The United Kingdom adopted far-reaching labor market reforms in the 1980s. The aim of the reforms was not just to raise the employment level and reduce the unemployment rate; they were intended primarily to change the nature of industrial relations prevailing in the United Kingdom, decentralize wage bargaining, and make the labor market more flexible and more resilient in the aftermath of adverse shocks. This paper evaluates the dynamic behavior of the U.K. labor market and the extent to which it was altered by the policy changes of the 1980s, using the methods described in Chapter 1 of this volume. The U.K. labor market is particularly suited for an analysis of policy change.
A number of policy changes were proposed in the 1980s as part of the labor market reforms launched by the Conservative Government led by Mrs. Thatcher. An important policy shift related to legislation that would reduce the incidence of industrial disputes by raising the costs to unions of going on strike. The legislative changes in this period also banned “secondary action,” so that workers could not participate in the industrial disputes of enterprises to which they were not directly connected. Furthermore, secret balloting of employees was made mandatory before the employees could go on strike. The strengthening of the prerogative of employers through legislation was perceived as an important way of enhancing flexibility in the labor market. In addition, the Employment Acts of 1980 and 1982 and the Trade Union Act of 1984 decentralized the level at which wage bargaining was conducted.1 Decentralization of wage bargaining was conceived as the appropriate strategy for moderating the United Kingdom’s traditionally high wage inflation.
The reforms of the 1980s also aimed to remove a number of other restrictions that hampered flexibility in the labor market. Restrictions on hiring and firing were eased significantly. There are currently no restrictions on the dismissal of workers who have worked fewer than two years for an employer, and the United Kingdom does not practice affirmative action programs for hiring except in the case of handicapped persons. In addition, the labor market reforms made significant changes to the benefits system. The maximum duration of unemployment benefits was reduced to one year, and the replacement ratio was reduced to one of the lowest in Europe.2 The reforms of the early 1980s also considerably reduced the scope of wage councils, which were instrumental in setting minimum wages; the wage councils were completely abolished in 1993, and there are currently no minimum wages in the United Kingdom except those for agricultural workers.3 The removal of minimum wage laws was expected to improve the employment prospects of unskilled labor.
This paper explores the effect of these market reforms on the lag structures determining employment, wage, and labor force participation behavior. This topic is important because the U.K. unemployment problem—like that of most continental European countries—has entailed not only a high average rate of unemployment, but also a slow decline of unemployment after recessions in the product market are over. The reaction of unemployment to both temporary and permanent shocks (unemployment persistence and imperfect unemployment responsiveness in the terminology of Chapter 1) is estimated with a view to determining if the reforms made a significant difference to the dynamic behavior of the labor market.
One finding of this paper is that the responsiveness of employment to changes in output may have increased after the reforms were instituted. The increased flexibility has been associated with a decline in the time it takes unemployment to respond to shocks. Surprisingly perhaps, in light of the reforms of the 1980s, this analysis does not reveal evidence of significant changes in the wage formation process.
In comparing the recession of the early 1980s with that of the early 1990s, it is striking that the lag in the response of unemployment to changes in national output has quickened much more than the response of employment to changes in output. (See the OECD Economic Survey for the United Kingdom (1995).) However, this observation suggests that, over the past decade, U.K. employment behavior may have changed less than labor force participation behavior. But, because the U.K. labor market reforms were designed to alter employment and wage setting behavior, not to reduce labor force participation in the aftermath of recessions, it is difficult to interpret changes in participation as evidence of improved labor market flexibility as envisaged in the policy reforms.
Some Stylized Facts About the Labor Market
One of the striking features of the U.K. labor market is the dramatic increase in unemployment in the early 1980s (Chart 1). The unemployment rate increased from just over 4 percent in 1979 to about 11 percent by the mid-1980s. This increase was also associated with a substantial increase in the long-term unemployed, who, in 1980, constituted just about 23 percent of the total unemployed. By the mid-1980s, long-term unemployment had doubled to almost 45 percent of the total unemployed, raising concerns about hysteresis in the labor market.
United Kingdom: Unemployment
Sources: Sources: Central Statistical Office. Monthly Digest of Statistics; and Employment Department. Employment Gazette.However, unlike in the other European countries,4 the unemployment rate in the United Kingdom came down sharply following the consumption-led boom associated with financial liberalization in the mid-1980s. In fact, the unemployment rate had fallen to less than 6 percent by the end of 1989 (Chart 1). One question that is analyzed in this paper is whether the fall in the unemployment rate in the late 1980s was also associated with a better trade-off between wage inflation and unemployment. This trade-off is generally considered to be worse for European countries than for the United States. However, because the record of employment creation over this period was significantly better in the United Kingdom than in continental Europe, it is not wise to assume that its trade-off between wage inflation and unemployment necessarily follows the European pattern.
Another notable feature of the U.K. labor market is the significant increase in the participation rates for women (Chart 2), which increased steadily from just over 50 percent in 1970 to 65 percent in 1994. However, this phenomenon was associated with a gradually declining participation rate for men, so that there has not been a significant change in overall participation rates.
United Kingdom: Labor Force Participation and Part-Time Employment
Sources: Central Statistical Office, Monthly Digest of Statistics; Employment Department, Employment Gazette; and OECD, Quarterly Labor Force Statistics.The increased participation of women in the labor force has also been associated with a rise in part-time employment for women (Chart 2) and a rapid growth of employment in the services sector (Chart 3). Labor force surveys in the United Kingdom indicate that an increasing proportion of part-time job choices are becoming voluntary.5 Both employers and employees have the option of “trying out” jobs on part-time or temporary contracts before entering into permanent contracts, so that there is a greater possibility of matching labor supply more closely to labor demand. These developments indicate a greater degree of flexibility in the types of labor contracts that are being offered in the United Kingdom, but not, as argued later in this paper, an improvement in the wage formation process.
