Estimating the Impact of Labor Taxes on Employment and the Balances of the Social Insurance Funds in Turkey

Transcription

1 Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Report No TR Estimating the Impact of Labor Taxes on Employment and the Balances of the Social Insurance Funds in Turkey Synthesis Report April 2009 Human Development Sector Unit Europe and Central Asia Region Document of the World Bank

2 The International Bank for Reconstruction and Development The World Bank 1818 H Street, NW Washington, DC 20433, USA All rights reserved The World Bank enjoys copyright under protocol 2 of the Universal Copyright Convention. This material may nonetheless be copied for research, educational or scholarly purposes only in the member countries of The World Bank. Material in this report is subject to revision. ii

3 Currency Equivalents Exchange rate effective as of March 31, 2009 Currency Unit: New Turkish Lira (YTL) 1.70 YTL = US$1 YTL 2.24 = 1 Euro Government Fiscal Year January 1 - December 31 ABBREVIATIONS AND ACRONYMS EU GDP HLFS IMF ISKUR OECD PROST SSK TISK TURK-IS TURKSTAT UI UISIM European Union Gross Domestic Product Household Labor Force Survey International Monetary Fund Turkish Employment Organization Organisation for Economic Co-operation and Development Pension Reform Options Simulation Toolkit Social Security Institution Union of Turkish Employer Associations Conferedations of Turkish Trade Unions Turkish Statistical Institute Unemployment Insurance Unemployment Insurance Simulation Model Vice President: Country Director: Sector Director: Task Team Leader: Shiego Katsu Ulrich Zachau Tamar Manuelyan Atinc Gordon Betcherman iii

4 Estimating the Impact of Labor Taxes on Employment and the Balances of the Social Insurance Funds in Turkey PREFACE V Table of Contents EXECUTIVE SUMMARY...VI I. INTRODUCTION...1 II. III. A BRIEF SUMMARY OF THE LITERATURE ON THE EMPLOYMENT EFFECT OF LABOR TAXES AND SUBSIDIES... 4 STUDY METHODOLOGY...7 (a) Establishment-level analysis of labor demand...8 (b) Analysis of regional subsidies...10 (c) Analysis of changes in labor costs using Household Labor Force Survey data...13 IV. RESULTS ON EMPLOYMENT IMPACTS Estimates of labor demand elasticities and pass-through rates Evidence on employment impacts from the regional subsidy programs Evidence on employment impacts from household data The modeling exercise...25 VI. POLICY IMPLICATIONS List of Tables TABLE A1. REAL LABOR COSTS FOR MINIMUM WAGE WORKERS, (MILLION TL)...34 TABLE A2. EXPENDITURES UNDER LAW 5084 IN THE 15 NEWLY SUBSIDIZED PROVINCES AND COST PER JOB CREATION UNDER DIFFERENT ASSUMPTIONS, JANUARY 2004 TO APRIL TABLE A3. EXPENDITURES UNDER LAW 5350 IN THE 13 NEWLY SUBSIDIZED PROVINCES AND COST PER JOB CREATION UNDER DIFFERENT ASSUMPTIONS, MAY 2005 TO DECEMBER iv

5 PREFACE High labor tax wedges and slow formal employment growth have combined to make labor tax reform an important economic policy issue in Turkey. However, solid evidence does not exist to confidently assess the likely employment gains and fiscal implications of reduced taxes on wages or social insurance contributions. This synthesis report presents the results of a series of empirical studies of the impact of a labor tax reform. It should be noted that the analysis was undertaken before the social contribution reforms that were introduced as part of the 2008 employment package. Using data from firms, households, and social insurance files, the research finds that employment does respond to changes in labor costs at levels that are comparable to those found in other middle-income and OECD countries. The results show that reducing labor costs could significantly boost registered employment. However, the actual effect of lower taxes on employment would be diluted because a significant portion of the reduced tax would be captured by workers through higher wages rather than by employers through lower labor costs. While this pass-through effect would substantially limit the job gains of a general reduction in labor taxes, it is much smaller at lower wage levels where most of the effect of a tax cut would be to lower total labor costs and, thus, increase labor demand. As a result, tax cuts targeted towards low-wage labor would be more cost-effective than across-the-board reductions. Indeed, this is confirmed by simulations of the implications of various tax reform options on the fiscal position of Turkey s social security and UI funds. Options designed to target contribution reductions on young workers (predominantly employed at low wage levels) turn out to be considerably more cost-effective than across-the-board reductions. However, under all scenarios, the registered employment gains of a cut in social insurance contribution rates do not broaden the tax base enough to compensate for the reduced contribution rates. To achieve overall fiscal neutrality, compensating additional revenues from other sources or reduced expenditures would need to accompany lower contribution rates. v

6 Estimating the Impact of Labor Taxes on Employment and the Balances of the Social Insurance Funds in Turkey 1 EXECUTIVE SUMMARY Job creation has been a major challenge for Turkey. The current economic crisis has caused rapid increases in unemployment. However, even before the slowdown started, employment creation had been slow despite strong economic growth in the years after the last financial crisis in The Government therefore identified increasing employment as one of the development axes in its Ninth Development Plan, covering The Plan sets a target of 2.7 percent employment growth per year during this period, a rate well in excess of the 1.1 percent annual increase registered between 2002 and Turkey s employment performance has lagged well behind European comparators, especially in the case of women. In 2006, only 46 percent of the working-age population was employed in Turkey. This rate had been relatively constant since the 2001 crisis and remained well below the employment performance in EU-member countries. In fact, employment rates for prime-age males are only slightly below EU averages. However, Turkey s employment is substantially lower for younger and older workers and especially for women: less than one in four Turkish women are employed, compared to 57 percent in the EU countries. The other dimension of Turkey s jobs deficit is the large share of employment in the informal sector close to 50 percent of the employed labor force is not registered with the social insurance institute. The Government recognizes that a range of factors are constraining employment growth. The Ninth Development Plan raises a number of issues that are negatively affecting the functioning of the labor market and hampering job creation. These include low participation rates especially for women, a lack of responsiveness of the education system to labor demand, underdeveloped active labor market programs, insufficient flexibility in the labor market, and high non-wage labor costs, including social insurance contributions and personal income taxes. 1 This synthesis paper was written by Gordon Betcherman (task team leader) and Carmen Pagés (co-task team leader). Other members of the research team included Erol Taymaz, Meltem Daysal, and Kerry Papps. Zafer Mustafaoglu and Rodrigo Chaves contributed to the overall design and policy implications, as well as taking responsibility for liaisons with the relevant Government agencies. Ufuk Guven, Tomaz Rejec, Milan Vodopivec, and Anita Schwarz were responsible for the pension fund and unemployment insurance fund simulations. Necdet Kenar provided helpful comments and policy advice. The comments of the peer reviewers, Bill Maloney, Ana Revenga, Jan Rutkowski, and Mathew Verghis, have been very helpful. The project benefited greatly from collaboration with many Government agencies including Treasury, State Planning Organization, Ministry of Labor and Social Security (including the Social Security Institute and ISKUR), Turkstat, Ministry of Finance, SSK, and Bag Kur. The project team also benefited from discussions with TISK and TURKIS as well as academic and research experts. vi

