Paper_Capital Structure and Employment Flexibility

546 views

Published on

NES 20th Anniversary Conference, Dec 13-16, 2012
Article "Capital Structure and Employment Flexibility" presented by Olga Kuzmina at the NES 20th Anniversary Conference.
Author: Olga Kuzmina, Assistant Professor, New Economic School

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
546
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
8
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Paper_Capital Structure and Employment Flexibility

  1. 1. Capital Structure and Employment Flexibility Olga Kuzmina New Economic School November 2012 Abstract This paper uses a unique panel dataset to establish a causal relationship between the use of ‡exible contractual arrangements with labor and capital structure of the …rm. Using the exogenous inter-temporal and cross-regional variation in government labor policies, I …nd that hiring more temporary workers leads …rms to have more debt. Characterized by much lower …ring costs, temporary employment contracts allow …rms to adjust the labor force and pro…ts upon negative shocks realizations and reduce the default risk on the margin, thereby promoting debt …nancing. I …nd the supporting evidence of this mechanism and also interpret it as a substitution between operating and …nancial leverage. Given the overwhelming extent of labor reforms in continental Europe in recent years that touch upon the incentives to use di¤erent employment contracts and are aimed at o¤ering more job security to workers, it is important to understand how such policies would a¤ect …rms. Keywords: Capital Structure, Fixed-term Contracts, Operating Leverage JEL codes: G32, J47, M55 Most of the work on this paper was carried out during my doctoral studies at Columbia Graduate School ofBusiness. I wish to acknowledge with deepest gratitude the never-ending support of my advisor Maria Guadalupe. Iwould like to specially thank Ignacio García Pérez for sharing the data on subsidies. I am much obliged for many con-versations and suggestions to Patrick Bolton, Wei Jiang, Sarmistha Pal, Daniel Paravisini, Bernard Salanié, CarstenSprenger, Sergey Stepanov, Catherine Thomas, Vikrant Vig, Daniel Wolfenzon, and to the seminar participants atColumbia Graduate School of Business, Goethe University Frankfurt, New Economic School, Universtity of Mary-land –Smith School of Business and University of Surrey Business School. I am also thankful to the participants ofthe 2012 International Finance and Banking Society meeting in Valencia. All remaining errors are my own. Emailfor correspondence: okuzmina@nes.ru; http://pages.nes.ru/okuzmina. Address: O¢ ce 1721, Nakhimovsky pr. 47,Moscow 117418 Russia. 1
  2. 2. 1 IntroductionThis paper studies how a …rm’ composition of contractual arrangements with factors of production sa¤ects its capital structure. In an uncertain environment the option to adjust input costs aftershocks have realized has value. By using ‡exible contract arrangements with capital and labor,which e¤ectively convert some of the …xed operating costs into variable costs, a …rm can lowerits operating leverage. The optimal level of overall risk that a …rm is willing to tolerate de…nesthe total amount of …xed costs that the …rm can potentially cover with its variable cash ‡ow, sothat operating and …nancial leverage are substitutes (Mandelker and Rhee, 1984). Since the useof ‡exible contracting with factors of production reduces operating leverage, it should also increasethe debt capacity of …rms. Given the natural existence of various types of contractual arrangements with labor that di¤erdramatically in terms of employment ‡exibility that they provide, it is surprising that the e¤ects ofcomposition of employment contracts have not been evaluated in the context of the choice of …nancialstructure. My paper attempts to …ll this gap. I use a unique panel dataset of manufacturing …rmsto show that the use of more ‡exible contract arrangements with labor increases debt …nancing.Importantly, I build the identi…cation strategy on the exogenous inter-temporal and cross-regionalvariation in government programs that aimed at increasing job security among workers by promotingless ‡exible contracts with labor for …rms. This setting, akin to a natural experiment, allows toidentify the causal e¤ect of employment ‡exibility on capital structure. The two main types of employment contracts that exist across European countries are permanentand temporary employment contracts. The main distinctive feature of temporary contracts isa much lower …ring cost, as compared to permanent contracts. When workers are hired undertemporary contracts, …rms can adjust their total labor force, and hence labor costs and pro…ts,faster and more cheaply when responding to idiosyncratic shocks, as compared to employing workerson permanent contracts. Consider, for example, an adverse shock to product demand or businessconditions that lowers average labor product and …rm’ cash ‡ s ow. If the shock is su¢ ciently bad,and workers are costly to …re, the …rm might not be able to meet its debt obligations (in the formof the interest payment or principal due) and go bankrupt. Absent …ring costs, however, the …rmwould …nd it optimal to lay o¤ some of the workers, thereby increasing its pro…t on the marginand meeting the debt obligations. The absence of …ring costs e¤ectively makes total labor costsless …xed and more variable by allowing …rms to match them to idiosyncratic shocks, especially onthe downside where bankruptcy is a concern. This is precisely the implicit option provided by the 2
  3. 3. more ‡exible –temporary –contracts. By decreasing the probability of bankruptcy on the marginthey reduce the expected costs of …nancial distress. This enhances the debt capacity of …rms andenables them to support a higher level of debt that may be advantageous due to di¤erent motives,such as tax shield considerations (DeAngelo and Masulis, 1980) or reductions in the free cash ‡owthat is available for overinvestment by managers as in Stulz (1990), or others. Therefore, otherthings being equal, …rms that employ relatively more workers on temporary contracts should usemore debt …nancing. The results of my paper indeed support this hypothesis. The economic magnitude of the causale¤ect of ‡exible employment on capital structure is large. A thought experiment of completelyprohibiting an average …rm from o¤ering temporary employment contracts would suggest that sucha …rm should reduce its debt level by 3.6 percentage points, which is about 6.3% of the average debtlevel across …rms. Given the overwhelming extent of labor reforms in continental Europe in recentyears that aim at o¤ering more job security to workers by touching upon the …rm-level incentivesto use di¤erent employment contracts, it is important to understand how such policies would a¤ect…rms. My paper contributes to several strands of literature. Firs of all, it relates to the research onthe interactions between corporate …nance and labor economics that has recently attracted someattention (see a survey by Pagano and Volpin, 2008). A large strand of this literature has focused onthe e¤ects of labor policies, typically related to unionization, on …rms’real decision and outcomes,such as pro…tability and market values (Ruback and Zimmerman, 1984, Abowd, 1989, and Hirsch,1991), cost of equity (Chen, Kacperczyk and Ortiz-Molina, 2009), innovation (Acharya, Baghai andSubramanian, 2010), investment and economic growth (Besley and Burgess, 2004). The generalconclusion of these papers is that pro-worker regulation a¤ects …rms in a negative way. As forcapital structure, Perotti and Spier (1993), Matsa (2010), Simintzi, Vig and Volpin (2010) haveexplored the strategic e¤ect of debt …nancing when workers are unionized, suggesting that a …rmmay ex ante choose the level of debt in a such way so as to preclude workers from bargaining overtheir wages ex post (with the direction of the e¤ect depending on whether debt is renegotiable ornot). My paper di¤ers from this literature in an important way in that it looks at the e¤ects of‡exible employment contracts per se, rather than those of the union-level bargaining. By exploringthe setting of the Spanish institutional environment, where all workers in a …rm irrespective of theircontract type are covered by collective bargaining agreements, and these agreements are generally 3
  4. 4. set at the levels above …rm (industry provincial or industry national), one can …lter out the e¤ectof union-level bargaining and illustrate a di¤erent causal e¤ect – the one of ‡exible employmentcontracts –that e¤ectively goes through the changes in …rm’ operating leverage. s To shed light on this precise mechanism behind the e¤ect of employment ‡exibility on capitalstructure, I complement my main results with additional tests. In particular, I provide evidencethat temporary workers are used as a margin of adjustment when negative shocks arrive, so thattemporary contracts are indeed the more ‡exible contract arrangements with labor. Then, I revealthat the use of temporary contracts reduces probability of bankruptcy by exploring cross-sectionalheterogeneity in the levels of bankruptcy costs and comparing the e¤ects on capital structure forthese di¤erent types of …rms. Finally, I look at the data from yet another angle by providing somesuggestive evidence that being unable to adjust capital structure to the changes in employmentstructure is related to default. Overall, these results taken together strongly indicate that themechanism behind the causal e¤ect of employment ‡exibility on debt …nancing is the one of theoperating to …nancial leverage substitution. Although the idea of the substitution between operating and …nancial leverage has been aroundfor many decades, the empirical evidence of …rms changing their …nancial leverage to match changesin operating leverage remains scarce. Several notable exceptions include Petersen (1994) who ex-amines the role of operating leverage in the …rm’ pension choice. In particular, he …nds that the sprobability of a …rm choosing a more ‡exible de…ned contribution plan, rather than a less ‡exiblede…ned bene…t plan, is higher on average for …rms with more variable cash ‡ows. He interpretsthis result as …rms e¤ectively reducing their operating leverage by selecting a de…ned contributionplan, although not explicitly relating it to the changes in …nancial leverage in the empirical setting.Hanka (1998) explores employment decisions in U.S. …rms and …nds that having more debt is corre-lated with reducing …rm employment more heavily and relying more on part-time labor force. Hisconjecture is that this may be due to higher incentives of …rms to make labor costs variable ratherthan …xed. Nevertheless, the exact direction of the causal e¤ect in many of the proposed empiricalrelationships between variables that potentially a¤ect operating leverage and debt …nancing, is notalways clear, because …rms choose their real and …nancial policies endogenously. My paper adds tothis literature by quantifying the magnitude of the causal e¤ect in a particular direction: the onefrom employment ‡exibility, which reduces operating leverage, to …nancial leverage. Finally, my paper contributes to the empirical literature examining the relation between variousreal ‡exibilities and …nancial structure. Mauer and Triantis (1994) provide a dynamic model and use 4
