1 The Dynamics of European Unemployment
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
  • | 2 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Abstract

This Book Examines how the dynamic interactions among labor market activities—employment, wage setting, and labor force participation—have contributed to the high levels of unemployment in five European countries—France, Germany, Italy, Spain, and the United Kingdom. It also investigates how labor market policies have affected the resilience of these countries’ labor markets.

This Book Examines how the dynamic interactions among labor market activities—employment, wage setting, and labor force participation—have contributed to the high levels of unemployment in five European countries—France, Germany, Italy, Spain, and the United Kingdom. It also investigates how labor market policies have affected the resilience of these countries’ labor markets.

This approach differs substantially in emphasis from the dominant way of analyzing unemployment nowadays, which concentrates on long-term labor market equilibrium. According to this latter approach, movements of unemployment over time may be understood as random variations around a reasonably stable long-run rate, the “nonaccelerating inflation rate of unemployment” (NAIRU), or the “natural rate of unemployment.” The major policy implication of this theory is that, barring some temporary policy shocks that may contribute to the random variations of unemployment, the effectiveness of labor market measures depends entirely on how they influence unemployment in the long run.

What motivates the book is the view that, although the concept of a NAIRU has helped identify many important determinants of unemployment, it does not capture an important European phenomenon, namely, those mechanisms in many European labor markets that prevent employers and employees from adjusting promptly to changing market opportunities and that keep these labor markets from recovering quickly after recessions. It is well known that unemployment is a serious problem in Europe not just because it rises in times of recession, but also because it does not fall readily when the recessions are over. It is this aspect of European unemployment that is the focus of the present study.

The book explores the labor market implications of lagged responses of employers, employees, the unemployed, and the inactive to new economic circumstances. In the countries investigated, these lagged responses—in employment, wage setting, and labor force participation—can make unemployment diverge from its long-run rate for substantial periods of time. The policy implication is that labor market measures not only change unemployment in the long run (as highlighted by equilibrium theory), but also affect the process whereby unemployment approaches its long-run rate. This book suggests that unemployment rates may diverge significantly from their long-run values for a decade or more, and it implies that policies that improve the flexibility of labor markets can play an important role in tackling the European unemployment problem, even if they are ineffective in the long run.

The purpose of this book is to give a more balanced appraisal of the sources of European unemployment. In addition to oil prices, interest rates, taxes, and unemployment benefits, which have tended to raise the long-run unemployment rate, there are important dynamic features of unemployment—such as inertia and overshooting—that have contributed significantly to the European unemployment problem over the past 25 years.

The five countries studied in this book are analyzed in the same methodological way, using a broadly common model as the starting point. The resulting empirical models of course differ between countries, but by following a generally common methodology in each country, the differences can be attributed to underlying differences in country behavior. The empirical results for each of the countries are reported in Chapters 26 of this book. This chapter presents the reasoning underlying the study and the methods used and gives a short overview of the results of the countries studied.

European Unemployment Problem

In the European Union (EU) as constituted in 1994, unemployment rose rapidly from about 3 percent in the early 1970s to reach about 6 percent at the end of the 1970s, reaching a local peak of over 10 percent in the mid-1980s before falling to a little under 9 percent at the end of the 1980s. By 1994, it had risen again to stand at just under 12 percent. In this respect, its behavior contrasts with that in the United States, where unemployment also rose rapidly in the 1980s, but has since declined to levels seen at the end of the 1970s (see Chart 1). But it would be wrong to assume that the pattern of unemployment changes in Europe was homogeneous. Austria, Switzerland, and the Nordic countries maintained unemployment at relatively low levels over much of the 1970s and 1980s. Even within the EU, unemployment experiences differed substantially during the 1980s. In the United Kingdom, unemployment fell from about 11 percent in the early 1980s to about 6 percent toward the end of the 1980s. In west Germany, unemployment remained more stable over the same period, declining to about 5.5 percent at the beginning of the 1990s from its peak of 8 percent in the early to mid-1980s. The reduction in unemployment in France (where it fell from 11 percent to 9 percent) was similarly relatively small as it was in Italy (unemployment there only falling from 12 percent to 9.5 percent). Experiences in Spain were rather different, with unemployment peaking at 21 percent in 1985, before falling to 15.5 percent at the beginning of the 1990s. Subsequently, unemployment rose again in each of these countries as recession took hold, falling noticeably more recently only in the United Kingdom, where recovery predated that in the rest of the countries reviewed here.

Chart 1.
Chart 1.

Unemployment in the European Union and the Untied States

(In percent)

Since unemployment is the difference between the labor force and employment, cross-country differences in unemployment movements can, by definition, be attributed to intercountry differences in employment creation and labor force adjustment. In the United States, for example, employment grew on average by 1.6 percent a year during the 1980s, compared with 0.5 percent in the EU (see Organization for Economic Cooperation and Development (1994)). But intercountry differences in the degree to which the labor force responds to employment opportunities have also affected unemployment patterns over the business cycle. In Europe, movements of participation rates have differed markedly: France, Spain, and, until quite recently, Germany have had steady increases in the overall labor force, owing in part to increases in female participation rates. The rapid increase in the participation rate for women in the United Kingdom from the end of the 1980s may have contributed to increased unemployment during this period. More recently, declining U.K. participation appears to account for the more rapid onset of falling unemployment in this recovery when compared with previous ones. In Italy, the labor force has declined rapidly during the recession, following substantial increases in female participation in the 1970s and 1980s.

These contrasts have been commented upon at length.1 In much of this discussion, considerable emphasis has been given to the purported “inflexibility” of European labor markets compared with the U.S. market as the explanation of these broad continental differences (see, for example, International Monetary Fund (1994)). This book takes a more precise look at the phenomenon of inflexibility by identifying the factors that affect the process whereby unemployment adjusts to the long-run equilibrium rate over time.

Considerable attention has been lavished on the role of structural factors, such as the restrictive practices of European trade unions, the relative generosity of European unemployment benefits, and the regulation of European product markets, in bringing about high equilibrium unemployment rates. Relatively little attention, however, has thus far been devoted to the ability of labor markets to respond readily to changing economic circumstances in the aftermath of recessions. This book is a step toward righting the balance, by focusing not only on the structural deterioration of European labor markets (as in Layard, Nickell, and Jackman (1991)), but also on the determinants of sluggish labor market adjustment.

Thus far, the literature on unemployment has concentrated on two polar extremes of unemployment dynamics: the natural rate and hysteresis models. According to the natural rate models, unemployment can be divided into two major components: (1) the long-run equilibrium rate of unemployment, which is the unemployment rate determined by the structural characteristics of the economy: technologies, endowments, preferences, and the degree of competition; and (2) random short-run deviations from that rate caused by temporary frictions.

In contrast, the hysteresis hypothesis asserts that the unemployment rate tends to “get stuck” at whatever its current rate happens to be. (More precisely, the current unemployment rate is asserted to be the best predictor of the future unemployment rate.) Here every unemployment rate can be a long-run rate and the effects of temporary labor market shocks persist indefinitely.

We will argue that neither of these hypotheses squares well with the central features of European unemployment experience. On the one hand, European unemployment rates can diverge from their long-run “natural” rates for substantial periods of time; on the other, unemployment rates show too great a tendency to return to a limited range of values to be explained by pure hysteresis. The argument we advance is that adjustment dynamics are a neglected dimension of the unemployment problem. Following Snower and Karanassou (1995), this chapter contends that the sluggishness with which the labor market responds to shocks can be a major force keeping unemployment high following a shock, and this tendency often persists for long periods of time. To improve one’s understanding of the importance of labor market dynamics and to design appropriate policy responses, one must explore how the major lags in labor market activities arise, identify how they interact with each other, and establish how labor market policies may affect them.

The model used in the individual country studies to advance this argument is grounded fully in real terms and thus focuses exclusively on the dynamics of the labor market following real shocks. An important question that the chapter ignores, therefore, is the role of nominal-real interactions, where nominal shocks may have long-lasting real effects owing to sluggish adjustment of wages, prices, or other nominal variables. This source of labor market sluggishness lies beyond the scope of this study, as does the possibility that labor markets may not respond symmetrically to favorable and unfavorable shocks.2

This chapter builds on the theoretical and empirical research in labor economics, which suggests that there are significant lagged responses in various labor market decisions. Among the most important examples of “within equation” lagged responses are the presence of lagged employment responses attributable to employment adjustment costs, wage staggering effects attributable to the presence of overlapping wage contracts, insider-outsider effects in wage setting, and labor force lags attributable to costs of entering the labor force. The model of employment, real wages, and the labor force in this chapter covers these and other dynamic effects.

