Youth Unemployment in Advanced Europe
Okun’s Law and Beyond
Author: Angana Banerji, Ms. Huidan Huidan Lin, and Mr. Sergejs Saksonovs

Contributor Notes

Author’s E-Mail Addresses: ABanerji@imf.org; HLin@imf.org; SSaksonovs@imf.org
Publication Date:
21 Jan 2015
eISBN:
9781498382519
Language:
English
Keywords:
WP; business cycle; product market
Download PDF (2.3 MB)

The crisis has intensified what was previously a chronic unemployment problem in Europe; youth unemployment is now at unprecedented highs in some European countries. This paper assesses the main drivers of youth unemployment in Europe. It finds that much of the increase in youth unemployment rates during the crisis can be explained by output dynamics and the greater sensitivity of youth unemployment to economic activity than adult unemployment. Labor market institutions also play a significant role in explaining the persistently high levels of youth unemployment, especially the tax wedge, minimum wages relative to the median wage, spending on active labor market policies, the opportunity cost of working (measured by the unemployment benefits), vocational training, and labor market duality. This suggests that policies to address youth unemployment should be comprehensive and country-specific, focused on reviving growth and advancing labor market reforms.

Abstract

The crisis has intensified what was previously a chronic unemployment problem in Europe; youth unemployment is now at unprecedented highs in some European countries. This paper assesses the main drivers of youth unemployment in Europe. It finds that much of the increase in youth unemployment rates during the crisis can be explained by output dynamics and the greater sensitivity of youth unemployment to economic activity than adult unemployment. Labor market institutions also play a significant role in explaining the persistently high levels of youth unemployment, especially the tax wedge, minimum wages relative to the median wage, spending on active labor market policies, the opportunity cost of working (measured by the unemployment benefits), vocational training, and labor market duality. This suggests that policies to address youth unemployment should be comprehensive and country-specific, focused on reviving growth and advancing labor market reforms.

I. Introduction1

Youth unemployment has surged in Europe during the global crisis, and is at historic highs in many countries. Several papers have highlighted the deleterious consequences of high youth unemployment. These include the impact of “scarring” that leads to a lower probability of future employment and/or lower wages, which has been found to be persistent (Dao and Loungani, 2010; Ellwood, 1982; Gregg and Tominey, 2005; Kahn, 2010; Oreopoulos et al., 2012). High levels of youth unemployment have been shown to erode social cohesion and institutions (Giuliano and Spilimbergo, 2009; Altindag and Mocan, 2010; ILO, 2013), and foster crime (Fougère et al., 2009; Carmichael and Ward, 2001). In addition, young unemployed workers may have a higher propensity to relocate abroad because of lower “sunk costs” and greater potential for education and, subsequently, work abroad.2 The outward migration of the youth combined with the general ageing of the population may lower the potential output of the euro area since a significant share of the outward migration is to countries outside the euro area (OECD, 2013a). High youth unemployment and an outward migration of younger members of the workforce could also undermine the sustainability of spending on social safety nets in euro area countries, which is of particular concern given their demographic challenges.

In response, specific policies have been formulated at both the European Union (European Commission, 2012 and2013a) and national levels to deal with youth unemployment. The most notable examples are the EU’s Youth Guarantee Scheme, aimed at ensuring that all young people get a job offer within four months of leaving formal education or becoming unemployed, and the Youth Employment Imitative, aimed at supporting active labor market policies for young people not in education, employment or training in regions with high youth unemployment. Policies at the national level include, for instance, Italy’s July 2013 measures to support youth unemployment, including tax breaks for employers hiring under-30s on permanent contracts, and the increase in training, apprenticeship and internship schemes. Such policies aim to smooth the job-searching process and promote youth employment.

But what drives youth unemployment? Despite the focus on the harmful effects of pervasive youth unemployment and policies to address the problem, studies devoted to understanding the main drivers of youth unemployment are relatively limited. The large literature on labor markets usually examines overall unemployment, which has also been an important concern during the global crisis.

The few papers that do investigate youth unemployment and employment in advanced economies all point to the important role of economic activity and labor market institutions. Some find significant effects from multiple factors—including labor market reforms, economic freedom, education, part-time employment, and active labor market policies (Choudhry et al., 2012a). Others highlight the role of individual labor market institutions such as the union involvement in wage setting (Jimeno and Rodriguez-Palenzuela, 2002; Bertola et al., 2007), labor market flexibility (Bernal-Verdugo et al., 2012a), or employment protection legislation (Bassanini and Duval, 2006) for youth labor market outcomes.