7 Employment rates are far below EU averages and the Lisbon targets* Overall employment rate Male employment rate Female employment rate Turkey EU-27 * Employment rates for the population aged in Lines denote Lisbon targets for overall employment and for female employment. Even with the recent reforms, the gap between total labor costs and take-home pay for lowwage workers and workers with families in Turkey is among the highest of all European countries in the OECD. This large tax wedge stems from high social insurance contributions (by employers and employees) and income taxes levied on workers and the lack of progressivity in the labor tax system. For a single worker in Turkey paid at the average wage or above, the size of this tax wedge about 40 percent of gross wages -- is in the middle range among OECD countries elsewhere in Europe. However, for low-wage workers and workers with dependents, Turkey s relative position worsens considerably. For example, the tax wedge for an employee with two children earning 67 percent of the average wage is higher than anywhere else in the OECD. This is because, unlike higher-income countries, Turkey s tax burden is not progressive but remains relatively constant regardless of income level or family situation. The minimum living relief introduced at the beginning of 2008 did introduce some modest improvement in this situation. This high tax wedge is a potentially important obstacle to employment, especially in the lowwage sector that includes many women and young people. A high tax wedge can be expected to reduce employment in the formal sector. To the extent that social contributions and personal income taxes raise total labor costs, they decrease employer incentives to hire new workers. At the same time, by cutting take-home pay, a high tax wedge reduces incentives for workers to participate in the labor market. A number of countries, including the Netherlands and Belgium, for example, have introduced measures to lessen the tax burden, especially in the case of lowwage workers and women and young people, not only for equity reasons but also to encourage their employment. Very few reforms have been systematically evaluated and, where careful studies have been undertaken, the estimated impact can vary considerably: for example, researchers found that the cuts increased employment in Belgium but not in the Netherlands. The availability of solid evidence on the likely employment and fiscal effects is a key challenge for the consideration of labor tax reform options in Turkey. Although lower taxes on labor would encourage job creation, the magnitude of the effect is not known. At the same time, a vii

8 direct effect of reduced labor taxes would likely be to lower fiscal revenues. This is especially important because of Turkey s high social insurance deficit. This empirical study aims to estimate quantitatively the employment and fiscal implications of reductions in labor taxes. Based on data from enterprise surveys, the Household Labor Force Survey, and social insurance administrative data, the analysis uses various methodologies to estimate how much new employment would be created if labor taxes were reduced. Then, using social insurance simulation models, the study estimates how increased registration with lower contribution rates would affect the fiscal position of the pension and unemployment insurance funds under various reform scenarios. It should be noted that this analysis was carried out before the social contribution changes introduced as part of the 2008 employment package. Conclusion 1: Employment in Turkey is responsive to changes in labor costs. This is the common conclusion of various analytical exercises undertaken using different data and methodologies. Our estimates indicate that elasticity of labor demand how much employment changes as labor costs change -- is in the -0.4 to -0.6 range, comparable to what has been observed in other middle-income and developed countries. Turkey s labor market also responds relatively quickly: most of the employment adjustment to changes in labor costs occurs in less than 18 months. In addition, an evaluation of subsidies in low-income provinces concluded that reductions in total labor costs have significantly increased registered employment in eligible provinces. Finally, employment probabilities for low-wage workers especially for women and youth -- are affected by policy changes that alter the minimum wage and the social insurance contribution base. Reducing labor taxes, then, would increase registered employment in Turkey. Conclusion 2: However, the effect of lower taxes on employment would be diluted because a significant portion of the reduced tax would be captured by workers through higher wages rather than by employers through lower labor costs. While our results show that reducing labor costs can significantly boost registered employment, the full impact of changes in labor taxes is complicated by the fact that employers and workers actually share the burden of the taxes, regardless of who they are statutorily levied on. The analysis indicates that much of the effect of a tax reduction would be passed through to workers in the form of higher wages, rather than to employers through reduced labor costs. Because of this, a one percentage point cut in social insurance contributions for an average-wage worker would reduce total labor costs by less than 0.50 percent with the result that employment would increase by under 0.25 percent. This passthrough effect would limit the overall employment gains of a general reduction in labor taxes. Conclusion 3: Consequently, across-the-board reductions affecting all employment would not likely be cost-effective as an instrument for increasing formal jobs. Because much of the effect of an across-the-board tax cut would be absorbed through increased wages especially for higher-wage workers, it could achieve a major employment impact only if it is large. A more modest across-the-board reduction would only generate limited employment gains and at a significant cost. We simulated the effects of reducing employer pension contributions by 5 percentage points and eliminating their 2 percent contribution to the UI fund. Our estimate is that this would increase registered employment by about 1.1 percent (85,000 jobs), which would reduce the unemployment rate by about 0.35 percent (based on 2005 HLFS numbers). However, viii

9 the cost of this reform to the pension fund would be considerable, initially increasing the fund s deficit by 0.75 percent of GDP and worsening somewhat over much of the simulation period. The goal of cost-effective job creation would be better served if cuts were targeted at lowerwage labor. At lower wage levels, much more of the effect of a tax reduction decreases labor costs for the employer. According to our estimates, about three-quarters of a tax cut targeted at minimum-wage labor would be captured by employers through lower labor costs, and only a quarter by workers through higher wages. As a result, the employment effect would be comparatively stronger than when general tax reductions are offered. In other words, tax cuts targeted towards low-wage labor would be more cost-effective in terms of stimulating employment gains. Indeed, simulations show that reductions in contributions oriented towards low-wage workers, compared with across-the-board cuts, increase employment more at the same cost or achieve comparable employment increases at lower cost. The simulations to assess the impact of focusing tax cuts on low-wage labor assumed contribution reductions targeted at workers under the age of 30. The decision to target youth in these simulations stems from the concern that explicit targeting on low-wage workers would create incentives for employers to officially register employees at low wage levels, even if their real earnings were higher. Youth are a good proxy because they are identifiable and are heavily represented in the low-wage sector. (In 2005, over 50 percent of workers earning less than 1.25 times the minimum wage were under 30.) The analysis suggests that the targeted reduction again reducing employer pension contributions by 5 percentage points and eliminating their 2 percent contribution to the UI fund - - would increase employment (of young workers) by 70,000, not much less than the 85,000 with the general tax reduction. At the same time, the targeted option would have only a relatively small negative effect on the balance of the pension fund (about 0.2 percent of GDP). Under all scenarios, the registered employment gains of a cut in social insurance contribution rates do not broaden the tax base enough to compensate fully for the reduced contribution rates. Even under the more cost-effective targeted options, Turkey would face some additional fiscal pressures because employment gains would not be sufficient to make up for lower contributions per worker. These pressures would be greatest with general tax reductions. In Turkey s current fiscal situation, this is an important factor if the authorities were to consider, for example, reductions in social insurance contributions. In 2005, general budget transfers equal to 4.8 percent of GDP were needed to meet social insurance fund obligations, largely for pensions. Ceteris paribus, this transfer would need to increase under the types of scenarios evaluated in this study. To achieve overall fiscal neutrality, policy-makers would need to find revenues from other tax sources or decrease public expenditures to compensate for the costs of labor tax reductions. The adjustments on either the revenue or expenditure side would, in turn, have implications for employment. These second-order effects have not been included in this study but would need to be considered for a complete assessment of the implications of reducing taxes on labor. If the authorities were to consider reductions in Turkey s tax wedge, this study suggests that targeted cuts, especially for low-wage workers such as youth or women, would be the ix

10 preferable option. This conclusion is based on the fact that the relative burden of employment taxes is greatest for low-wage workers and our finding that reductions targeted at this group would have the most concentrated employment impact. This study only explicitly considered cuts in social insurance contributions. However, income tax reforms to make the taxation of personal income more progressive could also be considered. In any event, there is a strong case for reducing Unemployment Insurance contributions which would contribute modestly to employment -- because of the huge surplus in the Unemployment Insurance Fund. Over the longer run, as the fiscal position of the other social insurance funds and the general budget becomes more favorable, there will be additional scope for further reductions. The Turkish authorities might consider an effectively targeted reduction of social insurance contributions that increases employment as part of a comprehensive reform strategy to address the country s jobs deficit. Although this study has focused on labor tax reductions, a comprehensive reform program is required to increase formal employment, as recognized in the Ninth Development Plan. Analytical work undertaken by the Government, the World Bank, and others suggests that this might include not only efficiently-targeted reductions in social security contributions, but also more flexible contracting arrangements, reduced severance obligations, improvements in unemployment insurance and active labor market programs, and stronger labor market institutions to administer and enforce policies and regulations. x