Present Approach Compared with Some Alternatives

Much recent discussion about the source of high unemployment in Europe has emphasized the long-run equilibrium properties of the labor market, leading, arguably, to a lack of recognition of the importance of the adjustment of the labor market when subject to shocks. Those who pursue long-run equilibrium in the labor market have also tended to draw attention to the supply side of the labor market as being the most likely area where the explanation of high unemployment is to be found. Hence “structural” supply-side changes play a major role in this proposed explanation. The structuralist interpretation emphasizes the notion that the higher rates of unemployment actually observed are largely due to changes in its long-run equilibrium rate. In other words, the steady state or nontransitory part of unemployment has risen because of worsening structural factors, such as increasing trade union power, increases in out-of-work benefits, or an increasing dislocation—or mismatch—of unemployed workers to available jobs. A prominent version of this view is the NAIRU equilibrium, where the long-run equilibrium unemployment rate is that achieved in the absence of misperceptions and adjustment costs. Deviations of unemployment from its NAIRU are generally explained in terms of errors in wage-price expectations or intertemporal substitution.

The NAIRU theory leads to a clear-cut diagnosis of European unemployment. Since the long climb in European unemployment over the past two decades cannot realistically be ascribed to growing wage-price misperceptions or to long-term intertemporal substitution of leisure for work, the theory implies that it must be due to changes in the factors determining the long-run equilibrium rate, that is, to a structural worsening of the sort noted earlier.

One of the main reasons for adopting an approach different from the NAIRU is that, over the past two decades, this hypothesis has found it increasingly difficult to explain the movements of European unemployment. As noted in Chapter 5, estimates of the NAIRU for the United Kingdom are surprisingly variable over time. Chapter 6 questions whether the NAIRU in Spain can be as high as the 18 percent at which some recent estimates place it. The European unemployment rate increased markedly with the first oil price shock of 1973 and remained at unprecedentedly high postwar levels until the second oil price shock of 1979, when it rose to yet higher levels. The level of unemployment has gone up again following the 1990s recession. It has been difficult, if not impossible, to find reasons why the NAIRU should have risen by so much during this period. Thus, the model does not seem to provide a satisfactory empirical explanation for these changes in unemployment in European countries when the NAIRU is treated as the sole explanation of high unemployment. This is simply because the major variables used to account for the changes in long-run equilibrium unemployment, such as union power, changes in replacement ratios (working through an effect on the reservation wage), taxes on employers and employees, and mismatch, do not move enough to provide a satisfactory quantitative account of the trend increase in unemployment that has actually occurred.3 The large demographic changes that could have raised the NAIRU—the increase in female labor force participation and the influx of young people into the labor force—did not occur during the time of largest unemployment increases. The important institutional changes pulling in the same direction—the expansion of the unemployment benefit systems and associated welfare state benefits and the rise of union power—occurred, for the most part, in the 1960s. Extremely long lags in people’s adjustment to institutional changes, combined with significant overshooting behavior, would be required to rationalize the upward climb of European unemployment between the mid-1970s and mid-1980s for these to be an explanation.4

It has been increasingly recognized that the persistence of the labor market is an important aspect of its behavior. First, in some labor markets it is becoming evident that the effects of temporary shocks may last for years. Second, it is clear from recent international comparisons that different economies have differing degrees of persistence (see, for example, Barrell, Morgan, and Pain (1995) and Nickell (1995)). These features point to the need to account for not only the presence of such inertia in the labor market, but also for the degree of persistence in any national labor market in terms of its institutional and other features, such as the degree of centrality of its bargaining structure (Calmfors and Driffill (1987)) and the importance of overlapping in wage contracts in different sectors (or bargaining units) in the economy. In this light, a rigid distinction between structuralist and persistence explanations of rising unemployment is difficult to sustain. For example, if bargaining structure affects persistence, as it appears to, this is evidently a “structural” factor, which affects both the persistence and the long-run equilibrium level of unemployment.

Furthermore, international differences in persistence imply that it is crucial to investigate persistence mechanisms using behavioral models of the labor market rather than to simply rely on a time-series analysis of the unemployment rate itself. It is sometimes concluded that, when relating unemployment to its lagged values, the coefficients in the autoregression estimate the degree of persistence. (See Blanchard and Summers (1986) and Alogoskoufis and Manning (1988) for examples of this.) But this technique almost certainly leads to an underestimate of persistence, because considering unemployment alone does not allow for the possible influence on it of other variables.5

At the opposite extreme from the NAIRU in the literature on unemployment dynamics is the hysteresis approach, whereby every unemployment rate that the labor market produces is a long-run equilibrium rate. An adverse temporary labor demand shock, such as an oil price increase or an exchange rate appreciation, simply pushes the economy from one long-run equilibrium unemployment rate to a new, higher one. Consequently, temporary shocks have permanent unemployment effects.

The hysteresis approach, like the NAIRU, is also at variance with some important facts of unemployment experience in the United States, Europe, and elsewhere. Most important, if unemployment were to depend on its past value with a unit root plus a random error, then the variance of unemployment would clearly be unbounded. This means that unemployment would eventually hit zero or 100 percent with certainty. There is, needless to say, no evidence of such behavior anywhere. In practice, unemployment in the member countries of the Organization for Economic Cooperation and Development (OECD) tends to return repeatedly to values lying within a narrow range of between, say, 2 percent and 12 percent over the long run. The hysteresis approach is unable to account for this tendency.

The approach here builds on these considerations and postulates that, although a long-run equilibrium value of unemployment exists, the actual unemployment rate may deviate from it for substantial periods of time. In practice, it must be recognized that it is difficult to distinguish between protracted persistence and the case where unemployment is responding to permanent changes in structural factors, Z (the basis of NAIRU models). Models built around a long-run equilibrium view attribute all permanent changes in unemployment to the permanent components of Z and serial correlation in unemployment to serial correlation in these exogenous variables. In this way, equilibrium models have tended to understate the role of labor market lags in accounting for dynamic features of unemployment. As we previously suggested, it is important to recognize that these two alternatives—long-run equilibrium and persistence—are not mutually exclusive; European labor markets may have undergone structural worsening and exhibit prolonged persistence. Indeed, one of the more important findings in the present study is that of differing degrees of persistence versus structural change among the countries under review. France is at one end of the scale with relatively short persistence, while Germany is estimated to have considerable persistence, being possibly twice as persistent in the face of shocks. The main point is that it is difficult to identify these two developments—structural change and changes to persistence—unless the model used postulates that both are occurring.

The degree to which sluggish labor market behavior is due to labor market lags as distinct from changes in structural variables is a matter of serious policy concern. Different policies may be required to alter the equilibrium levels of unemployment models (such as the duration and size of social security benefits and union density), than are needed to reduce the dependence of wages, employment, and the labor force on their past values. Policies directed toward altering the long-run equilibrium level of employment in an economy where a high degree of unemployment persistence is actually the problem risk at best being ineffective in reducing high unemployment. That said, there are examples where each approach points to similar factors; the degree of centralization in wage bargaining, for example, probably affects both.

As emphasized already, there has recently been growing awareness of the importance of persistence, and the possibility that it may be a major part of the explanation of rising unemployment in Europe. What, then, does the present study provide that the existing literature does not? The main answer is that the present research analyzes the dynamics in each of the labor markets while providing a statistically acceptable explanation of their long-run behavior using cointegration analysis. This marks it out from other studies. Alogoskoufis and Manning (1988) take the long-run equilibrium rate to be exogenous in their model. The thoughtful statistical analysis of the sources of persistence by Nickell (1995) concentrates on correlating measures of labor market inertia, including employment sluggishness and change effects of unemployment in wage inflation, with intercountry measures of labor market tenure, severance conditions, and indices of union and employer coordination. His measures of persistence in employment and wage setting (the dependent variables in his regressions) are, however, those previously reported in Layard, Nickell, and Jackman (1991). It is not evident that these earlier estimates are based on acceptable econometric estimates of long-run behavior, because they combine stationary and nonstationary variables. The long-run part of the relevant equations does not cointegrate (that is, the implied equilibrium relationships may not have stationary errors), and hence the estimated short-run dynamics of the models may well be misleading.

As well as providing an econometric explanation of transitional and long-run unemployment dynamics, the study also derives summary statistics of the dynamic unemployment responses to temporary and permanent shocks. These statistics draw attention to the movement of unemployment through time, which depends not only on the constellation of lags in labor market behavior, but also on the interaction between this constellation of lags and the dynamic structure of the shocks to which the labor market is subject. These shocks—such as oil price changes, movements in the exchange rate and interest rates, changes in taxes, and so on—have both temporary and permanent components. The existing literature on persistence and hysteresis has, however, focused almost exclusively on temporary shocks.

