Incidence of corporate tax credit on profits, wages and employment: evidence from a French reform 1

Transcription

1 Incidence of cororate tax credit on rofits, wages and emloyment: evidence from a French reform 1 First draft: February 2017 PRELIMINARY Clément Carbonnier Clément Malgouyres Loriane Py Camille Urvoy Université de Cergy- Banque de France Banque de France Sciences Po Pontoise, THEMA Abstract This aer exloits a far-reaching French reform as well as a very rich set of administrative data to evaluate the imact of a cororate tax credit aimed at reducing labor costs on several outcomes: emloyment, rofit and wages. The e ects of the Cometitiveness and Emloyment Tax Credit (CETC), a refundable tax credit based on the wagebill, introduced in France in 2013, are estimated thanks to double (and trile) di erence methodologies, instrumented by the intensity of the intention to treat, thanks to data at the firm and individual levels on the eriod Our results show that this relatively large tax break - about 17 billion euros er year - does not succeed in boosting emloyment in the first two years after being set. However they suggest that firms used the CETC to restore their margins. Moreover wages have increased significantly in more intensively treated firms, articularly those of white-collar emloyees. These results cast doubts regarding the e ectiveness of such tax credits to boost emloyment. More imortantly, they also rovide new quasi-exerimental evidence regarding rent sharing in the labor market. Keywords: tax credit, incidence, rent sharing; JEL codes: D22, H25, H32. 1 We would like to thank institutions that built the databases (Insee, DGFiP, ACOSS, Douanes) and made them available for research on the CASD, and articularly their agents who diligently answered our questions about their content. We also thank Marc Ferracci, Sara Guillou, Yannick L Horty, Etienne Lehmann, Benoit Ourliac, Bruno Palier, Etienne Wasmer for their comments in ad hoc meeting organized by France Strategy or in academic seminars. This work is suorted by ublic grants overseen by the French National Research Agency (ANR) as art of the Investissements d avenir rogram within the frameworks of the Centre d accès sécurisé aux données -CASD(ANR- 10-EQPX-17) and the LIEPP center of excellence (ANR-11-LABX-0091, ANR-11-IDEX ). Researchers from the Banque de France articiated to this work, the views exressed in this aer are their own and do not necessarily reflect those of the Banque de France or the Eurosystem. 1

2 1 Introduction In the context of a globalized economy in which many develoed countries face sluggish economic growth and low emloyment rates, governments have sought to increase firm cometitiveness and boost economic activity by imlementing olicies aiming at reducing labor costs. In France, between 1993 and 2004 a series of direct ayroll tax cuts have been imlemented, targeting the bottom of the wage distribution, artly in order to o set the imact of the minimum wage on labor cost (Bunnel and L Horty, 2012). In 2013, this set of olicies was comlemented by a large cororate income tax (CIT) credit whose amount is roortional to the wage bill. This olicy, called the Cometitiveness and Emloyment Tax Credit (CETC) was conceived as a mere continuation of the re-existing ayroll tax cuts. It is well ossible however that firms resond to wage bill-based CIT credits di erently than to ayroll tax cuts. They could for instance resond to a CIT credit by increasing their net rofit or by sharing the benefit with their emloyees through wage increases. These e ects are likely to be all the more di erent that labor and good markets are not cometitive and that there exists rent sharing between emloyers and emloyees within the firm. In this aer, we evaluate this far-reaching reform, taking advantage of a very rich set of administrative data, in order to understand how firms react in resonse to tax cuts and labor cost shocks. The cometitiveness and emloyment tax credit (CETC) was introduced in France in January 1 st This scheme consists of a CIT credit equal to 4% of the eligible wagebill in 2013 and 6% of that eligible wagebill in Crucial for identification strategy, the eligible wagebill corresonds to the sum of gross wages for emloyees aid less than two and half-time the hourly minimum wage. In other words, the wages of salaried just above this threshold are not eligible to the CETC and this discontinuity generates imortant variation in the intensity of treatment between firms (even for firms with very similar wage structure) that we are going to exloit. Our emirical analysis relies on a very rich dataset at the firm and individual levels. Very recise data on firms wage structure are found in annual social declarations database (DADS) at the level of each job (one observation er osition for each comany and one observation er job for each emloyee), built by Insee (French statistic agency). This database, exhaustive at the level of salaried jobs, also informs about the tye of osition held both in contractual terms and in terms of tasks. General information on the cororate structure of roduction and rofits come from FARE database and consist in cororate tax return which are collected by the General Directorate of Public Finance - DGFiP - matched with survey data build by Insee. This database is exhaustive at the level of the comanies. DGFiP also secifically builds MVC file informing the comany s entitlements to CETC and its imutations, deferrals or reimbursements. Our emirical strategy consists in estimating the e ects of the CETC by comaring the evolution of emloyment, wage and rofits (in levels or in growth rates) for comanies more or less beneficiaries of the CETC due to the existence of this threshold in wagebill eligibility (double and trile di erences estimations). To ensure the exogeneity of the treatment the magnitude of CETC received as a share of labor cost we also instrument the CETC actually received by the CETC that firms 2

