3.1 Equations

Theories of unemployment may be divided into two broad groups, depending on whether they are based on stocks or flows. The specification of the labor market adopted in this paper focuses on flows. It builds on the standard matching model developed by Pissarides (2000). Suppose that the size of the labor force is constant¹. We let u_t and v_t at time t denote the unemployment and vacancy rates, both expressed as percent of the labor force. The number of job matches taking place per unit of time is assumed to follow a Cobb-Douglas specification of the form

where µ denotes the degree of efficiency in the matching technology and ρ ε (0, 1) represents the unemployment elasticity of job matches. The job vacancies and unemployed workers that are matched at any time t are randomly selected from the sets u_t and v_t. Hence the hazard rate out of unemployment follows a Poisson process with rate

where θ_t = v_t/u_t is the number of vacancies per unemployed workers or labor market tightness. This generates an exogenous hazard rate into employment equal to f_tu_t.

The flow into unemployment results from job-specific productivity shocks arriving to firms at Poisson rate s_t. When the shock arrives, the net product of the job changes to some new value drawn from a general probability distribution G (x_t), where 0 ≤ x_t ≤ 1 and w (x_t) stand for the job’s productivity and its corresponding wage. Let us denote by J (x_t) the firm’s expected profit from an occupied job. The firm chooses a reservation productivity R_t (with J(R_t) = 0) and destroys jobs whose productivity fall below it because of the shock. In words, job destruction operates when the reservation productivity is such that the expected profit from an occupied job is 0. This generates an endogenous hazard rate into unemployment equal to the product of the fraction of firms hit by a shock s_t, and the probability that a shock is below reservation G(R_t), with associated job destruction of s_tG(R_t)(1 − u_t). The out-of-steady-state unemployment dynamics is therefore given by

At equilibrium the unemployment rate is constant, thus

with the interpretation that the mean number of workers who enter unemployment is equal to the mean number of workers who leave unemployment. Put another way,

which implies that, at equilibrium, the expected duration of unemployment is equal to the expected duration employment. Substituting (2) into (4) and solving for the u yields

which tells us that the equilibrium unemployment rate is the product of the hazard rate into unemployment sG(R) times the expected duration of unemployment $\left(1-u\right)\frac{1}{\mu {\theta }^{1-\rho }}$.

Solving the model requires obtaining the equilibrium conditions for the two endogenous R and θ, to which we now turn. Suppose that job creation satisfies the zero-profit condition², that job destruction operates when J(R) = 0, and that the wage sharing rule is set in a Nash bargaining fashion. Let β ε (0, 1) be the labor’s relative strength in wage bargaining, z the unemployment income, c the hiring and firing costs of firms, κ an indicator of the regulatory stance in the market, tw the tax wedge, and r the rental cost of capital. One can then express labor market tightness and reservation productivity (characterizing job creation and job destruction respectively) as a function of the exogenous variables in the model, that is

It is convenient to represent this system in the space (R, θ) while viewing the remaining parameters as constant (Figure 3). The model behaves according to the following logic:

• Expression (7) on job creation corresponds to the free-entry condition postulated by search models and reflects that firms post vacancies as long as the marginal recruiting cost per vacancy is equal to the expected value of holding it. At equilibrium, the expected gain from a new job to the firm must be equal to the expected hiring cost paid by the firm. It slopes down in the space (R, θ) because at higher reservation productivity, the expected life of (and gain from) a job is reduced. Firms create fewer jobs as a result, leading to a fall in market tightness θ.

• Expression (8) on job destruction slopes up in the space (R, θ) because at higher the worker’s outside opportunities are better and wages are higher. Higher wages translate into a lower net product of labor. Thus more marginal jobs are destroyed. A noteworthy property implicit in the equilibrium condition (8) is that the reservation productivity R is less than the reservation wage w (R). This implies some labor hoarding in equilibrium: because job productivity might change, the firm keeps some currently unprofitable jobs occupied.

View Full Size

Model’s predictions

Citation: IMF Working Papers 2012, 054; 10.5089/9781463937140.001.A001

Source: The authors.

