Sonia Alexandra Agudelo Ayala Universitat Autònoma de Barcelona

Abstract

This study analyzes the impact of payroll taxes (paid by firms) on employment decisions of Colombian manufacturing industry during the period 1974-2009. To estimate empirically this effect, I consider an extended version of a standard labor demand equation. This equation is estimated as an autoregressive distributed lag model by GMM. The empirical methodology has a main advantage over previous Colombian studies because of the statutory payroll tax rates, the effect of social security contributions (health and pension) and parafiscal contributions can be identified individually. The main findings are (i) a 1% increase in the social security contributions rate causes a 1.2% reduction in the labor demand ii) a 1% increase in the parafiscal contributions rate causes a 2.3% reduction in the labor demand.

JEL Classification: J23, J32. Keywords: Employment, Labor Demand, Payroll Taxes.

1. INTRODUCTION The impact of pay roll taxes on labor markets is a hotly debated issue. The critical point of most studies has been to test the presence of tax shifting and its extent. The standard partial equilibrium analysis states that the extent of tax shifting depends on the elasticities of labor demand and supply. However, most changes in labor taxation are often justified by their expected effects on employment. The Colombian labor market provides a recent example of how policy makers expect boosting employment by cutting payroll tax. The Colombian government issued a new law in 2010 which created some exceptions to the payment of parafiscal contributions (a type of payroll tax only paid by firms). Even more, in early 2013, these contributions were eliminated through a new tax reform. These measures policy become a kind of experiment due to that, first, during last four decades the Colombian labor market had only experienced increases in payroll taxes, especially in 1990s. Second, their potential negative effects on employment have not been assessed. The 1990s was a crucial decade for the Colombian labor market. During this period, the Colombian labor market experienced a process of structural reform whose main aim was to expand the coverage in health and pension services. As consequence of the reform payroll taxes were increased, including social security contributions (the part paid by the employees and the employers) and the parafiscal contributions. Nevertheless, the main source of the increases arose from the increases in the social security contributions paid by employers. Bear in mind these reached a rate of 20.5 % by 2009 whereas the parafiscal contributions 9%. In this context, the target of this paper is to analyze the impact of payroll taxes (paid by firms) on labor demand during the period 1974-2009. This study enables to assess the institutional changes up to 2009 and shed light on the expected effects of the recent policies. I perform this analysis for the manufacturing industry because first, it is the only sector with panel data information with a long time dimension and containing data on wages. Second, the Colombian manufacturing industry generated a great share of the total urban employment, 23% on average, by the period 1976-2009. And third, manufacturing industry is a specific sector where the Colombian institutional changes of 2

the 1990s could have further destroyed employment levels. The Colombian manufacturing industry was characterized by significant job cuts, 2% on average. The theoretical effect of payroll taxation on labor demand in the long-run can be analyzed by using the standard microeconomic theoretical framework of factors demand. According to this, the impact of payroll taxation might be explained as consequence of two effects: a substitution effect and an output effect. The substitution effect reflects how firms substitute capital for labor when face a payroll tax, for a given level of output; the output effect, in turn, represents the fall in labor demand due to the output reduction holding production technology constant. This study focuses on estimating the total and the substitution effects. To capture empirically these effects, I consider two extended versions of a standard labor demand model. In specific, I take as reference the models of Karanassou et.al (2007) and Antràs (2004). I include explicitly the following considerations in both models. (i) The government sets a payroll tax rate which must be paid out by firms. This tax is assumed as ad valorem and proportional to wage. (ii) The technological change increases at a constant growth rate. Regarding methodological matters, both equations are estimated as autoregressive distributed lag models by using GMM. I must highlight that, as I use statutory payroll tax rates, this study has two advantages: first, it avoids econometric problems by using an effective tax rate, such as the simultaneity in the determination of wages and payroll tax rates, and the spurious variability of payroll tax rate. Second, the effect of payroll taxes can be broken down by the social security payroll tax (health and pension) and the parafiscal contributions. Previous literature has focused on estimating the net effect on employment and wages while I focus on the effect on labor demand. Recall that when a payroll tax is imposed to firms, these can shift its burden to workers through lower net wages. This implies that the net effect of payroll taxation on employment is determined by two opposing forces: i) the contraction of labor demand ii) a compensatory effect caused by net wage cuts. Studies such as Cruces et al. (2010); Bennmarker et al.(2009); Bauer and Riphahn (2002); and Gruber (1997) found that reductions in payroll tax rates do not generate

