Abstract
The empirical model is divided in two different stages of transition (see Figure 4) to address the questions: Who is entering self-employment, how successful is selfemployment and can the grant of bridging allowance affect the success of the chosen employment opportunity. In a first stage, unemployment duration is investigated to obtain the determinants of self-employment. As possible exit states from unemployment self-employment, paid-employment and out-of-the-labour-force are regarded. It is also possible that the individual stays unemployed during the whole observation period. In such a case the unemployment spell is rightcensored. Therefore, in the first stage of the analysis unemployment and out-oflabour force are different employment states.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
For an extensive discussion of evaluation methodologies see Chapter 4.
For a survey on duration models see e.g. Petersen (1995).
The derivation of the discrete hazard rate model follows Steiner (1997). He applies a similar model on unemployment duration. For the application of this model see also Steiner (2001).
The non-proportional specification has less restrictive assumptions than proportional hazard rate models, which assume that effects of the covariates on the hazard rate are independent of duration. See e.g. Kalbfleisch and Prentice (1980) or Lancaster (1990) for more details.
For an application see Jenkins (1995).
It should be well noted that transition only occurs during t =T k i and that for right-censored spells δ ij for t =T k i .
See e.g. Stata (2000), Handbook Release 6.0, commands “xtlogit” or “quadchk” for a discussion on the reliability of the Gauss-Hermite-Quadrature method.
To facilitate writing the following refers to the unemployment model. Hence index j is suppressed for the unobserved heterogeneity.
See Sin and White (1996) for a survey on various IC.
The estimator “mlogXmX” is programmed using Stata’s (Version 6.0) d2-method. Computation time on a Athlon 1.3 GHz is about 5 days for the unemployment model with four mass points for the Sample-All. The programme code together with a help-file is available by the author upon request.
Lee (1996) suggests a correction for the computation of the difference of the vari-ance-covariance matrices which always yields a positive test statistic. However, Lee’s procedure is much more complex than the procedure proposed by Small and Hsiao (1985).
For the derivation of the fixed-effects estimator see e.g. Greene (1997). According to the Hausman test the fixed-effects estimator is preferred over the random effects es-timator for this analysis.
For a detailed description of the GSOEP see http://www.diw.de/english/sop/index.html.
Hence, it should be well noted that the region of residence of an individual is con-structed as time-varying variable (see below).
As the results for the Sample-West are similar to those for the Sample-All the discussion is concentrated on the Sample-All. Descriptive statistics for the Sample-West can be found in the Appendix in Table 41 and Table 42.
Earnings of the paid-employed were estimated using 16 cross-sections obtaining 16 different parameter vectors. Due to a smaller sample size, self-employment earnings were estimated on a pooled sample using GEE methods (see Liang and Zeger, 1986). These estimates were obtained using a reduced form instead of a structural selection model to control for labour market participation. A possible heteroscedasticity prob-lem may arise from the fact that SPIR and IRR, respectively, are estimated variables. Therefore, the standard errors for these variables may be inconsistent in the duration analysis. To address this problem bootstrapped standard errors could be used in prin-ciple. However, bootstrapping is not applicable to this analysis as computation of the multinomial logit model with random effects takes between one and five days per model. See Appendix A.l for the calculation of the SPIR and IRR.
I.e. unemployment-benefits/[Probability(being self-employed)*expected-self-employ-ment-income + Probability(being paid-employed)*expected-employment-income)].
The UV-ratio and the GDP growth rate proved to approximate demand pull and un-employment push best. Several other measures were also tested, like the unemploy-ment rate, GDP growth on quarterly data, number of insured employees in a certain sector etc.
The further discussion of Section 6.5 is concentrated on the results for the Sample-All. Results for the Sample-West are similar. The interested reader is referred to Table 43 and Tables 45 to 50 in Appendix A.2.
To save time of computation, the testing of the significance of specific variables was done within the model of no unobserved heterogeneity. For the exit into self-employment a significance level of 20% is used, because of the relative small number of observations for this state.
See Table 44 in the Appendix A.2.
For the estimation results neglecting unobserved heterogeneity see Table 44 in Appendix A.2.
Part of this effect may be due to legal requirements according to the German economic and trade regulations. In the craft sector as well as in some professional occu-pations (for example lawyer, doctor) special examinations or vocational degrees, and in the banking and insurance sector some minimum requirements with respect to initial capital, are a necessary precondition for becoming self-employed. However, in principle everybody is allowed to start a business in most parts of the private sector of the economy.
Doubling the expected income from self-employment is equal to doubling the SPIR and halving the IRR.
For the construction of the GSOEP samples see the GSOEP Desktop Companion (2000).
The corresponding graphs for the Sample-West can be found in the Appendix, Fig-ure 25 and Figure 26.
The studies by Taylor (1998, 1999 and 2001) on the duration of self-employment do not test for duration dependence, as he employs the Cox proportional hazard model.
1,900 DM represents the 10%-and 6,000 DM the 90%-Percentile.
Estimation results are displayed in Table 13.
Based on several simulations, it was not possible to isolate a specific individual or a specific type of individuals which cause these implausible values.
Kernel density estimations for the other samples can be found in the Appendix, Fig-ure 27 to Figure 29.
Results for the sample of the initially paid-employed are omitted because of the implausible estimation results for the hypothetical income from self-employment.
The results for the regression of the time-invariant variables on the fixed effects can be found in the Appendix in Table 47.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2004 Springer-Verlag Berlin Heidelberg
About this chapter
Cite this chapter
Reize, F. (2004). Determinants and Success of Self-Employment. In: Leaving Unemployment for Self-Employment. ZEW Economic Studies, vol 25. Physica, Heidelberg. https://doi.org/10.1007/978-3-7908-2685-2_6
Download citation
DOI: https://doi.org/10.1007/978-3-7908-2685-2_6
Publisher Name: Physica, Heidelberg
Print ISBN: 978-3-7908-0168-2
Online ISBN: 978-3-7908-2685-2
eBook Packages: Springer Book Archive