Abstract

The Substantial And persistent rise in Italian unemployment since the mid-1970s has prompted attempts by academics and policymakers to characterize its causes and provide possible policy responses. The research approach followed in this paper relies on two insights that appear not to have been fully explored to date either in Italy or in any other European economy that has experienced persistent unemployment. The main insight is that lags in labor market behavior are important and may reinforce one another. Thus, the full effects of a labor market shock on unemployment may take much longer to appear than the length of any simple lag, and unemployment may even overshoot its long-run equilibrium in response to either temporary or permanent shocks.

The Substantial And persistent rise in Italian unemployment since the mid-1970s has prompted attempts by academics and policymakers to characterize its causes and provide possible policy responses. The research approach followed in this paper relies on two insights that appear not to have been fully explored to date either in Italy or in any other European economy that has experienced persistent unemployment. The main insight is that lags in labor market behavior are important and may reinforce one another. Thus, the full effects of a labor market shock on unemployment may take much longer to appear than the length of any simple lag, and unemployment may even overshoot its long-run equilibrium in response to either temporary or permanent shocks.

The second insight is that labor market policies can affect the length of the lags and thereby influence the speed with which unemployment recovers from a recession. This feature of policy is significant, because an important aspect of the unemployment problem in Italy, as in other European countries, is not just that unemployment is high on average, but that it takes a long time to fall after adverse shocks have occured. A number of academic papers have pointed out potential sources of lagged responses of unemployment, such as hiring and firing costs, labor force adjustment costs, and insider employment effects.1 Thus, legislation on wage indexation, legal obstacles to firing and hiring, or even direct lagged responses to changes in policy variables, such as tax rates, can have an important impact on the speed with which unemployment adjusts to macroeconomic shocks. For time periods over which policymakers can have an impact (including not only on unemployment but also on inflation and other important macroeconomic aggregates), quantifying the sources and lengths of the various lags can be very important. The relevant lags can be quite long, but to the extent that they depend on policy, policies can to some degree reduce them: one can even argue that increasing labor market flexibility is synonymous with reducing the length of the lags. This is important in the present context, where the Italian authorities are adopting policies that are designed to increase labor market flexibility and ultimately reduce unemployment.

Overview of Macroeconomic, Labor Market, and Institutional Developments

Aggregate unemployment in Italy has followed a pattern similar to that in a number of other European economies. From relatively low levels of about 6 percent in the early 1970s, it began increasing after the first oil shock in 1974, reaching levels of 8 percent by 1980 (Chart 1). Unemployment accelerated after the second oil shock and continued to increase until about 1988, peaking at over 13 percent. It fell by about 2 percentage points following the economic recovery, but lost all gains during the most recent recession, which started at the end of 1991. From about 6 percent in 1971–72, inflation also increased significantly during 1973—80, indicating the presence of adverse macroeconomic supply shocks and accommodative aggregate demand policies by the authorities (Chart 2). Inflationary pressures have eased steadily since 1981, with inflation returning to 6 percent by 1986. Since 1986, inflation has tended to track the economic cycle.

Chart 1.
Chart 1.

Italy: Adjusted Versus Unadjusted unemployment

(In percent of adjusted/unadjusted labor force)

Sources: Italian authorities; and IMF staff estimates.1Unemployment adjusted for the impact of the wage supplementation fund (WSF).
Chart 2.
Chart 2.

Italy: Unemployment Versus Inflation

(In percent)

Sources: Italian authorities; and IMF staff estimates.Note: Unemployment adjusted for the impact of the WSF.

That important adverse macroeconomic shocks contributed to the rise in Italian unemployment is not in dispute. Real business sector capital stock has been in continuous decline in percentage terms (albeit punctuated by short-lived periods of resurgence) since about 1970. From a rate of increase of more than 1.2 percent a year, capital accumulation has slowed to just over 0.5 percent (Chart 3). Real GDP growth has also fallen, although there is also some evidence that its variability declined during 1983–90. Competitiveness, defined as the ratio of the import price deflator over the GDP deflator, also turned sharply negative in 1974–83, reflecting movements in oil prices (Chart 4). In general, however, the Italian economy has benefited from movements in competitiveness, especially since 1985. At the same time, social security contributions have risen, punctuated by plateaus (Chart 5), increasing both the costs of employment to employers and the wedge between pre- and post-tax returns to work.

Chart 3.
Chart 3.

Italy: Real Capital Stock Versus GDP

(In percent)

Sources: Italian authorities; and IMF staff estimates.
Chart 4.
Chart 4.

Italy: Competitiveness

(Import deflator over GDP deflator, index)

Sources: Italian authorities; and IMF staff estimates.
Chart 5.
Chart 5.

Italy: Social Security Contribution Index

(SSC divided by a wage bill index)

Sources: Italian authorities; and IMF staff estimates.

Over this period, the labor force participation rate has also varied over and above the change that would normally be expected as a result of movements in employment. During 1972–76, participation was relatively low at 56.5 percent of the working-age population, moderating the initial impact of the oil shock on unemployment (Chart 6). However, from 1976 to 1981, the participation rate rose in two steps, reaching more than 59 percent by 1980, substantially exacerbating unemployment. This rise was entirely due to the increase in the female participation rate, which increased from 30 percent to 36.7 percent. The female employment rate jumped from 26.7 percent in 1972 to 31.4 percent in 1981, even as the female unemployment rate increased to 14.4 percent from 10.9 percent. The male participation rate, in contrast, fell to 75.1 percent from 78.2 percent. The male employment rate declined to 71.1 percent from 74.6 percent over this period, while the male unemployment rate increased marginally to 5.4 percent from 4.6 percent. (See de Luca and Bruni (1993), Table 8, p. 22.) By 1991, the female participation rate stabilized at 40.1 percent, and the female employment rate increased to 33.6 percent, while the male participation rate continued to fall, reaching 70.5 percent, as the male employment rate fell to 65.5 percent. Thus, aggregate unemployment at least partly reflected shocks arising from an increasing labor force participation rate for females; it will be seen later that this increase in the participation rate was partly caused by factors exogenous to the labor market.

Chart 6.
Chart 6.

Italy: Participation Rate

(Labor force over working-age population)

Sources: Italian authorities; and IMF staff estimates.

A comparison of unemployment in Italy and in other countries of the European Union (EU) reveals that, although the pattern of time variation is similar, Italian unemployment is higher than the average: in fact, Italy has the third highest rate of unemployment after Spain and Ireland. This is the result of a striking regional dualism that has taken root in Italy, where the north is more industrialized and has relatively low unemployment rates, while the central region and the south are relatively more agrarian, more dependent on public employment, and have substantially higher rates of unemployment. In 1995, for example, the unemployment rate (according to the national unemployment definition) in the south was over 20 percent, a rate almost three times higher than in the north and the central region.2 An important explanation for the dual unemployment rate has in fact to do with the relatively small wage differentials between north and south despite wide productivity differentials.3

Adverse macroeconomic shocks and the marked regional duality provide only one part of the explanation for Italy’s higher unemployment rate. Another part has to do with the institutional framework in the labor market and various policies that determine the response of the labor market to macroeconomic shocks. Clearly, if the labor market were perfectly flexible—except perhaps for purely frictional unemployment—then one would expect unemployment to be less persistent than the macroeconomic shocks that are affecting it. But the institutional and policy framework of Italian labor markets is far from being perfectly flexible: according to the recent jobs study of the Organization for Economic Cooperation and Development (OECD) (1994) and Demekas (1994), the Italian labor market is one of the least flexible in Europe.4 To facilitate discussion and later econometric work, Appendix I provides a table of labor market and institutional and policy events for Italy.