The present study attempts to right the balance by considering not only the persistent unemployment effects of temporary shocks, but also the delayed unemployment effects of permanent shocks. We will call the former phenomenon “unemployment persistence” and the latter “imperfect unemployment responsiveness.” We will derive measures of these phenomena in the countries under consideration.

Behavioral Lags and Unemployment Dynamics

Earlier, we criticized simple autoregressions of unemployment as ways of estimating persistence. Suppose, for example, that unemployment can be characterized by the first-order autoregressive AR(1) model,

Ut=ρUt1+ɛt,(1)

where ρ is the measure of persistence. This usually shows values insignificantly different from unity for the five countries under consideration. Table 1 shows typical results.

Table 1.

AR(1) Models of Unemployment

article image

Equations run with constants, where DF is the Dickey-Fuller test of the hypothesis that (ρ-1) is zero. (The DF is -2.89 at the 95 percent level.)

Such tests are highly fallible of course, and we do not rely on them here.6 Moreover, autoregressions are themselves not particularly helpful in understanding the causes of high unemployment. A finding of persistence in a simple autoregression is of little help in formulating policy, because the model would not identify the reasons underlying that persistence. Thus, a behavioral dynamic model is necessary, and the basic form of the one used in the country studies in this project is described below.

It is a commonplace of empirical research that employment, wage determination, and labor force participation include significant lagged effects.7 We argue that such lags—and particularly the interactions among them and with shocks containing temporary and permanent components—play a key role in determining persistent movements of European unemployment rates in the aftermath of severe labor market shocks, such as the oil-price shocks of the mid- and late 1970s.

This feature—which arises from our multiequation approach—mirrors the insight from early business cycle models that emphasized dynamic interactions, such as multiplier-accelerator mechanisms, in accounting for macroeconomic cycles. The interactions among lags may be analyzed in terms of the model we use, which takes the following general form:

β0Yt=β1Yt1++βnYtn+Γ1Xt++ΓmXtm+ηt,(2)

where Y is a vector of labor market variables—employment, the real wage, and the labor force. X is a vector of exogenous variables, such as labor market policy variables, the real oil price, and the real interest rate; ηt is a vector of white-noise error terms. This model allows for within-equation and between-equation dynamics. The classic distinction by Frisch of “impulse” and “propagation” mechanisms in producing economic fluctuations plays the same role in this model. The dynamics inherent in equation (2) determine the propagation mechanisms of the labor market, which parallel the same concept in business cycle models. This chapter focuses on the reasons for, and the importance of, these propagation mechanisms.

We describe next a portfolio of the important behavioral lags that, we argue, have played a significant role in determining unemployment dynamics, and these lags will be the centerpiece of the empirical work that follows. For convenience, lags are grouped under five headings, the names of which refer to their common microeconomic causes. They are not of course intended to be a comprehensive description of these causes—lagged variables may occur in equations for reasons other than the ones we describe, but the headings are proposed merely as a labeling device.

Employment Adjustment Effect

The employment adjustment effect is represented by the lagged employment terms in the employment equation. This is the familiar lagged response in employment, often caused by the presence of adjustment costs on labor inputs. The underlying idea is straightforward: when firms face costs of adjusting their employment—such as hiring, training, and firing costs—their current employment will depend on their past levels of employment. The literature on optimal intertemporal employment plans in the presence of quadratic adjustment costs on employment is extensive. Sargent (1978) provides an extension to the case where firms are assumed to have rational expectations. Berndt and Fuss (1986) survey recent theory and applications of temporary equilibrium models in which factor demand equations for quasi-fixed factors such as employment are subject to adjustment costs. The traditional literature in this area focused exclusively on continuous and infinitely divisible adjustment costs, whereby the costs approached zero as the size of the employment adjustment fell. More recent literature, however, also considers “lumpy” adjustment costs that are of finite size for each worker (for example, Lindbeck and Snower (1988)). The assumption of lumpy adjustment costs introduces discontinuities into firms’ employment decisions: there is now a range of productivities over which it is in the firms’ best interests to remain inactive (refraining from both hiring and firing), whereas hiring and firing occur at productivities lying above and beneath this range, respectively. However, with the assumption of either divisible or lumpy adjustment costs, current employment depends on past employment.8

Insider Membership Effect

The insider membership effect can be proxied by lagged employment terms in the wage setting equation. The underlying hypothesis is that past employment determines the size of the current incumbent workforce (that is, the current stock of “insiders”), which, in turn, affects the insiders’ objectives in wage bargaining. This effect may be positive or negative, depending on the relative strength of two countervailing effects: (1) for any given distribution of labor demand shocks, the smaller the insider workforce of a firm, the greater the insiders’ job security at any given real wage, and consequently the higher the negotiated wage (see, for example, Blanchard and Summers (1986) and Lindbeck and Snower (1987a)); (2) the smaller the insider workforce, the smaller the bargaining power of the insiders (because, for example, the weaker the threats that the insiders make to the firms in case of bargaining disagreement), and therefore the lower the negotiated wage (see, for example, Lindbeck and Snower (1987b)). If the size of the current insider workforce depends on past employment, the negotiated wage will depend on past employment as well.

Wage Staggering Effect

This effect is represented by lagged wage terms in the wage setting equation. It has been recognized from at least the time of Taylor’s (1980) seminal article that current wages may depend on lagged wages, through the effects of overlapping wage contracts, where one set of bargainers makes a wage claim in the light of past and expected future settlements by comparator wage bargainers. In practice, these effects may reflect concerns with the maintenance of wage differentials. To take one of many examples, Wadhwani (1985) explores this hypothesis at length in estimated aggregate wage equations for the United Kingdom. (See also Foster and Henry (1984), who estimate sectoral wage equations—again for the United Kingdom—where they identify intersectoral wage effects as evidence of such overlapping wage contracts.9) Future expected real wage effects are not explicitly included in these specifications, although, on the assumption of the rational expectations hypothesis (REH), where real wages can be characterized by an autoregressive process, according to the substitution method of estimating an REH model, lagged real wages include effects of expected future real wages.

Long-Term Unemployment Effect

The long-term unemployment effect is represented by lagged unemployment terms in the wage equation. There has been considerable discussion of the weakening of the unemployment effect on wage outcomes as the duration of unemployment increases. This diminished effect occurs for several reasons: the skills of the unemployed deteriorate as their unemployment lengthens, making them less attractive to employers; the motivation to search for employment may also wane the longer the unemployment spell; and, finally, the employer may prefer to hire an individual who has been unemployed for only a short period.

A number of empirical studies have found support for the existence of weakening effects on wage settlements as unemployment duration increases. Hall and Henry (1987) and Layard, Nickell, and Jackman (1991) show there is a significant decrease in the bargaining power of the long-term unemployed in the United Kingdom.

Labor Force Adjustment Effect

This effect is represented by the lagged labor force terms in the labor force equation. These terms can be explained by the costs of entering and exiting the labor force. The costs of entering include the costs of registering as unemployed, of engaging in job search activities, and—perhaps most important—of changing one’s lifestyle to adopt the characteristics necessary to become employable (for example, reliability, punctuality, and initiative) as well as acquiring the basic skills needed in the workplace. These costs—whether infinitely divisible or lumpy—make people’s current labor force participation decisions depend on their past decisions in much the same way that the costs of employment adjustment make firms’ current employment decisions depend on past employment.

Interaction Among the Lagged Effects

An obvious but neglected point is that the lags above can operate much more powerfully in conjunction with one another than in isolation. Treating each of the lags described above alone as the principal source of unemployment persistence is, in our judgment, quite wrong. An example of this is provided in the otherwise excellent survey by Bean (1994), which, after describing persistence mechanism in seriatim, concludes that each alone does not appear to be large enough to account for Europe’s unemployment. While this inference appears to be valid, it is important to add the qualification that together these mechanisms may indeed be an important reason why unemployment is so high. Moreover, the approach followed in this book should be distinguished from those in many previous studies of unemployment persistence, which are highly aggregative;10 they generally take the form of autoregressions of the aggregate unemployment rate alone, rather than decomposing the constituent lags into the lagged effects on employment, wage determination, and labor force participation. Consequently, as previously emphasized, these studies can do little more than provide a summary description of sluggish unemployment adjustment; they cannot shed light on the economic causes of such sluggish adjustment.

