Explaining Long-term
Unemployment
An investigation into the effects of labour
market institutions on unemployment
Mark Jansema
Mark Jansema
Mark Jansema
Mark Jansema
1383590 1383590 1383590 1383590Master’s Thesis in Economics Master’s Thesis in Economics Master’s Thesis in Economics Master’s Thesis in Economics
Keywords: unemployment, long-term unemployment, institutions, error correction model
Supervisor: dr. J.P. Elhorst
page 1
page 2
SUMMARY
Over the last four decades unemployment rates in Europe have reached a higher plateau, while the United States succeeded in withstanding adverse shocks and adapting to a changing climate through its labour market institutions. The difference in labour markets also lies in the share of long-term unemployment. Most European countries, the Netherlands in particular, have higher shares of long-term unemployment than the United States. This paper aims to find which institutional characteristics of the labour market are responsible for the development of long-term unemployment.
Most economic scientific research focuses on finding the determinants of the aggregate unemployment rate, while mostly neglecting the specific case of long-term unemployment and the particular issues associated with it. Using an error correction modelling approach, we investigate the effects of institutional characteristics of the labour market on long-term unemployment, while also considering the aggregate unemployment rate. Here, the focus lies on a sample of twenty OECD countries, which is divided in a group of countries with high shares of term unemployment and a group of countries with low shares of long-term unemployment.
page 3
CONTENTS
1. Introduction ... 4
2. Explaining long-term unemployment: What do we know? ... 9
3. Theoretical background ... 16
4. Empirical approach ... 20
5. Variables and data ... 25
6. Overview of findings and diagnostic tests ... 33
7. Concluding remarks ... 42
Acknowledgements ... 46
References ... 47
Annex I – Graphs ... 51
Annex II – Tables ... 55
page 4
INTRODUCTION
The thirtieth president of the United States, John C. Coolidge, once approached the concept of unemployment as follows:
‘More people out of work leads to higher unemployment.’
He was, of course, accurate in making this statement, but the problems associated with unemployment comprise a much larger area than just being “out of work.” This paper explores the institutional characteristics of the labour market that cause people to be unemployed, and is particularly concerned with long-term unemployment.
Observations
Unemployment rates have fluctuated dramatically in OECD countries in the last few decades with rates as high as 19.9 percent in Poland in 2002 and rates as low as 0.1 percent in New Zealand in 1974. Fluctuations include large differences between decades in one country and large differences between countries for long periods. The average rise of the unemployment rate in these countries between 1960 and 2006 is 4.2 percentage points with a median rise of 4.0 percentage points. OECD countries had an average unemployment
rate of 6.6 in 2006.
Up until about 1990, the unemployment rate of the Netherlands exceeded the rates of institutionally similar countries like Germany, Norway, and Sweden. When considering all
OECD countries, the Dutch unemployment rate is among the lowest in the sample. This is especially clear at the end of the 1990s and the beginning of the 21st century. In this period, the Dutch unemployment rate fell below the rates of Norway and Sweden. Over the entire period, the Netherlands’ unemployment rate seems to follow the overall trend. The difference in unemployment rates between European countries and the United States is striking. Although the United States have experienced similar shocks as Europe, the US unemployment rate has been relatively stable, whereas the European rates have reached a higher plateau. Figure 1 shows the unemployment rates in the EU13 countries and the United States.1 Figure 1 also shows that about halfway through the 1980s Europe’s
unemployment rate became permanently higher than that of the United States, which had a higher rate before 1983.
page 5
Figure 1: Unemployment rates in the EU13 countries and the United States.
The development of the ratio of long-term unemployment to total unemployment is even more interesting to note. Long-term unemployment refers to people who have been unable to find a job for over 12 months. For the Netherlands, this ratio stayed just below 50 percent until the beginning of the 21st century, when it plunged below 30 percent. A few years after the plunge the ratio rose again and according to the latest figures, the incidence of long-term unemployment is almost back at its 50 percent level. When compared to other
OECD countries, the Netherlands has one of the highest shares of long-term unemployment
in total unemployment. From 1960 to 2006, only Belgium, Ireland, Italy, and Spain have had higher shares of 55 percent, on average. In contrast, Canada, Korea, Norway, and the United States had the lowest shares of long-term unemployment in the sample of 5 to 15 percent to total unemployment. Figure A1 in the appendix displays the ratio of long-term unemployment to total unemployment of the countries in the sample used in the empirical part of this paper.
0,0 2,0 4,0 6,0 8,0 10,0 12,0 1 9 6 0 1 9 6 2 1 9 6 4 1 9 6 6 1 9 6 8 1 9 7 0 1 9 7 2 1 9 7 4 1 9 7 6 1 9 7 8 1 9 8 0 1 9 8 2 1 9 8 4 1 9 8 6 1 9 8 8 1 9 9 0 1 9 9 2 1 9 9 4 1 9 9 6 1 9 9 8 2 0 0 0 2 0 0 2 2 0 0 4 2 0 0 6
page 6
Figure 2: Decomposition of the unemployment rate by duration for the Netherlands.
Figure 2 shows a decomposition of the Netherlands’ unemployment rate by duration. Although the Dutch share of long-term unemployment rate has decreased over the years, it is still almost twice as large as any of the other duration categories is. The unemployment rate can be divided into five different duration categories: people who have been unemployed for less than one month, between one and three months, between three and six months, between six months and a year and more than a year.
Figure 3: Long-term unemployment rates in the EU13 countries, the Netherlands and the United States.
Figure 3 displays the evolution of the long-term unemployment rate from 1983 to 2006 for the Netherlands, Europe and the United States. It is immediately clear from Figure 3 that although the long-term unemployment rate of the Netherlands has almost decreased
0 2 4 6 8 10 12 > 1 year > 6 months < 1 year > 3 months < 6 months > 1 month < 3 months < 1 month 0 1 2 3 4 5 6
page 7
threefold over the period, it is still higher than the rate of the United States, but lower than the European average. While the share of long-term unemployed is higher in the Netherlands than in most of the rest of OECD Europe, its long-term unemployment rate is
lower in this figure, because the Netherlands’ aggregate unemployment rate is lower. In addition, even though the aggregate unemployment rate of the United States is about one percentage point higher than that of the Netherlands, its long-term unemployment rate is lower, because it has a much lower share of long-term unemployed than the Netherlands.
