Hiring costs for skilled workers and the supply of firm-provided training

Transcription

1 Oxford Economic Papers, 68(1), 2016, doi: /oep/gpv050 Advance Access Publication Date: 29 July 2015 Hiring costs for skilled workers and the supply of firm-provided training By Marc Blatter a, Samuel Muehlemann b, Samuel Schenker c, and Stefan C. Wolter d a Department of Economics, University of Bern b University of Munich, Munich School of Management, Geschwister-Scholl-Platz 1, D Munich; muehlemann@bwl.lmu.de c State Secretariat for International Financial Matters, Bern d University of Bern, IZA Bonn, and CESifo Munich Abstract This article analyses how the costs of hiring skilled workers from the external labour market affect a firm s supply of training. Using administrative survey data with detailed information on hiring and training costs for Swiss firms, we find evidence for substantial and increasing marginal hiring costs. However, firms can invest in internal training of unskilled workers and thereby avoid costs for external hiring. Controlling for a firm s training investment, we find that a 1 standard deviation increase in average external hiring costs increases the number of internal training positions by more than half of a standard deviation. JEL classifications: J23, J24, J32 1. Introduction Finding the right workers with the appropriate skills to perform a particular job can be difficult for a firm, thus the search process may consume a considerable amount of time and resources. Even though unemployed workers are looking for jobs, a firm may at the same time have difficulties filling its vacancies because the skills of the unemployed workers do not match the job requirements. A firm basically has three options to meet its demand for skilled labour. First, it can simply keep searching for a qualified applicant who possesses all the required skills (which may take a very long time). Second, a firm can hire an experienced worker from the external labour market who does not yet possess all relevant skills. Thus, there will be an adaptation period during which the worker is not yet fully productive due to participating in formal or informal workplace training. Third, a firm may hire and train a young and inexperienced worker internally with the intention to retain that person as a skilled worker at some point in the future. In practice, the first scenario is very unlikely, as most jobs require some firm-specific component, so that it will take new hires some time VC Oxford University Press All rights reserved.

2 M. BLATTER ET AL. 239 (and training) to become fully productive. Moreover, firms may be reluctant to keep a vacancy for an extended period, as the costs of a vacancy are reflected in forgone profits. Thus, the decision of a firm essentially boils down to hiring experienced workers externally (and if necessary to provide training at the workplace) or train unskilled or low-skilled workers internally. In times when the overall willingness of firms to provide such skills at the workplace declines (e.g., as argued by Cappelli, 2014 for the USA), firms in general will find it more difficult and hence more expensive to recruit adequately skilled workers from the external labour market. Put differently, whilst the overall unwillingness of firms to provide training may cause a skills mismatch in particular industries, an individual firm can overcome that problem by providing the necessary skills at the workplace. Empirical studies show that hiring costs constitute an important component of nonwage labour costs, averaging between one and two quarters of wage payments (Manning, 2006; Merz and Yashiv, 2007; Ejarque and Nilsen, 2008; Blatter et al., 2012; Muehlemann and Pfeifer, 2015). In addition, these studies point towards a convex structure of hiring costs, that is, marginal hiring costs increase with the number of hires, which is in line with the assumptions in many theoretical search models of the labour market (e.g., Mortensen, 2005). Thus, hiring exclusively from the external labour market can become very expensive, particularly for large firms that hire many workers in each period. However, firms can reduce external hiring by training unskilled workers internally and retaining them after training (Stevens, 1994). Nevertheless, such a strategy can only be profitable if the (net) costs of providing workplace training are lower compared to external hiring costs. So far there is a paucity of large-scale and representative data that allows analysing to what extent a firm s provision of workplace training depends on the relative costs of external hiring compared to internal training and retention. Our empirical analysis uses representative Swiss data that include detailed and direct measures of hiring costs for skilled workers and training costs for young and initially unskilled workers (apprentices) for a particular establishment. The case of apprenticeship training is particularly suitable for analysing a firm s decision to train unskilled workers internally or to hire skilled workers externally. Apprentices accumulate a substantial amount of human capital over three to four years, making them viable substitutes for skilled workers that are available on the external labour market. Thus, depending on the magnitude and structure of external hiring costs compared to the internal provision of the required skills, a firm may opt either for external hiring, internal training, or both. We carry out our empirical estimations in a generalized Tobit model, accounting for (i) truncation at zero in the supply of training, (ii) selectivity in the firm s training and hiring decisions, and (iii) simultaneity issues in observed hiring and training costs. First, as found in earlier studies, we show that hiring costs are substantial and have a convex cost structure, which means that firms hiring many workers each period will face higher average costs to fill a vacancy compared with firms that recruit less frequently. Second, we find that the structure of (net) training costs is concave, implying economies of scales in the provision of workplace training. Finally, we can show that firms facing high external hiring costs supply more apprenticeship positions, as training reduces or for some firms even eliminates the need to hire externally. Thus, firms can save future hiring costs for skilled workers by recruiting and training young workers internally. Moreover, we find a positive association of external recruitment costs and the share of graduated trainees who stay with the firm as skilled workers. Our results can at least partly explain why large firms that have a high demand for skilled workers almost always offer apprenticeship training programs, whereas many small firms are only willing to