11 Estimating the Impact of Labor Taxes on Employment and the Balances of the Social Insurance Funds in Turkey I. INTRODUCTION Job creation is a major challenge for Turkey. Certainly, the current economic crisis has caused rapid increases in unemployment. However, even before the slowdown started, employment creation had been slow despite strong economic growth in the years after the last financial crisis in This has been a major concern for the Government. As a result, increasing employment has been identified as one of the development axes in its Ninth Development Plan, covering (Government of Turkey 2006). The Plan sets a target of 2.7 percent employment growth per year during this period, a rate well in excess of the 0.4 percent annual increase registered during the period covered by the Eighth Development Plan. Turkey s employment performance lags well behind European comparators. The employment rate (15-64 years) in 2006 was only 46 percent, well below any EU member country, and far from the Lisbon targets for In fact, employment rates for men are only slightly below European averages (Table 1). However, Turkey s disadvantage is substantial for younger and older workers and especially for women: less than one in four Turkish women are employed, compared to 57 percent in the EU countries. Table 1: Turkey s employment rates (15-64 years) relative to the Lisbon targets and EU members, 2006 Overall employment rate Female employment rate Male employment rate Turkey EU EU Lisbon target Source: Eurostat A second dimension of the jobs deficit is the large employment numbers in the informal sector. Although formal employment has rebounded somewhat since late 2004, the fundamental picture remains concerning. 2 In 2005, one-half of the employed labor force was not registered with a social security institute; this figure is about one-third even when agriculture is excluded. Over 30 percent of regular (i.e., non-casual) wage employees in the private sector are not registered. Moreover, these figures on non-registration actually underestimate informality because of underreporting of wages for those who are registered. 3 Studies undertaken for the Government (e.g., Tunali 2003; Government of Turkey 2006) and the World Bank (2006a, b) have concluded that a range of factors are constraining employment 2 In 2005, according to HLFS data, growth in employment registered with SSK was 12 percent. 3 For about 50 percent of employees enrolled in SSK, earnings as reported by employers are at the minimum insurable level. 1

12 growth. The analysis underlying the World Bank reports emphasizes that many factors come into play, some outside the labor market and others directly related to labor market policies, institutions, and practices. The Ninth Development Plan raises a number of issues that are negatively affecting the functioning of the labor market and hampering job creation. These include low participation rates especially for women, a lack of responsiveness of the education system to labor demand, underdeveloped active labor market programs, insufficient flexibility in the labor market, and high non-wage labor costs, including social insurance contributions and personal income taxes levied on workers. The high levels of labor taxes figure prominently in policy discussions on employment. 4 Indeed, after the analysis reported in this report was completed, the Government in 2008 introduced changes in social contributions that were not incorporated into our calculations. Prior to these changes, combined employer-employee contributions to finance pensions and disability insurance, health insurance, unemployment benefits, and workers compensation were percent of gross wages. Income tax ranges from percent of the gross wage. 5 Using contribution and tax rates prior to the 2008 reforms, Table 2 compares the tax wedge on labor income in Turkey with the EU-15 countries (pre-2005 members) and a selection of (new accession) EU-10 countries for workers at different earnings levels and with different family characteristics. 6 Turkey s relative position varies, depending on family status and earning levels. In the cases of single individuals and married couples with no children at or above the average production wage, Turkey s tax burden is in the middle ranks in Europe. But for families and singles with children, Turkey s taxes on labor are among the highest. This is especially the case for low-wage workers with children where Turkey has the highest tax wedge of all of the European OECD countries. A high tax wedge can be expected to reduce employment in the formal sector because of its unfavorable effects on labor demand and labor supply. To the extent that social contributions and personal income taxes raise total labor costs, they decrease employer incentives to hire new workers. At the same time, by cutting take-home pay, a high tax wedge reduces incentives for workers to participate in the labor market. Many countries have introduced measures to lessen the tax burden, especially in the case of low-wage workers and women and young people, not only for equity reasons but also to encourage their employment. Indeed, studies internationally have shown that employment for these groups is particularly sensitive to the size of the tax wedge. 4 Throughout this report, labor taxes is used as a term to include both social security contributions (levied on employers and employees) as well as personal income taxes levied on employees. 5 Between 2000 and 2004, income tax rates ranged from percent. In 2005, the top rate was cut to 35 percent and the number of brackets was reduced from six to five. 6 The tax wedge is defined as income taxes and combined (employer-employee) social security contributions, minus cash benefits, as a percentage of total labor compensation. The calculations of the tax wedge are based on OECD estimates with additional calculations made by the World Bank to take into account Turkey s consumption tax credits which were not included by the OECD. Note that payroll taxes account for about 70 percent of Turkey s overall labor taxes. 2

13 Table 2: Comparing average tax wedges 1 in Turkey with European countries by Family Type and Wage Level, 2004 Single Single Married Family type No children 2 children 2 children Wage level Turkey tax wedge (%), OECD methodology Number of EU-15 countries with higher tax wedge Number of EU-4 countries 3 with higher tax wedge Turkey tax wedge (%), with consumption tax credit Number of EU-15 countries with higher tax wedge Number of EU-4 countries 3 with higher tax wedge Turkey tax wedge (%), with minimum 38,18 40,28 43,10 37,03 37,44 39,03 39,90 32,57 living relief 7 Turkey tax wedge (%), 2008 (after intended reduction of social contributions) 35,50 37,71 40,66 34,33 33,60 36,42 37,32 29,67 1. Income tax plus employee and employer contributions less cash benefits; consumption tax credits added for Turkey. 2. Figures in this row indicate wage level as a percent of average production wage. In the married family examples, wage levels for each adult are given (e.g., 67-0 means that primary earner has wages at 67 percent of average production wage and the other adult has no earnings). The last example (44-0) represents the case of a family with one minimum wage worker and a non-earner. 3. Includes EU-10 (new accession) countries that are also OECD members -- Czech Republic, Hungary, Poland, and Slovak Republic. Source: OECD (2004), with calculations by IMF and World Bank staff. 7 The calculations presented in the table do not incorporate the reforms introduced in As of January 2008, the practice of Minimum Living Relief replaced the practice of Special Expenditure Relief in the Income Tax Law, which resulted in a decrease in the tax wedge. Furthermore, additional relief in the tax wedge was introduced after the 5 percent reduction in the social security contributions of employers that took effect in October The table below shows the tax wedges after the introduction of new practices, as calculated by the Government. 3