This has serious consequences for policy analysis. It is impossible to derive valid policy conclusions from the mere fact that the unemployment rate displays a high degree of serial correlation. Policies must be designed from an empirical assessment of the individual dynamic effects that interact to produce the observed serial correlation in unemployment. In the latter case, different employment policy instruments may be identified that give the policymaker scope to influence different lagged effects. For example, changes in job security legislation may be expected to affect the degree to which current employment depends on past employment, while tax breaks for hiring the long-term unemployed will influence the degree to which current wages depend on past unemployment.

For this reason, it is important to identify the behavioral lags in labor market activity and analyze how the various lags interact with one another and with the dynamic structure of labor market shocks in generating the time path for unemployment. Charts 2 and 3 illustrate lagged interactions between employment and wage setting. The employment equation, representing the aggregate level of employment at any given real wage, is pictured as the downward-sloping labor demand curve LD in Chart 2. The wage setting equation, representing the negotiated wage at any given level of employment, is given by the upward-sloping WS curve. Finally, the labor force participation equation, representing the aggregate labor force at any given real wage, is given by the LF curve. The intersection of the labor demand curve (LD) and the wage setting curve (WS) yields the equilibrium real wage (w*) and the equilibrium level of employment (N*). The size of the labor force at the equilibrium real wage is the equilibrium labor force (L*). The difference between the equilibrium labor force (L*) and the equilibrium level of employment (N*) is the equilibrium level of unemployment (U*).

Chart 2.
Chart 3.
Chart 3.

Labor Market Response to Adverse Shocks

In this context, lags in employment adjustment owing to hiring and firing costs make the position of the labor demand curve (LD) depend on last period’s level of employment. Similarly, the labor force may be subject to adjustment lags that make the position of the labor force participation curve (LF) depend on last period’s labor force. Furthermore, there can be wage staggering, insider membership, and long-term unemployment effects that, in turn, make the wage setting curve depend on last period’s wage, employment, and unemployment, respectively.

Suppose that the labor market is initially (in period 0) at its long-run stationary state, denoted by point A in Chart 3, and then (in period 1) a temporary, adverse labor demand shock occurs, shifting the labor demand curve downward from LD0 to LD1 for one period. Thus, employment declines from N0 to N1. In the following period (period 2), when the adverse shock has disappeared, the labor demand does not shift all the way back to LD0 again; rather, it shifts part of the way back, say, to LD2, on account of lagged employment adjustment (the employment adjustment effect). The resulting equilibrium level of employment in period 2 is N2, and the process of adjustment continues in subsequent periods. This illustrates how the employment adjustment effect causes a temporary labor demand shock to have persistent effects on employment.

But that is not all. The adverse labor demand shock raises the level of unemployment from U0, to U1 in period 1. This may be expected to raise the unemployment durations of those without jobs. If effort on job search declines as the average unemployment duration rises (the “long-term unemployment” effect), then the rise in unemployment will raise the wage setting curve in period 2, for now the unemployed have become less effective in competing for jobs, and thus a higher wage is negotiated at any given level of employment. The rise in the wage setting curve, say, from WS0 to WS2 in Chart 3, puts further downward pressure on employment. Thus, employment will rise by less than the distance from N1 to N2 in Chart 3, and may even fall. This shows how the long-term unemployment effect reinforces the employment adjustment effect in prolonging the effects of a temporary shock.

Furthermore, the temporary labor demand shock reduces the real wage from w0 in period 0 to w1 in period 1. If current wage setting depends on lagged wages (the “wage staggering effect”), the wage setting function in period 2 will be lower than it would otherwise have been. Consequently, the wage setting function will rise by less than the previously described shift from WS0 to WS2. In this way, the wage staggering effect can be seen to dampen the influence of the employment adjustment effect, reducing the degree to which the effects of a temporary shock persist.

The influence of the insider membership effect in this context obviously depends on the relative strengths of its components. If the job security component dominates the bargaining component, then the fall in employment from N0 to N1 will push the wage setting function upward (because the reduced number of insiders now have greater job security than previously and are thus in a position to make more ambitious wage claims). In this case, the insider membership effect reinforces the employment adjustment effect, augmenting the persistent effects of the temporary shock. Conversely, if the bargaining component dominates the job security component, then the fall in employment in response to the adverse shock will push the wage setting function downward (since the reduction in the size of the insider workforce reduces the insiders’ bargaining power), and then the insider membership effect dampens the employment adjustment effect.

The illustration above was conducted entirely in response to a temporary shock. A permanent shock could elicit a quite different set of dynamic responses. There is no one-to-one relationship between the dynamic responses of unemployment to temporary and to permanent shocks. Economies in which temporary shocks have comparatively persistent effects on unemployment are not necessarily also the economies in which the full unemployment effects of permanent shocks are comparatively slow to manifest themselves. In short, inertia in the aftermath of temporary shocks does not necessarily imply inertia in the aftermath of permanent ones (see Karanassou and Snower (1993) and Snower and Karanassou (1995)).

This example illustrates our view that a fruitful way to understand the movement of unemployment in different European countries is by estimating the lags in employment, wage setting, and labor force participation equations, investigating how these lags interact with one another, and analyzing the implications of these interactions in the presence of shocks with both temporary and permanent components.

Empirical Model

This section specifies the empirical model that underlies the studies reported in the companion country studies, providing the basis for measures of unemployment persistence and responsiveness presented next.

As already anticipated, the model used is consciously synthetic, building on much previous work on the labor market. The framework is a one sector model of the labor market with emphasis on dynamics. It comprises the following equations.11

Employment Equation

This employment equation is the empirical representation of the aggregate short-run labor demand curve. In general, it can be expressed as

A0(L)ηt=A1(L)Y1t+A2(L)X1t+ɛ1t,(3)

where nt is total employment, Y1 is other endogenous variables (including the real product wage), X1 is a vector of exogenous variables, and ∈1 is a white-noise error term. Ai(L) (i = 0, 1, 2) is a polynomial in the lag operator L. The lagged employment terms in this equation are the “employment adjustment effect.” The underlying model is based on familiar dynamic optimizing models of the firm, where the firm faces adjustment costs on labor. Assuming quadratic adjustment costs coupled with assumptions that employment comprises several types (such as insiders and entrants) with differing cost functions, second-order lagged equations in employment result (see Sargent (1978), and Nickell (1978) for further discussion). With imperfect competition, the first-order profit-maximizing conditions may, under specific circumstances, lead to an employment equation conditioned on aggregate demand, among other things, and hence aggregate demand would then be a natural variable to include in the Y vector in equation (3).12 In the light of the endogeneity of real demand, an instrumental variable (IV) estimation would be needed. Alternatively, aggregate demand may be modeled using other exogenous variables, such as measures of aggregate policy and the external environment. (See Henry and Wren-Lewis (1983) for an early example. See also Bean (1994).)

Wage Setting Equation

This equation represents the real wage that emerges from bargaining between employers and employees. It can be written in general as

B0(L)wt=B1(L)X2t+B2(L)Y2t+ɛ2t,(4)

where w is the real consumption wage; Y2 represents other endogenous variables, including unemployment and employment; X2t is a vector of exogenous variables; and ∈2t is the error term. Again, this model encompasses models that have been extensively used in empirical work. Specifically, X2 could include tax and price wedge terms, and other structural factors, such as union membership or mismatch. The employment terms in the equation are associated with the insider membership effect, since the size of the firms’ insider workforce affects the insiders’ objectives in the wage setting process. The set of lagged real wage terms can reflect effects of overlapping wage contracts and produces the wage staggering effect, because staggered wage setting makes current real wages depend on their past values. Finally, current unemployment can capture the effect of an excess supply of labor on bargains, and lagged unemployment terms in the equation can proxy long-term unemployment effects, because the long-term unemployed tend to search less intensively for jobs and thus have less influence on the wage setting process than the short-term unemployed.

Labor Force Participation Equation

This equation describes the size of the labor force and takes the following form:

Γ0(L)lt=Γ1(L)X3t+Γ2(L)Y3t+ɛ3t,(5)

where lt is the size of the labor force, and X3 represents exogenous variables such as the working population, and Y3 represents endogenous variables including unemployment and the real wage. The lagged labor force terms in the equation may be associated with the “labor force adjustment effect,” since costs of entry to and exit from the labor force often make the current labor force depend on its past magnitudes. Discouraged-worker effects are proxied by unemployment. Thus, if unemployment is high, fewer workers are expected to enter the workplace because the probability of finding a job is low.

