Andreu, 2018).

Thus, the labour reforms and the Second Employment and Collective Bargaining Agreement (II AENC in Spanish) initiated a wage devaluation process, as illustrated in Figure 1. Indeed, real hourly wages grew by 0.9% annually from the introduction of the euro until 2010q1 and fell by –0.1% yearly from the implementation of the first labour market reform until 2018q4.

Figure 1. Labour share 2010q1–2018q4.

Source: Quarterly National Accounts.

Social dialogue during the economic crisis was focused on concertation and nominal wage moderation, as established by the Interconfederal Agreement for Collective Bargaining (ANC) initiated in 2002 and extended until the first years of the crisis (2008). Inflation forecasts, productivity growth and wage revisions, including the difference between the variation in the actual consumer price index and the inflation forecast (in order to avoid a price/wage spiral), were considered during the wage-setting negotiations. That is, social dialogue accounted for flexibility mechanisms designed to adapt to the business cycle and macroeconomic shocks, prior to the labour reforms.

The negative relationship between negotiated wages in collective bargaining and the unemployment rate during the economic crisis is depicted in Figure 2. A 1 percentage point (pp) increase in the unemployment rate reduced the negotiated wage by –0.14 pp. This reduction provides evidence of downwards negotiated wage flexibility in the wage-setting mechanism prior to the labour reforms, contrary to the arguments of the proponents of the labour market reforms.

Figure 2. Negotiated wages Phillips curve in Spain (2008–2018).

Source: The negotiated wage is measured as the seasonal growth rate of the (nominal) negotiated wages, this data come from the Spanish Collective Bargaining Agreements Statistics. The monthly unemployment rate is measured as the unemployment registered in the State Employment Public Service over the labour force (number of employees, self-employed and unemployed).

Nevertheless, the labour reforms had a strong impact on the wage-setting mechanism through a modification of the relationship between wages and productivity. Figure 3 evidences an increasing nominal wage and nominal productivity growth gap from 2010q2 onwards, whereas during the early years of the crisis, both variables grew at a similar pace (Reference Álvarez, Uxó and FebreroÁlvarez et al., 2018; Reference Cruces, Álvarez, Trillo, Gyes and SchultenCruces et al., 2015).

Figure 3. The wage-productivity gap since the economic crisis (cumulative per cent change since 2008q1).

Source: Quarterly National Accounts, National Statistics Institute (INE).

From 2010q2 (the starting point of the wage devaluation strategy) to 2013q4 (the trough quarter of the business cycle), the labour share decreased by 2.87 pp (see Table 4).

Revisiting Okun’s law

The pro-flexibility hypothesis holds that recent labour market reforms modified Okun’s law, that is, the relationship between unemployment rate variation and GDP growth rate, so that the post-labour reform Spanish economy requires lower GDP growth rates to reduce its unemployment rate than in the previous expansionary cycle, thanks to the wage moderation strategy and the increased flexibility in the labour market.

This hypothesis is derived from the assumptions of the NAIRU model (Reference Carlin and SoskiceCarlin and Soskice, 2005). Accordingly, the institutional aspects of the labour market (such as strict employment protection and hiring laws and high unemployment benefits coverage and intensity) increase workers’ bargaining power. As a response, firms are forced to increase their prices in order to protect their margins (profit share) from falling. When the real unemployment rate is below the NAIRU, a process of accelerated inflation reduces demand and increases unemployment, until the NAIRU is reached. Since the labour reforms were aimed at reducing these labour market institutions and shifting the wage bargaining curve, the NAIRU is expected to fall. As a result, the growth threshold above which the unemployment rate is reduced is lower.

The evolution of the Okun’s law coefficient in recent years has been increasingly controversial in Spain. On the one hand, Reference Fernandez-KanzFernandez-Kanz (2016) held that the economic growth threshold for job creation is significantly lower after wage moderation policies. Similarly, Reference Doménech, García and UlloaDoménech et al. (2018) argued that labour market reforms avoided a larger GDP and employment contraction, which implied a reduction of the output growth threshold for the reduction of unemployment. In addition, Reference Cuerpo, Geli, Herrero, Manasse and KatsikasCuerpo et al. (2018) estimated several Okun’s law equations, finding that the Spanish economy requires lower GDP growth rates to reduce unemployment or to avoid its increase. However, these scholars found that there is a higher atypical employment elasticity to GDP – in other words, linking the reduction of unemployment to an increase in non-standard jobs.