• Download Figure
• Download figure as PowerPoint slide

3.2 Model’s predictions

Comparative statics implied by equations (7) and (8) is as follows (Figure 3):

• At given θ,

  $\begin{array}{cc}\frac{dR}{d\mathit{\beta }}>0,\frac{dR}{dz}>0,\frac{dR}{dtw}>0,\frac{dR}{dk}>0,\frac{dR}{dc}<0,\frac{dR}{ds}\lessgtr 0,\frac{dR}{dr}>0& \left(9\right)\end{array}$

• At given R,

  $\begin{array}{cc}\frac{d\theta }{d\mathit{\beta }}<0,\frac{d\theta }{dz}<0,\frac{d\theta }{dc}<0,\frac{d\theta }{dtw}<0,\frac{d\theta }{dk}<0,\frac{d\theta }{ds}<0,\frac{d\theta }{dr}<0,\frac{d\theta }{d\mu }<0& \left(10\right)\end{array}$

Consider now the influence of parameter changes on the job destruction and job creation equilibrium values, the latter being inversely related to our second variable of interest, the expected duration of unemployment. Bearing in mind that unemployment will change only if the new flows implied by the change in R and θ are not equal, the model’s predictions can be summarized as follows:

• Both a higher unemployment income z and a higher labor share in bargaining β raise reservation wages. This reduces the expected gain from a job, shrinking job creation (increasing duration) and increasing job destruction. As a result, the unemployment rate will be higher in the new equilibrium.

• An increase in the tax wedge tw makes leisure more attractive relative to work, putting upward pressure on bargained wages. This renders marginal workers unprofitable and, all else being equal, more jobs are destroyed. At the same time, higher wages reduce the expected value from job matches to the firm, thereby inhibiting their recruiting activities. Higher job destruction and lower job creation therefore lead higher unemployment rates.

• Stricter regulation κ tends to lower labor productivity in equilibrium, hence dampening job creation. A deterioration in competition-restraining regulations conveys higher mark-ups to producers, who will now seek to sell their goods at higher prices. As there is excess supply at the new prices, some unprofitable firms will exit the market and jobs will be destroyed until a new equilibrium is reached characterized by lower production and employment levels. Overall, the equilibrium unemployment rate is hypothesized to increase following a deterioration in the regulatory conditions.

• Job creation increases (duration falls) if productivity changes more frequently (higher s). Job destruction is subject to two opposing forces. On the one hand, job destruction increases because there are more shocks on average, but on the other hand, it decreases because firms are now more willing to hold on to labor if they expect a quick arrival of better conditions. Thus the effect of increased s on unemployment is ambiguous.

• Higher interest rate r reduces job creation (increases duration) as future profits on new jobs are discounted more heavily. At the same time, firms respond to higher interest rates by destroying more jobs. This is so because the option value of keeping unprofitable jobs is lower, given that the returns from a productivity change only accrue in the future. Less job creation and more job destruction combine into higher unemployment rates.

• By reducing the expected duration of vacancies, higher matching efficiency µ reduces hiring costs for firms, increasing job creation (reducing duration) and bringing down the unemployment rate.

• Higher hiring and firing costs c lower the expected gain from a job and reduce job creation. Likewise, larger firing costs make firms more conservative in their firing decisions and limit job destruction. This makes for a more stagnant labor market, but the effect on lower inflows and higher durations on equilibrium is itself ambiguous.

4.1 Data and Empirical Strategy

In order to assess the impact of labor policies on job destruction rates and the duration of unemployment spells, we estimate empirical versions of (7) and (8) of the form:

Equations (11) and (12) control for out-of-steady-state dynamics by entering the output gap as an indicator of cyclical position in the regressions. In our notation t, i and j are indexes for time, countries and institutions, and x represents a vector of institutions. The specification allows for country fixed and common time effects. The two dependent variables are extracted from available data on unemployment rates following the methodology described in the Appendix.

All institutional variables in the model are normalized to have zero mean and unit standard deviation. They are mapped into empirical concepts in the way we turn to describe. Reservation wages z are proxied by benefit replacement rates during the first year of unemployment, as averaged over various family types and earning levels (Nickell and others, 2001; and OECD, labor market policies database). Firms’ firing costs c are approximated by the OECD indicators of Employment Protection Legislation for regular workers (EPL) and the incidence of temporary employment.

The matching efficiency is supposed to be influenced by three categories of Active Labor Market Policies (ALMPs). Given that heterogeneity in the synthetic indicator of ALMPs is one reason for non-significance of these policies in previous studies (Bassanini and Duval, 2006), we proceed with a disaggregated analysis whereby ALMP spending is decomposed into three main categories: job seeker support by Public Employment Services (PES), training policies and financial support to labor-demand. Data on ALMPs programs is also taken from the OECD labor market policies database. As suggested by Estevao (2003), we express ALMP expenditures per unemployed person as a percentage of GDP per capita to ensure cross-country comparability.

We use union centralization and coordination measures (Visser, 2007) to calibrate workers’ strength (β) in wage bargaining. Tax wedges comprise social security contributions, personal income and indirect tax rates (Nickell and others, 2001; and OECD, labor market policies database).

