3 Labor Market Dynamics and Economic Policy in France
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund


As in many other member countries of the Organization for Economic Cooperation and Development (OECD), the unemployment rate in France has risen during each cyclical downturn, but has not returned to its prerecession levels during the subsequent recoveries. As a result, unemployment has gradually increased over the last 30 years, from about 2 percent of the labor force in the 1960s to about 12 percent in 1993 and 1994 (Table 1 and Chart 1).1

As in many other member countries of the Organization for Economic Cooperation and Development (OECD), the unemployment rate in France has risen during each cyclical downturn, but has not returned to its prerecession levels during the subsequent recoveries. As a result, unemployment has gradually increased over the last 30 years, from about 2 percent of the labor force in the 1960s to about 12 percent in 1993 and 1994 (Table 1 and Chart 1).1

Table 1.

Unemployment Rates in Industrial Countries

(In percent of labor force)

article image
Sources: OECD, Analytical Database; and International Monetary Fund (1995).
Chart 1.
Chart 1.

France: Unemployment Rate

(In percent)

Source: IMF, World Economic Outlook.

The costs of high unemployment have been substantial. First, output has been lost, with real GDP several percentage points lower than its full-employment level. Second, the income support provided to more than 2 million persons (out of about 3 million unemployed) has placed additional strains on public finances.2 Third, the human and social costs of rising dependency on public assistance cannot be dismissed; these include the depreciation of human capital, the loss of social standing, and the degradation of family structures.

This paper establishes a relationship between economic policies and unemployment in France by constructing a macroeconomic model of the labor market. The analysis of the aggregate data suggests that the rise of unemployment has been related mainly to the increasing size of the social security system, which is financed by heavy taxes on labor. More specifically, high levels of social contributions and benefits appear to have severely dampened the growth of employment.3 A second and equally important result is that the minimum wage has a strong influence on aggregate wages, employment, labor force participation, and unemployment.

Labor Market Developments and Economic Policies: An Overview

A salient feature of the French labor market in the last two decades has been the slow growth of employment relative to population. As shown in Table 2, the working-age population of France has increased by about 20 percent since 1970, while employment—especially in the private sector—has grown very little by comparison. Put differently, labor force participation fell while unemployment and nonemployment increased sharply.4 These developments took place against a backdrop of substantial economic growth. With employment essentially stagnant, the increase in real GDP can be accounted for almost entirely by higher labor productivity, which in turn reflects some combination of capital deepening and technological change (Chart 2).

Table 2.

Demographic and Labor Market Flows

(In millions of persons, unless otherwise noted)

article image

Individuals aged 16 to 64.

Unemployed persons in percent of labor force.

Persons of working age not employed, in percent of working-age population.

Labor force in percent of working-age population.

Chart 2.
Chart 2.

France: Labor Market Indicators and Economic Performance

Sources: Institut national de la statistique et des études économiques database; Organization for Economic Cooperation and Development, analytical database; and IMF staff calculations.1 Labor force in percent of working-age population.2 In percent of working-age population.3 Real GDP divided by employment.

Wages also increased substantially over this period, although their growth rate at times diverged markedly from that of labor productivity (Chart 3). For example, from 1970 to 1983, the ratio of real producer wages to labor productivity (which is equivalent to the share of labor income in GDP) rose markedly. In contrast, during the second half of the 1980s, wages increased much more slowly than productivity. It is not a trivial task to explain these changes in the share of labor. However, they appear to have been strongly correlated with the real price of oil—perhaps because higher energy prices sharply depressed the rate of return on the existing capital stock or because workers succeeded, in the wage bargaining process, in being compensated for the increase in real fuel prices.

Chart 3.
Chart 3.

France: Wage Developments

Sources: Institut national de la statistique et des études économiques database; Organization for Economic Cooperation and Development, analytical database; and IMF staff calculations.1 Gross income from dependent employment divided by GDP deflator.2 Income from dependent employment net of taxes, deflated by consumer price index.3 Equivalent to share of labor in GDP.4 In Francs per barrel, divided by GDP deflator.

The period since 1970 has also been characterized by substantial changes in economic policy, many of which fit into a larger pattern of expanding government social programs (Chart 4). This has been reflected in a remarkable rise in the ratio of general government current expenditure and revenue in relation to GDP. Most of this increase is accounted for by higher social spending—on health care, pensions for regular and early retirement, unemployment compensation, and welfare—and has been financed largely by continuing increases in social security taxes levied on labor income. It is of course legitimate to ask whether higher social spending and higher social security taxes were principally a cause or an effect of higher unemployment—a question that is further discussed in the next section.

Chart 4.
Chart 4.