3 could get, according to the characteristics of their roduction structure the years receding the introduction of the CETC. To check for the robustness of this identification strategy, fixed e ects and controls are introduced to ensure the common trend assumtion between the treated and control firms. Finally, the otential existence of remaining diverging trends is also directly tested through lacebo regressions. Our results suggest that CETC did not succeed in boosting emloyment in the two first years (2013 and 2014). However the results suggest that firms used the CETC to restore their margins as the CETC has a ositive and significant imact on three rofit indicators (which tends to increase over time). Moreover the estimates show that the cororate income tax credit has been artially shifting on to wages. Di erencing the e ect on wages er tye of worker, it aears that white collar are the main indirect beneficiaries of the scheme. These results bring new evidence on the existence of rent sharing between caital and labor and in favour of insiders (esecially for whitecollar workers). An imortant literature has been dedicated to the evaluation of olicies aiming at reducing labor cost. Many studies focus on social contribution cuts. Bohm and Lind (1993) and Bennmarker et al. (2009) for Sweden, and Korkeaäki and Uusitalo (2009) for Finland have develoed estimates based on geograhical di erences in rates (taking advantage of regional reforms in social security contributions), and conclude to the absence of e ect on emloyment. Social contribution cuts in France aear to have had more favorable e ects on emloyment (Créon & Deslatz 2001, Kramarz & Philion 2001, Chéron et al., 2008). A otential reason could be that they are more targeted on low wages. However, Huttunen et al. (2013) use double di erence estimation method (by age grou) to assess the imact of social contribution cuts targeting low-wage workers in Finland: they found no imact at the extensive margin and only a very limited imact at the intensive margin. One reason of the imortance of low-wage targeting for exlaining the imact of social contribution cuts comes from wage incidence. Gruber (1994), Anderson and Meyer (1997, 2000), and Murhy (2007) demonstrate, through natural exeriments in the United States, that the share of social contributions actually aid by emloyers is inversely roortional to the level of wages. Moreover, taxation incidence on wages is not limited to social contributions: three recent emirical analyses (Arulamalam et al., Dwenger et al., 2011, Liu & Altshuler 2013) found that about half of the CIT rate cuts were assed on to emloyees through wage increases. Our contribution to the literature is threefold. First, we evaluate the imact of the CETC jointly on three di erent outcomes (emloyment, wage and rofits) and thus rovide new quasiexerimental evidence on the incidence of cororate income tax credit. Moreover using detailed information on individual workers, we are able to document the existence and magnitude of rent sharing in the labor market and more imortantly to which categories of emloyees it is most relevant (job stayers versus new hires, white-collars or blue-collars). Finally, our estimates, based on a large and still ongoing CIT credit rogram, oint to weak emloyment e ects, which casts new doubts on the relative e ectiveness of such incentives. Given the oularity of cuts in cororate 3

4 income tax as a way to boost economic activity, our results are informative to the current olicy debate. Our current results are based on a linear di erence-in-di erence. However, we are currently imlementing a matching estimator that allows us to isolate the variation in treatment intensity that comes from the discontinuity in the eligibility at the 2.5 minimum wage threshold. The matching estimator consists roughly in matching firms rior to treatment on several oint of their cumulative density function ensuring that the remaining variation in treatment intensity stems from di erence in wage structure around the 2.5 minimum wage threshold that we consider as good as random. The re-treatment eriod allows us to erform several auxiliary exercises (lacebo tests). The rest of the aer is organized as follows. The databases are resented in section 2 and the identification strategy is detailed in section 3. The results of the estimates are resented and discussed in section 4. Section 5 concludes and ut into ersective the imacts of CETC onto the di erent outut in order to draw the global icture of the CETC aftermaths. 2 Data This emirical analysis is based on three administrative databases, built from firms returns to the tax agency (DGFiP, the French General Directorate for Public Finance) and to the institution resonsible to collect social contribution (ACOSS). DGFiP has comuted, since the reform, a database secifically informing about the amount and use of the CETC at the firm-level (MVC database). They also rovide in association with Insee (the French statistical agency) a database on firms accounting (FARE database). ACOSS rovides in association with Insee a database on workers and wages at the contract level (DADS database). We got access to these databases for the years 2010 to MVC database DGFiP secifically built the MVC files informing the firms initializations of CETC rights. This database, which began to be created for the 2013 vintage, contains five variables for all firms likely to benefit from the CICE - i.e. more than 800,000 observations. These five variables are: initialization, the amount of tax credit to which the comany is entitled, initialized on its tax returns; increase, uward adjustments given the evolution of the comany s wage structure; decrease, similar downward adjustments; imutation, the amount of CETC that comanies were able to deduct from their CIT. These variables allow us to understand the CETC distribution. After airing with the other databases (and the loss of some comanies absent from certain bases), the total amount initialized in 2013 is 9.8 billion euros. A large number of comanies benefit from a relatively small amount of CICE, with about EUR for micro-enterrises and EUR for SMEs, whereas the amounts received by large comanies are ten to one hundred times larger: the 288 large comanies 4

