THE WAGE CURVE: FINNISH EVIDENCE

Transcription

1

2 THE WAGE CURVE: FINNISH EVIDENCE TUOMAS PEKKARINEN * ABSTRACT In this paper we study the wage curve with both cross-sectional and pooled Finnish data. Hourly wage is used as a dependent variable throughout the paper as well as a fairly detailed definition of the local labour market. Results indicate that there exists a relationship similar to downward sloping wage curve in the Finnish data. Results are sensitive to the inclusion of regional fixed effects. Keywords: Wage curve, local unemployment. JEL Classification: J31, J64. * I would like to thank Erkki Koskela for helpful comments as well as the Confederation of Finnish Industry and Employees (Teollisuus ja työantajat), Juhana Vartiainen and Aila Mustonen at the Labour Institute for Economic Research, Pekka Myrskylä at the Statistics Finland and Ilkka Nio at the Ministry of Labour for providing the data. The usual disacclaimer applies. LABOUR INSTITUTE FOR ECONOMIC RESEARCH DISCUSSION PAPERS 144 HELSINKI 1998

3 ISBN ISSN

4 1. INTRODUCTION The wage curve is an empirical observation originally reported by Blanchflower and Oswald (1990, 1994, 1995), that describes a relationship between a worker s pay and the unemployment rate in the local labour market. The causality is thought to run from the unemployment rate to the wage rate. The wage curve is estimated using microeconomic data and a standard microeconometric wage function with the local unemployment rate as an additional independent variable. Original studies by Blanchflower and Oswald use repeated cross-sectional data from 12 different countries, containing information on approximately 3,5 million individuals. 1 The results achieved by Blanchflower and Oswald indicate that there exists a negative relationship between local unemployment and wages. A worker who is employed in a high unemployment region is expected to be paid a lower wage than a correspondent worker in a region with low unemployment. According to Blanchflower and Oswald (1994) the wage curve relation is well-approximated by a simple log-linear function of the following form: ln w = -0.1 ln U r + other variables In other words the coefficient of the logarithm of the local unemployment rate U r takes approximately the value of -0.1 in an equation where the 1 The countries studied by Blanchflower and Oswald are United States, Great Britain, Canada, South-Korea, Austria, Italy, Netherlands, Switzerland, Norway, Ireland, Australia and Germany. The largest single regression uses data of over 1,7 million individuals. 3

5 dependent variable is the logarithm of the individual s wage and other variables are a typical set of measured individual characteristics, such as gender, age and education. This relationship can be drawn as a negatively sloped, convex curve in a wage-unemployment -space and, following the example of Blanchflower and Oswald (1990), it has been called the wage curve in the literature. The finding of a negative relationship between local unemployment and the worker s wage contradicts traditional thinking on the wage-unemployment relationship, which is based on the theoretical model of Harris and Todaro (1970). The model predicts that wages and unemployment are positively correlated across regions due to the compensating differentilas 2. Furthermore Blanchflower and Oswald argue that this negative relationship is an international phenomenon. The estimated coefficient of the local unemployment rate takes approximately same values regardless of the country or period which the data are from. After the original contributions of Blanchflower and Oswald (1990, 1994,) a number of recent studies have documented the existence of the wage curve in micro data sets from several countries (Groot et al. 1992; Wagner 1994; Bratsberg and Turunen 1996; Hoddinot 1996). The wage curve seems to be a general result which holds regardless of the institutional structure of the labour markets under study. This has led authors like Blanchflower and Oswald (1994, 1) and Card (1995, 798) to claim it an empirical law of economics. In this paper we set out examine whether there exists a similar kind of relationship in Finnish data. There is only one earlier example of Finnish wage curve estimations in the literature. The Finnish data used here allow us to distinguish between workers annual and hourly wage. This distinction is important since for example commentators like Card (1995) have pointed out that the original results of the studies by Blanchflower and Oswald 2 As an interpretation of the wage curve results Blanchflower and Oswald (1994) present three different theoretical models that give an outcome which is similar to downward sloping wage curve. All the models presented by Blanchflower and Oswald are based on the assumption of imperfect competition in the labour market 4