3

significant effects on employment but they have negative effects on net wages. These findings have been interpreted as evidence of full tax shifting. Some recent evidence for Colombia is provided by Botero (2011) and Hernández (2011). Both found that eliminating the parafiscal contributions will lead to increases in the level of formal employment. Likewise, Kugler and Kugler (2009) showed that payroll tax increases have negative effects on formal employment and net wages. These results have been interpreted as evidence that there is partial shifting. That means, employers partially shift payroll tax burden to workers as lower wages, then there is a negative effect on the employment level. They argue that the partial shifting arise from the presence of the downward wage rigidity which does not allow net wages to decline enough to offset the effects on employment. In conclusion, there is a virtual lack of studies about the impact of payroll taxation on the Colombia labor market and the international empirical evidence is far from reaching a consensus, thus, the objective of this paper becomes relevant. The results of this study suggest that the institutional changes brought by the government up to 2009 affected negatively the labor demand of the Colombian manufacturing industry. Even more, they give empirical support to the policy implemented by the Colombian government in 2010. In particular, the main findings are: First, the total effect of social security and parafiscal contributions are 1.21% and 2.33%, respectively. Second, the effect of substitution for social security and parafiscal contributions are close to 1.5%. This paper is divided into five sections including this introduction. Section 2 focuses on the theoretical effects of payroll taxes on labor demand and provides an empirical implementation. Section 3 discusses the data and empirical issues. Section 4 presents the empirical results. And Section 5 concludes. 1. EFFECTS OF PAYROLL TAXES ON LABOR DEMAND 1.1.

Theoretical Underpinnings

Payroll taxes (paid by employers) are argued to have a negative impact on labor demand because these taxes represent an extra labor cost for the employers. In the long-run, this

4

impact can be explained as consequence of two effects: a substitution effect and an output effect. These can be illustrated through Figure 1. On the top panel of figure 1, the horizontal axis measures the level of employment and the vertical axis measures the capital stock

. Point

of factors

is the production function or isoquant.

) in absence of taxes, where

For a given level of wage by point

represents the optimal choice

, the optimal choice of employment

on the labor demand curve

is also represented

of the bottom panel.

If the government sets a payroll tax rate , paid by the employer; it will increase labor costs. Given the wage

) then the slope of

, the labor cost increases to

budget constraint (straight line) contracts and the optimal choice of factors is given by point B. In turn, the labor demand curve shifts to the left,

. Note that the total effect

of the payroll tax on labor demand is given by the reduction from

to

. This

reduction, as pointed formerly, arises from: the substitution effect and the output effect. Figure 1. Substitution and output effects from payroll tax changes 𝐾

𝐶

𝐾𝑌

𝐴

𝐾

𝑌

𝐾 Tax 𝑌 𝑁

𝑁

𝑁

𝑊

𝑊

𝑨

𝐵

𝑪

𝑁 𝑁

𝑁

𝑁𝑌

5

𝐷

𝑁𝑌𝐷

𝑁𝐷 𝑁

The substitution effect corresponds to the movement from

to

labor demanded quantities, is equivalent to the movement from

which, in terms of to

. This effect

reflects how firms substitute capital for labor when face a payroll tax, for a given level of output (see top panel). This occurs because payroll taxes make the labor factor becomes relatively more expensive than capital. Nevertheless, the extent of the substitution effect depends on whether firms are labor or capital intensive. Capital intensive firms could more easily substitute capital for labor. In turn, the output effect is represented by the movement from

to , or from

to

.

It represents the fall in labor demand due to output reduction holding production technology constant (see top panel). Note that output reduces because increases in labor costs lead to higher output prices and therefore to lower sales. The idea behind this effect is that the real budget constraint is lower. This implies that the firm can hire less capital and labor units with the same monetary unit. An important issue in this analysis is that the short run response of labor demand to a payroll tax change is fully captured by the output effect, whereas substitution effect adds to the output effect in the long run the so that the total effect becomes the relevant measure of the impact of payroll taxation. This study empirically quantifies the total and the substitution effects. As a final point, the graphic analysis on figure 1 shows that the points A and B correspond to long-run labor demand curves pre-tax and post-tax, respectively (see bottom panel).While point C corresponds to a labor demand conditional on output. Therefore, the empirical estimation of unconditional labor demand might allow to capture the total effect of payroll taxation (distance AB, figure 1) and the empirical estimation of a conditional labor demand might capture the substitution effect (distance AC, figure 1). 1.2.

Empirical Implementation

In this section, I perform the derivation and the interpretation of the unconditional or capital constrained labor demand, equation (1) and the labor demand conditional on output, equation (2). )

6

( )

)

)

where,

denotes aggregate employment,

capital stock,

denotes the output,

the real wage,

a trend,

is the aggregate

is the payroll tax rate and

is a white

noise error term. Before explaining in detail how the equations are obtained and giving a plausible interpretation to their coefficients, let me introduce two general aspects about these equations. First, it is important to highlight that labor demand equations (1) and (2) are a version extended of the models presented by Karanassou et al. (2007) and Antràs (2004), respectively. In particular, I include explicitly the following considerations in both models. (i) The government sets a payroll tax rate which must be paid out by firms. This tax is assumed as ad valorem and proportional to wage. (ii) The technological change increases at a constant growth rate. Second, stand out that the extension made for both equations as well as the interpretation of the coefficients

and

make up one of the contributions of this

paper, since they become an original methodology to empirically approach the total and the substitution effect to payroll taxation on labor demand. 1.2.1.

