4.1. The Standard Turkish Wage Curve Revisited
The existing standard wage curve estimates for Turkey excludes some groups of individuals, such as workers in the agricultural sector (
Baltagi et al. 2012,
2013,
2015), and workers in the public sector (
Ilkkaracan et al. 2013). We present the standard wage curve estimates for the whole labor force without excluding any group of workers in
Table 2 using FE-2SLS in which we take into account the endogeneity of the unemployment rates.
We present only the estimates of the unemployment elasticity of pay, β, for the sake of saving space.
Table 2 shows the FE-2SLS estimations results of β for all workers as well as some subgroups. The unemployment rates used in the estimation are the regional overall unemployment rates except for column 2 in which we use non-agricultural unemployment rates. As has been mentioned, the standard errors need to be clustered for regions, because individual wages in the same region would be correlated. However, we assume that the regional unemployment rates are not independent across regions in the spatial wage curve estimates and thus clustering is not feasible. Hence, we present robust (heteroskedasticity-consistent) standard errors to be able to compare our results from the standard wage curve and the spatial wage curve. However, we checked our results clustering for regions. The results are not presented here, but available upon request. We found that the standard errors become higher and some estimates lose their significance. However, the FE-2SLS results for all workers, as well as for males and females remain same.
Column 1 of
Table 2 suggests that the unemployment elasticity of pay is −0.071. In the second column, we exclude workers in the agriculture sector and use non-agricultural unemployment rate in the wage curve estimation. The FE-2SLS results for the non-agriculture sectors with non-agricultural regional unemployment rate show that the unemployment elasticity of wages is −0.107. This is very similar to the estimates that found in
Baltagi et al. (
2012,
2013). The overall unemployment rate for the whole population is used in the estimation of the wage curve of disaggregated samples. The FE-2SLS results show that men are more responsive to the regional unemployment rate than women are; −0.073, −0.063, respectively. This result is in line with the international evidence, in general. However, previous studies on the Turkish wage curve either find a higher unemployment elasticity for women (e.g.,
Baltagi et al. 2012,
2013) or no statistically significant unemployment elasticity (e.g.,
Ilkkaracan and Selim 2003;
Konyali 2012). The first two studies use regional non-agricultural unemployment rates while the last two use regional unemployment rates. Note that these results can be attributed to the labor force characteristics of women in Turkey. The selection into wage employment for women is severe in Turkey. The female labor force participation is around 30 percent. The probability of labor force participation for women increases by education level. That is, the rate of labor force participation is the highest for women with college education. It is also found that the wages of highly educated people are less responsive to the regional unemployment rates. Hence, it is reasonable to find less elastic wage curve for the female workers.
Furthermore, we test our results for different time periods, i.e., periods 2005–2008 and 2005–2009, to be able to investigate the discrepancy between our results and previous results. Note that the wage curve indicates the cyclicality of the wages. Including time fixed effects in Equation (1), we control for the business cycle in the estimation of the wage curve. However, using data representing one segment of the business cycle will introduce a bias in the wage curve estimation. The years 2005–2008 can be characterized by a boom period of the business cycle (
Ilkkaracan et al. 2013). Hence, using the data for the years 2004–2013 can take into account for the complete business cycle and eliminates this bias.
We present the unemployment elasticity of pay for the comparison periods with the overall unemployment rates as well as with non-agricultural unemployment rates in the
Appendix Table A1. Panel A of
Table A1 suggests that there is no evidence for a wage curve in periods 2005–2008 and 2005–2009, except one case, for the male workers in the 2005–2009 sample, when the overall unemployment rate is used. This is similar to
Ilkkaracan et al. (
2013) and
Konyali (
2012). Furthermore, Panel B of
Table A1 shows the same comparison but using the non-agricultural unemployment rate. We were able to replicate the fixed effect results of
Baltagi et al. (
2012,
2013), such that the wage curve for women is more elastic than for men in the periods, 2005–2008 and 2005–2009. However, for the period 2004–2013, women have a lower unemployment elasticity of pay than men, −0.044, and −0.072, respectively. The FE-2SLS results are higher than fixed effects results using a non-agricultural unemployment rate. Comparing the results using the period 2005–2008 and the period 2005–2009 implies that inclusion of the crisis year 2009, at which the unemployment rate was the highest, the wage curve gets more elastic, −0.11 and −0.125, respectively. Consequently, using data for the period 2004–2013, the unemployment elasticity of pay gets closer to −0.10. This indicates that using data which covers a whole business cycle can eliminate the bias. The same patterns emerged for the disaggregated sub-samples. It is also found that female wages are more sensitive to unemployment rates in different segments of the business cycle.
