Gross Job Flows in Europe

Transcription

1 Gross Job Flows in Europe Ramón Gómez-Salvador (1) DG-Economics European Central Bank Julián Messina (2) DG-Research European Central Bank Giovanna Vallanti (3) * Centre for Economic Performance London School of Economics May 2003 Abstract We examine job flows in the 1990s for a sample of 12 European countries. Using a firm dataset that covers all sectors, we find firm characteristics to be important determinants of job flows, with smaller and younger firms within services typically having a larger degree of job turnover. Once controlled for firm and sectoral effects, the role of institutions in the dynamics of job creation and destruction is examined. As expected, employment protection is found to reduce job flows, although this is partially counterbalanced through the use of temporary employment. Similarly, countries with higher unemployment benefits, labour taxes and more coordinated wage bargaining systems present are characterised by lower job flows. Keywords: Gross Job Flows, labour market institutions. JEL-Codes: J23, J60 (1) Ramón Gómez-Salvador. European Central Bank, Kaiserstrasse 29, D Frankfurt am Main. Fax: Tel: ramon.gomezsalvador@ecb.int. (2) Julián Messina. European Central Bank, Kaiserstrasse 29, D Frankfurt am Main. Fax: Tel: julian.messina@ecb.int. (3) Giovanna Vallanti. Center for Economic Performance, 10 Portugal Street, London WC2. Fax: Tel: g.vallanti@lse.ac.uk. * We are grateful to Neale Kennedy, Alfred Stiglbauer and seminar participants at the ECB, Federal Reserve Bank of Boston and LSE for their comments and suggestions. We also thank Jean-Paul Genot for data assistance. The views presented in this paper are those of the authors and do not necessarily reflect the views of the European Central Bank. All the errors and omissions are our own.

2 1 Introduction Recent theoretical and empirical literature has stressed the importance of job reallocation in a world where agents (firms and workers) are heterogeneous and the matching process between vacancies and workers is costly. When a shock hits the economy, the desired allocation of jobs among firms and sectors changes, leading to job destruction on the one hand and the creation of new vacancies on the other. Because of heterogeneity and other labour market frictions, new vacancies and unemployed workers do not match instantaneously, implying spells of unemployment and vacant positions in the economy (Pissarides, 2000). Gross job flows may be considered a proxy for labour market flexibility to the extent that they provide a measure of the responsiveness of the labour market to changes in economic conditions. In recent years, several studies have estimated job creation and destruction from longitudinal data at plant or firm level. Studies on gross job flows have shown that a high number of jobs are simultaneously created and destroyed in the economy even when the employment growth is close to zero. 1 This provides evidence on the complexity of the dynamics underlying the adjustment process in the labour market and the heterogeneity in the behaviour of both workers and firms. The main limitation of the existing studies on job flows is the lack of internationally comparable job flows statistics (OECD 1994). A number of problems arise when using establishment/firm level data, which became of particular concern when doing international comparisons. Differences in definitions, sampling intervals, sectoral coverage and sampling frame may lead to misleading interpretations of the cross-country differences in estimated job flows. We examine time series and cross-sectional patterns of job flows for 12 European countries using a unique homogeneous firm dataset that covers the whole spectrum of productive sectors. First, we provide comparable estimates of job flows and examine cross-countries differences and regularities. Job flow measures in relation to firm characteristics are reported in order to identify the patterns of job reallocation among different groups of firms within and between the countries studied. We find important regularities across countries, where smaller and younger firms concentrated in services exhibit the larger job turnover. Next, we examine the relationship between job flows and the business cycle. As expected, job creation shows clear pro-cyclical patterns, while job destruction moves countercyclically. Within the countries studied, the first effect outweighs the latter, resulting in a pro-cyclical movement in job reallocation. This suggests an important role of firing restrictions in Europe, which by obstructing the layoff decisions of the firms as discussed in Garibaldi (1998) might make job destruction more time consuming than job creation. The last section of the paper focuses on the impact of institutional factors in the determination of job turnover. Once controlled for compositional effects we find persistent cross-country differences that can be partially explained by institutional features. As expected, we find a clear negative effect of policies aiming to protect jobs in the dynamics of job creation and job reallocation. These effects are reinforced by policies that might increase the cost of labour such as the tax burden and generous unemployment benefits. However, some countries with stringent employment protection like Spain seem to be making an extensive use of temporary contracts in order to gain flexibility at the margin, resulting in larger aggregate job flows. The remainder of the paper is organized as follows. In section 2, we present the theoretical motivations of our study and the most relevant empirical evidence. Section 3 describes the data set used in the analysis and defines concepts and measures of gross job flows. In section 4, we describe gross job flows for different firm characteristics and extend the analysis to the multivariate framework in order to uncover the main driving factors of labour dynamics. Once controlled for sectoral characteristics, Section 5 assesses the role of institutional features in explaining persistent cross-country differences in gross job flow patterns. Section 6 concludes. 1 For a thorough discussion of the results in this literature, see the excellent survey of Davis and Haltiwanger (1999).

