Do Active Labor Market Policies Increase Employment?
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund

Contributor Notes

Author’s E-Mail Address: mestevão@imf.org

Using panel data for 15 industrial countries, active labor market policies (ALMPs) are shown to have raised employment rates in the business sector in the 1990s, after controlling for many institutions, country-specific effects, and economic variables. Among such policies, direct subsidies to job creation were the most effective. ALMPs also affected employment rates by reducing real wages below levels allowed by technological growth, changes in the unemployment rate, and institutional and other economic factors. However, part of this wage moderation may be linked to a composition effect because policies were targeted to low-paid individuals. Whether ALMPs are cost-effective from a budgetary perspective remains to be determined, but they are certainly not substitutes for comprehensive institutional reforms.

Abstract

Using panel data for 15 industrial countries, active labor market policies (ALMPs) are shown to have raised employment rates in the business sector in the 1990s, after controlling for many institutions, country-specific effects, and economic variables. Among such policies, direct subsidies to job creation were the most effective. ALMPs also affected employment rates by reducing real wages below levels allowed by technological growth, changes in the unemployment rate, and institutional and other economic factors. However, part of this wage moderation may be linked to a composition effect because policies were targeted to low-paid individuals. Whether ALMPs are cost-effective from a budgetary perspective remains to be determined, but they are certainly not substitutes for comprehensive institutional reforms.

I. Introduction

The steady rise in unemployment rates in the 1970s and 1980s in Europe has been variously attributed to mismatches between labor skills demanded and supplied, excessive wages vis-à-vis productivity levels, over-generous out-of-work benefits, and rigid institutions designed to curb labor churning. Among the possible solutions was the introduction of government policies to better mold labor force characteristics to changes in demand, to lower firms’ labor costs directly and to increase job-search efficiency. These policies have been grouped under the label of “active labor market policies” (ALMPs).2 Indeed, during the second half of the 1990s, employment performance improved appreciably in several European countries raising the presumption that such policies actually worked.

This paper evaluates the aggregate effect of ALMPs on employment and finds a positive correlation between spending on ALMPs as a percentage of GDP and the employment rate in the business sector in the 1990s, but not in the late 1980s, when such expenditure was still relatively small. Among all the ALMPs, direct subsidies to job creation were the most effective in raising employment rates, while expenditures on training programs seem to have been largely ineffective. By estimating a wage-setting curve for the same sample of countries, it is also shown that substantial wage moderation (reduction in real wages adjusted by technology for a given rate of unemployment) was associated with increases in ALMPs in the 1990s. These results reveal one of the possible sources of the hitherto unexplained wage moderation in some European countries.3 However, even though ALMPs do increase employment, they also weigh heavily on the budget. Institutional reforms to lower production costs and enhance labor market flexibility and work incentives are a better way to increase employment rates.

The methodology used here addresses three key shortcomings of previous studies of the effect of ALMPs on the labor market using cross-country aggregated data, which have generally been inconclusive.4 First, the specification used in many of these studies tends to overestimate the effect of ALMPs on the unemployment rate (very few studies focus on the most appropriate measure of labor market performance, the employment rate). Second, other studies use either pooled cross-country regressions, or panel data with random effects, with no (or very little) within-country variation in ALMP spending. In particular, most of the literature has focused on the effect of institutions on unemployment rates, leaving ALMPs as a control variable. In many cases, such a focus has limited the amount of time variation allowed in the data as institutions tend to be quite constant in time. Third, data used in previous studies did not extend beyond 1995.

The remainder of the paper is organized as follows. The next section explains why ALMPs may be effective in raising employment rates. Section III uses the empirical strategy of previous work as a guideline for the identification challenges to be met in the simple econometric setup used here. Section IV sets up the empirical work and discusses estimates for the effect of ALMPs on employment rates. Section V provides evidence on the link between expenditures on ALMPs and wage moderation in the panel of countries under study. Section VI concludes this paper.

II. Why Might ALMPs Increase Employment?

ALMPs may affect employment through at least five channels. To catalogue these effects consider a simple labor market model with a downward-sloped labor demand and an upward-sloped wage-setting relationship resulting from the wage bargaining models discussed in Layard et al (1991) (Figure 1).5

Figure 1.
Figure 1.

How Might ALMPs Affect Employment?