One the other hand, following a regional approach, Reference Buendía and SánchezBuendía and Sánchez (2017) concluded that the output growth required for employment to grow is lower than the level required to reduce unemployment, thus pointing to the potential role of the labour force participation rate in Okun’s law. To tackle this issue, first we decompose the determinants of Okun’s law and assess their changes after the labour market reforms; second, we estimate the Okun’s coefficient.

First, we use the basic identity proposed by Reference GordonGordon (2010), decomposing real GDP (Y) into output per hour (Y/H), aggregate hours per worker (H/E), the employment rate (E/L), the labour force participation rate (L/N), and the working-age population (N). This five-component equation in logs can be expressed in terms of change of the unemployment rate, given that u ≈ −ln (E/L), as follows

(1) $\Delta u=-\Delta \ln \left(Y\right)+\Delta \ln \left(Y/H\right)+\Delta \ln \left(H/E\right)+\Delta \ln \left(L/N\right)+\Delta \ln \left(N\right)$

This decomposition is similar to the one provided by Reference Bentolila, Dolado and JimenoBentolila et al. (2012), but incorporates the evolution of the labour force participation rate and the working-age population. Table 1 compares the contribution of each component to unemployment changes over the different phases of the business cycle.

Table 1. Evolution of the unemployment rate in Spain (1995–2018).

Source: Own elaboration from the Quarterly Spanish National Accounts (update 07/2019) and Economically Active Population Survey (EAPS) (update 08/2019), National Statistics Institute (INE).

By comparing the unemployment rate in the different business cycle phases, we can appreciate the volatility of unemployment as well as the stronger negative reaction of unemployment to changes in GDP during the recession than during expansions. To be precise, the unemployment rate increased by 20.1 pp in 5 years. Despite the considerable reduction of unemployment during expansionary phases, the unemployment rate has remained rather high since the 1990s. This constitutes a stylised fact of the Spanish labour market (Reference Muñoz de Bustillo, Esteve, Piasna and MyantMuñoz de Bustillo and Esteve, 2017).

According to Table 1, the main difference between the expansionary period (1995q1–2008q1) and the recovery period (2013q2–2018q4) is that GDP growth reduced the unemployment rate to a greater extent, which implied a more dynamic response to changes in demand during the recovery period (and after the labour market reforms).

This is explained by the evolution of the remaining components of the equation. Growth in productivity per hour and increases in hours per worker have been limited. The lower the productivity growth, the lower its contribution in reducing the unemployment rate because this effect was not compensated for by a reduction in the hours per worker. It is noteworthy that productivity increased at a higher pace during the recessionary periods than during expansions. This countercyclical behaviour of labour productivity is another stylised fact of the Spanish labour market given its labour intensity. This had the effect that during the economic crisis half of the unemployment increases were due to productivity growth (9.5 pp of 20.1 pp). On the contrary, during the expansion phase, the stagnation of productivity entailed an accelerated unemployment contraction. In this sense, Reference Vergeer and KleinknechtVergeer and Kleinknecht (2014) held that labour market deregulation slowed down labour productivity growth because in deregulated economies with a labour-intensive organisation like the Spanish one, people work more hours to produce a similar output growth, and therefore reducing unemployment at the expense of productivity growth (Reference Storm and NaastepadStorm and Naastepad, 2017).

Hours per worker only fell by 1 pp, mirroring the weakness of the work-sharing schemes in Spain during the economic crisis, which as noted by Reference Amable and MayhewAmable and Mayhew (2011) may have long-term impacts due to hysteresis effects. Moreover, there was almost no change in the number of hours per worker in the expansionary phase.

The drastic fall in the labour force participation rate led to a rapid contraction of unemployment since it countervailed the effect of the working-age population increase. It is especially significant that the contraction of the labour force participation rate was larger during the recent expansion than during the economic crisis. During the previous expansionary phase (1995q1–2008q1), the labour force participation rate and the working-age population grew rapidly, slowing down the reduction of the unemployment rate. As a consequence, a larger output growth was required to reduce the unemployment rate.

It follows that in order to analyse GDP growth effects on the evolution of unemployment, it is convenient to control for changes in the labour force, thus avoiding a bias derived from its variations. Indeed, by decomposing the unemployment rate into the ratio GPD/labour force (Y/L) and labour productivity (Y/E), there is no significant difference between the expansionary periods (Table 2).

Table 2. Evolution of the unemployment rate in Spain (1995–2018), accounting for the labour force.

Source: ibid.