At the beginning of the 1970s, the Italian labor market already had an institutional and legal structure that significantly inhibited flexibility. Nonagricultural firms were required to hire from a “public list” maintained by public employment agencies with a monopoly on employment services. The rank order of employees to be hired was determined by the state, taking into account such criteria as status of employment, duration of unemployment, and social factors, including the number of dependents. Wage indexation, the scala mobile, had already evolved to a national system of uniform adjustments based on a specially calculated index (the indice sindacale). Worker unrest during the fall of 1969 (the “hot autumn”) led to legislation (Charter of Workers’ Rights, or Statuto dei Lavoratori), which further expanded employment regulations, while Law No. 300/1970 gave special privileges to the three largest national trade unions. These reforms, together with the already significant dominance of the national trade unions, resulted in a centralized wage bargaining system whereby national wage agreements would be negotiated at three-year intervals: in general, longer contracts tend to result in more persistence and longer adjustment lags. The bargaining sessions of 1970 and 1973 resulted in wage increases that some observers characterized as “huge.” Firing workers was costly not only because of strong legal protection accorded employees and the presence of implicit contracts with them, but also because of a system of severance pay whereby wage deductions from employees were used to finance a severance pay fund that entitled employees to automatic severance payments upon termination of employment. Another institution peculiar to Italy was the wage supplementation fund (WSF, or cassa integrazione guadagni). Industrial workers who would normally be laid off would instead receive 80 percent of their wages from the WSF, almost entirely paid for by the state. Since official Italian unemployment statistics exclude these workers, all data used in this paper (in text, charts, and model estimation/simulation, except where otherwise indicated) have been adjusted to include them.

Before the first oil shock, labor market restrictions were not especially detrimental, as unemployment remained low because of the favorable macroeconomic conditions. It began to rise, however, after the first oil shock, when macroeconomic conditions became less favorable (GDP growth and capital accumulation slowed, while import prices grew faster than domestic prices), suggesting that labor market inflexibility was indeed costly. Despite the persistent rise in unemployment, labor market participants continued to seek—and achieved—further changes in labor market institutions that resulted in even greater inflexibility. A 1975 agreement between trade unions and Confindustria (the main national employers’ association) that was fully implemented by 1977 extended the coverage of the scala mobile to agricultural and service agreements (previously it had applied mainly to industrial northern regions); it also increased the frequency of indexation by adjusting wages on a quarterly basis. The increase in indexation implied by these changes was a reaction to the mounting inflation, yet it also contributed to a nominal inertia in wages, which raised the costs of disinflation. This would become apparent after 1981. Another important effect of the extension of the coverage of the scala mobile was to reduce wage differentials, both between industries and between regions, which tended to disproportionately penalize employment in the south, where productivity was relatively lower. Wage differentials, as measured by the coefficient of variation of interindustry wages, fell from about 23 percent in 1970 to about 16 percent in 1975 and 11 percent in 1977. Compared with other countries, Italy went from a position of relatively high wage differentials (above Germany, France, and the United States and below Japan) to a position of very low wage differentials (lower than all the aforementioned countries).

In 1977—83, unemployment continued to rise, and the authorities attempted without much success to introduce reforms in the labor market. Trade unions generally resisted, and the resulting reforms were piecemeal and contradictory. For example, the 1979 wage agreement indicated that more wage differentiation would be allowed in the next wage agreement: this might be taken as a sign of progress. However, the 1982 wage bargaining session failed to be concluded because of a stalemate on reform proposals for the scala mobile and wage increases, and wage differentials remained at the same low levels between 1981 and 1984. (See de Luca and Bruni (1994), Fig. 17, p. 77.) In 1980, the maximum payout from the WSF was increased in nominal terms for salaried employees, although average coverage fell to 76 percent from 80 percent. The severance pay system was relaxed somewhat by reducing the constant of proportionality for payments into the severance pay funds to less than 100 percent, but the scala mobile wage component was now included in the bases. In general, during this period there were complicated changes at the margin that were the result of compromise and did not address any of the fundamental institutional issues and policies.

In 1983–84, there was a more serious attempt to deal with the scala mobile. The 1983 wage bargaining agreement reduced the effective degree of indexation by 15 percent, while a once-off reform of incomes policy was introduced in 1984, which limited the scala mobile points that were allowed to be counted for wage indexation purposes. These reforms had some success in reducing overall wage indexation, as measured by the average degree of coverage in the manufacturing sector. This measure declined from over 70 percent in 1982 to about 60 percent (with some fluctuations) between 1983 and 1991. (See de Luca and Bruni (1994), Fig. 12, p. 63.) In 1984, hiring and firing regulations were changed to allow employers to hire a limited number of workers without regard to the rank order contained in the public list. Some improvement was also made regarding firms’ internal flexibility—their ability to reorganize internally—by introducing “solidarity contracts” that would permit workers in firms with more than 1,000 employees to share the work. These contracts were apparently only lightly used, however.

Labor market reform did not begin in earnest until 1991–92, when continuing high unemployment finally resulted in a political consensus to put in place a number of reforms designed to remove some of the impediments to labor market flexibility. The long-standing system regulating the rank order of hires was abolished in 1991. Some exceptions remained for the disabled and the disadvantaged (that is, the long-term unemployed). In addition, firms were allowed to place workers who were unemployed because of mass layoffs on a “mobility” list rather than on the WSF. Although workers on the mobility list and the WSF receive approximately equivalent benefits, and workers on the mobility list are supposed to receive priority in hiring, the effective employment protection accorded workers on the mobility list was substantially reduced relative to the protection enjoyed under the WSF. The 1991 wage accord suspended wage indexation, and the 1992 scala mobile adjustment was suspended. By 1993, the scala mobile was also definitively abolished and the wage bargaining system was reformed. Two levels of wage bargaining were defined (national/sectoral, and regional/firm) to take into account national benchmarks based on targeted inflation. Labor contracts were extended to four years, although nominal wages would be renegotiated every two years.

The effect of these policies was to improve wage flexibility somewhat, but also, more important in the Italian context, to improve the ability of firms to hire and fire. The recession, which began at the end of 1991, also resulted in an unprecedented shakeout in services (including the public sector). Unlike in previous recessions, employment in the services sector declined. The result of these changes was an unprecedented decline in employment and an increase in unemployment by about 2 percent. Comparing the behavior of employment during the most recent recession with that of earlier recessions shows that employment was much more cyclical and responsive to output than before (Chart 7). Clearly, firms used the opportunity of lower firing restrictions to restructure, and productivity increased substantially as a result (which is atypical for an Italian recession). The difficult question is whether firms will prove equally flexible in hiring people during the upswing.

Chart 7.
Chart 7.

Italy: Behavior of Employment During the Business Cycle, 1973–94

(Index; trough = 100)

Sources: Bank of Italy; and IMF staff estimates.Note: Employment series reconstructed before 1992 and spliced before 1981.

Finally, it is worth mentioning that a number of policies that are often important when considering the unemployment experience of other countries have not played an important role in Italy. Minimum wage legislation, for instance, does not exist at the national level. Unemployment benefits have also been relatively unimportant, with a replacement ratio of 10 percent throughout most of the period under review. However, the WSF in large part substituted for the absence of a more typical system of unemployment benefits, which is why this paper uses labor market figures adjusted for the WSF.5 In addition, the Italian authorities have recently increased the replacement ratio to 25 percent.

A Model-Based Analysis of Historical Developments

The analysis presented in the previous section was suggestive of the causes of unemployment in Italy, but, in the absence of a quantified framework, it is not possible to pinpoint the relative importance of the various factors or the time scale over which they operate. In this section, two structural vector autoregressive (VAR) models of the labor market in Italy are estimated for the period 1977–94 on the basis of quarterly data available as of April 1995. The first model will be referred to as the “basic” model. It will be used to calculate measures of persistence and imperfect responsiveness. The second model is an expanded, 11-equation version of the basic VAR model. The added equations are simple autoregressions for the exogenous variables in the basic VAR (so that the additional equations in the 11-equation VAR remain block-exogenous to the basic set of equations). The purpose of the expansion is to be able to perform a variance decomposition for unemployment (also for employment, real wages, and the labor force) that includes the exogenous variables.

