Davud Rostam-Afschar, IV.2 Entry regulation and entrepreneurship: a natural experiment in German craftsmanship in:

Oliver Holtemöller (Hrsg.)

How Can We Boost Competition in the Services Sector?, Seite 89 - 133

1. Auflage 2017, ISBN print: 978-3-8487-4676-7, ISBN online: 978-3-8452-8902-1, https://doi.org/10.5771/9783845289021-88

Bibliographische Daten
Zitation Download Teilen
Entry regulation and entrepreneurship: a natural experiment in German craftsmanship59 (Davud Rostam-Afschar) Dr Davud Rostam-Afschar, University Stuttgart-Hohenheim Abstract This paper uses the 2004 amendment to the German Trade and Crafts Code as a natural experiment for assessing the causal effects of this reform on the probabilities of being self-employed and of transition into and out of self-employment. This is achieved by using repeated cross-sections (2002–2009) of German microcensus data. I apply the difference-in-differ‐ ences technique for three groups of craftsmen which were subject to dif‐ ferent intensities of treatment. The results show that the complete exemp‐ tion from the educational entry requirement has fostered self-employment significantly by substantially increasing the entry probabilities, while exit rates have remained unaffected. I find similar, though weaker relative ef‐ fects for the treatment groups that were subject to a reduction of entry costs or a partial exemption from the entry requirements. Moreover, I con‐ sider effect heterogeneity within each of the treatment groups with respect to gender and vocational training, and show that the deregulation of entry requirements has been most effective for untrained workers. Introduction How does entry regulation influence entrepreneurship? In an attempt to answer this question, many different kinds of regulation, such as the regu‐ lation of product and labor markets, have been investigated. The theoreti‐ cal predictions of the effects of these kinds of regulations are ambiguous. On the one hand, the public choice theory argues that regulations lead to socially inefficient outcomes, while on the other hand the public interest IV.2 1 59 Rostam-Afschar, D. (2014): Entry regulation and entrepreneurship: a natural ex‐ periment in German craftsmanship, Empirical Economics, 47(3), 1067–1101. IV.2 Entry regulation and entrepreneurship 89 theory of regulation claims that regulations serve to cure market failures. With a focus on entrepreneurship, Branstetter et al. (2013) predict that a reform which reduces the fixed costs of setting up a business leads to an increase in the number of firms as well as in employment but the addi‐ tional firms will have entrepreneurs with relatively lower entrepreneurial ability. This study further shows that these firms will be smaller and have a lower probability of survival. Empirical evidence tends to support the view that various implementa‐ tions of entry regulation have detrimental effects. Most of these studies re‐ ly on aggregate data from many countries, as in the influential work by Djankov et al. (2002) and subsequently in research by Klapper et al. (2006); Ciccone and Papaioannou (2007), and van Stel et al. (2007). In ad‐ dition, evidence based on microdata (cf. Bruhn 2011; Ardagna and Lusardi 2010, 2009; Branstetter et al. 2013) enforces the conclusion that lower en‐ try costs increase entry into (formal) entrepreneurship. Moreover, there is a set of studies that, while not directly focusing on entrepreneurship, investigates the effects of entry regulation with microda‐ ta, including, e.g., Bertrand and Kramarz (2002); Sadun (2008); Viviano (2008). In an important contribution, Bertrand and Kramarz (2002) evalu‐ ate a commercial zoning regulation implemented via regional zoning boards in French retailing. This study finds that greater entry regulation reduced employment growth in the retail sector, while concentration and prices increase. Other work connected to this study investigates the effects of product market deregulation for industry dynamics (Aghion et al. 2009; Cetorelli and Strahan 2006; Kerr and Nanda 2009). From studying a deregulation of the French banking industry in the 1980s, one of the find‐ ings in Bertrand et al. (2007) corroborates the notion that less state inter‐ vention is associated with increased firm entry and exit rates. Briefly, the empirical literature almost unanimously comes to the conclusion that entry regulation in its various forms restrains entrepreneurship and similarly other economic outcomes. One particularly interesting implementation of entry regulation is the requirement of the Meister degree in German craftsmanship, as required by the German Trade and Crafts Code (HwO) for registration as an en‐ trepreneur. Prantl and Spitz-Oener (2009) and Prantl (2012) explicitly con‐ sider the entry requirement for craftsmanship to discuss regulatory effects in the wake of German reunification in 1990. The aim of this study is to evaluate a change in the regulatory require‐ ments empirically. It contributes to the literature on entry regulation and IV Past Reforms in the Services Sector and their Effects 90 entrepreneurship by providing evidence of the causal effects of entry regu‐ lation, exploiting this change to the HwO as a natural experiment. Dating back to the late nineteenth century, this latter entry requirement, called Meister (see Sect. 2), underwent a dramatic change: the amendment to the HwO in January 2004 decreased the number of occupations in which craftsmen were required to hold a Meister degree in order to start a busi‐ ness from 94 to 41. Moreover, the entry requirements for the remaining 41 occupations were relaxed. To the best of my knowledge, this is the first paper to use this setting as a natural experiment. The reform was the result of a passionate debate in which proponents of the entry requirement (e.g., German Confederation of Skilled Crafts 2003) cited market failures resulting from information asymmetries and external effects, while opponents (e.g., German Deregulation Commission 1991; German Monopolies Commission 1998, 2002) objected, in the spirit of the public choice theory, that these regulations would lead to greater ineffi‐ ciencies. The government justified the regulation primarily as a means to prevent health related dangers. This argument, in turn, was itself contro‐ versial because there was no agreement as to whether the costs of regu‐ lation would outweigh the costs incurred by careless craftsmen doing haz‐ ardous jobs, for example barbers or chimney sweeps. Focusing on entrepreneurship, in addition to credit constraints (e.g., Evans and Jovanovic 1989, Blanchflower and Oswald 1998, Hurst and Lusardi 2004, Fossen 2011), the entry requirement is regarded as a key impediment to starting a business. For instance, Holtz-Eakin and Rosen (2005) point to the entry requirement as a disincentive to taking up selfemployment in German craftsmanship. To shed some light on the effects of this regulation on entrepreneurship, I use repeated cross-sections (2002–2009) of German microcensus data on self-employment to proxy for business creation. I apply the difference-indifferences (DID) approach to estimate the effects of the policy change for three distinct occupational groups on the probability of self-employment, as well as the probability of transitioning into and out of self-employment. The empirical results provide evidence that the probability of being self-employed increased in line with the amendment to the HwO. The strongest relative increase significantly raised the probability of self-em‐ ployment to a level more than 40 % higher than a hypothetical situation without the reform for an occupational group with a relatively low propen‐ sity to engage in entrepreneurship. This group, hereafter referred to as the group of B1-occupations, has been completely exempted from the entry IV.2 Entry regulation and entrepreneurship 91 requirement. The reform also seems to have increased the probability of being self-employed for professions that experienced only a reduction of or a partial exemption from the entry requirement. The effects for these groups are also positive, although weaker. The analysis shows further that these increases resulted from increasing the probability of entry, while the probability of exit from self-employment has remained virtually unaffect‐ ed by the policy change. The reforms seem to have affected individuals across professional qualifications differently; the deregulation of entry has been most effective for the group of untrained workers who are disadvan‐ taged in the labor market. Below, in Sects. 2 and 3, respectively, I describe the institutional frame‐ work of the natural experiment and outline the empirical approach. In Sects. 4 and 5, I describe the data and discuss the results. Section 6 con‐ cludes. The amendment to the German Trade and Crafts Code in 2004 as a natural experiment Over the course of time, three key institutions of German craftsmanship have emerged: the small proof of competence (Kleiner Befähigungsnach‐ weis), the greater proof of competence (Großer Befähigungsnachweis), and the register of self-employed craftsmen (Handwerksrolle). The small proof of competence restricted the training of apprentices to craftsmen who held a Meister certificate, though such a degree was not required to start a business. However, the greater proof of competence mandated that craftsmen obtain a Meister certificate for both activities, to train and to have a new business listed in the register. Since 1965, legislation has distinguished between restricted regular craftsmanship (Vollhandwerke), which requires a greater proof of compe‐ tence, and unrestricted trades similar to crafts (Handwerksähnliche Gewerbe), referred to in this text as A-occupations and B2-occupations, respectively. In this study, the focus is on craftsmen in A-occupations who remained regulated by a form of the greater proof of competence, in con‐ trast to those in B2-occupations. The Meister title is the highest professional degree in craftsmanship. To attain it, a person must complete several levels of training and pass exami‐ nations. Having obtained the qualification level called Geselle, a crafts‐ man could be employed in a business or continue on to a Meister degree. 2 IV Past Reforms in the Services Sector and their Effects 92 Full-time courses to prepare for the Meister exam take 1–3 years, and the occupation-specific overall costs range, according to the Chambers of Crafts and Trade, from 4,000 to 10,000 Euros. The Meister exam tests both occupation-specific skills and general education in business and com‐ mercial knowledge, as well as law. Moreover, the exam contains a peda‐ gogical component, as holding a Meister degree makes the craftsman eli‐ gible to train apprentices. Those who have passed the examination and started a business are recorded in the register; though in rare exceptional cases, some people may be recorded in the register without a Meister de‐ gree. In the situation immediately prior to the amendment to the HwO in 2004, the options available to a crafts-person were to get hired in a busi‐ ness or to set up a business after having obtained the Meister degree. This analysis exploits this reform to assess the causal effects of entry regulation on entrepreneurship. In the next section, I describe how the different com‐ ponents of this reform altered the options available to a craftsman, and de‐ fine treatment groups and a control group accordingly. The natural experiment Before Requirement After Requirement A (Meister) AC (Meister) A (Meister) A1 (Altgeselle) A (Meister) A2 (Altgeselle, no requirementa) A (Meister) B1 (no requirement) B2 (no requirement) B2 (no requirement) Notes: This table describes the requirement before and after the reform in descending order of a priori supposed intensity of entry regulation. The control group comprises pre- and post-reform occupations that turned out to belong to the AC-occupations. Each treatment group includes pre- and post-reform occupations that turned out to be‐ long to the B1-, A1-, and A2-occupations, respectively. The occupational groups B1, A1, A2, and AC are defined to be mutually exclusive. However, non-craft occupations and B2-occupations within these groups are not always discriminable due to data pro‐ tection, and have been excluded from the analysis where possible. The main results re‐ main unchanged when these occupations are included in the samplea For A2-occupa‐ tions, no requirement is imposed after the reform if a prospective entrepreneur com‐ mits to limit the range of the activities of his firm to tasks that can be learned within 3 months Table 4.2.1: IV.2 Entry regulation and entrepreneurship 93 Empirical specification Definition of the treatment and control groups How did the reform alter the options available to a craftsman? After the reform, a crafts-person could choose to seek employment in a business, re‐ gardless of her obtained professional degree, just as before the reform. The choice to start a business on her own, in contrast, was facilitated by the amendment to the HwO. The amendment came into effect on January 1, 2004, in the context of a series of reforms aimed at the German social sys‐ tem and labor market called Agenda 2010. It defines certain occupational groups which are subject to different degrees of regulation. I matched each reported occupation of an individual in the German microcensus with the respective occupation listed in the law, with examples of these vocations provided below. From this information, I was able to construct four occu‐ pational dummies that reflect the different intensities of the treatment, as outlined in Table 4.2.1. The deregulation of the Meister degree requirement, which is the main element of the policy change, generated a group of 53 B1-occupations by dividing up the former 94 A-occupations. After the reform, craftsmen be‐ longing to the group of B1-occupations were allowed to start businesses without a Meister degree, but still had to demonstrate their ability to train apprentices. These B1-occupations represent the treatment group that was deregulated most, referred to