Okun’s law is usually formulated following either the growth rate version or the gap version (Reference Ball, Leigh and LounganiBall et al., 2017). The first establishes a negative relationship between the quarterly real growth rates of GDP and the quarterly change of the unemployment rate. This version is composed of a static formalisation of contemporaneous relations and a dynamic formalisation, captured by the lagged variables. The second version, known as the gap version, employs filter techniques to obtain the trend of both variables, so that the Okun’s coefficient can be calculated in terms of differences with respect to their trend (cyclical component). Despite the extensive developments in ways to estimate this empirical regularity, we use the first formulation as it is the most intuitive and appropriate for our aims

(2a) $\Delta U_t=\alpha +\beta Y_t+{\omega }_t$

(2b) $\Delta U_t=\alpha +\beta (Y_t/La)+{\omega }_t$

On the one hand, there is the traditional Okun’s law that links GDP growth rates to the variation in percentage points of the unemployment rate (equation (2a)). On the other hand, the modified Okun’s law is presented in terms of the growth rate of the ratio GDP/labour force, with the variation of the unemployment rate also in percentage points (equation (2b)). Figure 4 plots both estimates.

Figure 4. How has the unemployment/GDP relationship changed after labour market reforms (2010–2012)?

Note: The period is 1995q1–2018q4, data after 2012 are plotted in grey.

Source: Own elaboration from the Quarterly Spanish National Accounts (QSNA) (update 07/2019) and Economically Active Population Survey (EAPS) (update 08/2019), National Statistics Institute (INE).

The left-hand panel provides evidence in favour of the advocates of the labour reforms, since the grey line for the period after 2012 indicates that the real GDP growth threshold to reduce unemployment in 1 pp has shifted with respect to the previous period. According to the traditional formulation, an increase of almost 3.5% of GDP was necessary to reduce the unemployment rate in 1 pp during the period 1995–2012, whereas in the second period (2013–2018) the GDP growth threshold was 1.2%.

The right-hand panel, where real GDP has been modified to account for changes in labour force size, does not point to a change in the threshold. According to the modified equation, the growth of real GDP adjusted by the labour force necessary to reduce 1 pp of the unemployment rate was 1.25% during the period 1995–2012. After the labour markets reforms (2013–2018), the GDP/Labour force threshold to reduce 1 pp of the unemployment rate rises to 1.74%. This illustrates that when controlling for the labour force, the real GDP growth threshold to reduce unemployment does not vary in the recovery period.

In accordance with these results, we may argue that in order not to overestimate the effect that the institutional changes have on the reduction of unemployment it is necessary to incorporate in the analysis the evolution of the labour force. This evidence suggests that after 2012, job destruction was similar to the previous crisis and the reduction of the unemployment rate was analogous (or even lower) in the expansionary phase, contrary to the narrative of the labour market proponents.

Indeed, employment recovers rapidly during the 2013–2018 period due to the labour-intensive idiosyncrasy of the Spanish economy, which is mirrored in the stagnation of labour productivity growth, and as a result of the fall in the labour force participation ratio. This cast serious doubts about the usefulness of labour market reforms to achieve a higher employment rate, as the pro-flexibility policy claims.

Second, we estimate the Okun’s coefficient following the methodology proposed by Reference HartwigHartwig (2014) and IMF (2010), yet we introduce the labour force (L), as the previous evidence suggests this variable explains the variations of the unemployment rate. We formulate the following equation^{Footnote 1}

(3) $log{U}_t=u_o+{\beta }_{yt}log{Y}_t+\sum _{i=1}^n{\beta }_{yi}log{Y}_{t-i}+{\beta }_{lat}logL{a}_t+\sum _{i=1}^n{\beta }_{lai}logL{a}_{t-i}+\sum _{i=1}^n{\beta }_{ui}log{U}_{t-i}$

U_t denotes the unemployment rate and Y_t denotes real GDP. ${\beta }_{yt}$ captures the contemporaneous effect – or short-run effect – of 1 pp in real GDP on the unemployment rate. The long-run effect or dynamic effect ${\beta }^{*}$ is given by equation (4). The estimation results are reported in Table 3 and will be used in the section headed ‘The effects of the growth of the labour share on the unemployment rate’.

(4) ${\beta }^{*}=\frac{{\beta }_{yt}+{\sum }_{i=1}^n{\beta }_{yi}}{1-{\sum }_{i=1}^n{\beta }_{ui}}=\frac{-0.472}{(1-0.354)}=-0.73$

Table 3. Estimation results (1995q1–2018q4).