In
Table 2, we also present the wage curve estimates for subsamples based on age, skill, and sector disaggregation.
4 In columns 5 and 6, the FE-2SLS results indicate that the young workers who are aged 15–24 have a higher elastic wage curve than workers aged 25 and older do, −0.096, and −0.065, respectively. This is an expected result, because workers who are aged 15–24 are more likely work in part-time jobs or the lower end of the job distribution. Thus, the wage of this group of workers is more fragile and sensitive to the fluctuations in the regional unemployment rates relative to those workers who are 25 years old or older. We further disaggregate our sample as workers in the public and private sector.
5 The FE-2SLS results suggest that the regional unemployment rates positively affect the individual wages with an elasticity of 0.02. This confirms the expectations that workers in the public sector would not be affected by the fluctuations of regional unemployment rate due to the nationwide bargaining process. Column 8 shows that the wage curve phenomenon is more about the private sector. The unemployment elasticity of pay is −0.098 using the FE-2SLS. Furthermore, we find that the responsiveness of the real wages decreases by skill/education levels. The FE-2SLS results indicate that the less educated/low-skilled workers have a more elastic wage curve than more educated/high-skilled workers do, −0.099 and −0.02, respectively. This corroborates the conclusion of
Card (
1995) such that the wages of highly educated workers are less responsive to the regional unemployment rates.
Appendix Table A2 shows unemployment elasticities for more disaggregated samples. It is found that temporary and less-experienced workers have more elastic wage curves than permanent and more experienced counterparts, respectively. Furthermore, the unemployment elasticity of pay is higher for those who work in the small firms than those who work in the medium-sized or large firms. The general patterns in
Table 2 and
Appendix Table A2 suggest that the more fragile the worker group, the more elastic the wage curve they have.
The wage curves for the disaggregated samples in
Table 2 are estimated using the regional overall unemployment rate. The results show different elasticities for different groups. However, some groups of workers might be more affected by the group-specific unemployment rate rather than the overall unemployment rate (
Card 1995). In other words, if the wage curve results in efficiency wage or monopsonistic competition, workers might compete for each other within a group rather than all workers, then the group-specific unemployment rate would be more appropriate. It is likely that workers compete for each other within skill groups. We test this using group-specific regional unemployment rates that are available for gender, education level, and age groups in TURKSTAT’s website. We make use of these disaggregated unemployment rates to estimate wage curves for various group of workers.
Table 3 presents wage curve estimates using group-specific regional unemployment rates of different groups; gender, skill, and age groups. These estimates should be compared to those in
Table 2. The second column of
Table 3 shows the unemployment elasticity of pay for women using women-specific unemployment rates. The FE-2SLS results show that there is no significant wage curve for women, unlike the highly significant elasticity of −0.063 in
Table 2. The third column, on the other hand, shows that the unemployment elasticity of pay for men is very similar in
Table 2 and
Table 3. These findings indicate that male wages are responsive to the overall and male-specific unemployment rates by around the same magnitude. However, women’s wages are more responsive the overall unemployment rate rather than women-specific unemployment rates. In other words, the significant wage curve for females in
Table 2 might result in the different elasticity of female unemployment rate to the overall unemployment rather than the unemployment elasticity of pay.