Definition of Unemployment

While unemployment is the difference between the labor force and employment, to preserve the linearity of our system, we take the following approximation:

ut=ltηt.(6)

The lags identified above are not the only ones that may be significant in labor demand, wage setting, and labor force participation decisions, but they occupy a particularly prominent place in the empirical and theoretical labor market literature, which is one reason they are emphasized here.

Estimation Methods and Overall Model Behavior

Estimating the Labor Market Model

The models described in this study are joint models of employment, wages, and labor force determination. A number of issues need to be addressed, including the following:

  • At least some of the labor market variables are nonstationary (namely, I(1), needing to be differenced once to render them stationary). To allow for this, equilibrium relationships between the nonstationary variables need to be sought so that, in combination, the variables have a stationary error (the combination of the variables are then cointegrating vectors). Using these results, the dynamic relationship between all the relevant variables—whether stationary or nonstationary—can be estimated in a statistically consistent way.

  • The labor market models entail the solution of simultaneous equations, with employment depending on current values of the real wage, for example. To avoid possibly biased estimates of the coefficients in the model, these simultaneous effects need to be explicitly allowed for. There are a large number of possible ways of dealing with this, depending on the precise forms of model involved, and some of these are illustrated in the country papers.

Some of the technical issues arising in estimating models of this sort are discussed more fully in the appendix.

Indices of Unemployment Persistence and Imperfect Responsiveness

The literature on unemployment persistence and hysteresis 13 has concentrated almost exclusively on temporary shocks. In practice, however, labor market shocks have both temporary and permanent components. For example, whereas some exchange rate fluctuations and movements in the prices of raw materials are temporary, some changes in productivity, taxes, and real interest rates are long lasting. Supply-side changes such as deregulation, the decline of union power, and the easing of job security legislation are further examples of longer-term shocks.

The study provides a methodology for exploring these issues. It recognizes two dimensions of the unemployment problem: (1) the persistent effects of temporary labor market shocks, or unemployment persistence, and (2) the delayed effects of permanent shocks, or imperfect unemployment responsiveness.

While it is possible to get some idea of the effects of lags in particular equations, interest is primarily in the dynamic behavior of the entire model. It is standard to use dynamic simulations of temporary and permanent shocks to estimate how long it takes for unemployment to reach the neighborhood of equilibrium (for example, the half-life of a delay is often used, being the time taken to get halfway to equilibrium).

The simulations reported are fully dynamic ones. These first solve the complete model forward in time, treating the exogenous variables as fixed at their 1994Q2 values. This gives a base solution. Next, a further simulation applies a shock of a 0.01005 reduction in the constant term of the employment equation, which is equivalent to a 1 percent reduction in the level of employment, and recomputes the solution. This simulation can be thought of as an exogenous productivity shock to the system. The difference in the two solutions then gives the full model dynamic response to the exogenous shock. We report two sets of summary indices concerning unemployment persistence and imperfect unemployment responsiveness. The first indicates how long the model’s predicted unemployment rate takes to get halfway back to equilibrium after it has been disturbed by a temporary shock (that is, one that lasts for only one quarter) or a permanent one. The second set is the sum of the deviations of the predicted unemployment rate from its long-run equilibrium value normalized by the size of the shock, whereas the first set focuses on how long the behavioral lags keep unemployment from its stationary state. The second set is a proxy for the resulting waste of labor resources over time.

Overview of Country Results

Country results are described at length in Chapters 26. Here, we draw together the key results, highlight the differences that have been found between the countries, and discuss their policy implications.

There is a growing awareness of the possible importance of employment and unemployment dynamics as key features of OECD unemployment. It is increasingly recognized that an explanation of the response of employment and unemployment to shocks is crucial to the understanding of high levels of unemployment in Europe. In comparisons with the United States, European countries show considerable persistence (see Table 2), and there are significant differences in the pattern of persistence among European countries themselves.

Table 2.

Persistence Measures of Selected European Countries and the United States

(Mean lag in years)

article image

These are ε neighborhood calculations, not mean lags. They are thus higher than the mean lag, which gives the time taken to go halfway to equilibrium.

Why is there such an apparent profound difference between the United States and Europe on the one hand, and what is the explanation of inter-European differences on the other? Before turning to the results of our empirical work, we note that the literature on European labor market inflexibility tends to have concentrated on a number of distinct themes: hiring/firing regulations and other costs of adjusting employment, wage rigidities, and the structure of bargaining (specifically with respect to its degree of centralization). These phenomena affect unemployment dynamics in that they are responsible for generating lags in labor market behavior.

The evidence that hiring/firing regulations interfere with the functioning of most European labor markets is relatively unambiguous. About 50 percent of employers in most EU countries claim that these costs constitute a serious impediment to increasing employment (see Table 3 and Pujol (1995) for further discussion). Interestingly, the United Kingdom appears to be an exception in this sample.

Table 3.

Firms Citing Hiring/Firing Procedures as a Factor for Not Hiring

(In percent)

article image
Source: Pujol (1995).

The findings from existing empirical studies analyzing the influence of wage rigidities have been more contentious. Layard, Nickell, and Jackman (1991) adopt a fairly simple measure of real wage rigidity in terms of the effects of unemployment on the real wage and the markup of prices over wages. The aggregative nature of this study and others makes it difficult to assess what the sources of this rigidity may be—in terms of lagged responses of labor market decisions to external shocks—and how these sources interact with one another. Other studies are somewhat more precise. Alogoskoufis and Manning (1988), for example, attempt to identify the importance of insider effects in wage setting, with the recently employed changing the unemployment effect on real wages. They find that this insider effect is not important. Nickell (1995), on the other hand, relates cross-sectional estimates of the effect of wages on the change in the unemployment rate (as an effect from the change in unemployment on wages is often taken as a measure of wage inertia) to a number of indices of insider power and bargaining coordination. He finds, in contrast, that insider power is important in accounting for real wage persistence.

As noted previously, none of these studies systematically analyzes the time-series evidence, such as presented here. What the studies cited so far confirm, however, is that variations in persistence are substantial between Europe and the United States. For example, Nickell (1995) estimates a mean lag of just over three years in unemployment following a shock for the United States, whereas it is between seven and nine years for the United Kingdom (p. 31). See also Karanassou and Snower (1993).

Table 2 places these results in a wider context, presenting summary statistics on persistence measures in selected European countries and in the United States. The table illustrates the familiar result that persistence in unemployment is estimated to be much lower in the United States than in European countries. Although the above calculations are based on different models, estimation methods, and sample sizes, each is nonetheless consistent with this general pattern.14 In turn, employment is estimated to be more responsive in the United States, a finding that appears to be an important determinant of low unemployment persistence there. In similar fashion, there is some evidence that the persistence of German unemployment may be higher than in other European countries.

Summary of Findings on Labor Market Dynamics

For the five countries studied, the findings suggest that labor market lags play an important role in determining how unemployment responds to labor market shocks. This conclusion is supported by the econometric results for single equations, which have significant lagged effects, implying slow responses in employment, wages, and the labor force following a shock when each of these variables is considered separately. Single-equation information, of course, is only a rough guide to the dynamics of the whole model, because it does not include interequation dynamics. Only full model simulations account for all the dynamic feedbacks in the model, and these are reported next.

Model simulations of a temporary labor demand shock in France show that unemployment takes approximately three years to recover halfway to equilibrium.15 For Germany, this response may take as long as six and a half years. Equilibrium unemployment may be higher in France, however. In Italy and the United Kingdom, the response appears to be about the same at five years. (See Chapters 26 and Table 4.)

Table 4.

Unemployment: Indices of Persistence

article image
Note: The mean lag is the familiar half-life of the response to the shock. The sum of the absolute changes of unemployment from the base following the shock is given as an alternative. For Spain, (A) refers to the basic model, (B) to the policy model. (See Chapter 6 for the definitions of these two scenarios.)

These results indicate that considerable dispersion exists in the overall dynamic behavior of these selected European economies. France appears to respond quickly to shocks, but Germany appears to be at the other end of the scale, with Italy and the United Kingdom somewhere in between. Although other researchers have drawn attention to the differences between European and U.S. persistence, the empirical results given here show that intra-European differences are also profound.

The empirical results for German employment, wage, and unemployment equations are consistent with the existence of relatively strong persistence mechanisms there; that is, the tests reveal significant employment persistence, wage staggering, and labor force adjustment effects. More generally, these findings appear consistent with institutional features of the German labor market, namely, its centralized bargaining system and strong employment protection (which is probably coupled with the presence of extensive risk sharing by employers, thus ensuring that employees are shielded from adverse temporary real shocks).