Background for Equation (1)

In their analysis of the macroeconomics of labor market, Karanassou et al. (2007) depart from assuming a competitive labor market containing a fixed number

of identical

firms with symmetric production and cost conditions, and monopoly power in the product market. The i’th firm has a Cobb Douglas production function: ̅

, where

is output supplied,

is employment, ̅

is capital stock,

) is a parameter accounting for relative influence of capital and employment, and

is the technological change where is a growth rate.

As each firm chooses its employment at the profit maximizing level (for a given capital stock) and government sets a payroll tax rate

which must be paid out by the employer.

Then, the following aggregate labor demand can be obtained by solving the first order condition and aggregating across the firms. 7

[ where

(

)

)]

| |

[

)

)]

,

is the price elasticity of product demand,

employment and

is aggregate

is aggregate capital. ) to

Taking natural logarithms, introducing a white noise error term capture supply and demand shocks, and making the following rearrangements: [

(

)]

)

)

)

,

)

)

)

; I obtain the

empirical equation (1). 1.2.2.

Background for Equation (2)

In Antràs (2004), an aggregate labor demand is derived from a profit maximization problem where the real output

of the economy is described by the following

production function CES: [ ( where

)

)

is aggregate capital stock;

)

]

,

aggregate employment,

substitution between capital and labor,

the elasticity of

and

represent the

technological change of capital and labor which grow at constant rates

and

and

is a distribution parameter. Assuming that government sets a payroll tax rate . Solving and manipulating the two first-order conditions yield: ) [

)

]

[

Taking natural logarithms, adding the noise error follows

[

)

)

];

;

)] , and rearranging the terms as ;

)

.;

;I

obtain the empirical equation (2). Although, both equations can be extended by adding price controls such as the price of capital; I do not add such extra price controls to avoid measurement problems. For example, Clark and Freeman (1980) showed that labor demand elasticities tend to be biased upwards when the price of other factors are considered in this type of equations. Additionally, this has been the route followed by recent labor demand empirical studies (i.e Hijzen and Swain (2010), Sala and Trivín (2012)). 8

1.2.3.

Interpreting the coefficients

The coefficients

in equation (1) and

this paper. The coefficient

in equation (2) are the relevant parameters in

captures the total effect of payroll taxes on the labor

demand (given a wage level). A negative sign of with taxes, unless

indicates that the labor demand falls

in which case the employers bases their demand decisions on

real wages independently of the payroll tax. Then, the tax causes no fall on labor demand. The coefficient

measures the substitution effect. Thus a negative sign of

shows how firms substitute capital for labor when firms face a payroll tax increase. The rest of coefficients have a standard interpretation which does not change through extension of the models. The coefficients to real wage, whereas labor

is the labor demand elasticity with respect

is the standard elasticity of substitution between capital and

). Regarding coefficients

and

they represent the labor demand elasticity

with respect to capital and output, their signs are positive. Finally,

and

reflect the

influence of technological change on labor demand. Note that given the a-priori modeling assumptions, there are two key equalities: first,

, this means that

the total effect is determined by the elasticity of the labor demand with respect to the real wage. Second,

, the substitution effect is determined by the standard

elasticity of substitution between capital and labor

).

1.3. The Shifting of the Payroll Tax A key issue in the labor demand analysis performed so far is the fact that employers are assumed to equate the gross marginal labor costs with the marginal revenue of producing an extra unit of output, where the gross labor cost consists of the basic nominal labor wage

) along with the employer’s payroll tax contribution

Thus, it can be expressed as

). The crucial issue is whether

).

is assumed as

exogenous (i.e. constant). The standard incidence analysis sets that when a payroll tax is imposed to the firms, the employers can shift their tax burden to workers in the way of lower nominal wages or even as higher output prices. Hence, tax shifting affects the gross wage (nominal and real) through the net wage or prices. By so doing it induces reactions on both sides of the labor market. So, the net effect of a payroll tax on employment is uncertain and determined by two opposing forces: first, a reduction of employment measured by the 9

contraction of the labor demand curve, given a wage level. And second, an increase of the level of employment due to the reduction of net wages which might offset the negative effect on labor demand. As I will focus in compute the negative effect on labor demand. I must, therefore, deal with the endogeneity of wages. 2. DATA AND EMPIRICAL ISSUES The functional form of the equations to be estimated is based on the derived demand functions (1) and (2). I extend them in the following directions. First, due to the relevance of adjustment costs in labor demand decisions, I also add lags of explanatory variables. This enables to perform a dynamic analysis of the labor demand. In other words, it allows to shed light on the shape and speed of the convergence processes towards the optimal levels of labor demand, and also to analyze the short-run and long-run effects generated by payroll taxation. Second, I control for the degree of economic openness as the trade liberalization of 1990s might have affected the labor demand in the Colombian manufacturing sector. Third, I take the natural logarithm of all variables including the payroll tax rate. In particular I replace

) by

, so that, all the coefficients can be interpreted as

elasticities. This transformation facilitates the comparing of the coefficients in equations (1) and (2). Finally, as I work with a two-dimensional panel data, I also add fixed effects which allow me to control for unobserved heterogeneity among sectors. Therefore the equations (1) and (2) will be estimated as an autoregressive distributed lag (ARDL) model that takes the following general form: ∑

where the subscripts

∑

and

,

denote sector and time index, respectively,

represent the dynamic structure of the model, sectorial cross-section intercept,

)

(3), and

is the dependent variable,

is a

is the inertial (or persistence) coefficient,

is a

10

vector of explanatory and control variables, where their influence on dependent variables, and 2.1.

is a set of parameters that reflects

is the error term.

