5 Arquitectura de la red óptica de acceso
5.5 Bloque funcional de terminación de línea óptica
We closely follow Thurik et al.’s (2008) model that builds upon the work of Carree et al. (2002, 2007). Broadly speaking, the model builds the framework for how actual self-employment and optimal self-employment rates can influence economic performance, here the unemployment rate. Assume that for each country 𝑖 in year 𝑡 an optimal self-employment
rate 𝑆𝐸𝑖,𝑡* exists as a function of the stage of a country’s economic
development. 𝑆𝐸𝑖,𝑡* is optimal in the sense that deviations from that
level in either direction decrease economic performance: If a country’s
self-employment rate 𝑆𝐸𝑖,𝑡−1 is lower than the optimal one (𝑆𝐸𝑖,𝑡−1<
𝑆𝐸𝑖,𝑡* ), the economy’s competitiveness and dynamic efficiency are likely
diminished which negatively affects growth. Too high self-employment
rates (𝑆𝐸𝑖,𝑡−1 > 𝑆𝐸𝑖,𝑡* ), on the contrary, absorb too much (human)
capital in a large number of marginal entrepreneurs which induces an underutilization of economies of scale and scope (Carree et al., 2002; Thurik et al., 2008).
A country 𝑖’s unemployment rate 𝑈𝑖𝑡 in year 𝑡 can be decomposed
into two components: first, the unemployment rate 𝑈𝑖,𝑡0 that would be
present if the actual self-employment rate would equal the optimal one
(𝑆𝐸𝑖,𝑡−1= 𝑆𝐸𝑖𝑡*) and, second, the impact on unemployment stemming
from deviations of 𝑆𝐸𝑖,𝑡−1 is from the country-specific optimal self-
employment rate; represented by equation (5.1)
𝑈𝑖,𝑡= 𝑈𝑖,𝑡0 + 𝛾|𝑆𝐸𝑖,𝑡−1− 𝑆𝐸𝑖,𝑡* | with 𝛾 > 0, (5.1)
where the coefficient 𝛾 is assumed to be positive, that is, any positive or negative deviation from the optimal self-employment rate increases unemployment. Taking the first differences of equation (5.1) leads to
Δ𝑈𝑖,𝑡= 𝑈𝑖,𝑡− 𝑈𝑖,𝑡−1
= Δ𝑈𝑖,𝑡0 + 𝛾(︁|𝑆𝐸𝑖,𝑡−1− 𝑆𝐸𝑖,𝑡* | − |𝑆𝐸𝑖,𝑡−2− 𝑆𝐸𝑖,𝑡* |
)︁ .
(5.2)
The optimal self-employment rate is assumed to vary only little over time given its dependence on firmly established institutional and socio- economic factors (Thurik et al., 2008; Prieger et al., 2016). In case
both the self-employment rate in period 𝑡 − 1 and that in 𝑡 − 2 (𝑆𝐸𝑖,𝑡−1
and 𝑆𝐸𝑖,𝑡−2) are above the optimal self-employment rate (𝑆𝐸𝑖,𝑡* ), the
term in brackets in equation (5.2) reduces to Δ𝑆𝐸𝑖,𝑡. Adding start-
ups to an economy with an already larger than optimal start-up rate
therefore increases unemployment. In case both 𝑆𝐸𝑖,𝑡−1 and 𝑆𝐸𝑖,𝑡−2
are below 𝑆𝐸𝑖,𝑡* , the term in brackets reduces to −Δ𝑆𝐸𝑖,𝑡−1. Increasing
the self-employment rate in an economy that consists of too few start- ups thus decreases unemployment. In case the self-employment rate in
one period is lower whereas in the other it is higher than 𝑆𝐸𝑖,𝑡* , and
both are close to the optimal one (𝑆𝐸𝑖,𝑡−1≈ 𝑆𝐸𝑖,𝑡−2≈ 𝑆𝐸𝑖,𝑡* ), the term
in brackets vanishes. Then, 𝑆𝐸𝑖,𝑡* does not affect Δ𝑈𝑖,𝑡.
