Assessing the Impact of Bridging Allowance at the Firm Level

  • Frank Reize
Part of the ZEW Economic Studies book series (ZEW, volume 25)


To assess the quantitative impact of bridging allowance on firm success, a simple regression approach may be used, including an indicator whether the firm was founded by an unemployed or not. But as studies investigating the direct impact of social programmes have shown, participants and non-participants in the programme systematically differ in their observable and unobservable characteristics. As discussed in Section 2.3 this seems also to be true for the subsidisation of firm foundation by the unemployed. Such a selection bias may occur either through programme selection, e.g., when a competent authority has to assess the sustainability of the self-employment envisaged, or through self-selection. Besides observable characteristics such as the legal form of the firm or industry which may determine these selection processes, there are also, at least for the researcher, unobservable characteristics, such as personal abilities and employment history of the founder, capital endowment or transition costs, which may influence both, programme participation and firm success. Neglecting such selection effects can result in inconsistent estimates. Therefore, the econometric framework of programme evaluation seems to be appropriate when studying the impact of bridging allowance on the success of the firm. Whereas selection on observable characteristics can be handled more or less easily, the existence of selection on unobservables requires either panel data or econometric models assuming a certain joint distribution of the selection process and the measure of success (see Section 4.2). As a consequence, a dummy endogenous regression model seems to be appropriate to assess the impact of bridging allowance.