The basic VAR consists of three behavioral equations, shown below:

nt=fn(t)+ann(L)nt1+anωωpt+ankkt+anrrt+anyyt+ans(L)SSCt,(1)
ωct=fω(t)+aωn(L)nt1+aωωωct1+aωppt+aωs(L)SSCt,and(2)
lt=fl(t)+atn(L)nt+an(L)lt1+alωωct+ataWAPt,(3)

where n denotes employment, wc the consumption wage, wp the product wage, l the labor force, k business sector capital stock, r competitiveness, y real GDP, p productivity, WAP the working-age (14–64) population, and SSC the ratio of social security contributions to the wage bill (in index form). L denotes the lag operator (for any time series xt, Lxt = xt-1), a(L) a polynomial in the lag operator, and f(t) a deterministic function that includes a constant, a time trend, and quarterly dummies, because all data are seasonally unadjusted. All variables, except SSC, are measured in logarithms. Appendix II provides further details on data sources and definitions.

Equation (1) is an employment equation (which may be interpreted as a labor demand equation), equation (2) a wage setting equation, and equation (3) a labor force (participation, or labor supply) equation. The unemployment rate is then determined as a log-linear approximation to the identity linking unemployment, employment, and the labor force (ut = lt – nt). Because the appropriate wage concept for the labor demand equation is the product wage, but for the wage setting/labor supply equation it is the consumption wage, an additional equation is appended for the product wage (in effect, capturing the wedge, which is shown in Chart 8). For the work reported in this section, the product wage is computed as an identity that uses the within-sample wedge together with the simulated consumption wage.

Chart 8.
Chart 8.

Italy: Wedge

(Log CPL-log PPI)

Sources: Italian authorities; and IMF staff estimates.Note: CPI refers to the consumer price index, and PPI to the producer price index.

In choosing the specifications included in equations (1)(3), a number of alternatives were considered, especially regarding which explanatory variables to include and whether to use the unemployment rate directly rather than employment or the labor force or both. The statistical significance of the coefficients as well as the overall fit of the system and its dynamic properties were considered for inclusion. Variables such as oil prices and direct tax rates were not statistically significant in the employment equation. The labor force participation equation also tended to have worse autocorrelation properties when unemployment was used instead of employment and lagged labor force.6 Table 1 summarizes the best specification achieved for each equation.

Table 1.

Regression Results, Basic VAR

article image
Notes: n denotes employment, wc the consumption wage, wp the product wage, l the labor force, k business sector capital stock, r competitiveness, y real GDP, p productivity, WAP the working-age (14-64) population, and SSC the ratio of social security contributions to the wage bill (in index form). All variables, except SSC, are measured in logarithms. Appendix II provides further details on data sources and definitions.

In the employment equation, the coefficients on the explanatory variables have the expected signs. According to the results, increases in the product wage decrease employment, as do unfavorable competitiveness shocks and increases in social security contributions.7 They show that faster capital stock accumulation and GDP growth increase employment. The inclusion of GDP in the employment equation requires some explanation because, according to the standard competitive model of the firm, output does not enter the factor demand equation (although it does enter the conditional factor demand equation, in which case capital does not enter; see Varian (1984), Section 1.7). One interpretation is that the model of the firm that implicitly underlies the employment equation considered here is one of imperfect competition. As in Layard and Nickell (1987), an activity variable (or, in their model, an expected activity variable, which could be proxied by contemporaneous or lagged GDP) plays a role in a firm that is not perfectly competitive and that chooses prices, output, and employment on the basis of its expectation of activity.

The phenomenon of labor hoarding (whereby the firm hoards labor in anticipation of increases in activity, which again is signaled by increases in contemporaneous activity) is an instance of this. The results show that employment appears strongly correlated with output and has an implicit coefficient of 1.9, a reasonable number.8 In the short run, however, the estimated responsiveness of employment to output is very low, which also corresponds to conventional wisdom.

Interestingly, after the effects of the explanatory variables are accounted for, a statistically significant trend decline in employment remains. This could be the result of an omitted, and as yet unidentified, explanatory variable. Alternatively, it could indicate the effect of a continuous introduction of labor-saving technology. Significant lagged employment effects are estimated, which are normally interpreted as capturing employment adjustment costs. Overall, social security contributions also affect employment negatively with a lag pattern that is related to employment adjustment costs: not all of the adjustment to a change in social security contribution rates can take place immediately.

For the wage equation, it proved difficult to obtain significant and plausible effects from a set of potential explanatory variables. Unemployment, either contemporaneously or with a lag, appears to affect consumption wages only weakly, and the results indicated that it could be dropped from the final specification. This points either to significant insider power (because insiders are able to insulate themselves from market forces in the wage setting process) or to a weak effect of the excess supply of labor on wage settlements (a Phillips curve effect). This effect is strong in Italy, but it operates through the participation rate rather than through wages. Productivity affects wages positively, but does not appear to be statistically significant. This may be partly because the productivity variable has been smoothed by applying a centered five-quarter moving-average filter, which may be too short to smooth out cyclical movements in productivity (a nonsmoothed productivity variable yielded similar results). However, other researchers have also noted the lack of a strong link between real wages and productivity. (See de Luca and Bruni (1994), pp. 69–71.) The social security contribution variable also does little, and there is no residual trend movement in wages. Direct taxes also proved insignificant.

Most of the important effects in the wage equation are lagged effects. There is a significant lagged real wage coefficient, which can be theoretically linked to wage staggering effects. Because of the operation of the scala mobile in Italy throughout most of the sample period, one would expect strong wage staggering effects, which indeed are shown to exist. The other interesting lag effect is the statistically and economically significant three-quarter lagged effect of employment on wages, which can be interpreted as the “insider employment” effect. Recalling the earlier discussion of hiring and firing rigidities, it can be inferred that insiders will have an unusually strong bargaining position in the Italian labor market. This means that as employment increases, so does the strength of insiders, perhaps after some lag, before the new insiders become fully entrenched. According to the estimated effect, a 10 percent increase in employment results in an almost 2 percent increase in real wages after three quarters. Overall, the results for the wage equation are consistent with the conventional description of wage bargaining in Italy; namely, centralized trade unions negotiated with centralized employers under conditions that have tended to favor insiders in the past and resulted in wage bargains that did not take market conditions into account.

The labor force equation is notable for the strength of the discouraged-worker effect. Variations in employment appear to cause significant variations in labor force participation; the long-run elasticity of the labor force with respect to employment is 0.576. These results suggest that a significant part of labor market clearing takes place through variations in labor supply. Consequently, the unemployment series underestimates the amount of excess capacity in the labor market in the aftermath of a recession. Considerable labor force adjustment effects are also evident in the equation, with significant lags for both employment and the labor force. The working-age population variable appears to be only marginally significant, perhaps because the relatively large shifts in labor participation caused by female entrants into the labor force—which are unrelated to the increase in working-age population—are masking its effect. Nevertheless, the coefficient has the correct sign. The consumption wage has the expected positive effect on labor participation, although its statistical significance is marginal. Finally, there remains a statistically significant residual trend reduction in the labor force, which may be related either to an omitted variable or to the reduction in the male participation rate noted earlier.

Perhaps more interesting than the individual equation results is the way in which the equations interact in the context of a system. VARs were in fact emphasized by Sims (1980), among other researchers, with a view to examining cross-equational effects and dynamic impulses. However, VARs also raise some difficult issues of interpretation. One is the identification problem, which stems from the fact that in general the VAR covariance matrix is nondiagonal (so that the shocks to the individual equations tend to be contemporaneously correlated) and a unique method for “diagonalizing” a general VAR is not available. Typically, researchers using VARs accept them either as “atheoretical,” which means that they are meant to capture generic dynamic responses that may not correspond to easily understood economic shocks (for example, demand or supply), or as “structural” that is, they identify the sources of shocks using restrictions. A good example is the analysis of demand and supply shocks by Blanchard and Quah (1989). These researchers were able to identify the demand and supply shocks by a priori imposing a restriction on the lag structure of the moving-average representation of the VAR. (They assumed that the demand shock had a purely temporary effect, so that the long-run response of the VAR to a demand shock was zero, while a supply shock had a permanent effect.) However, in this project, the aim is to estimate the lag structure directly.