United Kingdom: Employment by Sector
(In percent of labor force)
Sources: Central Statistical Office, Monthly Digest of Statistics; and Employment Department, Employment Gazette.Accounting for the Rise in U.K. Unemployment
The methodological approach focuses on major lags in the labor market in order to identify prominent dynamic processes at work. This approach helps overcome some important limitations of the standard model of the nonaccelerating inflation rate of unemployment (NAIRU), which devotes little attention to the dynamics of change. NAIRU models are particularly unconvincing in explaining the unemployment trend in the United Kingdom in the 1980s. The NAIRU, as is well known, is influenced by such factors as the degree of union power, the generosity of the benefits system, and the degree of mismatch in the labor market. The 1980s in the United Kingdom were characterized by declining union power and a tightening of social welfare in general. One would expect the NAIRU to decline on these counts. However, most estimates of the NAIRU show it rising in the 1980s, although it is generally estimated to have fallen more recently.
Quite apart from the general lack of conformity of movements in the NAIRU with a priori expectations, there is also a disturbingly wide range of estimates for it depending on which model of the labor market is used and on precise assumptions made about the behavior of variables, such as the real exchange rate. A recent paper provides a comprehensive survey (Coulton and Cromb (1994)). Table 1, taken from one of the tables in the paper, illustrates the variation in estimates both over time and between modelers. Not surprisingly, the authors conclude that estimates of the NAIRU for the United Kingdom are not reliable enough to serve as a firm basis for economic policymaking or forecasting (Coulton and Cromb (1994), p. 50).
NAIRU Estimates
(In percent standardized on current definition)
NAIRU Estimates
(In percent standardized on current definition)
| 1974–80 | 1981–87 | 1988–90 | |
|---|---|---|---|
| Liverpool (1990) | 7.3 | 7.0 | 3.5 |
| Layard and Nickell (1986) | 4.5 | 5.4 | n.a. |
| Layard, Jackman, and Nickell (1991) | 5.3 | 7.9 | 8.1 |
| Wren-Lewis and others (1990) | n.a. | 9.0 | 7.0 |
| Joyce and Wren-Lewis (1991) | 7.0 | 6.0 | 4.0 |
NAIRU Estimates
(In percent standardized on current definition)
| 1974–80 | 1981–87 | 1988–90 | |
|---|---|---|---|
| Liverpool (1990) | 7.3 | 7.0 | 3.5 |
| Layard and Nickell (1986) | 4.5 | 5.4 | n.a. |
| Layard, Jackman, and Nickell (1991) | 5.3 | 7.9 | 8.1 |
| Wren-Lewis and others (1990) | n.a. | 9.0 | 7.0 |
| Joyce and Wren-Lewis (1991) | 7.0 | 6.0 | 4.0 |
Instead of emphasizing long-run behavior, the present paper tests the extent to which the employment, wage setting, and labor force equations exhibit persistence and imperfect responsiveness. The reason for the emphasis on the effects of lags is that it is difficult to identify a sequence of negative shocks that can fully explain the rising NAIRU in the 1980s in the absence of long-lasting dynamic responses to these shocks within the economy. So, it is necessary to look for potential explanations when temporary shocks have possibly long-lasting effects on the labor market, as well as when permanent shocks hit the labor market.6
There is, in linear models at least, no incompatibility between the dynamic characteristics of the labor market and its long-run behavior, and analysis of one does not preclude the other. Generally, the long run is simply the steady-state part of a dynamic model. However, an important question of emphasis is involved. Many theories of labor market behavior—of wage setting, employment, and participation—suggest that there are considerable time delays in the evolution of wage changes, in the adjustment of employment to shocks, and in labor force responses to higher unemployment. Each of these mechanisms implies that the labor market will show considerable inertia when hit by shocks—a widely recognized phenomenon. In their recent book (summarizing the arguments of their earlier, more extensive, study), Layard, Nickell, and Jackman (1994) emphasize the role of persistence in accounting for the sluggish improvement in unemployment in the member countries of the European Union (EU) after 1982, as contrasted with that in the United States, where it fell rapidly (Layard, Nickell, and Jackman (1994)). They further suggest that this persistence is due exclusively to the duration of benefits and the coordination of wage bargaining. However, this view of the sources of persistence seems narrow and does not emphasize the role of other important lags operating in the labor market.
It is clear that other persistence mechanisms may be at work, ensuring that the long-lasting effects of temporary shocks are an important feature of the labor market. A temporary negative shock that increases unemployment, for instance, also has long-lasting effects. The unemployed miss the opportunity of augmenting human capital through firm-specific training and also tend to lose the habit of working, thereby reducing their probabilities of finding future employment. Unemployment thus has a tendency to perpetuate itself. “Insider-outsider” mechanisms act further to reinforce the persistence of unemployment, as the “insiders,” or the employed workers, collude to prevent real wages from falling. (See Lindbeck and Snower (1988) for a discussion of the insider-outsider theories of unemployment persistence.7) In turn, there is considerable evidence that wage staggering effects are long lasting in the United Kingdom. (See Wadhwani (1985), for example.) Finally, there is substantial evidence of significant employment inertia, owing in turn to sizable adjustment costs faced by firms both in recruiting for and in reducing their workforce. (See Henry (1981) and Nickell (1984), for example.) All this suggests that the evolution of unemployment may be highly “path dependent,” in that the trend in unemployment depends crucially on the starting point and on the nature of the shocks that the labor market is initially subject to.
From the modeling point of view, to capture the sources of inertia in the labor market, it is preferable to focus on the interaction of all the lags in the employment, wage, and labor force equations—as is done here—rather than simply using a single equation for unemployment and testing whether it shows path dependency. (An influential example of the latter used to test for persistence appears in Blanchard and Summers (1986).)