3 2 Theoretical motivations and empirical evidence 2.1 Job flows: international comparisons There is a large literature aiming to explain the magnitude and cyclical behaviour of job reallocation and its components. Empirical studies on job flows include Davis and Haltiwanger (1992), Davis et al. (1996) and Haltiwanger and Schuh (1999) for the US manufacturing industry, Blanchflower and Burgess (1996) for the UK, Broersma and Gautier (1997) for the Netherlands, Albaek and Sorensen (1998) for Denmark, Lagarde et al (1994) for France, Dolado and Gomez (1995) for Spain, Contini et al. (1991) for Italy, Stiglbauer et al. (2002) for Austria, Faggio and Konings(2001) for 5 accession countries and Contini et al. (1995) for countries of the European Union. In addition OECD (1994) and OECD (1996) reports results on job flows for 10 OECD countries between the late 1980s and early 1990s. The main findings of this literature can be summarised as follows: 1. A high number of jobs are simultaneously created and destroyed in all countries and sectors regardless of the cycle phase; 2. Job creation and destruction are negatively correlated but not perfectly. This implies that, although job creation is clearly pro-cyclical and job destruction is counter-cyclical, the volatility of the two flows over the business cycle may differ. Estimates for the US, Canada and the UK show that the increase in job destruction during economic downturns tends to be stronger than the increase in job creation during upturns, resulting in counter-cyclical movements in job reallocation. The cyclical behaviour of job reallocation is less clear for countries in Europe, where job reallocation tends to be a-cyclical or slightly procyclical and job flows are, on average, smaller; 3. Job reallocation is inversely correlated with capital intensity, more jobs being created and destroyed in services than in manufacturing; 4. The intensity of job reallocation depends on some firm-specific characteristics, in particular job creation tends to be negatively associated with firms age and size; 5. Job reallocation is a persistent phenomenon. This implies that the observed job flows can not be accounted for by temporary layoff and recall policies; 6. There are remarkable similarities in job reallocation rates across countries, even if they differ substantially in the degree of institutional flexibility. 2.2 Job flows and labour market institutions: theory and empirical evidence Cross-country comparisons of job flows provide the basis for a more formal investigation on the link between job turnover and labour market institutions and policies. The focus on the institutional determinants of gross job flows allows testing sharp theoretical predictions, which have ambiguous effects on net changes. A typical example is employment protection legislation. Barriers to the layoff of workers are expected to hinder both job creation and destruction, having ambiguous effects on the average level of labour demand (Bertola, 1990). Garibaldi (1998) focuses on the effect of dismissal costs on the cyclical behaviour of job creation and destruction. Introducing firing restrictions in a quite standard matching framework with endogenous job destruction, he argues that when costs associated with dismissals are negligible, job destruction is instantaneous while job creation takes time. As a consequence job destruction varies more than job creation and job reallocation should move counter-cyclically. This prediction is supported by the counter-cyclical pattern of job reallocation observed in US manufacturing (Davis and