Columns 4 and 5 show wage curve estimates for workers disaggregated by age using age group-specific regional unemployment rates. The FE-2SLS results indicate that unemployment elasticity of real wages is −0.131 for workers who are aged 15–24, and −0.054 for workers who are aged 25 and above. Comparison of these results with those found in
Table 2 indicates that wages young workers are even more sensitive to the youth unemployment rates whereas wages of workers who are 25 years old or older are less sensitive to their group specific unemployment rate. These findings suggest that the use of overall unemployment rate relatively masks the real relationship between wages of younger workers and unemployment rate.
Next, we present, the wage curve estimates for the workers disaggregated by three skill levels using group-specific regional unemployment rates in the last three columns of
Table 3.
Ilkkaracan et al. (
2013) used regional unemployment rates disaggregated by three education/skill level to estimate the wage curve of females and males in the private sector. However, they did not provide wage curve estimates for the skill subsamples. The FE-2SLS estimations result in unemployment elasticities for low-skilled, medium skilled, and high-skilled workers with the magnitudes, −0.054, −0.094, and −0.076, respectively. Comparing wage curve estimates for skill subsamples in
Table 2 and
Table 3 suggests different conclusions. The wages of the low-skilled workers have the highest responsiveness to the regional overall unemployment rate in
Table 2, whereas the high-skilled workers are the least responsive group. However, the FE-2SLS results in
Table 3 indicate that the slope of the wage curve is the steepest for the medium-skilled workers. Also, the wages of high-skilled workers are more responsive to the group-specific unemployment rates than that of low-skilled workers. These findings confirm the expectations that the workers are more likely compete within the skill groups. Also, they indicate that the use of the regional overall unemployment rate instead of skill-specific unemployment rates masks the true relationship between unemployment rate and real wages to some extent.
Overall, we find that the wage curve estimates are sensitive to the exclusion of some groups that was the case in the existing wage curve estimates for Turkey. Moreover, the wage curve for Turkey is about −0.07. This corroborates the finding of
Nijkamp and Poot (
2005) rather than the elasticity of −0.10 that is found in
Blanchflower and Oswald (
1990,
1994) and other studies. Our findings also corroborate the findings that wages of the more fragile worker groups are more responsive to the fluctuation in regional unemployment rates (e.g.,
Card 1995;
Baltagi et al. 2012,
2013). Furthermore, the use of group-specific unemployment rates in estimation of the wage curves for disaggregated samples suggests that the use of the regional overall unemployment rates might mask the true relationship between wages and unemployment rates of certain groups.
4.2. The Spatial Turkish Wage Curve
The wage curve estimations in which spatially weighted unemployment rate is included as an independent variable are presented in
Table 4 and
Table 5. We report only unemployment elasticity of pay, β, and spatial spillover unemployment elasticity of real wages,
. The standard errors are robust (heteroskedasticity-consistent) standard errors.
Table 4 presents the FE-2SLS estimation results of Equation (2) using the row-normalized contiguity matrix as the spatial weight matrix. Moreover, we take into account the endogeneity of regional unemployment rate and spatially weighted unemployment rate. We use the FE-2SLS method using one-year lagged values of unemployment variables as instruments. The FE-2SLS results show that own-region unemployment elasticity of pay for is −0.056, and spatial spillovers of unemployment elasticity to contiguous regions is −0.087. These results indicate that unemployment rate of contiguous regions has a significant effect on individual wages. This effect is higher than the effect of the own-region unemployment rate on real hourly wages. Comparing FE-2SLS results for the unemployment elasticity of real wages in
Table 2 and
Table 4, we find that ignoring the spatially weighted unemployment rate introduces a downward bias, such that inclusion of spatially weighted unemployment rate lowers the unemployment elasticity in absolute terms. These findings are in line with
Baltagi and Rokicki (
2014) for Poland, and
Fingleton and Palombi (
2013) for the UK. The higher effects of contiguous regions’ unemployment rates might indicate the monopsonistic competition explanation of the wage curve. That is, employers determine the wage offers considering unemployment conditions in the local labor market as well as the adjacent regions’ unemployment rates. In the second column, we estimate the spatial wage curve using the non-agricultural unemployment rate and excluding the workers in the agriculture sector in order to compare our results with those found in
Baltagi et al. (
2015). The FE-2SLS results show that the wage elasticity with respect to own-region unemployment is higher than spatial spillover unemployment elasticity with magnitudes of −0.084 and −0.162, respectively. The FE-2SLS results of
Baltagi et al. (
2015) show that the wage elasticity with respect to the local unemployment rate is higher than the spatial spillover unemployment elasticity of the proximate regions. This is contradictory to our results. However, our findings are in line with their estimations, using full inverse distance matrix as a spatial weight matrix. Note that our results are based on the contiguity matrix, and our data covers one additional year, whereas
Baltagi et al. (
2015) define spatial weight matrix as a function of the inverse distance.