Our paper examines the main cyclical and structural explanatory factors behind the youth unemployment dynamics during the global crisis. It covers 22 advanced European countries—18 in the euro area, as well as Denmark, Sweden, Norway and the United Kingdom, but the discussion places a special emphasis on the euro area. The sample period is from 1980 to 2012, with the actual size subject to data availability, in particular for labor market features.

The paper makes several contributions to the empirical literature. First, our analysis compares and contrasts the effects of various factors on both youth and adult unemployment, and the changes and levels of the unemployment rate. Second, our methodology allows us to take fuller account of output fluctuations on youth unemployment. Like many other studies on similar issues, we also utilize panel data to control for unobserved time-specific or country-specific characteristics. However, unlike these studies which assume that output fluctuations have a common effect across countries, we estimate the country-specific sensitivity of unemployment to output fluctuations and also control for such country-specific output-sensitivity when studying the role of labor market institutions. There is evidence that the sensitivity of unemployment to output varies significantly across countries (e.g., Ball et al., 2013).

Moreover, this paper investigates examines whether labor market factors amplify or dampen the sensitivity of unemployment to output fluctuations in each country. In the context of the current crisis, a few papers have studied how financial crises affect youth and how labor market institutions affect the reaction of youth unemployment to crisis (e.g., Choudhry et al., 2012b). However, they mainly focus on recessions (rather than output fluctuations in general) and only incorporate select labor market factors into the analysis (e.g., labor market flexibility in Bernal-Verdugo et al., 2012b, and employment protection legislation, O’Higgins, 2012). Our paper covers a broad range of labor market institutions that affect both labor demand and supply. Also, it looks at how each of a broad set of labor market factors affect the sensitivity of youth unemployment rates to the output gap, including during downturns. We do so by introducing an interaction term of output gaps and labor market institutions.

The results suggest that youth unemployment rates are more sensitive to output fluctuations than adult unemployment, and the significant increase in youth unemployment during the crisis, especially in vulnerable euro area economies, was largely due to the decline in economic activity. In addition, the persistently high levels of youth unemployment could be also explained by a host of labor market institutions, including: the opportunity cost of working (measured by the unemployment benefits), labor costs (measured by the tax wedge and minimum wage relative to the median wage), spending on active labor market policies (ALMPs), vocational training, and labor market duality. Finally, labor market features do not generally affect the sensitivity of youth unemployment to output gaps.

The remainder of the paper is structured as follows: Section II documents stylized facts about labor market dynamics in advanced European countries. Section III describes empirical methodology and data. Section IV summarizes the empirical results, and, finally, Section V concludes.

II. Stylized Facts

The crisis has left a sizable dent in euro area output and investment, which dropped by 2 and 19 percent respectively since their pre-crisis peak. The hardest hit sectors were construction, retail and wholesale sectors, and manufacturing, with sectoral output lower than their pre-crisis peaks by 20 percent, 5 percent, and 4 percent, respectively.

Labor market outcomes have been dismal: about 4 million jobs were lost across the advanced Europe during this time period, with unemployment rates—youth unemployment in particular—rising to unprecedented levels.3 The crisis has reversed a decade-long trend of modest declines in youth unemployment—the youth unemployment rate in the euro area remains elevated at 24 percent in mid-2014, well above the 15 percent rate in 2007. The adult unemployment rate has also risen significantly, but less so than youth unemployment (10¼ percent in mid-2014, from 6½ percent in 2007). The higher youth unemployment rate only partly reflects the fact that the youth labor force is typically smaller than the adult labor force; the gap between youth and adult unemployment rates has increased significantly after the crisis.4 The NEET rate, defined as the share of the youth population that is not employed or involved in further education or training, is less influenced by measurement biases inherent in determining the size of the youth labor force. Even the NEET rates show that the mild improvements that had started before the crisis have now been reversed.

A01ufig01

Euro Area: Unemployment Rates and NEET

(Percent)

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: Eurostat

Across the euro area, there are large cross-country differences in the level and change in youth unemployment, and these differences have increased during the crisis. The current unemployment situation reflects a complex mix of developments during the crisis, combined with a variety of pre-crisis unemployment rates. Taking these differences into account, euro area countries can be placed into four buckets: (i) those that witnessed large increases in the youth unemployment during the crisis from relatively low pre-crisis levels (Ireland, Cyprus); (ii) those with above-average unemployment rates before the crisis, but small increases since the crisis (Belgium, France, Finland and Sweden); (iii) those with the worst of both of the above categories—large increases in youth unemployment after the crisis, but relatively high rates to begin with (Greece, Spain, and Italy); and, (iv) those with the best of both worlds (Austria, Netherlands, Germany), with small increases in unemployment (or even a decline in the case of Germany) on top of low pre-crisis unemployment rates to begin with.