4 Haltiwanger, 1992). However, when firing is costly and time consuming the asymmetry in the job flows cyclical behaviour disappears or might even be reversed for stringent enough dismissal restrictions. Thus, this would provide a rationale for the a-cyclical pattern in job reallocation found in Austria (Stiglbauer et al., 2002) and Germany (Boeri and Cramer, 1992) and pro-cyclical pattern found in France (Lagarde et al., 1994) and Sweden (OECD, 1994). Mortensen and Pissarides (1999b) and Pissarides (2000) incorporate the effects of both passive and active labour market policies in a fairly general search-equilibrium framework. Both unemployment benefits and employment taxes decrease job creation and increase job destruction through an increase in labour costs. In equilibrium, the economy settles at a higher unemployment rate while the effect on the overall job reallocation remains ambiguous. Job subsidies reduce the cost of matching inducing higher job creation. But job destruction rate increases as well because of the increase in market tightness, which improves the worker s options in the labour market. In contrast, Leonard and Audenrode (1993) argue that subsidies to declining firms must be supported by taxes on growing firms, which overall will reduce job creation and destruction and therefore job reallocation. The role of wage setting institutions on employment dynamics has been emphasised in a number of studies. It has been argued that union may influence worker exit behaviour through keeping wages above the market clearing level and through other non wage aspects (Faber, 1986; Freeman, 1980). In both cases the presence of unions contributes to improve the employee-employer relationship, making job separation more costly and consequently reducing job turnover. Regarding the level of wage bargaining, one could argue that more co-ordinated wage negotiations combined with wage drift policies might impose an additional restriction to plants when negotiating wages, reducing job creation and therefore gross job flows. On the other hand, Bertola and Rogerson (1997) show how wage compression induced by either a centralised bargaining system or by the presence of wage floors, may be conducive to higher job turnover through an increase in job creation by the more productive firms and job destruction by the less productive ones. From an empirical point of view, a preliminary attempt to relate facts with theory is due to Garibaldi, Koenings and Pissarides (1997). They pool summary job turnover measures from previous studies and present cross-country bivariate relationships with several measures of labour market institutions and policies for selected OECD countries. As suggested by the theory, the cross-country comparison reveals a negative correlation between job reallocation and the strictness of employment protection and unemployment benefits (but the correlation changes sign when unemployment benefits are measured in level rather than in duration). The correlation between active labour market policies and job turnover is slightly positive. OECD (1999) also finds a negative association between the strictness of protection legislation and job turnover rates. Regarding wage setting institutions, Lucifora (1998) for Italy and Blanchflower and Burgess (1996) for the UK find a lower rate of job turnover in unionised sectors. On the other hand, Heyman (2001) finds a positive association between job reallocation and the degree of wage compression on a panel of Swedish manufacturing establishments, supporting Bertola and Rogerson s (1997) hypothesis. The only study to our knowledge that presents multivariate analysis on the effect of product and labour institutions on cross-country labour market dynamics is Salvanes (1997). Pooling cross-sectional data from previous studies at the sectoral level for seven OECD countries characterised by different degrees of regulation, he tests the role of employment protection, wage bargaining centralisation and industry subsidies on job flows. The findings on employment protection are in line with previous studies, with stricter dismissal costs having a negative impact on job creation and destruction rates. Interestingly, the degree of centralisation also has a negative impact on labour market dynamics by reducing job creation. With regards to industrial subsidies, he finds a positive impact on job reallocation that contrasts with the negative effect found by Leonard and Van Audennrode (1993) when comparing the US and Belgium labour markets. However, these results should be read with caution, given the limited number of countries considered and the heterogeneity of data sources and definitions. Thus, despite the growing number of studies on this area, there is still little consensus over the observed facts and no clear pattern emerges looking at the cross-country job flow developments. The difficulties in international comparisons partly reflect the lack of homogeneous data. Available data are usually drawn from national sources, which differ in the way information is collected, definitions, sectoral and time coverage, etc. These may largely have affected the empirical results presented so far.