France: Indicators of Public Sector Activity

Sources: Institut national de la statistique et des études économiques database; Organization for Economic Cooperation and Development, analytical database; French authorities; and IMF staff calculations.1 Ratio of social security taxes to gross income from dependent employment.2 Social security benefits per capita in percent of average net wage.3 Social security benefits per capita in percent of minimum wage.

Another way of looking at the expansion of social expenditure is to calculate the ratio of social benefits per capita to the average wage net of social security contributions (Chart 4).5 It appears that, on average, a French resident (of any age) in 1993 received social support payments equivalent to almost 30 percent of the average net wage and to almost 90 percent of the minimum wage. This represents a massive increase compared with 1970, when these ratios were about half as high as they are now. A possible implication could be that unemployment, or nonwork, has become materially more tolerable than it was a generation ago and may even be an attractive alternative to certain low-paying jobs. The data support this interpretation.

An important example of the trend toward more remunerative government social programs, and one that is thought to affect the labor market directly, is the average replacement rate of the unemployment insurance system.6 This rate became substantially more generous from 1970 until about 1982 (Chart 5).7 A sharp turnaround took place under the government of socialist Prime Minister Laurent Fabius in 1983. However, even under the conservative governments that held office from 1986 to 1988 and after 1993, benefits were never scaled back to the levels that prevailed until the early 1970s.

Chart 5.
Chart 5.

France: Indicators of Labor Market Policy

Sources: French authorities; and IMF staff calculations.1 Unemployment benefits per unemployed person in percent of average wages.2 Minimum wage in percent of average gross wage.

The legal minimum wage has also risen substantially in real terms.8 However, it has remained in a narrow band relative to average gross producer wages, ranging from a high of 45 percent in 1970 to a low of 40½ percent in 1980 (Chart 5).

Single-Equation Models of Unemployment and Nonemployment

One way to investigate general economic influences on the unemployment and nonemployment rates is to seek statistically significant correlations between these rates and other economic variables.9 Such an exercise—based on an ad hoc specification search—is preliminary to the more systematic structural investigation to be described in the next section.10

The statistical determinants of the unemployment rate and of the non-employment rate are analyzed using a general autoregressive distributed lag (ARDL) model that includes a large number of explanatory variables, including real GDP, real interest rates, the employment rate, the real price of oil, social security benefits, the replacement ratio of the unemployment insurance system, and minimum wages.11

The principal result is that, for both the unemployment rate and the nonemployment rate, the only policy variable that matters is the ratio of social security expenditure to gross income from dependent employment (the “social benefit ratio”), and this variable has an extremely strong and robust influence.12 Granger causality tests suggest that the direction of causation runs both from social security expenditure to the dependent variable and the other way around. An equally strong result is obtained if the social expenditure ratio is adjusted for the effects of the business cycle.

A Structural Model of the Labor Market

Modeling Aggregate Labor Market Variables

The single-equation models of unemployment (and nonemployment) presented in the previous section offer an incomplete account of labor market behavior. In particular, the model of the unemployment rate (relative to the labor force) contained the employment rate (relative to working-age population) as an explanatory variable, while the model of the nonemployment rate included wages. Both of these variables should be viewed as endogenous to the labor market.

This underscores the need for a structural, multiequation model that simultaneously explains labor force participation (labor supply), employment (labor demand), and the wage bargaining process. In deriving and estimating the model, particular attention is given to identifying the dynamics inherent in the labor market, not just in each equation individually, but also in the interactions between the equations. The hope is that modeling these interdependencies will deepen our understanding of the sources of unemployment persistence and imperfect responsiveness. The estimated simultaneous model is later used to examine how the labor market responds to exogenous shocks.

The theoretical basis for the model is the widely used “competing claims” approach, which treats the determination of wages as the outcome of a bargain between unions and firms, where the latter operate in imperfectly competitive markets. Although there is a wide variety of different versions of the union-firm bargaining model, there appears to be a broad similarity in the implications they have for wage and employment behavior (see Bean (1994) on this point). Thus, in these models, the real wage equation is generally of the form


where unemployment (U) enters the equation as a determinant of the unions’ fallback wage and captures the effect on the real wage of excess supply in the labor market. The term θ(.) is a polynomial in the lag operator, which in turn allows for inertia in the real wage. The most important element in the equation is a set of variables (Z) that typically includes variables from objective functions of unions and firms (and thus depends implicitly on the relative strength of unions in the bargain).