14 Consideration of labor tax reform options in Turkey is hampered by the fact that solid evidence does not exist on the likely employment and fiscal impacts. Although lower taxes on labor would encourage job creation, the magnitude of the effect is not known. At the same time, there would also be serious financial considerations to take into account. This is especially critical because of the social insurance deficit and the need to generate a sizeable primary surplus. At the request of the Government, the World Bank has undertaken an empirical study of the impacts of reductions in labor taxes on employment and on the fiscal positions of the social insurance funds. Based on data from enterprise surveys, the Household Labor Force Survey, and social insurance administrative data, the analysis uses various methodologies to estimate how much new employment would be created if labor taxes were reduced. Then, using social insurance simulation models, the study estimates how increased registration with lower contribution rates would affect the fiscal position of the pension and unemployment insurance funds under various reform scenarios. The results and policy implications are summarized in this synthesis report. The detailed analysis on employment impacts is presented in three technical reports (Taymaz 2006; Betcherman, Daysal, and Pagés 2007; and Papps 2007). The rest of this paper is organized as follows. Section 2 provides a brief literature review on the employment effects of labor taxes and subsidies. The methodologies used for the different empirical studies included in this project are described in Section 3. Section 4 summarizes the results of our estimations of the employment effects of changes in labor taxes. In Section 5, we present a set of simulations to demonstrate what the effects of cuts in social security contributions rates would be on the fiscal situation of the pension and UI funds. Finally, policy implications are drawn in Section 6. II. A BRIEF SUMMARY OF THE LITERATURE ON THE EMPLOYMENT EFFECT OF LABOR TAXES AND SUBSIDIES As background, the key concepts and existing literature on the employment effects of labor taxes are summarized in this section. Labor tax reductions have been implemented in many countries at different times to encourage employment. Their effect on employment depends on the elasticities of labor supply and demand, and how the tax cut is distributed between lower labor costs for the employer and higher wages for workers. Economists have measured the impacts of taxes in various ways. While there is no consensus, the most common finding is that tax reductions have a small but significant effect on formal employment. However, further research is needed to generate more solid conclusions. Labor tax reductions (including social security contributions) and employment (or wage) subsidies aim to encourage employment by decreasing the cost of labor and increasing labor demand and/or by increasing take-home pay which will improve labor supply. Tax cuts and subsidies can be applied to all employees or only to new hires. They can also be general, in the sense of applying to all workers and establishments, or specific, if only certain types of workers or firms qualify. Although labor tax cuts and employment subsidies have been implemented in many countries at different times, the quantitative evidence of their impacts on employment is limited. To the extent that these impacts have been estimated, two broad approaches have been 4

15 used. The first is to make inferences based on estimates of the elasticity of labor demand. The second is to directly estimate the effects of actual tax cuts or subsidies. Regarding the first approach, labor demand elasticity estimates provide a measure of the expected change in employment in response to a change in labor costs. However, when the reason for a change in labor costs is a tax change (or a subsidy), this approach is confounded by the fact that, a priori, it is not clear what the tax incidence is i.e., whether the tax is actually being paid by the employer or by the employee. This depends fundamentally on the elasticity of labor demand and labor supply. When the incidence is fully on the employer, the result of a tax cut or a subsidy will be lower total labor costs and increased labor demand. On the other hand, when the incidence is fully on the employee, the result will be higher take-home pay for the worker but no effect on labor demand. However, in this case, there could also be an effect on employment depending on the elasticity of labor supply. These principles are the same whether the tax is statutorily imposed on the employer or on the employee (or the subsidy given to one or the other). In reality, tax reductions or subsidies are typically shared by the two parties and, assuming competitive labor markets, both employment and wages will increase. Other factors can also come into play. One is whether minimum wages are binding. In that case, an excess supply of labor would imply that firms can recruit more workers without having to increase wages. This situation is similar to the case when the labor supply is fully elastic: a tax cut or an employment subsidy yields maximum employment expansion and no effect on wages. As we will see later in this report, this has potentially important implications for thinking about labor tax reforms that will maximize employment gains. Second, in non-competitive labor markets, the tax incidence can also be determined by the relative bargaining power of employers and employees. Another key issue is how workers perceive the value of the benefits financed by their social contributions or income taxes. If they fully value these, then in the case where taxes been reduced, they will be more likely to demand higher wages as compensation for the expected lower social benefits. The existing literature offers some guidance on the plausible range of labor demand elasticity estimates, although most of the studies are based on data from industrialized countries. The international evidence suggests that the likely range is between and (i.e., a 10 percent decrease in the cost of labor would cause employment to rise from between 3 percent and 5 percent). 8 Recently, there have been a growing number of studies from developing and transition countries, with most of the (long run) elasticities estimates in the to range. 9 There have been a small number of labor demand studies in Turkey, which also yield results that are roughly consistent with findings elsewhere (Taymaz 2006, Table 1). Estimates of (long-run) elasticities for Turkish manufacturing employment range from to -0.57, while one study also calculated estimates for services (-0.20) and construction (-0.27). 8 This range is based on the literature review in Vroman and Brusentsev (2005), who rely heavily on Hamermesh (1993). They present these estimates with some caution, because of the lack of empirical evidence from middleincome countries. A more recent study of EU-8 countries (World Bank 2005) finds a higher elasticity of employment of between There is, however, some possibility of bias in these estimates relying on macro data, although it is not clear in which direction any bias might go. 9 For a list of these studies, see Taymaz (2006). 5

16 As noted above, labor demand elasticities do not fully capture the employment effects of changes in labor taxes because that depends also on the tax incidence. Thus, researchers need to take into account what economists call the pass through i.e., the extent to which labor taxes are shifted on to employees. Studies in middle-income countries provide a wide range of estimates which indicate that, in some cases, the pass through can be quite large. For example, research in Latin America suggests that anywhere from percent of the employer s social security contributions are passed on to the worker (Heckman and Pagés 2004). At the upper end of this range, most of the effect of a tax reduction would be captured by employees through higher wages, with little effect on employment. Indeed, this is what Gruber (1997) found to be the result of a major payroll tax cut (of 25 percentage points) in Chile between 1979 and In an analysis using firm-level data, he found that wages adjusted fully to the tax cut so that there was no employment effect. However, the most reasonable assessment based on the complete literature is that labor taxes do have a modest effect on employment. This is the conclusion drawn by Nickell (2003) who assembled the results of a number of studies, albeit only from OECD countries. He concludes that a 10 percentage point change in the tax wedge can be expected to affect employment by between 1-3 percent, a relatively small but by no means insignificant effect (p. 8). On the basis of cross-country regressions for Eastern European and Central Asian countries, Rutkowski (2007) estimates that a one percentage point change in the tax wedge results in a percent change in the employment rate. It should be noted, however, that these studies refers to tax changes affecting all workers. The limited evidence available does suggest that, because passthrough rate seems to decline around the minimum wage, the effect of tax reductions might have larger effects for low-wage workers. The second approach directly estimating the employment effects of different tax cuts or subsidies -- has been used on a limited basis only. Bishop (1981) employed a time-series methodology to evaluate the employment impact of the U.S. New Jobs Tax Credit and found that the program increased aggregate employment by percent. More recently, Katz (1998) evaluated another U.S. scheme, the Targeted Jobs Tax Credit which was available to employers hiring workers defined as vulnerable and disadvantaged. Using a difference-indifferences methodology, Katz estimated that a 15 percent reduction in labor costs because of the credits yielded a net employment effect of 7.7 percent; under the assumption of an infinitely elastic labor supply, this implies an elasticity of labor demand of -0.5, a value within the range found by Hamermesh (1993). In the context of middle-income countries, Galasso, Ravaillon, and Salvia (2001) evaluate Proempleo, a wage subsidy scheme targeted to workers in temporary employment in Argentina, and find that the program provided assistance to low-wage workers in finding regular wage employment, although effects were only statistically significant among women and youth. On the other hand, Muhlau and Salverda (2000), using time-series analysis, found no employment effects of SPAK, a Dutch program to reduce taxes and social security contributions paid by employers for workers with wages around the minimum wage It should be noted, however, that studies using a time-series approach may not provide a good identification of the effects of the scheme relative to the effects of other factors that also influence employment. 6