3 240 EVIDENCE FROM SWITZERLAND provide training if the provision of apprenticeship training itself is profitable. Moreover, the effect of expected hiring costs on the supply of apprentices is stronger in establishments that incur net training costs. The article is organized as follows: Section 2 gives a brief overview of the literature on hiring and training costs. Section 3 describes the institutional background in Switzerland and the data sources, and Section 4 presents the econometric modelling and the empirical results. Section 5 concludes. 2. Literature on hiring and training costs Firms may offer training positions because they expect some future return that offsets initial net training costs (Oatey, 1970; Lindley, 1975; Merrilees, 1983). This idea was first formalized by Stevens (1994) in an investment model for a firm s supply of training. Whilst the costs associated with internal training constitute the employer s investment, returns on the investment are given by reduced hiring costs for skilled workers. However, given the lack of data on both hiring costs and net costs of training, Stevens (1994) had to make strong assumptions about the structure of both types of costs. For example, data on wages serves as a proxy for training costs, whereas a variable indicating a shortage of skilled labour in the firm serves as a proxy for hiring costs. 1 Before focussing on our own contribution, we review the empirical evidence on hiring and training costs. 2.1 Evidence on hiring costs Few empirical studies contain direct data on hiring costs. In a first study, Oi (1962) shows that hiring costs are equal to about three weeks of wage payments in the USA, whereas Barron et al. (1985) reports hiring costs of one week s pay. More recent estimates for the USA report average hiring costs of $4,000 per vacancy (Dube et al., 2010). Blatter et al. (2012) use the Swiss costs-benefit survey data from 2000 and 2004 and show that hiring costs in Switzerland range between 10 and 17 weeks of wage payments. Finally, in a recent study based on the latest cost-benefit survey in Germany, Muehlemann and Pfeifer (2015) find average hiring costs of more than eight weeks of wage payments for skilled workers in Germany. 2 The magnitude of hiring costs is crucial for a firm s supply of training. If training costs exceed hiring costs, a firm will not supply any internal training. However, it will offer training positions if training costs are lower than hiring costs, and if it expects to retain a 1 There may be other reasons for firms to invest in general skills of their employees, such as the existence of a compressed wage structure (Acemoglu and Pischke, 1998, 1999). Although it would be of interest to analyse whether training firms can retain their former apprentices and extract a posttraining benefit by offering a post-training wage below productivity, our data do not provide adequate information to quantify this type of post-training benefit. Muehlemann et al. (2013) show that although Swiss employers indeed have some monopsony power, the size of the effect is quite modest, suggesting that potential post-training benefits due to compressed wages structures in Switzerland are rather low. 2 A recent survey on hiring costs appears in Manning (2011), with the earlier literature summarized in Hamermesh and Pfann (1996).

4 M. BLATTER ET AL. 241 sufficiently high number of trainees. Thus, if training costs were sufficiently low, a firm would not hire externally but instead only retain former trainees as skilled workers. Depending on the structure of hiring costs, a firm s optimal recruitment strategy may consist of both hiring externally and supplying some internal training. Such a situation can arise if hiring costs are convex, that is, if hiring more workers in a given period becomes increasingly expensive. Many search models assume convex hiring costs (e.g., Mortensen, 2005), with the reasoning that it is easier for a firm to hire few compared with hiring many workers within a given period, because suitable workers that fulfil the firm s job requirements may be scarce. Hence, a firm needs to invest more resources in the search process, for example, post more job advertisements in newspapers or pay for headhunter services. Conversely, if the hiring process involves a large fixed-cost component, then average hiring costs would decrease as a firm hires more new workers. Firms would then find it cheaper to hire many workers at once making internal training less beneficial to the firm. In two recent contributions, Muehlemann and Pfeifer (2015) and Blatter et al. (2012) show that hiring costs in Germany and Switzerland are convex. Manning (2006) provides evidence that hiring costs are convex in the UK, and Pfann and Verspagen (1989) find increasing marginal hiring costs for Dutch firms that increase their labour force significantly. Similarly, Dube et al. (2010) provide evidence for convex hiring costs in the United States. For France, however, Abowd and Kramarz (2003) report that hiring costs are concave, whereas Kramarz and Michaud (2010) later find evidence for linear hiring costs. 2.2 Evidence on training costs Direct and detailed data on training costs are relatively scarce. Much of the training literature focusses on a firm s training decision rather than on its training costs. A series of surveys contains information on the costs and benefits of training apprentices in Germany and Switzerland, two countries with a long tradition of apprenticeship training. Using the two comparable cost-benefit surveys for Germany and Switzerland in 2000, Muehlemann et al. (2010) show that the costs of apprenticeship training are outweighed by short-term training benefits for Swiss firms, whilst apprenticeship training results in net costs of more than e5,000 per apprentice and year of training (Jansen et al., 2015). A recent survey on apprenticeships, including both a detailed treatment of all cost-benefit studies in Germany and Switzerland and the general features of apprenticeship markets, is provided by Wolter and Ryan (2011). 3 Without considering hiring costs, Muehlemann et al. (2007) find that expected net training costs are an important determinant of a firm s decision to train apprentices based on the second cost-benefit survey in Switzerland. Muehlemann et al. (2010) show that expected training costs are significantly higher in Switzerland for non-training firms than for firms that offer training, whereas they find no significant difference in Germany. A number of studies indirectly assess post-training benefits by analysing retention rates of apprentices, tenure, or wages of movers and stayers (e.g., Booth and Satchell, 1994; Winkelmann, 1996; Euwals and Winkelmann, 2002; Cappelli, 2004; Mohrenweiser and Backes-Gellner, 2010). The main result of these studies is that firms are indeed able to retain a significant fraction of their former trainees after training, and firms tend to be able to retain the more able apprentices. Thus, even if the nature of the training is general, 3 For our analysis we use the first two cost-benefit surveys in Switzerland that are based on the years 2000 and 2004.