Even if there is a wide consensus in the research that leads to specifications of the general form of equation (1), there is debate about the variables to include in the set Z. For example, there has been much discussion about the statistical importance of the replacement ratio, including how this variable should be measured (see Layard, Nickell, and Jackman (1994)). Earlier, it was noted that the replacement ratio in France had increased, so it might be expected to have an important effect here. It turns out that there is little evidence that the replacement ratio has a direct effect on wage setting. Rather, the data strongly support the view that an indicator of the fiscal weight of the social security system has a powerful effect on employment.

Minimum wage provisions are another potentially important influence upon aggregate wages. There has been considerable controversy about whether minimum wages have deleterious effects on wages and employment. For example, in efficiency wage models it is possible that a minimum wage could help, not hinder, employment. More conventional labor market theory implies that minimum wage provisions, by preventing perfect downward flexibility of wages, raise aggregate wages and hence lower employment. In the wage equation above, this is tested by including a measure of minimum wages in the set of determinants of the real wage. As will be seen below, this variable was highly significant.

Employment is determined within the same imperfectly competitive framework, to which is added the assumption that lagged employment can have an important effect on the current level of employment. This dynamic effect is due to the presence of adjustment costs: the number of persons employed is seen as difficult and costly to change in the short run. In contrast, labor input is somewhat more flexible. For example, the utilization of employees can be changed through variations in the work week or through increasing the intensity of effort.

Finally, the model is closed with a labor force participation equation. This allows for the simultaneous determination of real wages, employment, and unemployment. The labor force equation allows for variations in participation rates in response to changes in demographic factors and wages. Overall, then, the system of equations takes the following form:


where yt is a vector of the three endogenous variables, namely, employment (LE), the labor force participation rate (LFR), and real producer wages (RWAGR). A(L) and B(L) are matrix lag polynomials, and zt is a vector of exogenous variables.13 The reduced form of this model is simply a vector autoregression (VAR) that also includes contemporaneous and lagged exogenous variables.

The model as a whole may be represented most concisely by the static long-run solution to the three-equation model, written in the following stylized form:


where LFR is the labor force participation rate, LERPW is the ratio of employment to working-age population, WNET is the real wage net of social security contributions, LE is employment, WGR is the real gross wage, Y is real GDP, G is the ratio of social security expenditure to gross labor income, LPR is labor productivity, LUR is the unemployment rate, WMIN is the real minimum wage, and POILR is the real oil price. All variables, except the ratio G, are given in terms of natural logarithms.

Thus, labor force participation depends on the prospects of obtaining employment and on net wages. Employment depends on gross real producer wages, real GDP, and the social benefit ratio. Real producer wages depend on labor productivity, the unemployment rate, the minimum wage, and the price of oil.

On the further simplifying assumption that social security taxes are equal to social security benefits (and expressing the ratio appropriately), one obtains the following identities:


Substituting these identities in the system of equations set out above, one obtains a system with a rich structure of simultaneous relationships among the endogenous variables. Thus, wages and employment affect the labor force in the first equation; wages affect employment in the second equation; and employment and the labor force affect wages in the third equation.

Estimation Results

The issues that arise when estimating a model of the kind presented here are discussed at some length in Karanassou and Snower (1993) and Chapter 1 of this volume. Briefly, the main problems are those of dealing with non-stationary variables and of ensuring that the model, which is a simultaneous one, is estimated consistently. To allow for the presence of nonstationary variables, tests of cointegration among these nonstationary variables have been conducted.14 The estimated models are dynamic structural equations, and each equation is estimated as an “error correction” model (ECM), where the cointegrating relationship between the levels of the variables appears as the long-run equilibrium part of the equation. To allow for the presence of contemporaneous values of other endogenous variables on the right-hand side of an equation (for example, where the current real wage is a determinant of employment), this equation needs to be estimated by a method that allows for this to avoid bias in the parameter estimates. In what follows, this issue is addressed through the use of instrumental variables (IV) estimation. Where appropriate, tests of these and other characteristics of the estimated equations are shown in the tables of empirical results.

All the variables used in this section were first tested for their order of integration.15 All the relevant variables were found to be nonstationary, although in certain cases, this result was not clear cut (Table 3). Subsequently, tests for the presence of cointegration among the levels of the variables that compare the long-run parts of the participation, employment, and wage equation, respectively, showed that each had one cointegrating vector. The subsequent estimation of the dynamic model equations then builds on this finding. The results for each equation are discussed in somewhat greater detail next.

Table 3.

Tests of Orders of Integration

article image
Note: The 95 percent critical value is 3.45 for the Dickcy-Fuller (DF) statistic, and 3.46 for the Augmented Dickcy-Fuller (ADF) statistic.