5 resent in the base have initiated in 2013 a tax credit aroximately equal to that of the 496,750 micro-enterrises. Initializations, besides being highly variable, reresent relatively small amounts for comanies: it exceeded one ercent of turnover for only one quarter of the comanies in 2013 and less than half in 2014 (which also illustrates the increase in CETC amounts between 2013 and 2014). Moreover, the collection of these amounts remains very sread out over time because of the nature of the tax credit nature of the CETC. The econometric results deend on this since any CETC imacts which would occur through the relaxation of budgetary constraints could only be observed after some years and cannot be estimated in our framework. Note that these mechanisms should not be at stake given that French firms are not budget constraints (Krem and Sevestre 2013). On the contrary, incentive e ects - esecially in terms of emloyment linked to labor costs - are quicker to occur and our econometric framework should therefore be adated to estimate them. 2.2 FARE database General information on the roduction structure of comanies and their benefits is resented in the FARE database of the ESANE system (annual business statistics). It is built by Insee on the basis of the tax data, social declarations and a survey. The urose of the survey questionnaire is to roduce structural business statistics; it should be noted that the questionnaire sent to comanies was amended in This database covers all firms (including firms without emloyees) with the excetion of the financial sector and farms. Our deendent variables regarding the average workforce or the rofitability of the firm come from this dataset. On this last oint, the accounting entry for the CETC is unclear and it is likely that the di erent comanies have accounted for it di erently (deduction of labor costs, oerating subsidies, other oerating income or CIT deduction). Thus, the various measures of rofit may or may not take into account the CETC according to how it is accounted for. We have tried to address this roblem by considering three rofit indicators: EBIT and EBITDA as a roortion of turnover, as well as oerating income as a roortion of oerating costs. We also use variables from FARE as controls: roductivity (value added divided by average workforce) and caital stock (tangible and intangible assets). As a robustness test, we also consider the set of control variables used by Gilles et al. (2016): margin rate (EBITDA / VA), economic rofitability (EBITDA / fixed assets), roductivity (VA / workforce), caital intensity (tangible assets / workforce), share of exorts in turnover, investment rate (tangible investments / VA), debt ratio (borrowings and debts on the share caital, emission remiums, income from oerations, investment subsidies on liabilities and other equity), financial drawdown rate (interest on loans / EBITDA). While all other databases are defined at the firm level (with the SIREN number identifying them), the FARE files comute combinations for some of them, which is called rofiling. Indeed, some major grous have transformed arts of their roduction chain into indeendent legal units, 5

6 while decisions remain at the central level. In order to rovide a better overview of the roductive structure, Insee gathers di erent legal units (with di erent SIREN numbers) into a single entity. For the six historical rofiled comanies, which only aear in the database in their rofiled form, we consider the rofiled comany and similarly rofile the other databases. The hundreds of other rofiled comanies are resent in the database both under their individual SIREN and under their rofiled SIREN. We consider for them only the individual SIREN. 2.3 DADS database The annual social data declaration (DADS) files contain information on each salaried contract in each comany: net and gross wages, working time, socio-rofessional categories, tyes of contracts, sex of the emloyee... There is one observation er contract for each comany and emloyee. Thus, the same emloyee can be found several times in the dataset if she has contracts with several comanies. It is therefore a database to be used from the oint of view of the comanies and not of the emloyees. In addition, it is imortant to know that DADS are resented in the form of regional files and that observations concerning emloyees of an enterrise located in one region but residing in another region are resent in the regional files of the two regions. A first work before starting the analysis therefore consisted in urifying these databases from double accounts. Moreover, for each item, the values of the variables are also given for the revious year. This makes it ossible to construct changes in the variables from one year to the next for each item. Indeed, the identifiers of the contracts are not recognizable from one vintage to the other and it is therefore not ossible to build a anel of contracts. On the other hand, the comany identifiers are the SIRENs, stable over time, and we therefore constitute anels of comanies. Thus, as far as wage increases are concerned, we have oerated in two ways. On the one hand we calculated the average wages er firm each year, and comared them from one year to the next. Since the changes can be due both to changes in the wages themselves or to the structure of emloyment in the comany, we have also calculated the growth hourly wage for each osition resent two following year in the same firm. Then, we aggregated by calculating for each year the average of individual wage growth. Pay data is accurate in that it is at the job level, but relatively imrecise as to what it covers. Gross remuneration includes all remuneration received by the emloyee under her contract of emloyment, before deducting comulsory contributions. For instance, it includes bonuses for end of fixed-term contracts (corresonding to 10% of the amounts received during the contract). This can lead to biases in the observation of hourly wage growth since these bonuses inflate the total gross earnings for the contract year but not the number of hours worked. In order to measure the growth of hourly wages, it is therefore necessary to urge the bases of these observations at the end of the contract. In order to carry out our identification strategy, it is necessary to be able to measure the otential CETC to which an enterrise would have been entitled before the actual reform imlementation, according to its roductive structure. However, this is not ossible on the basis of actual tax data, 6