5 242 EVIDENCE FROM SWITZERLAND employers have an incentive to pay for such training because they are able to recruit suitable skilled workers. However, if qualified skilled labour is readily available at low cost on the external labour market, a firm would not have any incentives to bear substantial net costs from training apprentices. In particular, if labour markets are competitive, former apprentices would eventually leave the training firm if they received higher wage offers by competing firms. Mohrenweiser and Zwick (2009) analyse the impact of a firm s share of apprentices on firm profit and conclude that German firms are only willing to pay for training if they expect to retain a high share of apprentices after graduation and if they find it difficult to hire skilled workers externally. In contrast to previous studies on firm-sponsored training, we carry out our analysis using detailed and direct firm-level data on both hiring and training costs. Thus, our contribution is to simultaneously estimate the effects of hiring costs and training costs on a firm s supply of apprenticeship training. 3. Institutional background and data 3.1 The Swiss apprenticeship system The apprenticeship system has a long tradition in German-speaking countries. It is characterized by the so-called dual education system, that is, a training program combining training and working within a firm and vocational education at school. In Switzerland, this path is chosen by around 60% of young people who complete their compulsory schooling. From the remaining 40%, about one half attend grammar school to prepare them for an academic education, whereas the remainder opt either for other entirely school-based forms of education or pursue no form of post-compulsory education. Apprentices can choose from over 200 different professions. Usually, an apprenticeship training program lasts three to four years. During this program, an apprentice spends about one or two days a week in a public vocational school. During the rest of the week, an apprentice participates in the production process or receives further training within the firm. After completing the training program, apprentices receive a diploma recognized throughout the country. Graduated apprentices who have acquired an additional qualification (professional baccalaureate) have access to third-level studies at a university of applied sciences. Thus, an individual with a completed apprenticeship has various perspectives for further professional development. In 2004, Swiss establishments offering apprenticeships invested 4.7 billion Swiss francs in their training programs, which corresponds to about 1% of GDP. Overall, the apprentices created a value of 5.2 billion Swiss francs during the training program. Hence, an apprenticeship training program is profitable for the firm on average. However, approximately one third of the apprenticeships end with net costs for the firm. If the employer and the apprentice want to continue their employment relationship after the training program, they have to negotiate a new labour contract. In 2004, only 37% of apprentices remained with the firm where they received their training one year after graduation. By continuing the employment relation with the graduated apprentices, the firm can reduce its hiring costs for skilled labour. This way, establishments may offset some training costs against the costs of hiring skilled workers on the external labour market. 3.2 Survey design and data We use pooled data from two administrative and representative establishment-level surveys conducted in Swiss establishments in 2000 and 2004 by the Centre for Research in

6 M. BLATTER ET AL. 243 Economics of Education at the University of Berne and the Swiss Federal Statistica Office. 4 Overall, we have information on 6,739 Swiss establishments, of which we have information on hiring costs for 4,052 establishments (of which 1,671 establishments do not offer any internal training), as well as information on training costs for 3,644 establishments (of which 1,263 establishments do not hire externally). 5 The survey contains information on the number of skilled workers (in this case with a vocational qualification) that an establishment has hired in the previous three years. Management or the person in charge of human resources supplied the information in a paper-based survey that the Swiss Federal Statistical Office (FSO) sent by regular mail. The survey provides information on the average costs of hiring a skilled worker with a vocational qualification from the external labour market. For ease of comparison across establishments, establishments had to fill out the survey for only one occupation. To avoid selectivity issues, the FSO randomly assigned each firm an occupation based on the occupation s relative importance before sending out the survey. A main advantage of this data is that we observe detailed information of an establishment s training expenses, and wage payments across different groups of employees and separately for apprentices. Moreover, we observe how much time apprentices spend at the workplace, and what type of tasks an establishment allocates to the trainees in each year of training. Importantly, the data are available at the occupation level, which reduces heterogeneity in training investments, as the establishments need to follow the national training regulations for a particular occupation. Finally, we can observe the costs that arise when an establishment hires a skilled worker in the same occupation from the external labour market. This allows us to compare the costs of internal training and then retaining an apprentice with the situation where an establishment hires a worker from the external labour market. To our knowledge, representative data on such variables are currently only available in the cost-benefit surveys in Switzerland and Germany. However, there are also some disadvantages to our data source. We observe average outcomes within establishments, thus we are not able to identify the hiring and training process of an individual employee or apprentice over time and to follow their future employment patterns. In particular, we are not able to include a worker fixed effect in our estimation to allow for heterogeneity in the ability of workers within establishments. Moreover, we cannot distinguish hiring costs by the age of new hires, and we do not have information about individual wage patterns. To carry out such an analysis, it would be necessary to match a cost-benefit survey with existing linked employer-employee data. However, such data are currently not available in Switzerland. 4 This data were originally collected for estimating total private sector expenditures on vocational education and training in Switzerland, as those expenditures are part of official OECD statistics on educational expenditures (OECD, 2009). Potterat (2006) provides detailed information about the sampling procedure. 5 We excluded public establishments and non-profit organizations from the sample, as the principle of profit maximization does not fully apply to such establishments. Nonetheless, using the full sample, our main results remain robust, although the magnitude of the effects is somewhat smaller. Results are available on request.