6 (1994) may be biased due to the authors failure to control for the changes in the number of working hours by using simply annual or monthly wages as dependent variables. This possible source of bias can be avoided with the use of hourly wages. The data also make it possible to define the worker s local labour market as a regional county (seutukunta) of which there are 88 in Finland. This may be considered as a fairly detailed definition of the local labour market and it should allow for a efficient control of region-specific fixed effects. First the earlier results relevant to this paper are studied shortly. These include the small body of Scandinavian results as well as the only earlier Finnish study by Parjanne (1997). The data used here are then presented in the third section. The fourth section presents the wage curve results. In the concluding section the interpretation of the results is discussed. 5

7 2. SCANDINAVIAN WAGE CURVE RESULTS There exists only one earlier wage curve study conducted on Finnish data. Parjanne (1997) studies the flexibility of the hourly wages with four different cross-sectional data sets. The data are from the annual Labour Market Surveys conducted by Statistics Finland. Both regional and industry-specific unemployment rates are used as independent variables. Parjanne divides Finland into 13 regional labour markets. The results from the crosssectional regressions obtained by Parjanne (1997) support the hypothesis that there exists a relationship similar to downward sloping wage curve in the Finnish data. Estimated coefficients of the regional as well as the industry-specific unemployment rates are negative, significant and relatively close to the standard result of Blanchflower and Oswald (-0.1). The study does not include results from a pooled data. Small body of Scandinavian results also exist. As Finnish labour markets are often grouped together with other Scandinavian countries and Austria as representing a typical corporatist labour market structure, it is useful to review other Scandinavian results before examining the results obtained here. 3 Original contributions by Blanchflower and Oswald (1994) contain results estimated on Norwegian data that support the wage curve hypothesis. The results from pooled data, covering the period of , are sensitive to the inclusion of regional fixed effects in the regression: the coefficient of the local unemployment rate remains negative but does not reach significance. However according to Blanchflower and Oswald (1994, 335) only a part of the regional dummies used to control for the fixed effects are significantly different. When the regression is repeated using only these significant regional dummies the coefficient of the regional 3 See for example Pekkarinen, Pohjola and Rowthorn (1992). 6

8 unemployment rate is again significant and negative: Blanchflower and Oswald do not have their own results from the Swedish data to present. Instead they refer to the work done in Sweden. According to results of for example Holmlund and Skedinger (1990) there are some grounds to conclude that the Swedish data supports the wage curve hypothesis. Contradicting results have however also appeared. 4 Blanchflower and Oswald nevertheless conclude that there is evidence on the existence of the wage curve in Scandinavian data, even though these countries with their centralised bargaining systems and nationwide wage parities are not necessarily likely to exhibit signs of a regional wage curve (Blanchflower and Oswald 1994, 355). 4 See for example Blomskog (1993). 7

9 3. FINNISH WAGE CURVE The results presented by Blanchflower and Oswald give relatively strong reasons to consider the wage curve as a stylised fact of the labour markets. Scandinavian results however play only a small role in their work and they are based on a small sample sizes. Therefore new Scandinavian results are interesting from the point of view of the generality of the wage curve. In the following we present a set of results from estimations where the motivation has been to replicate the original studies of Blanchflower and Oswald using Finnish data Data The data used here are a sample from manufacturing workers wage statistics by the Confederation of Finnish Industry and Employers Teollisuus ja työnantajat (TT) from the period of Wage statistics contains data on all the manual workers who are above 15 years old and are employed in a firm affiliated with TT and have worked during the calendar year. The sample examined here has been used in a number of earlier Finnish studies on the Finnish wage structure, wage drift, and the wage discrimination. It has been restricted to include only the five most important industries in Finland: metal, paper, timber, textile and clothing industries. According to Vartiainen (1994) the sample is supposed to represent the structure of the manufacturing sector s labour force in each year of crosssection as well as its dynamics. The sample has been picked by choosing the year 1990 as a base year and ordering the 1990 wage statistics by firms. Every fourteenth worker is then picked for the sample. A longitudinal data base has been constructed from 1990 onwards and backwards all the 8