  5. 5. numerical analysis to suggest that production ‡exibilities of …rms can enhance their debt capacity.MacKay (2003) shows that investment ‡exibility in workforce, estimated by the ratio of actual toshadow rents of the workforce input, is positively related to leverage ratios. In contrast to thesepapers, I do not need to rely on calibrated measures of real ‡exibilities, but can directly observeemployment ‡exibility at the …rm level, as measured by the composition of employment contracts,and importantly use an exogenous shock to this composition to establish a causal relationshipbetween employment ‡exibility and capital structure. The rest of the paper proceeds as follows. Section 2 outlines the details of the institutionalenvironment, building in the identi…cation strategy and describing the data. Section 3 presents theempirical results on the e¤ect of employment ‡exibility on capital structure. Section 4 providesevidence of the mechanism of operating to …nancial leverage substitution behind these results byshowing that temporary workers provide the margin of adjustment to negative shocks and by ex-ploring the cross-sectional heterogeneity in bankruptcy costs. Section 5 concludes and discussespolicy implications.2 Institutional Environment and Identi…cation StrategyThis section presents a brief description of the institutional environment in Spain and the govern-ment policies promoting the use of permanent employment contracts at the expense of the temporaryones. It then discusses how the di¤erential implementation of these policies is used to constructan exogenous source of variation in employment ‡exibility within and across …rms to test for thecausal relation between ‡exible employment and capital structure of the …rm.2.1 Dual Labor MarketA dual labor market consisting of workers characterized by di¤erent degrees of job security existsin virtually every country, either informally (with "under-the-table" payments) or formally (withdi¤erent legal contract arrangements with employees). One particular country that provides anexcellent opportunity to study the e¤ects of employment ‡exibility on …nancing decisions of …rmsis Spain. First of all, the labor market in Spain is a formal dual market, which enables one toaccurately measure the composition of the labor force by the type of employment contract and thecorresponding degree of job security. Second, there is an extraordinary level of temporary employ- 5
  6. 6. ment there1 , which has been the highest among the OECD countries (OECD, 2002), suggestingthat temporary employment contracts are not an artifact and indeed are commonly used in prac-tice. Finally, the only legal di¤erence across the two types of contracts is the implicit …ring costassociated with each type of contract. Temporary employees (especially those hired under a particular type of temporary contract –…xed-term employment contract) often perform the same job within a …rm as permanent employees.Most importantly, all workers in a …rm irrespective of their contract type are covered by collectivebargaining agreements, and the agreement a given …rm faces does not allow it to discriminatebetween workers based on their contract type, e.g. by paying di¤erent wages to workers on di¤erentcontract types (Jimeno and Toharia, 1994). No worker can be excluded from the provisions of thecollective bargaining agreement. Furthermore, the agreements for 85% of …rms in manufacturing,and especially the smaller ones, are not at the …rm level, but rather at a more aggregated level (suchas industry provincial or industry national, as reported by Izquierdo et al., 2003). These agreementsapply to all …rms equally irrespective of whether they participated in the actual bargaining processor not, and given that smaller …rms which I have in the sample are generally not in the core of thebargaining process, those agreements are arguably exogenous for them. This means that workerswithin the same …rm cannot be characterized by di¤erent degrees of wage bargaining. On the other hand, the di¤erence in …ring costs across the two types of contracts is quitedramatic, meaning that on the side of the …rm it is much easier to lay o¤ workers on temporarycontracts rather than on permanent contracts. When a temporary worker is dismissed (or whena …xed-term contract worker is not converted into a permanent one at the end of the three-yearmaximum tenure) a …rm pays 0 to 12 days’wages in severance payments as opposed to 33 to 45days’wages for permanent workers (Jimeno and Toharia, 1994). Since both …gures are per year ofseniority, the e¤ect is further ampli…ed by the observation that a permanent worker is more likely tohave worked in the …rm for a longer time, given the three-year legal limit for workers on …xed-termand apprenticeship contracts and the short nature of contracts for temporary jobs. Hence the costdi¤erential in absolute terms is even larger. Moreover, …ring a permanent worker can involve a courtprocedure for "unfair dismissal" with substantial administrative costs, while a temporary workerdoes not have the right to sue her employer for lay-o¤. Finally, a …rm may simply choose not toprolong the …xed-term contract upon expiration and anecdotal evidence suggests that …xed-termcontracts with some employees get renewed every week, providing the …rm with an option to adjust 1 24% of all salaried workers as of 2010 according to the Encuesta de Población Activa 2010 (Economically ActivePopulation Survey), conducted quarterly by the National Institute of Statistics (INE). 6
  7. 7. its labor force and total wage bill for the next period almost immediately.2 Ultimately, investigating the e¤ects of employment ‡exibility on capital structure is very ap-pealing in the framework of Spanish institutional setup, because the di¤erence across …rms in thecomposition of employment contracts can fully characterize the di¤erence in the degree of ‡exibilityon the side of the …rm in adjusting the labor force following negative shocks, keeping the e¤ectsof union-level bargaining constant. Finally, the large di¤erence in …ring costs across the two typesof contracts implies that …rms with di¤erent contract compositions will be far apart in terms ofemployment ‡exibility, so that it will be easier to detect its e¤ect on capital structure of the …rm.2.2 Panel FrameworkThe main hypothesis of this paper is that composition of employment contracts a¤ects capitalstructure of …rms. In particular, the use of more ‡exible employment, i.e. hiring workers that canbe easily …red when needed, enables …rms to take more debt. This relationship of interest can beoperationalized in the following way: 0 Dit = rt + T empit 1 + Xit 2 + i + it ; (1) where Dit is the ratio of total debt to total assets, rt are the region-year …xed e¤ects, T empit 1is the proportion of workers on temporary contracts in the prior year3 , Xit 0 2 are various …rm-levelcontrol variables (taken with a two-period lag to make them predetermined4 ; included in somespeci…cations to account for past …rm-speci…c shocks), and i are …rm …xed e¤ects. The panel structure of the dataset allows me to control for any time-invariant observed andunobserved di¤erences that …rms may have with respect to their capital structure. Some examplesinclude whether the …rm in general has a more variable cash ‡ whether it is a small business with ow,a distrust in credit and banking, whether its tasks generally require more human capital speci…city,or whether its assets are on average more tangible. This provides an opportunity to explore whatdrives within-…rm changes in …nancing decisions holding constant any time-invariant heterogeneity 2 For the operating vs …nancial leverage story to work it must be the case that wages are senior to debt repayments,since otherwise they can be abandoned when bankruptcy becomes a concern, so that hiring temporary vs permanentworkers does not make a di¤erence in terms of shifting the bankruptcy threshold. Indeed, in Spain wages are seniorclaims to non-collaterized debt. 3 I have allowed for a one year lag in the independent variable, because it may take time for the …rm to changeits capital structure policy upon changes in employment policy, given that these decisions are likely to be made bydi¤erent divisions in the company. Empirically contemporaneous and lagged values of Temp are highly correlated,and the results are qualitatively similar to using contemporaneous values. 4 This avoids the "bad control" problem: if contemporaneous values are not truly exogenous and can themselvesbe outcome variables, the estimate of is biased. 7
  8. 8. across …rms. In addition, including region-year …xed e¤ects allows one to make sure that the di¤erences inleverage ratios are not explained by …rms potentially having di¤erential access to credit over timeinduced by their location in more or less credit-abundant regions, or macroeconomic e¤ects drivingthe cost of …nancing. Moreover, if there is generally more pressure from the society against …ringworkers in regions with higher unemployment rates and …rms take more conservative debt policiesthere, region-year …xed e¤ects will also capture such di¤erences. It may also be the case that …rms in certain regions are able to employ di¤erent proportionsof temporary workers over time (due to e.g. worker migration) and at the same time raise debt atbetter terms. The correlation between the error term and the independent variable that could stemfrom this argument and bias the estimate of can be eliminated by region-year …xed e¤ects as well. Although equation (1) already picks up a lot of endogeneity as outlined above, there is still apossibility for the time-varying unobserved component of the error term being correlated with the…rm’ choice of employment composition. In this case, estimating (1) using ordinary least squares swould yield an inconsistent estimate of . Below I provide several potential reasons for such acorrelation and discuss the direction of the bias.2.3 Endogenous Choice of Employment FlexibilityOne of the primary reasons for endogenous choice of employment ‡exibility is the (unobservableto econometrician) …rm’ investment opportunity set, which can be changing over time at the …rm slevel and hence cannot be taken into account by saturating the model with …rm and region-year…xed e¤ects. In particular, when …rms have a large range of uncertain projects that they caninvest into and expect there to be a scope for ex post project substitution, they may prefer hiringworkers at more ‡exible temporary arrangements ex ante. On the other hand, when investorsrationally anticipate such risk-shifting behavior, they supply less debt. This would show up as anegative correlation between the error term and the independent variable, thereby biasing the OLScoe¢ cient downwards. In order to identify the causal e¤ect of ‡exible labor force one then needsto introduce a shock to this variable that would be orthogonal to …rm’ investment opportunities. s Another idea, brought about by Caggese and Cueat (2008) points out that …nancially con- nstrained …rms are more vulnerable to liquidity shocks and therefore generate "demand for ‡exibil-ity", hiring more temporary workers than …rms that are not …nancially constrained. In contrast tomy paper, they look at the e¤ect of …nancial constraints on ‡exible employment. Their argument, 8