Youth employment conditions are fragile, tenuous and concentrated. The young are far more likely to be hired on temporary contracts than adults in all countries, but particularly in Spain, Italy and Portugal, countries which experienced some of the largest increases in youth unemployment during the crisis. Youth employment is also concentrated in sectors which tend to be more sensitive to the business cycle: manufacturing, wholesale and retail trade, and hotels and restaurants. Before the crisis, these sectors comprised 65–75 percent of youth employment in countries where youth unemployment increased the most after the global crisis. In the aftermath of the crisis, the top three sectors below together still employed on average 50 percent of the youth and 40 percent of adults. As of 2013, wholesale and retail trade was the dominant sector for youth employment, followed by manufacturing and accommodation and food services.5

A01ufig02

Share of temporary employment, 2013, percent

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: Eurostat
A01ufig03

Youth Employment By Sector, 2000-07, percent

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: Eurostat
A01ufig04

Share of employment in top 3 areas of economic activity, 2013, percent

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

III. Data and Methodology

A. Variable Selection and Data Coverage

Measure. In line with the literature, this paper uses the unemployment rate as a main measure of youth and adult unemployment. This enables us to exploit existing theoretical frameworks for our analysis, and allows easier comparison with the literature. Some have argued that the unemployment rate is not informative and is a biased measure of the incidence of unemployment given the smaller and more volatile size of the youth labor force. But other measures also come with their own challenges.

Alternative measures. Other unemployment indicators—the NEET rate or the unemployment ratio6—can be problematic to use. First, it is challenging to construct the appropriate analytical framework that could allow these measures to be used for empirical analysis, which is, perhaps, also why they are seldom used in the literature. Second, these indicators pose statistical challenges. The NEET rate does not allow for a comparison with adult unemployment as it is only defined for the 15–34 years old age group. The definition of NEET varies across countries,7 making cross-country comparisons difficult. As for the unemployment ratio, it is perhaps as biased as the unemployment rate—both indicators ignore considerations such as: the unemployed not joining the labor force and remaining in education because they could not find a job, or returning to university to avoid a potentially scarring unemployment spell, even though they would have preferred to work; those without a job who are not registered as unemployed; or, those that migrate abroad to study and work. Moreover, the unemployment rates and ratios are closely correlated (with a correlation coefficient of 0.89), suggesting that using one over the other is unlikely to fundamentally change any conclusions regarding the dynamics of youth unemployment.

Measurement of labor market factors. For robustness, a number of different measures are used for each of the following categories of labor market features—labor costs; the opportunity cost of working; collective bargaining; labor market duality; education and training; and spending on ALMPs. In each category, the analysis relied on several indicators and different sources (see Table 1 for definitions). For instance, labor costs were measured by the tax wedge and the ratio of the minimum wage to the median wage under the assumption that the former was most relevant for the youth. Several countries in the euro area do not have minimum wages and this ratio was set to zero in the data base. Similarly, the opportunity costs of employment are measured by gross benefit replacements rates, net replacement rates (benefit replacement rates net of taxes) and “the inactivity trap” which arises when tax and benefit systems interact to reduce the financial incentive of individuals who qualify for social protection benefits to work.

Table 1.

Data Definitions

article image

Data challenges. There are several constraints. Some arise from the challenges inherent in capturing the structural characteristics of labor markets in a manner that captures country-specific institutional context. Others arise from the need to measure labor market factors consistently across a broad range of countries and over a long period of time to maximize the sample size.

  • Sample size. Data on the selected labor market institutions is usually not available for the full sample period, and is especially limited for new entrants to the euro area. The smaller sample size implies that country-by-country empirical analysis combining the institutional variables and measures of the business cycle together would not yield robust results. For instance, the data series for some variables (e.g., ALMPs) starts from late 1990s for most countries, and even later for smaller countries; some variables (e.g., OECD’s index for employment protection legislation) are missing in some countries, and other series (e.g., tax wedge) only start from early 2000’s for all countries. This cuts the total sample size significantly, and in some cases, we have to use fewer variables than preferred.

  • Duration of the analysis. Given the long lags in data availability, it is difficult to extend the analysis beyond 2012 at this juncture since only a subset of the labor market variables is available. Thus, labor market reforms since the crisis are not fully captured in the dataset, e.g., Spain’s major labor market reforms of 2012 (Ley 3/2012 de medidas urgentes para la reforma del Mercado laboral) which included reforms to collective bargaining, opt-outs and firm level internal flexibility regarding working conditions (wages, hours), and dismissals (conditions, costs).