Following Audretsch et al. (2002) and Thurik et al. (2008), we under- take some further transformations of equation (5.2) in order to derive
an empirically estimable equation. First, we assume 𝑈𝑖,𝑡0 to be idiosyn-
cratic with respect to country and time (captured in the error term
business cycle effects common to all countries. Since we use first dif- ferences, country-specific effects are differentiated out which is why an inclusion of country dummies is not needed. Second, we include the lagged difference in unemployment rates as an explanatory variable to test for the direction of causality. Third, we include multiple time lags in a one-by-one manner so as to test whether the expected effects are of short-run or medium-run in nature. Lastly, we additionally run the complementary model where the difference in the self-employment rates are related to a function of its lagged differences and the difference in unemployment rates (and time dummies and the error term) to test for the case of reverse causality related to the entrepreneurial effect. The resulting models (5.3a) and (5.3b) then read
𝑈𝑖,𝑡− 𝑈𝑖,𝑡−𝐿= 𝛼 + 𝐽 ∑︁ 𝑗=1 𝛽𝑗 (︁ 𝑆𝐸𝑖,𝑡−𝑗𝐿− 𝑆𝐸𝑖,𝑡−(𝑗+1)𝐿 )︁ + 𝐽 ∑︁ 𝑗=1 𝜁𝑗 (︁ 𝑈𝑖,𝑡−𝑗𝐿− 𝑈𝑖,𝑡−(𝑗+1)𝐿 )︁ + 𝑇 ∑︁ 𝑡=1 𝛿𝐷𝑡+ 𝜖𝑖,𝑡 (5.3a) 𝑆𝐸𝑖,𝑡− 𝑆𝐸𝑖,𝑡−𝐿= 𝛼 + 𝐽 ∑︁ 𝑗=1 𝜑𝑗 (︁ 𝑈𝑖,𝑡−𝑗𝐿− 𝑈𝑖,𝑡−(𝑗+1)𝐿 )︁ + 𝐽 ∑︁ 𝑗=1 𝜙𝑗 (︁ 𝑆𝐸𝑖,𝑡−𝑗𝐿− 𝑆𝐸𝑖,𝑡−(𝑗+1)𝐿 )︁ + 𝑇 ∑︁ 𝑡=1 𝜓𝐷𝑡+ 𝜈𝑖,𝑡 (5.3b)
Based on equations (5.3a) and (5.3b) we run a population-weighted Vector Autoregressive Model (VAR). In order to establish whether un- employment causes self-employment and/or vice versa, we perform
Granger-causality tests (Granger, 1969). Hereby we can test how much of the variation in self-employment (unemployment) can be ex-
plained by previous self-employment (unemployment). In a second
step, it can be found out whether lagged values of unemployment (self- employment) can increase the explanatory power of the model. To be precise, if the coefficients of lagged unemployment (self-employment) are significant, then self-employment (unemployment) is said to be Granger-caused by unemployment (self-employment). In the choice of the lag length, we follow Thurik et al. (2008) and include, four, eight, and twelve-year lags. Likelihood ratio tests are applied in order to test whether the inclusion of extra lags improves the model significantly. Finally, as to test whether the effects found differ by the gender of the founders, we additionally run sub-sample analyses for males and females.
5.4 Data
To estimate equations (5.3a) and (5.3b), we construct a country-level dataset encompassing 23 OECD countries for the 1991 to 2015 period. In most related empirical studies, the Compendia database (Van Stel, 2005) is used. The advantage of this data is that the rates of self- employment have been harmonized over years and countries. This is necessary because countries report self-employment rates to the OECD (the underlying data source of Compendia) based on different defini- tions. The disadvantage, however, is that the data are not put out sep- arately for men and women. We therefore use the data provided by the World Bank (2017) for self-employment rates, which are part of the ILO (2017) estimates. Like Compendia, these data are harmonized across countries and years “by accounting for differences in data source, scope
of coverage, methodology, and other country-specific factors” (World Bank, 2017). In principle, there are rates of self-employment for far more than 23 countries, especially for many developing countries. How- ever, the World Bank (2017) warns against using the self-employment rates of women from certain, mostly developing countries for analyses or comparing these rates with those of women from developed coun- tries, as social, legal and cultural disagreement may exist over the def- inition of women’s self-employment. Given this reference, we limit our empirical analysis to those countries that are also used in Compendia. These all belong to the developed, western world, which is why possible differences in the definition of women’s work should be rather small. The share of self-employed workers 𝑆𝐸 is defined as the percentage of own-account workers (or those who work with one or a few partners or in cooperative) among the total labor force (World Bank, 2017). This self-employment measure comes at the cost that it neither differenti- ates by type of self-employment (start-up or already established) nor by high- or low-tech sectors, or by qualification needed. Still, opera- tionalizating entrepreneurial activity by self-employment rates is well established given its comparability over countries and time (Storey, 1991; Thurik et al., 2008). Unemployment rates 𝑈 are defined as a country’s share of the labor force that is without work but available for and seeking employment. In the subdivision of self-employed and un- employment rates by gender, the corresponding Figures 5.1a and 5.1b
are given as a percentage of this subgroups’ labor force.68
Figure 5.1 shows the average self-employment and unemployment rates of all 23 countries for the period from 1991 to 2015, with both measures also being broken down by gender. The average total, male, and female
68
For example, the female self-employment rate is the number of self-employed females expressed as a percentage of the female labor force (World Bank, 2017).