AIC: Akaike information criterion; BIC: Bayesian information criterion; AR: autoregressive.

Standard errors in parentheses.

^∽ p < 0.10

^* p < 0.05

^** p < 0.01

^*** p < 0.001

Basing on Quarterly National Accounts data, we calculate the difference of the logarithms of the variables in equation (3), except for the unemployment rate. We run the cumulative sum test for parameter stability to check for the presence of a structural break after the labour market reforms. The presence of a structural break would imply a change in Okun’s coefficient, as the labour market reform advocates hold. According to the CUSUM test, we cannot reject the null hypothesis of no structural break (see Table 3).

These results reinforce the presented evidence and cast serious doubts about the usefulness of labour market reforms to achieve a higher employment rate, as the pro-flexibility policy claims.

An extended Bhaduri–Marglin model with labour market for the Spanish economy

There are previous Bhaduri–Marglin estimations for the Spanish economy, the most relevant being Reference Naastepad and StormNaastepad and Storm (2006), Reference Storm and NaastepadStorm and Naastepad (2012), Reference Onaran and ObstOnaran and Obst (2016), Reference Álvarez, Uxó and FebreroÁlvarez et al. (2018) and Reference Villanueva, Cárdenas and UxóVillanueva et al. (2020). All of them find the demand regime in Spain to be wage-led. Nonetheless, our purpose is to use the estimation results to measure the effect of the wage devaluation on unemployment. Our Bhaduri–Marglin model consists of a set of equations in which changes in the labour share affect different components of private aggregate demand, following the structural or single-equation approach proposed in Reference Álvarez, Uxó and FebreroÁlvarez et al. (2018). As the fall in the labour share during the wage devaluation period has been directly attributed to the labour market reforms (as seen in Figure 2), income distribution is assumed to be exogenous. Critics of this approach as opposed to the aggregative approach have argued that this assumption leads to a simultaneity bias. Nevertheless, by estimating a generalised method of moments estimation for the US to correct for such a bias, Reference Blecker, Cauvel and KimBlecker et al. (2020) find similar results to an ordinary least squares estimation. Given that we included the unemployment rate in our model, the exogeneity of the labour share implies no dynamic interactions between the demand regime and the unemployment effect on income distribution, as a Goodwin-type model would.

Changes in the labour share directly affect household consumption, as the propensity to consume out of wages is usually higher than the same propensity out of profits. Private investment is also affected directly by these changes as this variable is sensible to the profit rate (which includes the profit share). On the contrary, net exports are affected indirectly through changes in domestic or export prices via unit labour costs (trade competitiveness).

In this section, we follow the single-equation estimation procedure to obtain the value of all the elasticities that determine the total effect of 1 pp increase of the labour share on private consumption, private investment and net exports, and therefore on private aggregate demand.

We use quarterly data from Eurostat, the OECD and the Ministry of Economy and Business from 1995q1 until 2018q4. We transformed all variables into logarithms, except for the nominal long-term interest rates and the unemployment rate. Hence, the estimated coefficients are elasticities. In addition, all variables are expressed in real terms. After performing the augmented Dickey–Fuller test, we differentiated the integrated variables.^{Footnote 2} Long-term interest rates, the profit share and the unemployment rate do not have a unit root.

We have included as explanatory variables both the contemporaneous value and the first lag of the variables, keeping finally those that were statistically significant. Autocorrelation has been corrected through the Cochrane-Orcutt transformation (Reference Villanueva, Cárdenas and UxóVillanueva et al., 2020) when the autoregressive term is AR(1). We have estimated the following six equations

(5) $logC=c_o+c_{w}logW+c_{r}logR+c_{dh}logDh$

(6) $logI=i_o+i_{y}logY+i_{\pi }log\pi +i_{r}r+i_{dh}logDh$

(7) $logX=x_o+x_{y*}log{Y}^{*}+x_{reer37}logREER37$

(8) $logM=m_o+m_{y}logY+m_{ppm}logPPM$

(9) $logP=p_o+p_{ulc}logULC+p_{pm}logPM+p_{U}U$

(10) $logPX=p{x}_o+p{x}_{ulc}logULC+p{x}_{pm}logPM$

First, equations (5) and (6) measure the extent to which consumption and investment are affected by changes in functional distribution. Household consumption (C) growth is determined by adjusted employee compensation (W)^{Footnote 3} and the adjusted operating surplus (R), as in the traditional Bhaduri–Marglin model. In addition, we include household debt (Dh) as a control variable, following Reference Onaran, Stockhammer and GraflOnaran et al. (2011), Reference Stockhammer and WildauerStockhammer and Wildauer (2016), Reference Onaran and ObstOnaran and Obst (2016) and Reference Álvarez, Uxó and FebreroÁlvarez et al. (2018).