Survival Probability Employment Growth Variance Covariance Matrix Firm Growth Full Information Maximum Likelihood 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 94.
    See e.g. Heckman et al. (1999) or the discussion in Chapter 4.Google Scholar
  2. 95.
    Firm foundation by the unemployed and subsidisation with bridging allowance is synonymous, as hardly any unemployed would start a business without bridging al-lowance. For a detailed discussion on this, see Pfeiffer and Reize (2000a,b), Reize (2000) as well as Section 2.3.Google Scholar
  3. 96.
    The FIML estimation of dummy endogenous bivariate probit models has some ad-vantages compared to the two step estimation (see Greene, 1998), e.g., the estimation of the standard errors is quite easier in the FIML than in the two step model.Google Scholar
  4. 97.
    This specification has been used in literature on firm growth before. See e.g. Evans (1987a).Google Scholar
  5. 98.
    To correct the standard errors, derivatives of Mill’s ratios (which are non-linear func-tions of the cumulative bivariate normal distribution) are needed.Google Scholar
  6. 99.
    For the derivation of the log-likelihood function of the traditional Heckman selection model see, e.g., Amemiya (1985) or Nawata (1994). Di Tomasso (1999) employs a similar log-likelihood function for a trivariate structure. However, she estimates two probits as output equations including a continuous endogenous variable.Google Scholar
  7. 100.
    The estimator “biheck” is programmed in Stata’s Version 6.0. The programme code together with a help-file can be obtained by the author upon request.Google Scholar
  8. 101.
    The data in the ZEW Start-Up Panel has been made available to the ZEW every six months since 1989 by the Association “Verband der Vereine Creditreform” (VVC). For further details, see Harhoff and Steil (1997).Google Scholar
  9. 102.
    A company may be registered outside the 15 labour market districts of the IAB census, as the address of the firm can differ from the address of the shareholder(s).Google Scholar
  10. 103.
    This means that the period analysed by the IAB is always extended by a quarter, as there is the possibility that an application for a trade register entry was filed before the subsidies were granted, or that a company was formed some time after the support was granted.Google Scholar
  11. 104.
    Often a boundary of only 50 employees is used. See, e.g., Fritsch (1992).Google Scholar
  12. 105.
    Thus 1,741 additional firms are identified of which 45 are subsidised as compared to a previous study (Pfeiffer and Reize, 1998 or Pfeiffer and Reize, 2000b). As start-ups can enter the panel with some delay, the six additional waves of the ZFPS which are available for this study, increase the number of firm start-ups for all cohorts.Google Scholar
  13. 106.
    To prevent a large reduction (almost 50%) of the sample size in East Germany, a category “age of most important shareholder is missing” is constructed. For more de-tails see the next sections.Google Scholar
  14. 108.
    Table 14 shows the definitions of the variables used in the different equations.Google Scholar
  15. 109.
    Table 15 provides some descriptive statistics on firm heterogeneity for West and East Germany.Google Scholar
  16. 110.
    Estimations with different measures of the tightness of the labour market showed that the unemployment to vacancy ratio is the most suitable. This is also shown by Pfeif-fer and Reize (1999a), where the impact of the push-and pull-factors on the intensity of firm foundation is tested for West and East Germany.Google Scholar
  17. 111.
    Most of these variables are typically found in studies examining firm survival and employment growth on the basis of the ZFSP (see e.g. Harhoff et al., 1998). Table 14 shows detailed definitions of these variables.Google Scholar
  18. 112.
    See Stiglitz and Weiss (1981). For an empirical examination of the influence of the legal form on insolvency, see Harhoff et al. (1998).Google Scholar
  19. 113.
    The exact date of closure is only available for a few firms, see Section 8.2.4.Google Scholar
  20. 114.
    This gender specific difference disappears in the multivariate analysis of Sec-tion 8.3.3. Hence, the difference in growth rates between males and females is based on different types of firm started by each gender.Google Scholar
  21. 115.
    Between 10 and 20% of closures within the first five years and 45% of all closures were involuntary in nature. The percentage is higher in the new German federal states with a share of involuntary closures between 19 and 34% for the first five years and of 58% for the whole observation period.Google Scholar
  22. 116.
    Brüderl et al. (1996) employed the same measure. In principle, it is also possible to build up a duration model in discrete time, using yearly time intervals. Hence, the gain of additional information from such a model would be small, since a discrete du-ration model only pools the separate estimations, including additional parameters for time dependence. However, the main interest of our analysis is to assess the impact of brid ging allowances on firm survival and not duration dependence. Moreover, a sepa-rate estimation allows for the measurement of different parameters for different peri-ods of survival. To achieve this in a discrete duration model, an interaction of all co-variates with the baseline hazard rate would be necessary.Google Scholar
  23. 117.
    This measure indicates the share of start-ups, for which an employment growth rate can be calculated, in the speech of the “survivor bias”.Google Scholar
  24. 118.
    For the sake of clarity, the subsidisation equation (8.6) is estimated as a single equation using a probit model. The differences from the results of a joint estimation with the various success variables are minor. The estimates of the subsidisation equation for the different survival models are displayed in Appendix A.4 through Table 51 to Table 60 and for the growth models through Table 32 and Table 33.Google Scholar
  25. 119.
    The Tables display the coefficient, standard error and mean as well as the marginal effect, which is calculated at the mean of the exogenous variables.Google Scholar
  26. 120.
    Start-ups are promoted in Germany by over 100 programmes and many more local initiatives. See also Footnote 23.Google Scholar
  27. 121.
    In the following, (weakly) significantly different from zero refers to a (10%) 5% level of significance.Google Scholar
  28. 122.
    A variety of estimations with other measures of labour market tightness which were used in earlier drafts of the study did not yield different results. If, e.g., the unem-ployment ratio is used instead of the unemployment to vacancy ratio, the results are not affected (see Pfeiffer and Reize, 2000b).Google Scholar
  29. 123.
    Estimations of the four-year survival probability and the employment growth rate yield a significant and positive effect on the probability that a derivative start-up will be subsidised (see next Section and Table 59 in the Appendix A.4).Google Scholar
  30. 124.
    Table 22 to Table 31 display only the coefficients, standard errors, marginal effects and means for the survival equation, whereas the marginal effects are calculated in accordance with equation (8.10), (8.11) and (8.12), respectively. The corresponding subsidisation equations are reported in the Appendix A.4 in Table 51 to Table 60. If not otherwise stated, the results described in this section are based on the bivariate probit model of Section 8.1.2. For the sake of consistency and efficiency only vari-ables which proved to be significantly different from zero on the 10% level on the ba-sis of a Wald test are included. Further results of estimations, of the overall survival probability, i.e. survival without restriction to a certain period, can be found in the Appendix A.4 in Table 62 and Table 63.Google Scholar
  31. 125.
    See Lechner and Pfeiffer (1993) and Pfeiffer (1999) for more detailed studies on the determinants of self-employment and the search for self-employment after unifica-tion.Google Scholar
  32. 126.
    Estimations in Table 32 and Table 33 only include coefficients, which are significantly different form zero on the 10% level, based on the Wald test for the FIML estimation and on the F test for the two-step model.Google Scholar
  33. 127.
    It should be well noted that the standard errors of the two-step estimation are only in-consistent for the growth equation. Those errors in the first step (the bivariate selec-tion equations) are also obtained using FIML estimation. Therefore, the standard er-rors of the selection equations can even be smaller for the two-step estimation as compared to the FIML estimation.Google Scholar
  34. 128.
    Table 32 and Table 33 show the estimated coefficients of the inverses of Mill’s ratio (σεu and σεv) for the two-step model, instead of pεu and pεv For the definition of the correlation coefficients, see Section 8.1.3.Google Scholar
  35. 129.
    For the discussion of the subsidisation equation, see Section 8.3.1.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2004

Authors and Affiliations

  • Frank Reize
    • 1
  1. 1.KfW BankengruppeKSb Volkswirtschaftliche AbteilungFrankfurt am MainGermany

Personalised recommendations