17 Not only are there few studies directly estimating the employment effects of tax cuts and subsidies, but the literature has some important gaps. There appear to be no attempts to assess whether employment growth that does occur is due to the creation of new jobs or the formalization of existing jobs. Another issue that has not been analyzed is whether job gains are the result of the expansion of existing firms (i.e., in the intensive margin) or because of an increase in the number of firms (extensive margin). Another parameter of key importance and about which little is known is the deadweight loss i.e., what share of subsidized employment would have been created in the absence of the incentives. Most estimates are obtained from interviews with employers rather than from quantitative estimates. Even though employers may not have the right incentives to report on the actual numbers they would have employed in the absence of subsidies, such estimates still suggest large deadweight losses. Estimates range from around percent for marginal subsidization under targeted programs up to 93 percent for non-targeted, across the board measures (Marx, 2005). Another important gap in the literature is the effect of tax cuts and subsidies on wages. III. STUDY METHODOLOGY This section describes the methodology and data used in the different lines of analysis that comprise the complete study. The approach involves two stages (Figure 1). The first analyzes how employment in Turkey responds to changes in labor taxes. Three methods are used: (i) calculating labor demand elasticities and pass-through rates based on establishment data; (ii) evaluating the number of jobs created due to subsidies offered under regional development programs using social insurance administrative files; and (iii) estimating how employment probabilities for low-wage workers respond to changes in the cost of their labor because of changes in the minimum wage and social insurance contribution base. The rationale for this multi-pronged approach is to increase the robustness of the findings. In the second stage, with our estimates of the employment effects of changes in labor taxes, the financial impacts of the new employment levels and the reduced tax rates on the pension fund and the UI Fund are projected using social insurance simulation models. 7

18 Figure 1: Summary of study methodology Reduction in labor taxes Employment impacts Establishment data, SSK data, Household Labor Force Survey PROST, UISIM models Methodology and data for estimating employment impacts The estimates of employment impacts of tax reductions are based on three different methodologies and data sources. (a) Establishment-level analysis of labor demand This part of the study analyzes how employment in enterprises responds to changes in the cost of labor. The methodology and results are reported in detail in Taymaz (2006). The empirical work is based on Turkstat datasets for the manufacturing industry and construction. These sectors were selected because of the availability of panel data at the establishment level. Information on the data sources is provided in Box 1. Unfortunately, appropriate data were not available for the service sector. The enterprises included in these datasets can be assumed to operate in the formal sector; as a result, the analysis cannot provide insights into how informal employment responds to labor cost changes, nor how employment shifts between the informal and formal sectors. 8

19 Box 1: Establishment data used for labor demand and pass through analysis The econometric modeling of labor demand elasticities and labor tax pass-through is based on three Turkstat datasets. For more details, including variable definitions see Taymaz (2006). The Annual Survey of Manufacturing Industries (ASM) covers all private establishments employing 10 or more people and all public establishments, and primarily includes information on production-related aspects (number of employees, wage payments, costs of inputs, sales, investment, etc.). The dataset used in the study covers the period from 1992 to The empirical analysis reported in this paper is based on all private establishments only. Establishments are classified into three sub-sectors: i) capital goods and consumer durables, ii) consumer goods, and iii) intermediate products, in accordance with the European Commission definition. The Short Term Production Surveys were conducted quarterly (large sample) and monthly (small sample) until 2005, and are now being conducted on a monthly basis for the large sample. The larger sample covers about 3,000 establishments that produce 90 percent of output at the 4-digit industry level. The content of the quarterly survey is more limited than that of the Annual Survey, but does include data on production, sales, employment, and wages. Data from the first quarter of 1988 to the last quarter of 2005 are included in the analysis. While establishments in the sample change over time, our panel analysis is restricted only to those establishments included for at least 68 quarters. The construction survey is conducted annually, and our analysis of establishments in this sector covers the period The basic structure of the construction survey is similar to that of the manufacturing survey, and includes questions on the number of employees, wage payments, costs of inputs, construction services provided, investment, etc. The analysis involves calculations of labor demand elasticities and pass-through rates of employer social security contributions. As we will see in the next section when we discuss results, the pass-through is a major factor in determining the magnitude of the employment gain from a labor tax reduction. In this study, we measure the pass-through as the percentage change in labor costs for a 1 percent change in employer social security contributions. 11 Taymaz (2006) formally develops and specifies the econometric models used for calculating the elasticity and pass-through estimates. 12 Briefly, labor demand elasticities are estimated using the various panels described in Box 1 according to a dynamic fixed effects model that follows the common practice in the literature. This model estimates the number of employees (or hours 11 This is actually the inverse of the pass through. For example, if three-quarters of a 1 percent reduction in employer contributions is passed on to employees through higher wages, we calculate the pass-through measure as 0.25 (i.e., labor costs are reduced by 0.25 percent). 12 To handle various econometric issues related to endogeneity of some variables and the use of lagged dependent variables, Taymaz (2006) uses the GMM-System estimator. The models are estimated using first differences (fourth differences in the quarterly models) to eliminate unobserved firm-specific effects. 9

20 worked) in an establishment at a point in time as a function of its average real cost of labor, real cost of capital, and real output, as well as unobservable firm-specific effects. The labor demand elasticity is calculated on the basis of the estimated coefficient for the labor cost variable. The preferred specification of that variable is total labor costs. This is available in the annual data but gross wages are used in the quarterly model because total labor costs are not reported in the quarterly surveys. 13 The lag value of employment is included in the labor demand model to estimate the speed of employment adjustment. To calculate the pass-through rate, Taymaz estimates a separate wage equation using the annual data which include employer social security contributions. The model estimates the real gross wage in an establishment at a point in time as a function of the social security contribution rate (the ratio between the employer s contribution and gross wages), and various control variables including the real minimum wage rate, the market share of the firm, the proportion of female employees in labor force, the capital/labor ratio, and a leading indicators index. The passthrough calculation is based on the coefficient for the social security contribution rate. This model also accounts for unobservable firm-specific effects. The lagged depended variable is included into the model to allow for a partial adjustment process. (b) Analysis of regional subsidies This study attempts to directly measure the employment impacts of subsidies offered by the Government of Turkey to encourage investment and employment in low-income provinces. The program details, and the methodology and results are reported in full in Betcherman, Daysal, and Pagés (2007). The research covers three different incentive regimes, legislated through Law 4325 (1998) which covered 22 provinces, Law 5084 (2004) which expanded coverage to an additional 15 provinces, and Law 5350 (2005) which added 13 more provinces. 14 Given that our data do not cover the period before Law 4325 was enacted, the econometric analysis of the net employment impacts focuses on the subsidies offered under Laws 5084 and Although there have been some differences in qualification requirements and the actual subsidies, these two laws have included four subsidy components: (i) reductions in employer social security contributions; (ii) credits on income taxes on wages; (iii) subsidies for electricity consumption; and (iv) land subsidies. Employers in eligible provinces have qualified for the various subsidies on the basis of meeting new job creation thresholds, either by opening new establishments or by expanding employment in existing ones. Since these subsidies reduce the cost of labor in some provinces but not in others, the programs can be examined to estimate how much new employment is likely to be created when taxes or social security contributions are reduced. Often the impact of programs such as these is assumed by policy-makers to be simply equal to the number of new jobs that have been subsidized. However, this does not take into account the job creation that would have occurred if the 13 One potential issue concerns the underreporting of wages. It has often been observed that a large number of firms operating formally tend to underreport wages to avoid taxes. Underreporting can lead to downward bias in the estimated values of wage elasticity because the actual wage cost is likely to be higher. However, if the behavior of firms to underreport wages is consistent over time, the firm-specific unobservable effects in the model can capture these effects as well and lead to unbiased estimates. 14 One of the 22 provinces covered under Law 4325 (Tunceli) was not covered under Law 5084 but was covered again under Law