Labor Force Equation

It is reasonable to hypothesize that the labor force participation rate depends on the probability of obtaining employment and on the prevailing net wage rate. Policy variables may also affect the decision of individuals to join or leave the labor force. For example, minimum wages may induce some individuals to offer their services who might otherwise not find it worthwhile to seek employment. Similarly, the replacement ratio of the unemployment insurance system might increase participation by offering an incentive to stay on the unemployment rolls instead of dropping out of the labor force.

Systematic testing and reduction yielded the model shown in Table 4. In it, the labor force participation rate is positively correlated with the employment rate (or negatively correlated with the unemployment or nonemployment rate). In addition, the findings suggest that higher net wages encourage labor force participation. These variables were found to form a cointegrating set (the Johansen likelihood ratio test was 31.9 compared with a 95 percent critical value of 29.7, confirming the presence of at least one cointegrating vector). However, none of the policy variables considered in this study (social security benefits, social security taxes, direct taxes, real interest rates, minimum wages, or the replacement ratio of the unemployment insurance system) was found to have a significant effect on the labor force participation rate when they were added to the basic economic variables already considered. Only the replacement ratio of the unemployment insurance system comes close to being significant, but it entered the equation with a negative sign—a counterintuitive result—and so was discarded as a probable influence upon participation.

Table 4.

Labor Force Equation

(Dependent variable LFR; sample period 1970Q3–1991Q4)

article image
Note: In all the reported results, figures in parentheses are t-statistics.Other summary statistics quoted are defined as follows: χ12(.) is a Lagrange multiplier test of residual correlation; χ22(.) is the Ramsey RESET test; χ32(.) is a test for normality of residuals; and χ42(.) is a test for hereroscedasticity. χ52(.) is Sargan’s test that the instruments are valid. Instruments used were lagged values of the right-hand-side variables.

Employment Equation

As expected, employment is affected negatively by real producer wages and positively by a measure of output. In addition, there is a somewhat unusual result, in that an indicator of the fiscal weight of the social security system has a strong negative effect on employment. This effect might be expected to be intermediated through the real wage, with the social security variable tending to increase real gross wages, which in turn would reduce employment. This issue is discussed more fully below.

Table 5 shows the parameter estimates. Tests confirmed the presence of a single cointegrating vector in this set of variables (LR = 38.0 compared with 27.1 at the 95 percent level). As with the other equations presented in this paper, this specification was obtained by systematically testing and reducing a general model. A full set of policy variables was included initially, notably the real interest rate, oil prices, a measure of international competitiveness, the replacement ratio of the unemployment insurance system, and the minimum wage. None of these—with the exception of the social security burden—was found to have a significant influence.

Table 5.

Employment Equation

(Dependent variable LE; sample period 1970Q3-1991Q4)

article image
Note: For definitions, see Table 4. Instruments used were lagged values of GDP, the labor force, and wages.

In particular, higher real interest rates appear to have no measurable direct, adverse effect on employment. This result runs counter to the hypothesis that an anti-inflationary monetary policy is associated with higher unemployment, at least in the short run. However, real interest rates may be shown to affect investment and output, and so affect employment indirectly.

From a policy standpoint, the key conclusion would appear to be that labor demand is sharply dampened by an increase in the overall burden imposed by the social security system. This burden is measured by social security expenditure, which by definition is the sum of social security taxes and the financial balance of the social security system. This result raises several difficult questions. The first two are statistical, while the third concerns the economic interpretation of the result.

First, it is somewhat surprising that the effect of the social security system on employment is not fully captured by producer wages, which already include social security taxes. To investigate this point further, wages net of social security taxes were substituted for gross wages in the employment equation. The result was that the adverse effect of the social security variable on employment became even larger, while the size of the coefficient on wages dropped sharply. The overall size of the effect of the social security system on employment was thus unchanged. There is, according to this result, evidence of a larger (negative) effect on employment of social security payments than is captured by their effect on the gross real product wage alone.

Second, it might be hypothesized that causality runs in the opposite direction, from employment to social security spending. Indeed, when employment declines, as in a cyclical downturn, social security spending tends to increase. 1 However, much of the effect of the cycle on employment is already accounted for by the inclusion of GDP in the equation. Furthermore, the adverse and separate effect of social security spending on employment is still found if the social expenditure variable is adjusted for cyclical deviations of actual GDP from potential GDP.

If it is granted that the effect of social expenditure on employment is statistically robust, what is the channel through which it exercises its influence? In particular, economic theory provides little support for the notion that social security benefits have a direct effect on the hiring decisions of employers. One possible explanation is that the employment equation is not a labor demand equation: this follows from the definition that any equation for employment is also an equation for nonemployment.16 In the preceding section, the nonemployment rate was shown to depend strongly and positively on social expenditure. The economic rationale for this effect appears straightforward: higher benefits encourage workers to remain non-employed.17 This effect, which seems empirically well founded and rests on a plausible behavioral theory, will therefore also be reflected in the equation for employment.