The five countries’ responses to permanent shocks reveal important differences between Italy and the United Kingdom. (See Chart 4 and Table 5.) In both countries, the equilibrium level of unemployment rises, but it rises much more in the United Kingdom than in Italy. For a protracted interval (of more than a decade), the increase in unemployment in Italy exceeds its long-run effect. Gradually, the level of unemployment adjusts downward, largely because of falling participation rates, but these processes take a long time to work through. For the United Kingdom, although increases in unemployment appear more regular, the timescale of the change is significant. Thus, it takes about five years to get halfway to the full equilibrium effect. The pattern of unemployment in France following the shock is also one of overshooting, so that the worsening of short-run unemployment exceeds that of long-run unemployment (as with Italy). The process is much quicker in France, also because the persistence of unemployment appears short lived. Germany shows a most singular oscillating response to the productivity shock, with unemployment over- rather than undershooting its long-run value. The time taken to get to the new equilibrium is also very long, with the labor market approaching equilibrium in a cycling manner, with cycles each lasting over a decade.

Chart 4.
Chart 4.

European Union: Imperfect Responsiveness

Table 5.

Unemployment: Indices, of Imperfect Responsiveness

article image

Finally, in Spain, the response to a permanent shock is estimated to take a very long time in the “basic” model, but appears much shorter in the “policy” model. This point is discussed further in the next section.

Are Labor Market Lags Affected by Policy?

The research described in the country studies shows that important differences may have emerged between European countries during the 1980s as a result of labor reforms introduced in certain countries. The United Kingdom reformed its industrial relations procedures in the 1980s to improve both wage bargaining and the flexibility of hiring and firing regulations. In Italy, far-reaching reforms were initiated in the mid-1980s and again at the beginning of the 1990s, both times to reform wage bargaining procedures and, in the 1990s, hiring and firing regulations. Spain has experienced profound changes in its labor market structure. After the demise of the Franco regime in 1975, labor relations and social protection changed significantly, although restrictions on working conditions and labor mobility continued. The power of unions increased dramatically and entailed substantial severance costs. The country has an active minimum wage policy, although the real value of wages did not grow in the 1980s at rates achieved in the 1970s. In the last couple of years, Spain has undertaken reforms to reduce unemployment compensation and increase labor mobility, in part by reducing firing costs. Last, and in contrast to the previous examples, France and Germany have attempted relatively little deep-seated reform of the labor market.

Although any results must be regarded as tentative, some quantitative assessment of the effects of these reforms is possible. According to our econometric results, there appears to have been little change in the underlying behavior of employment, wage setting, or labor force participation in France and Germany. In Italy, Spain, and the United Kingdom, however, reforms implemented over the last decade or so appear to have discernable, quantitative effects on certain aspects of labor market behavior.

The case of Italy provides two examples of policies designed to reform the labor market: policies that were undertaken in 1984 and those undertaken in 1991. According to results reported in the study on Spain (Chapter 6), the policy initiatives in 1984 failed (in the sense that they appear not to have affected underlying behavior), whereas those of 1991 appear thus far to have succeeded. Although tentative, these latter results suggest that

  • the employment adjustment effect has declined;

  • the responsiveness of employment to output variations has fallen, and that with respect to wages has increased;

  • the wage staggering effect has fallen (although it is too early to judge whether this is a significant change); and

  • the labor force is significantly less prone to discouraged-worker effects.

These changes all point to more flexibility in the labor market. The changes in the effect of lagged employment and output on current employment are consistent with a decrease in labor hoarding as well as with the hypothesis that employment adjustment costs, both actual and implicit, have fallen.16 Increases in the elasticity of employment with respect to real wage changes also point to increasing flexibility in the labor market. In turn, the decline in the discouraged-worker effect and the decrease in lagged employment effects in the labor force equation indicate possible increases in the flexibility of labor supply. Finally, wage bargaining patterns may have changed, as the evidence suggests that wage staggering effects have decreased.

The United Kingdom also shows that the underlying patterns of behavior in the labor market seem to have changed under the impetus of policy initiatives taken during the 1980s. These were mainly in the sphere of reform of industrial relations, which raised the costs of strikes, eased hiring and firing costs, and promoted the decentralization of wage bargaining (see Henry and Karanassou (Chapter 5) for details). These reforms might be expected to affect both employment flexibility and the responsiveness of wages.

Unlike in Italy, labor market reforms in the United Kingdom have been in place long enough to enable direct tests of structural change in employment and wage equations to be made. The results of these tests show that the principal effects of the reforms have been on employment. This is consistent with there being some easing of hiring and firing constraints on firms. Ascertaining the scale of this improvement is not straightforward, however. The evidence from the employment demand equation is that the overall effect, although not statistically significant on conventional tests, may nonetheless be quantitatively important. This result may be affected by the level of aggregation used, however, and effects at the manufacturing level, for example—where adjustment costs are probably significantly larger than elsewhere—could be larger.

Tests were also conducted of possible effects of the reforms on wage bargaining. These concentrated on whether there is evidence of a change in the wage-unemployment relationship. Whereas in a more flexible labor market, the unemployment effect might be expected to increase, little evidence of a discernable change in the effect of unemployment on real wage developments is found. Our tests thus support the interpretation of no change in underlying wage behavior.

France and Spain fall into a different category in that policy is found to have important effects on the equilibrium level of unemployment. In Spain, policy also appears to have had important effects on the dynamic behavior of the labor market. In France, social security spending is found to have an important effect on unemployment through its effect on employment, reducing this below what it would otherwise be. This appears to operate directly on employment, not, as might be expected, through an effect on the real wage. The analysis is further complicated by the finding of an important effect on the aggregate real wage of the minimum wage level. At the same time, persistence in France is among the lowest of the countries reviewed here. Thus, although France appears to be suffering from policies that lead to a high equilibrium level of unemployment, it appears to respond quickly to shocks.

Spain is not only the country with the highest rate of unemployment in the countries investigated but its labor market is the most complex. The judgment on developments in Spain (see Chapter 6) is that over the 1980s a number of changes in the labor market raised the equilibrium rate of unemployment and probably decreased its (substantial) persistence. It seems that the increases to disability pensions, higher levels of unemployment benefits, and the increased incidence of industrial disputes tended to lower employment and that the minimum wage tended to raise the aggregate wage. All these factors tend to raise the trend rate of unemployment. Some amelioration in these tendencies is due to the increased share of the salaried workforce on temporary contracts, which appear to reduce the real wage. This reduction is small, however, and affects only a small part of the workforce. Overall, these factors appear to have decreased the persistence of the labor market as well. Nonetheless, it appears a reasonable characterization of developments in the Spanish labor market to interpret the policy changes over the 1980s as both leading to higher equilibrium rates and, through effects on employment and wage adjustment, reducing persistence.

Conclusions

While it is important not to overstate the conclusions drawn from our models, most of the major European economies exhibit a substantial degree of unemployment persistence and imperfect responsiveness. Although preliminary, the empirical results suggest that the degree to which each country is characterized by these features differs, apparently quite significantly. One finding of this study is that there appear to be substantial differences in these countries’ dynamic responses to shocks. It is therefore misleading to lump them together as having common dynamics of a single “European” variety. In turn, the analysis suggests some of the key areas of the labor market in which these differences arise. Although in a study such as this it is difficult to quantify the effects that particular institutions and policy characteristics have upon the dynamic behavior of the labor market as a whole, some indications of the role that policy may play can be obtained. In France, persistence seems relatively low, but the minimum wage and social security policy have probably led to high trend rates of unemployment. Over the 1980s, unemployment in Spain has risen, but persistence has declined. In Germany, on the other hand, labor market policy has changed little, and the responsiveness of the general labor market to shocks appears slow despite, for example, the coordination of pay bargaining and the level of training, which might be expected to lead to low persistence.

Italy and the United Kingdom exemplify policy initiatives that can improve the resilience of the economy to shocks. The evidence that this is happening is not clear, however, even in the United Kingdom, where reform has been in train for over a decade. This finding, by itself, suggests that improving the patterns of dynamic responses in labor markets may be a very long process.

Appendix: Issues in Estimating Dynamic Systems

This appendix discusses issues in estimating the models described in this chapter and used in the country studies, which are in the traditional form of structural dynamic models. Repeating equation (2) from p. 11 for convenience, they are of the following general form:

β0Yt=β1Yt1++βpYtp+Γ1Xt++ΓqXtq+ηt,(2)

where Yt is an m-dimensional vector of endogenous variables, Xt is a k-dimensional vector of exogenous variables, and ηt is an m-dimensional vector of independently and identically distributed (iid) random disturbances. The model allows for the possible presence of simultaneous effects between the endogenous variables; hence, the matrix β0 will have restrictions that derive from economic theory (that is, β0I, the unit matrix). To estimate this model, appropriate simultaneous estimators of either a limited-information or a full-information variety are required. (This is discussed further below.)