  • Measurement and relevance. The analysis had to use data series that were available for a large set of countries over a long period of time. But some of these variables are not always relevant for the young unemployed (e.g., the unemployment benefit system), nor do they capture the country-specific institutional context that may be driving unemployment. Some variables only partially capture the elements that might have an impact on unemployment. For example, the OECD’s indicator on union density measures the incidence of unionization among the employed, but does not measure the degree of centralization or the coverage of unions. Another example is measuring the skill levels of unemployed workers. The level of formal education may not provide a complete picture of a worker’s skills; vocational training and apprenticeships are also important but are difficult to measure. The analysis was based thus on both measures of skills, and survey data on the reasons for temporary work contracts was used as a proxy for the incidence of vocational training. Last but not least, product market flexibility was measured using the Global Competitiveness Index, but this index includes components such as the intensity of local competition, extent of market dominance and anti-monopoly policy, whose effect on unemployment rates may be ambiguous.8 On the other hand, alternative measures such as the progress in reducing the cost of starting a business according to the World Bank Doing Business indicators is less comprehensive.

  • Slow moving. Finally, by their very nature to some extent, structural variables change very slowly from year to year, which makes it difficult to study the impact of reforms on labor market dynamics. Also, labor and product market features have remained relatively unchanged since 2000 for many of the countries in the sample (Figure 1) because of lack of major reforms in several countries, regardless of which indicator is used. In general, these indicators seem to have remained broadly the same or even deteriorated in countries that have witnessed the highest levels and the biggest increases in youth unemployment. Germany and the Scandinavian countries are among the few that have registered improvements.

Figure 1.
Figure 1.
Figure 1.

Labor and Product Market Institutions

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: Eurostat, European Commission, OECD, World Economic Forum, World Bank Doing Business Report, authors’ calculations
A01ufig05

Reason for Temporary Work Contract Among Youth, percent of unemployed, 2012

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: EurostatNote: missing red dot indicates missing data.

B. Methodology

There were several methodological constraints. These limitations arose as a result of the data challenges discussed in the previous section and the uncertainties involved in estimating equilibrium values of potential output and unemployment. As a result, the specifications considered below did not take into account non-linear and lagged effects of output and labor market indicators. In addition, significant simplifying assumptions had to be made about the country-specific effects of labor market features.

Okun’s Law estimation

The Okun’s Law, proposed by Arthur Okun in 1962, is the empirical regularity that changes in unemployment rates and output growth are negatively related.9 Many studies confirm this relationship for overall unemployment but research on youth unemployment is less common. Some authors highlight the higher sensitivity of youth unemployment to the business cycle (OECD, 2006; Scarpetta, 1996; Scarpetta et al., 2010; and European Commission, 2013b).

Standard specification. The Okun’s coefficient for individual countries is estimated on the basis of the following specification, where β2,i is the estimated value of the Okun’s coefficient for country i (Table 2):

Table 2.

Okun’s Law Estimates

article image
Sources: Authors’ estimates
Δui,t=β0+Σi=222β1,ici+Σi=122β2,iciΔyi,t+εit,(1)

  • – where Δui,t is the change in youth or adult unemployment rates in country i, year t

  • – ci is the dummy variable for country i (other than for country 1 to avoid perfect collinearity)

  • – Δyi,t is output growth, and,

  • – εit is the error term with standard assumptions.

Alternative specification. We also consider an alternative specification using the output gap (yi,t−yi,t*) and deviation from natural rates of unemployment (ui,t−ui,t*), respectively, as independent and dependent variables. β2,i is the Okun’s coefficient in this model (with estimates provided in Table 2):

ui,t−ui,t*=β0+Σi=222β1,ici+Σi=122β2,ici(yi,t−yi,t*)+εit,  (1a)

Ball et al. (2013) argue in favor of the specification (1a) on the grounds that specification (1) is only theoretically valid if one is willing to assume a constant equilibrium unemployment rate as well as constant potential growth. However, estimating the output gap and deviation from structural unemployment in specification (1a) is difficult. We use output gap estimates from the IMF’s World Economic Outlook database and apply the HP filter (with a smoothing parameter of 100) to get the natural rate of unemployment. The HP filter suffers from the well-known problems of unreliability at the end of the sample, and using output gap measurement from WEO allows us to sidestep this problem somewhat for potential output, but not for the natural rates of unemployment. Therefore, the specification in differences in equation (1) is our referred method for estimating the Okun’s coefficient.

Disaggregation. The idea of disaggregating the Okun’s Law specification into the impact of various expenditure components was proposed in Anderton et al. (2014), which, however, did not differentiate between youth and adult unemployment. We apply this methodology as well to see which component of growth—consumption, investment and exports—has the largest effect on unemployment. For practical purposes, this involves replacing Δyi,t growth of various expenditure components (namely, growth in consumption, investment and exports). The results are listed in Table 3.

Table 3.