21 program had not been introduced. To do this requires a control group that approximates a reasonable counterfactual, while controlling for other factors that determine employment levels. Eligible provinces under Law 5084 had a per capita GDP of less than $1500 (in 2001) or had been designated as priority development regions. All provinces (but one) covered under the existing regional incentives law (4325) qualified for this new program and any firm that received subsidies under Law 4325 could choose to continue under the old regime or switch to the new one. Beginning in January 2004, employers could qualify for monthly subsidies under Law 5084 based on the additional registered employment they reported above a benchmark based on their October 2003 level. New start-ups were eligible for subsidies for their total registered workforce. The first subsidies were paid in March Law 5350 came into effect in May 2005 and modified a number of provisions in Law 5084, in addition to extending coverage to 13 more provinces with low socio-economic development according to an index elaborated by the Turkish State Planning Organization. Again, any firm that received subsidies under the previous law could choose to continue to receive subsidies under that law or switch to the new law. One major difference between Laws 5350 and 5084 was that an employment threshold of 30 employees was set as a condition to qualify for subsidies under the new law. 15 At the same time, Law 5350 provided for social insurance and income tax subsidies equal to three times the amount payable to a new hire, up to specified maximums. This implied that, for every eligible worker, the law could potentially subsidize two already-employed workers. The employment impacts of the regional incentives are analyzed using difference-in-difference models that estimate registered employment (levels and growth) in a province in a given month as a function of whether the regional incentives program is in force, the time period, and provincial variables that cover province-specific effects. Various alternative specifications incorporating these determinants in different ways are estimated. 16 The estimation approach used is to compare the change in employment between the period before and after the introduction of subsidies in the provinces that benefit from them ( treated provinces ) with the change in employment in provinces that are not covered ( control provinces ). To clarify the specification of control and treatment groups, the following groups of provinces are defined: (i) 4325-subsidized : the 22 provinces that were subsidized under Law 4325; (ii) 5084-subsidized : the 15 provinces added under Law 5084, but not covered by Law 4325; (iii) 5350-subsidized : the 13 provinces added under Law 5350, but not covered by Law 5084; and (iv) never-subsidized : those provinces that were never covered under any of the three programs. To estimate the net employment effect of Law 5084, two control groups are used subsidized and never-subsidized. The former has the advantage of providing a better counterfactual in that it includes provinces that are similar in terms of income, geographic 15 Newly-created firms were now required to have at least 30 registered employees, while existing firms also had to have at least 30 employees and to have increased employment by at least 20 percent above a benchmark based on their January 2005 level. 16 For an example of the general methodological approach, see Autor, Donohue III, and Schwab (2006). 11

22 location, and other characteristics to the provinces in the 5084-subsidized group, but fell above the minimum income per capita threshold to qualify for subsidies under Law On the other hand, it could be argued that substitution effects will be stronger between similar and geographically close provinces than between provinces that differ substantially in income and other attributes, which suggests that the never-subsidized group may give a better idea of the effects of the subsidies net of substitution effects. The options are more restricted for estimating the effect of Law 5350, with the never-subsidized group specified as the only control group. In fact, it should be noted that, although the study includes estimates of the effects of both Law 5084 and Law 5350, the analysis is on stronger footing for the earlier law than the later one. The impacts of Law 5084 can be estimated under two different control groups, while the Law 5350 analysis can only use one. Moreover, given the available data, the observation period for Law 5350 is short (only 8 months). The models are estimated with three dependent variables. The first, registered employment, is used to estimate the net employment impact of the subsidies, as already discussed. The second, registered workplaces, allows for an assessment of how much of the estimated employment gains have been due to the addition of new firms (extensive margin) or the expansion of existing ones (intensive margin). The third, average taxable earnings, is used to estimate what effect the subsidies have had on wages and the actual incidence of the subsidies (i.e., who benefited, employers or workers). Estimations are based on administrative data provided by the Social Security Administration (SSK). These data are reported by provincial unit for every month and cover the number of registered workers, their insurable earnings, SSK premiums, number of workplaces registered, the number of newly registered employees and workers, as well as other variables. 17 The SSK data also include files with a number of variables describing new registrations under the regional incentives laws, which we use in the cost calculations. However, the data on total registrations are used for the econometric modeling described above. The models are estimated using the monthly data for the period covering The calculation of the costs incurred under the program includes the social security, income tax, and energy subsidies. The social security costs are provided in the SSK database described above. We do not have data on the income tax subsidy costs but they have been estimated as a proportion of the social security subsidies. 18 The costs for the energy subsidies by month and province have been provided by the Treasury. Land has not been included in the cost calculations because neither data nor a method for approximating these costs is available. 17 The data are actually provided on a sub-provincial basis (i.e., SSK reporting unit). The sub-provincial data were aggregated up to a provincial basis for each month. 18 During the period between January and June 2004, the income tax subsidies under the program are assumed to be 56 percent of the observed social security subsidies. Between July 2004 and December 2005, this figure is calculated at 73 percent. These proportions are calculated on the basis of differences between the employer social security contribution rate on the minimum contribution base and the income tax rate on minimum wages. The change in the ratio between the two tax subsidies reflects the fact that the minimum contribution base was higher than the minimum wage in the first period in the first period and then the two were set at the same level from July 2004 on. Actual calculations are shown in Betcherman, Daysal, and Pagés (2007). 12

23 One additional aspect of the analysis was to assess the extent to which gains in registered employment because of the subsidies were due to real increases in economic activity or to the registration of previously unregistered activity. Data availability limited our ability to examine this issue. 19 The approach used was to estimate a model of electricity consumption, which is a commonly used proxy for economic activity, using annual provincial data between 2002 and It should be recognized that, even here, data are a limiting factor since we are only able to use three observations per province, with only one after Law 5084 was introduced. Moreover, the effect of Law 5350 cannot be tested with the available data. (c) Analysis of changes in labor costs using Household Labor Force Survey data This study analyzes how employment probabilities change for low-wage workers in response to policy-induced shifts in the cost of their labor. The data for this research come from the quarterly Household Labor Force Survey (HLFS). In addition to providing another line of evidence on how employment is affected by changes in labor costs, the HLFS analysis can offer some insights into the interactions between the formal and informal sectors and how different types of workers (e.g., women, young) are affected by changes in labor costs. Details on the methodology and results are reported in Papps (2007). The policy changes considered in the analysis are the social security contribution base and the minimum wage. In recent years, there have been regular changes in these parameters that have implied non-trivial variations in the cost of employing low-wage workers. Annex Table A1 summarizes these changes and their implications for the real labor costs of workers at the minimum wage. 20 The impacts of these changes can be evaluated by estimating models that relate the probability of low-wage workers continuing in employment as a function of changes in the cost of their labor. The analytical approach is based on (i) observing an individual s employment status and earnings at one point in time (a particular quarter); (ii) calculating how total labor costs in employing that individual at a second point in time (a later quarter) would have changed because of changes in minimum wages and the contribution base; and (iii) observing the individual s actual employment status during the second observed point (quarter). 21 The probability of being employed during the second observed quarter is estimated in a model which includes individual characteristics and a treatment variable that measures the change over the two quarters in the total labor cost required to employ the worker at his or her gross wage (as observed in the first 19 Available data from the HLFS were not disaggregated to the provincial level so comparisons of formal and informal employment between covered and uncovered provinces were not possible. Another possible approach would be to use province-level GDP data to examine whether economic activity has increased in the treated relative to the control provinces but these data were not yet available for the period under analysis. 20 It should be noted that prior to July 2004, the contribution base was sometimes at a higher level than the minimum wage. In fact, this was the case in When this occurred, in addition to their contributions, employers were required to pay the employee s social security contributions in the earnings zone between the employee s actual wage and the contribution base. In July 2004, the contribution base and the minimum wage were synchronized. Actually, for employers, they were effectively synchronized in the January-July 2004 period because the government paid the employee contributions in the zone between the minimum wage and the contribution base. 21 This methodology has been used by economists most commonly to assess the impacts of minimum wage changes. For example, see Currie and Fallick (1996) and Kramarz and Philippon (2000). 13