7 244 EVIDENCE FROM SWITZERLAND 3.3 Calculation of hiring costs The calculation of hiring costs can be divided into two parts: the costs of recruiting a worker, denoted by r, and the costs of the initial training necessary for the worker s adapting to the new job, denoted by a. Recruiting costs r i consist of the costs for posting a vacancy v i and interview costs J i c ai, where J i denotes the number of applicants per vacancy invited for an interview, and c ai denotes the costs of conducting an individual interview, that is, the product of interview time and the wage of the interviewer(s). 6 In addition, we denote the costs for external advisors or placement agencies by e i. Summarizing, recruitment costs are r i ¼ v i þ J i c ai þ e i Adaptation costs arise because a newly employed skilled worker is not immediately fully productive. 7 In the survey, establishments report for how many days d ai a new worker is less productive than an average skilled worker in the firm. The relative productivity of a new recruit is denoted by p i. There are several reasons for a new recruit being initially less productive. One is firm-specific human capital, such as learning the firm culture, learning production processes, and becoming acquainted with work colleagues. 8 Another reason for lower worker productivity is that new recruits receive training outside the workplace. First, the firm has to pay the worker the daily salary w di during the number of training days d ti. Second, there are direct training costs c ti for training personnel, travel costs, or course fees. Thus, adaptation costs a i can be written as a i ¼ d ai ð1 p i Þw di þ d ti w di þ c ti Total hiring costs H for filling a vacancy are the sum of recruitment and adaptation costs H i ¼ r i þ a i 3.4 Calculation of the net costs of training apprentices We define net costs of apprenticeship training as the difference between the firm s costs and the benefits of training. Training costs c consist of the wages of apprentices w a and the costs of training personnel w T, which add up in equal share to about 90% of total costs. We denote the remaining costs by x, which includes costs for materials, infrastructure, external courses, hiring, and administrative tasks. Thus, the average training costs for firm i are c i ¼ w ai þ w Ti þ x i The calculation of training costs suggests that the main part of it is given by training hours, wages of instructors, and wages of apprentices. Thus, whilst firms are legally required to 6 There are five job categories for interviewers: management, skilled workers with a vocational qualification (by the sub-categories of administration, technical or social, or crafts), and workers with no vocational qualification (i.e., unskilled workers). 7 Adaptation costs can also be interpreted as a reduced benefit from employing the new hire compared to an internal skilled worker. 8 Our estimates correspond to the true adaptation costs in the case where skilled worker wages are equal to skilled worker productivity. However, when wages are set below productivity, our estimates provide a lower bound for the true adaptation costs, as the firm only obtains the fraction ð1 p i Þ of the difference between productivity and wage during the adaption period.

8 M. BLATTER ET AL. 245 offer a minimum training standard, it is always possible to provide more training, particularly if an establishment trains with an investment motive. The wages of apprentices are largely pre-determined by wage recommendations of professional associations. Thus, the variation of wages within a profession is relatively small. Larger establishments typically offer higher wages to apprentices, which is consistent with the fact that larger establishments offer higher wages to all categories of workers. Similarly to standard wage regressions, the average wages of the training instructors can be best explained using variables such as firm size, industry, and regional characteristics. 9 The remaining costs for material, infrastructure, and external courses are essentially given by the training profession. Training benefits b are calculated by the type of work an apprentice performs. An apprentice spends a fraction a of his work hours performing unskilled tasks that require no formal qualifications. During the remaining time ð1 aþ, the apprentice carries out skilled tasks. Whilst we assume that the apprentice s performance is equal to that of an unskilled worker ðw u Þ, the apprentice s performance in skilled tasks, 1w, is lower compared to that of a skilled worker. Finally, we denote the wage of a skilled worker by w. b i ¼ aw ui þð1 aþ1w i Thus, the average net costs of training C constitute the difference between the costs c and the benefits b of training, so that C i ¼ c i b i 3.5 Descriptive statistics In this section we provide descriptive statistics on the external and internal hiring process, as well as the weighted proportion of firms hiring externally, train internally, or do both (or none). Establishments with internal training and external hiring activity report the costs for internal training and external hiring for the same occupation, facilitating the comparison of internal versus external training costs. 10 We provide detailed descriptive statistics of the different components of average hiring costs in Table 1. Distinguishing by training and hiring activities, Table 2 shows that hiring costs amount to 11,888 Swiss francs (CHF 11 ) in establishments that both hire externally and offer internal training. Such establishments hire on average 3.5 new workers each period. Establishments that solely hire externally fill on average 2.6 vacancies each period at an average cost of CHF 14,286. Establishments with both external hiring and internal training hire more workers each period, however, the separation rate of skilled workers d in these establishments is only 10.2%, whilst the separation rate is 14.7% in establishments without internal training. These results seem surprising at first sight, as establishments without internal training even pay slightly higher wages to skilled workers (albeit this difference is not statistically significant at the 5% level). Nonetheless, establishments without 9 Muehlemann et al. (2013) find that Swiss establishments have some monopsony power over skilled workers and apprentices, whereas unskilled pay is set competitively. However, the size of the effect of monopsony power is fairly modest, implying that an establishment s possibilities to set wages are limited. 10 Establishments without internal training report hiring costs for the occupation in which they would most likely offer internal training if they decided to do so. 11 At current exchange rates, CHF 1 ¼ US$0.9, and CHF 1 ¼ e1.05.