Okun’s Law: Decomposition of Expenditure Components

article image
Note: To conserve space intercepts for decomposition equations are not reported, but are available upon request.Source: Authors’ estimates

The role of labor market institutions

Due to the previously discussed data challenges, it is not feasible to estimate country-specific effects of several labor market factors simultaneously, while also allowing for individual country effects of the business cycle (measured by the output gap). As a result, this paper considers two model specifications.

Multivariate approach. This specification considers the effect of several labor market features (and output gap) simultaneously, but restricts their impact to be the same across countries, while allowing the impact of the business cycle to vary across countries. More formally, the specification for the multivariate approach is given by the following model with the estimates provided in Table 4:

Table 4.

Multivariate Model Estimates

article image
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.Homoskedasticity is tested for each regression. If rejected, robust standard errors are estimated. Equation (2) was estimated for each column. This table reports only the coefficient before structural variable x.Source: Authors’ estimate
ui,t=β0+β1,ici+Σjγjχi,j,t+Σi=122β2,ici(yi,t−yi,t*)+εit,  (2)

  • – where ui, t is the level of the youth or adult unemployment rate,

  • – ci is the dummy variable equal to 1, if dependent variable is from country i,

  • – (yi,t−yi,t*) is the output gap, and

  • – xi, j, t represents labor market institution j, in country i, year t.

Univariate approach. An important drawback of the multivariate specification (equation (2) above) is the reduction in sample size as additional labor market variables are added to the list of x. To circumvent this limitation, equation (2) was also estimated for each labor market feature individually. The results of this approach are listed Table 5.

Table 5.

Univariate Model Estimates without Interaction Term

article image
Standard errors in parentheses.Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1Homoskedasticity is tested for each regression. If rejected, robust standard errors are estimated. Each structural variable enters the following regression individually. The regression below was run with each of the 13 measures of labor market features included one at a time. This table reports only the coefficient before structural variable x due to space constraint.ui,t=β0+β1,ici+β2χi,j,t+Σi=222β2,ici(yi,t−yi,t*)+εit,   where  j=1, …,13Source: Authors’ estimate

Univariate approach with interaction term. The trade-off between maximizing the sample size and incorporating variable effects of labor market factors in each country also motivates our second univariate approach.

This approach also considers the impact of one labor market feature at a time but allows its impact to vary across countries via its interaction with the business cycle. We consider the following specification, with the same notations as in (2):

ui,t=β0+β1,ici+β2χi,t+Σi=122β3,ici(yi,t−yi,t*)+Σi=122β4,iciχi,t(yi,t−yi,t*)+εi,t,  (3)

This specification includes one additional term—the interaction of the labor market variable institutional variable, output gap, and the country dummy. It is based on the assumption that (i) labor market features may affect the way unemployment rates responds to the business cycle, and this impact may vary across countries; and the simplifying assumption that (ii) the effect of the labor market factor remains the same across countries, except indirectly via its impact on the business cycle.

From equation (3), the marginal impact of the change in labor market feature xi, t on the level of unemployment is given by the partial derivative:

∂ui,t∂xi,t=β2+β4,ici(yi,t−yi,t*),

Thus, the impact of a change in any labor market feature differs across countries. Crucially, the marginal effects of a unit change in variable labor market factor will then depend on the value of the output gap yi,t−yi,t* at which they are evaluated. This point is calculated at the country-specific average output gap during the sample period. The standard errors for the marginal effects are computed using the delta method. The results of this approach are summarized in Figure 2.

Figure 2.
Figure 2.
Figure 2.

Marginal Impacts of Labor Market Institutions on Unemployment Rates, percentage point

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Source: Authors’ estimates

Robustness check. A variety of additional specifications were considered, e.g., using the total unemployment rate and employment rates for different age groups as dependent variables. In addition, we have considered models with changes in labor market institutions as independent variables and output growth instead of the output gap as the independent variable, to mimic the standard Okun’s law formulation in (1) more closely.

Levels versus changes. Our preference ultimately lies with models that feature levels (not changes) of unemployment rate as dependent variables. A number of theoretical models of unemployment are consistent with using unemployment rate in levels as dependent variable in reduced-form equations. For example, Nickell and Layard (1999) develop a wage bargaining model with numerous identical firms, showing that equilibrium level of unemployment rate will be decreasing in any exogenous factor that increases job separation rate (represented in our case by the output gap), increases the search effectiveness of the unemployed (represented by ALMPs), lowers the benefit replacement ratio, lowers the strength of the workers in the wage bargain (union density) or raises the elasticity of product demand facing the firm. The latter argument even suggests some scope for including variables associated with product market regulation into unemployment equations. Other examples of similar models include Scarpetta (1996) and Bassanini and Duval (2006), who estimate a specification very similar to ours.10

IV. Empirical Results

This section discusses the main results from the various specifications described in Section III. Some of these results need to be interpreted with caution given the methodological and data constraints already discussed. Moreover, it would be important to remember that these results do not measure the impact of the reforms undertaken in countries after the crisis as they fall outside the sample period. Some results are counterintuitive and difficult to explain without additional granular information about other relevant and country-specific factors.