  9. 9. however, can be elaborated on to explain why OLS would underestimate the true e¤ect of temporaryemployment on capital structure: if more …nancially constrained …rms indeed prefer more ‡exibleemployment and are at the same time less levered being unable to get su¢ cient funds, the errorterm will be negatively correlated with the independent variable. Therefore, to uncover the causale¤ect the shock to ‡exible labor force should also be orthogonal to …rm’ …nancial constraints. s These concerns suggest that if OLS estimation uncovers a positive relation between employment‡exibility and capital structure, then the true causal e¤ect is only larger. In this case OLS canbe used to argue for the existence of the e¤ect, while exogenous variation helps in quantifying thetrue magnitude. However, there are also reasons why simple panel estimates can actually be biasedupwards. In this case one needs the quasi-experimental setting to validate the very presence of thee¤ect. A potential reason for positive correlation between the error term and employment ‡exibilityis …rm’ desire to stimulate human capital investment. In particular, Butt Jaggia and Thakor s(1994) and Berk, Stanton and Zechner (2010) argue that since …rm-speci…c human capital is lostin bankruptcy, …rms that wish to induce employees to invest in human capital can o¤er longer-term contracts and precommit to more conservative debt policies. In this case, if the reason foro¤ering permanent employment contract is the (unobserved to econometrician) need for humancapital investment, which also prescribes …rms to take less debt, the OLS speci…cation would showa spurious positive relationship between ‡exible shorter-term employment and debt. To the extentthat the willingness of the …rm to stimulate human capital investment is a …xed characteristic ofeach …rm, this concern will be already taken care of by the introduction of …rm …xed e¤ects intothe speci…cation. If it is time-varying, however, one needs to use a shock to labor force compositionthat is orthogonal to …rm’ current needs for human capital in order to …nd out whether there is a scausal e¤ect between ‡exible labor force and capital structure.5 Finally, suppose that …rms indeed use temporary workers as the margin of adjusting labor forcein order to meet their debt obligations when faced with negative shocks – the very mechanismbehind the operating to …nancial leverage substitution. Then, once in a while when a negativepro…tability shock arrives, the …rm will exercise the option of …ring temporary workers, while at thesame time debt-to-assets ratio will go up, somewhat mechanically since equity in the denominator 5 A similar concern arises if workers, willing to invest into human capital, self-select into permanent contract onlywhen …rms carry relatively low debt levels. Given high unemployment rates in Spain during the period considered,it is unlikely that workers had much of the bargaining power in choosing the type of the contract. Still, even if theydid, the instrumental variable approach that I use further in the paper would cure this reverse causality concern aswell. 9
  10. 10. will be hurt by the same negative shock. This means that if the hypothesis of this paper is true, thenestimating (1) by OLS will bias the coe¢ cient of interest downwards, since part of the time …rms willbe changing the proportion of temporary workers endogenously due to the unobserved pro…tabilityshocks, generating a spurious negative correlation between the variables of interest. Therefore, if oneaims at estimating the true causal e¤ect of ‡exible employment on capital structure, it is importantto introduce shocks to the proportion of temporary workers that would be uncorrelated to …rm-levelpro…tability or demand shocks. In fact, …nding that the magnitude of the panel OLS coe¢ cient islower than the true causal one could provide some indirect evidence towards the mechanism of thepaper. All the above concerns illustrate the importance of the use of exogenous variation in proportionof temporary employment in order to reveal the magnitude of the causal e¤ect of ‡exible laboron debt …nancing, i.e. to make …rms change their workforce compositions not endogenously , forexample due to risk-shifting, …nancial constraints or demand for human capital investment, but dueto exogenous reasons. In my paper the exogenous incentives are provided by time-varying region-level government policies that promoted the use of permanent employment contract at the expenseof temporary employment contract. Hence I proceed with a brief description of these policies andthen discuss their implementation in the instrumental variable framework.2.4 Region-Speci…c Government Labor Policies in SpainThe origin of a dual labor market in Spain lies in the 1984 reform which recognized the needfor ‡exibility in the labor market by largely extending the applicability of temporary employmentcontracts. As a result, their use quickly rose up to 35% in 19956 . Empirical evidence for someof the European countries7 suggests that such dualism in the labor market may have negativee¤ects on the economy. And indeed in the late 1990s the Spanish government partially reversedthe employment liberalization policy by introducing subsidies to …rms for converting temporarycontracts with existing workers into permanent ones and for hiring new workers on permanentcontracts. The Spanish government has subsidized the creation of permanent contracts both at the nationaland at the regional levels. Since the national reform a¤ects all …rms equally and at the same time,one would not be able to attribute within-…rm changes in employment composition to the e¤ect ofthe reform itself, rather than, for example, to some global macroeconomic shocks. On the other 6 Encuesta de Población Activa 1995. 7 Blanchard and Landier (2002) for France; a survey by Dolado, García-Serrano and Jimeno (2002) for Spain. 10
  11. 11. hand, the reforms at the regional level show much more variation due to the di¤erent timing oftheir implementation, distinct worker eligibility criteria (such as gender) and di¤erent amounts tobe paid to …rms in case of a new permanent contract creation.8 These regional subsidies were paid to the …rm once at the time of conversion of an existingtemporary contract into a permanent one (or creation of the new permanent contract), either asa direct transfer to the …rm or as a reduction of payroll taxes, per each contract9 . Althoughautonomous regions introduced subsidies both for creating new permanent contracts and subsidiesfor converting existing temporary contracts into a permanent ones, the two types of subsidies werehighly correlated and in many region-years exactly identical. This makes me not di¤erentiate acrossthe two in my empirical analysis, so that I record one subsidy value for each region-year-gender –the maximum across the two if they are di¤erent. I summarize these maximum statutory subsidy amounts that a given …rm could receive per 10contract by region, year and gender of the worker in Table 1 . As can be seen from this table,the time pro…le of the policies is diverse: some regions, such as Andalucia, implemented thesesubsidies in every year from 1997 onwards, some — only in certain years, while Catalonia did notintroduce any regional-level subsidies at all during the sample period considered. One can also notea considerable variation in subsidy amounts across regions, years, and worker’ gender that range sfrom 1653 to more than 15000 Euros per contract.2.5 Implementation of the Identi…cation StrategyIn order to establish a causal relationship between employment ‡exibility and capital structure ofthe …rm, one needs to hold other determinants of capital structure constant. I implement this byestimating in equation (1) through IV-2SLS using an exogenous source of variation coming fromgovernment labor policies. In this respect, Spain represents a unique opportunity to study thiscausal e¤ect, because subsidization programs promoting permanent employment at the expense oftemporary employment were introduced in various regions of the country in di¤erent years and 8 I would like to thank Ignacio García Pérez for sharing the detailed data on subsidies that he and YolandaRebollo Sanz have assembled from multiple public sources. More information on these regional policies may be foundin García Pérez and Rebollo Sanz (2009). 9 The scope for manipulation on the part of the …rm aimed at obtaining the subsidy without any real changes inemployment is limited: only workers who have held a temporary contract within the same …rm (or were unemployed)for a certain period of time, usually at least a year, are eligible for new permanent contract creations. 10 Sometimes it was not clear what the maximum Euro value could be (e.g. Valencia in 1998-2000 o¤ered subsidiesas percentages of payroll tax). For these region-years I recorded a missing value. In my empirical analysis I alsodid a robustness check imputing values from total wage bill information that I have and the results were similar.Given that such imputation has to rely on additional assumptions, I opted to exclude such region-years from themain analysis. 11
  12. 12. amounts, depending on worker’ exogenous characteristics (such as gender). Such policy implemen- station helps me to construct a valid instrument. Firms were a¤ected by the policies di¤erentially depending both on the statutory amount ofthe subsidy in their region and on how many eligible temporary workers these …rms had accordingto that particular region’ criteria. To exemplify the source of identi…cation, let’ consider, for s sexample, a …rm located in Baleares autonomous region. In 2000 such …rm was eligible to get aone-time 1653 Euro subsidy for every female worker it converted from temporary to permanentcontract. But if the …rm did not employ women on temporary contracts, this subsidy would nota¤ect its proportion of temporary workers. The intuition behind the identi…cation strategy can be further illustrated by the similarity with adi¤erence-in-di¤erences approach. A given increase in the statutory subsidy amount brings a largerincrease in incentives to substitute away ‡exible contracts to …rms that employ more workers thatare eligible for subsidization (women in the above example). The e¤ect of the reduction of ‡exibleemployment can then be estimated by comparing capital structures across …rms that have highand low eligibility to substitute ‡exible contracts. Forasmuch as the cross-sectional variation in theproportion of eligible workers is driven by predetermined …rm characteristics, their potential directe¤ect on capital structure can be controlled for with …rm …xed e¤ects. At the same time, region-year …xed e¤ects capture all time-series variation in temporary employment within regions, whichcould be related to the relative size of regional budgets and corresponding governmental choices ofsubsidy amounts, as well as region unemployment rates and other macroeconomic conditions. The identi…cation assumption of such a test is that the remaining variation is not correlatedwith things such as …rm’ investment opportunity set, …nancial constraints or other …rm-speci…c sshocks. Importantly, this also means that the actual amounts of subsidies received by …rms wouldnot constitute a valid instrument, since …rms may endogenously self-select into participating in theregional subsidy program depending on their current unobservable characteristics. The expectedamount of subsidy that a given …rm in a given region was eligible to receive in a given year, on theother hand, is by construction unrelated to …rm’ current conditions.11 In other words, one can use sthe following expected subsidy amount to predict the shift in …rm’ demand for temporary labor: s X T ExpectedSubsidyit = wig Subsidygrt , (2) g 11 A similar dichotomy between expected and actual values is present in Paravisini (2008) who studies the e¤ectof bank …nancial constraints on lending: although actual amounts of external bank …nancing are endogenous, theexpected ones can be used as a valid instrument for bank sources of capital. 12