9 246 EVIDENCE FROM SWITZERLAND Table 1. Descriptive statistics Variable Mean S.E. Minimum Maximum Obs. Costs for job postings v (CHF) 1,103 1, ,000 4,052 Costs for interview per applicant c a (CHF) ,844 4,052 Number of interviewed applicants J per vacancy ,052 Personnel costs for interviews J c a 2,009 3, ,586 4,052 Costs for external advisors/headhunters e (CHF) 414 1, ,000 4,052 Recruitment costs r ¼ v þ J c a þ e (CHF) 3,878 5, ,117 4,052 Duration of adaption period in days d a ,052 Average decline in productivity ð1 pþ during ,052 adaption period (%) Daily wage w of a skilled worker with vocational ,052 qualification (CHF) Duration of training courses in days d t ,052 Direct training costs c t (CHF) 550 1, ,000 4,052 Adaption costs a ¼ d a ð1 pþwþd t w þ c t (CHF) 9,688 11, ,779 4,052 Average hiring costs H ¼ r þ a to fill a vacancy (CHF) 13,570 13, ,575 4,052 Table 2. Hiring and training costs External hiring No external hiring Training No training Training No training Hiring costs H 11, , (11, ) (14, ) Number of recruits R (4.369) (2.919) Separation rate d (16.629) (19.907) Net training costs C 7, , (34,908.28) (33,170) Number of trainees L (3.205) (1.864) Number of trainees retained (4.078) (1.386) Trainee separation rate c ( ) Number of skilled workers N s ( ) (6.6944) (9.7787) (2.7852) Number of other workers N o (115.4) (37.62) (59.41) (15.04) Monthly skilled worker pay w 6, , , ,229.0 (1,417.5) (1,460.8) (1,281.0) (514.6) Observations 2,381 1,671 1,263 1,424 Weighted proportion of firms Notes: Standard deviation in parentheses. The number of recruits refers to the previous three years within a randomly assigned occupation.

10 M. BLATTER ET AL. 247 internal training are typically much smaller than training establishments, both in the number of skilled workers in the occupation of interest, but also in the number of other workers. Therefore, the observed differences in descriptive statistics may be attributed to economies of scales in external hiring for large establishments, allowing large establishments to achieve a better match quality at lower costs. 12 Internal training results on average in a net benefit of CHF 7,304 for establishments that also hire externally (Table 2). Establishments that train internally but do not hire externally during our observation period generate on average a net benefit of CHF 9,959. However, about one third of all Swiss training establishments bear substantial net training costs. The separation rate of apprentices c, that is, the fraction of trained apprentices who leave the firm within the first year after training, is on average 68.3% for training establishments that also hire externally, and 74.05% for training establishments that do not hire externally. The number of new hires in an establishment consists of the number of external hires as well as number of trainees that were retained as skilled workers after completion of training. For establishments that neither hired externally nor retained any trainees, we estimate their potential number of hires by ordinary least squares, using firm size, wages, occupation, and industry as independent variables. Our main findings remain robust to alternative specifications of the number of hires, and also if we drop observations of establishments without new hires in the period of observation. All results are available on request. The summary statistics for all other variables are provided in Online Appendix Table A1. Training establishments are on average larger than non-training establishments and are located more frequently in the construction and industrial sector. The share of foreign establishments in our sample is on average 11%. However, foreign establishments are less likely to offer internal training compared to domestic establishments. Muehlemann (2014) shows that the main reason foreign establishments refrain from offering apprenticeships is that they are often very small and specialized establishments that to not have a high demand for workers with a vocational qualification. 4. Econometric modelling and empirical results Our aim in this section is to estimate the causal effect of hiring costs for skilled workers on an establishment s supply of apprenticeship positions. We are interested in testing whether expected external hiring costs lead to an increase of an establishment s supply of apprenticeship positions, whilst holding net training costs constant. We have to deal with two problems for which we need an econometric solution: selectivity and endogeneity. First, we observe hiring costs only for an establishment that actually recruited skilled workers externally in the previous three years, thus establishments that did not hire externally could potentially face different hiring costs. Second, we observe training costs only for an establishment that actually trains apprentices, thus training establishments may be a selected sample of the entire population of establishments (as previously shown in Wolter et al., 2006). Third, the (average) hiring costs that we observe for those 12 Idson (1993) provides evidence for lower turnover rates in large establishments. Blatter et al. (2012) also show that large establishments find it less costly to hire a given number of new workers than do small establishments. However, marginal hiring costs for large establishments are convex if hiring is above an establishment s expected hiring rate.