The VAR estimated here is of the structural type. The case rests partly on theoretical grounds, which provide guidance as to which variables to include in a labor demand or supply equation (see Chapter 1 in this volume). However, because of data limitations or an unsuccessful specification, the estimated VAR may not possess the requisite properties. Two pieces of evidence support the structural interpretation of the estimated VAR for Italy. First, the estimated covariance matrix for the VAR is “near diagonal” (see Table 2 for the expanded version of the VAR), indicating that contemporaneous covariances between the endogenous variables are successfully captured by the VAR specification and obviate the need for complicated—and possibly unfounded—identifying assumptions. Second, the impulse responses of shocks to the employment equation are damped, and the long-run response of employment to a temporary shock on employment is zero (see Charts 9-11), as would be expected if the employment equation corresponded to a demand equation.

Table 2.

Correlation Matrix of Residuals, 11-Equation System

article image
Notes: For definitions, see Table 1; see also Appendix II.
Chart 9.
Chart 9.

Italy: Persistence Versus Imperfect Responsiveness

Sources: Italian authorities; and IMF staff estimates.
Chart 10.
Chart 10.

Italy: Impulse Responses and Labor Demand

(Temporary shock)

Sources: Italian authorities; and IMF staff estimates.
Chart 11.
Chart 11.

Italy: Impulse Responses and Labor Demand

(Permanent shock)

Sources: Italian authorities; and IMF staff estimates.

The covariance matrix elements (in fact, reported as correlations) are fairly small in almost all cases, and a correlation of –0.49 between working-age population and competitiveness is not considered to signify substantial misspecification. A few exceptions may not represent sampling error; these include a negative residual correlation between the product wage and social security contributions and competitiveness, a negative correlation between the labor force and social security contributions, and an apparently strongly negative correlation of –0.72 between unemployment and capital stock. But, overall, most of the correlations reported appear small enough to justify the characterization of the covariance matrix as near diagonal.

To further check the result of the estimated VAR, its out-of-sample forecasting performance was examined and dynamic simulations were performed. The out-of-sample forecasting performance was examined systematically by calculating the root-mean-squared (RMS) error statistic and the Theil U statistic over different forecast steps. (See the results in Table 3.) The dynamic performance of the model can be seen in the simulations presented in Charts 12-15. Generally, the model has good out-of-sample forecasting properties, with RMS increasing only gradually as the forecasting horizon increases. The calculated Theil U statistics, which are well below 1, indicate that the basic model easily outperforms a “naive” random walk model. The dynamic simulations (which will be discussed more fully in a later section) also seem to indicate that the model tracks well within sample and forecasts reasonably well out of sample. Appendix III provides further tests of the statistical adequacy of the model.

Table 3.(RMS error in parentheses)
article image
Notes: n denotes employment, wc the consumption wage, l the labor force, and u the unemployment rare. The Theil U statistic computes rhe ratio of the forecast error to the forecast error obtained by a random walk model in each variable.
Chart 12.
Chart 12.

Italy: Employment and Unemployment

Sources: Italian authorities; and IMF staff estimates.
Chart 15.
Chart 15.

Italy: Behavior of the Labor Force During the Business Cycle, 1973–94

(index; trough = 100)

Sources: Bank of Italy; and IMF staff estimates.Note: Labor force series reconstructed before 1992 and spliced before 1981.

Given that the model adequately captures some of the dynamic interrelationships that seem important for the Italian labor market, it is possible to use the model to gain a quantitative understanding of the sources of unemployment. The main tool will be a variance decomposition for unemployment that uses the extended version of the model. Although this method has the disadvantage (when compared with the direct measures of persistence and imperfect responsiveness to be provided below) of not disentangling the lagged effects from the effects of exogenous variables, it is nevertheless a convenient way to characterize the “gross” sources of unemployment. The variance decomposition is a way of determining the fraction of the innovations in employment, real wages, labor force, and unemployment explained by each endogenous variable in the model. To implement this, the variables that appeared as exogenous in the basic model must be endogenized. This is accomplished by appending autoregressions for the exogenous variables to the basic model and re-estimating the extended model (in effect, the additional variables are treated as being block-exogenous). The results of the variance decomposition in the short run (1 step ahead, or one quarter) and in the medium to long run (24 steps ahead, or six years) are collected in Table 4.

Table 4.

Variance Decomposition, 11-Equation Model

(1 step ahead; 24 steps ahead in parentheses; in percent)

article image
Notes: For definitions, see Table 1; see also Appendix II.

A number of results become apparent. Unemployment is significantly affected by changes in the labor force (in fact, labor force appears to dominate employment), a result that holds in both the short and the long run. This result seems plausible for Italy, given the large exogenous changes in the participation rate over the sample period; it may be less significant for other countries.9 In the short run, all series are dominated by their own innovations. However, in the long run, the labor force retains about 33 percent explanatory power, and the consumption wage about 20 percent, while employment shocks become negligible. These findings are consistent with the interpretation that employment represents short-run demand by firms, whereas wage setting and labor force participation represent supply-side effects, which are expected to persist in the long run—as argued, for example, by Blanchard and Quah (1989). The product wage seems to behave differently, retaining a 20 percent long-run effect from the consumption wage, but a 60 percent effect from its own innovations.

This surprising result is probably related to the rather persistent behavior of the wedge over the sample period (Chart 8).

One result with far-reaching implications, however, is that variables treated as exogenous in the extended model (capital, GDP, working-age population, and competitiveness) ultimately become predominant, explaining 64 percent of employment, 60 percent of the consumption wage, and 54 percent of the labor force variance in the long run. This result is consistent with a view that attributes much of the rise in unemployment to the weak macroeconomic performance of the Italian economy together with unfavorable external shocks. However, it must be kept in mind that the impact of the unfavorable macroeconomic environment was magnified and sustained by the inflexibility of Italian labor market institutions, which kept real wages high and made it difficult for new entrants into the labor force to find jobs.

Another significant result is the effect of the increases of the social security contributions over the sample period. This variable alone explained 16 percent of the variance of employment and 13 percent of the variance in the consumption wage. Perhaps more important, this effect largely persists into the long run. This finding is important, however, not only because it indicates that policy had a significant detrimental effect, but also because of the implied statistical properties of a policy variable. Social security contribution is a persistent variable—it is 1(1). Thus, to the extent that the statistical properties of social security contributions are representative for other policy variables, the effects of policy (both detrimental and favorable) can be more powerful and long lasting.

Evidence on Labor Market Policies and Lagged Effects

In the framework adopted in this paper, policy can have three distinct roles. One, it can affect unemployment directly (for example, social security contributions were shown to reduce employment in both the short and the long run). Two, it can affect unemployment indirectly, by influencing variables that were considered exogenous from the point of view of the model (for example, GDP growth, capital formation, and competitiveness). Three, it can affect the lags with which both policy and nonpolicy variables affect unemployment, an area of particular importance under the framework of Karanassou and Snower (1994). For example, social security contributions affect employment both contemporaneously and with a lag; the scala mobile resulted in more rigid wages and a longer wage adjustment lag; and firing costs resulted in longer employment adjustment lags. Each will be discussed in turn.

The direct effects of policy on unemployment are the simplest to describe. As already shown, social security contributions increased the cost of employment to employers and introduced a wedge between the value of work to employees. The overall impact on employment and wages was significant and persistent, as shown by the variance decomposition. The reason for the continuous rise in social security contributions is probably related to two factors: demographics (Italy has one of the highest old-age dependency ratios in the OECD), and increasing spending on pensions, which have grown from some 5 percent of GDP in the 1960s to more than 15 percent in the 1990s. Compared with other countries, Italian spending on pensions in the 1960s was close to the average of the seven major industrial countries, but increased to the highest in the OECD (except Austria) by the 1990s. Although the increase in pensions is partly related to demographics, the generosity of the Italian pension system is also responsible: Italy’s old-age pension benefits, when expressed as a ratio to per capita GDP, are the highest of the EU countries (see Canziani and Demekas (1995)).