7 which did not collect such information rior to However, the DADS database allows us to aroach these values thanks to the recision on wage structures comanies. It is indeed ossible to calculate the share of wagebill below 2.5 minimum wage, and to comute a otential CETC. This calculation for the years 2013 and 2014 is very close to the amounts of CETC actually initialized with the tax deartments and resented in the MVC database. Similar wage structure indicators are also built at other thresholds: 1.5, 2, 3 and 3.5 minimum wage. 2.4 Building the merged database We are working on the matching of the three reviously resented databases. After the DADS-MVC matching, we calculate the ratio of the CETC imuted on the basis of the DADS database over the actual CETC initiated in the MVC database. We exclude from the samle the firms whose ratio belongs to the uer ercentile in 2013 or in We then erform the matching with the FARE database. Only the comanies resent in the 3 bases (DADS-FARE-MVC) are ket. We then make two selections. First, we kee only comanies that have at least one full-time equivalent job over the year, so as not to be biased by the emty shells. Then, we constructed the balanced anel database over the eriod There are then slightly fewer than 500,000 firms in the final database used for the estimates. 3 Identification Strategy The aim of the resent evaluation is to use the French CETC reform to estimate the imact of wage costs decreases on firms behavior. Calling Y the deendent variable and C the cost variable (deending on the tyes of behavior studied, we can look at the total roduction costs TC or wagebill only WB only). We intend to measure the elasticity Y,C ln(y )/@ ln(c). This elasticity may result from various economic mechanisms. This may be a form of rent sharing in the case of wage increases resulting from CIT rates decreases (Arulamalam et al. 2012, Liu and Altshuler 2013, Dwenger et al. 2011), substitution between roduction factors whose relative rices have been modified (Chirinko et al., 2011, Karabarbounis and Neiman 2013) or changes in the volume of outut due to lower rices associated with lower costs. Obviously, firms roduction costs are strongly related to firms behavior, and it is not ossible to directly estimate the link between costs and the various deendent variables. The reform of the CETC not only needs to be evaluated er se, it also rovides an oortunity to assess how firms react to an (indirect) decrease in wage cost. Indeed, it exogenously alters the roduction costs through a tax credit based on wages lower than 2.5 minimum wage. We use this exogenous variation in roduction costs to imlement a double di erence estimation of the imact of wage costs on firms behavior. However, such an estimate generally requires dividing firms in two grous: the treatment grou containing firms imacted by the reform and the control grous containing those which are not. 7

8 It is not ossible here to set u a control grou because virtually all comanies benefit from the reform. However, the extent to which CETC reduces roduction costs varies widely among firms, including between like-minded firms in the same economic sector. Figure 1 resent the distribution of treatment intensity within two category of firms (categorization with resect to four sizes and six industries), which corresond to the two extreme in terms of distribution of treatment intensity. Both show a large variation in treatment intensity. We therefore imlement double di erence estimations on the treatment intensity, considered as a continuous variable. The logarithm of the deendent variable is regressed on the logarithm of the roduction cost (less the CETC from 2013). This intensity of treatment can be considered on the basis of wage costs only (regressions on emloyment and wages) or on total roduction costs (rofits). However, two main reasons may cause estimation bias: 1/ Reverse causality: any comany increasing its ayroll one year - for reasons indeendent of the CETC - de facto increases the intensity of its treatment. Thus, treatment intensity is fundamentally endogenous to ayroll growth (as long as it remains below 2.5 minimum wage). 2/ Common trend assumtion: any double di erence estimate requires that the common trend assumtion between more intensively and less intensively treated grous be verified. Here, this means that firms that are more or less intensely treated by the CETC (i.e. comanies whose eligible wagebill is more or less imortant comared to roduction costs) would have behave the same way in the absence of the CETC. This is not the case and it is therefore necessary to ensure that regressions are e ectively controlling for otential heterogeneity of trends. Figure 1: Distribution of treament intensity among manufacturing and ersonal services SMEs Source: DADS-FARE-MVC

9 The statistical treatment of these two otential biases will be di erent and resented in the next two subsections. 3.1 Reverse causality In order to tackle the reverse causality issue, a common solution used in the literature is to assign treatment on the lagged values of the variables which constitute the tax base or the subsidies. This strategy was used by Auten and Carroll (1999) in their estimation of the imact of the taxation of earned income, by alying the variation on the rate of earned income the year receding the reform. In our case, it comes to use the relative stability in the roduction structure and to consider the ratio of eligibility the years receding the introduction of the CETC as a roxy of the ratio of eligibility ex ante. In other words, we use the ratio of eligibility that firms would have had given their roduction structure before the imlementation of the CETC so as not to take into account their endogenous resonse behavior in the comutation of the ratio of eligibility and avoid the reverse causality issue. The same tye of methodology was used in the case of France by Créon et Deslatz (2001) to evaluate the imact of social contribution rebates in France. In order to ensure that the reverse causality is not anymore an issue, one has to check for the validity of the instrument. A good instrument must fulfill two conditions: it has to be exogenous (contrary to the regressor that it is intended to instrument) and is has to be highly correlated with this regressor. Regarding the first criterion, temoral lags ensure exogeneity. The intensity of treatment is calculated with revious year wage structure, which has been chosen by firms in order to adat the 2012 economic situation (with ast deendency), without knowing about the existence and future introduction of the CETC. Indeed, the tax credit has been voted at the very end of 2012, and was resented and discussed in very short time at the end of this year. Consequently, the intensity of the intention to treat (otential treatment rior to firm behavior in resonse to treatment) can be considered as exogenous. Regarding the second criterion, one can analyze the ower of rediction of the e ective treatment (i.e. of the CETC initialized in 2013 according to the tax records in the MVC database). In our case, it comes from the relative stability of firm roduction structure over time. Indeed, regressions of e ective treatment (tax credit e ectively initialized by comanies in 2013 according to the MVC) on the instrument (i.e. tax credits redicted according to the wagebill of wages inferior to 2,5 the minimum wage the year receding the reform), reveal the redictive ower of this instrument. The coe cient is always highly significant and very near from unity. Besides, these first stage regressions contribute to exlain most of the variance of the instrumented variable : more than 90% of the CETC initialized when measured as as a share of wage cost and between 66% and 80% of CETC initialized when measured as a share of total costs. The main rincile of the double di erence estimation is to comare the evolution of treated and control grous before and after a reform. In order to do so, we estimate a anel regression with individual fixed e ects, a time dummy and a time dummy interacted with the treated grou. 9