  13. 13. where Subsidygrt is the maximum statutory subsidy allowed by the government in region r inyear t for a worker of gender g 2 ffemale; maleg12 , and wig is the …rm-speci…c proportions of Ttemporary workers by gender (which is held constant at the year the …rm enters the data to avoidany endogenous gender substitution; that year is subsequently dropped from the estimation)13 . Ideally, I would like to use the …rm-speci…c workforce composition by gender, however thedata allow me to observe only the overall proportion of temporary workers, which is already animprovement upon many available datasets. In order to overcome this data limitation, I use theindustry-speci…c gender intensities to proxy for …rm-speci…c gender intensities. These industry-speci…c gender intensities, as summarized in Table 3, provide a considerable variation. For example,more than three quarters of all employees in the "Apparel" industry are female, while only around5% in the "Other transport equipment" industry. These industry ratios are quite stable over time,but in order to mitigate any endogeneity concerns arising from the di¤erent eligibility criteria ofsubsidies and possible gender substitution within industries, in the empirical analysis they are kept…xed at the pre-sample, 1993, year. Furthermore, the potential di¤erences in capital structuresacross industries due to their intrinsic gender compositions will be …ltered out by the industry …xede¤ect (subsumed by the …rm …xed e¤ect in all speci…cations). The …nal instrument for employment ‡exibility is given by the lagged value of14 X T ExpectedSubsidyit = wi wsg Subsidygrt , (3) gwhere wsg is the industry-speci…c use of female and male employees, …xed at the pre-sample year, Tand wi is the …rm-speci…c proportion of temporary workers at the year it enters the data (thatyear is subsequently dropped from the estimation). Finally, I also express the subsidy amount inreal 2006 Euros by de‡ating it using the industry-level producer price index. This instrument calculates the expected total real Euro value of subsidies that a given …rmwould receive if it converted all of its temporary workers into permanent ones, per employee, andcan be further described as the expected wage bill reduction per employee. As summarized in Table 12 The maximum statutory subsidy amounts, Subsidygrt , are listed in Table 1 and their summary statistics areprovided in Table 2 under "Maximum Statutory Subsidy per Eligible Worker". 13 Given that subsidies were also di¤erentiated by age of the worker in some regions, I have experimented with X Tone more layer of worker heterogeneity – age: ExpectedSubsidyit = wiag Subsidyagrt , where workers are also agcharacterized by their age cohort a 2 fless than 25; 25 to 30; 30 to 40; 40 to 45; 45 to 50; above 50g. The results ofthe estimation were similar both qualitatively and quantitatively. 14 The subsidy is either received in the year of the actual conversion, or it reduces the tax burden paid in the nextyear. Therefore, there is no presumption on whether the lagged or contemporaneous value should be used. Thelagged value, however, turns out to be more signi…cant in the reduced form estimation. 13
  14. 14. 2 under "Expected Subsidy per Employee" this expected per-employee wage reduction amounted,on average, to 816 Euro. Although this variable may appear to implicitly assume that all eligibleworkers are converted, this does not have to be the case in reality for the instrument to work, sincethe instrument can also be interpreted in the intention-to-treat framework. In fact, as mentionedabove, it would not be possible to use actual subsidy amounts received as an instrument dueto endogenous participation in the government program, potentially correlated with …rm-speci…cshocks (e.g. related to …nancial health of the company). The expected subsidy, on the otherhand, is arguably exogenous for the …rm since it combines predetermined …rm eligibility, de…nedby its pre-existing practices and the intrinsic characteristics of its industry, with the exogenousgovernment interventions, making it uncorrelated to current …rm-speci…c shocks (conditional onthe unobserved heterogeneity and the region’ choice of policies over time), thereby validating its suse as an instrument. The regional level variable, Subsidygrt , has been used in the literature on temporary employmentin other contexts to instrument for the worker’ probability of being converted into a permanent semployee on the worker-level data (Fernández-Kranz et al., 2010, Barceló and Villanueva, 2010).To the best of my knowledge my paper is the …rst one to construct the …rm-level subsidy from theregional data and use it as an instrument for the overall use of temporary contracts within a …rm.2.6 Data Description and Variables De…nitionThe results in this paper are based on three sets of data. I combine …rm-level data with regional dataon subsidies promoting permanent employment contracts with industry-level data on the gendercomposition of workforce. The …rm-level data come from the Encuesta sobre Estrategias Empresariales (ESEE) and spanthe years from 1994 to 2006. This is a panel dataset of Spanish manufacturing …rms collected bythe Fundación SEPI (a non-government organization) and the Spanish Ministry of Industry. It isdesigned to be representative of the population of Spanish manufacturing …rms and includes onaverage about 1700 …rms per year. The response rate in the survey is 80% to 100% across yearsand, when …rms disappear over time due to attrition, new …rms are re-sampled to ensure the panelremains representative.15 The dataset contains information on both private and public …rms. 14% of …rms that enterthe data with more than 200 employees will at some point trade on an exchange. Among smaller 15 Details on the survey characteristics and data access guidelines can be obtained athttp://www.fundacionsepi.es/esee/sp/svariables/indice.asp. 14
  15. 15. …rms this percentage is less than 1%. Firms in the sample represent all 17 regions (autonomouscommunities) and 2-digit NACE industries. Following the literature, I use the ration of otal debt to total assets as a measure of leverage.Total Debt is de…ned as the sum of short-term and long-term liabilities, while Total Assets is thebook value of assets, also equal to the sum of Total Debt and Book Equity. As reported in Table2, around 57% of …rm …nancing comes from debt. Although the survey is anonymous and the datacannot be matched to market values of equity, this does not pose a problem given that the vastmajority of …rms are private for whom such data would not exist by de…nition. This is a unique dataset in that, on top of the basic balance sheet information and the totalnumber of employees, it contains the information on the composition of contract arrangements withemployees. In particular, I can directly measure the …rm-level ‡exibility of labor force over time bythe proportion of workers on temporary contracts (measured at the end of the year). As can be seenfrom Table 2, 269 employees work in an average …rm, and 24% of them have temporary contractsin the year the …rm enters the data (this percentage is lower in subsequent years, in particulardue to the subsidies promoting permanent employment implemented by the government). Firmsize, measured as the natural logarithm of …rm’ real sales, is equal to 16, which corresponds to sapproximately 8.8mln in real 2006 Euros. Average pro…tability, measured by …rm’ operating pro…t smargin (which is de…ned as the ratio of sales net of purchases and labor expenses to sales), is equal to23%. To proxy for growth opportunities I also measure research and development intensity de…nedas the ratio of R&D expenditures over sales. These variables are typically found to be determinantsof capital structure choice (Titman and Wessels, 1988) and will be used as …rm control variablesin some of the analysis. Some speci…cations will also include tangibility, measured by the share ofgross buildings and land in total assets16 , and average wage, measured as the total wage bill peremployee in real 2006 Euros, as control variables. All …rms report the location of their industrial plants and I use the region of the …rst plant tomerge …rm-level data with the data on regional subsidies promoting permanent employment. Giventhat 85% of …rms have just one plant and additional 6% of …rms have two plants with the two mainplants in the same region, this constitutes the exact merge for the majority of …rms (the results 16 Unfortunately, the survey does not measure the value of depreciation and amortization separately for di¤erenttypes of assets, but rather records its total value. This prevents me from constructing a somewhat neater measure oftangibilitity, such as net …xed assets over total assets. In speci…cations with tangibility I opted to de…ne it in termsof buildings and land since these typically do not lose their collateral value when depreciated and they are morelikely than equipment to be redeployable, which is essentially what matters for the amount of debt. The results inthe paper are, however, robust to allocating all depreciation to all tangible assets, as well as proportionally to grosstangible and intangible assets. 15