Data: Description, Sources and Treatment

The definition of variables and its sources is provided in Table 1. The main source of my database is the Annual Survey of Manufacturing (EAM by its acronym in Spanish) which is supplied by the National Administrative Department of Statistics (DANE by its acronym in Spanish). From this survey I take data on paid employment, wages, output and net capital stock. This data is available from 1974 to 2009 and is disaggregated by 19 sectors according to the International Standard Industrial Classification (ISIC) revision 3.AC. Table 1. Definitions of variables

⁄

Variables

Sources

Real output Net real capital stock

EAM EAM

Formal employment

EAM

Average real wage

EAM

Statutory parafiscal contributions rate Statutory social security payroll tax rate

LEGIS LEGIS

Openness (

DANE

)

Linear time-trend

Sub-indices

Constructed

Note: All nominal variables are deflated with price index of manufacturing (base: June 1999)

It is important to provide some details about these variables. First, average real wages are calculated as the real wage bill over total paid employment. Second, I use the value of real fixed assets1 as a proxy for net capital stock. To test if this is a good proxy, I checked that the series normalized for the variation of real fixed assets and real net investments were correlated and quantitatively close. Third, I use as output the real value added. Regarding payroll tax rates I have available two types of data. The first one is the nonwage costs obtained from EAM and the second one is the statutory payroll tax taken from LEGIS. In the first case, I could calculate payroll tax rate as total non-wage costs over wage bill. However, non-wage costs include concepts such as: severance payment, 1

Fixed assets include office equipment, transport equipment, industrial equipment, buildings and structures, constructions in progress and lands.

11

settlement, paid vacations, and other kinds of payment, which by definition cannot be considered strictly as payroll taxes. The problem is that it is not possible to break down each of these costs and, as a consequence, a correct payroll tax rate per sector cannot be calculated. In this context, to identify the individual effect of payroll taxes is not possible, which is disappointing given the purpose of this paper. The second option is the statutory tax rate. This data is uniform for all sectors and has two main advantages. First, it avoids econometric problems by using an effective tax rate. In particular, the simultaneity in the determination of wages and payroll tax rates and the spurious variability of payroll tax rate are not an issue. Second, I can break down the effect of payroll tax by type of tax: social security (health and pension) and parafiscal contributions. Likewise, using statutory payroll tax has a weakness: the little variability of data might lead to multicollinearity problems. On the other hand, to measure the degree of economic openness I take data from DANE. Finally, all nominal variables are deflated with the price index of manufacturing (base: June 1999). In summary, I work with a panel model with a cross-section dimension of sectors and a time dimension of 2.2.

years covering the period 1974-2009.

Colombian Payroll Taxation and Manufacturing Industry

In Colombia when a company hires a worker, according to labor law, it assumes the following mandatory payments besides of wages: a social security contributions (health insurance and pension schemes), parafiscal contributions2, paid leaves, a severance payment, and occupational hazards, among others. During the period 1970-1990, the Colombian Social Security System was characterized by institutional disintegration and low coverage. First, there were different institutions (private and public) providing separately the following services: pension schemes, savings, health insurances, occupational hazards and social solidarity services. And second, the employees formally linked to the labor market were unique members of the system and in some cases their family group. These features were identified as its main 2

The parafiscal contributions are a type of payroll tax which is only paid by employers. These contributions have been used to finance the Family Compensation System, the National Service of Learning (SENA by its acronym in Spanish) and the Colombian Family Welfare Institute (ICBF by its acronym in Spanish).

12

weaknesses due to the system could not ensure the welfare of the population. Hence, they carried out structural reform in 1993. A new social security system was conceived; it brought together all services: pension schemes, savings, health insurances, occupational hazards and social solidarity services. As a result of the reform, the social security system was taken as a compulsory public service, and the payroll taxes were increased, including social security contributions (the part paid by the employees and the employers) and the parafiscal contributions. Nevertheless, the main source of the increases arose from the increases in the social security contributions paid by employers. Bear in mind non- wages costs reached a rate of 61% over wages by 2009 (53% paid by employers) where the social security payroll tax rate paid by employers along to the parafiscal contributions were in total 29.5%. Table 2 summarizes the evolution of the social security payroll tax rate paid by employers and the parafiscal contributions. The rest of information, except economic openness rate, corresponds to aggregate averages of Colombian manufacturing industry for the relevant period of analysis. The average openness rates are given for the overall economy. Table 2. Economic scenario in the Colombian manufacturing industry. )

Years 1974-1979

6.23

-2.74

3.32

2.25

3.97

7.44

6.67

18.44

1980-1989

3.55

3.98

0.90

-0.05

3.59

8.33

7.90

22.04

1990-1999

1.87

17.21

5.12

-2.00

3.87

15.52

9.00

28.78

2000-2009

5.28

2.33

1.20

1.15

4.12

22.07

9.00

35.65

1974-2009 4.23 5.19 2.64 0.34 3.89 13.34 8.14 26.22 Notes: is the difference operator and indicates average growth rates. All variables are expressed in percent.