10 way to the year The period was chosen here because the data on individuals county of residence are available only from 1992 onwards. The sample sizes for the years 1992, 1993, and 1994 are 7 399, 6 807, and respectively. Furthermore a balanced panel was constructed for the pooled analysis: workers were present in the each year of the panel, which therefore contains observations. Data do not contain a very detailed information on the individual characteristics of the worker. Only age and gender can be identified. However the data on the working hours and the different compensation forms are very detailed. Working hours are divided into normal working hours, piecework hours, added pay hours, extra time hours, and working hours on Sunday. Worker s earnings form each of these hours are also separated. Data also include the wage group of the individual. The wage group is determined in the wage negotiations and it serves as a proxy of the job s requirement level Variables of the wage equation In their original studies Blanchflower and Oswald (1994) use monthly or annual wages as dependent variables. This method may be a source of bias in the wage curve result since the annual earnings are a product of annual working hours and hourly wages. It is to be expected that the annual hours may react to changes in the local unemployment rate. This has led commentators like Card (1995) to express their concern that Blanchflower and Oswald have not distinguished between an hours curve and the wage curve. According to the results of Card (1995) and Bratsberg and Turunen (1996), the coefficient of the local unemployment rate is closer to zero in an equation where the dependent variable is an hourly wage than when the annul earnings are used as a dependent variable. We chose to use hourly wages as a dependent variable. First annual earnings from normal working hours, piecework hours and added pay hours were calculated. Hourly wage was given by dividing this sum by the number of hours worked. Since 9

11 Blanchflower and Oswald end up preferring the log-linear equation form, the dependent variable was included as a logarithm. Naturally the most interesting independent variable in the equation is the logarithm of the regional unemployment rate. The original data included the worker s county of residence. Using a county (kunta) as a definition of the local labour market isn t however reasonable, since there are over 300 counties in Finland. Here we chose to use the regional counties (seutukunta) of which there are a total of 88 in Finland. Regional county is usually seen as a typical go-to-work area in Finland. The sample contained observations from 75 regional counties. Unemployment rates of the regional counties were provided by the Statistics Finland. The chosen period was rather exceptional in the Finnish unemployment history, since the Finnish unemployment experienced it s most dramatic rise in the whole 20 th century during this period: the sample average reached a peak of 23 % in As already mentioned above, the data did not include a rich set of variables describing individual s characteristics. Dummy variables controlling for the individual s gender and age were included in the equation. Individual s wage group was included to control for the job s requirement level. A set of variables controlling the way in which the individual has been compensated for the hours worked was also used. These variables included the proportion of the piece rate hours worked by the individual to the total number of hours worked and dummy variables controlling for work on extra time, Sundays or in shifts. Although the earnings from these hours were not included in the dependent variable, we considered that these variables control for the possibility that the individuals who work on extra time or on Sundays tend to earn larger wages also from normal working hours. Different working schedules there are 10 of them in the official contracts were controlled for as for the industry in which the individual works. In some of the pooled regressions a full set of regional county -dummies was used to control for the region specific fixed effects. Naturally in the crosssectional regressions this wasn t possible. However the data included a variable that splits Finnish counties in two according to their price level. 10

12 This variable could be used as a rough approximation of the regional characteristics in the cross-sectional regressions. In the pooled regression the yearly dummies were included to control for the changes in the price level On the results We present here results obtained from the estimation of the wage equation of the type described above. 5 The estimation procedures are ordinary least squares and regional fixed effects model. There are results from the crosssections 1992, 1993 and 1994 as well as from the pooled sample of As the hourly wage clearly is the theoretically correct dependent variable, the following results are mainly from equations that use hourly wages. 6 We shall first focus on the results from the cross-sections. Table 1 presents the estimated coefficients of the regional unemployment rates from the cross-section equations that use hourly wage as a dependent variable. 5 To save space, only the results relevant to the wage curve debate are presented here. The complete regression results are available on request. For the comparison with the results of Blanchflower and Oswald (1994) the regressions that use annual wages as a dependent variable were also conducted. To save space the results are not included in the current version of the text. They are however generally in line with the results from hourly wage equations and are also available on request. 6 Blanchflower and Oswald (1994) present three different tehoretical models that generate a downward sloping wage curve-type realtionship between wages and regional unemployment. The relationship is explicitly defined to exist between wages and unemployment, not between hours of work and unemployment. 11