  16. 16. are also robust to estimating everything using the sample of these …rms). Table 2 also reportsthe average values of maximum statutory subsidy amounts per eligible worker (i.e. per each newpermanent contract created), as well as expected subsidy per employee, which are equal to 3523 and816, respectively. It should be noted that these averages correspond to all years from 1994 to 2006,so also include a period when regional subsidies were equal to zero. When conditioning on the periodafter 1997, these averages become 4288 and 994, respectively. This means that given the averageyearly wage of about 29 thousand Euros, the one-time subsidy covers about 52 4288=28790 = 8weeks of salary for the worker that was actually converted from temporary to permanent, roughlycorresponding to the numbers reported in García Pérez and Rebollo Sanz (2009). Finally, I use the data on intensities of the use of female and male employees in di¤erent man-ufacturing industries, as provided by the Encuesta de Población Activa, and merge them to the…rm-level data at the industry level. These gender intensities, measured as of the 4th quarter of1993, are listed in Table 3 and have been discussed in detail in Section 2.5.3 The E¤ect of Employment Flexibility on Capital Struc- tureIn this section I apply the estimation strategy outlined in Section 2 to study the e¤ect of employment‡exibility on capital structure. I defer evidence on the exact mechanism driving these results untilSection 4.3.1 Main ResultsBefore turning to the formal analysis, I …rst use ESEE data to explore the relationship between theproportion of temporary employees and capital structure graphically. Figure 1 plots the averages ofthe two variables across di¤erent industries for the period from 1994 to 2006. As can be seen fromthis …gure, the industries that employ larger proportions of temporary workers, such as Leather andFootwear or Timber, are also characterized by higher ratios of debt to assets than industries thatemploy relatively lower proportions of temporary labor force, such as Chemicals or Beverages. Figure 2 plots the time-series relationship between the two variables and again a positive re-lationship can be deduced. One can notice a striking drop in the use of temporary labor forcestarting approximately in 1997. One of the possible explanations for this drop is the country-wideimplementation of subsidies promoting the use of permanent employment contracts, as described in 16
  17. 17. Section 2. Interestingly, consistent with my hypothesis, the drop in temporary employment is alsoaccompanied by a fall in average debt to assets ratio. Although these …gures provide interesting visual correlations, there may be various unobservedcharacteristics of industries or common macroeconomic factors that may show up as a positiverelationship between temporary labor force and debt …nancing either across industries or acrossyears. Therefore, I now turn to a more systematic …rm-level analysis by employing the panelstructure of the dataset and the exogenous variation arising from government labor programs toestimate the causal e¤ect of the use of temporary employment contracts on capital structure, using…xed-e¤ects and instrumental variable approaches. Table 4 shows the OLS and IV-2SLS estimates of the coe¢ cient of interest in di¤erent speci…ca-tions. The standard errors throughout the paper are two-way clustered at the …rm and region-yearlevels, so that all statistics are robust to heteroskedasticity and arbitrary within-…rm and within-region-year correlation (which could potentially arise from the same statutory subsidy amounts…rms in general face in a given region-year).17 The coe¢ cient in column 1 means that a one per-centage point change in the proportion of workers on temporary contracts is associated with a 0.06percentage points higher leverage ratio. I have also calculated the average within-…rm standarddeviation of proportion of temporary workers, which equals 0.11 in my data. Therefore, when agiven …rm changes its proportion of temporary workers by 1 standard deviation, it also increases itsleverage by 0.66 percentage points. This speci…cation accounts for time-invariant …rm heterogeneityand region-year …xed e¤ects, so that the results illustrate within-…rm di¤erences in leverage andare not driven by region-speci…c variables, such as credit abundance across and within regions ormacroeconomic e¤ects. Column 2 adds several …rm-level control variables that have been identi…ed in the literature asdeterminants of capital structure: size, measured by log of …rm’ real sales, average pro…tability, sand share of R&D expenses over sales. Both the magnitude and the signi…cance of the coe¢ cientof interest stay similar, so that the observed di¤erences in debt ratios cannot be explained by …rmsbecoming larger and more pro…table over time or changes in their growth opportunities.18 Althoughthese speci…cations already pick up all time-invariant di¤erences across …rms, as well as the e¤ectsof size, pro…tability and growth, employment composition and capital structure are still likely to be 17 The two-way clustered standard errors, proposed by Cameron, Gelbach and Miller (2006) and Thompson (2011),were obtained using the Scha¤er (2010) xtivreg2 command in STATA. 18 I have also tried including accumulated pro…ts during the previous 3 years, since …rms are likely to pay out debtwhen they have had a positive shock to their cash‡ ow. The results were similar. I opted to exclude this variablefrom further analysis to keep more years of observations. 17
  18. 18. chosen endogenously due to within-…rm changes in investment opportunity set, …nancial constraintsor demand shocks. To identify the causal e¤ect I exploit the exogenous variation induced by thedi¤erential implementation of government labor policies and report with the IV-2SLS results incolumns 3 to 6.19 Columns 3 and 5 report the results of regressing proportion of temporary employment on theexpected subsidy instrument, …rm and region-year …xed e¤ects, and additional …rm-level controls(in column 5). These regressions correspond to the …rst stage of the IV-2SLS estimation of (1) andare given by 0 T empit = rt + ExpectedSubsidyit 1 + Xit 1 + i + it ; (4) The estimate of in column 3 is signi…cant at 1% level and shows that an expected per-workersubsidy of 1000 Euro incentivizes a …rm to reduce its proportion of temporary workers by 3.8percentage points. Since conditional on the …rm and region-year …xed e¤ects the variation inthe expected subsidy amount is reasonably exogenous, the change in employment ‡exibility canbe entirely attributed to changes in the extent of government incentives to promote less ‡exibleemployment contracts. For each …rst-stage speci…cation throughout the paper I also report theweak identi…cation Kleibergen and Paap (2006) F-statistic. It exceeds the Stock and Yogo (2002)weak identi…cation critical value of 16.38 (for 5% maximal size distortion for 1 instrument and 1endogenous regressor) in all speci…cations, suggesting that my instrument is also strong.20 Although the use of predetermined …rm-level controls is not necessary with this identi…cationstrategy, it helps in corroborating the exclusion restriction. In column 5, I add a range of …rm-level control variables to account for time-varying …rm-speci…c shocks (the model is even furthersaturated in the robustness tests). The estimate of remains similar and is still signi…cant at 1%level, suggesting that the instrument is uncorrelated with the range of included variables. The coe¢ cients in columns 4 and 6 report the corresponding second-stage IV-2SLS estimatesof . As outlined in Section 2.3, the presence of unobserved …rm-speci…c shocks to investmentopportunity set, …nancial constraints or demand shocks, that can be correlated with employment 19 Since I am able to directly observe the proportion of workers on temporary contracts and can use governmentsubsidies to construct an instrumental variable, I do not have to rely on purely reduced form estimation (e.g. debton employment laws). For completeness, I report the reduced-form regression results (debt on subsidy) for allspeci…cations in Tables 4 and 5 in Appendix Table 1. They all have predicted coe¢ cient signs and are signi…cantat conventional levels. The estimates suggest that a 1000 Euro expected per-worker subsidy leads to 0.42-0.77percentage point reduction in the debt to assets ratio. 20 Stock and Yogo (2002) critical values are derived under the assumption of homoskedasticity and no autocorrela-tion, so that their comparison to Kleibergen and Paap (2006) F-statistic, which is robust to heteroskedasticity andwithin-cluster correlation, should be interpreted with caution, as suggested by Baum, Scha¤er, and Stillman (2007). 18
  19. 19. composition and choice of …nancing, would bias the OLS estimate of downwards. Indeed, I …ndthat the magnitude of the IV-2SLS estimates is larger. The preferred estimate of in column 4(0.149 with a standard deviation of 0.0582) means that a one standard deviation increase in theproportion of ‡exible employment leads to a 1.64 percentage points higher leverage ratio. This resultis economically and statistically signi…cant. In particular, such magnitude suggests that prohibitingan average …rm from hiring temporary employees (i.e. reducing its proportion from the averageof 23.9% to 0%) would lead to a 3.6 percentage points reduction in debt level, or about 6.3% ofthe average. Finally, in column 6 I add …rm control variables, and the resulting estimate is alsostatistically signi…cant and similar in magnitude.3.2 Robustness to Additional Speci…cationsTo corroborate the exclusion restriction further and show robustness of the results I saturate myempirical speci…cation even more in Table 5. Columns 1 and 2 introduce an additional controlvariable of tangibility, measured by the share of gross land and buildings in total assets. If theinstrument were in fact picking up some of the …rm-level time-varying shocks related to the natureof …rm’ assets, then we would not observe a signi…cant and large e¤ect in the …rst stage of this sspeci…cation. The coe¢ cient in column 1 suggests that the …rst-stage results are similar, implyingthat the instrument is orthogonal to the set of included controls. Furthermore, the magnitude ofthe causal e¤ect in column 2 is also unchanged. This means that the results are not driven by …rmschanging their capital structure as a response to changes in the nature of their assets over time. Columns 3 and 4 include average wage, de…ned as total wage bill per employee, as an additionalcontrol variable. This speci…cation allows to rule out the possibility that …rms respond to …rm-speci…c time-varying shocks in wages (e.g. arising from new bargaining agreement at the industrylevel) by changing contract composition and adjusting capital structure, and that the instrumentaccidentally picks up this variation. The results indicate that this is not the case either. Both the…rst-stage and second-stage results are the same in both magnitude and signi…cance with the mainresults, providing evidence against wage e¤ects. In columns 5 and 6, I take yet another approach and saturate the model with region-industry-year, rather than simple region-year, …xed e¤ects. This speci…cation provides a very tight iden-ti…cation. In particular, it allows to control for potential time-varying industry-level lobby powerthat could a¤ect the amounts and timing of subsidy introduction. It also captures the non-uniformdistribution of industries across regions and allows for a separate non-parametric trend in capital 19