The indirect effects of policy on unemployment, which operate through such variables as GDP, capital accumulation, and competitiveness, are harder to pin down because they touch upon a number of difficult macroeconomic issues that cannot be analytically resolved in the absence of a fully specified macroeconomic model.10 However, a number of comments can be made. Demand-management policies were shown to be ineffective after the first oil shock. Academics and policymakers alike no longer believe that it is possible for the authorities to raise GDP (or capital accumulation) in a sustained fashion through demand-management policies.11 However, structural policies can influence GDP and capital accumulation. Two such sets of policies that, overall, could be expected to have detrimental effects on GDP and capital accumulation in Italy can be mentioned. The first concerns regional policies vis-à-vis the south, while the second concerns the net effect of overall revenue and expenditure policies—the continuing accumulation of a large stock of public debt.

In the 1960s and 1970s, regional development in the south was predominantly attempted through the installation of capital-intensive industrial enterprises, which were at the time expected to contribute to employment through linkages with other sectors. (See de Luca and Bruni (1993), pp. 28–33.) This policy, by effectively subsidizing capital-intensive activities at the expense of relatively abundant labor, had precisely the opposite effect from what was intended. (Another way of describing the effect of the policy is that, for a given level of investment, fewer jobs were created.) In addition, the south was granted easy access to public sector employment, but wages were set in a centralized fashion without adjusting for local conditions, including cost of living, wages, and unemployment. As previously discussed, this was exacerbated by the effect of centralized bargaining and wage setting. The overall effect was to raise wages for the whole region well above market-clearing levels.

The other major policy influence on both GDP and capital accumulation is the continuing accumulation of public debt, which raised long-term real interest rates. Although academics tended to challenge this point, recent evidence suggests that the levels of public debt have risen throughout the industrial world, with statistically and economically significant effects on real interest rates (see Ford and Laxton (1995)). In addition, Italian public debt has risen faster than in the rest of the EU, and real interest rate differentials between Italy and Germany are wider. Clearly, these phenomena will continue to depress capital formation in Italy until the level of public debt is brought down.

The third way in which policy in Italy has affected unemployment is through its influence on the delay with which both policy and nonpolicy variables affect unemployment. These lags have received special attention in this study. Italy appears to exhibit a full complement of these effects: an employment adjustment effect (lagged employment in the employment equation), a wage staggering effect (lagged real wages in the wage setting equation), an insider membership effect (lagged employment in the wage setting equation), and a labor force adjustment effect (lagged labor force in the labor force equation). A fifth effect, arising from the fact that the long-term unemployed discourage employment, seems important for Italy, although a similar effect appears to operate through lagged employment terms in the labor force equation rather than through lagged unemployment terms in the wage setting equation. In other words, the long-term unemployed tend to leave the labor force altogether, with little effect on wages; in contrast, insiders appear to have an effect on wages.

Policy in Italy has tended overall to act in a way that would be expected to magnify these lags. The main policies were described in greater detail in an earlier section. To recapitulate, wage indexation in the form of the scala mobile would be expected to result in more rigid wages and a longer wage adjustment lag (a stronger wage staggering effect); firing costs and employment protection would be expected to result in a stronger insider membership effect as well as longer employment adjustment lags; hiring costs (the requirement that hiring be done through public employment agencies in the order determined by the state) would be expected to increase employment adjustment lags, strengthen the insider membership effect, and perhaps increase the discouraged-worker effect.

Not all of these effects are necessarily expected to affect the average level of unemployment. The employment adjustment effect, the wage staggering effect, and the labor force adjustment effect would tend to increase the lag with which unemployment adjusted after being subjected to shocks, but would not in themselves increase unemployment. On the other hand, the insider membership effect and the discouraged-worker effect can affect average unemployment by, respectively, increasing real wages and lowering the labor force. There is some theoretical controversy about whether employment protection legislation (hiring and firing costs) would be expected to affect average unemployment. Bertola (1990) constructs models that imply that the variance of employment would tend to be reduced, but average unemployment remain largely unchanged. Snower and Lindbeck (1990), on the other hand, argue that because the bargaining power of insiders would increase in the presence of hiring and firing costs, and because the presence of insiders would tend to increase wages, average unemployment would also be expected to increase.

Estimating the quantitative impact of policy on the various lag effects for Italy is complicated by a number of factors. First, it is generally impossible to measure policy directly: policy changes tended to be of a complicated, legal nature that is not amenable to direct measurement. Second, not all policies changed in a discrete manner over the sample period. Finally, when policy did change discretely, it tended to happen in a number of areas simultaneously, which complicates the interpretation of the results. Nevertheless, the effect (or the absence thereof) was estimated through the construction of dummy variables that captured the effect of policy changes that occurred at known times; Appendix I lists known policy changes and their timing. Two potentially significant break points were found in the sample when policy changed: 1985, when the scala mobile was somewhat weakened and internal labor flexibility increased; and 1992, when the scala mobile was abolished, the national wage bargaining system was revised, the public list system of hiring was abolished, and firing costs were generally reduced.

A statistical investigation of coefficient stability before and after 1984 was carried out to see whether policies (or, indeed, other factors) had significant effects on labor market behavior. Given preliminary results for other countries,12 and some existing research results for Italy,13 it was expected that a structural break would be found. Surprisingly, the evidence points instead to a rather impressive degree of parameter stability and a relative lack of policy-induced changes. First, F-tests of overall parameter stability were performed for each of the three equations (employment, wage setting, and labor participation). For all three equations, the hypothesis of joint parameter stability failed to be rejected at conventional significance levels.14 A visual examination of the residuals of each equation also failed to reveal large outliers or any other evidence of misspecification (such as heteroscedasticity, or lack of stationarity of the residuals). Then, the equations were re-estimated, allowing the major policy-related coefficients to differ across the two subsamples.15 The relevant results are reported in Table 5.

Table 5.

Parameter Stability, 1977-84 Versus 1985-94

article image
Notes: For definitions, see Table 1; see also Appendix II.

Most coefficients change remarkably little from one subsample to the next. The main exception is a weakening of the discouraged-worker effect in the labor force equation, a change that is relatively large and approaches statistical significance.16 The only statistically significant change was in the first employment lag in the employment equation, but the economic significance of the magnitude of the difference was negligible. In all other cases, statistical tests failed to reject the hypothesis of individual coefficient equality across subsamples. The principal conclusion then is that policy changes after 1984 were largely ineffective, failing to reduce labor market rigidities.

Finally, the model was simulated to investigate the degree of inertia and the sources of lags. To conduct the simulations, the three basic labor market equations were augmented by the unemployment and product wage identities, and by an estimated production function to capture the fact that output would be expected to change from the actual path because the simulation resulted in employment paths that differed from actual.17 The model was subjected to a 1 percent transitory negative labor demand shock. The model exhibited persistence in the sense that the path of the variables settled down to equilibrium after a considerable lag (approximately seven to eight years; see Charts 9-11). The long-run effect, however, was negligible, as would be expected for a stable system. Equally, when the model was subjected to a 1 percent permanent negative shock, it exhibited imperfect responsiveness in that the endogenous variables reached their new equilibrium level after a considerable lag (also approximately seven to eight years; see Chart 9). The impulse responses of the system are shown in Charts 10 and 11. Measures of these effects, and their sources, are contained in Table 6.

Table 6.