In estimating the models in the country papers, a sequence of steps is followed:

  • (1) The relevant variables are tested for orders of integration, and tests are conducted for the presence of cointegration in the subset of I(1) variables.

  • (2) Cointegrating vectors are estimated, using either Johansen’s maximum likelihood procedure or Pesaran and Shin’s (1995a) autoregressive distributed lag (ARDL) approach.

  • (3) The model is estimated in its I(0) form, as a vector error correction model (VECM).

Step (1) is clearly necessary to ensure that issues of spurious regression and inconsistent estimation are avoided; it is familiar enough not to require extensive justification. The justification behind step (2) involves identifying cointegrating vectors and is more unfamiliar. Step (3) is also a familiar procedure, most often involving the two-step procedure originated by Engle and Granger (1987), where cointegrating residuals are entered into a dynamic equation that, by construction, is composed of I(0) variables. In a single-equation ARDL (1,1) model

yt=α0+α1+yt1+β0xt+β1xt1+ɛt,(1)

where yt and xt are scalar I(1) variables. In a two-step procedure, estimates of the long-run relation are based on the presence of a cointegrating vector between yt and xt. Then, equation (1) may be estimated in its ECM form as

yt=α0(1α1)(yt1θxt1)+β0Δxt+ɛt.(2)

The long-run multiplier yt with respect to xt

θ=(β0+β1)/(1α1),

is estimated directly from equation (1) by Pesaran and Shin (1995a), who have shown that by estimating a suitably augmented ARDL, all the short-run parameters (α0, α1, β0, β1) are T-consistent, while the long-run parameter, θ, is T-consistent. Equation (1) can then be used to produce standard errors of the model’s short-run parameters as well as the long-run parameters such that standard inference on both parameters is possible.

Integrability

Dividing by β0, equation (2’) can be written as

Yt=φ1Yt1++φpYtp+ψqXtq+μt,(3)

where φi=β01βi,i=1,,p,ψj=β01Γj,j=0,,q,andμt=β01ηt. At this point, it is important to define the order of integrability of the variables in the model. If they are all I(0) stationary variables, then equation (2’) or, if the model is just identified, equation (3) may be estimated directly, either by using a simultaneous estimator on equation (2’) if the model is overidentified or by using indirect least squares on equation (3) in the just-identified case. If all the identifying restrictions from the reduced form are ignored, the model can be consistently estimated by single-equation least squares, although it has no structural interpretation.

This vector autoregression (VAR) model is then the theoretic model proposed by Sims (1980), for which he recommended the application of “minimal” restrictions in order to identify shocks. Where the variables are nonstationary, they may be rendered stationary by differencing, which is standard practice for estimating both structural models like that in equation (2’) or, in the unrestricted case, VAR models of the same general form as in equation (3), with the change that all variables are now replaced by first differences (ΔYt, ΔXt). Inducing stationarity by differencing the relevant variables the appropriate number of times loses potential information and is not appropriate where there is potential cointegration between the nonstationary variables. Where the I(1) variables are cointegrated, estimating the model in first differences is invalid and will result in misspecification and inefficiency (see Robertson and Wickens (1994)).

The appropriate form of model is then a VECM. Assuming that the k-dimensional exogenous variables, Xt, follow the multivariate independent unit root processes, ΔXt = et, and p = q, we can write equation (2’) as the VAR(p) model,

A(L)Zt=ut,(4)
A(L)=[β(L)Γ(L)01L],

where β(L)=β0β1LβpLp,Γ(L)=Γ0Γ1LΓpLp,u1=(ηt,et),andZt=(Yt,Xt) is the n (= m + k)-dimensional vector of combined endogenous and exogenous variables. Noting that A(L) = A0 - A1L - … - APLP, and dividing equation (4) by A0, we obtain

Π(L)Zt=vt,(4)

where Π(L)=InΠ1LΠpLp,Πi=A01Ai,i=1,,p,andνt=A01ut.. Reparameterizing equation (4) we finally obtain the (reduced-form) VECM as

ΔZt=Π1LΠpLp,Πi=A01Ai,i=1,,p,andvt=A01ut.(5)

where Π(1)=InΠ1ΠpandΠi*=Σjp=i+1Πj.. If there are r cointegrating vectors between the n-dimensional Zt variables, then the rank of Ï(1) is r, and Ï(1) can be decomposed as αβ’. where β is the n × r matrix of r cointegrating vectors, and α is the corresponding matrix of factor loadings (see Engle and Granger (1987)). If ht = β’Z, is the r-vector of cointegrating residuals, then the appropriate stationary form for estimating the model is

ΔZt=αht1+ΠiΔZt1++Πp1*ΔZtp+1+vt.(6)

Structural Versus Reduced-Form Models

Reorganizing the model above into a VECM can be done directly on the structural form of the model (2’) Thus, if we take the simpler first-order model for convenience

β0Yt=β1Yt1+Γ0X1+Γ1Xt1+ηt,(7)

an ECM with endogenous regressors results, that is,

β0ΔYt=(β0β1)(Yt1θXt1)+Γ0ΔXt+ηt,(8)

where θ=(β0β1)1(Γ0+Γ1)..

Estimation of equation (8) then needs to allow for the possible presence of simultaneously determined variables in each equation. In the examples provided in the country papers, equations with endogenous regressors have been estimated using instrumental variables to ensure independence of right-hand-side variables and the equation error. An alternative is to use full-information methods, although the trade-off between these and limited-information methods is well known (full information is sensitive to misspecification of any single equation, whereas limited information is robust to this). However, there is no clear-cut way to resolve these differences, and examples of each approach are given in the country papers.

Identification

The principal method of estimating the model, Johansen’s maximum likelihood approach, gives the long-run relations ht = β’ZT, the r linear combinations of the Z, variables, which are cointegrated. This, however, is a purely “empirical” identification scheme. For identification, estimation, and hypothesis testing under the general nonhomogeneous and possibly nonlinear overidentifying restrictions on the cointegrating vectors, see Pesaran and Shin (1995b). In addition, in Johansen’s VAR approach, the lag length on each variable is assumed to be the same, and in many cases this lag length may be long, thereby severely constraining degrees of freedom. This is one justification for estimating the long-run parameters using the ARDL approach. Pesaran and Shin (1995a) have shown that this approach is appropriate even when the variables are I(1), provided that the lag lengths are appropriate as judged by information criteria. This procedure may be an advantage over the Johansen one because it is clearly possible to both incorporate a priori information readily on signs of individual parameters and to allow differing lag lengths to be estimated for individual variables. Pesaran and Shin show that such ARDL models can yield superconsistent estimates of the model’s long-run parameters.

There are arguments, however, in favor of using a system maximum likelihood approach, because the set of cointegrating vectors it identifies may then be tested for economic-theoretic restrictions (see Johansen (1991) and Pesaran and Shin (1995b)).

References

  • Alogoskoufis, George S., and Alan Manning, 1988, “On the Persistence of Unemployment,” Economic Policy: A European Forum, Vol. 3 (October), pp. 42869.

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

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

  • Bean, Charles, and Richard Layard, eds., 1987, The Rise in Unemployment (New York: Blackwell).

  • Berndt, Ernst, and Melvyn Fuss, 1986, “Productivity Measurement with Adjustments for Variations in Capacity Utilization and Other Forms of Temporary Equilibrium,” Journal of Econometrics, Vol. 33, pp. 729.

    • Search Google Scholar
    • Export Citation
  • Blanchard, Olivier J., and Lawrence H. Summers, 1986, “Hysteresis and the European Unemployment Problem,” NBER Macroeconomics Annual, Vol. 1 (Cambridge, Massachusetts: MIT Press).

    • Search Google Scholar
    • Export Citation
  • Calmfors, Lars, and John Driffill, 1987, “Centralization of Wage Bargaining and Macroeconomic Performance,” Seminar Paper No. 402 (Stockholm: University of Stockholm Institute for International Economic Studies).

    • Search Google Scholar
    • Export Citation
  • Commission of the European Communities, 1993, Growth, Competitiveness, Employment: The Challenges and Ways Forward into the 21st Century (Luxembourg: Office for Official Publications of the European Communities).