Wage Equation

In the long run, wage developments in France appear to be accounted for mainly by increases in labor productivity, unemployment, and the minimum wage. Again, it was found that a cointegrating vector exists in this set (LR = 30.3 compared with a 95 percent critical value of 29.7). The results of estimating an ECM with this vector as its equilibrium term are shown in Table 6.

Table 6.

Wage Equation

(Dependent variables WGR, LPR; sample period 1970Q3-1991Q4)

article image
Note: For definitions, see Table 4. Instruments used were lagged values of the labor force, the real oil price, unemployment, and the minimum wage.

The minimum wage appears to have a strong influence on real wages as a whole, tending to raise them above what can be accounted for by productivity and unemployment. This result stands in contrast to some earlier studies of the French labor market.18 However, the work of other authors supports the conclusion reached in the present paper. Bazen and Martin (1991) show that a higher minimum wage raises the cost of youth labor. Moghadam (1995) finds a direct effect of the minimum wage on the real wage. He notes that the proportion of individuals earning the minimum wage is particularly high for those who are less than 26 years old (35.5 percent of wage earners in 1992), which strongly suggests that the overall effect is transmitted through younger workers.

Finally, the unemployment rate is an indicator of outsider pressure in wage negotiations; higher unemployment tends to act as a constraint on the wage demands of insiders.19

Dynamic Analysis

This section presents an analysis of the aggregate dynamics of the labor market. A system consisting of the behavioral equations and the identities described in the previous section was used to conduct a variety of simulations.

A first set of simulations examined the effect on unemployment of both a permanent and a temporary shock to (increase in) labor demand. These shocks took the form of a change in the constant term of the employment equation equal to 1 percent of the equilibrium level of employment. A permanent shock of this type permanently lowers the unemployment rate by about 1¾ percentage points (see Chart 6). As may be seen, the adjustment of employment and the labor force—and hence of unemployment—to a new equilibrium is fairly rapid. The response of wages is considerably slower. The reaction of the system to a temporary shock is somewhat different: the labor force initially responds more strongly than does employment, reflecting the strong effect of higher employment on wages. Thus, the unemployment rate rises before settling back to its equilibrium value (Chart 7).

Chart 6.
Chart 6.

France: Response of Labor Market to Permanent Labor Demand Shock

Source: IMF staff calculations.Note: Shock to labor demand equivalent to 1 percent of employment.1 Deviation from baseline; in percentage points.2 Employment in percent of working-age population.3 Labor force in percent of working-age population.4 Real producer wages; percentage deviation from baseline.
Chart 7.
Chart 7.

France: Response of Labor Market to Temporary Labor Demand Shock

Source: IMF staff calculations.Note: Shock to labor demand equivalent to 1 percent of employment.1 Deviation from baseline; in percentage points.2 Labor force in percent of working-age population.3 Employment in percent of working-age population.4 Real producer wages; percentage deviation from baseline.

The dynamic system was also used to simulate the effect on unemployment of a permanent change in the level of social protection. Chart 8 shows the impact on the unemployment rate of a 1 percentage point reduction in the social security tax and benefit ratios (these ratios are computed relative to gross income from dependent employment). There is some overshooting of the unemployment rate early on, reflecting a stronger short-term effect on labor force participation. In the longer run, the effect on employment predominates, and the unemployment rate declines by about 0.2 percentage point. The incidence of the reduction in social security taxes is mainly on workers: producer wages increase by only ½ of 1 percent in the long run, largely reflecting the lower equilibrium unemployment rate, while net consumption wages rise by more than 1½ percent.

Chart 8.
Chart 8.

France: Response of Labor Market to Permanent Alleviation of Social Security Taxes and Expenditure

Source: IMF staff calculations.Note: Reduction in revenue and expenditure by 1 percent of gross income from dependent employment.1 Deviation from baseline; in percentage points.2 Employment in percent of working-age population.3 Labor force in percent of working-age population.4 Percentage deviation from baseline.

The last simulation examined the effect of a permanent 10 percent reduction in the real minimum wage (Chart 9). Again, there is some overshooting of the unemployment rate. In the long run, however, employment rises substantially and the unemployment rate drops about ½ of 1 percentage point. As would be predicted by theory, a reduction in the minimum wage leads to a somewhat smaller decline in average wages (the elasticity appears to be in the neighborhood of ½).

Chart 9.
Chart 9.