Finally, the Colombian government issued a new law in 2010, which created some exceptions to the payment of parafiscal contributions. Furthermore, in early 2013 with the approval of the tax reform eliminated these contributions. The main aim of Colombian government is to boost employment.

13

On the other hand, as shown table 2 real growth of the Colombian manufacturing industry was around 4.2% on average since the mid-1970s until 2009. The main source of growth in the last three decades was capital investment which was benefited from the trade and financial openness process started in the late 1980s. Note, for example, in the 1990s there was a high capital investment growth (17.2%) and at the same time there was a high external trade flow (28.78%). Although, the manufacturing industry grew; the net formal job creation was low 0.3%. Even more, the 1980s and 1990s were mainly characterized by formal job cuts (-0.1% and -2.0%) and lower output growth (3.6% and 1.9%). This situation could reflect the slowdown of the Colombian economy in the early 1980s and the second half of 1990s. The first one was due to the debt crisis and the second one due to the financial crisis. Nevertheless, part of the employment decrease could also be explained as a result of the payroll tax increases described at the beginning of this section. As a final point, in 2000-2009, the manufacturing industry was characterized by the deepening economic openness rate (35.6%), the improvement on labor productivity (4.1%), and the growth of social security costs (from15.5% to 22.1%). 2.3.

Panel Unit Root Tests and Panel Cointegration Tests

As I have a dynamic panel, I must ensure there is a long run equilibrium relationship among the variables. In other words, to determinate if there might be causal relationships. That implies testing that all variables will be stationary I (0). If two or more variables are non-stationary but integrated of order 1 (1); I must check their cointegration to avoid the peril of having spurious relationships. Table 3. Panel Unit Root Test ⁄

Statistic p-value

27.74

48.60

31.37

47.59

(0.865)

(0.116)

(0.768)

(0.137)

0.13

0.09

0.06

0.11

Notes: All variables are expressed in logs. The 5% of critical value of KPSS test is 0.146 using intercept and trend, and testing in first differences.

In order to check the order of integration of the variables, I perform a series of unit root test. These tests are different depending on the type of variables to be dealt with. In 14

particular, I use the KPSS unit root test 3 for the variables that are common across sectors; while, I used the AD Fisher unit root test for the variables that are sectorspecific. I used the AD Fisher unit root test, since, as Maddala and Wu (1999) point out, this test is simple and straightforward to use and is a better test than the Levin and Lin (1993) and the Im, Pesaran and Shin (2003) tests. The test has two attractive characteristics: first, it does not restrict the autoregressive parameter to be homogeneous across sector under the alternative of stationary. And second, the choice of the lag length and of the inclusion of a time trend in the individual ADF regressions can be determined separately for each sector. Table 3 shows the tests. The test statistic and the p-value are provided for the AD Fisher test, while for KPSS test only the test statistic is provided. The results indicate that all variables follow a unit root process. Therefore, I test if there are cointegrating relations among variables. Specifically, I use the Johansen Fisher panel cointegration test which is shown in Table 4. Table 4. Panel Cointegration Equation ( Hypothesized No. of CE(s)

None

Fisher Statistic (from trace test) 355.04

)

0.00

Fisher Statistic (from maxeigen test) 219.72

p-value

Equation (

0.00

Fisher Statistic (from trace test) 278.62

p-value.

p-value 0.00

)

Fisher Statistic p-value. (from maxeigen test) 225.83 0.00

At most 1

165.94

0.00

107.38

0.00

105.82

0.00

83.51

0.00

At most 2

82.63

0.00

65.06

0.00

45.73

0.18

39.75

0.39

At most 3

38.38

0.45

33.94

0.66

22.27

0.98

17.42

0.99

At most 4

22.95

0.97

17.92

0.99

17.72

0.99

15.21

0.99

At most 5

23.68

0.97

23.68

0.98

18.37

0.99

18.37

0.99

Note: All variables are expressed in logs. Test computed using intercept and trend

These results indicate that the existence of a single cointegrating vector cannot be rejected. Actually, the first equation has three cointegrating relations and the second one has two. This means there is at least a long run relationship between the variables included in each specification for formal employment.

3

See Kwiatkowski, Phillips, Schmidt and Shin (1992) for details.

15

2.4.