11 248 EVIDENCE FROM SWITZERLAND establishments that hire externally and at the same time also offer apprenticeship training are different from the hiring costs such establishments had faced in the absence of apprenticeship training, unless marginal hiring costs were constant. Establishments that train apprentices need to hire fewer workers externally as long as they retain some apprentices as skilled workers after training. As a consequence, if hiring costs are convex, having to hire fewer workers from the external labour market results in lower observed average hiring costs. Such a situation would not arise if both hiring and training costs were constant. In that case, an establishment would simply choose the cheaper recruitment option, but never both. Endogeneity problems arise because an establishment can influence both hiring and net training costs. An establishment can influence its hiring costs by adjusting time and effort for screening job applicants (thereby also affecting the turnover rate). Moreover, training regulations require establishments to enforce minimum training standards. However, it is always possible for an establishment to provide excess training, for example if firm-specific skills are of particular importance. 13 A final complication is that the dependent variable of interest, the supply of training positions, is truncated at zero. We use a type 4 Tobit model to account for selectivity and endogeneity of hiring and training costs and the truncation of training supply at zero. The following sub-section describes the econometric model in detail Econometric modelling To estimate our econometric model, we assume that hiring costs can be described as H T ¼ h 0 R a N b s Nc o wd (see, e.g., Manning, 2011). We thereby account for the dependence of hiring costs H on the number of recruits R and the number of employees N s (skilled workers in the corresponding occupation) and N o (the number of other employees), as well as the wage of skilled workers w. Our specification therefore accounts for the stylized fact that larger establishments typically hire more workers in a given period and pay higher wages. Thus, marginal hiring costs are given ¼ ah 0R a 1 N b s Nc o wd In our data we observe average hiring costs, which are given by H ¼ HT R ¼ h 0R a 1 N b s Nc o wd Taking logs on both sides, we estimate a regression model of the form ln H ¼ ln h 0 þða 1Þln R þ b ln N s þ c ln N o þ d ln w: 15 Marginal hiring costs are increasing if a > 1, and decreasing if a < 1. The supply of training also depends on net training costs C. We estimate a linear regression model, because net training costs can be negative (i.e., establishments can generate 13 Whilst observing establishments across time would enable us to eliminate unobserved firmspecific effects, we can only identify a very small and non-random sub-sample of fewer than 300 establishments in our data in two time periods. We thus refrain from presenting those results. 14 For a classification of different types of Tobit models, see Amemiya (1985). 15 As noted in Manning (2011), marginal hiring costs can be described in terms of average hiring costs, ¼ a HT R.

12 M. BLATTER ET AL. 249 a net benefit from training if the productive contribution during training exceeds the training costs). 16 Consider the structural model of an establishment s supply of training positions. ln H ¼ x 1 b 1 þ e 1 (1) C ¼ x 2 b 2 þ e 2 (2) L ¼ max½0; x 3 b 3 þ ah þ dc þ e 3 Š (3) where L denotes the number of trainees, H the average costs of only hiring skilled workers externally, and C the average net costs of training. We always observe (x, L), H is observed if L ¼ 0 (because for establishments with L > 0, average hiring costs H in the absence of apprenticeship training could be higher than observed hiring costs due to a convex cost structure, as internal training reduces the need to hire externally), and C is observed if L > 0. The error terms (e 1 ; e 2 ; e 3 Þ are independent of x, with a zero-mean trivariate normal distribution. The vector x 1 contains firm characteristics, including the number of recruits, the number of employees, skilled worker wage, firm ownership, occupation, and industry. Furthermore, x 1 needs to contain at least one element that is not in x 3 but has a significant effect on hiring costs (the first exclusion restriction). We use the binary variable difficulties in finding skilled workers as a measure for the tightness of the labour market. Because establishments have to spend more resources on the search of appropriate job candidates when the labour market is tight, hiring costs increase. However, we assume that the firm s supply of training positions is not directly related to labour market tightness, conditional on other observed firm characteristics. Similarly, x 2 must contain at least one element that is not in x 3 but that significantly affects net training costs (the second exclusion restriction). We use the variable population share of potential trainees, which is defined as the product of local share of young people age (1 share of young people in grammar school ). 17 Thus, the variable population share of potential trainees captures the variation in the supply of local trainees due to demographic and institutional differences across local labour markets. 18 A higher share of suitable candidates reduces the net costs of training by increasing the probability that an establishment finds a suitable trainee, that is, a good match. Compared to a situation in which an establishment needs to hire a less suitable candidate, a good match will reduce the firm s required effort for a trainee to reach the desired skill level. The coefficients of interest are the structural parameters a and d. For identification, we follow the procedure described in Wooldridge (2002). 16 We also estimate a model with interaction terms of firm size and the number of trainees, as large establishments offer more training positions. However, we find no statistically significant effects. Results are available on request. 17 We follow Muehlemann and Wolter (2011), defining a local labour market as containing all cities an individual can reach by car from the centre of a local labour market within 30 minutes. The variable share of young people in grammar school is a lagged variable that measures the local share of youth enrolled in grammar schools the year before establishments in our sample decided to offer internal training. 18 As a robustness check we have used each of the two variables separately as an individual instrument. For both cases, the results remain qualitatively similar, but of slightly lower statistical significance.