Consequences
Being long-term unemployed can have severe effects on a person’s well-being, as well as influence the economy as a whole. Turning to personal distress first, long-term unemployment is related to a heightened level of depressive symptoms and anxiety, even after controlling for individuals’ childhood and adolescent characteristics (Kokko, Pulkkinen, and Puustinen, 2000). Furthermore, a high long-term unemployment rate indicates that unemployment is disproportionately concentrated on a few individuals, leading to more income inequality, given that employment is the main source of income among households.
A large share of long-term unemployed people also has consequences for a country’s economy as a whole. Depending on how the national unemployment benefit system is organised, a large share of long-term unemployed people can put a strain on the national budget, on employers and/or on employees.
The continued absence of this group of potential workers from the labour market, leads to them being demoralized and stigmatized in the eyes of employers (Jackman and Layard, 1991). Demoralization may lower the search intensity of the unemployed and stigmatization could be the result of the perception that people who are absent from the labour market slowly lose some of their skills and work ethic (Phelps, 1972). Employers could further use the history of unemployment as a signal of employee productivity. Budd, Levine and Smith (1988) use an analogy with a flower shop to illustrate stigmatization: only new flowers and those that fade least rapidly are sold every day. These two factors cause the long-term unemployed to have a lower probability of finding work.
page 8
than they could, or even should be. As a result, high long-term unemployment is one of the causes of high unemployment.
Understanding the factors that determine why one person is still waiting for a job two years after he lost his previous job, while another person is able to find a job within a month, is therefore key to understanding how to reduce unemployment in general and long-term unemployment in particular. After all: data for OECD countries in 2006 show that the
average duration of unemployment is 15.8 months.
Objectives and organisation
This paper investigates for two samples of OECD countries which factors influence the development of long-term unemployment, and which influence they have. The two samples are separated on the basis of their share of long-term unemployment being higher or lower than the sample average. This paper also tries to find answers to the following sub-questions: Why is there such a large difference between the share of long-term unemployment of low-share countries (like the United States and Scandinavian countries) on the one hand, and high-share countries (like the Netherlands and Germany) on the other hand? Is a government able to influence the share of long-term unemployment? If so, what can a government do to reduce this share?
page 9
EXPLAINING LONG-TERM UNEMPLOYMENT: WHAT
DO WE KNOW?
A possible answer to the question posed in this section’s title is ‘Not much.’ Although there is a growing literature on long-term unemployment, not many papers have tried to find the causal factors of long-term unemployment and most consider only hazard rates. There is, however, a wealth of papers that describe and explain aggregate unemployment from a theoretical point of view. Because we believe total unemployment and long-term unemployment could very well be influenced by the same factors, we consider papers on unemployment highly relevant to our research. This section provides a non-technical summary of these papers and discusses papers dealing with long-term unemployment when relevant.
Bean (1994) uses a Phillips curve that represents a fairly larger assortment of variables in the natural rate. His model includes a number of variables to explain unemployment. He considers the effect of contractionary demand policies through the price-surprise term, the relationship between the reservation wage and worker productivity, the reservation wage itself, changes in the tax and import price wedge, union power, changes in the wage mark-up due to changes in matching success between the demand and smark-upply of different types of labour, unemployment benefits, changes in the price mark-up mainly due to changes in the interest rates, and demographic developments. Regrettably, Bean finds that no single variable stands out as the cause of the rise in European unemployment. He does not believe that the terms-of-trade shocks of the seventies and the counter-inflationary policies, associated with high interest rates, of the eighties are relevant. One noteworthy result is Bean’s finding that real wage rigidity is relatively high in the European Community and the United States, because wage settlements seem less responsive to the level of unemployment in these countries. In addition, the persistence of European unemployment is rationalized by a high degree of propagation mechanisms of unemployment. This explains the difference between the European and American unemployment rates, despite being exposed to similar economic shocks.2
Siebert (1997) argues that institutional changes over the last 25 years can explain Europe’s poor labour market performance and the difference in unemployment figures between Europe and the United States can be explained by institutional differences. He then
2 See Figure 1 on page 3: Europe unemployment reached a higher plateau, while the United States’
page 10
embarks on a lengthy discussion about institutionary characteristics that have played a role in Europe and the United States. He discusses wage and employment flexibility (measured by the wage elasticity of labour demand and the unemployment elasticity of the wage rate), wage differentiation (where more centralization and coordination lead to less inequality and thus less differentiation), wage bargaining (a high degree of coordination and centralization leads wage formation away from a market process), tax structure, job protection legislation and the reservation wage. The author concludes with a policy advice:
‘Unemployment can easily become persistent, and to overcome rising unemployment in a self-enforcing trap, it takes a comprehensive push of institutional change’ (p. 53)
In a paper that looks at labour market differences between Europe and North America, Nickell (1997) argues that there are rigidities that serve a purpose and do not cause high unemployment. Nickell discusses eleven institutional factors as having an influence on unemployment. The first is employment protection legislation. The second is a labour standards index where countries are scored on five different points: working time, fixed-term contracts, employment protection, minimum wages, and employees’ representation rights. The third through fifth factors fall under the heading “Treatment of the Unemployed”: the benefit replacement ratio, benefit duration, and active labour market policies. Sixth through ninth are factors that summarize the structure of wage determination: union density, union coverage, and union and employer co-ordination. The last two factors are tax-related factors: the payroll and total tax rates. The data show large differences between Europe and North America, but between European countries as well. Nickell regresses the logarithms of total unemployment, long-term unemployment and short-term unemployment on these variables, supplemented by the change in inflation and a dummy for his second sample period, using Generalized Least Squares (GLS) with an estimated covariance matrix. His results are not remarkable; the effects of the explanatory variables are highly significant and have the expected signs. Employment protection has no significant effect on unemployment, but has a negative impact on employment.
page 11
unemployed to take jobs) and high levels of unionization and union coverage (if these are accompanied by high levels of employer coordination in wage bargaining).
Baker, Glyn, Howell, and Schmitt (2002) summarize the findings of other authors who have tried to explain the unemployment rate. In an assessment of these findings, they conclude that although:
‘[the] literature is widely viewed to provide strong evidence for the labour market rigidity view, a close reading of the leading papers suggests that the evidence is actually quite mixed (…)’ (p. 43)
The authors themselves regress the standardized unemployment rate of 20 OECD countries
on a number of variables for four five-year periods from 1980 to 1999. In a follow-up paper (2004), the authors extend the dataset to include the period 1960 to 1979. The factors they include are a measure of employment protection legislation, the replacement rate, a measure of benefit duration, union density and coordination of wage bargaining, the tax wedge, union coverage and the change in inflation. The results of these regressions are, in the authors’ words, lacking robustness and seem to depend on the measures of the institutions used and on the time period covered. Baker et al. add:
‘(…) there is little evidence here of the consistency of results which could convincingly underpin sweeping recommendations for labour-market reform’ (p.52)
When trying to mimic Nickell’s (1997, 1998) results with a more extensive database of institutional variables, Baker, Glyn, Howell and Schmitt find none of the variables to be significant. A second regression including interaction terms provides counterintuitive results concerning benefit generosity, while employment protection legislation, union density and the size of the tax wedge are found to be insignificant. A measure of bargaining coordination and its interaction with union density are highly significant and have the expected signs. A third setup, where the authors divide the sample period into two sub-periods, produces even less convincing results.