Estimation Method

Given the panel structure of my database, I will estimate the respective versions of specification (3) as a dynamic one-way fixed effect model (FE). However, these estimates are likely to be biased due to two main problems. First, the potential endogeneity caused by the introduction of a lagged dependent variable. And second, the well-known simultaneity of wages and capital (or output). Regarding endogeneity caused by the introduction of a lagged dependent variable, Nickell (1981) points out that when

is small and

is large, specifically when

,

the within or fix effect estimator will be biased and inconsistent even if there is no serial correlation of the error term. Nevertheless, Álvarez and Arellano (2003) show that when and

grows fast enough with respect to

, the FE estimator will be

consistent. This is relevant for me because I work with a large , large , and but

does not grow fast enough with respect to

;

. Therefore, I do not expect FE

estimator will be consistent. To deal with this problem I will estimate the respective equations by the Generalized Method of Moments (GMM) or Arellano and Bond (1991) estimator. The advantages of using GMM are that I can obtain consistent estimates and at the same time I can take into account the endogeneity of wages and capital (or output). Thus, as a first step and for comparison purposes, I estimate the formal employment specifications by FE and GMM assuming wages and capital (or output) as exogenous variables, and then I estimate by GMM endogenizing these variables. Additionally, with the aim to raise the efficiency of the FE estimator, I compute white cross-section standard errors (clustering by period) correcting for the possible presence of cross-section specific heteroskedasticity. I also compute a white covariances matrix whose estimates are robust to arbitrary heteroskedasticity and within cross-section serial correlation. In the case of the GMM estimator, I control for arbitrary heteroskedasticity and within cross-section serial correlation.

16

3. RESULTS As noted, equations (1) and (2) are estimated using a dynamic one-way FE model, GMM one-step (assuming wages and capital or output as exogenous) and GMM onestep (assuming wages and capital or output as endogenous). Comparing estimates by FE and GMM one-step in which wages, capital, output, taxes, and openness are assumed as exogenous; I should not expect large differences in the estimated coefficients, as they turn out to be (see Table 5). Nevertheless, given the potential heterogeneity bias, there are unavoidable differences between both sets of estimates which, in turn, yield significant differences in the long run elasticities, especially in the equation that include output as explanatory variable (see Table 5 and 6). Likewise, if I put together the two estimations made by GMM, I realize that endogenizing wages and capital (or output) does not generate a high variability in the estimated coefficients but it does in the long run elasticities. Therefore, given that these differences among the three estimations might arise from the heterogeneity bias and the simultaneity of some regressors, I take as reference the results of the second estimation by GMM (third column in Table 5) since they combine the characteristics of dynamic panel data estimation and endogeneity control.4 Table 6 presents the crucial elasticities (short run and long run) for my analysis. Recall that the total effect of payroll taxes on the labor demand and the wage elasticity are obtained from equation (1) while the substitution effect of payroll taxes and the standard elasticity of substitution are obtained from equation (2). So qualitatively, the three estimation methods deliver a similar picture but quantitatively they do not. The main difference is in the long run elasticity of substitution estimates. To check the robustness of the GMM estimator from equation (6) (third column in Table 6), I take as reference some estimates of the labor demand in the Colombian manufacturing industry similar to equation (2) and I compare the long run substitution elasticity. For example, Arango and Rojas (2003) and Vivas et al.(1998) found that the substitution elasticity is around -0.7. Likewise, Robert and Sckofias (1997) found that it is-0.4 for skilled workers, and-0.6 for unskilled. 4

I also resorted to other panel data techniques such as Two Stage Least Square (2SLS) and Three Stage Least Square (3SLS) in order to control for the endogeneity of wages and capital (or output). But these did not yield reasonable results.

17

Table 5. Estimated labor demand, 1974-2009. Dependent variable: FE

GMM one-step *

0.24

0.11

(0.67)

(0.76)

-0.08

-0.07

-0.10

-0.08

-0.09

-0.07

(0.00)

(0.00)

(0.00)

(0.03)

(0.00)

(0.03)

0.04

0.04

(0.00)

⁄

(

(0.07)

(0.06)

0.02

0.02

(0.01)

(0.56)

(0.61)

0.17

0.15

0.17

(0.00)

(0.00)

(0.00)

-0.15

-0.11

-0.12

-0.06

-0.12

-0.05

(0.00)

(0.00)

(0.01)

(0.02)

(0.00)

(0.00)

0.13

(0.00)

ADF Fisher Test

0.03

0.04

0.14

⁄ )

0.16

(0.00)

(0.00)

-0.09

-0.06

-0.11

-0.09

-0.11

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

-0.19

-0.12

-0.22

-0.13

-0.22

-0.13

(0.00)

(0.03)

(0.00)

(0.02)

(0.01)

(0.00)

0.01

0.01

0.01

0.01

0.01

0.01

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

-0.17

-0.06

-0.15

-0.07

-0.15

-0.07

(0.01)

(0.14)

(0.01)

(0.09)

(0.01)

(0.12)

0.35

0.27

0.35

0.30

0.36

0.32

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

0.17

0.16

0.17

0.16

0.17

0.16

(0.00)

(0.02)

(0.03)

(0.02)

(0.03)

646 116.56

646 134.74

627 407.04

627 405.25

608 395.23

608 392.08

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

498.35

484.33

486.30

477.00

(0.30)

(0.41)

(0.20) (0.35) Notes: All variables are expressed in logs. p-values in brackets. FE: Fixed Effects. GMM one-step: Generalized Method of Moments or Arellano and Bond (1991) estimator. * All regressors, except , are assumed as exogenous. in

model

with

K:

.In model with Y: **

-0.10

(0.00)

Sargan Test

Instruments

GMM one-step **

⁄

⁄

⁄

⁄

are assumed as endogenous. ⁄

Instruments in model with K:

⁄ ⁄

In model with Y ADF Fisher Test computed using intercept and trend.