Labor Market Model and Estimation Methods
The model and estimation methods are briefly described below. Further detail on estimation methods is to be found in Chapter 1 of this volume.
Model Equations
The model is one of labor demand, wages, and the labor force. Equations for each are given next, together with a short motivation.
Labor Demand Equations
It has been recognized for some time that employment behavior in the United Kingdom is characterized by a distinct lag pattern, with employment depending upon its first and second lagged values. Employment should depend on past employment in this way for a number of reasons. The reason most favored by economists is that it is costly to change employment, either when hiring or when firing employees. Among these costs are search (by the employer) and training costs when the firm is expanding its workforce, but there are also costs associated with decreasing the workforce, including institutional restrictions on firing and redundancy payments. To allow for such costs, a dynamic model of the following form is used:
where e is employment,y is output, rw is the real wage (all in logs), and the equation assumes that exogenous technical change is taking place, represented by the time trend (t) in the equation. An equation of this form may be derived from the first-order maximizing condition for the profitmaximizing imperfectly competitive firm. Barrell, Morgan, and Pain (1994), for example, derive such an equation, assuming a constant elasticity of substitution (CES) production function. They include materials in the production function, implying that the prices of material inputs affect employment in addition to the terms included in equation (1). The evidence for this is briefly discussed below. The choice of two lagged values of employment reflects the assumption that adjustment costs are quadratic. (See Hall and Henry (1988) for a fuller discussion of adjustment cost models.) Although it is clear that this assumption could be violated by, for example, asymmetric costs (such as when the costs of expanding the workforce exceed those of an equivalent decrease, as described in Burgess (1993)), the assumption of quadratic costs is nonetheless both plausible and convenient.
Wage Equations
The wage equation used here is based on a “competing claims” model of union-firm behavior, as is the labor demand equation above. The wage (and price or, alternatively, labor demand) is then the outcome of a bargain between the firm and the union, where—in the “right to manage” version—once the wage has been determined, the firm sets the level of employment according to an equation like (1) above. Underlying the union’s behavior is a utility function that depends on employment, the real wage, and earnings elsewhere—the reservation wage, which depends, in turn, on unemployment benefits. Models of this sort produce an equation for the real wage dependant upon real unemployment benefits, productivity (or real profits per head), unemployment, and indices of union strength (see Layard, Nickell, and Jackman (1991)). One addition emphasized here is given by the phenomenon of wage staggering, which occurs when there are overlapping contracts in different sectors (or unions) in the economy. As emphasized in the seminal article by Taylor (1980), wage staggering is one reason why aggregate wages will then depend on past wages, and, if unions and firms are forward looking, on future expected wage settlements also. Although forward-looking expectations are not explicitly incorporated, lags in real wages are allowed for with a dynamic equation for the real wage of the following form:
where Z is a set of variables affecting real wage bargaining outcomes, such as unemployment, union strength, and productivity. In equation (2), two lags are proposed, but, as shown, the lag length will be determined by statistical criteria because, although theory suggests there may be lags, it does not indicate what the lengths of the lags are likely to be.
Labor Force Equation
The labor force equation is taken to be a simple equation where the labor force depends on the working population and unemployment, which depends on the discouragement effect—whereby higher levels of unemployment discourage workers from participating in the workforce—and the added-worker effect—which entails higher participation as unemployment increases. Hence, a priori, the sign of γ3 is ambiguous. Lagged effects can be expected to operate through the lagged effects of unemployment on the cumulative experience of would-be labor market participants. Such dynamic possibilities are allowed for in the labor force equation (3) below, which is the basis of the empirical models reported subsequently:
Estimation Methods
Each of the equations described above is in the form of an autoregressive distributed lag (ARDL) model, which in its simplest form has only one lag on each variable (that is, it is of the order (1.1)) and may be written:
where |α1|<1, and €t is assumed to be white noise.
In this equation, Y and X are assumed to be nonstationary variables. In current parlance, this means that they are at least I(1) variables, and they need to be differenced once to make them stationary. The following steps are taken to estimate this equation:
(1) Test for the presence of cointegration between all the I(1) variables using Johansen’s maximum likelihood procedure.
(2) Estimate the long-run relationship between the variables using the method of identifying a cointegrating relation from an ARDL equation as recently proposed by Pesaran and Shin (1995).
(3) Using this relationship as the long-run equilibrium term, estimate the short-run parameters of the model using its equivalent error correction model (ECM) representation, comprising stationary variables throughout, thus ensuring that the short-run parameters are estimated consistently. Where there are current values of other endogenous variables on the right-hand side of an equation, they are taken into account by using instrumental variable (IV) estimation.
Estimation Results
This section describes first the time-series characteristics of the data and then the estimated model equations.
Time-Series Properties of Variables
The orders of integrations of the variables are first tested using Dickey-Fuller (DF) and Augmented Dickey-Fuller (ADF) (of order 4, ADF (4)) tests. The results are shown in Table 2. Unless otherwise stated, the sample period is 1968Q3–1994Q3.