Measures of Persistence and Imperfect Responsiveness

article image
Note: Persistence is defined as the sum of deviations from the base path when unemployment is subjected to a temporary shock. Absolute persistence sums absolute deviations. Imperfect responsiveness sums deviations from the base path after subtracting the change in the long-run equilibrium resulting from subjecting unemployment to a permanent shock. Both measures are standardized by the size of the shock, which is a 1 percent reduction in labor demand. The sources of persistence and imperfect responsiveness are calculated by leading all lagged terms (setting L-kxt-k) of the source variable while leaving all other lags intact.

The model exhibits positive persistence, as unemployment is driven up for up to two years following a temporary shock. However, the discouraged-worker effect eventually drives so many workers out of the labor force that measured unemployment undershoots its long-run path (Chart 10).18 In fact, the sum of deviations is approximately zero, as is shown by the aggregate persistence measure. The absolute persistence measure is strictly positive, reflecting the undershooting. Employment also exhibits overshooting, although the effect of a temporary’ shock lasts longer. Some overshooting occurs because of output effects. The inclusion of a production function magnifies swings in employment, because as employment comes down, output is also reduced, which further reduces employment; the opposite occurs as employment increases. In terms of both persistence measures, employment lags in the employment equation contribute negatively. This can be explained by the fact that lags in the employment equation delay the initial fall in employment, hastening the reversal of employment to its eventual equilibrium level. Similarly, the wage staggering effect contributes to persistence positively, albeit by extending the time period over which wages adjust; in addition, wages also overshoot slightly, and the absolute persistence measure is higher.

In contrast, the insider employment effect tends to moderate persistence. An intuitive explanation is that wages fall as the stock of insider employees is reduced, which tends to moderate the fall in employment, which therefore adjusts faster back to its initial equilibrium level. The lag effects stemming from the labor supply equation are both more complicated and difficult to separate into their constituent parts, because they tend to interact intimately with lag effects from the employment equation. More intuitive results are obtained by considering these effects together. Clearly, although the labor force adjusts temporarily downward, as does employment, the net effect on unemployment remains positive. Hence, the net effect of the lagged terms plus their interaction is to contribute significantly to both persistence and absolute persistence.

The model also exhibits positive imperfect responsiveness, to which all but one term contribute. Clearly, employment lags cause employment to fall sluggishly to its new equilibrium level, contributing to imperfect responsiveness. The wage lags also cause wages to adjust downward toward their new equilibrium level with a lag. The two most interesting reactions to permanent shocks come from the insider employment and labor supply effects. The employment lags in the wage equation contribute positively to imperfect responsiveness because, by reducing wages, they tend to increase employment and hence delay its reduction to the new, lower equilibrium level. The discouraged-worker effect, together with employment interactions, tends by contrast to speed adjustment strongly because in Italy labor force participation effects are especially strong.19 Some workers are so discouraged as to leave the labor supply permanently: this causes equilibrium unemployment to shrink, which makes the transition to the new equilibrium level occur faster.

The Recent Recession and Increased Labor Market Flexibility

Whereas the labor market policies of the mid-1980s failed to have much of an impact, the same cannot be said for the reforms after 1991. Employment appeared much more flexible, although evidence of this exists only for the downward direction. This section will attempt to interpret events, although the short sample and some data uncertainties mean that the conclusions reached here should be treated with caution.

To gain an understanding of changes in labor market behavior after 1991, the model was estimated only up to 1991 and was used to produce forecasts up to 1994 (see Charts 12 and 13). Employment fell much more than would be expected based on the historical relationship between output and employment during a recession (see also Charts 7, 14, and 15). A formal statistical test of parameter change along the lines pursued earlier is not possible, because there are not enough degrees of freedom. However, it is possible to perform Chow’s predictive test, which compares the forecasting performance of the model with actual developments. When this test was performed, the hypothesis of parameter constancy in the employment and labor force participation coefficients was decisively rejected, whereas for the wage setting equation it was accepted.20 The results for the wage setting equation are somewhat surprising, because wage behavior was thought to have changed markedly after the recent reforms. The results suggest that wage setting behavior has not changed very much, although labor demand could have become much more sensitive to wage variation. It should also be kept in mind that the model is couched in terms of real wages and that nominal wages could have improved substantially.

Chart 13.
Chart 13.

Italy: Labor Force and Real Wages

Sources: Italian authorities: and IMF staff estimates.
Chart 14.
Chart 14.

Italy: Behavior of Industrial Production During the Business Cycle, 1973–94

(Index; trough = 100)

Sources: Bank of Italy; and IMF staff estimates.

Although we do not have sufficient degrees of freedom to test for parameter inconstancy in general, it is possible to get a sense of the sources of the changes by imposing some prior conditions. This is accomplished by re-estimating the equations for two subsamples, 1977–91 and 1992–94, but allowing only a small number of the most policy-relevant coefficients to change. The employment equation had difficulty distinguishing between changes in the relationship between employment and output, changes in the effect of wages on employment, and changes in the effect of a level shift, which is not surprising given the shortness of the sample. When all three were allowed to change, the effect of wages on employment increased from -0.11 to -0.87. The change was not statistically significant at conventional levels of confidence (the t-statistic was 1.3), but, if held over a longer period, this result would be a very important development. The employment adjustment effect was lower than it was when the equation was estimated over the full sample, with lagged employment coefficients at 0.41 and 0.08, at the first and third lags, respectively. Moreover, a statistically significant reduction in the third lag was observed after the reforms. In addition, a statistically significant upward level shift seemed to take place, while the sensitivity of employment to output appeared to weaken.21 The wage setting equation showed a decline in the wage staggering effect: the coefficient on lagged wages fell from 0.75 to 0.52. Again, the change was not statistically significant at conventional levels of confidence, but, with a t-statistic of 1.5, could prove more significant in the future. Finally, the labor force equation showed both economically and statistically significant reductions in the discouraged-worker effect, with lower coefficients on lagged employment.

In summary, preliminary results suggest that labor market behavior changed markedly after the reforms. Employment was much more flexible, with a weaker employment adjustment effect and a weaker discouraged-worker effect. Firms became much more sensitive to wages and the wage staggering effect became less pronounced. It was not possible to detect statistically significant changes in real wage setting behavior, but this could change as more data become available (reforms on wage bargaining have been in place only since 1994). These results accord well with the nature of the reforms, which eliminated the scala mobile and concentrated on removing impediments to hiring and firing.

The Future

To investigate further the possible future effects of a more flexible labor market, the model (estimated on the basis of the full sample) was simulated to produce forecasts for employment, unemployment, wages, and the labor force (see Charts 16 and 17). The underlying assumptions, including output and capital formation projections, were consistent with the assumptions contained in the most recent version of the IMF’s World Economic Outlook.22 The simulation shows that, although employment is projected to increase substantially and real wages are projected to decrease somewhat before they rebound, unemployment falls by less than 1 percent over the forecast period. This result is consistent with the average behavior of unemployment over the whole sample, but does not take into account the possibility of continuing shedding of employment by firms. The previous section presented evidence of a regime shift as a result of the labor market reforms. Future developments will therefore hinge on which of these two forces predominates and on whether the labor market reforms are maintained or, better yet, strengthened.

Chart 16.
Chart 16.

Italy: Employment and Unemployment

Sources: Italian authorities; and IMF staff estimates.
Chart 17.
Chart 17.

Italy: Labor Force and Real Wages

Sources: Italian authorities; and IMF staff estimates.

It is important to attempt to investigate the potential positive impact of making the labor market more flexible and reducing the various adjustment lags. This will provide some indication of the “upside” potential of labor market reforms. To investigate this, an additional simulation was performed that assumed that labor shedding did not continue.23 The equations were simulated with lagged coefficients on employment and wages consistent with the reductions suggested by the preliminary estimates reported earlier. The lagged employment terms in the employment equation were reduced by 0.20, the wage elasticity of employment increased by -0.76, the lagged wage coefficient in the wage equation was reduced by 0.25, and the lagged employment terms in the labor force equation were reduced by 0.02.24 (See Charts 18 and 19.) With the aforementioned caveats in mind, they indicate that impressive gains in employment, wages, and the labor force are possible and that unemployment could fall to less than 10 percent—translating the results of the WSF-adjusted unemployment figures to the national definition—by 1999.