The objective of a paper by Nickell, Nunziata, Ochel, and Quintini (2002) is to perform an empirical analysis of unemployment patterns for twenty OECD countries from the 1960s
page 12
from its equilibrium level: money supply shocks, productivity shocks, labour demand shocks, real import price shocks, and the real interest rate. These variables act as indicators for macroeconomic shocks. The results indicate that, in general, changes in labour market institutions can explain the broad pattern of unemployment shifts in the sample countries reasonably well. Also:
‘[Labour market institution’s] impact on unemployment is broadly consistent with their impact on real wages’ (p.17)
Nickell, Nunziata, and Ochel (2005) perform another analysis of unemployment patterns in twenty OECD countries. In a first specification, a basic fixed effects empirical model
estimated by Feasible GLS (FGLS), they explain the aggregate unemployment rate using the lagged unemployment rate, employment protection legislation, measures of the benefit system, union characteristics, employer coordination, a measure for the mobility of labour, and several macroeconomic shocks, equal to those described above for the paper by Nickell, Nunziata, Ochel and Quintini (2002). The regression also includes several interaction terms, time and country dummies and country specific trends. Again, changing labour market institutions provide an acceptable explanation of the broad pattern of long run unemployment shifts for these countries. Real interest shocks and employment protection have no significant impact. In a second regression a number of institutional variables from the first specification are duplicated, but now interact with time dummies. On their own, these interactions give reasonably well explanations for the shifts in unemployment in the sample (benefit generosity and the tax wedge provide the most valuable contribution), but when used to augment the first model, they are not of much use.
Belot and Van Ours (2001) use fixed effects OLS for a dataset spanning 1960 to 1995 for 18 OECD countries to gauge the effect on the aggregate unemployment rate of the
page 13
In addition to the aggregate unemployment rate, Bertola, Blau, and Kahn (2002) consider the youth unemployment rate. In a sample containing 5-year observations from 1960 to 1995 for 20 OECD countries, the authors find all observable shocks (country-specific or
macro-economic) significant with the expected sign, with the exception of a variable that gives the youth share in total population. Institutional determinants (including unemployment benefit replacement rates, benefit duration, and active labour market policies) also explain unemployment trends reasonably well. However, a model with interactions between shocks and time-varying institutions explains a significantly larger share of diverging unemployment trends than a model with shocks only or shocks and institutions entered separately. The robustness of these models must be questioned, because most interactions are insignificant in regressions with time-invariant institutions.
Blanchard and Wolfers (2000) find that macroeconomic shocks account for most of the general evolution of unemployment over the period 1960 to 1995. Interactions between shocks and institutions account for most of the heterogeneity of unemployment over the same period. The interacted terms in their fixed effects OLS estimation are significant (with the exception of union coverage); hence, institutions contribute to shaping the ultimate impact of a particular shock on unemployment. The explanatory power of the interaction terms diminishes significantly when time-varying measures of employment protection and unemployment benefit replacement rate are used.
Using fixed effects OLS and random effects FGLS for a sample of 19 OECD countries and a
data set spanning 1985 to 1999, Boone and Van Ours (2004) estimate the effects of institutional variables (time- varying for fixed effects, time-invariant for random effects) and a macroeconomic proxy on the aggregate unemployment and non-employment rates. The product of unemployment benefits and active labour market policy is added as an interacting term. The results indicate a positive and significant relationship for unemployment benefits and union density. Active labour market programmes can be ordered by effectiveness: labour market training is most effective, while subsidised jobs are not effective at all. Public employment services are positioned somewhere in between. Nicoletti and Scarpetta (2004) find that unemployment benefit generosity, union density, an indicator for product market regulation, and strictness of employment protection legislation all have significantly negative coefficients in fixed effects estimations for 20
OECD countries for 1980 to 2002. They also find evidence of a hump-shaped corporatism
page 14
employment rate in the non-agricultural business sector. The negative coefficients found are therefore as expected.
Nunziata (2002) takes into account a range of factors influencing the aggregate unemployment rate. He considers the change in Total Factor Productivity (TFP) growth, the change in real import prices, the real interest rate, a labour demand shift, and the change in money supply growth as country-specific macroeconomic shocks. Institutional variables include the unemployment benefit replacement rate and duration, strictness of employment protection legislation, the tax wedge, union density, and coordination. Other variables considered are fixed term contracts regulation and the owner occupation rate. Interactions between these variables are also included. Of the macroeconomic shock variables, only the money supply shock is insignificant. Statistical significance of the real interest rate shock varies, while labour demand, TFP growth, and real import price shocks is high, as is the unemployment benefit replacement rate. Unemployment benefit duration, coordination and the change in union density are significant in most specifications. The tax wedge is significant in only some estimates. Fixed term contract regulations raise unemployment. Interactions between institutional variables are significant and have the expected sign. Nunziata’s sample includes 20 OECD countries for a period of 1960 to 1995. Machin and Manning (1998) review the literature on the causes and consequences of the high incidence of long-term unemployment on European labour markets. They use a hazard function to assess the effects of changes in outflow rates at different durations on the incidence of long-term unemployment. The authors find that a decrease of outflow rates at all durations of unemployment caused the incidence of long-term unemployment to rise. Therefore, it is not only a low outflow rate of long-term unemployment itself that is responsible for a high incidence of long-term unemployment. Machin and Manning stress that this hazard rate is not even the main contributor to a large share of long-term unemployment. They further find that the degree of duration dependence contributes to the share of long-term unemployed; as the unemployment spell increases, the probability of finding a job decreases.3 Nevertheless, this has always been the case and the relative outflow
rate of short-term and long-term unemployed has not fallen over time.