as B1-craftsmen. This category includes tile and mosaic layers, coppersmiths, turners, tailors, millers, and photogra‐ phers. The remaining 41 A-occupations comprise three more groups: AC, A1, and A2. The AC group is comprised of strictly regulated occupations that remained subject to virtually the same requirements as before the policy change; they had already needed a mandatory Meister certificate to enter entrepreneurial activities. These vocations serve as the control group. They include chimney sweeps, optometrists, hearing aid audiologists, or‐ thopedic technicians, and dental technicians. The remainder of the A-occupations had their entry restrictions loos‐ ened by receiving permission to start a business without a Meister degree after having reached the level of an Altgeselle, i.e., by having proven 6 years of work experience as a Geselle, four of these in a decision-making position, in his or her prospective occupation. This Altgesellen rule de‐ fines the third treatment group (A1-occupations) which includes profes‐ 3 3.1 IV Past Reforms in the Services Sector and their Effects 94 sions such as roofers, surgical instrument makers, gunsmiths, plumbers, gas and water fitters, joiners, and pastry cooks. Workers in A1-occupations can start a business without providing proof of any qualification, provided they commit to limiting the range of their activities to tasks that can be learned within 3 months. This partial exemp‐ tion from the already reduced entry regulation aims particularly at sup‐ porting the establishment of small businesses. However, for a prospective entrepreneur who plans to carry out the full range of activities, obtaining vocational training according to the Altgesellen rule is still mandatory. In‐ dividuals in occupations that use this so-called easy-job-rule are grouped separately into the A2 group (cf. Müller 2006), including masons and con‐ creters, painters and varnishers, metalworkers, motor vehicle body and ve‐ hicle construction mechanics, bike mechanics, information electronics technicians, vehicle technicians, and butchers. In summary, the three treatment groups are described in descending or‐ der of their expected treatment intensity: the B1-, A2-, and A1-occupa‐ tions, while the AC-occupations are used as the control group. Having de‐ fined the three treatments and the control group, I describe in the follow‐ ing the development of the level of self-employment and self-employment rates for these groups. Trends in craftsmanship Between 2002 and 2009, the period relevant for this analysis, the number of self- employed craftsmen remained stable in the control group, while this number increased in the treatment group that has experienced the strongest treatment, i.e., the B1-occupations (see Figure 4.2.1), after the reform in 2004. This growth pattern can also be observed for the A1 and A2 groups, though it is less pronounced. In contrast to the B1 group’s al‐ most monotonic increase, the number of A2 craftsmen reverted to its prepolicy level. The number of A1 craftsmen also declined from 2007 to 2008 but nevertheless remained at a substantially higher level than before the reform. These facts may indicate that the reform had a positive impact on the self-employment rate in the treatment groups. 3.2 IV.2 Entry regulation and entrepreneurship 95 Self-employment in treatment groups and control group: number of self-employed craftsmen in B1, A1, A2, and AC occupations in thousands Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Figure 4.2.2 depicts the time trends in the self-employment rates, defined as the ratio of the number of self-employed craftsmen to the number of both self-employed and employees in the treatment groups and the control group, respectively. Figure 4.2.1: IV Past Reforms in the Services Sector and their Effects 96 Self-employment rates in treatment groups and control group: percentage share of self-employed among B1, A1, A2, and AC occupations Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Before the policy change in 2004, the differences between the time trends of the treatment groups and the control group remained steady. In subse‐ quent years, however, the differences between the self-employment rates of each of the treatment groups and the control group decreased substan‐ tially. This may again support the hypothesis that the 2004 reforms in‐ creased the probability of self-employment for the treatment groups. Note that the dip in the share of self-employed craftsmen in AC-occupations is due to a temporary increase in the number of employees. See Sect. 5.4 for a robustness check that shows that the results are not driven by this fluctu‐ ation. Interestingly, the trajectory prior to the treatment is very similar for craftsmen compared to non-craftsmen; in my sample, the rate of self-em‐ ployed among working persons raised from 11.17 % in 2002 gradually to 12.00 % in 2004. In 2005 this figure peaked at 12.44 % and remained thereafter relatively steady around 12 % (see Table 4.2.9). One explanation Figure 4.2.2: IV.2 Entry regulation and entrepreneurship 97 for this development might be the substantial increase in the number of self-employed in East-Germany. Identification of causal effects The empirical strategy outlined here exploits the reform of the regulatory framework of entrepreneurial craftsmen in 2004 as a natural experiment. To this end, I calculate the differences in the changes in average outcomes of employment status choices across each treatment group both before and after the reform. I then measure the changes in average outcomes of em‐ ployment status decisions of the control group before and after the reform. The differences in these changes is known as the DID estimator, and rep‐ resents the average treatment effects on the treated group (ATT) (e.g., Blundell and Costa Dias 2009). I use data from 2002 to 2009 for the three occupational groups (B1, A1, A2) subject to different intensities of regulation changes, as detailed in Sect. 3.1. These three groups are used as the treatment groups (cf. Meyer 1995) while the group of AC-occupations is used as the comparison group. To determine the ATT with the DID approach means specifically comparing the difference in the average self-employment probability of each of the three treatment groups before and after the reform with the av‐ erage self-employment probability of the AC-occupation group before and after the reform. Therefore, for this 8-year period, I have been able to quantify the ef‐ fects of the reform on the probability of self-employment. The main hypo‐ thesis, based on the theory of public choice, suggests that the policy change could have influenced the self-employment rate negatively or not at all. However, the direction of the effect depends on how the new policy has caused the entry and exit rates to change. Generally, an increase in the self-employment rate could result from either a higher entry rate, a lower exit rate or both. However, an increase could also result from a higher exit rate, which in turn is exceeded by an even higher entry rate. Another pos‐ sibility is that the self-employment rate overall remains unchanged if the policy shifts the entry rate as well as the exit rate equally in the same di‐ rection or has no effect at all. Therefore, with this analysis, I investigate not only the probability of being self-employed but also the probability of entry into and exit from self-employment. 3.3 IV Past Reforms in the Services Sector and their Effects 98 Identifying the ATT using the DID approach requires the assumption that the treatment groups and the control group are subject to common trends. This implies that macro shocks exert the same effects on both groups. For example, a sudden decrease in the interest rate should influ‐ ence trades related to health and hygiene, which are common among the AC group, just as it does the building and construction trades, which are part of the A2 group. If this is true, a hypothetical trend without a reform in the treated group would parallel the trend in the control group in the post-policy period. Otherwise, it would be unclear whether differences be‐ tween these groups are caused by the reform or by other factors. Section 5.4 provides evidence in favor of the identifying common trend assump‐ tion. Furthermore, this setting does not seem to be susceptible to what is a frequent concern in natural experiments. That is, the problem of self-selec‐ tion should not exist, because the different treatment groups are distin‐ guished by a law that was proclaimed for the first time in March 2003 (cf. Müller 2006), resulting in a relatively short time for workers to adjust and change occupations. Work in a specific vocation in craftsmanship, like individual character‐ istics, changes little over time. In the sample used for the estimation, 73.85 % of the individuals in B1-occupations had been working in their current occupation for 3 years or more in 2004 and 72.80 % in 2008. For the other groups of craftsmen, this figure is larger. Self-employed crafts‐ men tend to be less likely to change occupations. Again, the B1 group was the most dynamic, though in this group 83.42 % had run their business for 3 years or more in 2004 and 82.18 % in 2008. Therefore, adjusting behav‐ ior in expectation of the reform should not challenge the identification of the ATT parameter. Moreover, after the announcement of plans for the amendment to the HwO, a controversy arose with an unpredictable out‐ come. It was therefore not known what intensity of treatment each occupa‐ tion would receive before the reform actually came into being. Consider‐ ing this unpredictability, it seems unlikely that craftsmen would have changed jobs in anticipation of the effects caused by complicated new rules. Regarding changes between groups, the situation after the regulations were eased is somewhat different, as the B2-occupations could be substi‐ tuted for similar B1 or A-occupations more easily, which means that the compositions of the treatment and control groups might change systemati‐ cally. For instance, changing from a B2-occupation to engage in self-em‐ IV.2 Entry regulation and entrepreneurship 99 ployment in a B1-occupation might have been harder for an individual not having obtained the required degree before the reform. Conversely, a craftsman trained in a B1-, or A-occupation might have been more likely to move into an occupation from the B2-vocations because she wants to set up a business before the reform. This would bias the estimate of the treatment if these changes occur in anticipation of the reform. Moreover, the analysis includes a set of observable, time-varying co‐ variates and other characteristics to control for the potential for systematic differences in the populations over the two periods. I assume that changes in unobserved factors are the same between the treatment and control groups. Other entrepreneurship policies Some other major policies may also have interfered with the effects of the policy change. These are the enlargements of the European Union (EU) in 2004 and 2007 as well as some subsidies for entrepreneurship. The first relevant enlargement of the EU based on the 2003 Treaty of Accession took place in 2004, when ten countries became new member states. Moreover, the 2007 enlargement of the EU based on the 2005 Treaty of Accession saw Bulgaria and Romania join the EU. Although Germany restricted its labor market from workers from these 12 new member states, exceptions were granted to specific groups. Most impor‐ tantly, a person was permitted to engage in entrepreneurship immediately after her state of origin became member of the EU. Other important policy instruments include subsidies to entrepreneurs, such as the transitional allowance (Überbrückungsgeld, 1986–2006), the start-up subsidy (Existenzgründungszuschuss [EXGZ], 2003–2006), the entrance grant for entrepreneurs (Einstiegsgeld für selbständige Tätigkeit, since 2005), and another start-up subsidy (Gründungszuschuss, since 2006) (cf. Caliendo and Steiner 2005, Caliendo and Künn 2011). The years in which each of the programs was adopted and the year of its aboli‐ tion is given in parentheses. According to Baumgartner et al. (2006), the EXGZ in particular had significant effects on entrepreneurship, and thus could confound the main analysis. Although there are no reliable numbers, a surmise based on Müller (2006) would imply that just 2.93 % of the A-businesses established in 2004 received the EXGZ, and 2.13 % in 2005. For B1-businesses, less 3.4 IV Past Reforms in the Services Sector and their Effects 100 than 5.79 % of the start-ups in 2004 and 3.58 % in 2005 were subsidized by the EXGZ. This suggests that we should not be too concerned about the effects of these subsidies. Craftsmanship and entrepreneurship policies: total, unsub‐ sidized, and German self-employed craftsmanship in thou‐ sands Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Figure 4.2.3 shows three graphs from 2002 to 2009: the development of total self- employed craftsmanship, the number of craftsmen who did not report receiving SPP payments (a dummy for public payments for self-em‐ ployed), and the number of German self-employed craftsmen. All three se‐ ries experienced a substantial increase after the amendment to the HwO came into effect. The number of self-employed craftsmen jumped from 518,163, measured a year before the reform, to 579,036 in 2005 and then to 584,494 in 2006. This enormous change is also documented for the stock of businesses, with data taken from the register of craftsmen. Disre‐ garding the B2-occupations, these figures equal, in each year, approxi‐ mately 90 % of the stock of businesses reported in Table 4.2.8 which con‐ firms how well these occupations are represented in the data. Note that Figure 4.2.3: IV.2 Entry regulation and entrepreneurship 101 this result holds after accounting for the actual stock of businesses, which is approximately 15 % lower than the reported stock. Together with the number of self-employed craftsmen, the graphs for unsubsidized, self-employed craftsmen and for German self-employed