Estimating the model further requires identifying measures for productivity shocks and overall financial conditions, which we take to be the standard deviation of labor productivity³ and the real interest rate. As a measure of the regulatory stance in product markets, we use the OECD indicator of regulation in Energy, Transport and Communications (ETRC). This is narrower in terms of sectoral coverage than the OECD aggregate indicator for product market regulation but has the advantage of being available in long-time series. Moreover, as the EU Single Market Program has been more effective in liberalizing traditionally monopolistic sectors than in opening up services in general, the potential for reform presented in the following section can be interpreted as a lower bound to overall reform gains.

4.2 Results

We estimate the two-equation system on job destruction and unemployment durations given by (11) and (12) using data from the 20 OECD countries listed in Section 2 over the period 1985-2008. The Simultaneous Equations Model (SEM) estimator we use assumes that the two unemployment components are determined simultaneously by the model postulated in section 3. SEM further allows for cross-correlation in the error terms of the two variables of interest. Results are summarized in Table 1.

Table 1:

Regression Results

SEM Regression: Job Destruction Rate and Unemployment Duration

Note: The sample includes data for 20 OECD countries listed in Section 2. The estimation includes cross-country fixed effects (not displayed) and year dummies as common time effects. Dependent variables are in logs. The t-statistics are reported in parenthesis. Superscripts *, ** and *** indicate that the estimated coefficient is significantly different from zero at 10, 5 and 1 percent levels, respectively.

Table 1:

Regression Results

SEM Regression: Job Destruction Rate and Unemployment Duration

	Job Destruction Rate (% of Labor Force)	Unemployment Duration (Months)
Output Gap	-0.0522***	-0.0201***
	[-6.036]	[-2.829]
EPL	-0.1363**	0.0589*
	[-2.449]	[1.712]
EPL*Temporary Employment	0.0630**	0.0132
	[2.493]	[0.640]
ALMP
-Job Seeker Support	-0.0241	-0.1052*
	[-0.335]	[-1.789]
-Training	-0.0421	0.2043***
	[-0.436]	[2.581]
-Labor Demand Support	0.0791	-0.0573
	[1.064]	[-0.940]
Tax Wedge	0.1338***	0.1028***
	[3.664]	[3.437]
Output Gap*Tax Wedge	0.0055*	0.0017
	[1.796]	[0.294]
Product Market Regulation	0.1191**	0.0358
	[2.403]	[-0.881]
Replacement Rate	0.0624*	0.0705*
	[1.716]	[1.813]
Bargaining Coordination	-0.0085	-0.0396
	[-0.244]	[-1.379]
Real Interest Rate	0.0082	0.0897***
	[0.295]	[3.924]
Productivity Volatility	0.0042	-0.0274*
	[0.234]	[-1.862]
Total Observations	324	324
Adjusted R-squared	0.851	0.9

Note: The sample includes data for 20 OECD countries listed in Section 2. The estimation includes cross-country fixed effects (not displayed) and year dummies as common time effects. Dependent variables are in logs. The t-statistics are reported in parenthesis. Superscripts *, ** and *** indicate that the estimated coefficient is significantly different from zero at 10, 5 and 1 percent levels, respectively.

View Table

Table 1:

Regression Results

SEM Regression: Job Destruction Rate and Unemployment Duration

	Job Destruction Rate (% of Labor Force)	Unemployment Duration (Months)
Output Gap	-0.0522***	-0.0201***
	[-6.036]	[-2.829]
EPL	-0.1363**	0.0589*
	[-2.449]	[1.712]
EPL*Temporary Employment	0.0630**	0.0132
	[2.493]	[0.640]
ALMP
-Job Seeker Support	-0.0241	-0.1052*
	[-0.335]	[-1.789]
-Training	-0.0421	0.2043***
	[-0.436]	[2.581]
-Labor Demand Support	0.0791	-0.0573
	[1.064]	[-0.940]
Tax Wedge	0.1338***	0.1028***
	[3.664]	[3.437]
Output Gap*Tax Wedge	0.0055*	0.0017
	[1.796]	[0.294]
Product Market Regulation	0.1191**	0.0358
	[2.403]	[-0.881]
Replacement Rate	0.0624*	0.0705*
	[1.716]	[1.813]
Bargaining Coordination	-0.0085	-0.0396
	[-0.244]	[-1.379]
Real Interest Rate	0.0082	0.0897***
	[0.295]	[3.924]
Productivity Volatility	0.0042	-0.0274*
	[0.234]	[-1.862]
Total Observations	324	324
Adjusted R-squared	0.851	0.9

Note: The sample includes data for 20 OECD countries listed in Section 2. The estimation includes cross-country fixed effects (not displayed) and year dummies as common time effects. Dependent variables are in logs. The t-statistics are reported in parenthesis. Superscripts *, ** and *** indicate that the estimated coefficient is significantly different from zero at 10, 5 and 1 percent levels, respectively.