13 250 EVIDENCE FROM SWITZERLAND The reduced form for eq. (3) is L ¼ max½0; x 3 b 3 þ aðx 1 b 1 þ e 1 Þþdðx 2 b 2 þ e 2 Þþe 3 Š ¼ max½0; xq 3 þ ae 1 þ de 2 þ e 3 Š ¼ max½0; xq 3 þ u 3 Š (4) First, we regress L on x by a standard Tobit regression using all observations. Doing so enables us to generate a generalized residual (Vella, 1992): ^u 3i ¼ ^r 3 ð1 I i Þ/ðx 3i^b3 =^r 3 Þð1 Uðx 3i^b3 =^r 3 ÞÞ 1 þ I i ðl i x 3i^b3 Þ where I i is an indicator function denoting whether an establishment offers training. Second, we need a consistent estimate of u 3i to obtain estimates of b 1 and b 2,as and Eðln HjL ¼ 0; x; u 3 Þ¼x 0 1 b 1 þ Eðe 1 jx; u 3 Þ¼x 0 1 b 1 þ Eðe 1 ju 3 Þ¼x 0 1 b 1 þ c 1 u 3 EðCjL > 0; x; u 3 Þ¼x 0 2 b 2 þ Eðe 2 jx; u 3 Þ¼x 0 2 b 2 þ Eðe 2 ju 3 Þ¼x 0 2 b 2 þ c 2 u 3 where u 3 ¼ ae 1 þ de 2 þ e 3. We can use the coefficients c 1 and c 2 to test for selectivity (Vella, 1992). Using observations for establishments that hire solely externally, we regress ln H i on x i1 ; ^u 3i ; (5) yielding consistent estimates of b 1 and allowing us to test for selectivity. 19 Similarly, using observations of training establishments that do not hire externally, we regress C i on x i2 ; ^u 3i ; (6) yielding consistent estimates of b 2 and again allowing us to test for selectivity. Having obtained consistent estimates ^b 1 ; ^b 2 for b 1, b 2, we can estimate b 3, a and d using the reduced form of L in terms of the following structural parameters: L ¼ max½0; x 3 b 3 þ aðx 1 b 1 Þþdðx 2 b 2 Þþu 3 Š Using our consistent estimates ^b 1 ; ^b 2, we can estimate the following Tobit equation to obtain consistent estimates b 3 ; ^a, and ^d: L ¼ max½0; x 3 b 3 þ aðx 1^b1 Þþdðx 2^b2 Þþerror i Š (7) 4.2 Results Based on the reduced-form Tobit regression (Online Appendix Table A2), we obtain an estimate of the generalized residual (^u 3i ), which we need to obtain consistent estimates and test for self-selection bias in the hiring costs and net training costs equations. Regressing hiring costs on our vector of control variables x 1 and the generalized residual (Table 3), we find that the coefficient on the generalized residual is not statistically significant. Thus, we find no evidence that the sample of establishments that have hired skilled workers externally in the previous three years differs from those establishments that have 19 Our main results remain robust when running regressions with all available observations on hiring costs (i.e., including establishments that hire externally and also train apprentices). Results are available on request.

14 M. BLATTER ET AL. 251 Table 3. Hiring cost regression Dependent variable: ln Hiring costs H ln(number of hires R) * (0.0729) ln(skilled workers N s ) (0.0530) ln(other workers N o ) ** (0.0067) ln(wage of skilled workers w) *** (0.1214) Foreign firm ownership (0.0783) Aggregate cantonal income (in 1,000 CHF) (0.0024) Difficulties to find skilled workers *** (0.0558) Generalized residual (^u 3i ) (0.0329) Industry controls Yes Job controls Yes Constant (1.1153) R Observations 1,671 Notes: Cluster-robust standard errors in parentheses. Significance levels: * p < 0.05, ** p < 0.01, *** p < not hired externally, and we can obtain a consistent estimate of ^H. Confirming the earlier findings of Blatter et al. (2012), we find that the number of hires has a positive and significant effect on average hiring costs, implying a convex cost structure with a hiring costs elasticity of a ¼ 1:19. We also find a strong and statistically significant effect of our first exclusion restriction: establishments facing difficulties in hiring skilled labour externally exhibit hiring costs that are 20.4% above average. We further find that hiring costs increase with firm size and strongly depend on skilled workers pay. Regressing net training costs C on x 2 and the generalized residual (^u 3i ), we find that training establishments are a non-random subsample of the population of all establishments (Table 4), as indicated by the negative and statistically significant coefficient of the generalized residual. Thus, whilst training establishments on average generate a net benefit, nontraining establishments face significantly higher expected net training costs. In contrast to the results on hiring costs, we find that the structure of net training costs is concave, as hiring an additional apprentice decreases average net training costs by CHF 180. This result is mainly due to economies of scale in instruction time (training several trainees at the same time requires less time per trainee). 20 Furthermore, concerning our 20 Detailed results of regressing the different components of the net training costs (such as instruction time, material and infrastructure costs, and wages) on the number of apprentices are available on request.