5 3 Data and measurement issues 3.1 Data source Annual firm-level observations over the period are available from Amadeus produced by Bureau van Dijk (BvD). Amadeus contains comparable firm-level data for European countries and covers all the sectors with the exception of the financial sector, for a total of more than 3 million different firms. 2 Apart from employment data, the dataset includes a wide range of financial information (e.g. profit and loss account, balance sheet, etc.) and descriptive information (industry and activity codes, incorporation year, etc.) The database has several advantages, which make it especially well suited for international comparisons. 3 First, the data collection method is reasonably homogeneous across countries. Amadeus homogenises balance sheet data from the different countries in order to facilitate comparisons. This overcomes the problem of previous studies where available country data differed on the sources (administrative vs. survey) and unit of study (firm vs. establishments). Second, information is provided on narrowly defined sectors (2-digit NACE classification) and data on both manufacturing and non-manufacturing sectors are reasonably representative. The availability of data on service industries for all countries is an important advantage with respect to previous studies, where cross-country comparisons relied on information obtained from the whole economy in some countries and the manufacturing sector in others. The restriction of the analysis on job reallocation to a single sector can lead to misleading generalisations when the sector is not representative of other industries. Moreover, the fact that the manufacturing sector has experienced a stable contraction in employment in the last decade may have affected the dynamics of job flows. However, there are some limitations in our data. First, in order to calculate changes in employment at firm level we have to separate firms according to whether they are continuing firms, start-ups or shutdowns. Spurious start-ups and shutdowns may be accounted for, if we are not able to distinguish, for example, newly created firms from firms that simply enter the pool at a given period t but are already operating in the period before. The same problem arises if we cannot identify firms closures from firms that exit the pool for other reasons. Unfortunately, Amadeus does not provide any information in relation to the entry and exit of firms in and out of the sample. The strategy followed is to avoid the risk of false flows restricting our analysis to continuing firms, e.g. firms that are in the sample for at least two consecutive periods. This will lead, however, to a downward bias in the estimates of job flows, given that, according to previous studies, births and deaths of firms account for at least one quarter of the estimated job flows. Moreover the exclusion of job flows from new entry firms can penalise some sectors more than others given that the contribution of new entrants to job creation is higher in the innovative rather than in the traditional sectors. However, this should not pose serious problems in our cross-country comparisons as long as the bias incurred is similarly distributed across countries. Second, the data are available at the firm rather than the establishment level. There are several reasons why establishment data may be preferred to firm level data. Measuring job flows at firm level may understate the actual magnitude of total gross flows among plants. 4 The use of firm-level data may lead to longitudinal linkage problems (and to spurious flows) if ownership and organisational changes (i.e. mergers, acquisitions, etc) are not accounted for. 5 This may be less of a problem with plant-level data, plant being defined in terms of physical location and 2 To be included in Amadeus a company should satisfy the following size criteria: operating revenues equal to at least 1.5 million Euro, total assets equal to at least 3 million euro, number of employees equal to at least 15 for the UK, Germany, France and Italy; for all the other countries operating revenues equal at least 1 million Euro, total assets equal to at least 2 million Euro, number of employees equal to at least 10 3 Faggio and Konings (1999) study the dynamics of job creation and job destruction exploiting Amadeus. However, their focus is on transition economies. 4 Job creation and job destruction resulting from movement between establishments within the same firm offset each other at the firm level. As a result, higher job reallocation rates are expected at the establishment level. Schuh and Triest (2000) estimate for the United States that job flows between firms represent less than 60 percent of the total job flows between establishments owned by these firms. 5 See Davis et al. (1996) for a detailed discussion on problems arising from the measurement of employment changes at the establishment/firm level.

6 production. Moreover, estimates of job creation and job destruction flows based on year-to-year employment changes can lead to a miscalculation of the actual flows since short term jobs (i.e. seasonal jobs) are likely not to be accounted for. A final limitation is related to the inclusion criteria in Amadeus, which introduces a bias against very small firms. These aspects should be kept in mind when comparing our results with previous studies. 3.2 Measuring job flows The conventions of Davis and Haltiwanger (1990, 1992, and 1996) are followed in defining job flows statistics. Denote the level of employment at firm level in period t with n ft and let n ft be the change in employment between period t and t-1. Let S + be the set of firms in sector S with n ft>0 and S - be the set of firms in sector S with n ft<0. We calculate job creation by summing employment changes in S +. Correspondingly, job destruction is calculated by summing all the (absolute) changes in S -. Rates of job creation and job destruction are obtained by dividing by the size of sector. Firm size at time t is calculated as the average employment between period t and t-1, i.e. x ft=0.5(n ft + n ft-1). Accordingly, the sector size is defined as X. st x ft f S Job flow rates can equivalently be expressed as the size-weighted average over firms growth rates as follows JC JD st st where g f St f St ft g g ft ft x X n x ft ft ft st x X ft st Job Creation Rate Job Destruction Rate is the growth rate of employment in sector f and period t. 6 The sum of the job creation rate and job destruction rate is the job reallocation rate (JR). It gives the total number of employment positions reallocated in the economy. The difference between job creation and job destruction is the net employment growth (NET). 3.3 Sample description The analysis focuses on 12 European countries. 7 Table1 shows the final sample composition and the sample period for each country, after filtering the observations from possible outliers and missing values. 8 The period of observation 6 The growth measure defined above is monotonically correlated with the conventional measure defined as the change in employment divided by the lagged employment, and the two measures are approximately the same for small growth rates. Moreover, unlike the conventional measure, which ranges from 1 and, this measure of growth rate is symmetric around zero, being bounded in the interval [-2,2], allowing employment expansions and contractions to be treated symmetrically. 7 These include the EU countries with the exception of Italy, Luxembourg and Greece. Greece and Luxembourg are excluded from the analysis due to lack of institutional data. Initial exploration of Italian data showed important inconsistencies between annual employment growth rates computed from Amadeus and from the labour force survey. These inconsistencies did not seem to be related to specific outliers and therefore Italy was excluded from the sample. A potential explanation is the over-representation of small firms in Italy. Since Amadeus excludes small firms from the sample, this might give a more distorted representation of aggregate flows in Italy than in the other countries considered. 8 We leave a more detailed discussion of the cleaning of the data to the working paper version of the paper. Summarising, the following exclusion criteria have been imposed: 1.Firms for which only consolidated accounts are available; 2.Observations for which employment growth rate is missing value; 3.Observations for which the growth rates of compensation per employee is less then 50 percent or more than 50 percent within a given year; 4.For Austria and Germany there is very little information on wages