24 quarter). In order to focus the analysis on workers who would potentially be affected by the minimum wage and the contribution base, the estimation sample is restricted to wage employees who earned a real wage (at 2003 prices) of a maximum of 500 million TL. The analysis requires panel data in order to observe individuals at two points in time. While the quarterly Household Labor Force Survey follows a repeated sampling approach, 22 unfortunately it was not possible to obtain longitudinal household identifiers, except within 2004 and Accordingly, a longitudinal panel could not be created across these years or to include any earlier years. Given this situation, Papps (2007) takes two approaches. The first involves estimating the employment models using the panel data for The minimum wage and contribution base changed in July 2004 so, given the HLFS sampling procedure, some individuals can be observed before and after the policy change. Nonetheless, the observation period is short (at most 4 quarters) and the analysis can only consider one policy-induced change in labor costs. As an alternative approach, Papps creates a synthetic panel which does not follow individual workers per se but tracks demographic cells where each observation refers to a combination of (narrow) demographic characteristics that is repeated across quarters. This allows data from to be included. With this synthetic panel, what is being modeled is no longer the probability of a particular worker remaining employed across two quarters, but the change in the employment rate between two quarters for a particular demographic cell. Nevertheless, in most cases, the cells consist of a single person, meaning that the synthetic panel largely resembles an individual panel. Methodology for estimating the fiscal implications The second stage of the analysis (recall Figure 1) considers the fiscal implications of tax reductions. Although, in principle, labor taxes could be reduced by decreasing personal income taxes, we assume that any reform would focus on the social security contributions. The simulations are based on reductions in employer contributions since these are the basis of the estimates by Taymaz (2006) and since, as we argue later in this report, reductions in employer rates are likely to have a greater impact on employment than reductions in employee contributions. The future fiscal impacts of reform options for reducing social security contributions are simulated by using two World Bank social insurance simulation models the PROST model for pension insurance and the UISIM model for unemployment insurance. These computer-based models are designed to systematically model fund revenues and expenditures over a long time, according to program parameters, population and labor force trends, and macroeconomic conditions. These simulations are based on country-specific data and assumptions about the future. The PROST and UISIM models are designed to answer a range of policy-relevant questions including the one at the heart of the present analysis: What would be the implications of alternative scenarios regarding the number of contributors and program parameters (e.g., 22 Households are sampled for two consecutive quarters, then are out of the sample for two quarters, and then are brought back in for two final quarters. In principle, with consistent household identifiers, this should allow for tracking household members over an 18-month period. However, Turkstat reported that the availability of identifiers was limited, as described in the text. 23 There were no policy changes affecting the minimum wage or the social security contribution base in

25 contribution rates) for the future expenditures, revenues, and fiscal balance of the pension and unemployment insurance funds? 24 The approach for this part of the study is straightforward. First, different reform options for reducing employer pension and UI contributions are put forward. The specified contribution rates under each reform scenario are entered into the model, replacing the current (base case) rates. The employment effects through increases in the number of contributors are calculated using the estimates of labor demand elasticities and pass-through rates, based on the establishment and regional subsidy analyses. Then the model simulates the joint effects of the reduced contribution rates and the increased number of contributors on the pension and UI fund revenues, expenditures, and fund balance, relative to the existing base case projections. The UISIM projections go out to 2022 while the PROST projections extend all the way out to IV. RESULTS ON EMPLOYMENT IMPACTS This section summarizes the results from the studies on how employment reacts to changes in labor costs. All lines of analysis are consistent in finding that employment is quite responsive. Indeed, estimates of labor demand elasticity based on establishment and social insurance data are comparable to those found in other middle-income and OECD countries. The research also shows that low-wage labor is most responsive to changes in the cost of labor. The analysis of the establishment data demonstrates that the employment effects of changes in social insurance contributions are much greater at minimum-wage levels than at or above the average wage. The responsiveness of low-wage employment to changes in labor costs is confirmed by the HLFS analysis. These data also show that women and young people are particularly affected. 4.1 Estimates of labor demand elasticities and pass-through rates As noted in the previous section, Taymaz (2006) estimates a number of labor demand functions for manufacturing and construction based on establishment survey data. 25 Table 3 reports the constant output long-run total wage elasticity estimates for manufacturing and construction for production workers, administrative employees, and all employees. 26 These elasticities are calculated at the mean values of the variables. The manufacturing estimates are estimated for capital goods, consumer goods, and intermediate products. Separate labor demand functions for manufacturing were estimated, based on annual data and quarterly data. Although the quarterly 24 The study does not consider implications for the other major fund, for health insurance. This was done for practical reasons including the fact that a new health insurance regime was being instituted at the time the analysis was carried out. Accordingly, our methodology assumes that any social security contribution reductions would involve and affect only pensions and unemployment insurance. However, the World Bank has a newly-developed health insurance simulation model that, in principle, could be applied to this study. 25 Although the surveys include public-sector establishments, these functions are reported only for the private sector which is the focus for the study. The results are quite different. 26 Constant output estimates do not take into account second-order employment effects of labor cost changes associated with changes in product demand from price changes. Estimating labor demand elasticities without an output constraint is problematic. But it is reasonable to assume that constant output elasticities are downwardly biased since they do not account for output effects. Long-term elasticities refer to the completed employment effect, after the full adjustment to new labor costs have taken place. 15

26 data cover a longer period of time, the annual data are richer in terms of the variables covered, including the measure of labor cost. 27 For this reason, we use the estimates based on the annual data for assessing policy implications and, as a result, we only report these in this synthesis report. Table 3: Long-term total wage elasticities, manufacturing industries and construction, annual and quarterly data Production workers Administrative workers All workers Manufacturing -- Capital goods Manufacturing Consumer goods Manufacturing Intermediate products Construction Source: Taymaz (2006, Table A1a) Looking first at the elasticities for all workers, the estimates for the four sectors are roughly clustered in the -0.4 to -0.6 range. This means that total employment increases (decreases) by percent for a 1 percent decrease (increase) in labor costs. In all sectors, production employment is much more responsive than administrative employment to labor cost changes. These results are broadly consistent with the existing estimates for Turkey and internationally that were cited in section 2. In fact, our calculations are towards the higher end of the common range for developed countries (-0.3 to -0.5) and for lower- and middle-income countries (-0.2 to - 0.4). In other words, these results suggest that, by international standards, labor demand in Turkey is relatively responsive to changes in labor costs. Taymaz (2006) also calculates adjustment speeds i.e., how quickly it takes an establishment to adjust (half-way) to the new employment level after a change in labor costs. For the annual data, the median length of the adjustment lag is about 1.4 years (16 months). The estimates are quite similar across the three manufacturing industries and construction. These estimated adjustment speeds are relatively quick by international standards, again pointing to the relative responsiveness of the Turkish labor market. As we explained in section 2, the employment effects of changes in labor taxes cannot be calculated on the basis of labor demand elasticities alone because the actual incidence of a tax will be shared in some way by employers and employees. Accordingly, Taymaz also calculates pass-through rates which indicate how much labor costs will change as a result of a change in labor taxes. As described in the methodology section, this is based on a wage equation which estimates a coefficient for the employer social security rate (SSC rate), measured as employer social security contributions divided by gross wages The annual data provide information on both gross wages and employer social security contributions, so that total labor costs can be calculated. The quarterly survey did not include social security contributions until 2003 so the labor cost measure used is gross wages. 28 The computed SSC coefficient in the wage equation is transformed into a pass-through rate, defined as the percentage change in total labor costs as a result of a 1 percentage point change in the employers SSC rate, 16