10 Given that we our control grou is not comosed of firms which do not benefit from the tax credit (extensive margin) but rather by firms less intensively treated (intensive margin), we interact the time dummy with the intention to treat (i.e. the tax credit that firms could get given their roduction structure the years receding the reform). This is summarized in the following equation 1. ln(y i,t )= + 13.I i,t.1 [t=2013] + 14.I i,t.1 [t=2014] + X j.x j,i,t + X f.1 [f] + i,t (1) j f CICEi,t where I i,t = ln(1 i 1 C i,t 1 ) is the intention to treat comuted as a share of roduction costs. Note that deending on the secification, roduction costs corresond to the total wagebill (from the DADS database) or to the total roduction costs (from the FARE database). X j,i,t stand for the values of di erent controls j, 1 [f] for the di erent fixed e ects and i refers to firms and t to year. For ensuring exogeneity, as for treatment variables, control variable are also lagged one year. The coe cients 13 et 14 can be interreted as elasticities of the variable Y i,t relative to roduction costs for 2013 and Note that a negative sign was added in front of the intensity variable for the ease of interretation given that the CETC reresents a diminution in costs. This choice was made so that regressions coe cients can be easily interreted as the imact of the CETC: a ositive sign in result tables means that the CETC has a ositive imact on the outcome variable. 3.2 The common trend assumtion To be relevant, the double di erence estimation relies on a strong assumtion: the one of common trend assumtion. In other words, this method is valid if and only if firms with di erent intensity to treat follow the same trend before the introduction of CETC. This assumtion, when verified, means that di erences between the treated and less treated firms would have remained constant over time in the absence of the olicy under evaluation, and therefore ensure that relatively less intensively treated firms can serve as valid counter-factual for more intensively treated firms. In the oosite case, if those firms would not follow the same trend before the reform, it would be imossible for the econometrician to disentangle, in the ost-reform evolution between the treated and less intensively treated firms (double di erence estimation), what can be attributable to the olicy from what can be attributable to any other confounding factor. As this is not always the case, one can take into account any otential re-reform di erent trends by including several controls for caturing these trends. This becomes a common trend ceteris aribus. In this urose, we add may controls in the regressions. First, we introduce sector year and firm-size year fixed e ects in order to cature all secific sectoral and size class trends. We also control for firm fixed e ects to control for all the unobservable which are secific to a firm and constant over time. Moreover we add di erent time-varying controls at the firm-level which can also influence our outcome variables: roductivity (measured as a ratio of valued added on average emloyment), caital stock (tangible and intangible assets) as well as firm average wage. Moreover, we add many controls of a firm roduction structure. In order to cature most of the 10

11 intrinsic di erences re-reform between the firms which (will) become more intensively and less intensively treated after the introduction of the CETC, we control for the share of wagebill inferior to 2.5 the minimum wage, I i,t, without interacting it with the year dummies. Moreover, as in France, there are some yearly variations in the minimum wage and some exemtions which are roortional to the minimum wage, these variations can imact the total wagebill and therefore also imact our outcome variables. As a consequence, we also add controls of the share of wagebill exosed to the minimum wage variations to avoid estimation bias. In articular, we introduce IMISCi,t a = MS[1,5] i,t MS i,t 1 [t=a] for the di erent years a, wherems i,t is the gross wagebill of firm i in year t and MS [1,5] i,t its wagebill when workers ayed less than 1.5 the minimum wage are considered. Moreover, we also include secifications in which we relace our set of controls by the one used by Gilles et al. (2016): rofit margin (Gross oerating surlus/value added), economic rentability (Gross oerating surlus/tangible and intangible assets), roductivity (Value added/workers), caital intensity (tangible assets / workers), share of exort in the turnover, investment rate (tangible investments/value added), rate of debt (borrowing and debts on the sum of share caital, issue remium, investment subsidy in the liabilities and other equities), rate of financial burden (borrowing interest/gross oerating surlus), as well as di erent elements of the comosition of workers by gender (share of women) by socio-rofessional category (share of blue-collars, white-collars etc.) or by tye of contract (full-time, art-time, short-term or long-term contracts). In order to check that these controls roerly cature the intrinsic re-reform di erences between firms relatively more or less intensively treated, a usual test consists in oerating lacebo regressions. The idea is to roceed to the same regression as described in equation 1, but only on the years receding the e ective introduction of the CETC, in order to measure the fictive e ect of the introduction of the CETC in This comes to estimate the following equation 2. ln(y i,t )= + lacebo.i i,t.1 [t=2012] + X j j.x j,i,t + X f f.1 [f] + i,t (2) The lacebo test is valid only if the coe cient lacebo is not significantly di erent from zero; meaning that the deendance of the outcome variable in the structure of roduction which induces the intention to treat is stable before the introduction of the CETC. In other words, this is a test of the common trend assumtion. If the lacebo test validates the common trend assumtion, the coe cients 13 and 14 can be seen as unbiased estimates of the elasticity of the outcome variable Y on wage cost (or roduction costs deending on the outcome of interest). If the lacebo test rejects the common trend assumtion, one has to better control for this trend heterogeneity. One way to do it is to estimate a trile di erence rather than a double one, which reduced from is given by equation 3. ln(y i,t )= I i,t.1 [t=2013] I i,t.1 [t=2014] + X j j. X j,i,t + X f f.1 [f] + i,t (3) 11