A. Estimating the Okun’s Coefficient

Standard Okun’s coefficient. We find unambiguous evidence validating the Okun’s law across a wide range of countries. This is true regardless of how output fluctuations are measured—by real GDP growth or the output gap. In equations (1) and (1a), cyclical factors explain around 50 percent of the variation in the changes in youth unemployment rates and around 60 percent for adult unemployment rates across all advanced European countries, as measured by the R-squared (Table 2). Youth unemployment is more sensitive than adult unemployment in every country, reflecting their relatively fragile unemployment conditions.

Disaggregated Okun’s coefficient. In order to find out which component of growth relates most to youth and adult unemployment, the Okun’s law regression (1) was generalized and estimated by using growth rates of consumption, investment and exports as independent variables. The results indicate that, for most countries in the sample, youth and adult unemployment rates are most sensitive to consumption growth but much less sensitive to export growth in many countries (Table 3). This is broadly consistent with the findings by Anderton et al. (2014), which argues that “this reflects the highly labor-intensive nature of the services that represent the bulk of consumers’ expenditure, while the higher productivity manufacturing-related content of exports tends to be less labor intensive.”

A01ufig06

Euro Area: Okun’s Law Coefficients /1

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

/1 Shaded bars and circles reflect estimations that are not significant.
A01ufig07
Sources: Eurostate WEO; authors’ estimates. Unfilled shapes indicate statistically insignificant coefficients.

Greater sensitivity of youth. All estimates of the Okun’s coefficient suggest that youth unemployment is much more sensitive to the business cycle than adult unemployment. In every country, the estimated Okun’s coefficient is, on average, two to three times as large for the youth than for adults. It is partly because a smaller labor force means youth unemployment rates necessarily react more sharply than adult unemployment rates to a change in the GDP growth or output gap. However, it could also be explained by the greater concentration of youth employment in cyclically sensitive industries and perhaps in small- and medium-sized enterprises (SMEs) (Box 1).

SMEs: Financing Constraints and Youth Unemployment

Small- and medium-sized enterprises (SMEs)11 employ the majority of the labor force in Europe, with the employment share especially high in some stressed economies which experienced high increases in youth and adult unemployment rates—more than 75 percent for Spain, Italy and Portugal, for example. SMEs are usually more financially constrained than larger firms, particularly in the current environment of financial fragmentation and private sector deleveraging.

The Okun’s law framework is augmented with financial constraints—measured as either the share of firms in industry and services sectors which report financial constraints as a factor limiting production in surveys conducted by the EC or the average share of SMEs in an economy. The surveys do not list SMEs as a separate category, but evidence from other data sources, e.g. bank lending surveys, suggests that they are more affected by financial constraints than larger firms.1

Greater financial constraints are associated with higher youth unemployment.

  • Controlling for country-specific fixed effects and output gap, an additional percentage point of firms reporting financial constraints is associated with higher youth unemployment rates by 0.3 and 0.4 percentage points (for industrial and services sector respectively). The effect on adult unemployment rates is smaller (0.2 percentage points).

  • A 10-percentage-point increase in the average employment share in SMEs (or the SME share of value added) lowers the Okun’s coefficient by 0.1, making unemployment more pro-cyclical.

A01ufig08

SME Share in Employment, percent

(Average, 2008-2013)

Citation: IMF Working Papers 2015, 005; 10.5089/9781498382519.001.A001

Sources: European Commission; authors’ calcuations
1 There are data limitations. The average share of SME up to 2008 is used for each country because the data is available only from 2008 and stays broadly constant. The percentage of firms reporting financial constraints is small (often zero) and relatively unchanged, e.g., before the financial crisis only 2 percent of industrial firms reported financial constraints, on average, compared to 2012–2013 when the average was 9.3 percent.