15 252 EVIDENCE FROM SWITZERLAND Table 4. Net cost regression Dependent variable Net training costs C Number of trainees L ** ( ) Skilled workers N s ( ) Number of other workers N o * (4.7449) Wage of skilled workers w (1.0646) Foreign firm ownership (3,605.03) Aggregate cantonal income (in 1,000 CHF) ( ) Population share of potential trainees (%) 4, *** (1, ) Generalized residual (^u 3i ) * ( ) Industry controls Yes Job controls Yes Constant 18, (10,114.16) R Observations 1,263 Notes: Cluster-robust standard errors in parentheses. Significance levels: * p < 0.05, ** p < 0.01, *** p < second exclusion restriction, the results show that an increase in the local share of potential trainees by 10% (0.5 percentage point) leads to a statistically significant decrease in net training costs by CHF 2,180. Thus, an increase in the local supply of potential apprentices increases the match quality, and therefore results in lower net training costs. Having obtained consistent estimates of hiring costs ( ^H) and net training costs ( ^C), we estimate the structural effects of hiring costs and net training costs on the firm s supply of apprenticeship positions. The results show that both variables have significant effects on the number of training positions L (Table 5). An increase in average hiring costs by CHF 1,000 increases the number of training positions by 0.13, implying that an increase in average hiring costs by CHF 8,000 induces an establishment to hire one additional trainee. This effect is economically substantial, as an increase of H by 1 standard deviation (CHF 7,174) increases an establishment s supply of training by 0.52 standard deviation (0.93 training position). As expected, the net costs of training have a negative effect on a firm s supply of training. If the net costs C increase by CHF 1,000, the number of training positions L decreases by Put differently, if C increases by 1 standard deviation (CHF 13,251), a firm s number of internal training positions will decrease by 0.93 standard deviation (1.8 training positions). We then estimate the supply of training positions using only establishments with positive net costs ð ^C > 0Þ. We expect a stronger effect of hiring costs on training supply for this

16 M. BLATTER ET AL. 253 Table 5. Determinants of a firm s supply of training positions L Dependent variable Number of training positions L All establishments If ^C > 0 Hiring costs ^H (1,000 CHF) 0.127*** 0.185*** 0.236*** (0.013) (0.034) (0.050) Net training costs ^C (1,000 CHF) 0.134*** 0.234*** 0.194** (0.040) (0.069) (0.072) ^H ð 1Þ ^C (1,000,000 CHF) 3.210** lnð ^C= ^HÞ 0.364*** (1.040) (0.120) New hires (internal þ external) *** 0.247*** 0.295*** (0.039) (0.096) (0.088) (0.079) Number of skilled workers N s 0.055*** 0.020** ** (0.008) (0.009) (0.008) (0.008) Number of other workers N o *** *** *** *** (0.0015) (0.0015) (0.0016) (0.0016) Wage of skilled workers w (CHF) *** *** ** (0.0001) (0.0002) (0.0002) (0.0002) Foreign firm ownership 1.880*** 2.051*** 2.062*** 1.712*** (0.253) (0.622) (0.631) (0.471) Aggregate cantonal (i.e., state) income (1,000 CHF) (0.0102) (0.0148) (0.0105) (0.0116) Industry controls Yes Yes Yes Yes Job controls Yes Yes Yes Yes Constant (0.454) (1.066) (1.069) (0.824) Log pseudo likelihood 427, , , , Observations 6,739 2,354 2,354 2,354 Notes: Cluster-robust standard errors in parentheses. Significance levels: * p < 0.05, ** p < 0.01, *** p < sub-sample, as post-training benefits are crucial for establishments that incur net training costs (Table 5). We find that an increase in average hiring costs H of CHF 1,000 increases the supply of training positions by 0.17 (third column, Table 5). Put differently, an increase in H by 1 standard deviation leads to a 0.6 standard deviation (1.4 training positions) increase in the supply of training for establishments with ^C > 0. Including an interaction term of hiring costs and net training costs (multiplied by ð 1Þ), we find that the effect of hiring costs is inversely related to the level of net training costs (last column, Table 5). The coefficients on the individual variables remain significant. Our results show that the marginal effect of average hiring costs evaluated at the average of net training costs ¼ 0:236 0:00321 ^Ci Þ) on the firm s supply of training is Thus, an increase in H by 1 standard deviation leads to a 0.72 standard deviation (1.7 training positions) increase in the supply of training for establishments with ^C > 0. Finally, we run a specification for the share of net training costs and external hiring costs ð ^C= ^HÞ as a further robustness check (fourth column, Table 5). As this variable is skewed to the right, we use the logarithmic transformation ln ð ^C= ^HÞ in our regression