France: Response of Labor Market to Permanent Reduction of the Minimum Wage

Source: IMF staff calculations.Note: Reduction by 10 percent in real terms, relative to the baseline.1 Deviation from baseline; in percentage points.2 Employment in percent of working-age population.3 Labor force in percent of working-age population.4 Percentage deviation from baseline.

Concluding Remarks

The dynamic analysis is revealing, suggesting that dynamic responses to external shocks in France are relatively quick. Unemployment gets close to equilibrium within two years, a much shorter time than it takes other European countries to recover from a similar shock.

This is not a standard finding. Barrell, Morgan, and Pain (1995), for example, report that, if anything, France shows more inertia than do the other major European countries. There are, however, reasons for believing the present results are more likely. The model in this paper reveals powerful effects on the level of employment from social security payments and on wages from the minimum wage provisions. These are characteristics of the French labor market that are not found elsewhere in Europe. Hence, the role for a persistent movement in unemployment away from its equilibrium (NAIRU) rate is that much less in France, and, concomitantly, the structural rate of unemployment appears to have risen more in France than elsewhere.

For policy, these findings suggest that the share of the public sector in the economy, which has grown rapidly in the last quarter century and is now one of the highest among the industrial countries, may need to be reduced. Cutting the tax and social security burden should stimulate labor demand. Lower taxes, along with less generous benefits, can also be expected to give those who are not employed a greater incentive to seek work. The results presented in this study also provide support for a substantial reduction in the real minimum wage. Over time, it would be possible to lower the real minimum wage simply by avoiding any further increase in nominal terms.

Of course, policy recommendations based on macroeconomic results must be supplemented by a more disaggregated, microeconomic analysis. For example, because unemployment in France is concentrated among low-skilled individuals, measures targeted at this group could be relatively more effective. Thus, an across-the-board reduction in the level of social protection may be neither necessary nor sufficient to reduce unemployment.20 What presumably matters more is the high minimum level of social support, which is one of the principal determinants of the reservation wage. The level and comprehensiveness of welfare benefits are a key social choice, entailing a trade-off between the generosity of benefits, on the one hand, and employment objectives, on the other. The evidence in this paper suggests that, if unemployment is to be substantially reduced, the generous system of social protection must be trimmed. To be sure, more effective education and training as well as labor and product market liberalization are also called for as part of a comprehensive attack on the unemployment problem.

Appendix: Single-Equation Models of the Unemployment and Nonemployment Rates

Unemployment Rate

Formally, the model of the unemployment rate may be written as


where LUR is the unemployment rate, the Zj are N explanatory variables, and a(L) and bj(L) are lag polynomials. The static long-run solution of the model is simply


This general model was reduced by eliminating variables for which the parameter estimates showed the wrong sign or were statistically insignificant, or both. At each stage, the model reduction was tested using F-tests. While the minimum wage (relative to the average wage) was found to raise unemployment, the effect was small and statistically insignificant (p = 0.98), and this variable was rejected in the first stage of model reduction. The real price of oil had a negative effect whenever it was included. However, it was also insignificant (p = 0.40) and was rejected in the second stage of model reduction. Other variables that were excluded in subsequent stages generally showed the correct sign but were insignificant statistically and made only small contributions to the overall fit of the equation. Notably, higher real producer wages were associated with a higher unemployment rate, while higher real GDP was associated with a lower unemployment rate.21

In the final analysis, only two variables were retained in addition to lagged values of the unemployment rate: the rate of employment in the working-age population (LERPW), which measures the general prospects of obtaining work, and the social benefit ratio (RNYSS), which is a measure of the overall level of social support payments provided by the public sector. This model reduction was strongly accepted relative to the most general model (p = 0.53). It was also accepted relative to all rival models examined in the course of the reduction (p = 0.13 for the nearest rival). In addition to real wages, the nearest rival model also included real interest rates, which, however, again showed the wrong sign.

The parameter estimates for the final model are shown in Table 7. In the long run, a 1 percentage point increase in the employment rate (relative to working-age population) results in a 0.16 percentage point drop in the unemployment rate (the labor force is about two-thirds of the working-age population), while a 1 percentage point increase in the social benefit ratio (relative to wage income) leads to an increase in the unemployment rate of 0.27 percentage point.22

Table 7.

Model of Unemployment Rate

(Sample period 1970Q1-1993Q4)

article image
Source: IMF staff calculations.Note: The variables are defined as LUR, unemployment rate; LERPW, employment ratio; and RNYSS, social benefit ratio.

The basic model was further tested by adding other variables one by one. The results are presented in Table 8. None of the variables is significant, even at the 10 percent level. Most show the correct sign, with the exception of real GDP, the real interest rate, and the replacement ratio. Furthermore, both explanatory variables included in the previous model remain significant in all but one case. In summary, the basic model appears to be highly robust.