18

Given these results and the preference for the second econometric GMM estimate, I conclude that, first, there is a high level of persistence in employment decisions and, in turn, a low sensitivity of the demand for labor to wage shifts and payroll tax shifts in the short run. Second, the total effect (in the long-run) of social security payroll taxes and parafiscal contributions are respectively -1.2 and -2.3. That is, a 1% increase in the social security payroll tax rate will cause a 1.2% reduction in the demand for labor, while a 1% increase in the parafiscal contributions rate will cause a 2.3% reduction in the demand for labor. This means that demand for labor in the Colombian manufacturing industry is approximately twice as sensitive to a 1% shift in parafiscal contributions rates than to a 1% shift in social security payroll tax rate. Table6. Estimated labor demand elasticities 1974-2009. FE

Elasticity

Parafiscal contribution Social Security (health and pension) Real Wage Substitution

GMM one-step*

GMM one-step *

SR

-0.19

-0.12

-0.22

-0.13

-0.22

-0.13

LR

-2.37

-1.69

-2.18

-1.65

-2.33

-1.77

SR

-0.09

-0.06

-0.11

-0.09

-0.11

-0.10

LR SR LR

-1.18 -0.15 -1.89

-0.86

-1.14 -0.12 -1.18

-1.13

-1.21 -0.12 -1.31

-1.43

SR

-0.11

-0.06

-0.05

LR

-1.53

-0.72

-0.67

Notes: SR- Short run- and LR -Long run-

Third, the long-run substitution effect resulting from a payroll tax increase are -1.4 for the social security payroll tax rate and -1.8 for the parafiscal contributions rate. Nevertheless, as shown in the empirical derivation section, they are expected to be equal. Therefore, I check this hypothesis using a Wald-test and I cannot reject the equality in the short run and long run. This means substitution effect is the same for the social security and the parafiscal contributions. Quantitatively, they are approximately two point five times elasticity of substitution. As a final point, I also tested the equality of the labor demand elasticities with respect to real wages and with respect to the social security payroll tax (for the short and long-

19

run). I could not reject the null hypothesis of equality which means that firms respond in the same way to a 1% increase in wages or to a 1% increase in social security payroll tax rate. 4. CONCLUDING REMARKS In this study, I have analyzed the impact of payroll taxes on the labor demand to assess the effects of institutional changes of the Colombian labor market on the employment decisions taken in manufacturing industry (up to 2009). Through the standard profit maximization problem, I have obtained two empirical versions of labor demands, which incorporate the payroll tax rate. These labor demand equations have been estimated as autoregressive distributed lag models by GMM. The results obtained suggest that the institutional changes brought by the government in last decades (up to 2009) affected negatively the labor demand of the Colombian manufacturing industry. And, even more, they have two main implications in terms of the 2010 measures of economic policy. First, assuming that there is not full shifting as Kugler and Kugler (2009) pointed out, the reductions in parafiscal contributions rate issued by the Colombian government in 2010 might boost net job creation in the manufacturing industry, especially in the long run due to slow labor demand adjustment process. Second, the parafiscal contributions rate might be a fiscal instrument more powerful to create new jobs due to the fact that the demand for formal labor is approximately twice as sensitive to shifts in the parafiscal contributions rate than to shifts in the social security payroll tax rate. This result validates the policy implemented by the Colombian government. To further validate this conclusion, it is necessary that future studies estimate a dynamic multi-equation labor market system, which allows to simultaneously capture the effect of payroll taxes on wages and employment. In addition, adding more sectors of the economy could yield a more accurate picture on the effect on employment and wages. Controlling the type of employment (for example, by permanent and temporal personal or by production and administrative personal) could improve the estimates as there might be differentials of wages which might make firms react differently to changes in payroll taxes. Finally, two major challenges are: first, to design an approach to capture simultaneously the effects of taxation in the formal and informal sectors (for example through a CGE model), because Colombia is characterized by high and persistent levels 20

of informal activity (around 50%). Second, to design a model that also captures the effects of these policies on government revenue.

ACKNOWLEDGMENTS I would like to thank Hèctor Sala for his very useful comments and suggestions. I would also like to acknowledge financial support from the COLCIENCIAS -Departamento Administrativo de Ciencia Tecnología e Innovación- Colombia.