craftsmen evolve almost uniformly over time, though the effects of the 2007 enlargement of the EU is clearly visible. This suggests again at least that the subsidies did not affect the number of self-employed craftsmen systematically. However, to identify the effect of the amendment to the HwO separately from these policies, I include a dummy indicating EU cit‐ izenship and its interaction with the post-policy period in most of the spe‐ cifications. Moreover, in Sect. 5.3, I discuss the results, first by excluding all non-German craftsmen and then by excluding all craftsmen that receive any subsidy. Estimation procedure In estimating the effects of the reform for all treatments with repeated cross-sections from 2002 to 2009, all three treatment groups are included jointly in the regression models to yield more precise estimates. I present estimates of logit models using the maximum likelihood esti‐ mator in much of the rest of the paper, because predicted probabilities are not bounded by 0 and 1 and the partial effects of independent variables are constant in the linear probability model (LPM). However, I also employed LPMs for all of these specifications and the results remain essentially the same. In an LPM, the ATT equals the coefficient of the interaction term be‐ tween the treatment and the post-policy dummy. This interaction effect re‐ flects the comparison of the changes in predicted probabilities before and after the reform for the treatment and control groups. In a logit model, the outcome variable is assumed to be determined by the logistic function, and thus the model is nonlinear. In turn, the coeffi‐ cient of the DID interaction cannot be interpreted as the ATT, and the ef‐ fects of the reform must be computed as differences of predicted probabil‐ ities. The corresponding standard errors for the predicted probabilities can be obtained by applying the delta method. The dependent variable Yi for observation i is a binary variable that in‐ dicates self-employment in the stock models, and transition into or out of self-employment in the flow models. The conditional expectation of the 3.5 IV Past Reforms in the Services Sector and their Effects 102 binary outcome equals the probability Prob(·). In the main specification, given as Eq. (1) below, the regressors dPosti, d Oi, and Xi are included in zi, where dPosti is a dummy variable for individuals observed in the postpolicy period;60 d Oi = d B1i, d A1i, d A2i indicates an individual’s affilia‐ tion to one of the treatment groups; and X is the vector of control vari‐ ables. The specification includes interaction terms between the respective treatment group indicators and the post-policy dummy. Moreover, 0,  δω,  βω , and β4 , along with ω  =  B1,  A1,  A2 , represent the respective coeffi‐ cients or vector of coefficients, and β0  is a constant. Prob Yi  =  1 dPosti,  dOi,  Xi  = 1 1 + e−zi   with zi = β0 + δ0dPosti + βB1dB1i + βA1dA1i + βA2dA2i + δB1dB1i · dPosti + δA1dA1i  · dPosti + δA2dA2i · dPosti + Xiβ4 .   (1) In addition to dummies for the years 2003, 2004, 2006, 2007, 2008 and 2009, all models include in X variables for the following individual char‐ acteristics: age and its square, and dummy variables indicating gender, type of secondary schooling and professional qualification, nationality, re‐ gion of residence, the size of the respondent’s city of residence, marital status, number of dependent children, the branch of craftsmanship, the oc‐ cupation, and a constant. The included indicator d EU shows the citizen‐ ship of foreigners in an EU member state, and is included along with its interaction with the post-policy period, to separate the effects of the en‐ largements of the EU from the effects of the amendment to the HwO, as discussed in Sect. 3.4. Controlling for these characteristics is important for two reasons. First, the determinants of self-employment may have changed over the time. Second, including these control variables allows to obtain the estimate δω  more efficiently. The estimation sample consists of all craftsmen in a given year in the models for which the dependent variable is the self-employment probabili‐ 60 The post-policy period could be defined as the period from 2004 to 2009. How‐ ever, the data from 2004 refer to the beginning of this year, which basically repre‐ sents the status quo ante, so the post-policy period in the main specifications in‐ cludes only the years 2005 and 2009. Results from a specification where the postpolicy period is defined from 2004 to 2009 or 2004 is dropped are shown in Table 4.2.11. The post-policy dummy equals 1 for both years, which prevents the inter‐ action effect from differing in the post-policy periods. A more flexible specifica‐ tion is presented in Table 4.2.12 and discussed in Sect. 5.4. IV.2 Entry regulation and entrepreneurship 103 ty. The same population is used in the entry models. Note that employ‐ ment status in the previous year, used for the construction of the transition variables, is queried retrospectively, and it is not mandatory to respond. In contrast, the indication of the current employment status, which is used for the transition variables and the stock variable, is found in the mandatory section of the questionnaire. Moreover, some unemployed or inactive persons do not report a profes‐ sion, and it is thus unclear what proportion of these groups participates as a reserve in the labor market for craftsmen. Because the analysis excludes those who do not report an occupation, the results reflect an approxima‐ tion of the probability of entering self-employment from dependent em‐ ployment, unemployment, or inactivity, because not all potentially selfemployed persons are included in the estimation sample. In contrast, the estimation sample of the exit models comprises self-em‐ ployed craftsmen in the previous year. Therefore, it is the population that possibly could exit from self-employment within the given year. With this sample, it is appropriate to estimate the probability of exit, because the de‐ pendent variable clearly indicates whether a person is not self-employed after 12 months, but instead is an employee, unemployed, or inactive. Apart from these differences in the estimation population and the depen‐ dent variables, the econometric framework is identical in the stock models and the flow models. Data and descriptives Sample design This analysis uses data from the German microcensus (Mikrozensus), which is provided by the Federal Statistical Office. This official, represen‐ tative yearly household survey is comparable to the Current Population Survey in the United States and the Labour Force Survey in the United Kingdom. The German microcensus is a 1 % sample of all households in Germany. A subsample of 70 % or approximately 494,000 observations per year, is selected at random and provided to researchers as a scientific use file by the Federal Statistical Office. The large sample size is crucial to this analysis, because less than 10 % of the population are craftsmen. Most questions are compulsory; therefore, the German microcensus, a mandato‐ 4 4.1 IV Past Reforms in the Services Sector and their Effects 104 ry census, guarantees a low rate of item non-response and ensures that en‐ trepreneurs are adequately represented. This analysis uses pooled cross-sections of the German microcensus from 2002 to 2009. The years before 2002 are not considered for several reasons. First, effects of other policy changes, e.g., the amendment to the HwO from 1998, could still be significant at the beginning of 2001, inso‐ far as the process of adjusting expectations and changing occupations in response to the reform took some time. Second, training in some tradition‐ al occupations, such as blacksmiths and turners, ceased as of August 2002, superseded by more modern training structures with new fields of special‐ ization. However, Müller (2006) shows empirical evidence that suggests that these changes had no substantial effect on the transition rates. To avoid confusion due to these influences, I excluded the year 2001 from the analysis. Table 4.2.11 shows that the estimates from the main specification using the years 2001 to 2009 remain similar if 2001 is included. Other re‐ sults including 2001 are available on request. The transition variables reflect questions from the supplementary pro‐ gram that ask retrospectively for a person’s employment status in the year before the interview. Note that the supplementary questions were only posed to a 45 % random subsample of the microcensus up until 2004. Since the number of observations is still quite large, this does not influ‐ ence further analysis. However, this program is non-mandatory and there‐ fore non-response is higher. Indication of status as self-employed is used to measure entrepreneur‐ ship in German craftsmanship, because the HwO refers explicitly to selfemployment. While the majority of self-employed craftsmen run non-in‐ corporated businesses, the term self-employment can cover also incorp‐ orated businesses. The Appendix provides a description of how the key variables are constructed. The reader should keep in mind that indication of status as self-employed is based on self-assessment and thus is rather a proxy for the true number of self-employed. Because the focus of this study is on entrepreneurship among German craftsmen, I restrict the sample as follows, reporting the average number of dropped observations per year in parentheses: I exclude all individuals younger than 18 years, or older than 65 years (177,740). People whose employment status choice is determined by different factors are also omit‐ ted from the sample to avoid distortions. Thus, civil servants (11,978), ap‐ prentices (7,885), soldiers (968), conscripts (730), persons in education, or those drafted in the previous year (13,531 and 336, respectively), as well IV.2 Entry regulation and entrepreneurship 105 as all remaining non-craftsmen (254,571), are excluded. Moreover, family workers (1,981) helping in a family business are not included in the sam‐ ple, because they are not entrepreneurs in the sense that they run their own businesses. This process leaves me with a sample of about 25,000 obser‐ vations per year, which represent about 4 million craftsmen in the German population. To complete the picture, the following section shows how the transition variables used in the estimation evolved over time, and de‐ scribes the characteristics of the occupational groups. Descriptives Figure 4.2.4 shows how the number of B1 entries increased tremendously after 2004, returned to a somewhat lower level in 2006, peaked in 2007 and reverted in 2008, but still remained higher than in the period before the reform. The exits remained constant for a time, before declining in the aftermath of the policy change. Note that the balance (defined as entry-ex‐ it) exhibits a similar, though less wiggly, path than the number of self-em‐ ployed craftsmen in Figure 4.2.3, which implies that most of the variation stems from this particular group. The two peaks in 2005 and 2007 might reflect the effects of the enlargements of the EU on the entry rate on top of the effects of the reform to the HwO. A comparison of the path of the growth rate, measured as the annual change in the number of self-em‐ ployed craftsmen in percentage, and the balance shows how large the nonresponse bias in the transition variables is, since both variables should contain the same information. Indeed, in almost all of the graphs in Figs. 4.2.4, 4.2.5, 4.2.6, and 4.2.7, the growth rate seems to resemble the pattern of the balance, though only very roughly. Figure 4.2.5 illustrates that nei‐ ther the entries into nor the exits from the AC-occupations exhibit any sin‐ gularity until 2006. The subsequent peak might again stem from the en‐ largement of the EU. The path of the growth rate and balance correspond. Apparently, the numbers of entries and exits are both rather small. For sensitivity tests correcting for rare events see Sect. 5.4. 4.2 IV Past Reforms in the Services Sector and their Effects 106 Entries into and exits from self-employment and their differ‐ ence among B1-occupations, left ordinate number in thou‐ sands, right growth rate in percent Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Entries into and exits from self-employment and their differ‐ ence among AC-occupations, left ordinate number in thou‐ sands, right growth rate in percent Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Figure 4.2.4: Figure 4.2.5: IV.2 Entry regulation and entrepreneurship 107 In Figure 4.2.6, the transition variables do not exhibit any major oscilla‐ tion. In the post-policy period, the growth rate increases substantially and then slows down, but the balance contrasts with this development. Entries into and exits from self-employment and their differ‐ ence among A2-occupations, left ordinate number in thou‐ sands, right growth rate in percent Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) The series of transitional variables for A1-occupations, depicted in Figure 4.2.7, show that the entries increase modestly, whereas the exits remain roughly constant. Here, the balance series and the growth rate also show an increase in 2005 and a subsequent decrease in 2006. Again, entries peak in 2007. Figure 4.2.6: IV Past Reforms in the Services Sector and their Effects 108 Entries into and exits from self-employment and their differ‐ ence among A1-occupations, left ordinate number in thou‐ sands, right growth rate in percent Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Now that we know how the dependent variables developed, I will describe some of the characteristics of the four occupational groups included in the vector of control variables. Furthermore, I will show the share of self-em‐ ployed craftsmen among all craftsmen in each group, and the share of selfemployed craftsmen in each group among all self-employed craftsmen in Table 4.2.2 as weighted averages from the pooled cross-sections from both the pre-policy period (2002–2004) and from the post-policy period (2005– 2009). In all three treatment groups, the share of self-employed is higher after the reform than before, while this figure seems to remain constant in the control group. Again, this points to a positive effect of the reform. Figure 4.2.7: IV.2 Entry regulation and entrepreneurship 109 Weighted averages by treatment and control groups in preand post-reform (2002–2004; 2005–2009) samples B1 A1 A2 AC Pre Post Pre Post Pre Post Pre Post Self-employed (%) 7.87 9.69 15.96 17.67 12.89 14.05 19.94 19.47 Female (%) 58.81 58.97 16.90 17.59 3.08 2.82 41.24 43.80 Age (a) 42.77 43.66 38.97 39.82 39.46 40.14 38.99 40.13 East (%) 16.47 17.77 21.50 21.70 23.44 23.28 17.71 17.41 Nationality German (%) 80.89 80.60 90.17 90.66 90.84 90.50 95.95 96.14 EU (%) 4.64 6.36 3.57 3.98 3.04 3.98 1.80 1.90 Non-EU (%) 14.46 13.03 6.26 5.36 6.12 5.52 2.25 1.96 Professional qualification University (%) 1.08 1.31 0.78 0.99 0.25 0.41 1.05 0.71 UASa (%) 0.94 1.13 1.23 1.34 0.53 0.55 1.52 1.54 Meisterb (%) 5.64 5.24 17.99 17.64 16.37 17.41 27.23 28.85 Gesellec (%) 54.32 59.21 65.67 70.15 69.96 72.70 62.46 65.11 None (%) 31.09 32.60 8.50 9.41 7.04 8.45 2.38 3.51 Non-response (%) 6.92 0.52 5.84 0.47 5.85 0.49 5.37 0.28 Secondary school Abiturd (%) 4.88 5.78 4.74 5.44 2.65 3.23 13.98 18.06 Othere (%) 84.00 85.73 89.58 91.78 91.47 93.92 82.02 81.39 None (%) 5.76 7.33 1.54 2.01 1.52 2.12 0.22 0.21 Non-response (%) 5.36 1.16 4.14 0.76 4.36 0.73 3.78 0.34 Children under 16 (#) 0.72 0.64 0.68 0.61 0.65 0.61 0.59 0.55 Married (%) 70.50 68.26 60.01 57.36 60.14 57.81 57.68 57.99 City size >500,000 (%) 14.30 14.81 10.89 11.96 10.16 11.25 11.78 13.83 20,000–500,000 (%) 44.93 46.77 38.80 42.05 37.93 41.10 43.09 44.70 ≤20,000 (%) 40.77 38.42 50.31 45.99 51.91 47.65 45.13 41.47 % of all self-employed craftsmen 24.11 27.20 46.92 46.67 23.13 21.27 5.84 4.86 Observations 28,188 47,002 27,424 44,675 16,733 25,553 2,792 4,302 Notes: All numbers are weighted by survey weights provided by the microcensus. Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) a University of Applied Sciences b The Meister-craftsman degree certifies the highest professional qualification in craftsmanship. c The Geselle degree can be obtained by completing an apprenticeship. Table 4.2.2: IV Past Reforms in the Services Sector and their Effects 110 d Abitur refers to the higher secondary school degree that qualifies a student for univer‐ sity admission in Germany. e Other secondary school refers to a secondary school degree that does not qualify a student for university admission in Germany, typically obtained at a Realschule or a Hauptschule. A remarkable difference between the treatment groups is that the A2 group has almost no female workers, while the majority of B1 jobs are done by women. Another interesting point is that individuals working in a B1 vocation rarely engage in self- employment compared to the other groups. This is accounted for in the estimation by including the binary variables dOi . Moreover, it is noteworthy that persons working in a B1occupation are on average less qualified, as around 1/3 reports no profes‐ sional qualification. Further, the share of craftsmen that served as apprentices, and thus held the vocational degree Geselle, is substantially higher for the post-policy period across all four groups. However, even though the documented in‐ creases are large, in particular for the (B1- and) A1-occupations, one should be cautious about attributing this to the effects of the Altgesellen rule as the changes might simply reflect the fact that the survey response probability increased after the reform. Section 5.3 picks up on this in a de‐ tailed discussion of heterogeneous effects with respect to gender and dif‐ ferent levels of vocational training. Results Did the 2004 amendment to the HwO have the intended effects? Accord‐ ing to the plain DID results from an LPM using pooled cross-sections from 2002 to 2009, shown in the second column of Table 4.2.3, the answer for the B1-occupations is yes. A glance at the coefficient of the interaction term reveals that the reform increased the probability of entering self-em‐ ployment significantly, by 0.79 percentage points. This result does not change significantly when year and branch dummies and further control variables are added (third column of Table 4.2.3). Moreover, including the A1- and A2-occupations in the sample shows that while the A2-occupations seem not to be significantly affected (fourth column), the probability of entering self-employment is 0.69 percentage points higher for the A1-occupations. Note the large significant coefficient of the interaction of the EU dummy and the post-policy period, underlin‐ 5 IV.2 Entry regulation and entrepreneurship 111 ing the importance of controlling for the enlargements of the EU. This co‐ efficient shows that the 2004 enlargement of the EU raised the probability of entry by 1.52 (1.22) percentage points according to column three (four). In Sect. 5.3, I demonstrate that the principal results remain unchanged af‐ ter all non- German craftsmen are excluded from the sample. Table 4.2.10 presents the same specifications as used in Table 4.2.3, employing logit models. The estimates tell a consistent story: the signs of the interaction terms are the same across models, and, apart from the coef‐ ficient of the interaction between the post-policy dummy and the indicator for A2-vocations and between the post-policy and the EU dummies, the same interactions are statistically significant. Here, the functional form might help to identify the coefficients of the treatment interactions more precisely, whereas the coefficient of the post-policy period’s interaction with the EU dummy becomes insignificant at the 10 % level in column four of Table 4.2.10. While entry probabilities increased, the reform may have raised exit probabilities in the same way. This finding would be consistent with the view that a major portion of new entrepreneurs in the post-policy period use fly-by-night tactics, i.e., they set up a company, do business for a short period and then disappear suddenly. However, the results reported in col‐ umn five of Table 4.2.10 suggest rather that the policy change generated quite a sustainable number of start-ups. The negative, though highly in‐ significant point estimates for the interaction terms of the B1-occupations point to an interpretation that exit probabilities remained constant or may even have fallen in the post-policy period. For the A1 and the A2 group, the coefficients are negative and insignificant as well. Similarly, the corre‐ sponding LPM estimates shown in column five of Table 4.2.3 are insignif‐ icant throughout. Higher entry probabilities and roughly steady exit probabilities would imply that the stock of self-employed craftsmen should be higher after the reform. And indeed, the last column of Table 4.2.10 presents estimates that are in line with the earlier findings. The interaction term of being selfemployed has a significant positive coefficient for both the B1- and A1vocations, the coefficient for the A2 group is also positive, though in‐ significant. The results of the LPM presented in column six of Table 4.2.3 are again consistent with the logit estimates. IV Past Reforms in the Services Sector and their Effects 112 Estimation results of self-employment state and transition probabilities LPM Entry LPM entry LPM entry LPM exit LPM self-employed dB1 × dPost 0.0079*(0.0043) 0.0070* (0.0042) 0.0077* (0.0041) −0.0285 (0.0240) 0.0227* (0.0118) dA1 × dPost 0.0069*(0.0039) −0.0103 (0.0145) 0.0266** (0.0101) dA2 × dPost 0.0055(0.0038) −0.0067 (0.0136) 0.0161 (0.0110) dEU × dPost 0.0152**(0.0060) 0.0122** (0.0052) −0.0585 (0.0395) 0.0539*** (0.0164) dB1 −0.0161*(0.0081) −0.0012 (0.0058) −0.0019 (0.0050) 0.0330*** (0.0116) 0.0094 (0.0167) dA1 0.0060(0.0038) −0.2727*** (0.0137) 0.0407*** (0.0063) d A2 −0.0303***(0.0038) 0.0758*** (0.0106) 0.0133 (0.0105) dPost −0.0036(0.0037) −0.0005 (0.0038) −0.0018 (0.0036) −0.0182 (0.0123) −0.0083 (0.0088) dEU 0.0007(0.0038) 0.0034 (0.0036) 0.0039 (0.0449) 0.0200* (0.0106) Constant 0.0280***(0.0066) 0.0128 (0.0091) 0.0337*** (0.0083) 0.7161*** (0.0420) −0.2571*** (0.0470) Year dummies yes yes yes yes Occ. Dummies yes yes yes yes Branch dummies yes yes yes yes Controls yes yes yes yes Adj-R2 <0.01 0.03 0.02 0.08 0.26 Observations 64,842 64,842 154,940 17,211 196,669 Notes: Robust standard errors, clustered by occupation, are given in parentheses below the coefficients of the linear probability model (LPM). Controls included are age and its square, and dummy variables indicating gender, type of secondary schooling and professional qualification, nationality, region of residence, the size of the respondent’s city of residence, marital status, number of dependent children, citizenship of foreign‐ ers in an EU member state and its interaction with the post-policy period. Moreover, year dummies for 2003, 2004, 2006, 2007, 2008, and 2009, and indicators for the branch of craftsmanship, for the occupation, as well as a constant are included. Signifi‐ cance of the logit coefficients is indicated at the 10 %/5 %/1 % level by asterisks (*/**/ ***). Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.3: IV.2 Entry regulation and entrepreneurship 113 Treatment effects on transition probabilities To ascertain the quantitative effect of the amendment on the probability of entry and exit, I first predict the respective probabilities of a person with average characteristics before and after the reform, using the estimates from the preferred logit models reported in Table 4.2.10. Having obtained these, in a second step I calculate their differences. For the entry probabili‐ ties the results are reported in Table 4.2.4. The expected probabilities for each of the three treatment groups and for the control group before the re‐ form are shown in columns two to five of the first row, with their standard errors below. The same figure for the period after the reform is shown in columns two to five of the third row. The last three columns of row one and three report the differences in the expected probabilities of each of the treatment groups and the control group before and after the reform. Columns two to five of the last two rows in the upper panel present the differences in the same occupational group before and after the reform along with their standard errors. Finally, the last three columns show the difference in these differences, i.e., the cross differences. The lower panel shows how the counter-factual cross differences are obtained (see Puhani 2012). While the row displaying the expected proba‐ bilities before the reform is identical to the corresponding row in the upper panel, the expected probabilities for the post-reform period are predicted to constrain the reform’s effect to zero. Then, at the bottom of the table, the average treatment effects on the treated, i.e., the differences in the ac‐ tual cross differences from the upper panel and the counter-factual cross differences from the lower panel, are reported, in both absolute and rela‐ tive terms. The first thing that leaps out is that the probability of engaging in en‐ trepreneurship for individuals of the B1 group is substantially lower than that of the other occupational groups before the reform. From this compa‐ rably lower level, the entry probability resulting from the reform increased by 0.15 % points. This economically relevant effect is also statistically sig‐ nificant, with a standard error of 0.04 (p value <0.01). The probability of entering self-employment would have been 0.40−0.15 = 0.25 in the hypo‐ thetical situation without a reform. This shows that the entry probability has been increased dramatically with the reform; its relative effect amounts to 60.00 %. Effects of this kind are found in the A1- and A2-professions, too. The former group experienced an increase in the probability of entry of 0.56 5.1 IV Past Reforms in the Services Sector and their Effects 114 percentage points. This increase is significantly different from zero, with a standard error of 0.26 (p value 0.03). Consequently, this suggests that the opportunity to start a business without the Meister certificate provided by Probabilities of entry into self-employment (in %): differ‐ ence-in-differences B1 A1 A2 AC ΔB1 ΔA1 ΔA2 Panel A Before reform 2004 0.25*** (0.02) 1.86*** (0.16) 2.38*** (0.20) 3.73**** (0.49) −3.48*** (0.50) −1.88*** (0.45) −1.35*** (0.42) After reform 2004 0.40*** (0.02) 2.39*** (0.12) 3.01*** (0.16) 3.68*** (0.30) −3.28*** (0.31) −1.29*** (0.22) −0.67*** (0.19) Δ Between after and before reform 2004 0.14*** (0.02) 0.53*** (0.18) 0.63*** (0.21) −0.06 (0.48) 0.20 (0.48) 0.59 (0.47) 0.69 (0.49) Panel B Before reform 2004 0.25*** (0.02) 1.86*** (0.16) 2.38*** (0.20) 3.73*** (0.49) −3.48*** (0.50) −1.88*** (0.45) −1.35*** (0.42) After reform 2004 0.25*** (0.04) 1.83*** (0.25) 2.35*** (0.33) 3.68*** (0.30) −3.43*** (0.30) −1.85*** (0.30) −1.33*** (0.29) Δ Between after and before reform 2004 0.00 (0.03) −0.03 (0.24) −0.04 (0.31) −0.06 (0.48) 0.05 (0.45) 0.03 (0.24) 0.02 (0.17) Panel C Difference-in-differences 0.15*** (0.04) 0.56** (0.26) 0.67** (0.34) Relative difference-in-differences 60.00 30.60 28.63 Notes: Panel A shows the expected probabilities for the treatment groups (B1, A1, A2) and for the control group (AC) of a person with average characteristics before and after the reform, rounded to two decimal places. Moreover, it depicts the differences in the expected probabilities and the difference in these differences, i.e., the cross differences. The next part of the table shows how the counter-factual cross differences are obtained using the expected probabilities for the post-reform period, which result when the re‐ form’s effects are restricted to zero. Panel C reports the ATT, i.e., the differences in these cross differences. The relative differences in differences are computed, respec‐ tively, as the fraction of the treatment effect and the expected probability in the postpolicy period subtracted by the treatment effect. The same calculation, based on the av‐ erages of the respective probabilities among actual persons in the data instead of the expected probabilities