  20. 20. structure for each industry-region combination. The coe¢ cient of interest can still be identi…edbecause even within the same region-industry-year …rms with higher original (predetermined) pro-portions of temporary workers on average bene…t more from the same statutory level of subsidies.The coe¢ cient of interest in this most saturated speci…cation remains similar in magnitude and issigni…cant at 5% level. Finally, in columns 7 and 8 I explore the subset of …rms that are in the data in 1994. This allowsto hedge against a potential concern that the results are driven by …rms that were sampled by ESEEin later years when the government policies have already been announced or implemented (in thiscase their measured original level of temporary labor force may had already partially adjusted tothe reform). The results are robust. I …nd that even for …rms that determined their employmentpractices years in advance of government policies, the expected subsidy instrument is a good predic-tor of post-reform employment ‡exibility. Furthermore, employment ‡exibility signi…cantly a¤ectscapital structure and the magnitude of the coe¢ cient of interest is similar to previous speci…cations.3.3 The Role of CashOne important consideration to be analyzed is the observation that a subsidy promoting permanentemployment does not only in‡uence the composition of the labor force per se, but also provides the…rm with a cash in‡ (or reduces its cash out‡ ow ow). Firms may potentially use this cash to raiseeven more debt (Blanchard et al, 1994) or pay out the existing debt (Bates, 2005). In this respect,the exclusion restriction of the instrument would not be satis…ed. Given that the estimated e¤ectof proportion of temporary workers on debt levels is positive, we should be concerned only if cashfrom the subsidy is used to pay out debt (this will bias the estimated e¤ect upwards; if cash is usedto raise more debt instead, then the estimated e¤ect is biased downwards, which means that thetrue e¤ect of temporary employment on debt levels is even stronger). Interestingly, one would not be able to refute these concerns by looking at net debt levels, (TotalDebt - Cash) / Total Assets, as the dependent variable, because the cash from the subsidy wouldnot necessarily have to stay on the balance sheet under the cash item, but could immediately beused to reduce the total debt value. However, there are several other ways to look at whether cashin‡ plays a role in this particular situation and quantify its e¤ect if it does. ow First of all, I examine the e¤ect of temporary workers on capital structure for …rms that can beconsidered relatively cash-abundant. For these …rms it is unlikely that a marginal increase in cashfrom the subsidy could trigger paying out debt, since they could have done so without receiving 20
  21. 21. the subsidy, if they wanted to. Therefore, …nding a signi…cant e¤ect of temporary workers oncapital structure for …rms that can be characterized by relatively high cash holdings, would provideevidence against the cash story. Although ESEE does not contain a separate entry for cash and cash equivalents, I can use severalproxies for cash abundance based on the pro…t and loss items. I calculate an approximation to cash‡ as sales plus other income (e.g. from leasing and services provided) less material, personnel, owand other costs (e.g. advertising, R&D and external services), less 35% corporate tax rate, less netcapital expenditures (which is equal to purchases minus sales of tangible assets). Then I classify…rms as being relatively cash-abundant based on this measure and estimate speci…cation (1) forsubsamples of these …rms. The results are reported in Table 6, where the even- and odd-numbered columns correspond tospeci…cations with and without …rm-level controls, respectively. In columns 1 and 2 I de…ne …rmsas being rich in cash if their cash ‡ was above the industry median two years in advance. This owcorresponds to the time structure used throughout the paper when …rms …rst receive the subsidy,then adjust their labor force, and then change their capital structure. Since …rms are classi…edrelative to the yearly industry median, the results are not driven by accidentally capturing wholeindustries that got positively a¤ected by shocks in a particular year. The coe¢ cient in column1 suggests that among cash-abundant …rms a ten-percentage-point decrease in the proportion oftemporary workers leads to 2.54 percentage points lower debt ratio. This coe¢ cient is statisticallysigni…cant at 1% signi…cance level and robust to including …rm-level controls. I do a series of robustness checks by considering alternative de…nitions of cash-abundant …rms.Columns 3 and 4 classify a …rm as being rich in cash if its cash ‡ is above the industry median in owthe current year, i.e. when debt adjustment takes place. The results are very similar. In columns 5and 6 I use the ratio of cash ‡ to total assets as the measure of cash abundance (rede…ning the owindustry median accordingly) to make sure that the results are not driven by larger or smaller …rmsoverall. Finally, in columns 7 and 8 I classify …rms as having relatively high cash holdings if theirratio of cash ‡ows, accumulated over three years, over total assets is higher than the correspondingindustry median. This mitigates the e¤ects of transitory shocks and enables to look at …rms whichhave performed better than their peers over several years. Again, the results are very similar acrossall speci…cations. Overall, they indicate that even among cash-abundant …rms ‡exible labor forcehas a large and signi…cant e¤ect on capital structure, thereby providing evidence against …rmssimply using subsidies to pay out debt. 21
  22. 22. Another approach to look at the direct e¤ect of cash in‡ is to compare the magnitudes of owchanges in debt to the magnitudes of cash in‡ from subsidies. To do that one would need to owobserve the actual subsidies received at the …rm level, however, I am not aware of a dataset thatwould contain this information even at the region or industry level. Therefore, I do a back-of-the-envelope calculation of these magnitudes to quantify how much of the total change in debt levels,that is implied by my main results, can be attributed to purely repaying out using cash receivedfrom subsidies. The average within-…rm change in the percentage of temporary labor force is equal to 1.07percentage points per year in my data. Given the average size of the total labor force (269 fromTable 2) and the maximum subsidy for each eligible worker (3158 from Table 2), this amounts toreceiving 0:0107 269 3158 = 9090 Euro per year in subsidies21 . At the same time, my preferredestimate of 0.149 (Table 4 column 4) implies that such change in temporary labor force leads to1.07*0.149 = 0.16 percentage points change in debt-to-assets ratio per year or, given the averagetotal assets of 57.7 million Euro (from Table 2), to 0:0107 0:149 57:7 106 = 91991 Euro averagechange in debt level per year. These two numbers suggest that about 9090/91991 = 9.9% of thefound causal e¤ect can be due to cash considerations, i.e. the true causal coe¢ cient is not 0.149,but about 0.134. As a robustness check I also perform this calculation with medians instead of means, as wellas separately by region (weighted by the number of observations per region to obtain the overallmean). These results correspond to 6.5% and 10.3% of the total causal e¤ect being due to cashconsiderations, respectively, indicating that the true magnitude of the causal e¤ect of ‡exible laborforce on capital structure is potentially only slightly lower than the one reported in the main resultsof the paper.4 Flexible Employment Contracts Reduce Operating Lever- age and Default RiskThe results so far provide evidence of a positive causal relationship between the use of temporaryemployment contracts and …nancial leverage, but the exact mechanism is not yet identi…ed. In thissection I use further analysis to …rst demonstrate that ‡exible employment reduces the operating 21 Instead of taking the product of the averages, another option would be to average the product of total laborforce and maximum subsidy per eligible worker. This would amount to receiving 0.0107*839327 = 8981 Euro peryear in subsidies. 22
  23. 23. leverage by providing the margin of adjustment when negative shocks arrive. Then I present anadditional indirect test of this mechanism by exploring the cross-sectional heterogeneity and com-paring …rms with di¤erent levels of bankruptcy costs that implicitly de…ne the relative values ofthe use of ‡exible employment in reducing default risk for di¤erent …rms. Finally, I compare …rmscross-sectionally to illustrate that adjusting capital structure upon shocks to employment ‡exibilityis associated with surviving.4.1 Temporary Employees as a Margin of AdjustmentThe underlying assumption behind interpreting the causal e¤ect of ‡exible employment on …nancialleverage as the substitution between operating and …nancial leverage is that temporary workerslower the default risk of a …rm by providing the margin of adjusting the labor force and earningswhen faced with negative shocks. Although the recent evidence across a range of European countriessuggests that temporary workers absorb a higher share of the volatility of the output (Blanchardand Landier, 2002, for France; Alonso-Borrego et al., 2005, for Spain; Kugler and Pica, 2004, forItaly), it would be useful to see whether this happens in my data as well in order to corroboratethis assumption. It should be noted that to test the assumption that temporary workers reduce operating leverageand default risk one could not simply use a measure of the overall riskiness of the company (e.g.volatility of cash ‡ or probability of going bankrupt) as the dependent variable in a regression owsimilar to (1). The reason is simple: in such a test one needs to keep everything else constant,and …nancial leverage in particular, since this assumption implies that ‡exible employment reducesoperating leverage and probability of bankruptcy for a given level of …nancial leverage. In otherwords, the realized probability of default would also necessarily re‡ the endogenous adjustment ectof capital structure, that has been shown to adjust in Section 3. In fact, if companies have a targetlevel of overall risk and trade o¤ operating and …nancial risks by adjusting …nancial leverage to thechanges in operating leverage (Mandelker and Rhee, 1984), then empirically we should see no e¤ectof ‡exible labor force on the realized probability of default (still under the restrictive assumptionthat there is no e¤ect of temporary workers on survival other than through capital structure). However, there is yet another possibility to explore whether temporary workers provide theoption of adjusting the size of the labor force, the corresponding wage bill and earnings uponarrival of negative shocks. Speci…cally, one can attempt to test whether …rms indeed …re temporaryworkers when faced with negative shocks. To do that I use ESEE to measure the current state of 23