Citation: IMF Working Papers 2003, 234; 10.5089/9781451875645.001.A001

First, ALMPs may generate more efficient matching between job vacancies and unemployed workers because of adjustments in job-seekers’ skills (for instance, through training programs) or more effective searching (for instance, through more active employment agencies). The resulting smaller ratio between vacancies and unemployment reduces wage pressure, which causes a downward shift in the wage-setting curve, and, because vacancies are costly to employers, provides an outward shift in labor demand. Both effects will tend to raise employment with an uncertain final effect on real wages.

Second, labor force productivity may increase, owing to either training programs or on-the-job learning, in the case of direct subsidies to job creation. It may also affect non-program participants through an externality effect. This productivity increase would shift labor demand up and lift employment and wages.

Third, ALMPs may keep unemployed workers attached to the labor force, even after a longer period of inactivity. The resulting stronger competition for jobs would shift the wage-setting curve down, raising employment and reducing wages.

Fourth, job creation programs (e.g., direct subsidies to low-skill employment) may generate windfall effects, i.e., substitute for nonsubsidized employment, making ALMPs ineffective. However, the associated income effect from an overall reduction in labor costs could be large enough to increase labor demand, implying higher wages and employment in equilibrium.

Fifth, active policies may lower the disutility of being unemployed, as they provide an occupation to otherwise unemployed workers, some income, and a hope of keeping their labor skills. Workers would then demand higher wages during bargaining and, in equilibrium, employment would be lower.

Finally, an important caveat should be noted. Even if a positive effect on employment might be discerned, the fiscal cost of ALMPs may be very high, raising the question of their overall effectiveness in a general equilibrium or cost-benefit sense.

III. Identification Issues and a Critical Look at Previous Studies

Previous studies of the effect of ALMPs on labor market performance suffer from one or more of the following flaws: (i) the inability to separate the role of labor market institutions from the role of policies, whose resolution calls for using a panel database; (ii) small sample size that leads to insufficient time variation in ALMPs (quite related to (i)); (iii) unstable results depending on the metric used for ALMPs; (iv) the reverse causality from movements in employment to changes in expenditures in ALMPs (e.g., when employment is low, more people sign up for training and consult public employment services, while the government is more likely to enact new or more generous subsidy programs); and (v) a focus on unemployment, which leads to overestimation of the returns of ALMPs on employment and neglect of labor force participation effects.

The first studies used only a very limited number of observations (usually around 20) with countries as individual units and no time variation in the data.6 Since a few institutional controls cannot be expected to account for all cross-country diversity unrelated to ALMPs, this method is likely to wrongly attribute the influence of some unobserved institutional features on the unemployment rate to ALMP spending.

Subsequently, the work conducted during the second half of the 1990’s take advantage of the extended availability of data to use panel methods, therefore improving the identification of institutional effects on the unemployment rate.7 However, the same studies tend to pool the data in two periods or average the information on ALMP expenditures to minimize the reverse causality problem going from movements in the unemployment rate to variations in expenditures on ALMPs, which bias the estimates toward finding a positive correlation between both variables. In addition, this averaging neglects variation in the time domain as a source of parameter identification. Other studies have attempted to handle reverse causality by using the ratio of expenditures on ALMPs to unemployment but such a measure only changes the sign of the bias toward finding a negative correlation between ALMPs and unemployment rates.8

The focus on the reverse causality problem is not unwarranted, though, as it is likely behind the raw negative correlation between expenditures on ALMPs as a share of GDP and business employment rates (the labor utilization measure used in the next section) across countries (Figure 2).9 Similarly, passive labor market policies (PLMPs), comprising unemployment compensation payments and early retirement for labor market reasons, are negatively correlated to employment rates, and more so than ALMPs. This is likely because of the mechanical link between lower employment rates, larger unemployment rates and larger unemployment compensation outlays, although one should expect a negative effect on incentives to work from more generous compensation for jobless individuals.

Figure 2.
Figure 2.

Expenditure on Active and Passive LMPs, and Business Employment Rate

(Average from 1985 to 2000)

Citation: IMF Working Papers 2003, 234; 10.5089/9781451875645.001.A001

Source: OECD and author’s calculations.