REFERENCES Álvarez, J. and M. Arellano (2003) “The Times Series and Cross-Section asymptotic of dynamic panel data estimators”, Econometrica, 71(4), 1121-159. Antràs, P. (2004) “Is the U.S. Aggregate Production Function Cobb-Douglas? New Estimates of the Elasticity of Substitution”, Contributions to Macroeconomics, 4(1), Article 4. Arango, C. and A. Rojas (2003) “Demanda laboral en el sector manufacturero colombiano: 1977-1999”, Borradores de economía, 247. Arellano, M. and S. Bond (1991) “Some Test of Specification for Panel Data: Montercarlo Evidence and an Application to Employment Equations”, Review of Economics Studies, 58, 277-97. Bauer, T. and R. Riphahn (2002) “Employment Effects of Payroll Taxes, an Empirical Test for Germany”, Applied Economics, 34, 865-76. Bennmarker, B.; E. Mellander, and B. ̈ cker (2009) “Do regional payroll tax reductions boost employment?”, Journal of Labour Economics, 16(1), 480-89. Botero, J. (2011) “Impuestos al capital y al trabajo en Colombia: un análisis mediante equilibrio general computable”, The Review of Economics and Statistics, 62(4), 509-520. Clark,K.B and Freeman, R.B (1980) “How elastic is the demand for labor?”, ecos de Economía, 15(33), 49-69. Cruces, G.; Se. Galiani and S. Kidyba (2010) “Payroll taxes, wages and employment: Identification through policy changes”, Labour Economics, 17,743-49. Gruber, J. (1997) “The Incidence of Payroll Taxation: Evidence from Chile”, Journal of Labor Economics, 15(3), S72-S101. Hernández, G. (2011) “Impuestos parafiscales y Mercado laboral: Un análisis de equilibrio general computable?”, Archivos de Economía, (378), 480-89.

21

Hijzen, Alexander and Paul Sawim (2010) “Offshoring, labour market institutions and elasticity of labour demand”, European Economic Review, 54(8),1016-34. Karanassou, M.; H. Sala and D.J. Snower (2007) “The macroeconomics of labor market: three fundamental views”, Portuguese Economic Journal, 6(3), 1-20. Kwiatkowski, D.; P. Phillips; P. Smith; Y; Shin (1992) “Testing the null hypothesis of stationary against the alternative of unit root: how sure are we that economic time series are not stationary?”, Journal of Econometrics, 54, 159-78. Kugler, A. and M. Kugler (2009) “The labor market effects of payroll taxes in a middle-income country: evidence from Colombia”, Economic Development and Cultural Change, 57(2), 33558. Levin, A. and C. Lin (1993) “Unit Root Tests in panel data: New Results” Department of Economics, University California at San Diego, Discussion paper No.93-56. Madala and Wu (1999) “A Comparative study of unit root tests with panel data and new simple test”, Oxford Bulletin of Economics and Statistics, 61,631-52. Nickell, S. (1981) “Biases in dynamic models with fixed effects”, Econometrica, 49(6), 141726. Nickell, S. (2003) “Employment and taxes” CESifo Working Paper 1109. Pesaran M. and Y.Shin (2003) “Testing for unit roots in heterogeneous panels”, Journal of Econometrics, vol 115, 53-74. Roberts, M. and E. Skoufias (1997). “The long-run demand for skilled and unskilled labor in Colombian manufacturing plants”, The Review of Economics and Statistics, LXXIX (2), 330-34. Sala, Hector and Pedro Trivín (2012) “Structural changes in the Spanish labour demand: Does Rodrik’s conjecture hold?”, article presented at the IV Congreso Nacional sobre mercado de trabajo y relaciones laborales (Universidad de Valladolid, April 2012). Vivas A., S. Farne and D. Urbano (1998) “Estimaciones de funciones de demanda de trabajo dinámicas para la economía colombiana, 1980-1996”, Archivos de Macroeconomía, (092), 1-48.

22

APPENDIX Table A. Estimated labor demand, 1974-2009. Dependent variable:

⁄

(

GMM one-step *

GMM one-step **

-0.08

-0.07

(0.03)

(0.03)

0.17

0.18

(0.00)

(0.00)

-0.15

-0.16

(0.00)

(0.00)

-0.06

-0.05

(0.02)

(0.00)

-0.09

-0.10

⁄ )

ADF Fisher Test Sargan Test

(0.00)

(0.00)

-0.13

-0.13

(0.02)

(0.00)

0.01

0.01

(0.00)

(0.00)

-0.07

-0.07

(0.09)

(0.12)

0.30

0.32

(0.00)

(0.00)

0.16

0.16

(0.03)

(0.03)

627 405.25

608 392.08

(0.00)

(0.00)

484.33

477.00

(0.35) Notes: All variables are expressed in logs. P-values in brackets. FE: Fixed Effects. GMM one-step: Generalized Method of Moments or Arellano and Bond (1991) estimator. * All repressors, expect , are assumed as exogenous. ⁄

Instruments in model with K: ⁄

. In model with Y: **

⁄ ⁄

are assumed as endogenous. ⁄

Instruments in model with K:

⁄ ⁄

In model with Y ADF Fisher Test computed using intercept and trend.

23

(0.41)