As expected, unemployment inflows and durations are strongly affected by cyclical conditions. At times where the economy operates above potential output and labor markets are tighter, both job losses and the length of unemployment spells are reduced. The opposite holds during downturns, characterized by negative output gaps, slack labor markets, and higher unemployment inflows and durations.

For a right interpretation of the link between the set of institutional factors and the two endogenous variables, note that, by construction, the coefficients presented in Table 1 capture the estimated effect of each policy on unemployment inflows and durations when the economy is producing at full capacity (i.e. zero output gap).

Our estimates for job destruction suggest strong correlations with the tax wedge, EPL (alone and interacted with the incidence of temporary employment) and product market regulation, which we turn to rationalize.

By increasing the price of leisure relative to working activities, a higher tax wedge pushes up workers’ reservation wage. Workers at the margin are rendered unprofitable to firms and more jobs are destroyed. There is some evidence that this impact depends on the economic cycle, as indicated by the significance of the coefficient of the interaction term between the tax wedge and the output gap. A positive sign implies that the effect of the tax wedge on job destruction is augmented in periods when labor markets are tight (when the output gap is positive) and less marked in periods with considerable labor market slack (when the output gap is negative). This is consistent with workers attaching greater value to their leisure time during upturns as compared to downturns.

In keeping with previous findings in the literature, we find that stricter job protection rules tend to reduce job losses. However, as indicated by the positive sign of the interaction term between EPL and temporary employment, the dampening effect of EPL on job destruction is reduced in economies characterized by a high share of temporary jobs–where labor legislation tends to differ across permanent and temporary workers.

Hurdles to competition in the services sectors result into higher equilibrium levels of job destruction. This is because regulations that inhibit competition grant more market power to producers, each one charging now higher prices for the goods they sell. But, as aggregate demand is less than supply at the new prices, a number of unprofitable firms will exit the market. And more jobs will be destroyed. Thus stricter regulations lead to market configurations characterized by higher job destruction and lower employment levels.

Consistent with our theoretical priors, ALMP programs (our proxy to matching efficiency) do not appear to have a significant impact on job destruction. Neither could we find evidence of a significant relationship between bargaining coordination and the inflow rate. Using bargaining centralization to measure the capacity of the social dialogue to internalize the impact of wage demands on employment destruction preserved this result. And the coefficients of non-linear specifications testing the Calmfors-Driffill hypothesis turned out to be also non-significant.

Unlike direct indicators of wage bargaining, more generous replacement rates do seem to push up reservation wages and aggravate job destruction. Our estimates could not shed light on the theoretical ambiguity surrounding the influence of productivity changes on job destruction.

Turning to the duration of unemployment, high tax wedges appear to discourage job creation and prolong unemployment spells. By increasing equilibrium wages, higher tax wedges tend to depress labor demand and increase the flow of workers into unemployment. Likewise, the unemployment benefit replacement rate is found to have a positive impact on unemployment duration. This is likely to capture the negative impact of the replacement rate on search incentives, as stressed in earlier works (Bassanini and Duval, 2006, Blanchard and Wolfers, 1999). EPL also has a positive, albeit weakly significant, influence on unemployment duration.

Labor-demand support does not appear to influence unemployment durations while job-seeker support is only borderline significant. The former result is consistent with previous studies pointing to large dead-weight losses and substitution effects associated with employment subsidies. These studies have often been disappointing in terms of bringing the unemployed back into unsubsidized work (see, for instance, Martin and Grubb 2001).

In contrast with both theoretical priors and economic intuition, more spending on formation policies tends to result in longer unemployment spells. This could partly reflect reverse causality, as those programs can be adjusted more rapidly in response to an economic downturn and longer unemployment spells than, say, EPL or tax wedges. Moreover, training programs may reduce search efforts if not properly designed, and in the case where participation in such programs represents a more attractive alternative to workers than open unemployment, they could even augment wage demands. In any event, any beneficial effects of ALMPs need to be weighed against the costs of taxes required to finance them, which may in turn increase unemployment.

Our estimates confirm the theoretical prediction that job creation increases (duration falls) in response to more frequent changes in productivity while the time spent in looking for a job appears to be inversely related with real interest rates.