The larger cyclicality of expenditures on PLMPs sheds suspicion on other measures used to evaluate the effect of ALMPs on the labor market, e.g., ALMP expenditure as a share of total labor market expenditure (i.e., expenditure on active and on passive labor market policies).10 As long as an increase in unemployment leads to a higher increase in passive labor market policy expenditure than in spending on ALMPs, which is probable because of the more mechanical link between unemployment compensation outlays and unemployment rate movements, the effect of ALMP expenditure in reducing unemployment is likely overstated.

Most previous work focuses on identifying the effect of ALMP expenditure on the unemployment rate but because an increase in ALMP spending might affect labor force participation, it is hard to deduce the final employment creation effect.11 In addition, the focus on unemployment rates creates a bias because of the exclusion of program participants from unemployment statistics.12 Finally, total unemployment is not the right target variable when subsidies to private employment are included among ALMPs. Indeed, one would like to count the net job creation from subsidies to the private sector as an effect of the ALMP, unlike a pure accounting effect that would exclude participants in training programs from unemployment.

In summary, existing studies using OECD data might overestimate the actual effect of ALMP on labor market outcomes because the way they define the policy variable (expenditures on ALMPs per unemployed, or some variation of this measure) or because they do not correct for the decrease in unemployment due to program participation. On the other hand, many of the reviewed studies might not identify an effect because of their limited use of the variation in the yearly data and the short sample period: from the mid 1980s to the early 1990s in most studies. Finally, a few of these studies focus on employment rates and none of them, to my knowledge, focuses on employment rates in the business sector.

IV. Empirical Identification of the Effect of ALMPs on Employment Rates

The problems with previous studies determine my empirical strategy. First of all, the dependent variable used here is the share of the working-age population employed in the business sector, i.e., the employment rate in the business sector. By focusing on the employment rate, variations in labor force participation due to the effect of ALMPs are accounted for. Also, the focus on business employment rates avoids overestimating the policy importance of ALMPs by automatically excluding cyclical increases in public sector employment, which do not represent an improvement in labor market functioning through real labor productivity increases or cost reductions. Finally, while the unemployment rate is the focus of many macroeconomic theories, the employment rate is a better measure of utilization of able-to-work individuals.

Second, to avoid a bias toward estimating a positive effect of ALMPs on employment, expenditures on ALMPs are normalized by GDP and not by unemployment. Such a normalization may bias the results downwards because aggregate output shocks will change employment in the same direction and provoke a spurious negative correlation between ALMPs/GDP and the employment rate. Furthermore, for given GDP levels countries with lower employment rates spend more in ALMPs (Figure 2). However, these effects may be attenuated by carefully controlling for institutions and other country-specific factors, and economic shocks. In any case, the final estimate may be viewed as a lower bound for the effect of ALMPs on employment.

ALMP expenditures are defined as the sum of expenditures (as a share of GDP) on: public employment services and administration, labor market training, youth measures, subsidized employment and measures for the disabled. Alternative specifications excluding the measures for the disabled, as well as including each of the policy measures separately, are also used. These categories are described in detail in Martin (2000) or Martin and Grubb (2001). Complete data were available for 15 industrial countries between 1985 and 2000.13

The estimated equation should be interpreted as a reduced form of a model determining employment rates and wages. As discussed in section II, many of the expected effects of ALMPs on employment will occur through variations in wages, which are also a function of ALMPs. So, wages are excluded from the employment rate specification and the estimated effect of ALMPs on employment rates should already incorporate shifts in wage-setting. The benchmark equation is:

BEit=β1 ALMPit+β2 Xit+β3 Yt+β4 Ci+εit(1)

Where the indices i and t designate, respectively, year and country. BE is business employment as a share of the working-age population, ALMP is spending on active labor market policy (as a share of GDP), X a vector of control variables capturing changes in institutions and the business cycles, Y a vector of year dummies to control for common shocks, C a vector of country dummies, and ε the error term.

Time and country dummies are very important components of the specification. The time dummies may alleviate the reverse causality problem if the timing of adverse shocks is correlated between countries. Country fixed effects capture all time-invariant institutional and economic features explaining why one country has a different-than-average employment rate. Several studies focused on the effect of specific institutional differences (employment protection laws, extent of coordination in wage bargaining, union membership, and so on) on labor market performance, and often had to exclude country dummies when these institutional measures had no time variation. Obviously, other studies could not include country dummies when performing a simple cross-section regression. The importance of these country-specific effects cannot be minimized. For example, since the mid-1980s, Luxembourg spent on average a lower percentage of GDP on ALMPs than Belgium (1.2 percent for Belgium and 0.2 percent for Luxembourg), yet Luxembourg had a higher business sector employment rate in the sample period (66 percent compared to 45 percent for Belgium). If only variables capturing institutional effects (which, in general, are not very precise) were used to control for country-specific effects, part of the difference in employment driven by other institutional factors would be wrongly attributed to ALMP spending.