Tests of Integration
Tests of Integration
| Level | Differences | |||
|---|---|---|---|---|
| Variable | DF | ADF(4) | DF | ADR (4) |
| LE | -1.77 | -3.17 | -4.93 | -3.07 |
| LGDP | -2.04 | -2.71 | -11.85 | -4.44 |
| LRWP | -2.24 | -1.68 | -15.85 | -5.54 |
| LK | -1.74 | -1.85 | -4.55 | -2.29 |
| LRWC | -2.24 | -1.73 | -14.03 | -5.13 |
| U | -0.90 | -2.37 | -3.52 | -3.52 |
| LPROD | -2.18 | -2.30 | -13.3 | -4.86 |
| LPOD | -1.02 | -0.14 | -8.15 | -2.79 |
| LL | -1.3 | -0.89 | -9.07 | -2.91 |
| RR | -5.07 | -2.97 | - | - |
| REP | -6.03 | -3.49 | - | - |
| WEDGE | -3.36 | -3.72 | - | - |
| LCOMP | -1.4 | -2.1 | -8.9 | -5.4 |
| LROILP | -1.6 | -1.2 | -11.46 | -6.36 |
Tests of Integration
| Level | Differences | |||
|---|---|---|---|---|
| Variable | DF | ADF(4) | DF | ADR (4) |
| LE | -1.77 | -3.17 | -4.93 | -3.07 |
| LGDP | -2.04 | -2.71 | -11.85 | -4.44 |
| LRWP | -2.24 | -1.68 | -15.85 | -5.54 |
| LK | -1.74 | -1.85 | -4.55 | -2.29 |
| LRWC | -2.24 | -1.73 | -14.03 | -5.13 |
| U | -0.90 | -2.37 | -3.52 | -3.52 |
| LPROD | -2.18 | -2.30 | -13.3 | -4.86 |
| LPOD | -1.02 | -0.14 | -8.15 | -2.79 |
| LL | -1.3 | -0.89 | -9.07 | -2.91 |
| RR | -5.07 | -2.97 | - | - |
| REP | -6.03 | -3.49 | - | - |
| WEDGE | -3.36 | -3.72 | - | - |
| LCOMP | -1.4 | -2.1 | -8.9 | -5.4 |
| LROILP | -1.6 | -1.2 | -11.46 | -6.36 |
The variables are all in logs, except U, REP, WEDGE, and RR, and are employment (LE), output (LY), the real product wage (LRWP), the capital stock (LK), the real consumption wage (LRWC), the unemployment rate (U), labor productivity (LPROD), the working population (LPOP), the labor force (LL), the real interest rate (RR), the replacement ratio (REP), the tax wedge (WEDGE), competitiveness (LCOMP), and the real oil price (LROILP).
While most of these statistics are unambiguous, the results in a couple of cases are rather more difficult to interpret. For example, the capital stock has a low ADF when tested for integration in differences. This appears to be due to a pronounced long cycle in the data, attributable in part to the effects of premature scrapping of capital equipment in the first half of the 1980s. Nonetheless, the variable is treated as I(1). Similarly, the evidence that the unemployment rate may be integrated at an order greater than unity is discounted. Even treating it as I(1) raises familiar problems: in principle, the unemployment rate cannot be I(1) because it is bounded between 0 and 1. However, it can clearly appear I(1) in short samples, as here.
But one clear and important result is that the replacement ratio and the tax wedge are both stationary. The importance of this result is that neither of these variables is expected to be part of the explanation of the real wage in the long run, because the real wage is nonstationary, and stationary variables cannot statistically “account” for nonstationary ones. Hence, they do not figure in the tests reported later on the long-run real wage equation.
Labor Demand Equation
Tests for cointegration among the set of relevant variables (in logs) showed that a set involving employment (LE), output (LY), the real product wage (LRWP), the real oil price (LROILP), and a deterministic time trend (treated as a strongly exogenous variable) cointegrates. As noted earlier, inclusion of real oil prices implies a production function that includes energy as an input (and this is not assumed to be separable from other inputs). This version was estimated by IV owing to the presence of the current value of output and the real wage as independent variables in the employment equation. The IV estimates are shown in Table 3, which are used in the subsequent discussion.
Employment Equations
is a Lagrange multiplier test of residual correlation; 
is the Ramsey RESET test; 
is a test for normality of residuals; 
is a test for heteroscedasticity; and 
is Sargan’s test that the instruments are valid. LR(.) is the likelihood ratio given by Johansen’s trace statistic, that r = 0, where r is the number of cointegrating vectors, and where the value shown as the second number in parentheses is the 95 percent critical value.
Employment Equations
| Dependent Variable |
Independent Variables | ||||||
|---|---|---|---|---|---|---|---|
| Constant | LY | LRW | ROILP | T | (p, q) | LR | |
| LE | 0.13 | 0.81 | -0.25 | -0.001 | -0.002 | (2,0,00) | 22.4 (20.9) |
| (0.90) | (3.43) | (1.3) | (2.9) | (2.7) | |||
| Constant | ΔLE(-1) | ΔY | ΔLRW | ΔROILP | ΔECM | ||
| ΔLE | 0.13 | 0.55 | 0.05 | -0.02 | 0.0 | -0.06 | |
| (2.7) | (7.4) | (3.0) | (1.3) | (2.95) | (4.45) | ||
| R2=0.56, | |||||||
|
|||||||
is a Lagrange multiplier test of residual correlation; is the Ramsey RESET test; is a test for normality of residuals; is a test for heteroscedasticity; and is Sargan’s test that the instruments are valid. LR(.) is the likelihood ratio given by Johansen’s trace statistic, that r = 0, where r is the number of cointegrating vectors, and where the value shown as the second number in parentheses is the 95 percent critical value.
Employment Equations
| Dependent Variable |
Independent Variables | ||||||
|---|---|---|---|---|---|---|---|
| Constant | LY | LRW | ROILP | T | (p, q) | LR | |
| LE | 0.13 | 0.81 | -0.25 | -0.001 | -0.002 | (2,0,00) | 22.4 (20.9) |
| (0.90) | (3.43) | (1.3) | (2.9) | (2.7) | |||
| Constant | ΔLE(-1) | ΔY | ΔLRW | ΔROILP | ΔECM | ||
| ΔLE | 0.13 | 0.55 | 0.05 | -0.02 | 0.0 | -0.06 | |
| (2.7) | (7.4) | (3.0) | (1.3) | (2.95) | (4.45) | ||
| R2=0.56, | |||||||
|
|
|||||||
is a Lagrange multiplier test of residual correlation; is the Ramsey RESET test; is a test for normality of residuals; is a test for heteroscedasticity; and is Sargan’s test that the instruments are valid. LR(.) is the likelihood ratio given by Johansen’s trace statistic, that r = 0, where r is the number of cointegrating vectors, and where the value shown as the second number in parentheses is the 95 percent critical value.