of a person with average characteristics, yields similar results and is available upon request Cluster (occupation) robust standard errors calculated by the delta method are in parentheses. Asterisks (*/**/***) denote significance at the 10 %/5 %/1 % level Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.4: IV.2 Entry regulation and entrepreneurship 115 the Altgesellen rule has been used extensively in this group. The results further show that the craftsmen in A2-occupations responded to the reduc‐ tion and partial exemption of the entry barrier with an increase of 0.67 percentage points, which is significant with a standard error of 0.34 (p val‐ ue 0.05). In relative terms, the reform increased the entry probability of the A1 group by 30.60 %, while for A2-occupations, the entry probability was 28.63 % higher than the hypothetical situation without the reform. How sustainable are these entries? In Table 4.2.5, I present results that support the hypothesis that the amendment of the HwO did not signifi‐ cantly alter the probability of exit from self-employment. The reform’s ef‐ fect for the B1-occupations is −1.77 percentage points, with a standard er‐ ror of 3.19. This negative effect is insignificant (p value 0.58). Similarly, the effect of −0.22 percentage points for A1-vocations is highly insignifi‐ cant, with a standard error of 0.38 (p value 0.57). Thus, more sustainable business entries could be established after the deregulation. The results suggest that this is due to the reform, and support the findings in Prantl (2012) that this entry regulation suppressed long-living entrants. For A2occupations, the treatment effect of −0.11, though insignificant with a standard error of 0.37 (p value 0.76), points to a rather small decrease in the exit rate caused by the amendment. In fact, the point estimate is even positive once 2009 is excluded from the sample. One reason for this could be that in this group fly-by-night strategies might be more common. These, in turn, could be encouraged by the combination of the Altgesellen rule and the partial exemption for small businesses. For instance, splitting a firm up into one that runs the main business and another that serves as an ancillary business makes it easy to once more absorb the smaller one when it becomes convenient. However, on top of the fact that none of the effects on the exit probabilities is significant, the relative effects are rather small using the sample up to 2006 or 2008. Including 2009, they are −17.00, 20.37, and −11.83 % for the B1-, A1-, and A2-vocations, respectively. Treatment effects on self-employment probabilities As discussed above, the higher entry rates, together with constant exit rates, should raise the stock of self-employed persons. In fact, Table 4.2.6 shows that after accounting for the counter-factual situation without the reform, a person with average characteristics in a B1-occupation is 0.41 percentage points more likely to engage in entrepreneurship. This effect is 5.2 IV Past Reforms in the Services Sector and their Effects 116 significant, with a standard error of 0.11 (p value <0.01). The effect on the A1-occupations is larger. The probability of being self-employed in‐ Probabilities of exit from self-employment (in %): differencein-differences B1 A1 A2 AC ΔB1 ΔA1 ΔA2 Panel A Before reform 2004 19.69*** (3.23) 2.24*** (0.30) 1.94*** (0.29) 0.67*** (0.16) 19.02*** (3.21) 1.57*** (0.31) 1.27*** (0.31) After reform 2004 8.64*** (0.92) 0.86*** (0.07) 0.82*** (0.10) 0.32*** (0.04) 8.33*** (0.93) 0.54*** (0.08) 0.50*** (0.10) Δ Between af‐ ter and before reform 2004 −11.05*** (2.92) −1.38*** (0.36) −1.13*** (0.34) −0.35* (0.19) −10.70*** (2.88) −1.03*** (0.37) −0.78** (0.38) Panel B Before reform 2004 19.69*** (3.23) 2.24*** (0.30) 1.94*** (0.29) 0.67*** (0.16) 19.02*** (3.21) 1.57*** (0.31) 1.27*** (0.31) After reform 2004 10.41*** (3.31) 1.07*** (0.36) 0.93*** (0.35) 0.32*** (0.04) 10.09*** (3.28) 0.75** (0.33) 0.61** (0.31) Δ Between af‐ ter and before reform 2004 −9.28** (3.81) −1.16*** (0.43) −1.01*** (0.36) −0.35* (0.19) −8.93** (3.64) −0.81*** (0.26) −0.66*** (0.19) Panel C Difference-in-differences −1.77 (3.19) −0.22 (0.38) −0.11 (0.37) Relative difference-in-differences −17.00 −20.37 −11.83 Notes: Panel A shows the expected probabilities for the treatment groups (B1, A1, A2) and for the control group (AC) of a person with average characteristics before and after the reform rounded to two decimal places. Moreover, it depicts the differences in the expected probabilities and the difference in these differences, i.e., the cross differences. The next part of the table shows how the counter-factual cross differences are obtained using the expected probabilities for the post-reform period, which result when the re‐ form’s effects are restricted to zero. The Panel C reports the ATT, i.e., the differences in these cross differences. The relative differences in differences are computed, respec‐ tively, as the fraction of the treatment effect and the expected probability in the postpolicy period subtracted by the treatment effect. The same calculation, based on the av‐ erages of the respective probabilities among actual persons in the data instead of the expected probabilities of a person with average characteristics, yields similar results and is available upon request Cluster (occupation) robust standard errors calculated by the delta method are in parentheses. Asterisks (*/**/***) denote significance at the 10 %/5 %/1 % level Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.5: IV.2 Entry regulation and entrepreneurship 117 creased significantly by 2.80 percentage points, with a standard error of 1.08 (p value 0.01). A more flexible specification reported in Table 4.2.12 shows that this large effect is driven mainly by the years 2007, 2008 and 2009. Given that these years are relatively far from the date of the amend‐ ment to the HwO, one should be careful to attribute this effect to the re‐ form. Still, a marginally insignificant effect of 1.23 percentage points (p value 0.89) is observed when the years 2007, 2008 and 2009 are excluded (cf. Rostam-Afschar 2010). Further, the treatment effect for the A2-voca‐ tions including all years is 2.10 percentage points. This effect does not achieve statistical significance at the 10 % level with a standard error of 1.42 (p value 0.14). Note that the probability of being self-employed is substantially smaller for the B1-vocations in the first place. Therefore, the relative effect of 41.41 % is the largest compared to the other groups of craftsmen. For the A1-vocations the relative effect amounts to 20.59 % and to 14.21 % for the A2-professions. Heterogeneous treatment effects Who are these new entrepreneurs in craftsmanship? In this section, I take a closer look at the heterogeneity of treatment effects. This helps to deter‐ mine individual subgroups within the treatment groups on which the re‐ form had the greatest impact. Individuals, who are disadvantaged in terms of labor market opportunities, such as craftsmen with- out any profession‐ al qualification and female craftsmen, might see self-employment as a way out of unemployment (cf. Caliendo and Künn 2011). From Table 4.2.2, we know that treatment group B1, which ultimately showed the strongest rela‐ tive increase in the post-policy period, comprises more craftsmen without qualification, as well as more female craftsmen, compared with the other treatment groups. Thus, I expect the effects of the policy change to be highest for craftsmen with the above-mentioned characteristics in the B1 group. If the higher entries documented previously for the A1- and the A2-oc‐ cupations reflect the effects of the Altgesellen rule, this would be the re‐ sult of more Geselle- qualified craftsmen engaging in entrepreneurship. Thus, I expect that the largest effect for the groups of A1- and the A2-vo‐ cations will be observed for the subsample with this level of professional qualification. 5.3 IV Past Reforms in the Services Sector and their Effects 118 Moreover, I split the sample by nationality and by indication of having received public payments to show that the effects of the amendment to the Probabilities of being self-employed (in %): difference-in-dif‐ ferences B1 A1 A2 AC ΔB1 ΔA1 ΔA2 Panel A Before reform 2004 1.04*** (0.07) 14.21*** (0.71) 15.44*** (0.93) 22.62*** (1.99) −21.58*** (1.98) −8.41*** (1.59) −7.18*** (1.47) After reform 2004 1.40*** (0.06) 16.40*** (0.70) 16.88*** (0.95) 21.73*** (1.99) −20.33*** (1.98) −5.34*** (1.43) −4.85*** (1.41) Δ Between after and before re‐ form 2004 0.36*** (0.09) 2.19*** (0.76) 1.44 (1.22) −0.89 (1.47) 1.25 (1.46) 3.08** (1.48) 2.33 (1.69) Panel B Before reform 2004 1.04*** (0.07) 14.21*** (0.71) 15.44*** (0.93) 22.62*** (1.99) −21.58*** (1.98) −8.41*** (1.59) −7.18*** (1.47) After reform 2004 0.99*** (0.10) 13.59*** (1.15) 14.78*** (1.35) 21.73*** (1.99) −20.74*** (1.95) −8.14*** (1.44) −6.95*** (1.32) Δ Between after and before re‐ form 2004 −0.05 (0.08) −0.62 (1.00) −0.66 (1.07) −0.89 (1.47) 0.84 (1.38) 0.27 (0.46) 0.23 (0.39) Panel C Difference-in-differences 0.41*** (0.11) 2.80*** (1.08) 2.10 (1.42) Relative difference-in-differences 41.41 20.59 14.21 Notes: Panel A shows the expected probabilities for the treatment groups (B1, A1, A2) and for the control group (AC) of a person with average characteristics before and after the reform rounded to two digits after the decimal point. Moreover, it depicts the dif‐ ferences in the expected probabilities and the difference in these differences, i.e., the cross differences. The next part of the table shows how the counter-factual cross differ‐ ences are obtained using the expected probabilities for the post-reform period, which result when the reform’s effects are restricted to zero. The Panel C reports the ATT, i.e., the differences in these cross differences. The relative differences in differences are computed, respectively, as the fraction of the treatment effect and the expected probability in the post-policy period subtracted by the treatment effect. The same cal‐ culation, based on the averages of the respective probabilities among actual persons in the data instead of the expected probabilities of a person with average characteristics, yields similar results and is available upon request. Cluster (occupation) robust standard errors calculated by the delta method are in paren‐ theses. Asterisks (*/**/***) denote significance at the 10 %/5 %/1 % level Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.6: IV.2 Entry regulation and entrepreneurship 119 HwO are not distorted by the effects of other policies that potentially af‐ fect entrepreneurs. Table 4.2.7 shows the results of repeating the logit estimations from the main analysis for different subsamples, and then obtaining the absolute and the relative treatment effects. The first two columns present findings when the sample is restricted to German craftsmen and to craftsmen who indicated having not received substantial public payments (SPP). Appar‐ ently, for both subsamples the estimated coefficients are almost always slightly smaller compared with the overall results. This is true for the probabilities of entering self-employment and the probabilities of being self-employed. The probabilities of exit from self-employment are again insignificant (not reported, available on request). As the magnitudes and the significance of the effects are roughly the same, I conclude that the main results are not confounded by the enlargements of the EU or by sub‐ sidies for entrepreneurs. The next two columns display the treatment effects for female and male craftsmen. Surprisingly, the reform turns out not to have been effective for the entry probability of female craftsmen. Instead, the effects on the prob‐ abilities of entering self-employment are all positive and significant for male craftsmen. Moreover, the probabilities of exit are again highly insignificant for fe‐ male and male craftsmen of all vocational groups apart from females working in B1-occupations; their treatment effect does not fail signifi‐ cance at the 10 % level (p value 0.07). Here, the results indicate that the reform decreased the probabilities of exit from self-employment by 7.24 percentage points with a standard error of 3.98 implying a substantial rela‐ tive reduction, namely of 58.45 %. This means that a reduction of exits, together with a constant entry rate, increased the stock of female crafts‐ men. Indeed, the probability of being self-employed seems to be higher for female B1-craftsmen after the reform. This fact is intriguing, since the re‐ form should affect entries as it deregulates entry barriers but not exits. However, this result could stem from indirect effects of the deregulation, as the reform changed the competitive environment. To investigate this further is left to future research. Turning to the effects on the share of self-employed in the bottom part of Table 4.2.7, a different pattern is apparent for both sexes. For male craftsmen from the A2 group, the increases in the entry probabilities are not accompanied by a significant rise in the probabilities of being self-em‐ ployed. For females in this group, the entry probability seems to have been IV Past Reforms in the Services Sector and their Effects 120 unaffected, while the self-employment probability seems to have been in‐ creased, though not significantly. This could be due to increased exit rates although the positive point estimates are insignificant as well and very small. However, for females as well as for males in the B1- and A1-occu‐ pations, the treatment effects on the share of self-employed achieve signif‐ icance at the 10 % level (p values <0.01, 0.04, 0.04 and 0.01). Thus, while the evidence is not strong that both female in all and male craftsmen in the Treatment effects on entry into self-employment and on the share of self-employed for subgroups (in %): difference-indifferences Sample German Unsubsi‐ dized Female Male No quali‐ fication Geselle Meister Treatment effects on entry into self-employment DIDB1 0.13*** (0.03) 0.13*** (0.03) −0.01 (0.06) 0.36*** (0.12) 