The authors end by noting that although it is not clear that duration dependence in the exit rate from unemployment is worse in Europe than the United States, one should not conclude that long-term unemployment is not a particular problem in Europe. As they say:
page 15
‘The sheer numbers of people unemployed for long periods of time means that, if these individuals are less effective in competing for jobs, then unemployment is likely to be much more persistent in Europe thereby making it hard to reduce the level of unemployment.’ (p. 30)
page 16
THEORETICAL BACKGROUND
The previous section clearly indicates a number of variables that can be considered determinants of the aggregate unemployment rate. To put these variables in a broader perspective, we describe a framework for studying (long-term) unemployment in this section. The framework is based on the influential work of Layard, Nickell, and Jackman (1991). In a foreword to the 2005 reissue of their book, Layard et al. insist that after 14 years, their conclusions still hold up and their models are still valid. In addition, a number of studies reported in the previous section estimate empirical equations that have a similar functional form as the one Layard et al. end up reporting. This section is also inspired by Nickell (1998) and Nunziata (2002). What follows is a non-technical (where possible) overview of the framework used in this paper.
The economy is modelled by four equations in a simple but general setting. Equation (1) describes aggregate demand
(1) = − + ,
where y = real GDP, xn = exogenous nominal demand factors, p = GDP deflator, xr =
exogenous real demand factors. Lower case letters denote logarithms. The exogenous nominal and real demand factors should be thought of as exogenous demand factors that can be influenced by government macroeconomic policy. For example, Layard et al. propose the money stock as an exogenous nominal demand factor. The interest rate is another such factor. Exogenous real demand factors may include measures of fiscal stance, indicators of world economic activity or the relative price of imports.
Production is modelled by a (reversed) Okun’s law-type relationship. Equation (2) gives the relationship between the unemployment rate and demand
(2) = −,
where u = unemployment rate. Equation (2) states that a rise in unemployment is always
followed by a fall in demand. Demand shocks may be the initiating factor in changing unemployment, but more often a supply-side shock alters unemployment, with demand accommodating.
Price-setting behaviour is modelled by Equation (3) and takes the form of a price mark-up on wages
page 17
where w = wages, we = expected wages, pe = expected GDP deflator, k = capital stock, l =
labour force, zp = exogenous factors influencing the mark-up. In short, the price mark-up
on wages depends on the wage shock, the price shock, the capital-labour ratio, and other exogenous factors. The coefficients β1 and β2 capture nominal inertia in wages and prices.
Nominal inertia in wages exists because although wages might rise without much resistance, they will not go down easily. β1 captures this effect. β2 measures the extent of price
stickiness in the economy. Price stickiness varies with the level of product market competition. The capital-labour ratio captures trend productivity.
The fourth and final equation to complete the framework is the wage setting equation (4) − = − − ∆ − − −
− ,
where zw = exogenous factors influencing wage setting. The wage setting equation takes the
form of a wage mark-up on prices. The wage setting equation also includes price shocks and trend productivity. Current unemployment has an impact on wage setting, because the state of the labour market determines employer and employee bargaining power. Following Nickell (1998), we also assume changes in the state of the labour market have an additional impact on wages through insider dynamics or long-term unemployment effects. Institutional factors of the labour market enter the equation through elasticities γ1 and γ2.
As Nickell (1998) notes:
‘Obviously, unemployed individuals are likely to be more effective if they are ready, able, and willing to take up vacancies. Equally obviously, a generous and lax benefit system can easily reduce readiness and willingness, and a poor training system may lower ability. Similarly anything which reduces the effectiveness of the long-term unemployed as fillers of vacancies, such as long benefit durations, will lower the parameter [y2].’ (p. 803)
Equations (1) through (4) constitute a simple, stylized model of the economy. Now suppose that demand, price, and wage factors can be divided into anticipated and unanticipated elements. In other words:
= + , = + , = + and = + ,
where , , and are all anticipated and εn, εr, εp and εw are serially uncorrelated,
page 18
(5) , = + , !+ "ℎ , + $"% , + & + , ,
where u = unemployment rate, h = vector of labour market institutions, s = controls for
macroeconomic shocks, µ = country fixed effect, ε = random error. The subscript i denotes
country i and subscript t denotes year t.
The vector of labour market institutions, h, constitutes the following equation (6) "ℎ
, = '()* , + +, , + +-- , + ./00, , + 12*( , +
340 , + 560 , + 768 , + 9:, , ,
where ALMP = government spending on active labour market policies, BD = index for
benefit duration, BRR = unemployment benefit replacement rate, COOD = degree of
employer and union coordination, EPL = measure of employment protection legislation, HO = home ownership, TO = degree of trade openness, TW = tax wedge, and UD = union
density. All factors in equation (5) are institutional characteristics of the labour market. The vector of macroeconomic shocks, s, constitutes the following equation
(7) $"%
, = $,2)< , + $ ∆=>? , + $-=-( , + $.6?*< , + $1606< , ,
where D2MS = acceleration in money supply, ∆INF = change in the inflation rate, RIRL =
long-term real interest rate, TFPS = total factor productivity shock, and TOTS = terms of
trade shock. All shocks have a zero mean and die out after one period.
In the empirical model, the explanatory variables are represented by all factors influencing the equilibrium level of unemployment and the shocks that cause it to deviate from its equilibrium. The country specific intercept accounts for heterogeneity of unemployment rates that is not captured by institutional and control variables.
The model discussed above makes predictions about the unemployment rate. We are, however, especially interested in the statistical results for long-term unemployment. We see no intuitive theoretical objections to a straightforward substitution of the unemployment rate by long-term or, for that matter, any other duration category of the unemployment rate. Institutional characteristics influencing the inflow in and outflow out unemployment will also have an effect on long-term unemployment. An inspection of the average correlation coefficients reveals high enough values to justify this simple switch. Table 1 reports these coefficients.4 Nickell (1998) also follows this approach. Moreover, including
4 The correlation coefficient for Australia range is 0.310. Also, the coefficients for Greece for LTU-U and for
page 19
more than just aggregate unemployment as an independent variable leads to more insights into the workings of the labour market.
Table 1: Correlation matrix for aggregate, long- and short-term unemployment.
Aggregate unemployment rate Long-term unemployment rate
Aggregate unemployment rate 1 0.729
Long-term unemployment rate 0.729 1
page 20
EMPIRICAL APPROACH
Although we are primarily interested in the determinants of long-term unemployment, since, as we have explained earlier, it is a source of great economic inefficiency and personal hardship, it would be interesting to see the effect, if any, of the chosen determinants on other duration categories of unemployment. We estimate essentially three separate equations in this paper. The research questions will be answered through the estimation of these equations. The equations containing long-term unemployment as the dependent variables are of specific interest to us. The first of these equations takes the long-term unemployment rate as the dependent variable. Regressing this ratio on the explanatory variables shows which effects influence the rate of long-term unemployment. The second equation features the ratio of long-term unemployment to total unemployment as the dependent variable. The third equation is estimated to make clear the overall effect on the aggregate rate of unemployment.