Wage Equation
Cointegration appears to exist between the real consumption wage, unemployment, productivity, and competitiveness, which apparently gives a statistical explanation of the real wage in the long run. On estimating the long-run equation from an ARDL as described earlier, two extra refinements were included. The first was to impose the productivity effect at unity. This restriction was easily accepted and is plausible. The second was to include a dummy variable taking the value 1,1, -1 for 1974Q3 to 1975Q1. This was necessary to deal with outlining observations attributable to the effects of the three-day week. The equation estimated by instrumental variables is shown in Table 4.
Wage Equation
Wage Equation
| Dependent Variable | Independent Variables | |||
|---|---|---|---|---|
| Constant | UR | LCOMP | (p.q) | |
| WageProd | 3.5 | -0.95 | -0.11 | (1,0,1) |
| (14.1) | (5.54) | (2.27) | ||
| LR(69.4; 27.1) | ||||
| Δ(WageProd) | Constant | ΔUR | ΔLCOMP | ECM(-1) |
| 0.58 | -0.16 | 0.07 | -0.16 | |
| (4.9) | (3.9) | (2.1) | (4.6) | |
| R2= 0.33, | ||||
|
|
||||
Wage Equation
| Dependent Variable | Independent Variables | |||
|---|---|---|---|---|
| Constant | UR | LCOMP | (p.q) | |
| WageProd | 3.5 | -0.95 | -0.11 | (1,0,1) |
| (14.1) | (5.54) | (2.27) | ||
| LR(69.4; 27.1) | ||||
| Δ(WageProd) | Constant | ΔUR | ΔLCOMP | ECM(-1) |
| 0.58 | -0.16 | 0.07 | -0.16 | |
| (4.9) | (3.9) | (2.1) | (4.6) | |
| R2= 0.33, | ||||
|
|
||||
The wage levels equation is fairly satisfactory and includes a well-determined unemployment effect. Competitiveness also appears to be a significant influence on the level of wages according to this result. This term can be interpreted as arising from external effects when the wage bargaining model is extended to the open economy—see Layard, Nickell, and Jackman (1991).
The model thus suggests a well-determined effect from unemployment on the level of the real wage, and the effect for changes in unemployment is significant. A change effect from unemployment in the wage equation is often interpreted as showing hysteresis in wage behavior, but, as emphasized already, other persistence mechanisms are working in the model as a whole.
Labor Force Equation
General evidence of cointegration was found among the key variables—the working population, unemployment, and the labor force—although this effect was not strong.
The best results were obtained using the participation ratio (labor force/working population), and in what follows this is used as the dependant variable (defined as LPART). Again, the long-run and dynamic (ECM) version of the distributed lag model estimated by IV is provided (see Table 5).
Participation Equation
Participation Equation
| Dependent Variable | Independent Variables | |||||
|---|---|---|---|---|---|---|
| UR | Constant | (p.q) | ||||
| LPART | -0.28 | 0.31 | (2,2) | |||
| (1.1) | (14.7) | |||||
| LR (29.7; 29.7) | ||||||
| ΔPART | Constant | LPART(-1) | Δ UR | Δ UR(-1) | ECM(-1) | |
| 0.008 | 0.21 | 0.1 | 0.13 | -0.021 | ||
| (1.6) | (2.48) | (1.65) | (2.2) | (1.68) | ||
| R2= 0.18, | ||||||
|
||||||
Participation Equation
| Dependent Variable | Independent Variables | |||||
|---|---|---|---|---|---|---|
| UR | Constant | (p.q) | ||||
| LPART | -0.28 | 0.31 | (2,2) | |||
| (1.1) | (14.7) | |||||
| LR (29.7; 29.7) | ||||||
| ΔPART | Constant | LPART(-1) | Δ UR | Δ UR(-1) | ECM(-1) | |
| 0.008 | 0.21 | 0.1 | 0.13 | -0.021 | ||
| (1.6) | (2.48) | (1.65) | (2.2) | (1.68) | ||
| R2= 0.18, | ||||||
|
|
||||||
Although this is clearly a simplified equation, it has certain interesting features. The effects of unemployment on participation show some disparity between short- and long-run effects. In the short run, higher unemployment appears to encourage participation, suggesting an added-worker effect, although in the long run, this effect appears to go the other way, implying that a discouraged-worker effect dominates. However, this long-run effect is less well determined.
Evaluating the Effects of the Reforms
Subsample Estimates: Evidence for Structural Change
There has been considerable debate about the possible effects on labor market behavior of the reforms of the 1980s. That substantial changes in some key features of the labor market occurred over the 1980s is apparent: employment grew rapidly in the late 1980s, fueled in part by increases in part-time and self-employment. It also fell both more rapidly and earlier in the 1990s recession than in previous recessions. As noted earlier, policy initiatives in trade union reform, redundancy legislation, and the introduction of tighter unemployment legislation could be expected to produce greater labor market flexibility. The attendant movement toward more decentralized wage bargaining may, however, have neutralized this improved flexibility—as it is generally believed that more centralized wage bargaining can produce a better trade-off between unemployment and wage inflation. A seminal article on this issue is that by Calmfors and Driffill (1988), who found that the relation between centralization and labor market performance was hump-shaped, but their, and subsequent, discussion has suggested that the United Kingdom may be near the top of this hump, implying that either more or less centralization might improve the country’s performance.