Private gross fixed capital formation (I) growth is determined by real GDP (Y), profit share (π) and nominal long-term interest rates (r). National income proxies expected demand, which is the main determinant of gross fixed capital formation in OECD and Eurozone countries (Reference Álvarez, Uxó and FebreroÁlvarez et al., 2018). The profit share is introduced as a proxy for profitability. Nominal long-term interest rates should have a negative sign because they contain the financial cost of the investment (Reference Hein and OchsenHein and Ochsen, 2003). Likewise, household debt (Dh) has been included as a proxy of financial markets.

Equation (7) estimates the effects of foreign demand (total GDP of OECD countries, Y*) and price variables (relative effective exchange rate vis-à-vis 37 industrialised economies calculated using export prices, REER37)^{Footnote 4} on gross exports (X). Accordingly, this equation considers that exports of a country depend positively on the income level of the rest of the world – which is an exogenous variable – and negatively on the relative export price vis-à-vis its competitors. In addition, equation (8) shows the effects of domestic demand and price variables (the ratio of domestic prices over import prices, PPM) on gross imports (M). We expect the first factor to have a larger relevance, because of the high income elasticity of imports observed in the Spanish case.

In addition, the last two equations show how changes in unit labour costs (ULC) and import prices (pm) influences domestic prices (GDP deflator, p) and export prices (px). Equation (9) represents a price Phillips curve, where the GDP deflator (P) is a function of unit labour costs and import prices (PM) and the unemployment rate (U). We include the latter variable in order to control the effect of the unemployment rate fluctuations on prices because, as we have pointed previously, it is very volatile. The $p_{ulc}$ coefficient represents to what extent changes in unit labour costs are transferred to domestic prices. Furthermore, equation (10) estimates export prices as a function of both unit labour costs (ULC) and import prices (PM). Similarly, $p{x}_{ulc}$ shows the link between changes in ULC and changes in export prices. The other explanatory variable implies that the export prices could be influenced by import prices; predictably, this elasticity will show a positive sign.

Table 3 summarises the obtained results; the total effect of wage devaluation on the unemployment rate will be calculated in the following section.

The effects of the growth of the labour share on the unemployment rate

First, we calculate the effect of a change in the labour share on consumption (equation (11)).Using equation (5), dividing by GDP (Y) and deriving, we have the elasticity of consumption relative to the labour share, whose value can be obtained from Table 2. According to our results, the propensity to consume out of profits is not statistically significant. This goes in line with the results found in Reference Álvarez, Uxó and FebreroÁlvarez et al. (2018) but differs from those in Reference Onaran and ObstOnaran and Obst (2016). In addition, household debt (Dh) has the expected sign because credit can be a source of disposable income and hence drives consumption up

(11) $\frac{\Delta \frac{C}{Y}}{\Delta \Omega }=\left(c_w*\frac{Y}{W}*\frac{C}{Y}-c_r*\frac{Y}{R}*\frac{C}{Y}\right)=({\beta }_W-{\beta }_R)*\frac{C}{Y}$

(12) $\frac{\Delta \frac{I}{Y}}{\Delta \Omega }=\left(-i_{\pi }*\frac{I}{R}\right)+\left[\frac{\Delta \frac{Y}{Y}}{\Delta \Omega }\right]*i_y\frac{I}{Y}$

(13) $\frac{\Delta \frac{XN}{Y}}{\Delta \Omega }=\frac{\Delta \frac{X}{Y}}{\Delta \Omega }-\frac{\Delta \frac{M}{Y}}{\Delta \Omega }$

(14a) ${\left[\right(\Delta X/Y)/\Delta \Omega ]}_{Xcomp}={\epsilon }_{\Omega }^X*X/Y*1/\Omega =({\epsilon }_{\Omega }^{ULC}*{\epsilon }_{ULC}^{PX}*{\epsilon }_{PX}^X)*X/Y*1/\Omega$

(14b) ${\left[\right(\Delta M/Y)/\Delta \Omega ]}_{Msust}={\epsilon }_{\Omega }^M*M/Y*1/\Omega =({\epsilon }_{\Omega }^{ULC}*{\epsilon }_{ULC}^P*{\epsilon }_P^M)*M/Y*1/\Omega$

(15) $[\Delta U/\Delta \Omega ]=\left[\right(\Delta Y/Y)/\Delta \Omega ]*[\Delta U/(\Delta Y/Y\left)\right]=\left[\right(\Delta Y/Y)/\Delta \Omega ]*{\beta }^{*}$

Second, there are two effects of a change in the labour share on investment (equation (12)). On one side, the ex-ante (or direct) effect consists of the investment profit share elasticity (with a negative sign) multiplied by the reinvestment rate. Thus, even though the profit share could theoretically increase investment under a scenario of a low reinvestment rate the stimuli effect would be small.