0.23*** (0.06) 0.16*** (0.05) 0.07 (0.46) Relative DIDB1 57.90 68.18 −8.16 55.92 806.97 78.09 4.61 DIDA1 0.57** (0.26) 0.64*** (0.21) −1.34 (1.08) 0.62*** (0.20) 1.43* (0.79) 0.81*** (0.21) −2.33 (1.51) Relative DIDA1 32.25 43.27 −52.10 57.46 354.18 93.87 −30.87 DIDA2 0.66** (0.33) 0.60** (0.25) 0.00 (0.01) 0.57** (0.26) 6.40*** (2.44) 0.89*** (0.33) −1.87 (1.49) Relative DIDA2 30.01 32.38 0.00 39.23 573.94 71.37 −26.56 Treatment effects on the share of self-employed DIDB1 0.34*** (0.10) 0.34*** (0.10) 0.56*** (0.16) 0.51** (0.24) 0.50*** (0.12) 0.32 (0.20) 0.88 (0.55) Relative DIDB1 35.70 37.71 114.38 24.59 77.55 29.79 18.39 DIDA1 2.64** (1.05) 2.48** (1.14) 7.33** (3.55) 1.63** (0.65) 2.29 (1.77) 2.20** (1.12) 2.61 (2.33) Relative DIDA1 18.71 19.55 53.66 13.84 29.52 28.20 4.49 DIDA2 1.73 (1.37) 1.82 (1.41) 11.33 (9.42) 0.87 (0.86) 3.22 (2.58) 1.86 (1.37) 3.05 (2.62) Relative DIDA2 11.58 13.87 44.43 7.00 32.22 21.14 6.04 Notes: The treatment effects are based on the expected probabilities for a person with average characteristics. The relative differences in differences are computed as the fraction of the treatment effect and the expected probability in the post-policy period, subtracted by the treatment effect, respectively. Cluster (occupation) robust standard errors calculated by the delta method are in paren‐ theses. Asterisks (*/**/***) denote that a difference-in-differences is significantly different from zero at the 10 %/5 %/1 % level Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.7: IV.2 Entry regulation and entrepreneurship 121 A2 group experienced the intended effects of the reform, the results sug‐ gest that the increases reported in Tables 4.2.4 and 4.2.6 for the B1- and A1-occupations stem largely from male craftsmen engaging more often in entrepreneurship. The last three columns show the results obtained by splitting up the sample by professional qualification. The results show a clear picture: The amendment to the HwO had a positive effect on the entry probabilities of untrained craftsmen across all three treatment groups. These increases in entries also raised the probability of being self-employed for each of the groups. This implies that these businesses survived for some time. While the latter effects are insignificant for the A1- and A2-vocations, they are highly significant for the B1-occupations. Therefore, the expectation that craftsmen without professional qualifi‐ cation entered entrepreneurship more often in the B1-occupations is sup‐ ported by the results. Furthermore, the reform encouraged craftsmen who hold a Geselle de‐ gree to enter entrepreneurship. This was the purpose of the Altgesellen rule, and the objectives of this policy seem to have been accomplished. However, the fact that these entries could not increase the probability of being self-employed significantly, except for A1-vocations, favors the view that some of these new entrepreneurs used fly-by-night strategies. For example, a Meister could ask one of his Altgesellen to set up an ancil‐ lary business to drive a rival out of the market. The last column shows that for Meister craftsmen the reform had, as expected, no significant effect at all. Specification and sensitivity tests To assess the validity of the assumptions on which the DID approach is based, and to gage the robustness of the findings in this analysis, the logit models of the probability of being self-employed and of the transition probabilities are reestimated, varying the estimation sample, the definition of variables, and the specification. Column one in Table 4.2.11 shows the results of estimating the same specification as in the main analysis, in which the year 2001 is included. Obviously, the size and significance of the estimates are similar to those reported in Table 4.2.10. Hence, using this sample does not distort the main results. However, I decided to exclude 2001 from the sample be‐ 5.4 IV Past Reforms in the Services Sector and their Effects 122 cause a “placebo test” discussed below indicates significant coefficients of the interaction between a placebo reform dummy and the A1- and A2-vo‐ cation dummies, respectively. In columns two and three, I display the results when the year 2004 is omitted from the sample and when it is defined as belonging to the postpolicy period, respectively. Recall that the post-policy period was defined as being from 2005 to 2009 in the main analysis. I do this because up until 2005 the last week of April was usually the reference week of the survey, and the amendment to the HwO came into effect at the beginning of 2004. Apparently, dropping the year 2004 does not change the results a great deal, while defining 2004 as part of the post-policy period reduces the esti‐ mates somewhat. This shows that individuals needed some time to adjust to the new policy, as argued above. Next, in columns four to six, I scrutinized whether influences other than the actual treatment of the treatment groups were present but did not influ‐ ence the comparison group. Such influences would have confounded the analysis. In most settings, there is no way to test for these influences di‐ rectly, so placebo tests are based on the idea of reestimating the models while pretending that the policy event took place in a year prior to the ac‐ tual policy change. First, the post-policy period indicator is redefined to represent the period from 2003 to 2004, as if the policy change had taken place in late 2002. Second, the logit model for the probability of being self-employed is reestimated without the actual post-policy period to avoid measuring the true effect of the reform. These steps are repeated for a placebo policy reform in late 2003. In column four, the coefficients for the interaction terms turn out to be significant for the B1- and A1-occupations when the estimation sample in‐ cludes the year 2001— which is why the main analysis was based on the sample from 2002 to 2009. The interaction coefficients in columns five and six are insignificant, which would not be the case if confounding fac‐ tors existed before the policy change. Therefore, assuming this result ex‐ tends to the post-policy period, the validity of the identifying assumption of the DID analysis receives support. Furthermore, I examine the assumption of common trends more explic‐ itly by replacing the post-policy period dummies in the interactions with time dummies and all interactions involving the post-policy period dum‐ my with interactions using time dummies instead. Correspondingly, I in‐ cluded interactions between the branch dummies and time dummies. The results (see Table 4.2.12) are in line with the prior findings shown in Table IV.2 Entry regulation and entrepreneurship 123 4.2.10 that provided evidence that the probability of being self-employed increased significantly for B1-occupations; the coefficients of the interac‐ tions of the B1-dummy with year dummies from the post-policy period are individually positive, significant and of similar magnitude throughout. In‐ terestingly, the coefficients for the interactions in the A1-occupations are increasing in size and statistical significance over time. This suggests that the effects of the reform presented in Table 4.2.6 emerged only after some time for the A1-occupations. This is consistent with the picture in Figs. 4.2.1 and 4.2.2 where the lines representing the A1-occupations change slower than those of the B1-occupations. The fact that only very few entries are observed compared to non-en‐ tries could lead to a different problem highlighted in King and Zeng (2001). This study shows that applying the standard logistic regression po‐ tentially leads to significant distortion of the results as the finite sample bias is amplified by the rare occurrence of events. This should not be a problem because the sample size used should be sufficiently large— for the B1-occupations more than 700 entries are observed. Indeed, the differ‐ ences between the standard logit and a rare events logit model are very small (available on request); the general results of the main analysis re‐ main unchanged. In a further robustness check which is available on request, I collapse the sample by occupation and year. Then, I calculate the differences in dif‐ ferences as in the analysis based on individual data. The coefficient of the interaction between the B1-dummy and the post-policy period dummy of the same specification as in column four of Table 4.2.3 changes from 0.0077 (0.0041) to 0.0073 (0.0054). Running the specification of column six where the rate of self-employment is the dependent variable, the coef‐ ficient of the same interaction has a standard error of 0.0118 on the indi‐ vidual and of 0.0146 on the occupational level. The point estimate is 0.0179 on the occupational level. Moreover, I run regressions restricting the sample to individuals above various ages to see whether conditioning on vocational training could bias the results. The idea is that craftsmen who achieved their qualification ob‐ jectives long ago do not condition their decision to obtain a Meister degree on the desire to be an entrepreneur. Otherwise they would have had enough time to obtain the Meister degree if they had wanted to. On aver‐ age, since 2002 craftsmen have obtained the Meister degree at an age be‐ low 30, according to the Chambers of Crafts and Trade. Less than 7 % of the craftsmen that passed the Meister exam were above 40 years old in IV Past Reforms in the Services Sector and their Effects 124 2006/7. Restricting the sample to craftsmen older than 30, 35, and 40, I obtained treatment effects that are significant but slightly larger than those obtained from the entire sample. Complete estimation results are available on request. Summary and conclusions In pursuit of an answer to how the amendment to the HwO in 2004 influ‐ enced entrepreneurs in German craftsmanship, this paper evaluates the ef‐ fect of this reform on the probability of entering self-employment and of exiting from self-employment. Evidence is provided concerning how the probability of being self-employed changed as a result of the reform for three treatment groups that experienced different degrees of deregulation. Among other modifications, these legislative changes exempted the group of B1-craftsmen altogether from the requirement of passing a Meister ex‐ amination for admission to entrepreneurship, while for the A1- and A2-oc‐ cupations the entry requirement has been reduced; a lower level of voca‐ tional training has been required since the reform. This is known as the Altgesellen rule. Moreover, the amendment exempted a portion of the A1occupations from the Altgesellen rule under the condition of limiting busi‐ ness to simple activities that frequently take the opportunity to establish small businesses. This defines the A2 group. Apart from these deregula‐ tions, the HwO also provides a natural comparison group, because for some professions, the entry requirement remained mandatory. According to the legislation, four distinct occupational groups can be identified in the data from the German microcensus from 2002 to 2009. These groups are exploited within this setting in a natural experiment. The results of a DID analysis provide evidence that the probability of being self- employed in‐ creased significantly with the amendment to the HwO among B1- and A1occupations, while the positive effect fails to achieve significance for the A2- vocations. The strongest relative increase amounts to more than 40 %. This occurred in the group of B1-craftsmen that have received the strongest treatment. In A1- and A2-occupations, the results indicate weak‐ er, but still positive relative effects. The analysis shows further that these increases are caused by significant increases to the probabilities of entry across all three groups, whereas the probabilities of exit from self-employ‐ ment remained virtually unaffected by the policy change. 6 IV.2 Entry regulation and entrepreneurship 125 Two key findings that result from an investigation of heterogeneous treatment effects have important policy implications. First, the findings suggest that the increases in the entry probabilities result from male crafts‐ men who are significantly more likely to start businesses after the reform in all occupation groups. There is weaker evidence that, for these groups, the probabilities of being self-employed also increased after the reform. The results for female craftsmen are less clear: the entry probabilities seem not to have been affected at all, while the results for the self-employ‐ ment probabilities point to a positive effect. Second, untrained workers, mainly among the B1- and A2-vocations, have a significantly higher probability of starting a business after the re‐ form. Consequently, the probability of being self-employed is higher for this group, which is disadvantaged in the labor market. Craftsmen of all occupations that completed an apprenticeship also have engaged more in entrepreneurship since the reform, which was the intended effect of the Altgesellen rule. The increase in entries seems to have led to a greater probability of being self-employed for craftsmen trained in an apprentice‐ ship, though the evidence is weak. Interpreting these results, it is important to bear in mind that these re‐ sults focus only on engagement in entrepreneurship, and do not replace an evaluation of the reform in terms of its welfare effects on the German economy. Appendix Stock of businesses at the end of the year A B1 B2 Total 2002 590,146 76,044 177,471 843,661 2003 587,762 74,940 183,886 846,588 2004 595,309 102,568 189,216 887,093 2005 600,287 129,591 192,805 922,683 2006 603,443 149,981 193,474 946,898 2007 603,757 166,015 191,434 961,206 2008 602,605 175,557 188,526 966,688 2009 Data not available heretofore Notes: Müller (2006) argues that the actual stock of businesses is about 15 % lower than the reported stock due to registered but non-active businesses. Source: Own calculations based on Müller (2006) and data provided by the German Confederation of Skilled Crafts Table 4.2.8: IV Past Reforms in the Services Sector and their