  24. 24. the …rm’ main product market, proxying for demand shocks to its product. In particular, every syear …rms report whether the market for their good is in expansion, stable, or in recession. ThenI de…ne a dummy variable (N egativeShockit ) that equals 1 if the …rm reports that the market isin recession, and 0 otherwise. The idea of using this measure as a proxy for negative shocks relieson the observation that when the …rm’ product market is in recession, the average product of a sworker falls, so by …ring some of the temporary workers the …rm can save on the labor costs andhave a higher pro…t compared to the situation of keeping them employed. I estimate the following speci…cation: T empit = st + N egativeShockit + i + it ; (5) where st are the industry-year …xed e¤ects, N egativeShockit is the indicator variable de…nedabove, and i are …rm …xed e¤ects. The inclusion of …rm …xed e¤ects captures potential average di¤erences in assessing the state ofthe market across …rms as well as any other unobserved time-invariant heterogeneity that could berelated to the proportion of temporary workers. It is also important to include industry-year …xede¤ects in such speci…cation in order to control non-parametrically for the state of the business cyclein a given industry. This implies that N egativeShockit measures the …rm-speci…c demand shockover and above any industry-level shocks in the same year. The results of estimating (5) are presented in Table 7 column 1, while column 2 further saturatesthis speci…cation with region-industry-year …xed e¤ects. The latter identi…es this correlation verytightly, because …rm-speci…c demand shocks are now measured over and above any shocks to other…rms in the same industry in the region where the …rm is located. The coe¢ cients in both speci…ca-tions are highly statistically signi…cant and imply that when the market for …rm’ main product is sin recession, it employs a lower proportion of temporary workers. In particular, during an average…rm-speci…c negative demand condition, the proportion of workers on temporary contracts is 1.7percentage points lower than during normal demand conditions, which roughly corresponds to …ringabout one tenth of the total ‡exible labor force. This result is robust to clustering standard errorsat the industry level to account ‡exibly for the correlation of shocks within each industry, as well asto using a lagged indicator of …rm-speci…c negative demand shock instead of the contemporaneousone.22 22 In addition, I also tried including a leading indicator of N egativeShockit . This was not statistically di¤erentfrom zero, minimizing the concerns about reverse causality. 24
  25. 25. To the extent that …rms may have several product markets, measuring …rm-speci…c demandshock based on the main market can be noisy. Column 3 reports the results of estimating spec-i…cation (5) for a subsample of …rms with only one product. The results are robust. Although Ihave to admit of not having a quasi-experimental variation in the independent variable, includingthis large range of …xed e¤ects (…rm-level and region-industry-year) should capture a vast majorityof potential omitted variables that could be correlated with the independent variable. In order tofurther minimize the reverse causality concerns, I estimate speci…cation (5) for a subset of …rmsthat sell only one product and have a low share in that market (less than 5%). For these …rms, theextent to which a given …rm can a¤ect the state of its product market is very limited, making theindependent variable arguably exogenous for such …rms. The results are again robust. Overall, the results in Table 7 corroborate the assumption of temporary workers being a marginof adjustment when negative shocks arrive –the assumption that underlies the mechanism behind‡exible labor force a¤ecting capital structure.4.2 Cross-Sectional Heterogeneity in Bankruptcy CostsIf the mechanism is the one of ‡exible labor force reducing operating leverage and the incidence ofbankruptcy, then …rms with a relatively high level of bankruptcy costs should value the option to…re workers more.23 This means that for a given change in the proportion of temporary workersand the corresponding change in the probability of bankruptcy, the change in expected bankruptcycosts is higher for such …rms. Since expected bankruptcy costs matter for …nancial leverage, weshould …nd a higher causal e¤ect of ‡exible labor on corporate …nancing for …rms with higher levelsof potential bankruptcy costs. In order to examine this prediction I estimate the following equation using the instrumentalvariables approach: H L 0 Dit = rt + Highi T empit 1 + Lowi T empit 1 + Xit 2 + i + it ; (6) where Highi and Lowi are the indicator variables corresponding to …rms with high and lowbankruptcy costs24 . If indeed the mechanism is the one of temporary workers reducing operatingleverage and the probability of bankruptcy, then the causal e¤ect of ‡exible labor force on capital 23 Smith and Stulz (1985) present a similar agrument on hedging: it reduces the variability of cash ‡ows and theprobability of bankruptcy, and these reductions are valued most by …rms with higher bankruptcy costs. 24 The instruments for Highi T empit and Lowi T empit are Highi ExpectedSubsidyit 1 and LowiExpectedSubsidyit 1 . 25
  26. 26. Hstructure should be higher for …rms with high levels of bankruptcy costs, i.e. should be higher Lthan in the above speci…cation. What remains is to identify …rms with high and low levels ofbankruptcy costs. Bankruptcy costs are typically modelled as the loss of value in liquidation, when keeping the…rm alive would yield more. Williamson (1988) and Shleifer and Vishny (1992) have emphasizedthat the degree to which debt-holders can recover their assets in liquidation depends on the natureof these assets. In particular, when asset speci…city is low and assets can be easily redeployed forother purposes, the loss of value from liquidation is low and so is the bankruptcy cost of debt. Giventhat some of the least speci…c assets are buildings and land, I …rst classify …rms as having low levelof bankruptcy costs (Lowi = 1) if they have buildings or land on their balance sheets in the yearthey enter the data (to mitigate the concern of endogenous choice of assets speci…city over time).Likewise, …rms with no buildings and land are classi…ed as having high level of bankruptcy costs(Highi = 1). The results of these regressions are presented in Table 8 columns 1 to 4, with odd- and even-numbered columns corresponding to …rst- and second-stage results, respectively. Consistent withthe bankruptcy cost mechanism, the positive e¤ect of having a ‡exible workforce is pronouncedmostly within the high bankruptcy costs …rms. The coe¢ cient in column 2 means that for these…rms a one standard deviation increase in the proportion of workers on temporary contracts leadsto 3.3 percentage points higher debt ratio. Furthermore, the implied di¤erence in the coe¢ cientsbetween high and low bankruptcy cost …rms (-0.223 with a standard error of 0.128) is large andstatistically signi…cant, suggesting that …rms with high levels of bankruptcy costs are signi…cantlymore likely to adjust their capital structure in response to shocks to the ‡exibility of the labor force. Importantly, column 1 indicates that both types of …rms change their labor force compositionin response to subsidies promoting less ‡exible contract arrangements with workers. However, itis only …rms for whom this ‡exibility does matter (i.e. …rms that incur relatively high expectedbankruptcy costs) that indeed decrease their debt levels upon the reduction in the fraction of‡exible labor force. This cross-sectional comparison provides additional evidence on the mechanismbehind the e¤ect of ‡exible labor on debt …nancing being the one of reducing operating leverageand expected bankruptcy costs. Columns 3 and 4 report the results of a similar speci…cation with …rm-level control variables.The results are very similar. As a robustness check, I also reestimate speci…cation (6) for a di¤erentde…nition of high and low bankruptcy costs …rms. In order to take into account potential di¤erences 26
  27. 27. across industries in their usage of buildings and land, I now classify …rms as having high (low) levelof bankruptcy costs if they have less (more) buildings and land than the industry median in the yearthey enter the data. The results of these speci…cations are reported in columns 5 to 8 of Table 8,demonstrating similar patterns. Again, although both high and low bankruptcy cost …rms respondto subsidies by lowering proportion of ‡exible labor force (columns 5 and 7), it is only the former,i.e. the …rms for which ‡exible labor force is most valuable in reducing the expected bankruptcycosts, that adjust their debt levels accordingly.4.3 Firms in LiquidationThe mechanism of operating to …nancial leverage substitution implies that if …rms want to keepthe overall riskiness of the …rm constant, then they should match changes in ‡exible labor forceto changes in indebtedness. Section 3 has shown that indeed on average …rms reduce their debtlevels following reductions in employment ‡exibility (e.g. conversions of employee contracts fromtemporary to permanent). It would be also interesting to explore what happens to …rms that donot adjust their capital structure to the reductions in the proportion of temporary employment.Theoretically, these …rms should become riskier overall and hence be more likely to liquidate. Although it is not possible to identify the subsample of …rms that do not adjust their debtlevels and then explore their outcomes (and exit rates in particular), one possible alternative is tocompare the adjustment of debt to ‡exible employment of …rms that liquidate to the one of …rmsthat survive. To do that I de…ne an indicator variable Exiti that equals 1 if the …rm exits the databy the end of the sample (due to liquidation or switching to non-manufacturing activity), and 0otherwise. Similarly, Stayi is de…ned as an indicator variable that equals 1 if the …rm is still in thedata by the end of the sample (i.e. for each …rm Stayi + Exiti = 1). Then I estimate the following equation using the instrumental variables approach: S E 0 Dit = rt + Stayi T empit 1 + Exiti T empit 1 + Xit 2 + i + it ; (7) where Stayi and Exiti are the indicator variables as de…ned above25 . The results are reported in Table 9, with odd- and even-numbered columns corresponding to…rst- and second-stage results, respectively. Interestingly, column 1 indicates that both exitersand survivors respond to government subsidy program by reducing the proportion of workers on 25 The instruments for Stayi T empit and Exiti T empit are Stayi ExpectedSubsidyit 1 and ExitiExpectedSubsidyit 1 . 27