12 where stands for first di erence estimator. In this secification in trend, the firm fixed e ects measures the trend (out of treatment) in the growth rate of the outcome variable Y secific to each firm. 4 Results 4.1 Emloyment The objectives of the CETC was rimarily to boost emloyment; this is usually the main goal of olicies aiming at decreasing labor cost. There are several ways of counting emloyment: the two most usual being the number of workers emloyed (whatever their working time) and the number of full-time equivalent jobs (or equivalently the number of hours worked). We analyze these two measures as deendent variables. Furthermore, if the number of hours are only given by the DADS database, the workforce in each firm is given by two di erent sources: the DADS and the FARE databases. The two variables are regressed for checking the robustness of our results. Note that we are doubling the estimates: a series of regressions is weighted by workforce in 2012 and the other is not weighted. Unweighted regressions give more imortance to small firms because they are more numerous; they reveal the behavior of firms, considered as decision-making units. Conversely, weighted regressions give more weight to firms with a greater share of jobs; their coe cients are closer to an interretation of macroeconomic e ects. Besides, as highlighted by Solon et al. (2015), the comarison of weighted and unweighted regression is informative as it can reveal the existence of heterogeneity in firm behavior (according to firm size in our case) when coe cients di er. We therefore choose to resent both tyes of estimates because they are both relevant and comlementary. The main results of estimations are resented in tables 1. For robustness checks, alternative secifications are resented in Tables 6 to 11 of annex A.1. Results are quite stable in all secifications. The unweighted regressions validate the common trend assumtion, with lacebo tests very close to zero (not significant desite very small standard errors). However, lacebo tests are not validated for weighted regressions. Whatever the secification and the tye of emloyment measure used as deendent variable, results suggest that the CETC has had no imact on emloyment. Surrisingly, the e ect, if any, would rather be negative even though very small. Given that average e ects can hide heterogeneity between workers, we reroduce these estimates for di erent categories of emloyees aart. For these categorized deendent variables, we use only workforce extracted from DADS (as such details are not available in the FARE database). The first categorization concerns the socio-rofessional categories, the results of the regressions being resented in table 2 (and alternative secifications in annexe A.2 in tables 12 to 19). We find a small ositive e ect for executives and higher intellectual occuations, but with a lacebo test which fails for unweighted regressions. For weighted regressions, the lacebo test validates the estimations and the imact of the CETC is ositive in However, results are 12

13 Table 1: Imact of CETC on total emloyment Deendent variable Average emloyment Hours worked DADS FICUS-FARE DADS Unweighted regressions Placebo test (0.0537) (0.0438) (0.0467) Intention to treat intensity, *** *** *** (0.0547) (0.0444) (0.0416) Intention to treat intensity, *** *** *** (0.0470) (0.0400) (0.0386) Observations R Weighted regressions Placebo test * ** (0.435) (0.576) (0.558) Intention to treat intensity, *** (0.277) (0.289) (0.331) Intention to treat intensity, ** (0.347) (0.230) () Observations R Notes: Regression of the deendent variable (logarithm of workforce from DADS or FARE databases and yearly hours) on the intensity of the intention to treat, with controls for firm roductivity, caital stock, mean wage, wage structure, minimum wage exosure and fixed e ects: year industry, year size and firm. Robust standard errors in arentheses (firm level cluster), * <0.05, ** <0.01, *** <0.001 Sources: DADS, FARE, MVC not significant for intermediate rofessions. For blue collar workers, we find the same results as for global estimations, with validated lacebo tests and negative coe cients of imact estimation. As resented in tables 20 to 23 in aendix A.3, no di erence aears between male and female workers: the results for both are the same, and very close to those obtained for all workers estimations. Regarding contract tyes (tables 24 à 27 in aendix A.4), the results are mainly inconclusive but the lacebo tests fail. Overall the CETC has not had the exected ositive e ect on jobs, either because firms need more time before adjusting emloyment or because firms refer use the CETC for another urose. This is the toic of the next section. 4.2 Profits If the CETC has not been used to increase the workforce, keeing the tax credit as net rofit is another ossible use. However, it is not trivial to roerly measure it because there is no automatic way of including it in firm accounts. Therefore, we consider three rofit indicators: gross margins (EBIT as a roortion of turnover), net margins (EBITDA as a roortion of turnover), and 13