13 Table 1. Wage curve regressions with cross-sectional data log hourly wage Year log U r (t-statistic.) (8.69) (4.09) (4.16) R N Log U r is the estimated coefficient of the regional unemployment rate in the wage equation. The regressions include following control variables: constant, age, gender, proportion of the hours worked on piece rates, controls for work on Sunday, extra time and in shifts, controls for the work schedule, industry, wage group and the price level of the region. All the regressions use hetereoscedasticityconsistent covariance matrix. As can be seen from the table 1, the estimated coefficients of the regional unemployment rate are negative and significant throughout the crosssections. Especially the coefficient of the year 1992, -0.08, is particularly close to the stylised fact of Blanchflower and Oswald (1994): There are signs of a relationship similar to downward sloping wage curve in these results. The results presented above give evidence that there might exist a relationship similar to the downward sloping wage curve. However most of the results presented by Blanchflower and Oswald (1994) are from the pooled data. The pooled data also make it possible to control for the regionspecific fixed effects. For this we constructed a balanced panel data set from the sample, covering the whole sample period This could be done since the data give information on whether the individual is an entrant or a leaver in the sample workers were present throughout the whole sample, the number of observations was thus In six observations the individual had not worked normal hours at all so his hourly wage could not be calculated. This makes the number of observations that could be used for the estimation Table 2 presents the results from a 12

14 regression that uses the logarithm of the hourly wage as a dependent variable. Table 2. Wage curve regressions with the pooled data log hourly wage Year (1) log U r (t-statistic.) (10.13) (2) (0.8) (3) (7.203) Regional dummies None All (75) 64 R N Log U r is the estimated coefficient of the regional unemployment rate in the wage equation. The regressions include following control variables: constant, age, gender, proportion of the hours worked on piece rates, controls for work on Sunday, extra time and in shifts, controls for the work schedule, industry, wage group and region-specific fixed effects where indicated. All the regressions use hetereoscedasticity-consistent covariance matrix. When the region-specific fixed effects are not controlled for, the results give strong evidence on the downward sloping wage curve. The estimated coefficient of the regional unemployment rate in the first column is clearly negative and significant and relatively close to the standard of Blanchflower and Oswald (1995). The inclusion of a full set of regional dummies to control for the region-specific fixed effects drives the coefficient to insignificance, although it remains negative (second column). However only 64 of the regional dummies were found to be statistically significant. When the estimation was repeated with only 64 regional dummies, the estimated coefficient of the regional unemployment rate was once again negative and significantly different from zero. The result in the third column, -0.09, is also very near the result of Blanchflower and Oswald (1994). 13

15 3.4. Interpretation of the results The results presented above can be considered to be in accordance with the typical wage curve results as defined in Blanchflower and Oswald (1994). The estimated coefficients of the regional unemployment rate are all negative. With the pooled data the estimates tend to be driven to insignificance by the inclusion of a full set of regional dummies to control for the region-specific fixed effects. However when only initially significant dummies are used the estimate turns significant again and its value, -0.09, is surprisingly close to the typical results of Blanchflower and Oswald (1994). The results from cross-sectional data also support the existence of a relationship similar to the downward sloping wage curve. Certain reservations are in place however. Firstly, it is not at all clear that the exclusion of initially insignificantly different regional dummies is wholly justified. The role that regional fixed effects play in these results is an unsolved question. Secondly, the data on the individual characteristics of the workers are poor and it might be reasonable to control for the unobserved individual effects. The balanced panel constructed for this study makes it possible. Thirdly, it should be remembered that the period studied here is an exceptional one in the Finnish unemployment history. The results reported above may therefore fail to give a right account of how the wages react to changes in regional unemployment under normal conditions. The result in the third column of the table 2 with the pooled data and 94 regional dummies and hourly wage as a dependent variable may be considered a most representative one. Figure 1 then gives an idea of the shape of the Finnish wage curve. 14

16 Figure 1. Finnish wage curve : ln(w) = S 0.09 ln U r In figure 1 x-axis reports the regional county unemployment rates, which varied approximately between 12 and 30 percent in the sample, and y-axis the calculated hourly wages. The picture depicts the relationship of wages and unemployment for only those unemployment rates that are actually observed in the sample. If the x-axis had started from full employment the wage curve would have turned sharply steeper below the 10 percent unemployment rate. Figure 1 is thus the flat part of the wage curve. Some commentators, for example Paldam (1990), have suspected that the wage curve is being forced onto data by the chosen functional form. To correct for this, Blanchflower and Oswald (1994) estimate an unrestricted specification of the wage curve. Here the method is to use the usual control variables but to replace the regional unemployment rate with a series of dummy variables denoting intervals of unemployment rate. To get a fuller picture of the shape Finnish wage curve we ran a similar regression with 15