One interesting finding from our analysis is that, with the notable exception of EPL and ALMPs, all remaining labor market institutions seem to impact the job destruction rate and the unemployment duration in the same direction. Countries with both high inflows into unemployment and long durations may therefore find efficient to use the policies that affect both dimensions at the same time.

4.3 Policy Implications from Some Illustrative Simulations

To get an idea of the model’s projection capacity of both inflows and durations, we apply the coefficients estimated in Table 1 to output gap forecasts over the period 2010-12⁴. These estimates are based on the most updated (i.e. 2008) institutional configurations in each country. The projection exercise is meant to illustrate the link between the model’s institutional variables and the two unemployment components, rather than to provide accurate forecasts over the referred period. The projections presented in this section should therefore be taken with caution.

Figure 4 ranks countries according to their expected behavior in terms of inflows and durations. Projections reflect a wide variation in countries’ experiences. The duration of unemployment spells ranges from around 26 months in Greece to 5.6 in Norway. Belgium, Ireland, Italy and Portugal also feature comparatively long durations. The shortest unemployment spells are projected for the Nordic and Anglo-Saxon countries. Job destruction rates are envisaged to be the highest in countries as heterogeneous as Canada, Finland, Spain, and the US; and the lowest in Portugal, Netherlands, Belgium, Belgium and Switzerland.

View Full Size

Job Destruction Rates and Unemployment Duration: Baseline Projections

Citation: IMF Working Papers 2012, 054; 10.5089/9781463937140.001.A001

Source: Authors’ Calculations.1/ Projections are based on the estimated coefficients reflected in Table 1, 2010-2012 output gap data and 2008 institutional configurations in each country.

• Download Figure
• Download figure as PowerPoint slide

We now turn to calibrate the impact of changes in institutional variables on unemployment inflows and durations. To this aim, we quantify the (percentage point) difference from baseline projections induced by i) a move to best sample practices in services regulation; ii) a 5-percentage-point reduction in tax wedges; iii) a 5-percentage-point reduction in replacement rates; iv) a simulation combining all the above changes. Rather than as a literal description of what every country ought to do, the simulations presented here are meant to illustrate the scope for less employment destruction and lower unemployment duration, as well as to generate a benchmark against which to make cross-country comparisons. Clearly, countries need not to adopt such reform package in full, but may target specific policy areas depending on the nature of the unemployment problem and decide on the pace of reform that best suits their needs.

The results are indicative of substantial gains from reform (Figure 5). Our simulations imply that removing the hurdles to entering key services and alleviating the tax burden on labor can each reduce job destruction rates permanently by 0.05 percentage points on average for the whole sample. And by stimulating job search, lower replacement rates and tax wedges can limit the duration of unemployment spells up to 20 and 50 days, respectively. The combined impact of those changes would yield sizable unemployment reductions in all countries, with Greece, Spain, France, Finland, Ireland, Belgium, and Italy benefiting the most from the implementation of the combined package (Figure 6). Of course, actual unemployment impact will vary with the ambition of the reform agenda, the speed of its implementation, and the time needed for these reforms to take hold.

View Full Size

Reform Impact on Job Destruction Rates and Unemployment Duration

Percentage Points Reduction from Baseline Projections 2010-12 ^1/

Citation: IMF Working Papers 2012, 054; 10.5089/9781463937140.001.A001

Source: Authors’ Calculations.^1/ The heatmap illustrates changes in job destruction rates and unemployment durations assuming that (i) every country adjusts PMR to best sample practices; (ii) replacement rates are cut down by 5 percentage points across-the-board.2/ Percentage point reduction for job destruction rates and reduction in days for unemployment duration.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Combined Reform Impact on Unemployment Rates

Percentage Points Reduction from Baseline Projections 2010-12 ^1/

Citation: IMF Working Papers 2012, 054; 10.5089/9781463937140.001.A001

Source: Authors’ Calculations.^1/ The simulation illustrates the scope for reform by assuming that tax wedges and benefit replacement rates are each reduced by 5 percentage points in every country; and PMR is adjusted to match best sample practices.

• Download Figure
• Download figure as PowerPoint slide

These reforms could go hand in hand with reducing inequalities. In countries where labor market duality remains unacceptably high, measures to harmonize employment protection benefits between types of job contracts should reduce the disproportionate burden on temporary workers—those last hired and first fired. Our simulations suggest that a reduction in job protection rules for regular workers to cover only ¼ of the distance with the least regulated economies would have a neutral impact on unemployment, provided it is accompanied by a reduction in temporary unemployment to match best practices.