    • Search Google Scholar
    • Export Citation
  • Engle, Robert F., and C.W.J. Granger, 1987, “Cointegration and Error Correction: Representation, Estimation, and Testing,” Econometrica, Vol. 55 (March), pp. 25176.

    • Search Google Scholar
    • Export Citation
  • Foster, N., and S.G.B. Henry, 1984, “Public and Private Sector Pay: A Partly Disaggregated Study,” National Institute Economic Review, No. 107 (February), pp. 6373.

    • Search Google Scholar
    • Export Citation
  • Hall, S., and S. Henry, 1987, “Wage Models,” National Institute Economic Review, No. 119 (February), pp. 7075.

  • Hall, S., and S. Henry, 1988, Macroeconomic Modelling (Amsterdam; New York: North-Holland).

  • Henry, S., and S. Wren-Lewis, 1983, “The Aggregate Labour Market in the U.K.: Some Experiments with Rational Expectation Models,” in Contributions to Modern Macroeconomics, ed. by P. Muet and P. Malgrange (London: Basil Blackwell).

    • Search Google Scholar
    • Export Citation
  • Henry, S., J. Payne, and C. Trinder, 1985, “Unemployment and Real Wages: The Role of Unemployment, Social Security Benefits and Unionisation,” Oxford Economic Papers, Vol. 37 (June), pp. 33038.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 1994, “Fostering Job Creation, Growth, and Price Stability in Industrial Countries,” in World Economic Outlook (Washington), pp. 3453.

    • Search Google Scholar
  • Johansen, Soren, 1991, “Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,” Econometrica, Vol. 59 (November), pp. 155180.

    • Search Google Scholar
    • Export Citation
  • Karanassou, Marika, and Dennis J. Snower, 1993, “Explaining Disparities in European Unemployment Dynamics,” Rivista di Politica Economica, No. 2, pp. 3762.

    • Search Google Scholar
    • Export Citation
  • Layard, R., S. Nickell, and R. Jackman, 1991, Unemployment (New York: Oxford University Press).

  • Lindbeck, Assar, 1995, “Hazardous Welfare-State Dynamics,” American Economic Review, Papers and Proceedings, Vol. 84 (May), pp. 915.

    • Search Google Scholar
    • Export Citation
  • Lindbeck, Assar, and Dennis J. Snower, 1987a, “Union Activity, Unemployment Persistence, and Wage - Employment Ratchets,” European Economic Review, Vol. 31 (February), pp. 15767.

    • Search Google Scholar
    • Export Citation
  • Lindbeck, Assar, 1987b, “Strike and Lock-out Threats and Fiscal Policy,” Oxford Economic Papers, Vol. 39 (December), pp. 76084.

  • Lindbeck, Assar, 1988, “Cooperation, Harassment, and Involuntary Unemployment: An Insider-Outsider Approach,” American Economic Review, Vol. 78 (March), pp. 16788.

    • Search Google Scholar
    • Export Citation
  • Lindbeck, Assar, 1994, “How Are Product Demand Changes Transmitted to the Labour Market?” Economic Journal, No. 104 (March), pp. 38698.

    • Search Google Scholar
    • Export Citation
  • Nickell, S., 1978, “Fixed Costs, Employment, and Labour Demand over the Cycle,” Economica, Vol. 45 (November), pp. 32945.

  • Nickell, S., 1995, “Labor Market Dynamics in OECD Countries,” Discussion Paper No. 255 (London: Centre for Economic Performance, London School of Economics).

    • Search Google Scholar
    • Export Citation
  • Organization for Economic Cooperation and Development, 1994, The OECD Jobs Study, Part 2, The Adjustment Potential of the Labor Market (Paris).

    • Search Google Scholar
    • Export Citation
  • Pesaran, M. Hashem, and Yonagcheol Shin, 1995a, “An Autoregressive Distributed Lag Modelling Approach to Co-integration Analysis,” Working Paper No. 9514 (Cambridge, England: Department of Applied Economics, University of Cambridge).

    • Search Google Scholar
    • Export Citation
  • Pesaran, M. Hashem, and Yonagcheol Shin, 1995b, “Long-Run Structural Modelling,” Working Paper No. 9419 (Cambridge, England: Department of Applied Economics, University of Cambridge).

    • Search Google Scholar
  • Phelps, Edmund S., in collaboration with Hian Teekltoon, George Kanaginis, and Gylfi Zoega, 1994, Structural Slumps: The Modern Equilibrium Theory of Unemployment, Interest and Assets (Cambridge, Massachusetts: Harvard University Press).

    • Search Google Scholar
    • Export Citation
  • Pujol, T., 1995, “Labor Demand and Turnover with Adjustment Costs” (unpublished; Washington: International Monetary Fund).

  • Robertson, Donald, and Michael Wickens, 1994, “VAR Modelling,” in Applied Economic Forecasting Techniques, ed. by Stephen Hall (New York: Harvester Wheats heaf).

    • Search Google Scholar
    • Export Citation
  • Sargent, Thomas, 1978, “Estimation of Dynamic Labor Demand Schedules Under Rational Expectations,” Journal of Political Economy, Vol. 86 (December), pp. 100944.

    • Search Google Scholar
    • Export Citation
  • Sims, Christopher, 1980, “Macroeconomics and Reality,” Econometrica, Vol. 48 (January), pp. 148.

  • Snower, Dennis J., and Marika Karanassou, 1995, “A Contribution to Unemployment Dynamics,” CEPR Discussion Paper No, 1176 (London: Centre for Economic Policy Research).

    • Search Google Scholar
    • Export Citation
  • Taylor, John B., 1980, “Aggregate Dynamics and Staggered Contracts,” Journal of Political Economy, Vol. 88 (February), pp. 123.

  • Wadhwani, S., 1985, “Wage Inflation in the United Kingdom,” Economica, Vol. 52 (May), pp. 195207.

1

See Organization for Economic Cooperation and Development (1994) and the Commission of the European Communities (1993). both of which give a comprehensive survey of evidence and potential explanations. An influential academic study covering this ground is Layard, Nickell, and Jackman (1991).

2

We believe that the conventional treatment, which concentrates on price inflation “surprises” as the sole source of short-run nominal-real interaction, is oversimplified. However, to treat this issue fully would involve considerably more complex models than are presented here and would need to include equations for nominal exchange rates and the multiple stages of the pricing process.

3

Hall and Henry (1987) and Henry, Payne, and Trinder (1985) give some early evidence of this.

4

See, for example, Organization for Economic Cooperation and Development (1994), Lindbeck (1995). and Phelps (1994) for arguments along these lines.

5

It rules our the possibility that unemployment may be nonstationary or I(1) and that a cointegrating vector exists between unemployment and other I(1) variables.

6

Apart from the reasons advanced earlier, the tests themselves are problematic. Such tests based on Dickey-Fuller tests have been widely used in tests for unit roots in aggregate output. Apart from other problems, they may not be appropriate when the series is affected by a structural break.

7

The potential list of contributors would be extensive. Hut studies relevant to what is reported here arc Sargent (1978), Taylor (1980), Hall and Henry (1988), Lindbeck and Snower (1987a), Alogoskoufis and Manning (1988), Bean and Layard (1987), and Layard, Nickell, and jackman (1991).

8

Under quadratic adjustment costs, current employment is a linear function of lagged employment. Under other forms of divisible adjustment costs, the relation is generally nonlinear, and in this ease, the linear employment equations in the empirical models below may be intetpreted as linear approximations of such nonlinear functions. Under lumpy adjustment costs, current employment depends on past employment only in the range of inaction; but when the employment decisions of heterogeoeous firms are aggregated under stochastic conditions, a standard employment equation—in which current employment again depends on lagged employment—can be derived. There are also good reasons for believing that adjustment costs are asymmetric and that hiring new employees in a cyclical upturn can be more costly than reducing the workforce as output falls. The employment-output relationship may thus be nonlinear for this reason also.

9

In our applications, we allow tor the possibility that current real wages may depend on past real wages, reflecting this wage comparison phenomenon. Since the models covet the medium run, in which price misperceptions play no significant role, norminal wage staggering effects translate directly into teal wage staggering effects.

11

All variables that follow, except the unemployment rate, are in logs.

12

See, for example, Lindbeck and Snower (1994).

14

The Alogoskoufis and Manning results are probably overestimates, for reasons described earlier.

15

In what follows, half-lives are taken to be the time taken to get halfway to full equilibrium; in the case of a temporary shock, this means halfway to the base solution.

16

This needs to be interpreted with care, however, because the other contending explanation—that expected real demand has fallen—has nor been formally tested in this work.