14 Table 2: Imact of CETC on emloyment er sociorofessional category Deendant variable : aveage emloyment Executives, Intermediate Blue collars higher intellectual rofessions workers rofessions Unweighted regressions Placebo test 0.709*** 0.247* (0.111) (0.132) (0.119) Intention to treat, intensity, *** *** (0.102) (0.120) (0.110) Intention to treat intensity, *** *** (0.0871) (0.0953) (0.0839) Observations R Weighted regressions Placebo test (0.789) (1.243) (1.418) Intention to treat intensity, (0.490) (0.866) (1.060) Intention to treat intensity, * 1.834** *** (0.417) (0.730) (0.994) Observations R Notes: Regressions of the logarithm of average emloyment er socio rofessional category on the intensity of the intention to treat, with controls for firm roductivity, caital stock, mean wage, wage structure, minimum wage exosure and fixed e ects: year industry, year size and firm. Robust standard errors in arentheses (firm level cluster), * <0.05, ** <0.01, *** <0.001 Sources: DADS, FARE, MVC oerating margins (oerating income as a roortion of oerating costs). Comanies were advised to account for the CETC as deduction of ersonnel costs. If they did it in this way, then the CETC should aear in all three rofit indicators. Nevertheless, many other accounting entries were ossible, some involving not taking into account this writing in one or more indicators, so one should be aware that we might not fully cature the imact on rofits. The results of the estimations of CETC on each of the three rofit indicators are summarized in table 3. Regarding unweighted regressions, the lacebo tests validate the common trend assumtion uon which rely our double di erence estimations for gross margins and net margins and the CETC has a ositive and significant imact in 2014 in the case of net margins. Regarding weighted regressions, lacebo tests are validated for all rofits indicators, and results indicate a ositive and significant imact of the CETC on rofits in Overall, these results suggest that French firms used the tax credit to restore their margins and moreover that the imact of the CETC tends to increase over time. 14

15 Table 3: Imact of the CETC on rofits Deendant variables Net margins Gross margins Oerating margins Unweighted regressions Placebo test *** (0.0296) (0.0376) (0.0307) Intention to treat intensity, (0.0284) (0.0293) (0.0290) Intention to treat intensity, * (0.0213) (0.0222) (0.0222) Observations R Weighted regressions Placebo test (0.141) (0.116) (0.0872) Intention to treat intensity, * (0.125) (1.103) ( ) Intenstion to treat intensity, *** 0.168* 0.193* (0.0996*) (0.0723) (0.0732) Observations R Notes: Regression of the deendent variable on the intensity of the intention to treat, with controls for firm roductivity, caital stock, mean wage, wage structure, minimum wage exosure and fixed e ects: year industry, year size and firm, and weighted with 2012 workforce. Robust standard errors in arentheses (firm level cluster), * <0.05, ** <0.01, *** <0.001 Sources: DADS, FARE, MVC Wages If only a fraction of the CETC has translated into increases in firm rofits and if they did not use this credit to hire, firm might have shared the benefits of the tax credit with their emloyees. However, if any, the otential imact on wages is not clear. The tax credit scheme is very articular since it is based on the ayroll by generating a significant threshold e ect: the CETC rate is constant (4% in 2013 and 6% starting in 2014) and suddenly died down to 2.5 SMICs. Thus, a full-time emloyee aid two and a half times the minimum wage oened in 2014 an annual tax credit of 2,600 euros for his emloyer. If she had been aid even a few euros more, her emloyer could not have received any tax credit. Hence, emloyers may be articularly reluctant to grant increases to their emloyees close to the threshold or they will seek to set the wage of their new recruits as far as ossible below this threshold. In their extreme 15

16 configurations, strong emloyer reactions to these two issues (increases and hires) could lead to bunching at the threshold (Saez 2010). However, Carbonnier et al. (2014) showed that a ga discontinuity in a framework where the assignment variable was only imerfectly controlled could lead not to a oint of accumulation but to a discontinuity in the values of the variables involved. Carbonnier et al. (2016) study the two hyotheses and did not found the tiniest sign of wage setting behavior around the threshold. Therefore the wage behavior may be more sreadly distributed. The benefit of the CETC may be artially redistributed to emloyees indeendently on their wage osition vis à vis the threshold. In order to test this, we aly our identification strategy to di erent wage indicators. The main results for all tyes of emloyees are resented in table 4 and alternative secifications are reorted in annexe B.1 for robustness checks. Three di erent indicators are considered: mean yearly wage er emloyee (based on FARE workforce indicator), mean hourly wage of staying emloyees and the mean of individual hourly wage growth. Table 4: Imact of the CETC on wages Deendent variable Avergae Growth Average hourly in hourly yearly wage wage wage Unweighted regressions Placebo test 1.435*** 0.788*** 1.838*** (0.0332) (0.0303) (0.0629) Intention to treat intensity, *** 0.715*** 1.775*** (0.0292) (0.0258) (0.0570) Intention to treat intensity, *** 0.776*** 1.842*** (0.0932) (0.0175) (0.0480) Observations R Weighted regressions Placebo test 0.790*** * (0.165) (0.173) (0.349) Intention to treat intensity, *** 0.529*** (0.204) (0.134) (0.253) Intention to treat intensity, *** 0.456*** 0.572* (0.139) (0.0833) (0.258) Observations R Notes: Regression of the deendent variable on the intensity of the intention to treat, with controls for firm roductivity, caital stock, mean wage, wage structure, minimum wage exosure and fixed e ects: year industry, year size and firm, and weighted with 2012 workforce. Robust standard errors in arentheses (firm level cluster), * <0.05, ** <0.01, *** <0.001 Sources: DADS, FARE, MVC For yearly and hourly wages, double di erence lacebo tests fail but for the mean of hourly 16