This section builds upon the economic model set out in Section 3. A further discussion of the variables; their reason for inclusion, expected effect, and data sources can be found in the next section.
Error correction
page 21
For now, we assume all prerequisites for ECM estimation hold (that they hold, is proven below) and derive the functional form of the ECM. Using Equation (5) as the baseline equation, the ECM can be deducted as follows.
Equation (5) is repeated below for convenience
(5’) , = + , !+ "ℎ , + $"% , + & + , .
The long-term effect of the independent variables on the dependent variable is given by (8) = + Γ"ℎ
,
where the dependent and independent variables are assumed to be constant, the vector of macroeconomic shocks disappears (macroeconomic shocks have no long run effect), the error term is set to zero and country-specific intercepts disappear since the underlying economic principles that are employed to establish the long run equilibrium are assumed to apply similarly in all countries. The upper-case gamma is the long-run version of the lower-case gamma. The residual in this equation contains all short-run dynamics and is presumed stationary (proven below by a unit root test). The ECM also incorporates the short-run dynamics. Since there is no short-run stationary relationship, taking the first difference of (5) gives:5
(9) ∆ , = + "∆ℎ , + $"% , − Φ !+ , ,
where kt-1 is the residual of the long run relationship, or != , !− B"ℎ , !+ C.
Combining this residual with Equation (9) yields the ECM
(10) ∆ , = "∆ℎ , − ΦB , !− Γ"ℎ , !+ C + $"% , + , .
In Equation (10), the first term on the right-hand side denotes the short run effect of the institutional factors on unemployment. The parameter Ф is presumed positive so that deviations from the long run equilibrium are corrected by the terms between brackets with speed Ф. The terms between brackets denote the long run effect on unemployment, and the vector of institutional characteristics. The third term on the right-hand side denotes the macroeconomic shock variables that only have an effect in the short run; they are mean reverting. The last term on the right-hand side is the stochastic error term.
5 The macroeconomic shocks are not differenced, because the assumption that they are mean-reverting means
page 22
Again, the vectors of labour market institutions and macroeconomic shocks are as in Equations (6) and (7).6 They are repeated here for reference
(6’) "ℎ , = '()* , + +, , + +-- , + ./00, , + 12*( , + 340 , + 560 , + 768 , + 9:, , , (7’) $"% , = $,2)< , + $ ∆=>? , + $-=-( , + $.6?*< , + $1606< , .
Stationarity
Still, the variables must be tested for stationarity and cointegration to test whether reliable estimation can be performed. To test the dependent and independent variables for stationarity, two tests are usually recommended. The first is the Augmented Dickey-Fuller (ADF) unit root test and it tests the null hypothesis of non-stationarity against the alternative hypothesis of stationarity, by comparing the test statistic to the critical values as calculated by MacKinnon (1991, 1996).
The second stationarity test is the Phillips-Perron (PP) test. The PP test estimates the standard Dickey Fuller test equation, but modifies the test statistic in such a way that serial correlation does not affect the asymptotic distribution of the test statistic. The test statistic should be compared to the MacKinnon critical values.
We check for unit roots in each country series using both the ADF with trend and PP test. With the exception of the shock variables (D2MS, ∆INF, TFPS, and TOTS) and the
long-term real interest rate, all variables are non-stationary. Obviously, the shock variables are mean reverting and are therefore inherently stationary. The non-stationary variables are integrated of order one, meaning differencing the series once makes them stationary. Table A1 reports the results of the ADF unit root tests. Results of the PP test are not reported, as they do not lead to different conclusions.
One way to handle the presence of unit roots is by estimating the equation in first differences. However, taking first differences could lead to loss of information and subsequent underestimation of the relationships between variables. Ten Cate and Draper (1989) show that estimation in first differences leads to a variance of the first difference of the dependent variable that is smaller than that of the first difference of the independent variable(s). Consequently, the estimated coefficient is smaller than the correct coefficient,
6 The coefficients of the vectors of equation (6’) are denoted in lower-case gamma, denoting the short-run
page 23
leading to incorrect estimates of the long run effect. Therefore, estimating long run effects in first differences is not preferable.
However, if the variables are cointegrated, which means there exists a stable long-term relationship between these variables, non-stationarity should not be a big problem: the long-term relationship makes sure the time series do not diverge or converge, but move together. In other words, if all non-stationary variables are integrated of the same order, and there exists a linear combination that is of a lower order of integration, the variables are cointegrated. This argument originated with Engle and Granger (1987). The presence of cointegration is what we are looking for when estimating the error correction model. Cointegration can be tested with a Johansen (1991, 1995) cointegration test. The simpler way to test for cointegration is to test the errors of the baseline specification for unit roots.7
Rejection of the null hypothesis of the existence of a unit root leads to the acceptance of cointegration. We use the test suggested by Maddala and Wu (1999) which combines the results of N country-by-country unit root tests, each with a p-value Pi, in the statistic −2 ∑ ln * and is distributed as Chi-squared with 2N degrees of freedom. Table 2 reports the results of this procedure using both ADF with trend and PP tests for each variant of Equation (5) (variants are discussed in the next subsection). We reject the null hypothesis of a unit root in the residuals and conclude a cointegration relationship exists and estimating an ECM is valid.
Table 2: Results of the ADF with trend and PP tests
ADF test PP test
Aggregate unemployment rate 67.01 (0.00) 60.72 (0.00)
Long-term unemployment rate 64.02 (0.00) 68.18 (0.00)
Share of long-term unemployment 66.00 (0.00) 67.03 (0.00)
Notes: The test statistics are Chi-squared distributed. MacKinnon approximated P-values are reported in brackets.