Moreover, columns 3 and 4 present wage elasticities with respect to own-region and contiguous region unemployment rates for females and males, respectively. The FE-2SLS results suggest that wage elasticity with respect to own-region unemployment is −0.051, and elasticity with respect to contiguous region unemployment rate is −0.118 for females. In addition, the unemployment elasticity of pay for males is −0.058 and the spatial spillover unemployment elasticity of adjacent regions is −0.087. These results suggest that the wage elasticity of own-region unemployment for males is slightly higher than females. However, the effect of neighboring regions’ unemployment on real wages of females is larger than that of males. These results are similar to
Baltagi and Rokicki (
2014) for the Polish case. The higher spatial spillover unemployment elasticity of pay for females might indicate that the wages of the women workers are affected by unemployment rates in a wider labor market. This is also related to the low labor force participation of women and their probability of participation in wage employment is the highest among college graduates. Moreover, most of the college graduates work in public sector jobs. That is, there is a higher percentage of women with college degrees working in the public sector. Moreover, the wages in the public sector are subject to a collective bargaining and are affected by the nationwide unemployment rate, as has shown in
Table 2. Therefore, it is reasonable that the wages of females are more responsive to the contiguous regions unemployment rate.
Furthermore, we present the results for several disaggregated samples in
Table 4. There are statistically significant spatial spillover effects for all groups. The elasticities for further disaggregated worker groups are presented in
Appendix Table A3. We find statistically significant unemployment elasticities for all groups. Also, there is spatial spillover for unemployment elasticity of proximate regions for all groups except for workers who work in large firms.
It is found that workers aged 15–24 are more sensitive to adjacent regions’ unemployment rates than own-region unemployment rates with elasticities of −0.143 and −0.067, respectively. Both of these elasticities are higher than those found for workers aged 25 and older. It suggests that youth wages are depressed by the local and neighboring regions’ unemployment rate to a higher degree.
These findings reveal some interesting results. First, the workers in the public sector are negatively responsive to the unemployment rate in the neighboring regions and positively responsive to the local unemployment rate.
Card (
1995) points out that public sector wages are expected to be insensitive to the regional unemployment rate. However, the FE-2SLS results show that there is a positive effect of the local unemployment rate. This might result in the fact that the public sector jobs in less-developed regions of Turkey are paid compensating wage premiums (
Ilkkaracan et al. 2013). On the other hand, we find that the public sector wages are negatively affected by the proximate regions’ unemployment rates. Since public sector wages are determined by collective bargaining, it is expected that the wages in public sector are affected negatively by the aggregate unemployment rates. Second, we find that the private sector wages are more sensitive to the own-region unemployment rate than those of surrounding regions. These results seem surprising. Note that the spatial spillovers are found to be higher than own-region unemployment elasticity of pay in the FE-2SLS estimations for all groups except private sector workers. This confirms our interpretations for the public workers. Recall that all subsamples include public sector workers, so the effect of the aggregate unemployment rate on public wages might magnify the spillover effect of the unemployment rate.