17 wage growth, lacebo tests validate the common trend assumtion for the weighted secifications. In that case, the estimates are significantly ositive suggesting that the benefits of the CETC were artially transferred to some emloyees trough wage increases. Moreover, their value around 50% is very close to revious estimates of the CIT incidence on wages (Arulamalam et al. 2012, Dwenger et al. 2011, Liu & Altshuler 2013). However, these e ects on wages are not the same for all tyes of workers. Main results of regression by socio-rofessional category are reorted in Table5 (and alternative secifications in annexe B.2in tables 32 to 37. Regarding average hourly wage, weighted regressions suggest a ositive imact for executives and higher intellectual rofessions only. Regarding growth in wages, all lacebo tests reject the common trend assumtion in the case of unweighted regressions. However, all lacebo tests validate weighted regressions. Results suggest a ositive and significant imact of the CETC on wages for executive and higher intellectual rofessions and for intermediate rofessions and to a lesser extent for blue-collar workers in Table 5: Imact of CETC on wages er socio-rofessional catgeory Executives and Intermediate Blue collar intellectual rofessions workers rofessions Average hourly wage Placebo test (0.280) (0.300) (0.267) Intention to treat intensity, *** (0.229) (0.235) (0.376) Intention to treat intensity, *** 0.372* (0.137) (0.147) (0.184) Observations R Growth in hourly wage Placebo test (0.253) (0.255) (0.264) Intention to treat intensity, ** (0.194) (0.194) (0.184) Intention to treat intensity, *** 0.336** (0.126) (0.105) (0.128) Observations R Notes: Regression of the deendent variable on the intensity of the intention to treat, with controls for firm roductivity, caital stock, mean wage, wage structure, minimum wage exosure and fixed e ects: year industry, year size and firm, and weighted with 2012 workforce. Robust standard errors in arentheses (firm level cluster), * <0.05, ** <0.01, *** <0.001 Sources: DADS, FARE, MVC r 17

18 5 Conclusion In this aer, we exloit a large French CIT reform, introduced in 2013, (cometitiveness and emloyment tax credit, CETC) to assess the imact of cororate tax aiming at reducing labor costs on firm behavior. Our emirical analysis relies on three exhaustive databases which contain recise information at the firm and individual levels on the eriod We set an identification strategy in double (and trile) di erence based on the intensity of the intention to treat to quantify the imact of the introduction of this tax credit on three di erent outcomes: emloyment, rofit and wages. Our results suggest that the CETC has had no ositive imact on emloyment. Even more, some counter-intuitive results are found for lower socio-rofessional categories, whereas the coe cients are close to zero for uer socio-rofessional categories. Conversely, the imact on firm rofits is ositive and significant and tends to increase over time. Firms therefore mainly used the CETC to restore their margins. However, e ects on wage also aear ositive and significant : some of the benefits of the CETC have been distributed to emloyees through wage increases, a results which is close to revious estimates in the literature. Moreover, the results on wages di er deending on the socio-rofessional category of the emloyees. The stronger imact is found for executive and higher intellectual occuations while intermediate occuations benefit from mean wage increases. For blue collar workers and other emloyees, the results are not always significant and aear less robust. These results give new evidence about the imortance of taking into account rent sharing in favor of caital and in favor of white-collar emloyees when it comes to assess the e ectiveness of such tax incentives. References Anderson P.A., Meyer B.D. (1997) The e ects of firm secific taxes and government mandates with an alication to the U.S. unemloyment insurance rogram, Journal of Public Economics, vol. 65, Anderson P.A., Meyer B.D. (2000) The e ects of the unemloyment insurance ayroll tax on wages, emloyment, claims and denials, Journal of Public Economics, vol. 78, Arulamalam W., Devereux M.P., Ma ni G. (2012) The direct incidence of cororate income tax on wages, Euroean Economic Review, vol. 56, Bennmarker H., Mellander E., Ockert B. (2009) Do regional ayroll tax reductions boost emloyment?, Labour Economics, vol. 16, Bohm P., Lind H. (1993) Policy evaluation quality: a quasi-exerimental study of regional emloyment subsidies in Sweden, Regional Science and Urban Economics, vol. 23, Bunel M., L Horty Y. (2012) The E ects of Reduced Social Security Contributions on Emloyment: An Evaluation of the 2003 French Reform, Fiscal Studies, Vol. 33, 3 18