Variations
Equation (10) is the baseline equation and is used as the starting point for the other regressions. Three variants of this equation are estimated for the period 1960 to 2006 for both low-share countries and high-share countries. In each of the three variants a different dependent variable is used. These include the aggregate unemployment rate, long-term
7 This is the method Engle and Granger (1987) suggested. The entire estimation method is therefore known
page 24
unemployment rate (for people who have been unemployed longer than 12 months), and the share of long-term unemployed.8 We study the long-term unemployment rate and share
of long-term unemployment because these groups are of special interest to us. The aggregate rate is included for comparison, finding the overall effect and to make clear the possible existence of trade-off effects.9
The division between low-share and high-share countries is made by first calculating the average share of long-term unemployment over all countries and years. Next, the average share over all years per country is compared to this total average. If the average country share is lower than the total average share, that country is denoted a low-share country. Conversely, the same is true for high-share countries. Table 3 gives the sample composition per group.10
Table 3: Group composition
Low-share countries High-share countries
Austria Norway Belgium Italy
Australia New Zealand Germany Netherlands
Canada Sweden Ireland Portugal
Denmark Switzerland Spain United Kingdom
Finland United States France
Japan
The results of the regressions are reported in Section 6, where the focus lies on the long-term unemployed, and the difference between the low-share and high-share countries. First, Section 5 gives an in-depth description of the variables in each empirical equation and their expected effect on unemployment.
8 Following the substitution of the dependent variables, logically, the lagged dependent variable included in
each convergence term is substituted as well.
9 There should be a relationship between both rates and the share of long-term unemployment. Taking
account of this relationship by placing restrictions on the equations is beyond the scope of this paper and we have therefore not used any restricting conditions.
10 Basing the group composition on the long-term unemployment rate or aggregate unemployment leads to
page 25
VARIABLES AND DATA
This section contains a description of the variables used in the econometric estimation and gives the reasoning for their expected effect on the dependent variables. A summary of the hypotheses can be found at the end of this section in Table 3. The sample consists of annual observations for 20 OECD countries over the period 1960 – 2006. Table A3 in the
appendix contains an overview of descriptive statistics for the variables found in this section. The data appendix describes the construction of the variables in depth: it describes their sources, data manipulation and sample characteristics.
Unemployment
The dependent variable in the unemployment equation is the unemployment rate (u). This
is the ratio of unemployed people to the total labour force, where the latter is defined as the sum of employees, the self-employed, unpaid family workers and the unemployed. The unemployed are people of working age who, in a specified period, are without work, are both available for, and are actively seeking work.
The aggregate unemployment rate can be divided into a short-term and a long-term component. The short-term unemployment rate (stur) is defined as the ratio of people who
have been unemployed for less than one year to the total labour force. The long-term unemployment rate (ltur) is constructed in a similar way, but it concerns people who have
been unemployed for more than one year.
Because we are also interested in the shares of short-term and long-term unemployment, these shares are regressed on the explanatory variables. The share of short-term unemployed
(STU) is the ratio of long-term unemployment to aggregate unemployment (LTU).
Conversely, the share of long-term unemployed is the ratio of long-term unemployment to aggregate unemployment.
The coefficient on the lagged dependent variable gives the speed of convergence. This coefficient denotes the rate at which deviations from the long run equilibrium are corrected.
Active labour market policy
The extent to which active labour market policy (ALMP) conducted by the government is
page 26
when such programmes like training, job rotation and sharing and employment incentives are targeted on the unemployed and groups at risk of unemployment.11
Active labour market policy is considered beneficial for helping the unemployed back to work. These activities primarily consist of job placement services, vocational training, and hiring subsidies and should help registered unemployed connect to the labour market, while also encouraging inactive people to enter the labour market. ALMP thus contains two
opposing effects. Activating programmes that have an educational character usually take less than one period to complete, so the effects of this variable are noticeable within one period. Other activating programs are aimed at facilitating entrance to the labour market and have an immediate impact. As such, it is not necessary to incorporate ALMP as a lagged variable.
Other studies that have found this variable to be significant include Nickell (1997) and Campbell (2000). Campbell even argues that active policies are necessary to connect the
long-term unemployed to job opportunities.
In general, ALMP initially raises unemployment (the encouraging inactive people effect is dominant), but after this initial period and, presumably, after completing the programme, unemployment falls (the connecting to the labour market effect is dominant). Because activating programs are especially designed to activate the long-term unemployed, a larger beneficial effect is expected for this group.
Hypothesis: activation measures help the unemployed connect to the labour market, while
also welcoming inactive people to the same market. The overall effect is ambiguous ex ante.
Benefit duration
Benefit duration (BD) refers to the period during which the unemployed are eligible for
unemployment benefits. This index is constructed by taking a weighted average of the ratio of the benefit replacement rate in year two relative to year one and the ratio of the benefit replacement rate in year four relative to year one. If the unemployed are only entitled to one year of benefits, the index is equal to zero, while the index is equal to one when benefit provision is constant. A long benefit eligibility period discourages unemployed people to find a job quickly, i.e. the outflow out of unemployment falls. If the index is on the upper part of the range 0 to 1, the social security system is considered generous, while the opposite is true for the lower part of the range. We expect the benefit duration index to be positively related to unemployment. No duration specific effects are expected.
page 27
Hypothesis: generous benefit provision lowers search effort and willingness to work.
Benefit replacement rate
The benefit replacement rate (BRR) measures the percentage of wage income that is
replaced by unemployment benefits for the unemployed in the first year of unemployment. A high replacement rate corresponds to generous national social security system, while a low rate denotes a parsimonious system. A high rate therefore mitigates the monetary need for finding a paid job, as the unemployment benefit provides sufficient income. As a result, we expect the benefit replacement rate to impact positively on unemployment.
Hypothesis: generous benefits lower search effort and willingness to work.
Bargaining coordination
Bargaining coordination (COOD) measures the degree of coordination in the bargaining
process on the employers’ as well as on the union’s side. This index has a range 1 to 3 increasing in the degree of coordination. The first leg of this index corresponds roughly to coordination at the company or plant level without upper-level involvement. The second leg corresponds to coordination at the sectoral level with moderate involvement of major bargaining partners. The third and highest leg corresponds to coordination at a central level with an encompassing union confederation or governmental intervention. A higher level of coordination is associated with a higher degree of professionalism in wage setting, so a large number of external factors (the unemployment rate in particular) are considered when setting the wage. More coordination should lead to smaller wage increases and more wage moderation, so less inflows into and more outflows out of unemployment.12 The
relationship between this index and unemployment is therefore expected to be negative.
Hypothesis: a high degree of bargaining coordination is synonymous with professionalism in
wage-setting and internalising external factors such as the unemployment rate.
Employment protection legislation
Employment protection legislation (EPL) is a collective term for the set of regulations that
limit firms to hire and fire employees. Hiring restrictions reduce flows out of unemployment and firing restrictions reduce flows into unemployment. An example of a
12 Calmfors and Driffill (1988) argue the relation between bargaining coordination and unemployment is
page 28
hiring restriction is a provision favouring the employment of disadvantages groups in society. Firing restrictions include pre-notification periods and severance pay. It is not clear ex ante what the effect of EPL is on unemployment because the hiring and firing
restrictions have opposite effects. Hiring restrictions sustain unemployment during an economic boom, while firing restrictions sustain employment during a recession. In other words, in economies where strict EPL exists, labour markets are less flexible.