Effects 126 Self-employment rates in treatment groups and control group by year B1 A1 A2 AC WP 2002 7.88 15.20 12.41 19.26 11.17 2003 7.54 15.80 12.92 19.68 11.56 2004 8.20 16.98 13.38 20.91 12.00 2005 9.32 17.22 13.96 20.81 12.44 2006 9.48 17.39 14.20 18.49 12.27 2007 9.73 17.76 13.83 20.46 12.11 2008 9.78 17.24 13.69 19.30 11.95 2009 9.87 18.66 14.51 18.13 12.13 Percentage share of self-employed among B1, A1, A2, and AC occupations and per‐ centage share of self- employed among working persons (WP) Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.9: IV.2 Entry regulation and entrepreneurship 127 Estimation results of self-employment state and transition probabilities Logit entry Logit entry Logit entry Logit exit Logit self-em‐ ployed dB1 × dPost 0.4556*** (0.1466) 0.4152*** (0.1446) 0.4630*** (0.1440) −0.2057 (0.3426) 0.3528*** (0.1071) d A1 × dPost 0.2719* (0.1427) −0.2270 (0.3657) 0.2205** (0.0907) d A2 × dPost 0.2571* (0.1440) −0.1331 (0.4098) 0.1579 (0.1099) dEU × dPost 1.1145** (0.5305) 0.4527 (0.2855) −0.8180 (0.5496) 0.5809** (0.2667) dB1 −0.8704* (0.4738) 0.1464 (0.1935) −2.7259*** (0.1749) 3.5941*** (0.2985) −3.3255*** (0.1234) d A1 −0.7175*** (0.1290) 1.2221*** (0.2482) −0.5682*** (0.0852) d A2 −0.4634*** (0.1157) 1.0784*** (0.2652) −0.4706*** (0.0795) dPost −0.1400 (0.1310) 0.1150 (0.1951) −0.0309 (0.1346) −0.7154** (0.3469) −0.0709 (0.0831) dEU −0.1224 (0.5657) 0.1646 (0.2094) −0.2910 (0.5718) 0.4252*** (0.1453) Constant −3.5482*** (0.2416) −3.7692*** (0.7454) −1.4658*** (0.4525) −1.4618 (0.8993) −2.2017*** (0.5096) Year dummies yes yes yes yes Occ. dummies yes yes yes yes Branch dummies yes yes yes yes Controls yes yes yes yes Log likelihood −5,346.12 −4,577.49 −13,667.83 −2,159.01 −54,391.46 Pseudo-R2 <0.01 0.15 0.10 0.20 0.30 Observations 64,842 64,842 154,940 17,211 196,669 Notes: Robust standard errors, clustered by occupation, are given in parentheses below the coefficients of the logit models. Controls included are age and its square, and dum‐ my variables indicating gender, type of secondary schooling and professional qualifi‐ cation, nationality, region of residence, the size of the respondent’s city of residence, marital status, number of dependent children, citizenship of foreigners in an EU mem‐ ber state and its interaction with the post-policy period. Moreover, year dummies for 2003, 2004, 2006, 2007, 2008, and 2009, and indicators for the branch of craftsman‐ ship, for the occupation, as well as a constant are included. Significance of the logit coefficients is indicated at the 10 %/5 %/1 % level by asterisks (*/**/***) Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Table 4.2.10: IV Past Reforms in the Services Sector and their Effects 128 Timing sensitivity: Logit estimation results of self-employ‐ ment state probabilities Self-em‐ ployed: timing (2001–2009) I Self-em‐ ployed: timing 2004 ropped (2002–2009) II Self-em‐ ployed: timing 2004 as post (2002–2009) III Self-em‐ ployed: Placebo re‐ form in 2002 (2001–2004) IV Self-em‐ ployed: Placebo re‐ form in 2002 (2002–2004) V Self-em‐ ployed: Placebo re‐ form in 2003 (2002–2004) VI dB1 × dPost 0.4505*** (0.1029) 0.3853*** (0.1055) 0.3173*** (0.0839) 0.2432* (0.1427) 0.0829 (0.1373) 0.1087 (0.1163) dA1 × dPost 0.3183*** (0.0895) 0.2533*** (0.0850) 0.2295*** (0.0667) 0.2878** (0.1401) 0.1433 (0.1309) 0.1133 (0.0938) dA2 × dPost 0.2139** (0.1079) 0.1818* (0.1105) 0.1476* (0.0879) 0.2032 (0.1414) 0.1370 (0.1397) 0.0788 (0.0924) dEU × dPost 0.6124** (0.2497) 0.6695*** (0.2529) 0.5106** (0.2485) 0.2622 (0.1617) 0.3116* (0.1813) 0.2640 (0.1900) dB1 −1.0689*** (0.2088) −1.0644*** (0.2093) −0.9379*** (0.2159) −1.1287*** (0.2103) −1.0182*** (0.2296) −1.0017*** (0.2265) dA1 1.7002*** (0.2000) 1.6872*** (0.2046) 1.7793*** (0.2013) 1.6867*** (0.2032) 1.7667*** (0.2283) 1.8234*** (0.2108) dA2 −0.5579*** (0.0709) −0.5403*** (0.0841) −0.4517*** (0.0665) −0.6524*** (0.0891) −0.4838*** (0.1110) −0.4168*** (0.0645) dPost −0.0953 (0.0778) −0.0996 (0.0730) −0.1750*** (0.0593) −0.1052 (0.1364) −0.0492 (0.1261) −0.0258 (0.0898) dEU 0.4142*** (0.1370) 0.3323*** (0.1216) 0.5428*** (0.1302) 0.3276** (0.1317) 0.2344 (0.1482) 0.3574*** (0.1243) Constant −4.5595*** (0.4784) −4.4227*** (0.4945) −4.4666*** (0.4845) −4.8397*** (0.4515) −4.8152*** (0.4603) −4.8564*** (0.4536) Year dum‐ mies yes Yes yes Yes yes yes Occ. dum‐ mies yes Yes yes Yes yes yes Branch dummies yes Yes yes Yes yes yes Controls yes Yes yes Yes yes yes Log likeli‐ hood −60,898.75 −47,945.16 −54,411.65 −25,907.46 −19,434.86 −19,435.57 Pseudo-R2 0.30 0.30 0.30 0.32 0.31 0.31 Observa‐ tions 223,241 172,664 196,669 101,709 75,137 75,137 Notes: Robust standard errors clustered by occupation are given in parentheses below logit coefficients. Controls included are age and its square, and dummy variables indi‐ cating gender, type of secondary schooling and professional qualification, nationality, region of residence, the size of the respondent’s residence city, marital status, the num‐ ber of children, citizenship of foreigners in a member state of the European Union and Table 4.2.11: IV.2 Entry regulation and entrepreneurship 129 its interaction with the post-policy period. Moreover, year dummies and indicators for the branch of craftsmanship, for the occupation, and a constant are included. Significance of the logit coefficients is indicated at the 10 %/5 %/1 % level by asterisks (*/**/***) Source: Own calculations based on the scientific use file of the German microcensus (2001–2009) Robustness: Logit estimation results of self-employment state probabilities Self-employed: B1 Self-employed: A1 Self-employed: A2 dO × d2003 −0.0010 (0.1596) 0.0875 (0.1567) 0.0747 (0.1773) dO × d2004 0.2407 (0.1664) 0.1582 (0.1746) 0.1383 (0.1942) dO × d2005 0.4233*** (0.1608) 0.2287 (0.1626) 0.1826 (0.1912) dO × d2006 0.4283** (0.1824) 0.2445 (0.1762) 0.1815 (0.1968) dO × d2007 0.4755*** (0.1356) 0.3556*** (0.1324) 0.2027 (0.1667) dO × d2008 0.4566*** (0.1638) 0.3337** (0.1401) 0.1501 (0.1960) dO × d2009 0.5520*** (0.1957) 0.4807*** (0.1541) 0.3297* (0.1781) Notes: Robust standard errors, clustered by occupation, are given in parentheses below logit coefficients. Controls included are age and its square, and dummy variables indi‐ cating gender, type of secondary schooling and professional qualification, nationality, region of residence, the size of the respondent’s city of residence, marital status, the number of dependent children, citizenship of foreigners in an EU member state and its interaction with the post-policy period. Moreover, year dummies and indicators for the branch of craftsmanship, for the occupation, and a constant are included. The log-like‐ lihood value is −54,322.07, the pseudo-R2 0.3, and the number of observations 196,669. Significance of the logit coefficients is indicated at the 10 %/5 %/1 % level by asterisks (*/**/***) Source: Own calculations based on the scientific use file of the German microcensus (2002–2009) Description of key variables – Entrepreneur: Are you working as self-employed (with or without em‐ ployees)? This definition includes non-incorporated self-employed as well as incorporated self- employed. Table 4.2.12: IV Past Reforms in the Services Sector and their Effects 130 – B1, A1, A2, AC: Job title of most recent occupation. Occupational groups are constructed according to job titles in HwO. – Policy: Dummy indicating the post-policy period from 2005 to 2009. – Entry, Exit: Employment status in previous year. This non-mandatory question was included before 2005 for 0.45 % of the German popula‐ tion and for 1 % of the German population in 2005 and 2009. – SPP: Indicates receiving subsidies for self-employed. After excluding individuals eligible for child benefit, the dummy variable PP is restrict‐ ed to all recently (assuming start-ups are subsidized for at most three years) self-employed individuals, who earn below 26,076 Euros (close to the 25,000 Euro threshold of the EXGZ) per year and receive public payments. – EU: Indicates citizenship of foreigners in a member state of the Euro‐ pean Union (EU). References Aghion, P.; Blundell, R.; Griffith, R.; Howitt, P.; Prantl, S. (2009): The effects of entry on incumbent innovation and productivity. Review of Economics and Statistics 91(1), 20–32 Ardagna, S.; Lusardi, A. (2009): Where does regulation hurt? Evidence from new busi‐ nesses across countries. NBER Working Paper No. 14747 Ardagna, S.; Lusardi, A. (2010): Explaining international differences in entrepreneur‐ ship: the role of individual characteristics and regulatory constraints. In: Lerner, J.; Schoar, A. (eds.) International differences in entrepreneurship, NBER Conference Report, University of Chicago Press, Chicago and London, 17–62 Baumgartner, H.J.; Caliendo, M.; Steiner, V. (2006): Existenzgründungsförderung für Arbeitslose: Erste Evaluationsergebnisse für Deutschland. Vierteljahrs. Wirtschafts‐ forschung 75(3), 32–48 Bertrand, M.; Kramarz, F. (2002): Does entry regulation hinder job creation? Evidence from the French retail industry. Quarterly Journal of Economics 117(4), 1369–1413 Bertrand, M.; Schoar, A.; Thesmar, D. (2007): Banking deregulation and industry structure: evidence from the French banking reforms of 1985. Journal of Finance 62(2), 597–628 Blanchard, O.; Giavazzi, F. (2003): Macroeconomic effects of regulation and deregula‐ tion in goods and labor markets. Quarterly Journal of Economics 118(3), 879–907 Blanchflower, D.G.; Oswald, A.J. (1998): What makes an entrepreneur? Journal of Labour Economics 16(1), 26–60 Blundell, R.; Costa Dias, M. (2009): Alternative approaches to evaluation in empirical microeconomics. Journal of Human Resources 44(3), 565–640 IV.2 Entry regulation and entrepreneurship 131 Branstetter, L.G.; Lima, F.; Taylor, L.J.; Venâncio, A. (2013): Do entry regulations de‐ ter entrepreneurship and job creation? Evidence from recent reforms in Portugal. Economic Journal 124 (577), 805-832 Bruhn, M. (2011): License to sell: the effect of business registration reform on en‐ trepreneurial activity in Mexico. Review of Economics and Statistics 93(1), 382– 386 Caliendo, M.; Künn, S. (2011): Start-up subsidies for the unemployed: long-term evi‐ dence and effect heterogeneity. Journal of Public Economics 95(3–4), 311–331 Caliendo, M.; Steiner, V. (2005): Aktive Arbeitsmarktpolitik in Deutschland: Be‐ standsaufnahme und Bewertung der mikroökonomischen Evaluationsergebnisse. Zeitschrift für ArbeitsmarktForschung 38(2–3), 396–418 Cetorelli, N.; Strahan, P.E. (2006): Finance as a barrier to entry: bank competition and industry structure in local U.S. markets. Journal of Finance 61(1), 437–461 Ciccone, A.; Papaioannou, E. (2007): Red tape and delayed entry. Journal of the Euro‐ pean Economic Association5(2–3), 444–458 Djankov, S.; La Porta, R.; Lopez-De-Silanes, F.; Shleifer, A. (2002): The regulation of entry. Quarterly Journal of Economics 117(1), 1–37 Evans, D.S.; Jovanovic, B. (1989): An estimated model of entrepreneurial choice un‐ der liquidity constraints. Journal of Political Economy 97(4), 808–827 Fossen, F.M. (2011): The private equity premium puzzle revisited – New evidence on the role of heterogeneous risk attitudes. Economica 78(312), 656–675 German Confederation of Skilled Crafts (2003): Stellungnahme zum Themenkatalog zur öffentlichen Anhörung des Ausschusses für Wirtschaft und Arbeit des Deutschen Bundestages, Berlin German Confederation of Skilled Crafts (1991): Marktöffnung und Wettbewerb. Gutachten der unabhängigen Expertenkommission zum Abbau marktwidriger Reg‐ ulierungen, Stuttgart German Monopolies Commission (1998): Marktöffnung umfassend verwirklichen. Hauptgutachten der Monopolkommission, XII (1996/97), Baden-Baden German Monopolies Commission (2002): Reform der Handwerksordnung. Son‐ dergutachten der Monopolkommission, 31, Bonn Holtz-Eakin, D.; Rosen, H.S. (2005): Cash constraints and business start-ups: Deutschmarks versus Dollars. B.E. Journal of Economic Analysis & Policy 4(1), 1– 28 Hurst, E.; Lusardi, A. (2004): Liquidity constraints, household wealth, and en‐ trepreneurship. Journal of Political Economy 112(2), 319–347 Kerr, W.R.; Nanda, R. (2009): Democratizing entry: banking deregulations, financing constraints, and entrepreneurship. Journal of Financial Economics 94(1), 124–149 King, G.; Zeng, L. (2001): Logistic regression in rare events data. Political Analysis 9(2), 137–163 Klapper,L.; Laeven, L.; Rajan, R. (2006) Entry regulation as a barrier to entrepreneur‐ ship. Journal of Financial Economics 82(3), 591–629 IV Past Reforms in the Services Sector and their Effects 132 Meyer, B.D. (1995): Natural and quasi-experiments in economics. Journal of Business and Economic Statistics 13(2); 151–161 Müller, K. (2006): Erste Auswirkungen der Novellierung der Handwerksordnung von 2004 Göttinger handwerkswirtschaftliche Studien 74. Mecke Druck und Verlag, Duderstadt Müller, K. (2008): Auswirkungen der EU-Osterweiterung auf das deutsche Handwerk im Spiegel erster empirischer Erhebungen. In: Bizer,K. (ed) EU-Osterweiterung: Erste Zwischenbilanz für das Handwerk. Mecke Druck und Verlag, Duderstadt Prantl, S. (2012): The impact of firm entry regulation on long-living entrants. Small Business Economics 39(1), 61–76 Prantl, S.; Spitz-Oener, A. (2009): How does entry regulation influence entry into selfemployment and occupational mobility? Economics of Transition 17(4), 769–802 Puhani, P.A. (2012): The treatment effect, the cross difference, and the interaction term in nonlinear ‘difference- in-differences’ models. Economics Letters 115(1); 85–87 Rostam-Afschar, D. (2010): Entry regulation and entrepreneurship: empirical evidence from a German natural experiment. DIW Berlin Discussion Paper No. 1065 Sadun, R. (2008): Does planning regulation protect independent retailers? CEP Discus‐ sion Paper No. 888 van Stel, A.; Storey, D.; Thurik, A. (2007): The effect of business regulations on nascent and young business entrepreneurship. Small Business Economics 28(2–3), 171–186 Viviano, E. (2008): Entry regulations and labour market outcomes: evidence from the Italian retail trade sector. Labour Economics 15(6), 1200–1222 IV.2 Entry regulation and entrepreneurship 133

