Employment stability in newly founded firms: a matching approach using linked employer–employee data from Germany

International studies at the micro level include Dunne et al. (1989) for the US and Storey (1994) for the UK; macro analyses are provided, inter alia, by Davis et al. (1996) and Neumark et al. (2006) for the US and by Barnes and Haskel (2002) for the UK.

While unemployed job searchers may have no chance but to accept a risky job in a new firm, the decision of employed job searchers (who form the majority of entrants in our sample) is more open. Of course, successful matching also depends on the hiring decisions of employers.

Standard presentations of theories of job search, matching and compensating wage differentials can be found in labor economics textbooks, such as Cahuc and Zylberberg (2004) and Franz (2006). Relevant insights from industrial organization are discussed, inter alia, in Audretsch and Mata (1995), Geroski (1995) and Audretsch et al. (2001).

This sub-optimal scale of operation may be related to the fact that younger firms also face tighter financial constraints (in the form of lower ability to raise funds or higher cost of funds) than older firms. By paying lower wages today in exchange for higher future wages, new firms effectively borrow from their employees (see Michelacci and Quadrini 2005).

Since the factors influencing firm survival and failure are not a topic of our analysis, the interested reader is referred to the comprehensive overview on individual and firm-specific determinants of survival and exit by Parker (2004, 222 ff). He classifies the determinants of firm survival into eight broad categories: duration, human capital of the entrepreneur, size of the enterprise, financial variables, innovation, marketing, macroeconomic conditions and industry organization and ownership. Parker particularly emphasizes the role of macroeconomic conditions and of the human capital of the entrepreneur. For instance, the likelihood of employees entering new firms could be higher during an economic upturn, but unfortunately this cannot be analyzed with our data because we only can observe one cohort of entrants. Similarly, we do not have information about the characteristics and abilities of entrepreneurs that may influence an individual’s decision to join as well as a firm’s chance of survival. For discussions of age- and size-dependencies in firm survival, including various liabilities (such as newness, smallness, adolescence), see Carroll and Hannan (1999, part IV) and Heckmann (2008).

We are aware that the hypotheses tested and the corresponding empirical findings are the result of an interaction of supply and demand and will not enable us to clearly distinguish between the decision of an individual or a firm. Note that the preceding arguments also imply a sixth hypothesis on differences in wages, which is not postulated and investigated here. Since our dataset contains information only on daily, but not on hourly wages (there are no exact data on the number of hours worked by full- and part-timers and in different industries), the wage effects of working in new firms cannot be analyzed in this paper. Following an age cohort of firms over time, both Baldwin and Rafiquzzaman (1995) for Canada and Brixy et al. (2007) for Germany identify a negative wage differential of startups that becomes smaller over time. However, both studies rely on plant-level data and are not able to investigate the impact of working in a newly founded firm on individual employees’ wages.

The data generation process is described in more detail by Bender et al. (2000), although they concentrate on the IAB employment subsample, a related, publicly available dataset not used here.

IAB is an acronym for Institut für Arbeitsmarkt- und Berufsforschung, which is the research institute of the Federal Employment Agency in Germany.

Details regarding the IAB Establishment Panel are given in Kölling (2000).

Brixy et al. (2006) find that startups are characterized by higher labor fluctuation than incumbent establishments. However, they rely on plant-level data and are not able to investigate the impact of working in a newly founded firm on individual employees.

Using the employer side of our dataset, Heckmann and Schnabel (2006) report that 22% of firms that hired their first employee in 1995/1996 and were then integrated into the IAB Establishment Panel in 1997 do not survive from 1997 to 1998, and 43% do not reach the year 2000.

Matching analysis and the causal interpretation of the effects identified can be traced back to Rubin (1974). Latterly, the approach has become very popular in the evaluation of labor market programs; see, for example, Heckman et al. (1999).

In contrast to traditional OLS or 2SLS regression, the matching analysis is restricted to the region of common support, which means that the estimated treatment effect is restricted to the so-called region where data on the treated individuals as well as those from the control group are observed. Note also that the matching analysis (as well as OLS regression analysis) requires the decision to join a newly founded firm, conditional on the covariates, to be independent of the unobservables.

Since our control groups consists of 62,128 and 41,450 individuals in western and eastern Germany, respectively, and are thus much larger than our groups of treated individuals, nearest neighbor matching without replacement is feasible and appropriate. As a robustness check we also applied kernel matching, which led to similar results (that are available from the authors). Matching was performed in Stata 9.2 using the PSMATCH2 command (Leuven and Sianesi 2003).

In addition to our t-tests there exist other tests on balancing covariates in propensity score models, but in their comparative survey Smith and Todd (2005, p. 371) show and conclude that “these tests have a number of limitations”.

The lower employment stability in newly founded firms is also confirmed by the fact that the employment length of the median entrant is 575 days in new and 607 days in incumbent firms. This shorter length of employment in new firms should not be interpreted negatively since it may reflect employees’ move to other firms which offer better pay and working conditions.

Another possible approach might have been to restrict (before matching) the sample for the control group to firms founded before 1990, so that all individuals joining a newly founded firm would only be matched to individuals entering older firms. Such a severe restriction, however, would not adequately reflect the decision of individuals to enter different sorts of firms, which is the main focus of our analysis, and it would result in using statistical twins who are not really nearest neighbors in terms of individual characteristics, work history, etc.

One reason why the results are totally different once we include young firms of age one to six may be that these young firms were in a very special situation in eastern Germany during our period of observation. They had been founded after unification with massive financial support from German and EU sources, which may have helped them to survive the first year(s) even if their business did not prosper. Once financial resources dried up and the eastern German economy began to stagnate in 1997, however, employees in these firms experienced a high probability of being laid off (or quitting voluntarily), which dominated the results of our mixed control group of entrants in older and younger incumbent firms.