27 The interaction between the labor demand effect and the pass-through effect is illustrated in Figure 2. This figure uses the following average rates calculated on the basis of Taymaz s estimates for manufacturing: (1) An average pass-through rate of 0.30 which is calculated as the average of mean-wage production and administrative workers for the three manufacturing sectors, and (2) an average labor demand elasticity rate of 0.50 which is the mean of the elasticities in the three sectors weighted by their employment levels. It should be recalled that the wage and SSC data used for this analysis are establishment means and not individual worker data. To put the chart in words, for every percentage point reduction in employer SSC contributions, employment will increase by 0.15 percent. The major reason for this modest effect on employment is that most of a tax reduction goes to higher employee wages, with only a minority (30 percent) reducing total labor costs. It is important to emphasize that the effect illustrated in Figure 2 is calculated on the basis of the pass-through rate for workers at the mean wage (i.e., at the mean establishment wage). Figure 3 shows how different pass-through rates are for low-wage workers, defined as the establishment wage one standard deviation below the mean. For manufacturing as a whole, these wage levels are about.66 and.83 below the mean for production workers and administrative workers, respectively. These wage levels are slightly below the monthly minimum wage. 29 The passthrough rate is much lower for low-wage establishments in capital and intermediate goods. In two of the three sectors, then, the labor cost reduction due to a tax cut is much greater at low wage levels. Accordingly, the employment gains will be much greater for low-wage labor. At higher wage levels, much of the effect of tax reductions will be increased wages. Figure 2: The role of the pass-through on the employment effect of a labor tax reduction in manufacturing establishments 1 percentage point reduction in employer social security contribution PASS THROUGH EFFECT Reduces total labor cost by about 0.30% for average wage manufacturing worker Which increases employment by about 0.15% LABOR DEMAND EFFECT Source: Calculations based on Taymaz (2006) according to the following formula: dsη ssc + ln[1+ds/(1+s 1 )] ds[ η ssc + 1/(1+s 1 )] where η ssc is (the absolute value of) the SSC rate elasticity of wages, s 1 the initial SSC rate and ds the change in the SSC rate (Taymaz 2006). 29 This may be due to employees working less than full time (either part-time or part-month) or to employees actually being paid below minimum wages. 17

28 Three caveats should be reiterated with respect to the elasticity and pass-through findings from the establishment data. First, the estimates do not include the service sector so it is reasonable to question how well they approximate the complete labor market. Second, since the (constant output) elasticity estimates do not take into account any increases in employment that would result from greater production because of lower labor costs, they likely underestimate the actual employment effect. Third, the pass-through estimates are based on changes in employer social security contributions. Pass-through rates for changes in employee contributions and income tax have not been estimated. However, our best guess is that the rates would be similar. Figure 3: Estimations of pass-through rates for mean- and low-wage manufacturing workers 1.00% 0.90% 0.80% 0.70% 0.60% 0.50% 0.40% 0.30% 0.20% 0.10% 0.00% Percentage decrease in total labor costs resulting from a 1 percentage point reduction in employer social security contributions 0.16% 0.56% Capital goods 0.32% 0.26% Consumer goods 0.08% 0.80% Intermediate inputs mean wage minus 1 std.dev Source: Calculations based on Taymaz (2006) 4.2 Evidence on employment impacts from the regional subsidy programs As described in section 3, three laws between 1998 and 2005 provided for subsidies to firms in a progressively larger set of provinces in order to encourage investment and job creation. Administrative data provided by SSK were used to analyze the employment effects of the latter two of these subsidy programs, under Laws 5084 and 5350 (see Betcherman, Daysal, and Pagés 2007). After conducting various consistency and quality checks on the (monthly provincial) data, the analysis was restricted to 79 provinces covering the period from April 2002 to December Table 4 shows the registered employment and workplace growth rates by period and provincial coverage under each incentive regime, as defined by the legislation. The 4325-, 5084-, and 5350-subsidized groups include the provinces added under each respective law. Period 0 covers April 2002-December 2003 when 22 provinces received subsidies under Law Period 1 covers January 2004-April 2005, when 15 provinces were added under Law 5084 to those 18

29 already eligible under Law Period 2 covers May-December 2005, when 13 provinces were added under Law 5350 to those already eligible under the first two laws. In addition, the table also reports growth rates for those 32 provinces that have not been included under any of the laws ( never-subsidized ). Table 4: Mean monthly growth rates in registered employment and workplaces by type of province and period, Registered employment growth Period 0 Period 1 Period 2 (04/02- (01/04- (05/05-12/03) 04/05) 12/05) Registered workplace growth Period 1 (01/04-04/05) Period 0 (04/02-12/03) Period 2 (05/05-12/05) subsidized subsidized subsidized Never subsidized Source: Betcherman, Daysal, and Pagés (2007), based on SSK data These descriptive data show that the growth in registered employment and the number of registered workplaces increased in provinces when they became eligible for the subsidies (indicated by bold in table). The 5084-subsidized provinces experienced monthly employment and workplace growth rates of 2.0 percent and 1.3 percent, respectively during Period 1, when Law 5084 was introduced. These rates were much higher than the corresponding rates in Period 0 and the increases in this period compared to the previous one were greater than in the other province groups. A similar story describes the 5350-subsidized provinces in Period 2. The econometric analysis of the regional subsidies focuses on the effects of Law 5084 on the subsidized provinces during Period 1 and the effects of Law 5350 on the subsidized provinces during Period 2. To evaluate the impact of Law 5084, two control groups are used in our models the 5350-subsidized and the never-subsidized provinces. In order to estimate the effect of Law 5350, only the never-subsidized provinces are used as a control group. Unfortunately, the evaluation of that program cannot benefit from the comparison with a similar, though untreated, control group, as was the case with the earlier law. This, plus the fact that we can only observe the first 8 months of subsidies under Law 5350, means that our results for that scheme are based on less evidence than is the case for Law Various specifications of difference-in-difference econometric models were estimated in order to calculate the net (registered) employment impact of Laws 5084 and Figure 4 shows the magnitude of the calculated employment effect of Law 5084 under three different specifications that yield high-, mid-, and low-range estimates. As the chart shows, the size of the effect is sensitive to the model specification and the control group. The estimates of the net employment impact of the subsidies in the 5084-subsidized provinces range from 4.1 percent to 7.9 percent for the specifications using the never-subsidized provinces as the control group, to

30 percent when the 5350-subsidized provinces are the control group. The mid-point between the high and low estimates is around 8.2 percent. These effects translate into increases in the number of jobs created in the 5084-subsidized provinces ranging from close to 10,000 to 28,000, with the mid-point estimate around 19, Figure 5 shows the estimated employment effect of Law 5350 under two different specifications, with the never-subsidized provinces as the control group. The impact ranges from 9.0 percent to 14.9 percent, -- significantly higher than the comparable estimates for Law 5084 based on the never-subsidized control group. The mid-point estimate is just below 12 percent. These effects translate into a boost in the number of jobs in the 5350-subsidized provinces from just under 22,000 to over 34,000, with the midpoint estimate around 28, Figure 4: Econometric estimates of the net registered employment effect of Law 5084 in 15 added provinces, January 2004-April 2005 Source: Betcherman, Daysal, and Pagés (2007) The modeling exercise also addressed the question of whether the employment gain from the subsidies was due to new economic activity or the formalization of existing jobs. As discussed in the methodology section, this could only be tested partially for Law 5084, using electricity data which were taken as a proxy for economic activity. The estimations did not show any significant impact of the subsidies on the consumption of energy. While these results are limited by the factors discussed in the methodology section, they suggest that the subsidies increased 30 These percentage estimates were calculated as of April 2005 (at the end of Period 1). The formula used was number of new jobs created as a percentage of total employment in 5084-subsidized provinces in April 2005 minus new jobs created (i.e., what employment would have been without the subsidy effect). The actual registered employment in the 5084-subsidized group was about 248,000 in April These calculations are based on the same methodology used for Law 5084, using total employment in subsidized provinces in December 2005 (nearly 266,000) in the formula. 20