self-employment rate shows a decreasing trend over the whole obser- vation period. They were highest at the beginning of the 90s (ranging from 18 to 23%) and lowest in 2015 (ranging from 12 to 18%). The share of male self-employed is higher than that of females at each year throughout the observation period.
The average unemployment rate, on the contrary, ranges between 5 and 10%. It decreased after the early 1990s before it increased again at the start of the financial crisis in 2007/08. Until that latter event, women had consistently higher unemployment rates than men whereas since 2007/08, unemployment is rather similar across gender.
0 5 10 15 20 25
Share of self-employed (in %)
1990 1995 2000 2005 2010 2015
Year
Self-employment (Total) Self-employment (Males) Self-employment (Females)
(a) Average self-employment rate, 1991-
2015
0
5
10
15
Share of unemployed (in %)
1990 1995 2000 2005 2010 2015
Year
Unemployment (Total) Unemployment (Males) Unemployment (Females)
(b) Average unemployment rate, 1991-2015
Figure 5.1: Average self-employment and unemployment rates 1991-
2015 of 23 countries
Source: World Bank (2017)
Figure 5.1 does not, however, reveal variation in the self-employment and unemployment rates within countries over time. Given this pa- per’s focus on the effect of self-employment on unemployment rates,
in Table 5.1, we show 8-year changes69 of those six country/time com- binations with the highest and lowest changes in the self-employment rate from 1991 to 1999 and from 1999 to 2007. In the total sample, four out of the six country/time combinations with the largest increase in self-employment are characterized by a subsequent decrease in total unemployment whereas two show the opposite. Similarly, for four of the six country/year combinations with the strongest decline in self- employment rates an increase in unemployment rates in the following eight-year period is recorded. A similar, somewhat more mixed result can be seen when the self-employed rates are broken down by gender.
Table 5.1: Summary statistics
Self-employment Unemployment
𝑆𝐸𝑡− 𝑆𝐸𝑡−8 𝑈𝑡+8− 𝑈𝑡
Country Year 𝑡 Total Males Females Total Males Females
Canada 1999 2.2 1.8 2.8 -1.6 -1.5 -1.7 Netherlands 2007 1.7 3.0 0.3 3.7 3.7 3.7 Germany 2007 1.2 1.2 1.5 -4.1 -3.5 -4.6 Germany 1999 1.1 2.2 -0.2 -0.2 -0.1 -0.4 United Kingdom 2007 0.8 0.9 0.6 0.0 0.0 0.2 New Zealand 1999 0.7 1.0 0.6 -3.4 -3.9 -2.8 Austria 1999 -4.4 -4.8 -3.8 0.2 -0.2 0.5 Iceland 2007 -4.6 -5.2 -3.9 1.7 1.7 1.7 Greece 1999 -4.7 -3.8 -6.0 -3.5 -2.5 -5.3 Norway 1999 -6.1 -7.8 -4.0 -0.7 -0.7 -0.8 Iceland 1999 -6.8 -9.3 -3.5 0.1 0.6 -0.5 Greece 2007 -6.8 -5.4 -8.6 16.5 16.5 16.0
Notes: 𝑆𝐸𝑡and 𝑈𝑡denote the self-employment and unemployment rate in year 𝑡.
Source: World Bank (2017).
69 We use eight-year changes based on our regression results in Section 5.5. De- scriptive statistics for other time intervals can be obtained from the authors upon request.