Control variables include economic and country-specific institutional variables with time variation (for a detailed description, see Appendix II). The logarithm of per capita GDP in the business sector (in 1995 prices) is used to capture the level of economic activity in a country. Technological growth and the extent of economic openness may affect the level of employment and are also included. The share of GDP spent on PLMPs is an important control variable because of its positive raw correlation with expenditures on ALMPs (Figure 2). The inclusion of expenditures on passive policies will also capture some of the same cyclical factors affecting expenditures on ALMPs and attenuate the reverse causality bias. The institutional variables included in (1) and described in the appendix will capture institutional changes during the sample period. Other control variables commonly used in the literature (such as real long-term interest rates, the ratio of minimum to median wages, and other institutional variables) were not included in the final specification because their effect on employment rates was not significantly different from zero.

As a final note to the identification strategy, the conditional correlation between employment and ALMP expenditure as estimated in (1) could be due to a third variable not included in the regression, which would drive the levels of both ALMP expenditures and employment. Calmfors and Skedinger (1995) propose instruments for ALMP spending, but it is very unlikely that they are affecting unemployment only through ALMP. Lagged values of expenditures on ALMPs were used as instrumental variables for current expenditures on ALMPs in some of the specifications and do not change the results. Other specifications using lagged expenditures on ALMPs as regressors (instead of instruments for current expenditures) to check for dynamic effects were also used.

Estimates of the effect of ALMPs on the employment rate

ALMPs were very effective in increasing business employment rates in the 1990s but not before then. While there is no significant correlation between ALMP expenditures and business employment rates during the whole sample period (Table 1, column (1): coefficient of -0.13 with a t-statistic of -0.21), very different results are found when the sample period is split in two. The coefficient on ALMP spending is -0.12 and not significantly different from zero for 1985-1992 (column (2)), but it is 1.88 (and highly significant) for 1993-2000 (column (3)). Thus, for the 1993-2000 sub-sample, a 1 percentage point increase in ALMP spending (as a share of GDP) is associated with an increase in the business employment rate of 1.9 percentage points (for the 1993–2000 sub-sample).14

Table 1.

Active Labor Market Policy and Employment

Dep. Var.: Share of the Working Age Population Working in the Business Sector

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

Various specifications of the model confirm this pattern (Tables 1 to 4). The introduction of a lag in the measure of expenditures in ALMPs to check for dynamic effects shows a cumulative positive effect during 1993-2000 and insignificant effects in the earlier period (Table 1, columns (4) and (5)). Modifications in the cutoff point dividing the two periods (Table 2, columns (2) and (3)) and only using lagged expenditures on ALMPs or on PLMPs as the relevant policy variables (Table 2, columns (4) and (5)) do not change the results. Table 3 shows the importance of country dummies to the positive coefficient estimate for ALMPs (column (3)) and that excluding some particular groups of countries does not alter the basic results (columns (5) to (7)). Interestingly, the exclusion of Nordic countries augments the estimated coefficient of ALMP expenditures to 2.6, a result consistent with Elmeskov, Martin and Scarpetta (1998) and Scarpetta (1996).15

Table 2.

Robustness Check: Changes in Period Cutoff and Definitions of LMPs

Dep. Var.: Share of the Working Age Population Working in the Business Sector

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

Table 3.

Robustness Check: Effects of Ommiting Variables and/or Countries

Dep. Var.: Share of the Working Age Population Working in the Business Sector

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

Excludes Sweden, Norway and Denmark.

Table 4.

Robustness Check: Detailed Breakdown of ALMPs

Dep. Var.: Share of the Working Age Population Working in the Business Sector

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

PES: Public employment services and administration.

Training for unemployed adults and those at risk, and training for employed adults.

Measures for unemployed and disadvantaged youth, and support of aprenticeship and related forms of general youth training.

Subsidies to regular employment in the private sector, support of unemployed persons starting enterprises, and direct job creation (public or non-public).