Table 8.

Effect of Adding Variables to Basic Model of Unemployment Rate

article image
Source: IMF staff calculations.

Change in unemployment rate (in percentage points) in response to 1 percent (percentage point) change in dependent variable.

F-test for joint significance of current value and two lags.

Similar results were obtained when the overall social support ratio or the ratio of general government current expenditure to GDP (both shown in Chart 4) was substituted for the social benefit ratio in the basic model. The ratio of social security taxes to gross income from dependent employment also had a strong positive effect on the unemployment rate. This is not surprising because all of the substitute variables are highly collinear.

Nonemployment Rate

If it were true that more generous social expenditure has added to unemployment by making it more tolerable or indeed attractive in some cases, the rise of unemployment would in part reflect an increase in voluntary unemployment.23 In addition, one should be able to observe a decline in labor force participation. Thus, the effect of more generous social expenditure on the labor market might be better captured by the non-employment rate, which is defined as the percentage of working-age individuals who are not employed.

A single-equation model of the nonemployment rate is presented in Table 9.24 As with the unemployment rate, there is a strong positive effect of social benefits, both in the long run (the model in levels) and in the short run (the same model in first differences). Real consumption wages have a negative effect, both in the short run and in the long run, which means that people are drawn into the labor market by the prospect of higher net wages. Although stronger economic growth helps to reduce nonemployment in the short run, in the longer term, this demand effect appears to be outweighed by the tendency of nonemployment to rise in tandem with higher levels of real income. Again, the positive association, both in the short run and in the long run, between social benefits (however measured) and the nonemployment rate is highly robust to changes in the detailed specification of the model.

Table 9.

Model of Nonemployment Rate

(Sample period 1970Q3–1993Q4)

article image
Source: IMF staff calculations.Note: The variables are defined as LNER, nonemployment rate (in percent of working-age population); WNET, real net consumption wage; RNYSS, social benefit ratio; Y, real GDP; and POILR, real price of oil. One asterisk denotes significance at the 5 percent level; two asterisks denote significance at the 1 percent level.


  • Barrell, R., J. Morgan, and N. Pain, 1995, “Employment, Inequality, and Flexibility: A Comparative Study of Labour Markets in North America and Europe” (unpublished; London: National Institute of Economic and Social Research).

    • Search Google Scholar
    • Export Citation
  • Bazen, Stephen, and John P. Martin, 1991, “The Impact of the Minimum Wage on Earnings and Employment in France,” OECD Economic Studies, No. 16 (Spring), pp. 199221.

    • Search Google Scholar
    • Export Citation
  • Bean, Charles R., 1994, “European Unemployment: A Survey,” Journal of Economic Literature, Vol. 32 (June), pp. 573619.

  • Bolot-Gittler, Anne, “Le système d’indemnisation du chômage: évolution de ses caractéristiques entre 1979 et 1991,” working document of Ministry of Labor, Paris, France.

    • Search Google Scholar
  • Cornilleau, Gérard, Pierre Marioni, and Brigitte Roguet, 1990, “Quinze ans de politique de I’emploi,” Observations et diagnostics économiques, No. 31 (April), pp. 91120.

    • Search Google Scholar
    • Export Citation
  • Elmeskov, Jørgen, 1993, “High and Persistent Unemployment,” Economics Department Working Paper No. 132 (Paris: Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 1995, World Economic Outlook, May 1995: A Survey by the Staff of the International Monetary Fund, World Economic and Financial Surveys (Washington).

    • Search Google Scholar
    • Export Citation
  • Karanassou, Marika, and Dennis J. Snower, 1993, “Explaining Disparities in Unemployment Dynamics,” CEPR Discussion Paper No. 858 (London: Centre for Economic Policy Research).

    • Search Google Scholar
    • Export Citation
  • Layard, Richard, and Stephen Nickell, 1991, “Unemployment in the OECD Countries,” Applied Economies Discussion Paper No. 130 (Oxford, England: University of Oxford, Institute of Economics and Statistics).

    • Search Google Scholar
    • Export Citation
  • Layard, Richard, and Stephen Nickell, and Richard Jackman, 1994, The Unemployment Crisis (Oxford, England; New York: Oxford University Press).

    • Search Google Scholar
    • Export Citation
  • Lindbeck, Assar, and Dennis J. Snower, 1988, The Insider-Outsider Theory of Employment and Unemployment (Cambridge, Massachusetts: MIT Press).