7 varies across countries but information is available in most cases at least during The number of average valid observations per year ranges from almost firms in Germany to some 500 firms in Ireland. This implies an annual average employment coverage of 24 per cent when compared to figures in the Labour Force Survey. Table 2 shows the distribution of firms and employment by sector and country and compares the distribution of sectoral employment in our sample with the distribution calculated using information from the EU Labour Force Survey (LFS). The sample appears to be representative of both manufacturing and non-manufacturing sectors. Moreover the sectoral coverage is rather homogeneous across countries and stable over time. Figure 1 compares the evolution of employment growth from our sample with the LFS. On average, the employment figures in our sample follow quite closely those obtained from the survey (the average correlation is 0.8). However, in Portugal and Sweden the sample appears less representative of the aggregate evolution of employment. 4 Facts about job reallocation, creation and destruction 4.1 An Overview Table 3 reports the aggregate rates of job creation (JC), job destruction (JD) and job reallocation (JR) in each country, averaged within the sample period. First note the large flows, both regarding job creation and destruction, observable in all countries. Although all of the countries registered a net increase of employment within the period of study, the coexistence of significant job creation and destruction flows is a broadly based finding. Job creation rates moved between 4.6 percent in Germany and 9.2 percent in Spain, and job destruction rates from 2.8 percent in Ireland and 4.2 percent in the UK. These developments led to job reallocation rates of around 10 percent on average in the EU, Austria and Germany being the two countries with the lowest job reallocation (7.8%) and Spain that with the highest (12.4%). This means that, on average, one tenth of jobs were either created or destroyed per year. The rest of Table 3 presents summary statistics of flow rates by sector and firm s size, age and capital intensity pooling the information across countries and years. According to the sector, 9 service industries exhibit, on average, larger job flows. Business services is the sector with larger job flows in Europe during the period of study, a pattern mainly driven by the strong employment creation in this sector, while manufacturing presents the lowest JR rate. As regards size classes, the concept used in the analysis refers to the average size of the firm in two consecutive periods. The average size is used instead of the current size as it is expected to give a better indication of the intended scale of operations. We divide the sample in eight categories: 1-19 employees; 20-49; 50-99; ; ; ; 1,000-2,499; and 2,500 and over. The process of job reallocation is clearly stronger among smaller firms. In fact, there is an inverse relationship between the size of the firm and the intensity of job reallocation. Moreover, this inverted relationship is mainly due to the pattern of job creation, which shows a higher variation among firm size than the pattern of job destruction. Concerning the age of the firm, four groups are considered: 1 year old; 2-5 years; 6-10 years; and more than 10 years. Job flows are significant in all age groups but higher in the youngest firms than in the oldest. This fact is again related with a higher variation in job creation than in destruction. The indicator of capital intensity is obtained as the ratio of capital and value added. We distinguish four categories: below 20%; 20-30%; 30-40%; and more than 40%. According to the estimates in Table 3, there seems to be an U shaped relationship between capital intensity and JR, with firms with either low or high capital intensity exhibiting larger turnover rates. in Amadeus. Thus, in order to clean possible outliers we trimmed out the 1 st and 99 th percentile of employment growth for firms with more than 250 employees. 9 Sectors are defined according to the 2-digit NACE classification (NACE code, rev 1).