Kapitelvorschau

Literaturhinweise

Zusammenfassung

„Wie können wir den Wettbewerb im Dienstleistungssektor stärken?“ Dies ist eine Schlüsselfrage für eine größere Leistungsfähigkeit des ökonomischen Umfelds in Deutschland. Dieses Buch versammelt Konferenzbeiträge von Mitgliedern wissenschaftlicher Einrichtungen, von Ministerien, der EU-Kommission und anderen Organisationen zu Reformen im Dienstleistungssektor. Die Konferenz umfasste einen Eröffnungsvortrag zur Bedeutung und Durchführung von Strukturreformen in Europa und zwei Gesprächsforen zur Bewertung vergangener Reformen im Dienstleistungssektor und zur möglichen Reichweite sowie zu den möglichen Auswirkungen weiterer Reformen.

Die Zunahme der Produktivität ist seit den 1990er Jahren sowohl in Deutschland als auch in anderen Ländern der Europäischen Union deutlich geringer als in den USA. Es wird geschätzt, dass die Entwicklung des Produktivitätszuwachses im Dienstleistungssektor für zwei Drittel dieses zunehmenden Abstandes verantwortlich ist. Die Europäische Kommission spricht sich in ihren länderspezifischen Empfehlungen zu Deutschland für Reformen in diesem Sektor aus. Auf einer Konferenz im Juli 2016 in Berlin stellten Experten aus unterschiedlichen Bereichen Studien zu solchen Reformen vor und diskutierten deren Ergebnisse.

Mit Beiträgen von

Oliver Holtemöller, Brigitte Zypries, Joaquim Nunes de Almeida, Dirk Palige, Henrik Enderlein, Stefan Profit, Davud Rostam-Afschar, Paolo Mengano, Oliver Arentz, Erik Canton, Jochen Andritzky

Abstract

‘How Can We Boost Competition in the Services Sector?’ is a key question in the process of creating a more effi-cient economic environment in Germany. This book contains a collection of conference contributions on service sector reforms from members of academic institutions, ministries, the EU Commission and other organisations. The conference consisted of a keynote on the importance and implementation of structural reforms in Europe and two panels that dealt with the evaluation of past reforms in the services sector and the potential scope and effects of further reforms.

Since the 1990s, productivity growth in Germany and other Member States of the European Union has been significantly lower than in the US. The development of productivity growth in the services sector is estimated to account for two thirds of this widening gap. The European Commission advocated reforms in the services sector in its country-specific recommendations for Germany. At a conference in Berlin in July 2016, experts from various fields presented and discussed studies on service sector reforms.

With contributions by

Oliver Holtemöller, Brigitte Zypries, Joaquim Nunes de Almeida, Dirk Palige, Henrik Enderlein, Stefan Profit, Davud Rostam-Afschar, Paolo Mengano, Oliver Arentz, Erik Canton, Jochen Andritzky