Furthermore, the FE-2SLS results for the skill groups indicate that hourly wages are sensitive to the proximate regions’ unemployment rate more than own region’s unemployment rate. Results show that the low-skilled workers are the most sensitive skill group to the local unemployment rate and surrounding regions’ unemployment rates with elasticities of −0.075 and −0.132, respectively. Both elasticities get lower by skill level. Unemployment elasticity of real wages is not significant for high-skilled workers while the spatial spillover effect is significant, −0.067. In other words, high-skilled workers are more sensitive to the employment conditions of proximate regions than that of their own region. Again, this might be related to the interpretations about public workers, since a high percentage of high-skilled workers are in the public sector.
Appendix Table A3 present the FE-2SLS results for further disaggregated samples. Results indicate that the spatial spillover unemployment elasticities are higher than own-region unemployment elasticities except for those who works in large firms. Workers with low tenure have higher unemployment elasticity of earnings and spatial spillover effects. Workers in small firms have higher unemployment elasticity of pay and spatial spillover elasticity, in absolute value, than workers in medium or large size firms with fixed effects estimation. The FE-2SLS results suggest that there is no significant spatial spillover unemployment elasticity for workers in the large firms. These results show that workers in large firms are only affected by own region employment conditions, whereas workers in medium-size firms are the most affected group by unemployment rate in neighboring regions. We also find that temporary workers have the highest spatial spillover elasticity of −0.221, among all groups. However, there is no significant own-region unemployment elasticity for this group of workers.
Furthermore,
Table 5 presents the FE-2SLS estimates of the unemployment elasticity of pay
and spatial spillover unemployment elasticity
using group-specific unemployment rates. The FE-2SLS results indicate that there are no significant wage elasticities with respect to own-region and adjacent regions’ unemployment rates for the women workers. Note that the effect of own-region women-specific unemployment rate on the wages of females is similar to that found in
Table 3. This can be attributable to the selection into wage employment problem that females face and the lack of spillover effects for the high-skilled workers. These suggest that the significant effects in
Table 4 might be driven by the elasticity of female-specific unemployment rate with respect to the overall regional unemployment rate. However, there is significant male wage elasticity with respect to own-region and the neighboring regions’ male-specific unemployment rates with magnitudes of −0.056 and −0.061, respectively.
Columns 4 and 5 show that the spatial spillover effects are higher than own-region unemployment effects for the age groups using group-specific unemployment rates. The FE-2SLS results show that wages of young workers are more responsive to the own-region unemployment rate and contiguous regions’ unemployment rate than their counterparts who are aged 25 and above. The FE-2SLS results are of similar to those found in
Table 4.
The last three columns of
Table 5 present the elasticities for the skill groups using group-specific unemployment rates. The FE-2SLS results show that, among skill groups, the highest wage elasticities with respect to own-region and proximate regions’ unemployment rates are found for the medium-skilled workers with magnitudes of −0.07 and −0.12, respectively. Moreover, the low-skilled workers are similarly less sensitive to the own-region unemployment rate than that of contiguous regions with elasticities of −0.039 and −0.093, respectively. However, there is no significant spatial spillovers for unemployment elasticity for high-skilled workers. These results indicate that the slope of wage curve and spatial spillovers for the skill groups are sensitive to the use of skill-specific unemployment rates. Assuming that the wage curve results in efficient wages, the wage curve estimates using an overall regional unemployment rate may mask the true relationship.
Overall, we provide evidence that the spatially lagged regional unemployment rate has a significant effect on real wages. We find that the spatial spillover effects are higher than own region’s unemployment rate effect for disaggregated samples. However, this should be taken with caution since the subgroups include public workers and the public sector wages are mainly affected by aggregate unemployment rate rather than regional unemployment rates. Moreover, we show that the slope of wage curve and spatial spillovers are sensitive to the use of group-specific unemployment rate, especially for females. Thus, using the overall regional unemployment rates in the estimation of wage curve for disaggregated samples may mask the true variation in the slope of the wage curve and spatial spillovers.