8 4.2 The impact of firm characteristics and the cycle on job flows This section first studies the joint impact of the different firm characteristics considered in the descriptive analysis on the dynamics of job flows. We calculate JC, JD and JR rates for narrow sectors defined as the crossing of 4 age groups (those previously defined), 7 sectors (here the industrial sector includes Mining and quarrying, Manufacturing and Energy and water supply), 4 size groups (1-49 employees; ; ; 1,000 and over), 12 countries, 10 years (between 1992 and 2001) and 4 capital intensity groups (those previously defined). Then, we regress these estimates on dummy variables defined for each of these groups, with the aim to uncover the main determinants of labour market flows. We will consider two different specifications, depending on whether we include or not capital intensity groups. The reason is that Amadeus has very limited information on value added for firms in Austria, Germany and the Netherlands. Thus, considering capital intensity classes might affect significantly the estimates of these countries. When capital intensity crossings are excluded, the potential number of cells is 13,440, ascending to 53,760 if capital is included in the analysis. In the first case, we had about 7,000 valid observations, and almost 18,000 when capital intensity is considered. 10 OLS regressions on the sectoral dummies are weighted taking into account the employment size of each cell after removing the outliers. 11 Reported standard errors are robust to country clustering. Table 4 summarises the results of the OLS regressions for JR, JC and JD on the class dummies. Columns A to C do not include capital intensity groups, while these are included in columns D to F. First note that according to the goodness of fit in the regressions, the proposed models do a much better job in explaining the patterns of JR and JC than in explaining the sources of JD, suggesting a more important role to idiosyncratic factors in the determination of the latter. The results tend to be in line with the descriptive analysis discussed in the previous section. Thus, the age of the firm has a negative relationship both with job reallocation and creation, especially when firms are more than 5 years old, while there is no clear pattern for job destruction. Both job reallocation and creation are 3.5 percentage points lower in firms above 10 years old than in those only 1 year old. The sectoral dummies confirm a higher job reallocation and creation in construction and service sub-sectors compared with industry, while the latter has a higher job destruction. For instance, job reallocation and creation rates in Business services are more than 4 percentage points higher than in industry, while the difference in job destruction is negligible. Similarly, the negative relationship between the size of the firm and job reallocation is confirmed by the multivariate analysis. Indeed, both job creation and destruction rates are lower the bigger the firm is, although the differences are higher in job creation than in job destruction. A firm that has more than 1,000 employees has on average a job reallocation rate 4.3 percentage points lower than a firm with less than 50 workers, which is explained by 3.8 percentage points less in job creation and 0.7 percentage point less in job destruction. These differences fall to 1, 0.9 and 0 percentage points respectively if the firm has between 50 and 250 employees. Country differences also appear to be significant in the estimates. As regards job reallocation, only Spain shows a similar rate to that of the UK, while all the rest show significantly lower rates. The highest differences compared with the UK are observed in Austria and Germany, which have respectively 4.7 and 3.8 percentage points less JR than the UK. Apart from Spain, only Ireland shows a relatively higher job creation rate than the UK. Moreover, all countries show lower job destruction rates than the UK. These facts could be related with a relatively less regulated labour market in the UK, which would imply higher job reallocation and, especially, more variability in job destruction. 12 When ranges of capital intensity are taken into account, all previous results remain broadly unchanged (see Columns D to F in Table 4). In addition, it appears that capital intensity does not play a significant role in explaining differences in job reallocation rates, but it makes a small difference both for job creation and destruction. The intermediate levels 10 The main reason for missing observations is the variable sample period in each country. 11 We trimmed out outlier observations using the method of detection for the multivariate framework developed by Hadi (1992). In order to identify the outliers, we constructed categorical variables by age, size, industrial sector, country and capital intensity (if applicable). This implied the exclusion of 56 (147) cells in the case of JR, 65 (153) in the case of JC and 101 (237) in the case of JD in the sample without (with) capital intensity. The results presented in the paper refer to the regressions without outliers. Results including outliers, available upon request, do not differ importantly with respect to those presented in the text. 12 Interestingly, in previous international comparisons (e.g. OECD, 1994 and Contini et al., 1995) the UK appeared to have relatively low job flow patterns when compared to many Continental European countries. This apparently puzzling result, reversed in our study, might be due to the lack of homogeneous data.