  28. 28. temporary contracts. However, as column 2 shows, the e¤ect of employment ‡exibility on capitalstructure is only present among survivors. I.e., although …rms that eventually liquidate do changethe composition of labor force following subsidized contract conversions, they do not adjust theirindebtedness, unlike …rms that survive. The implied di¤erence in the e¤ect across the two subsets of…rms is large and statistically signi…cant26 . The results are qualitatively and quantitatively similarto the results in columns 3 and 4 that include …rm-level controls. Given that …rms that enter the data later in the sample are statistically more likely to surviveby any given date (and by the end of 2006 as they are currently measured), as a robustness check, Ireplicate the above analysis focusing on the subset of …rms that are in the data in 1994. The resultsof these speci…cations are reported in columns 5 to 8 and are again very similar. One certainly has to be very careful when interpreting these empirical results, and I do notattempt to claim that …rms liquidate precisely because they have not adjusted their …nancial lever-age after changing the proportion of temporary workers and the corresponding level of operatingleverage. There can be unobserved time-varying reasons that explain why some …rms could notor did not want to adjust …nancial leverage and that could at the same time be correlated withthe decision to liquidate. (The results of previous sections are of course still valid irrespective ofsuch reasons). Still, I think these results provide an interesting piece of suggestive evidence thatcan be used to complement the main analysis, and trigger further research on the channels of howemployment ‡exibility may a¤ect growth and survival of …rms in the long run.5 Concluding RemarksThis paper considers how a …rm’ use of di¤erent contractual arrangements with labor a¤ects its scapital structure. The di¤erence in …ring costs across employment contracts provides for a di¤erent‡exibility upon realization of …rm-speci…c shocks. Hiring workers on temporary contracts allows…rms to reduce the default risk on the margin and support higher levels of …nancial leverage, whichcan be bene…cial from tax shield or other considerations. Employment ‡exibility thus reducesoperating leverage which in turn promotes …nancial leverage. Indeed, I …nd that …rms that have ahigher proportion of workers on more ‡exible contracts have more debt. I exploit the di¤erential implementation of government policies promoting the replacement oftemporary contracts by permanent ones to identify the e¤ect of interest. This exogenous inter- 26 The non-signi…cant e¤ect of exiters (-0.0596 with a standard error of 0.120) is unlikely to be driven by the lackof power, since about one third of the …rms are classi…ed as exiters. 28
  29. 29. temporal and cross-regional variation allows me to construct a valid instrument for the …rm-levelemployment ‡exibility. At the same time the panel structure of the dataset enables me to controlfor any unobserved time-invariant characteristics of the …rm that may in‡uence its …nancing policy.The combination of these identi…cation strategies allows to evaluate the causal e¤ect of …rm’ use sof temporary employment contracts on its capital structure in a quasi-experimental setting. This e¤ect is economically large and, as expected, is more prevalent among the high bankruptcycost …rms, which should value the option to …re workers upon arrival of negative shocks more. Inparticular, a thought experiment of completely prohibiting an average …rm from o¤ering temporaryemployment contracts would suggest that such a …rm should reduce its debt level by 3.6 percentagepoints, which is about 6.3% of the average debt level across …rms. These empirical results imply that labor policies that promote more job security among workersby reducing the ‡exibility of …rm’ employment have signi…cant implications for capital structure of s…rms. They suggest that at least part of the cross-country di¤erences in …rm indebtedness could bedriven by the characteristics of the labor markets. This is an important evidence to provide in thelight of recent reforms across European countries that aim to promote more stable employment. Ifsuch reforms are made mandatory, then one needs to think carefully how this would a¤ect …rms. Inparticular, if …rms are constrained and cannot adjust their debt levels quickly to match the changesin employment structure, they may be forced to exit with potential consequences on the allocativee¢ ciency. This in turn opens up the directions for future research on the channels through whichtemporary employment may a¤ect …rms’real decisions, long-run growth and survival. 29
  30. 30. References [1] Abowd, John M., 1989, The E¤ect of Wage Bargains on the Stock Market Value of the Firm, American Economic Review 79(4), 774-800. [2] Acharya, Viral V, Ramin P. Baghai, and Krishnamurthy V. Subramanian, 2010, Labor Laws and Innovation, NBER Working Paper No. 16484. [3] Alonso-Borrego, César, Jesús Fernández-Villaverde, and José E. Galdón-Sánchez, 2005, Eval- uating Labor Market Reforms: A General Equilibrium Approach, NBER Working Paper No. 11519. [4] Barceló, Cristina, and Ernesto Villanueva, 2010, The response of household wealth to the risk of losing the job: evidence from di¤erences in …ring costs, mimeo. [5] Bates, Thomas W., 2005, Asset Sales, Investment Opportunities, and the Use of Proceeds, Journal of Finance 60(1), 105-135. [6] Baum, Christopher F , Mark E. Scha¤er, and Steven Stillman, 2007, Enhanced routines for instrumental variables/GMM estimation and testing, Boston College Working Papers in Eco- nomics 667. [7] Besley, Timothy, and Robin Burgess, 2004, Can Labor Regulation Hinder Economic Perfor- mance? Evidence from India, Quarterly Journal of Economics 119, 91-134. [8] Berk, Jonathan B., Richard Stanton, and Josef Zechner, 2010, Human capital, bankruptcy, and capital structure, Journal of Finance 65(3), 891–926. [9] Blanchard, Olivier, and Augustin Landier, 2002, The perverse e¤ects of partial labour market reform: …xed-term contracts in France, Economic Journal 112(480), F214–F244.[10] Blanchard, Olivier, Florencio Lopez-de-SiIanes, and Andrei Shleifer,1994, "What do …rms do with cash windfalls?", Journal of Financial Economics 36(3), 337-360.[11] Butt Jaggia, Priscilla, and Anjan V. Thakor, 1994, Firm-speci…c human capital and optimal capital structure, International Economic Review 35(2), 283–308.[12] Caggese, Andrea, and Vicente Cueat, 2008, Financing Constraints and Fixed-Term Employ- n ment Contracts, Economic Journal 118, 2013–2046. 30
  31. 31. [13] Cameron, Colin, Jonah B. Gelbach, and Douglas L. Miller, 2006, Robust Inference with Multi- way Clustering, NBER Technical Working Paper No. 327.[14] Chen, Huafeng, Marcin Kacperczyk and Hernan Ortiz-Molina, 2009, Labor unions, operating ‡exibility, and the cost of equity, mimeo.[15] DeAngelo, Harry, and Ronald W. Masulis, 1980, Optimal capital structure under corporate and personal taxation, Journal of Financial Economics 8(1), 3–29.[16] Dolado, Juan J. , Carlos García-Serrano, and Juan F. Jimeno, 2002, Drawing lessons from the boom of temporary jobs in Spain, Economic Journal 112(480), F270–F295.[17] García Pérez, J. Ignacio, and Yolanda F. Rebollo Sanz, 2009, Do wage subsidies a¤ect the subsequent employment stability of permanent workers?: the case of Spain, Moneda y Crédito, 228, 65-103.[18] Fernández-Kranz, Daniel, Aitor Lacuesta, and Núria Rodríguez-Planas, 2010, Chutes and lad- ders: dual tracks and the motherhood dip, IZA Discussion Paper No. 5403.[19] Izquierdo Mario, Esther Moral, and Alberto Urtasun, 2003, Collective bargaining in Spain: An individual data analysis, Documento Ocasional #0302, Banco de Espaea Occasional Paper n Series.[20] Jimeno, Juan F., and Luis Toharia, 1994, Unemployment and labour market ‡exibility: Spain (International Labour O¢ ce, Geneva).[21] Hanka, Gordon, 1998, Debt and the terms of employment, Journal of Financial Economics 48, 245-282.[22] Hirsch, Barry T., 1991, Union Coverage and Pro…tability among U.S. Firms, Review of Eco- nomics and Statistics 73, 69-77.[23] Kugler, Adriana and Giovanni Pica, 2004, E¤ects of employment protection and product mar- ket regulations on the Italian labour market, in (J. Messina, C. Michelacci, J. Turunen, and G. Zoega, eds.), Labour Market Adjustments in Europe, 107–137 (Cheltenham: Edward Elgar).[24] MacKay, Peter, 2003, Real ‡exibility and …nancial structure: An empirical analysis, Review of Financial Studies 16(4), 1131-1165. 31
  32. 32. [25] Mandelker, Gershon N., and S. Ghon Rhee, 1984, The impact of the degrees of operating and …nancial leverage on systematic risk of common stock, Journal of Financial and Quantitative Analysis 19(1), 45–57.[26] Matsa, David A., 2010, Capital structure as a strategic variable: evidence from collective bargaining, Journal of Finance 65(3), 1197–1232.[27] Mauer, David C., and Alexander J. Triantis, 1994, Interactions of corporate …nancing and investment decisions: A dynamic framework, Journal of Finance 49(4), 1253-1277.[28] OECD Employment Outlook 2002.[29] Pagano, Marco, and Paolo Volpin, 2008, Labor and …nance, mimeo.[30] Paravisini, Daniel, 2008, Local Bank Financial Constraints and Firm Access to External Fi- nance, Journal of Finance 63(5), 2161-2193.[31] Perotti, Enrico C., and Kathryn E. Spier, 1993, Capital structure as a bargaining tool: the role of leverage in contract renegotiation, American Economic Review 83(5), 1131–1141.[32] Petersen, Mitchell A, 1994, Cash ‡ variability and …rm’ pension choice A role for operating ow s leverage, Journal of Financial Economics 36, 361-383.[33] Ruback, Richard, and Martin Zimmerman, 1984, Unionization and Pro…tability: Evidence from the Capital Market, Journal of Political Economy 92, 1134-1157.[34] Scha¤er, Mark E., 2010. xtivreg2: Stata module to perform extended IV/2SLS, GMM and AC/HAC, LIML and k-class regression for panel data models. http://ideas.repec.org/c/boc/bocode/s456501.html[35] Simintzi, Elena, Vikrant Vig, and Paolo Volpin, 2010, Labor and capital: is debt a bargaining tool?, mimeo.[36] Shleifer, Andrei and Rovert W. Vishny, 1992, Liquidation Values and Debt Capacity: A Market Equilibrium Approach, Journal of Finance 47(4), 1343-1366.[37] Smith, Cli¤ord W. and Rene M. Stulz, 1985, The Determinants of Firms’ Hedging Policies, Journal of Financial and Quantitative Analysis 20(4), 391-405. 32
  33. 33. [38] Stock, James H., and Motohiro Yogo, 2002, Testing for weak instruments in linear IV regression, NBER Technical Working Paper No. 284.[39] Stulz, René M, 1990, Managerial discretion and optimal …nancing policies, Journal of Financial Economics, 26, 3-27.[40] Titman, Sheridan, and Roberto Wessels, 1988, The determinants of capital structure, Journal of Finance 43(1), 1-19.[41] Thompson, Samuel B., 2011, Simple formulas for standard errors that cluster by both …rm and time, Journal of Financial Economics 99(1), 1-10.[42] Williamson, Oliver E., 1988, Corporate …nance and corporate governance, Journal of Finance 43(3), 567-591. 33
  34. 34. Figures and TablesFigure 1Note: This figure plots the relationship between average firm-level leverage (defined as theratio of total debt to assets) and average firm-level share of temporary employees, computed fordifferent industries across all firm-years in ESEE. The time period covers 1994-2006.Figure 2Note: This figure plots the relationship between average firm-level leverage (defined as theratio of total debt to assets) and average firm-level share of temporary employees, computed fordifferent years across all firms in ESEE. 34

×