On the other side, the ex-post (or indirect) effect reflects the change in investment driven by the accelerator effect. To calculate it, the investment income elasticity is weighted by the investment share (investment over GDP ratio) and multiplied by the ex-ante effects of each component of aggregate demand. In this case, we include the positive impact of consumption and the negative impact of net exports. The total effect on investment will depend on the ex-ante and ex-post effects. If the ex-ante effect was higher than the ex-post effect, the investment would be profit-led even if aggregate demand were wage-led.

In our case, the direct profitability effect is not statistically significant; the same happens when alternative lag structures and specifications are tested. Therefore, investment is also wage-led as we find a strong positive relationship between income and private investment, hinting to a strong accelerator.

Although some previous papers have found a positive impact of the profit share on investment (Reference Storm and NaastepadStorm and Naastepad, 2012), we believe this might be due to the period used (1964–2000). There is another significant variable – household debt – which could play a determinant role in the high residential investment during the period 1995–2007. Long-term interest rates do not show a statistically significant relationship with investment.

The marginal effect of net exports on GDP (equation (13)) depends on the impact of the labour share on exports (equation (14a)) and imports (equation 14b)). In the first case, the main channel is the ‘price-competitiveness of exports’: if an increase of the labour share boosts unit labour costs, then the price of exports (PX) increases too. This price-competitiveness loss could imply a fall in gross exports. Thus, being ${\epsilon }_B^A$ the elasticity of variable A with respect to variable B, we should estimate the relation between Ω and ULC $\left({\epsilon }_{\Omega }^{ULC}\right)$ ; the elasticity of export prices relative to labour costs $\left({\epsilon }_{ULC}^{PX}\right)$ ; and the elasticity of exports with respect to export prices $\left({\epsilon }_{PX}^X\right)$ .

In the second case, if the increase in unit labour costs is transferred into domestic prices (P), this could foster a process of substitution of domestic production by imports, depending on the price elasticity of imports. Nevertheless, we have found that a 1% growth in ULC is only translated in a 0.14% growth in prices. For this reason, unit labour costs are not significantly transferred into export prices: a reduction in unit labour costs, and hence in the labour share, does not imply a price-competitiveness gain (see Table 5 in Appendix 1).

Table 4 reports the total marginal effects of each aggregate demand component on GDP and the unemployment rate. As the profit share does not significantly affect investment, the effect reported below captures the indirect effect on investment driven by changes in functional income.

Table 4. Labour share effect on GDP and Unemployment (2010q2–2013q4).

Source: Author’s own calculations, based on Eurostat quarterly data.

GDP: gross domestic product.

All in all, the marginal effect of 1 pp increase in the labour share on aggregate demand (0.62 pp), when the multiplier is taken into account, amounts to a positive impact of 0.91%. Once the effect of the labour share on GDP [(∆Y/Y)/∆Ω] and that of GDP on unemployment $\left({\beta }^{*}\right)$ have been calculated (the Okun’s law coefficient), we estimate the direct effect of wage devaluation on unemployment during the recession. During the period 2010q2–2013q4, the unemployment rate increased by 5.84 pp, whereas the labour share decreased by –2.87 pp. According to our estimations, a fall in GDP by –2.61 pp, derived from wage devaluation, lifted the unemployment rate by 1.9 pp. This means that 32.6% of the unemployment rate increment between 2010q2 and 2013q4 occurred directly as a consequence of wage devaluation. Contrary to the claims of the pro-flexibility literature, this evidence points to a negative effect of labour market reforms in the unemployment rate.

Finally, as discussed in the ‘Revisiting Okun’s law’ section, the evolution of the labour force during the recession had an effect on the unemployment rate. Given the fall of the labour force by 0.4% during that period and the labour force elasticity of the unemployment rate derived from Table 3 ^{Footnote 5} the evolution of the labour force partially compensated for the increase of the unemployment rate by –0.2 pp.