Empirical studies also seem to find ambiguous results. In a review by OECD (1999), results
corresponding to theory are found. On employment and unemployment:
‘[…] stricter EPL raises employment for prime-age men but lowers employment for youths and women, with the overall effect being a net reduction. Similarly, youths and perhaps women appear to bear a larger share of the burden of unemployment. However, these associations tend to be weaker or entirely absent when multivariate techniques are used to control for other factors that influence employment and unemployment levels.’ (p. 48)
Another result is that EPL tends to increase self-employment and lower turnover rates in
the labour market. These points imply that although fewer people become unemployed in countries where EPL is stricter, once unemployed, they have a higher risk of remaining
unemployed for a longer period. Everything considered this may point to a negative relationship between unemployment and EPL in the short run, while there is a positive
relationship between unemployment and EPL in the long run.
Hypothesis: strict EPL lowers unemployment in the short run, while raising unemployment
in the long run.
Home ownership
When a large share of people on the labour market owns a house, rather than renting it, the labour market will be highly inflexible as people are reluctant or unable to sell their home and move elsewhere. When these people become unemployed, they will remain unemployed relatively longer: decreasing the outflow rate out of short-term unemployment. This suggests the impediments to flexibility strongly hold for the short run, while the effect may lessen in the long run when people are more able and willing to change their housing preferences.
Hypothesis: being a homeowner decreases the willingness or ability to move and accept a job
page 29
Trade openness
The ratio of exports plus imports to GDP is used as a proxy for Trade openness (TO). An
open economy is more susceptible to movements in the world economy via trade partners. Therefore, shocks originating from other countries will affect an open economy more than a closed economy, which is more self-sufficient. In addition, trade openness lowers the bargaining power of employees, as they compete with a larger, global market.13 On the
other hand, because a closed economy is more isolated, it will not be able to profit from favourable export (increases demand) or import (reduces inflation) prices. The effect of trade openness on unemployment is therefore ambiguous.
Hypothesis: trade openness has two, opposing effects on unemployment. It is not clear ex
ante which effect dominates.
Tax wedge
The employment tax rate equals employers’ social security contributions divided by the sum of these contributions and total compensation for employees. The direct tax rate is the income tax plus employees’ social security contributions divided by households’ current receipts (the sum of compensation of employees, property income, social contributions and benefits and other current transfers). The indirect tax rate is the ratio of indirect taxes minus subsidies to households’ final expenditures. The tax wedge (TW) combines these
three tax rates (employment tax, direct tax, and indirect tax). The tax wedge acts as a mark-up on wages, thereby artificially raising wages, influencing firms’ hiring decisions and lowering outflow out of unemployment. As a result, a high tax wedge is associated with high unemployment.
Hypothesis: the tax wedge acts as a mark-up on wages and decreases firm’s willingness to
hire.
Union density
The ratio of total reported union members to salaried employees serves as a proxy for union density (UD). Union density gives the coverage of unions and is a direct measure of their
representativeness. According to economic theory, small and large unions are best for combating unemployment. Small unions are not important enough to wage-setters to have an impact on the decision, while large unions internalize the unemployment effect of wage
page 30
setting. Medium-sized unions are large enough to have an impact on wage setting, but are too small to internalize the unemployment costs. In our sample, the average union density is 40% with a median of 38%, so unions cover on average 40% of workers. This percentage can be classified as medium coverage. In effect, this means unions will do more damage than good and excessive wage claims will hurt unemployment.
Hypothesis: on average, unions are large enough to successfully negotiate higher wages, while
failing to take into account the effect on unemployment.
Acceleration in money stock
Changes in the rate of growth of the money stock affect interest rates in the short run. An acceleration in money stock growth could lead to an acceleration in spending growth, due to an increase in investment and consumption spending. In other words, when followed by an increase in spending, an increase in the money stock decreases unemployment.
Hypothesis: in the short-run, unemployment deviates from its long-term equilibrium
through changes in the acceleration of the money stock.
Change in inflation rate
Assuming a Phillips curve type relationship, there exists a trade-off between inflation and unemployment. Economic theory argues that this relationship only holds in the short run, because in the long run there is only one unemployment rate consistent with a stable inflation rate. The Phillips curve relationship is a negative one.
Hypothesis: in the short-run, unemployment deviates from its long-term equilibrium
through the inflation trade-off.
Long-term real interest rate
A rise in the real interest rate increases the user cost of capital and leads to lower capital accumulation and labour productivity. A secondary effect flows through the returns on non-human wealth. A high interest rate increases returns on non-human wealth and increases the reservation wage. These effects should be translated into positive coefficients.
Hypothesis: in the short-run, unemployment deviates from its long-term equilibrium
page 31
TFP shock
Because wages do not immediately react to productivity growth, surprises should create a temporary shift in structural unemployment. A positive shock, for example, causes employees to be paid in short of their productivity, leading labour to become relatively cheaper.
Hypothesis: in the short-run, unemployment deviates from its long-term equilibrium
through a TFP shock.
Terms of trade shock
If real wage resistance is present, a fall in the terms of trade is not accompanied by a fall in the real wage and unemployment rises. Real wage resistance is less prominent in the long run, so the terms of trade shock is temporary. We expect negative signs for the coefficient of the terms of trade shock.
Hypothesis: in the short-run, unemployment deviates from its long-term equilibrium
through real wage resistance.
Summary table
page 32
Table 4: Hypotheses summary regarding the independent variables’ effect on the dependent variables.
Expected relationship Expected relationshipExpected relationship Expected relationship Variable
Variable Variable
Variable Short run Long run
Labour market institutions
Active labour market policy + -
Benefit duration index + +
Benefit replacement rate + +
Bargaining coordination - -
Employment protection legislation - +
Home ownership + +
Trade openness unknown unknown
Tax wedge + +
Union density + +
Macroeconomic controls
Acceleration in money stock - no effect
Change in inflation rate - no effect
Long-term real interest rate + no effect
TFP shock - no effect
page 33
OVERVIEW OF FINDINGS AND DIAGNOSTIC TESTS
There is a range of tests that should be used to find out whether the specification of our empirical model is optimal and unbiased. The diagnostic and inference tests set loose upon the empirical testing are reported below, followed by a discussion of the regression results. In the discussion of the results, the results for the high-share countries are contrasted with the results for the low-share countries. Policy implications are discussed in the conclusion in Section 7.