Chart 18.
Chart 18.

Italy: Employment and Unemployment

(Post-1994 scenarios)

Sources: Italian authorities; and IMF staff estimates.
Chart 19.
Chart 19.

Italy: Labor Force and Real Wages

(Post-1994 scenarios)

Sources: Italian authorities; and IMF staff estimates.

Conclusions

The structural VAR model that was estimated in this paper appears to capture well a number of important dynamic and structural relationships in the Italian labor market. It was used to investigate Italian unemployment from different points of view. Over the long run, it was found that increasing Italian unemployment resulted from the interaction between adverse macroeconomic shocks and an inflexible wage setting mechanism. Over the short to medium run, substantial lags were found to exist that were linked to a number of policies and that interacted in significant ways. When subjected to a temporary shock, unemployment tended to settle down to equilibrium after a considerable lag (seven to eight years). Employment adjustment costs and the wage staggering effect tended to increase unemployment persistence, while the insider employment effect tended to reduce it. Unemployment also responded sluggishly to permanent shocks, owing to lags from all effects except the discouraged-worker effect.

The labor market reforms that were introduced after 1984 were found, as a result of statistical tests, to have been largely ineffective in reducing labor market rigidities. However, the reforms after 1991 appear to have resulted in a labor market that exhibits substantially more flexibility on the employment and labor-supply sides, although wage setting remains a question mark. There is, however, some evidence of a reduction in wage staggering, which is the expected result of the abolition of the scala mobile. Provided that the labor shedding witnessed since the last recession is discontinued and that the labor market flexibility that seems to have increased after 1991 persists or is strengthened, significant gains against unemployment could well take place.

Appendix I: Labor Market Institutional and Policy Events

article image
article image
article image
article image
article image
Notes: SM = scala mobile (sliding scale)WB = wage bargainingWSF = wage supplementation fund (cassa integrazione guadagni)UnB = unemployment benefits; WSF is a form of UnB

Appendix II: Data Sources and Definitions

Most of the raw quarterly data used in this paper were provided by the Bank of Italy, and were available as of April 1995. Exceptions include the capital stock and the fiscal variables (social security contributions and direct taxes), which were provided by the OECD, and exchange rate and oil price data, which were provided by the IMF. As mentioned in the text, the main labor market aggregates (labor force, employment, and unemployment) for Italy do not include those workers placed on the wage supplementation fund (WSF). Hence, all figures used in the paper have been adjusted to include them, except where otherwise indicated.

The Italian authorities recently revised the main labor market aggregates from 1992Q4 onward. This created a problem of interpretation, since the dramatic changes evident in labor market data after 1991 could possibly be attributed to the data revisions. The Bank of Italy has produced revised figures for labor force, employment, and unemployment going back to 1981, and these figures are used in the paper (further revised to include the impact of the WSF).25 The revised figures indicated that the evident break in the series after 1991 was not caused by the data revisions themselves. However, in 1992Q4 there is also an evident break in population data, and further revisions of the labor force statistics are likely as new census data are processed. Hence, the post-1991 results should be treated with caution.

The wage series used in this paper is wages in the manufacturing sector. Although an economywide aggregate would have been preferable, the manufacturing wage series goes back further in time and appears to track the economywide aggregate closely where the two overlap. The consumption wage is defined as the manufacturing wage deflated by the consumer price index (CPI), and the product wage is defined as the manufacturing wage deflated by the producer price index (PPI). Although these definitions are at best proxies for the theoretically indicated concepts, corrections were provided for any definitional imprecisions by including tax variables in the regressions—both social security contributions and direct taxes. Owing to data limitations, it was not possible to decompose social security contributions into employer and employee components (at least at a quarterly frequency for the time period under consideration).

Appendix III: Estimation and Testing of the Labor Market Model

All the series used in this paper were tested for stationarity using Augmented Dickey-Fuller (ADF) tests. Results indicated that all variables were integrated of order one, denoted as I(1). This result held even for the product wage. The null hypothesis assumed the existence of a deterministic linear trend. Given these results, the equations finally selected for inclusion in the main model (the employment, wage setting, and labor force equations) were tested for cointegration. For each equation, the hypothesis of cointegration among the included variables failed to be rejected, and the residuals were found to be stationary using ADF tests. The cointegrating vectors were also estimated using the Johansen procedure, and were found to have the same signs as those estimated for the main model. When the full set of variables was tested for the existence of cointegrating vectors using the Johansen procedure, the hypothesis that there existed six to seven cointegrating vectors failed to be rejected.

The main model was estimated using three different procedures: equation-by-equation using ordinary least squares (OLS); a seemingly unrelated regressions (SUR) technique; and a three-stage least squares technique (3SLS), which took into account estimated contemporaneous correlations between the errors (3SLS + SUR). OLS estimates are known to be at least consistent when variables are I(1) and cointegrated (see Chapter 1 in this volume for further discussion of these and related topics). Interestingly, for the final specification of the main model, different estimation methods yielded very similar results, provided that a full set of instruments was used in the 3SLS case.26 This result is probably related to the near diagonal nature of the residual variance-covariance matrix (which is reported in the main text). The results reported in the paper were obtained using the SUR procedure, which seemed somewhat advantageous from the computational point of view.

Finally, the individual equations of the main model were subjected to a series of tests additional to those reported in the main text, including further tests of serial correlation (LM tests with four lags), tests of normality (Jarque-Bera), and heteroscedasticity (ARCH tests with four lags). For each equation, the usual assumptions (no serial correlation, normality, and homoscedasticity) failed to be rejected.

Appendix IV: Inference Reliability in the Presence of Nonstationarity

Although it is clear that estimating relationships involving nonstationary variables is technically correct in the presence of cointegration—in the sense that the estimated coefficients converge to the correct value—it is less obvious that this is also true for the standard errors of these coefficient estimates. The literature on this topic provides some support that this will be so for many, but not necessarily all, of the coefficients. In addition, almost all distributional results from the cointegration literature rely on large sample approximations, which may or may not be applicable to the actual sample lengths available.

To address some of these concerns, a Monte Carlo simulation technique was employed to empirically calculate the distribution of the estimated coefficients. This included the mean and associated t-statistics, the minimum, median, and maximum, as well as various fractiles of the distribution. The specific technique used is known as the “bootstrap,” and works by resampling the estimated residuals a large number of times, shocking the model with the resampled residuals, and recomputing the variables of interest. This method is known to generate good approximations to the empirical distribution function of the estimated coefficients (or, indeed, to other statistical functions of interest).

The results of applying this technique are included in Table 7. The first entry, for example, indicates that the coefficient of the first employment lag in the employment equation was estimated as 0,556 using standard regression methods (in this case, the SUR technique), which also reported a t-statistic of 6.6. Using the bootstrap, the mean of the estimated coefficient is seen to be somewhat lower, at 0.4757, as is the corresponding t-statistic. However, the coefficient remains highly statistically significant, and in fact there is not a single replication among the 500 attempted that produced a coefficient estimate lower than 0.218.

Table 7.

Bootstrapped Coefficient Estimates versus Regression (SUR) Estimates

article image
article image
Notes: Five hundred replications. For definitions, see Table 1; see also Appendix II.

Overall, the bootstrapped t-statistics tend to be somewhat lower, but the correction is in no case so large as to overturn a conclusion of statistical significance reached using the regression-based estimates, except for some quarterly dummy coefficients, which have no policy significance. The one exception has to do with the effect of the real consumption wage on the labor force, whose statistical significance deteriorates to the point where the coefficient becomes suspect.

References

  • Bertola, Giuseppe, 1990, “Job Security, Employment, and Wages,” European Economic Review, Vol. 3 (June), pp. 85186.

  • Blanchard, Olivier, and Danny Quah, 1989, “Dynamic Effects of Aggregate Demand and Supply Disturbances,” American Economic Review, Vol. 79 (September), pp. 65573.