On the macroeconomic effects of the reforms, opinions appear to differ. Anderton and Mayhew (1994) are skeptical that a major improvement has occurred. In their view, the pattern of real wage growth has been difficult to alter, and they suggest that, in consequence of continued real wage resilience (and the failure of substantial measures to improve skills and training given the large structural changes in the economy away from manufacturing to service industries), excessively high levels of unemployment are needed to maintain downward pressure on wage inflation. (Similar skepticism on the effects of the 1980s policies on wage developments is found in Metcalf (1994).) Others have pointed to the existence of positive signs of improved flexibility attributable to policy, including decreasing costs of laying off workers as demand falls, without commensurately increasing the costs of taking workers on when demand increases, and reducing union power and increasing work incentives. The OECD Economic Survey for the United Kingdom notes possible improvements in labor market flexibility in the country’s recent behavior, including the faster response of unemployment (OECD (1995)).
The evidence for improved flexibility is considered using the estimated model presented earlier. The strategy adopted is to establish whether the model shows changes when estimated over subsamples and whether these changes could be interpreted as the effect of policy changes on employment, wage setting, and participation. Some caution must be applied to any such interpretation of the evidence. An apparent change in the adjustment of employment to changes in output may, for example, be prima facie evidence that employment decisions have become more flexible. There may, however, be an alternative explanation for it, such as labor saving technical change. That said, this discussion focuses on evidence for changes in the adjustment of employment, as shown by the output effect on employment and the estimated lag structure in the labor demand equation; and on alterations in the degree of real wage rigidity as estimated by changes in the unemployment effect on real wages.
We look for evidence of parameter change in the equations we estimated earlier when these are re-estimated over data samples that end in 1990, and, separately, in 1991. These dates are taken because they allow time for the effects of the policy reforms, if there are any, to be felt. We have also initiated the tests long enough after the onset of recession to prevent the tests from being contaminated by the model’s failure to predict the downturn. Although statistical methods exist for selecting dates when changes in underlying models occur, these are rather technical (see, for example, Banerjee and Urga (1995)), and the method employed here—although less refined—has the merit of being likely to capture significant changes in behavior if they have occurred.
The first test entailed re-estimating the employment equation over two periods beginning in the early 1990s, that is, dropping the observations from 1990Q1 from the data sample in one case and then redoing the test starting one year later (that is, dropping sample data from 1991Q2 onward, inclusive). The summary tests of model parameter stability were obtained (see Table 6). For the wage equation, only the first test was done, for reasons that are outlined below. In all cases, the model was used to predict the level of the variable (employment and the real wage, respectively).
Summary Tests of Parameter Stability
Summary Tests of Parameter Stability
| Estimation Sample Ending |
PRED | CHOW | |
|---|---|---|---|
| Employment | 1990Q1 | 18.67 (28.9) | 15.3(16.9) |
| 1991Q1 | 13.4 (23.7) | 11.7(16.9) | |
| Wages | 1990Q1 | 18.79(28.9) | n.a. (…) |
Summary Tests of Parameter Stability
| Estimation Sample Ending |
PRED | CHOW | |
|---|---|---|---|
| Employment | 1990Q1 | 18.67 (28.9) | 15.3(16.9) |
| 1991Q1 | 13.4 (23.7) | 11.7(16.9) | |
| Wages | 1990Q1 | 18.79(28.9) | n.a. (…) |
According to the tests, there is no significant evidence that the parameters of the employment equation have changed in the 1990s compared with the experience of the 1960s and 1980s. The tests are not powerful ones, however, and it is arguable from inspection of the predictions this model makes that a change in employment behavior has occurred, which is largely one of a stronger and more rapid response of employment to output.
Charts 4 and 5 show two ex ante predictions made by the employment model over the two periods—from 1990Q2 and from 1991Q2, respectively, until 1994Q3. These indicate that the model does indeed overpredict employment when its forecast is initiated from 1990Q2 (Chart 4). By 1994Q3, this overprediction amounts to almost 5 percent (of the employed labor force). As Chart 5 shows, however, if the model is initiated a year later (1991Q2), then its cumulative forecast error over the next 3½ years is negligible, amounting to just over 100,000 by the end of the period (less than ½ of 1 percent). Although the second forecast is begun well after the recession started, the first test encounters the well-known problem of the inability of time-series models to capture a cyclical downturn. But the discrepancy in the forecast performance of the model over these two periods is suggestive, if nothing more, of a break in the behavior of employment and its main determinants in the early part of the recession. Employment appears to have declined rapidly and by substantial amounts in the early 1990s compared with its previous behavior. It is too early to say whether this changed behavior will be repeated in the upturn, with increases in employment becoming more rapid and so matching the more rapid decreases in employment that occurred in the downturn. The evidence provided by the forecast performance of this employment model is consistent with the analysis that the change in employment is more of a step down, or onetime “shakeout,” of employees (which the first forecast suggests), but that once this change occurred, the previous pattern of behavior was re-established (that is, as the second forecast suggests).
United Kingdom: Employment Equation: Dynamic Forecast for 1990Q2–1994Q3
United Kingdom: Employment Equation: Dynamic Equation for 1991Q2–1994Q3
There is a further complication, which this exercise unfortunately does not shed light on, and that is whether the behavior of employment in this recession is different from that in previous ones. There is now increasing evidence that U.K. employment behavior is asymmetric. During recessions, in particular, decreases in employment are sharper and continue for shorter periods than increases in employment. In a recent study, Acemoglu and Scott (1994) estimate that the rate of job losses in a cyclical downturn may be as much as four times greater than job creation in an upturn. The more precise question raised by such evidence is whether the sharp rate of job losses in the 1990s recession is “normal” recessionary behavior or whether it marks a significant change from normal, with employment declining even more rapidly than in other recessions. The evidence is too limited to answer this question yet.