Diagnostics
Six equations are estimated in total: three for the sample of low-share countries and three for the high-share countries. The sample members can be found in Table 3 in Section 3. Because the countries are pooled in estimation, it is necessary to check whether a hypothesis of homogeneous coefficients is accepted. If coefficients are heterogeneous, the pooled estimator is inconsistent. To test the assumption of poolability, we conduct a poolability test of the kind suggested by Roy (1957), Zellner (1962), and Baltagi (1995). The Chi-squared test statistic of 81.03 (p-value: 0.99) for the low-share pool and 79.75 (0.99) indicates we cannot reject the null hypothesis of poolability for either sample. The data set is of a panel nature. Employing panel data with a substantial number of cross section units takes advantage of a much richer data source than using pure time series data.
Regular least squares estimation assumes homoskedasticity of the errors. Heteroskedasticity is a problem often found in cross section data and should therefore be tested for. Heteroskedasticity does not cause OLS coefficient estimates to be biased. However, the variance (and thus standard errors) of the coefficients tends to be underestimated, inflating t-scores and sometimes making insignificant variables appear to be statistically significant. We reject the null hypothesis of homoskedasticity in a Goldfeld-Quandt test for all equations.
We use a Breusch-Godfrey LM test to test for serial correlation. If serial correlation is present in the errors, ordinary least squares estimation is still unbiased, but no longer best. The Chi-squared test statistic is equal to 52.70 (p-value 0.00) in the best case among the equations and leads us to reject the null hypothesis of no serial correlation. This result warrants the adoption of an AR(1) structure in the disturbances.
page 34
estimator is not defined. We examine the sample correlation coefficients between pairs of explanatory variables. Because the highest absolute correlation coefficient found is 0.56, we reject the notion of exact collinearity in our sample. A collinear relationship might involve more than two explanatory variables and a simple inspection of correlation coefficients does not expose this relationship. Regressing each explanatory variable on the other explanatory variables in a so-called auxiliary regression and examining the adjusted R-squared, confirms the absence of multicollinearity in our sample. The highest R-squared obtained is 0.59. The diagnostics suggest the adoption of a feasible Generalised Least Squares estimation technique, with a covariance matrix that corrects for cross-section specific heteroskedasticity and an AR(1) structure in the errors. In iteration 0 of GLS, the estimated OLS residuals are used to estimate the error covariance matrix. Then in iteration 1, GLS estimation minimizes the sum of squares of the residuals weighted by the inverse of the sample covariance matrix. As the ECM is estimated in first differences, no country fixed effect may be imposed. First differencing already cancels out any individual specific component.
Explicit effects
As it can be difficult to grasp the difference between short run and long run effects and apparent contradictions like a short run effect on long-term unemployment, we will try to explain the explicit workings of these effects in this subsection.
For the aggregate unemployment rate, it should not be too difficult to distinguish between short run and long run effects. In the short run, which is within one year, the aggregate unemployment rate is affected by both temporary and inherently permanent factors that influence the inflow into and outflow out of unemployment. In the long run only the effects of permanent factors remain. This distinction between temporary and permanent factors is also valid for the other dependent variables, but it will not be mentioned again in this subsection.
Short run effects on long-term unemployment chiefly consist of measures that limit the outflow out of short-term unemployment, the inflow into long-term unemployment and the outflow out of long-term unemployment. Effects that influence the inflow into short-term unemployment do not influence long-short-term unemployment in the short run. This effect only adds to the mix in the long run, because people are only counted as long-term unemployed after they have been short-term unemployed for one period.
page 35
in inflow rates are associated with the attractiveness of the unemployment situation and firing decisions of firms.
The coefficients for the aggregate and long-term unemployment rates denote which part of the labour force enters or exits unemployment following a change in the institutional characteristics of the labour market. The coefficients for the share of long-term and unemployment explain how the composition of the unemployed changes following a change in labour market institutions.
Governments that aim to reduce long-term unemployment by setting a target of lowering the long-term unemployment rate and actually see a reduction in that rate, should check whether the aggregate unemployment rate decreased as well. If this is the case they were successful. If however, the aggregate unemployment rate remained stable or even increased, it means a lower long-term unemployment rate comes at a cost of a higher short-term unemployment rate.
In discussing the effects on the shares of short- and long-term unemployment, always keep in mind that one effect is always dominant in that a fall in the share of one duration category must cause a rise in the share of the other category, even if that other category also falls as an isolated effect.
To better understand the flows in and out of unemployment consider Figure 4 below.
Figure 4: Flow model of the labour market.
Note: The boxes denote states and the arrows denote flows.
short-term unemployment employment
long-term unemployment
page 36
Estimation results
Lagged dependent variable
Lagged dependent variables are included to account for the convergence effect. The convergence coefficient denotes the speed with which short run deviations from the long run equilibrium are corrected. A negative sign is expected for the coefficients. A coefficient that is equal to 1 implies instantaneous adjustment, but this is not expected. The results show that all coefficients are negative and highly significant. The coefficients range from -0.542 to -0.161 for the low-share countries and -0.485 to -0.252 for the high-share countries, meaning 0.542 to 0.161 points compensate every point of mismatch in the next period. The speed of convergence is higher for the long-term unemployment variables for the low-share sample. The speed of convergence is higher for the aggregate unemployment rate for the high-share sample. These findings confirm the convergence hypothesis.
Active labour market policies
Programmes and incentives to get people back into the workforce are considered active labour market policies. These activities should help registered unemployed connect to the labour market, while also encouraging inactive people to enter the labour market. For the high-share countries this variable is insignificant in all but two cases. Increasing expenditure on activating policies increases the long run aggregate unemployment rate and the long run share of long-term unemployment by 0.776 and 2.167 points, respectively, for each 1 point increase in spending. This result is remarkable because it suggests that ALMP are not effective at lowering unemployment in countries with already high unemployment figures. The fact that most ALMP are directed at the long-term unemployed and should therefore be especially effective in helping this group back to work is not reflected in the coefficients. The coefficients for the sample of low-share are more encouraging. With the exception of the share of long-term unemployment, activating measures are effective in lowering unemployment. This effect is larger in the long run for the long-term unemployment rate. The long run effect is insignificant for the aggregate rate. The positive relationship between ALMP and the share of long-term unemployment is, in light of the negative relationship of the rate variables, probably caused by a stronger negative reaction of shifts in ALMP on short-term unemployment.