9 of capital intensity show relatively higher job creation (0.3 percentage points) and relatively lower job destruction (around 0.5 percentage points). The next exercise measures the effect of the cycle on job flow rates. Three variables have been selected as alternative indicators of the cycle: a measure of the output gap, the rate of growth of GDP and the net change of employment in each cell. Table 5 summarises the results of regressions following the specifications presented in Table 4 including the cyclical indicators, both with and without time dummies. These seem to be quite consistent independently of the indicator of the cycle included. Job reallocation and job creation are found to be pro-cyclical, although the latter tends to react more to the cycle than the former, while job destruction appears to be countercyclical. The estimates are statistically significant in most cases, even if year dummies are included in the regression. As regards job creation and job destruction, these are the expected results. Job creation tends to increase in the expansions and decrease in recessions, while the contrary occurs to job destruction. However, previous studies have found a negative correlation between job reallocation and the cycle in some countries, like the US and the UK, while it has been found pro-cyclical in France, Italy and Sweden, or non-correlated with the cycle in countries like Germany, Austria and Spain. Garibaldi (1998) suggests a rationale for this divergence: in countries with low firing costs job destruction tends to show more variability than job creation and the correlation between job reallocation and the cycle is negative, while the contrary occurs in countries with high firing costs. Thus, we interpret the clear pro-cyclical pattern of JR in Europe as a sign of the importance of firing costs. Finally, in regressions not reported in the text we have interacted the cycle indicators with the UK country dummy to assess whether a different pattern emerges in this country with regards to JR. In most cases, the interaction effect was negative, as suggested by the theory, but not always statistically significant. However, our data are not very well suited to uncover individual country cyclical patterns given the limited number of years available. 5 Job Flows and Institutions The aim of our next set of regressions is to uncover the determinants of country idiosyncratic factors in the patterns of job reallocation. In the previous tables we have shown that compositional effects are very relevant for explaining the determination of labour market flows. Thus, failing to control for differences across countries in the size, age and sectoral distribution of firms might blur cross-country comparisons. In order to control for these differences, we repeat each of the specification A to C in Table 4 replacing the country dummies for a set of indicators of labour market institutions and policies. 13 According to our previous discussion, we concentrate on several institutional and regulatory aspects of the labour markets: Tax and benefits systems: This group includes an indicator of the duration and generosity of unemployment insurance systems and the tax wedge between the real (monetary) labour cost faced by the firms and the consumption wage received by the employees. The latter is normalized by GDP, while the duration of unemployment benefits ranges from 0 (if benefit provision stops after 1 year) to 1 (for a constant unemployment benefit after 5 years). Regarding wage-setting institutions, we include an indicator of the extent of coordination in the wage bargaining process. This is an index ranging from 1 to 3 increasing in the degree of wage setting coordination. 14 Restrictions to hiring and firing: We consider two alternative measures of the relative stringency of hiring 13 Although most of the institutional indicators are time-variant, the indicators of EPL and the degree of wage setting co-ordination are not, not allowing to control for fixed effects in the panel regressions. 14 Wage-setting co-ordination, unemployment benefits duration, and the tax wedge are taken from Nickell and Nunziata Labour Market Institutions database. The information is annual till When necessary, we extrapolated the variables for the period

10 and firing rules in the late 1990s collected by Nicoletti et al. (1999). These include an index of the overall stringency of employment protection legislation (EPL), and an index of the relative stringency of these contracts affecting permanent workers. Each index, ranging from 1 to 6, increases with the relative stringency of these contracts. Additionally, we include the share of workers holding temporary contracts. 15 Sectoral employment subsidies: We include an indicator of the share of sectoral and ad hoc state aid as a percentage of GDP. 16 Table 6 presents the results of OLS regressions of job reallocation, creation and destruction rates on the set of institutional variables and sectoral and firm characteristics. As expected, the strictness of employment protection legislation has a negative and significant impact on the extent of job reallocation. This result is robust no matter which indicator of EPL is considered. Interestingly, we find a negative and statistically significant (at the 10 per cent level in the case of EPL for all contracts) effect of firing restrictions in the creation of employment, suggesting that firms anticipate the cost burden of hiring additional workers in countries with tight dismissal restrictions. Similarly, there is some evidence of firing costs inhibiting job destruction, but this effect becomes insignificant in the case of EPL for regular contracts. The role of firing restrictions is partially counterbalanced by the use of temporary contracts, which clearly increase the job flows in the economy by boosting employment creation. Thus, according to the estimates in Column A an increase in 10 percentage points in the share of workers holding a temporary contract would increase JR in 1 percentage point within our sample. This result suggests that some countries might be using temporary and atypical forms of employment to overcome the cost burden imposed by firing restrictions. Thus, this can provide a rationale to the high rate of JR observed in Spain in spite of tight EPL in this country. Within the period, the average share of temporary employment in Spain was 33 per cent, more than 20 percentage points higher than the average for the other countries studied in the same period. Both the tax wedge and the duration of unemployment benefits have similar effects, reducing job reallocation by dampening job creation. The effects on job destruction are not statistically significant different from zero. This suggests that the costs imposed by these regulations are anticipated by firms, which respond by reducing the creation of new jobs. We find lower JR rates in countries with more co-ordinated wage setting institutions. This is in line with the findings of Salvanes (1997) for 7 OECD countries, suggesting that centralised systems impose an additional constraint for firms when negotiating wages. Again, the effects on JR are due to a reduction in JC but not in JD. Finally, we don t find statistically significant effects of sectoral employment subsidies on labour market flows. The two final tables provide some evidence of the robustness of the results presented. Table 7 presents sensitivity analysis with respect to the number of countries included in the regressions, by repeating each specification dropping one country at a time. Accordingly, the results presented above are not insensitive to the number of countries included in the analysis, something not very surprising given the limited number of countries under study. However, regarding JR the role of EPL, temporary contracts, the tax wedge and wage setting co-ordination are relatively robust, always with the expected sign and significant at the 90 per cent level in all but one of the regressions (in all cases for temporary contracts). Unemployment benefits are not statistically significant in 2 out the 12 regressions. Something similar can be observed in Columns D to F, which confirm the negative effects of EPL for regular contracts, the tax wedge and wage-setting co-ordination on JC and the positive effects of temporary employment. As expected by the lower fit in the regressions, the results concerning JD are less robust, with just the overall index of EPL keeping the expected sign statistical significance in all but one of the regressions. Table 8 looks at the effects of institutions when introduced one at a time. The direction of the effects is largely consistent with those previously presented. Moreover, both EPL measures, the tax wedge and the degree of wage setting co-ordination remain highly significant (the last two measures even become significant with respect to JD). However, the duration of unemployment benefits and the share of temporary employment contracts, which were highly significant in the multivariate framework, become insignificant when introduced on their own. Finally, sectoral employment subsidies on their own have a very significant negative 15 Source: Labour Force Survey. 16 Source: NewCronos Database.