Turning to the wage equation, according to the test in Table 6, there is little evidence of any change in the relationship between real wages, productivity, and unemployment. The relationship seems stable. Indeed, there is evidence that, over much of the 1990s, the model actually underpredicts the real wage (see Chart 6); the real wage in much of the 1990s has actually been higher than might have been expected from the relationships fitted to earlier data. But overall, and more precisely, the result indicates that there is little significant change in wage setting in the 1990s compared with earlier periods.8
United Kingdom: Wage Equation: Dynamic Forecast for 1990Ql–1994Q3
Simulations of Dynamic Responses
To obtain an overall quantitative measure of the dynamic behavior of the labor market, the complete model described earlier is used. As described in Chapter 1 of this volume, the method involves administering simple shocks to the model and calculating the response of wage employment and unemployment to the shocks using the estimated model. In each case, the solution is conducted over a long period of time to give the model time to settle down. Both a temporary and a permanent shock are applied, and their effects calculated, to estimate how the labor market responds to shocks that disturb it temporarily, such as a temporary productivity shock, but also to establish how the market moves from one equilibrium to another when one of the determinants of long-run behavior, such as a permanent change in competitiveness, changes.
Summary indicators of the dynamic behavior of the U.K. labor market, which our earlier statistical modeling implies, are shown in Table 7. These show the single-equation results for employment, wages, and participation. Whole model indicators of persistence, whereby for the whole model the mean lag following a temporary shock is cited—estimated using full model simulations—are given for unemployment.
Labor Market Dynamics
(Persistence in years)
Reference is to the present study.
Labor Market Dynamics
(Persistence in years)
| Henry and Karanassou (1996)1 |
Alogoskoufis and Manning (1988) | Barrell, Morgan, and Pain (1994) | Layard, Nickell, and Jackman (1991) | |
|---|---|---|---|---|
| Employment | 6 | … | … | 3 |
| Wages | 1.5 | 4.3 | 2.8 | … |
| Participation | 12 | … | … | … |
| Unemployment | 5 | 12.5 | … | 2.2 |
Reference is to the present study.
Labor Market Dynamics
(Persistence in years)
| Henry and Karanassou (1996)1 |
Alogoskoufis and Manning (1988) | Barrell, Morgan, and Pain (1994) | Layard, Nickell, and Jackman (1991) | |
|---|---|---|---|---|
| Employment | 6 | … | … | 3 |
| Wages | 1.5 | 4.3 | 2.8 | … |
| Participation | 12 | … | … | … |
| Unemployment | 5 | 12.5 | … | 2.2 |
Reference is to the present study.
The values in the table show that, on our estimates, there is evidence of substantial sluggishness in each part of the labor market except in wages. Our estimates tend to be higher than some others, although Alogoskoufis and Manning (1988) have higher estimates of persistence in real wages and in unemployment than we report. As we have made some effort both to identify behavioral models of the key labor market variables involved and to estimate these allowing for long-run equilibrium behavior, there are reasons for thinking our estimates are to be preferred. It is striking, however, that even after allowing for the presence of long-run equilibrium in employment, wages, and the participation rate, the evidence we find of considerable persistence in some single equations is strong.
Table 7 illustrates the diversity of views on the overall measure, that of unemployment persistence. The estimates by Alogoskoufis and Manning (1988) are flawed because they do not allow for the possible existence of cointegrating (or long-term equilibrium) relations in the labor market. In turn, Layard, Nickell, and Jackman (1991) suggest that persistence is relatively unimportant, although they use very simple specifications for the dynamics in their model and do not investigate the possibility of cointegration. According to our estimates of persistence, based on a simulation of the complete macro model, unemployment takes about five years to reach halfway to its equilibrium once the model is hit by either a temporary or a permanent shock. This estimate falls roughly halfway between the two estimates quoted in Table 7, but, for reasons already mentioned, we think it is a more refined and plausible estimate.
Following a permanent adverse shock to productivity, in our model, unemployment again approaches a higher equilibrium rate monotonically but slowly, as is to be expected. Again, it takes about five years to get halfway to the new equilibrium rate (see appendix to Chapter 1 in this volume).
Conclusions
The principal conclusion emerging from this paper is that the U.K. labor market is strongly subject to inertia. There is some evidence that the behavior of the labor market may have changed following the reforms of the 1980s, although the tests applied here are not able to pinpoint precisely what this change represents. However, it appears that employment decreased faster in the 1990s recession than would be predicted on the basis of previous behavior, at least initially. From 1991 onward, behavior consistent with previous relationships may have resumed. Thus, although there is some evidence of an improvement in employment flexibility, it is too early to rule out the possibility that it is an employment response to recession similar to that which occurred in the 1980s recession. While the analysis of the paper does not find evidence that the wage formation process changed significantly following the implementation of the labor market reforms, recent wage behavior suggests that this area merits further study.
Finally, the response of unemployment to the changes in output is very different from those in the last recession. The main reason for this difference is a large change in participation rates (as the OECD) points out in its comments on the apparent shortening of the lag between output and unemployment). There is still little evidence that employment has changed its adjustment to output in the upturn. Furthermore, although there have been substantial changes in participation, it is difficult to regard these as the objective of, or indeed the result of, the labor market reforms of the 1980s.
References
Acemoglu, Daron, and Andrew Scott, 1994, “Asymmetries in the Cyclical Behavior of U.K. Labor Markets,” Economic Journal, Vol. 104 (November), pp. 1303–23.
Alogoskoufis, George, and Alan Manning, 1988, “On the Persistence of Unemployment,” Economic Policy: A European Forum, Vol. 3 (October), pp. 428–69.