17 the data in hand. The unemployment intervals were constructed so that each cell contains approximately 5 % of the observations, so there are nineteen dummy variables for unemployment intervals in the regression the lowest interval 12.8 % 14.8 % being the omitted interval. Table 3 reports the results from this regression. The dependent variable is the logarithm of the hourly wage. Table 3. Wage curve regression with pooled data unrestricted functional form, log hourly wage Regional unemployment Coefficient (t-statistic) -14.8S (9.27) 16.7S (2.7) 17S (1.67) 17.9S (7.65) 18.5S (3.14) 19.1S (8.45) 19.6S (3.73) 20S (4.29) 20.6S (6.42) 21.5S (3.82) 21.8S (12.21) 22S (1.11) 22.7S (10.71) 23.2S (7.34) 23.7S (6.24) 24.3S (6.63) 24.8S (4.94) 25.1S (10.27) (10.83) The control variables are otherwise the same as in the regression of the first column of table 2. 16

18 If the results were to resemble the typical wage curve-results, we would obtain more negative estimates for the coefficients of the larger unemployment rates. This is not clear from looking at the table 3. As can be seen, the dummy variable for the interval obtains a positive but not highly significant coefficient. To make the interpretation of these results easier, we plotted the anti-logs of the estimates of the coefficients togethter with the mid-points of the unemployment rate intervals. 7 Figure 2. Non parametric wage curve 7 The estimated coefficient of the interval , which was positive but not highly significant, was left out of the picture. It would have been a clear out-lier: the anti-log of it s estimated coefficient is approximately

19 There are some signs of downward sloping wage curve in figure 2, especially if one takes into account that the estimated coefficient of the interval is clearly an outlier. Although the figure does not give such a clear support to the downward sloping wage curve as do for example the ones presented by Blanchfower and Oswald (1994, figures ) with US data, there are still no grounds to conclude that the results from the unrestricted estimation would somehow contradict the results presented above. 18

20 4. CONCLUSIONS It may be concluded that these results indicate that Finnish data do not present an exception to the argument of Blanchflower and Oswald (1994), according to which there exists a downward sloping curve, linking the regional unemployment rate and wages, that seems to be of approximately same shape regardless of the country or period under study. Regionspecific fixed effects are a problem that makes the interpretation of these results complicated, but it seems that this can be dealt with the same way as Blanchflower and Oswald (1994) have done in their estimations with the Norwegian data. When only significantly different regional dummies are used the estimated coefficient of the regional unemployment rate is negative, significant, and very close to the typical results obtained by Blanchflower and Oswald (1994):

21 REFERENCES Blanchflower, D. G. & A. J. Oswald (1990), The wage curve, Scandinavian Journal of Economics, 92, Blanchflower, D. G. & A. J. Oswald (1994), The Wage Curve, The MIT Press, Cambridge, Massachusetts. Blomskog, S. (1993), Is there a stable swedish wage curve?, Mimeo, Swedish Institute for Social Research. Bratsberg, B. & J. Turunen (1996), Wage curve evidence from panel data, Economics Letters, 51, Card, D. (1995), The wage curve: A review, Journal of Economic Literature, 33, Groot, W., E. Mekkelolt & H. Oosterbeek (1992), Further evidence on the wage curve, Economics Letters, 38, Harris, J. & M. Todaro (1970), Migration, unemployment and development: A two-sector analysis, American Economic Review, 69, Hoddinott, J. (1996), Wages and unemployment in an urban African labour market, Economic Journal, 106, Holmlund, B. & P. Skedinger (1990), Wage bargain and wage drift: evidence from the swedish wood industry, in Calmfors, L. (ed.), Wage Formation and Macroeconomic Policy in the Nordic Countries, SNS Förlag & Oxford University Press. Stockholm. Paldam, M. (1990), Comment on the wage curve, Scandinavian Journal of Economics, 92,

22 Parjanne, M-L. (1997), Työmarkkinat murroksessa, Elinkeinoelämän tutkimuslaitos, B 135. Pekkarinen, J., M. Pohjola & B. Rowthorn (ed.)(1992), Social Corporatism A Superior Economic System?, Clarendon Press, Oxford. Wagner, J. (1994), German wage curves, Economics Letters, 44, Vartiainen, J. (1994), Palkkaliukumat suomalaisessa teollisuudessa: yksilötason analyysi, Palkansaajien tutkimuslaitos, Tutkimuksia