    • Search Google Scholar
    • Export Citation
  • Moghadam, Reza, 1995, “Why Is Unemployment in France So High?” in France: Financial and Real Sector Issues, ed. by Paul Masson (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Schmitt, John, and Jonathan Wadsworth, 1993, “Unemployment Benefit Levels and Search Activity,” Oxford Bulletin of Economics and Statistics, Vol. 55 (February), pp. 124.

    • Search Google Scholar
    • Export Citation

Note: The views expressed are strictly personal. The authors wish to thank Jacques Artus. Willem Baiter, Thierry Pujol, Ramana Ramaswamy, Dennis Snower. and Uli Baumgartner for their comments on earlier drafts.


These issues arc discussed in Bean (1994). who provides a useful survey of labor market developments in Europe.

The staff estimates that the portion of the unemployment rate not attributable to the cycle amounts to 9–10 percent of the labor force. In the copious literature on labor marker issues, this noncyclical portion of the unemployment rate is generally attributed to structural factors that reduce the incentives of workers to accept employment and of employers to create jobs. These factors arc thought to include labor market mismatch, regulations governing hiring or dismissal, minimum wage laws, social benefits (especially unemployment benefits), and the tax system.


The expenditure of the unemployment insurance funds alone came to more than \ percent of GDP in 1994. This does not include the amounts allocated for early retirement and social welfare payments.


Schmttt and Wadsworth (1993) conclude that job searching is not much affected by the level of unemployment benefits. However, their investigation was limited to the United Kingdom and did not take into consideration the broader range of benefits available to non-employed individuals. Other studies, for example, that by Layard and Nickell (1991). show that the duration of benefits, rather than their level, is what most affects the labor market. Many social benefits in France are independent of efforts ro participate in the labor market and are of essentially unlimited duration.


The overall decline in the labor force participation rate masks a strong increase in the rate for females and an even sharper decline in the rate for males.


Ideally, one would want to calculate the benefits obtained by persons of working age who arc not employed and to set them in relation to the net wages (including benefits) received by those persons who are employed. These data are not readily available.


Useful descriptions of changes in labor market policy may be found in Bolot-Gittler and in Cornilleau, Marioni, and Roguet (1990).


The rate is calculated by taking the ratio of unemployment benefits per unemployed person to gross wages pet employed person. Of course, the replacement ratio is substantially higher for a newly unemployed person. Under the new system of allcation unique dégressive adopted in 1993, the maximal gross replacement ratio is equal to 75 percenr (for an individual previously earning the legal minimum wage). See Moghadam (1995) for further details.


The minimum wage rate may be increased for three reasons: (1) automatic increases whenever the consumer price index (CPI) has increased by more than 2 percent; (2) automatic increases equal to one-half the rate of increase in gross real wages; and (3) increases at the discretion of the Government.


The nonemployment rate is defined (in levels) as the share of the working-age population that is not employed.


Details are found in appendix.


All variables were tested for order of integration. With only one exception, a unit root cannot be rejected in levels, but is strongly rejected in first differences.


Suitable proxies of this variable, such as the ratio of general government expenditure to GDP, perform similarly well.


Exogenous here means nonmodeled bur not necessarily exogenous in a statistical sense.


Where nonstationary variahles cointegrate, at least one linear combination of these can be found that has a stationary error.


For example, an I(1) variable has to be differenced once to make it stationary, and the resulting stationary first difference is referred to as I(0).


The sum of employment and nonemployment is the working-age population, which is an exogenous variable over the time horizons studied here.


This argument can also he east in a somewhat different light. Higher social benefits reduce what might be called “effective labor force participation,” which is measured labor force participation less those unemployed persons who are voluntarily unemployed (and arc therefore not really seeking work). The smaller the effective labor force, the lower is employment in a search and matching framework, since the lower will be the probability that any given employer will find a new recruit (at any given wage offer).


Elmeskov(1993), in a review of explanations for unemployment in the OECD countries, concludes that “most empirical evidence points to rather modest effects [of the minimum wage] on total equilibrium unemployment.”


A systematic treatment of the insider-outsider theory of employment and unemployment may be found in Lindbeck and Snower (1988).


Reducing social contributions and benefits for the top 20 percenr of wage earners, or even the top 60 or 80 percent, is likely to have relativeiy little effect on either their employment or their unemployment.


However, real interest rates at times showed an incorrect sign.


The model was also expressed in error correction form in order to confirm the long-run results.


Distinguishing between voluntary and involuntary unemployment is not possible on the basis of existing statistics.


Again, the model was obtained by reducing a general autoregressive distributed lag to a form in which the retained variables were significant (or close to significant in the case of real GDP), and the model reduction could not he rejected at any stage.