    • Search Google Scholar
    • Export Citation
  • Canziani, Patrizia, and Dimitri G. Demekas, 1995, “The Italian Public Pension System; Current Prospects and Reform Options,” IMF Working Paper 95/33 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation
  • de Luca, Loretta, and Michele Bruni, 1993, Unemployment and Labour Market Flexibility: Italy (Geneva: International Labor Office).

  • Demekas, Dimitri G., 1994, “Labor Market Institutions and Flexibility in Italy: A Critical Evaluation and Some International Comparisons,” IMF Working Paper 94/30 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fachin, Stefano, 1991, “Is the Employment Function Structurally Unstable? An Empirical Test: Italian Manufacturing Industry, 1970–1984,” Labor, Vol. 5 (Spring), pp. 175214.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ford, Robert, and Douglas Laxton, 1995, “World Public Debt and Real Interest Rates,” IMF Working Paper 95/50 (Washington: international Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jaramillo, Fidel, Fabio Schiantarelli, and Alessandro Sembenelli, 1991, “Are Adjustment Costs for Labor Asymmetric? An Econometric Test of Panel Data for Italy,” Review of Economics and Statistics, Vol. 75 (November), pp. 64048.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Karanassou, M., and D.J. Snower, 1994, The Sources of Unemployment Persistence and Responsiveness (unpublished; London: University of London, Birkbeck College).

  • Layard, P.R.G., and S.J. Nickell, 1987, “Unemployment in Britain,” in The Rise in Unemployment, ed. by C.R. Bean, P.R.G. Layard, and S.J. Nickell (New York: B. Blackwell).

    • Search Google Scholar
    • Export Citation
  • Lindbeck, Assar, and Dennis J. Snower, 1994, “How Are Product Demand Changes Transmitted to the Labour Market?” Economic Journal, Vol. 104 (March), pp. 38698.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Modigliani, F., and F. Padoa Schioppa, 1986, “Aggregate Unemployment in Italy, 1960–1983,” Economica, Vol. 53 (Supplement), pp. S245S273.

  • Organization for Economic Cooperation and Development, 1994, The OECD Jobs Study: Evidence and Explanations, Part I: Labour Market Trends and Underlying Forces of Change (Paris).

    • Search Google Scholar
    • Export Citation
  • Sims, Christopher A., 1980, “Macroeconomics and Reality,” Econometrica, Vol. 48 (January), pp. 149.

  • Snower, D.J., and Assar Lindbeck, 1990, “Demand-and Supply-Side Policies and Unemployment: Policy Implications of the Insider-Outsider Approach,” Scandinavian Journal of Economics, Vol. 92 (No. 2), pp. 279305.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Varian, Hal R., 1984, Microeconomic Analysis (New York: Norton, 2nd ed.).

Note: The author wishes to thank Dimitrios Demekas, Jeffrey Pranks, Karl Habermeier, Timothy Lane, Paolo Mauro, Thierry Pujol, Ramana Ramaswamy, Massimo Russo, Tessa van der Willigen. and, especially, Brian Henry. Marika Karanassou, Alessandro Leipold. Alessandro Prati, and Dennis Snower for many valuable discussions and insights. Gislene Jeffers and Valerie Pabst provided excellent publication support. Any remaining errors are the author’s responsibility.

1

Karanassou and Snower (1994) provide a fuller explanation of these sources and how they are expected to affect the lagged responses of the labor market.

2

The reasons for the regional disparities are complicated; they include historical developments, public policies, weaknesses in the political structures of the south, and proximity to the center of gravity of Europe. For a brief overview, see de Luca and Bruni (1993), pp. 28–32

3

Studies quoted by de Luca and Bruni (1993) put the productivity gap in industry between north and south at 75–80 percent.

4

The OECD jobs study (1994) and Demckas (1994) provide a wealth of additional institutional detail on the Italian labor market.

5

The replacement rate applicable to the WSF has varied very little throughout the sample period.

6

Appendix III provides more detail regarding the estimation methods used, as well as further diagnostic tests. The nonstationarity of the variables requires that the estimated relationships be tested for Cointegration. This was done, and all the relationships reported here pass this test: this implies that the coefficients are estimated consistently. The nonstationarity of the variables could raise some concerns regarding the correctness of the estimated standard errors, however. A Monte Carlo simulation study was performed, the results of which are included in Appendix IV. The regression-based t-statistics were found to be close to the simulated t-statistics. Thus, the conclusions reported in the text were not significantly affected by the nonstationarity of the variables.

7

Because data on social security contributions decomposed by employer and employee are not readily available, the SSC variable is an imperfect proxy. Nevertheless, it appears that SSC approximates the employer contribution portion of social security contributions closely enough to allow the negative coefficient on this variable to be adequately measured.

8

Lindbeck and Snower (1994) explore the issue of the transmission of demand changes to the labor market in greater detail. Snower, especially, has argued that the effects of GDP on employment should be small, or that the model should be expanded to incorporate nominal effects. This suggestion is not pursued here.

9

Layard and Nickell (1987), for example, report that labor force movements were relatively unimportant in explaining U.K. unemployment.

10

The adverse movements in competitiveness for Italy were reversed after 1984, so they no longer represent an important impediment to employment creation.

11

Modigliani and Padoa Schioppa (1986) argue that unemployment in Italy can be traced to two primary factors: real wage rigidity and an inability to raise aggregate demand on account of leakages—in effect, an external constraint. They argue that what is required is a coordinated increase in aggregate demand by industrial countries. Of course, this begs the question of whether demand policies can affect output in a sustained fashion, which is valid regardless of the presence or absence of coordination (although the presence of an external constraint does tend to make the deficiencies of demand-management policies obvious faster).

12

For example, estimates of coefficients for France are generally thought to contain significant breaks in the early to mid-1980s.

13

Fachin (1991) found that an employment equation for Italy during 1970–84 exhibited structurally unstable coefficients, while jaramillo, Schiantarelli, and Sembenelli (1991) found evidence consistent with nonconstant adjustment cost parameters. However, because of different samples and sample periods, these studies are not directly comparable to the results contained here.

14

For the employment equation, F(14, 44) = 1.79; for the wage equation, F(11, 50) = 1.61, and for the labor force equation, F(16, 40) - 0.92. The 95 percent critical value for the tests was 1.90 or higher.

15

The sensitivity of employment to output was also allowed to vary.

16

The t-statistic on the test is 1.74 (without regard to sign).

17

The production function coefficients were constrained to preserve constant returns to scale.

18

Of course, to the extent that measured unemployment is reduced by the discouraged-worker effect, the social welfare effects may be worse than they appear.

19

In the early part of the twentieth century, the discouraged-worker effect was also very strong, resulting in significant outward migration flows.

20

Chow’s predictive test is an F-test: for the employment equation, F(12,58) = 3.12 and for the labor force equation. F(12,56) = 4.11, compared with a critical value of about 1.9, For the wage equation, however, F(12,61) = 1.47.

21

This last result is surprising, because a number of observers interpreted developments during the last recession as implying an increase in the sensitivity of output to employment. The regressions seem to indicate instead that firms continued to shed labor even as output rebounded.

22

The forecasts reported here use information available as of April 1995.

23

It is not yet possible to judge whether productivity gains have run their course. One can conceive of a realistic scenario whereby firms continue to shed labor, which would weaken output growth through the production function.

24

Tinkering with a dynamic model by changing only a few parameters will in general result in outcomes that do not make economic sense. The simulation reported here also required level shifts in the employment and labor force equations that were, however, much smaller than the estimated changes.

25

The Bank of Italy’s revised series were spliced with the earlier series so as to obtain consistent time series back to 1975.

26

And variables such as wages and output were instrumented using lagged wages and output, respectively. This procedure should successfully purge the biasing effects of endogene-ity, provided the error terms were serially uncorrected. Tests failed to reject the hypothesis of no residual serial correlation.