Anderton, R., and K. Mayhew, 1994, “A Comparative Analysis of the U.K. Labour Market,” in The U.K. Labour Market, ed. by R. Barrell (Cambridge, England: Cambridge University Press).
Banerjee, A., and G. Urga, 1995, “Modelling U.K. Trade: An Exercise in Sequential Structural Break Procedures,” Discussion Paper No. 24–95 (London: London Business School).
Barrell, R., J. Morgan, and N. Pain, 1994, “Employment Inequality and Flexibility” (unpublished; London: National Institute for Economic and Social Research).
Bean, Charles, 1994, “European Unemployment: A Survey,” Journal of Economic Literature, Vol. 32 (June), pp. 573–619.
Blanchard, Oliver, and Laurence Summers, 1986, “Hysteresis and the European Unemployment Problem,” NBER Macroeconomics Annual 1986, Vol. 1 (Cambridge, Massachusetts: MIT Press).
Brown, William, and Sushil Wadhwani, 1990, “The Economic Effects of Industrial Relations Legislation Since 1979,” National Institute Economic Review, No. 131 (February), pp. 57–70.
Burgess, S., 1993 “Nonlinear Dynamics in a Structural Model of Employment,” Nonlinear Dynamics, Chaos, and Econometrics, ed. by M. Pesaran and Simon Potter (Chichester, England; New York: Wiley).
Calmfors, Lars, and John Driffill, 1988, “Centralization of Wage Bargaining,” Economic Policy: A European Forum, Vol. 3 (April), p. 13.
Coulton, B., and R. Cromb, 1994, “The U.K. NAIRU,” Government Economic Service Working Paper No. 124 (London: H.M. Treasury).
Hall, S.G., and S.G.B. Henry, 1988, Macroeconomic Modelling (Amsterdam: North-Holland).
Henry, S.G.B., 1981, “Forecasting Employment and Unemployment,” in The Economics of the Labour Market, ed. by Z. Hornstein, J. Grice, and A. Webb (London: HMSO).
Layard, R., S. Nickell, and R. Jackman, 1991, Unemployment: Macroeconomic Performance and the Labor Market ((Oxford, England New York; Oxford University Press).
Layard, R., S. Nickell, and R. Jackman, 1994, The Unemployment Crisis (Oxford, England; New York: Oxford University Press, rev. ed.).
Lindbeck, Assar, and Dennis Snower, 1988, The Insider-Outsider Theory of Employment and Unemployment (Cambridge, Massachusetts: MIT Press).
Metcalf, D., 1994, “Transformation of British Industrial Relations? Institutions, Conduct and Outcomes, 1980–1990,” in The U.K. Labour Market, ed. by R. Barrell (Cambridge, England; New York: Cambridge University Press).
Nickell, S., 1984, “The Modelling of Wages and Employment.” in Econometrics and Quantitative Economics, ed. by D. Hendry and K. Wallis (Oxford, England New York; Blackwell).
Organization for Economic Cooperation and Development, 1993, Employment Outlook (Paris: OECD).
Organization for Economic Cooperation and Development, 1994, The OECD Jobs Study—Evidence and Explanations, Part 2, The Adjustment Potential of the Labor Market (Paris).
Organization for Economic Cooperation and Development, 1995, OECD Economic Surveys; United Kingdom (Paris).
Pesaran, H., and Y. Shin, 1995, “An Autoregressive Distributed Lag Modelling Approach to Cointegration Analysis,” DAE Discussion Paper No. 9514 (Cambridge, England: University of Cambridge, Department of Applied Economics).
Ramaswamy, R., and E. Prasad, 1994, “Shocks and Structural Breaks: Labor Market Reforms in the United Kingdom,” IMF Working Paper 94/152 (Washington: International Monetary Fund).
Robinson, Peter, 1994, “British Labour Market in Historical Perspective: Changes in the Structure of Employment and Unemployment,” Centre for Economic Performance Discussion Paper No. 202 (London: London School of Economics and Political Science).
Taylor, John B., 1980, “Aggregate Dynamics and Staggered Contracts,” Journal of Political Economy, Vol. 88 (February), pp. 1–23.
Wadhwani, S. B., 1985, “Wage Inflation in the United Kingdom,” Economica, Vol. 52 (May), pp. 195–207.
Note: Grateful thanks are extended to Ramana Ramaswamy for his help in producing this paper. We would also like to thank Professors W. Buiter, R. Smith, and D. Snower for comments.
A detailed analysis of the legislative changes in the labor market in the 1980s can be found in Brown and Wadhwani (1990) and Ramaswamy and Prasad (1994).
Also, unemployment benefits have been replaced by the Jobseekers’ Allowance, which has reduced the duration of benefits to six months for most applicants.
See Organization for Economic Cooperation and Development (1993 and 1994) for further details on labor market legislation.
See Bean (1994) and other chapters in this book.
See Robinson (1994) for a discussion of the long-term trends in participation rates and part-time employment. The proportion of part-time workers who say they are working part time because they are unable to find full-time work (involuntary part-time working) was found to be just 6 percent of the total in 1990, having declined from 10 percent in 1987. Robinson’s study shows that the extent of involuntary part-time work in the United Kingdom is below the average for most other industrial countries—a particularly favorable finding, given that the country has an above-average share of part-time workers.
Some of the conceptual issues relating to this theme are explored in Bean (1994). See also, in this context. Blanchard and Summers (1986).
Insider-outsider theories are based on the assumption that there are large transactions costs in replacing existing workers with the unemployed. Such costs include not only hiring and firing costs, but also more subtle mechanisms whereby insiders as a group refuse to cooperate with outsiders, thus increasing the effective costs of employing the latter.
This uses the model with productivity constrained to have a unit effect. Relaxing the constraint, however, gave a result similar to that described in the text, and the formal test still rejects parameter instability [X2(18) = 20.6].