Cross-country variation. The Okun’s coefficient, i.e., the sensitivity of unemployment rates to output fluctuations, varies across countries. Estimates range from not significantly different from zero (e.g., Austria, Germany12) to -1.9 in Spain, i.e., a one-percentage-point reduction in growth increases youth unemployment rates by almost 2 percentage points. The ranking in the size of the Okun’s coefficient of the countries is consistent with the estimates reported in Ball et al. (2013), who document the largest coefficient for Spain and the smallest one for Austria.13

Intercept. The intercept in the Okun’s law equations (1) and (1a) also has an interesting interpretation, which is sometimes overlooked. The intercept (including the fixed effect component) would equal the change in unemployment rate when economic growth is zero. Ball et al. (2013) point out that under the assumption of constant potential growth, the intercept equals the product of minus the implied potential growth rate and the estimated Okun’s law coefficient. In other words,

β0+β1,i=−β2,iΔy* in equation  (1)

Because the Okun’s law coefficient is empirically expected to be negative, this suggests that the intercept in equation (1) should always be positive. The estimates in Table 2 confirm that this is indeed the case. Intuitively, a positive intercept implies that if actual growth is zero, while potential growth is positive, the unemployment rate is expected to increase because of a wider output gap. High values of the intercept terms suggest greater difficulties in sustainably reducing unemployment, or, put differently, a stronger growth is required to prevent unemployment rates from rising. Spain has the highest Okun’s law intercept for youth unemployment exceeding 5 percentage points, with intercepts for Cyprus, Estonia, Greece, Ireland, Latvia, Portugal, Slovakia, and Sweden exceeding 2 percentage points. Only for Austria, Germany, Netherlands, and Norway are intercepts not significantly different from zero. On the other hand, the intercepts in equation (1a) (also including fixed effects) are almost all insignificant with the exception of Latvia. This is also consistent with economic theory—when output gap is zero, the deviation of unemployment rate from its natural level should be zero as well.

Business cycle effects across specifications. How does the impact of the business cycle or the output gap differ across specifications, i.e. between equation 1(a) which excludes the effect of labor market factors, and (2) and (3) that includes labor market factors? The estimations are not comparable because of very different sample sizes. The estimates of the Okun’s coefficient according to equation 1(a) are based on 565 observations (Table 2). The output gap coefficients in equation 2, on the other hand, are based on samples sizes that range from 99–160 observations (Table 4); these results are less robust and have therefore not been reported in this paper to avoid confusion. It is therefore not meaningful to compare the results across specifications to derive the additional explanatory power from the inclusion of labor market factors in the estimations, all the more so because of the simplifying assumption that labor market factors have the same effect across all countries.

B. Estimating the Role of Labor Market Features

Output changes, on average, explain about 50 percent of the increase in youth unemployment rates during the crisis, as shown in Table 2. Drawing on the labor market literature, this section examines the role of other possible explanatory factors, in particular the structural features of the labor market.

Empirical results. Changes in labor market variables were found to not be significant explanatory factors for the increase in youth unemployment rates. The lack of a robust relationship between labor market factors and changes in the unemployment rate could simply reflect the empirical, specification and data challenges that have already been discussed as well as the inability of the chosen labor market variables to capture country-specific institutional details that may also be relevant in individual countries.

However, the results of the multivariate approach (equation (2)) suggest that a number of labor market features are significantly associated with the levels of youth unemployment rate (Table 4). In particular, lower labor costs (measured by the tax wedge and minimum wages relative to the median wage) and higher spending on ALMPs, especially for training, are associated with lower unemployment. Higher opportunity costs of working (measured by gross and net benefit replacement rates) and stronger labor market duality (indicated by higher shares of temporary workers and lower employment protection of temporary workers) tend to raise unemployment, whereas collective bargaining (proxied by union density) has mixed effects which are not robust.

Robustness. The above results are, by and large, confirmed by the univariate approaches as well. However, there is a wide divergence in the range of estimates in part due to different sample sizes used for the different estimations because of data constraints. Considering one variable at a time allows the utilization of the whole sample for which data is available for the regressor variable whereas in a multivariate specification the length of the sample is determined by the sample size of the variable for which the data is least available. Thus, the sample size can generally be much larger for the univariate estimations than for the multivariate estimations. However, an important caveat to the results of the univariate regressions is that they may be prone to omitted variable bias when there are significant correlations among different labor market features which affect unemployment. Subsequent references to the estimation of the impact of labor market features generally refer to the multivariate specifications reported in Table 4, unless otherwise specified.

Greater hiring costs are associated with higher unemployment rates in the literature. Taxes on employers and employees, together with high statutory minimum wage rates increase labor costs and reduce labor demand. Our results indicate that a one-percentage-point increase in the tax wedge is associated with an increase in youth unemployment rates by 0.3-1.3 percentage points. The size of the correlation with adult unemployment is smaller (around 0.4–0.5 percentage points). A one-percentage-point increase in minimum wages (relative to median wages) is associated with an increase in youth unemployment by 0.4-1.2 percentage points, while the magnitude is smaller for adults. This is consistent with the finding of OECD (2012) that since 2007 the young have on average been at a big disadvantage in countries where the minimum wage is relatively high as a share of median pay.14

Generous unemployment benefits can raise unemployment by reducing the willingness to search for jobs and/or increasing the reservation wage to accept job offers.15 Our results indicate that youth and adult unemployment rates tend to be higher when the opportunity costs of working are high. A one-percentage-point increase in the gross replacement rates16 is associated with higher youth unemployment by 0.4–0.5 percentage points and higher adult unemployment rates by 0.1–0.2 percentage points.