Vocational rehabilitation and work for the disabled.

A detailed breakdown of the five policies within the ALMP aggregate provides some evidence that direct subsidies to employment creation and measures for the disabled were the main driving force behind the positive effect of ALMPs on the employment rate in the 1990s (Table 4). Public employment services and administration (PES), and youth measures were associated with lower employment rates while expenditures in training programs for unemployed and employed adults seemed irrelevant. Direct subsidies to job creation were even significant within a dynamic specification including lagged variables for the earlier period (Table 4, column (2)). Excluding the same groups of countries defined in Table 3 when analyzing the breakdown of ALMPs does not change the results (not shown).16

The control variables were in general estimated to have the expected signs. PLMPs have a consistent negative effect on employment rates in all periods but, as discussed above, part of this effect represents the expected reverse causality. When excluding it, the coefficient of ALMP expenditures is reduced from 1.88 in the benchmark specification to 1.36 (Table 3, column (1)), showing that including PLMP expenditures likely helps to attenuate the negative reverse causality bias on the coefficient of ALMPs.17 Technological growth affects employment rates negatively as it tends to save labor during the production process. Business GDP per working-age population, used to capture the level of economic activity, comes with a consistent positive sign. The coefficients for the extent of economic openness, unemployment benefits replacement rates, union membership, the tax wedge on labor income, and central bank independence tend to be negative.18 The share of public employment, the extent of wage bargaining coordination and the employment protection index change signs depending on the period under study. The first two variables have a negative effect on employment rates in the earlier period but a positive one during the 1990s. The employment protection index follows the same pattern but Table 2 shows that its effect may be sensitive to the cutoff point defining the latter period (columns (2) and (3)). Additional results not shown here include: 1) a positive effect of ALMPs on labor force participation; 2) a larger effect of ALMPs on total employment rates (to be expected since increases in ALMP expenditures tend to increase employment in the public sector mechanically); 3) indices of employment protection for both regular and temporary employment were used and generated the same results; 4) the coefficient of ALMP increases a bit when government current receipts, as a share of GDP, are included in the regression; and 5) surprisingly, the ratio of minimum to median wages does not affect equilibrium employment rates in most of the specifications tried.

V. ALMPs and Wage-Setting Behavior

To assess the effect of ALMPs on wage-setting, the following “wage curve” was estimated:

log(BWitPitAit)=α1log(uit)+α2 ALMPit+α3 Xit+α4 Yt+α5 Ci+ηit(2)

where BWit = wage per person in the business sector, Pit = consumer price index, Ait = technology, uit = unemployment rate, ALMPit = expenditures on ALMP as a share of GDP, Xit = vector of country-specific characteristics, Yt = year dummies, Ci = country dummies, and ηit = residual. This wage curve may be obtained theoretically using the same wage bargaining models behind the discussion in Section II. In these models, variables affecting workers’ utility from being employed vis-à-vis the alternative of unemployment shift this curve and should be included in X. All institutional variables included in (1) were tested but only the ones found (ex post) to be significant were retained in the final specification.

ALMPs are associated with wage moderation throughout the sample (Table 5). However, estimates for the first half are not significantly different from zero (column (2)). Correction for possible simultaneity between wages, unemployment and ALMPs yields identical results (columns (4) and (5)). Remarkably, the estimated elasticity of wages to the unemployment rate is exactly -0.1, confirming the claim in Blanchflower and Oswald (1994)—substantiated by their estimates of different “wage curves” using national micro data—that this is a universal value.19 The main results are insensitive to the inclusion of lagged ALMPs to capture possible dynamic effects (Table 6, column (2)).

Table 5.

Estimates of the Wage-Setting Curve

Dep. Var.: Logarithm of economy-wide wages deflated by the CPI and technology level

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.   3) Instrumental variables estimation uses lagged log unemployment rate and lagged ALMP expenditures as a share to GDP.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

Table 6.

Robustness Check: Effects of Ommiting Countries and Detailed Breakdown of ALMPs

Dep. Var.: Logarithm of economy-wide wages deflated by the CPI and technology level

article image
Sources: OECD - Analytical Database, Expenditure in Labor Market Policies Database, and Benefits and Taxes Database; some institutional variables from Nickell and Nunziata (2001) and Debrun (2003); author’s estimations.Notes: 1) Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Netherlands, Norway, New Zealand, Spain, Sweden, UK and USA.   2) t-statistics in parentheses. Bold figures are significant at least at a 5 percent level.