11 effect on JR by reducing both JC and JD, a result consistent with the theoretical view that providing subsidies to ailing establishments should lead to less job reallocation. 6 Conclusions This paper has presented an analysis of job flows for a panel of 12 European countries in the 1990s using a unique homogeneous firm level data set that covers the whole spectrum of productive sectors and permits cross-country comparisons given homogeneity in the definitions and sectoral coverage We have estimated cell regressions of gross job flow rates on different firm characteristics such as size, age, industrial sector of operation and capital intensity. We found that, on average, both the size and age of the firm have a negative impact on job flows, especially on job creation and reallocation, and firms located in service industries typically exhibit stronger patterns of job flows than firms in manufacturing sectors. Moreover, even after controlling for these compositional effects, we find significant cross-country differences in labour market dynamics. Thus, we have focused on institutional aspects of labour markets as potential explanatory sources of cross-country variation in job flow rates. The multivariate analysis suggests that, after controlling for sectoral and firms characteristics, the strictness of employment protection legislation has a negative and significant effect on job reallocation. Moreover, the impact of firing restrictions is partially mitigated by the presence of temporary contracts, which have a positive effect on job creation and reallocation. With respect to the role of wage setting institutions, wage bargaining co-ordination seems to reduce job creation resulting in lower job reallocation in highly co-ordinated countries. A similar effect is found for both the tax wedge and unemployment benefits, both reducing job creation and reallocation. Finally, introducing different cyclical indicators in the analysis confirms the positive and negative correlation, respectively, of job creation and destruction with the cycle. Job reallocation has been found to move together with the cycle in European countries, a finding that provides further support to the importance of tight employment protection in shaping employment dynamics in these countries, as suggested by Garibaldi (1998). This paper has focused on the dynamics of employment adjustment, leaving aside the response of wages in the joint determination of labour demand. Given the availability of wage data in Amadeus, this constitutes the main line for further research.

12 APPENDIX. Tables and Figures Table 1 Final Sample Composition Sample period Total number of observations per year Coverage (%) Austria Belgium Denmark Finland France Germany Ireland Netherlands Norway Portugal Spain Sweden UK Sources: Amadeus (2002) and Labour Force Survey.

13 Table 2 Distribution of firms and employment by sector (NACE code, rev. 1) Sectors Austria %empl (LFS) %empl %firms Belgium %empl (LFS) %empl %firms Denmark %empl (LFS) %empl %firms Finland %empl (LFS) %empl %firms France %empl (LFS) %empl %firms Germany %empl (LFS) %empl %firms Ireland %empl (LFS) %empl %firms Netherlands %empl (LFS) %empl %firms Portugal %empl (LFS) %empl %firms Spain %empl (LFS) %empl %firms Sweden %empl (LFS) %empl %firms United Kingdom %empl (LFS) %empl %firms Note: Agriculture, forestry and fishing; Mining and quarrying; Manufacturing; Energy and water supply; 45 Construction; Trade, Restaurants and Hotels; Transportation and communication; Business services; Community, social and personal services.

14 Table 3 Average job flow rates JC JD JR By country Austria Belgium Denmark Finland France Germany Ireland Netherlands Portugal Spain Sweden UK By NACE Agriculture Mining Manufacturing Energy Construction Trade, restaurants and hotels Transport and communication Business services Community, social and personal services By size 1-19 employees employees employees employees employees employees employees and more employees By age 1 year old years old years old More than By capital intensity 20% or less % % More than 40% Note: the figures in the table are average values over the sample period.