An inactivity trap can arise through the interaction of tax and benefit systems. High social protection benefits can reduce financial incentives to work because the combined effects of increased tax payments and withdrawn income-tested benefits offset the potential gain in disposable incomes from increased earnings. Higher inactivity trap is not significantly related to unemployment when other labor market features are controlled for (Table 4), but is associated with higher unemployment when investigated alone (Table 5).

Theoretically, labor market duality—a high share of temporary employment contracts together with lower employment protection for the temporary workers—has an ambiguous impact on unemployment. This is because stronger employment protection lowers not only job separation rates by increasing the cost of firing, but also the job finding rate by increasing the reluctance to hire workers in the first place. Our estimates suggest that a stronger labor market duality can have a detrimental effect on the employment of young people. Higher employment protection for temporary contracts is related with lower unemployment rates for youth and adults, with stronger effects for the youth given their higher share of employment on temporary contracts.17 A one-unit increase in the rating for employment protection legislation (EPL)18 is associated with a decline in youth unemployment rates by 2.5–5 percentage points and adult unemployment by 1.5–2 percentage points. A one-percentage-point increase in the share of youth on temporary contracts is associated with an increase in youth unemployment by 0.3–0.4 percentage points, but the relation with adult unemployment rates is not statistically significant.

Overall, the role of higher union density is found to be unclear and not robust. A one-percentage-point increase in union density could be associated with lower youth unemployment rates by 0.2–0.6 percentage points under some specifications.19 However, this finding is not robust to alternative specifications, including using different control variables or allowing country-specific interactions. The results from specifications with employment as dependent variable20 suggest that higher union density may be associated with an altered employment composition, perhaps because employers prefer to hire the non-unionized youth instead of the unionized adults in order to preserve the flexibility to adjust the work force as needed, leading to higher employment for the youth and lower employment for adults.

Educational attainment tends to improve employability (OECD, 2013b). Vocational training and expanded access to training could help make temporary jobs a stepping stone toward open-ended contracts (OECD, 2004). Our results indicate that access to vocational training—measured by the share of temporary workers under probation or vocational training—is associated with lower youth unemployment by around 0.3 percentage points, but is not statistically significant for adult unemployment. A higher share of individuals with low levels of education has generally no significant effects on youth unemployment or employment rates, but has a strong negative effect on adult unemployment and employment rates. Low education may be less of an obstacle for youth employment, perhaps because young workers can be more easily trained than adults.

Most empirical studies using macro-level data have found that ALMP spending, especially on training, reduces total unemployment (OECD, 2006). We find that higher spending on ALMPs, especially on training, is associated with reductions in both youth and adult unemployment rates. An additional thousand euros per unemployed increase in ALMPs spending is associated with lower youth and adult unemployment rates by around 0.3 and 0.1 percentage points respectively.21

Impact on sensitivity to business cycle. Allowing for interactions between labor market features and output fluctuations, as specified in equation (3), reveals significant cross-country differences in the role of labor market features. The coefficient before the interaction term of the output gap and the labor market variable would indicate the direction and magnitude of such effect. Given that the Okun’s coefficient is negative, a negative (positive) coefficient would imply that a particular labor market variable amplifies (dampens) the impact of the business cycle on unemployment (Table 6). Our analysis, for the most part, finds only limited evidence that labor market features change how youth unemployment responds to output fluctuations. For instance, a higher tax wedge is found to amplify the impact of the business cycle for Finland, Greece, Ireland, Slovakia and Spain, and dampen the impact for Denmark, Sweden and the United Kingdom (column 5, Table 6). Similarly, total ALMP spending is found to dampen the business cycle effect in Ireland, Slovakia, and Spain, but amplify the impact in Finland, France, and Malta. Some of these results may seem counterintuitive and cannot be properly explained without a detailed assessment of country-specific factors that may be influencing the results but are not adequately captured in the model specification.

Table 6.

Univariate Model Estimates with Interaction Term

article image
article image
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Homoskedasticity is tested for each regression. If rejected, robust standard errors are estimated. Equation (3) was estimated for each column. Each structural variable enters regression individually. Table reports only the coefficient before structural variable x and the coefficient before the interaction of country dummy, x, and output gap.Source: Authors’ estimate