Expenditures in labor market policies expressed as a share of GDP in the relevant fiscal year.

PES: Public employment services and administration.

Training for unemployed adults and those at risk, and training for employed adults.

Measures for unemployed and disadvantaged youth, and support of aprenticeship and related forms of general youth training.

Subsidies to regular employment in the private sector, support of unemployed persons starting enterprises, and direct job creation (public or non-profit).

Vocational rehabilitation and work for the disabled.

Excludes Sweden, Norway and Denmark.

Excludes Canada, USA and UK.

Within the components of ALMPs (column (3)), direct subsidies to employment creation and measures for the disabled (although the estimate for the latter policy has a higher standard error) have contributed to wage moderation, while larger expenditures on public employment services and administration (PES), and youth measures (the latter with a positive but statistically insignificant effect) had the opposite effect. These results show that PES and youth measures have a negative effect on employment likely because they lower the disutility of being unemployed, shifting the wage-setting curve upwards. At first sight, expenditures in training programs have kept wages low even though they have not affected employment rates. However, this result is sensitive to the exclusion of the Nordic countries from the sample (column (4)), while the sign of the remaining policies remained unchanged. The introduction of lagged policy variables (not shown) leaves these results unchanged.

The relevance of direct subsidies to job creation in shifting the wage-setting relationship downward suggests that part of the wage moderation might be due to a composition effect owing to a rise in the proportion of the employment of less-skilled workers. Following the schematic approach discussed in Section II, if effective, these subsidies should put upward pressure on wages since lower non-wage labor costs open up a margin for wage increases to attract more labor. However, many of the subsidies to direct job creation are targeted to low-paid workers, which would reduce average wages through a composition effect. The adjustment for technological growth in each country likely corrects only part of this bias.

The remaining control variables posted consistent effects throughout all the specifications studied although causality effects cannot be directly inferred from most of the estimates. A higher degree of economic openness may lower paid wages adjusted for technology and prices as international competition keeps wage demands in check. However, lower real labor costs also increase external competitiveness and, therefore, economic openness. A larger share of public employment wages are associated to higher business sector wages. That may be explained by a cost-induced displacement of workers from high-wage businesses to the public sector. Alternatively, countries with larger public employment may also have faced larger increases in public sector wages, which contaminated labor compensation in the business sector. More employment protection reduced wages, likely to keep labor attractive to employers. As expected, longer unemployment benefits boosted wages by lowering the costs of being unemployed.

VI. Final Remarks

ALMPs seem to have been effective, on average, in raising employment rates in the business sector of 15 industrial countries. Among such policies, direct subsidies to job creation seemed the most effective, which is consistent, at least, with recent studies using microeconomic data for Sweden and for France.20 Increases in ALMP spending were also correlated to less real wage growth after allowing for technological growth, changes in the unemployment rate and institutional and other economic factors. This wage moderation might have been one of the causes for the association of ALMPs with better employment rates. However, part of the causality going from larger ALMP expenditures as a share of GDP and wage moderation could be only apparent as some ALMPs are directed to low-skill low-paid workers, and the resulting composition effect toward lower average wages could be generating an spurious relationship.

Whether ALMPs are cost-effective from a budgetary perspective remains to be determined. Despite their overall positive impact on employment rates, their budgetary cost is high and they are likely to be subject to diminishing returns as employment rates rise. At the present level of employment rates, ALMPs could recoup their cost (or not) if they place benefit recipients into jobs and these benefits are phased out. However, this type of accounting does not consider, among other factors, the social benefits of lowering unemployment. In any case, given the negative effect of current institutional arrangements on European employment, institutional reforms seem to be a better bet for improving labor utilization without unduly high costs to the society as a whole. Among the many possible reforms, the estimated coefficients for the (admittedly rudimentary) institutional variables used in my econometric work suggest that reductions in tax wedges, in benefits replacement rates, in public sector employment, and, more generally, in insiders’ wage bargaining power, are a must.

Do Active Labor Market Policies Increase Employment?
Author: Mr. Marcello M. Estevão
  • View in gallery

    How Might ALMPs Affect Employment?

  • View in gallery

    Expenditure on Active and Passive LMPs, and Business Employment Rate

    (Average from 1985 to 2000)