Key variables
In this article, we assess whether and how the removal of transitional arrangements has affected the self-employment rates of EU2 and EU8 nationals in the EU15. Our dependent variables are, thus, the self-employment rates of EU2 and EU8 migrants, computed using the EU Labour Force Survey (EULFS) between 2004 and 2019, as the share of self-employed EU8 and EU2 migrants in the total population of employed EU8 and EU2 migrants, respectively. The EULFS allows us to distinguish between different groups of migrants by country of birth or nationality. We use nationality in this case, as the former is not available in the case of Germany. The quantitative differences between the two are very small or non-existent in most countries. Sweden and Finland do not distinguish between EU2 and EU8 migrants, likely because of the small sample size; therefore, we use the combined group for these two countries.Footnote 3 Although ideally we would distinguish between self-employment with and without employees, under the assumption that the latter is likelier to represent the type of bogus self-employment as hypothesized above, this distinction is not possible using our dataset.
The graphs in Appendix Fig. 1 seem to confirm a relation between the evolution of EU2 and EU8 self-employment rates and the presence or absence of transitional arrangements, at least in certain countries. For instance, in Austria, Belgium, and Germany, the self-employment rates of EU8 nationals increase after the enlargement in 2004 and decrease after 2011, with the removal of transitional arrangements. Similarly, the self-employment rates of EU2 nationals decrease in countries such as Austria, Germany, the Netherlands, and the UK after the removal of TAs, post-2014.
Our two main independent variables are meant to capture the effect of the removal of transitional arrangements on self-employment rates. To that end, we define them as two dummy variables equal to 1 for the period after the end of transitional arrangements for the EU8 and EU2 nationals, respectively, by country. Table 1 presents the year when each country removed the TAs for each migrant group.
Table 1 Period of transitional arrangements by country for EU8 and EU2 nationals Additionally, we control for a number of factors which might influence both the choice of destination and the opportunities and constraints on the path to become self-employed. We include in our models the share of EU8 and EU2 migrants in the receiving country’s population, to control for potential network and diaspora effects in attracting migrants towards a particular destination, and for the effects of networks on the likelihood to become self-employed. As the graphs in Appendix Fig. 2 illustrate, the EU enlargements led to significant increases in the share of EU8 and EU2 nationals in the EU15 receiving countries’ populations. The increase was mostly incremental in the case of EU8 nationals, although there were a few notable trends in Denmark, Ireland, the Netherlands, and the UK. In the UK and especially Ireland, both of which did not implement TAs, the share of EU nationals soared post-2004, while in Denmark and the Netherlands, it increased sharply post-2009 and post-2011, respectively, the years in which they removed the transitional arrangements. Similarly, the share of EU2 nationals in the total population of Italy and Spain grew significantly throughout the period analysed, regardless of the presence or the absence/removal of transitional arrangements in place. This development likely reflects the many TA exceptions present in these two countries, particularly in sectors experiencing labour shortages. On the other hand, in countries such as Belgium, Denmark, France, Greece, Luxembourg, or Portugal, which implemented more restrictive measures, the share of EU2 nationals increased sharply only after the removal of the transitional arrangements.
We also include the self-employment rate of the native population, as a proxy for the overall entrepreneurial culture and the friendliness of institutions and regulations to self-employment in the destination country. Furthermore, our models include three other variables which have been strongly linked to self-employment rates: unemployment, GDP per capita, and employment protection legislation. High unemployment may affect self-employment positively as the opportunity cost of starting a business decreases, or negatively, as it also entails fewer resources available which could undermine the creation of new businesses (see for example Blau 1987; Blanchflower and Meyer 1994; Audretsch et al. 2002; and for an extensive review Thurik et al. 2008). We obtain unemployment rates for the entire active population in each EU15 country, between 2004 and 2019, from Eurostat (2021b). The level of GDP per capita in purchasing power adjusted, a proxy for economic development, may affect self-employment negatively if it is associated with greater capital per worker, or if the returns from waged employment relative to self-employment are now higher (Lucas 1978). Conversely, it can have a positive effect on self-employment, when it is the result of increased economic growth, demand for goods and services and access to credit, encouraging business creation (Parker and Robson 2004). We obtain data on GDP per capita from the Eurostat (2021a) for the period 2004–2019. Self-employment rates might also be affected by the stringency of employment protection regulations (see Ulceluse and Kahanec 2018). By virtue of their role, that of protecting employees from dismissal, wage loss, or unfair treatment from employers, labour market regulations might make hiring and firing costlier. This in turn might incentivize employers to contract out work to individuals, therefore increasing self-employment rates. In order to control for this effect, we include in our models a variable reflecting the strictness of employment protection on individual and collective dismissals for regular contracts, obtained from the OECD database on employment protection. The indicators are compiled using the OECD’s own reading of statutory laws, collective bargaining agreements and case law as well as contributions from officials from OECD member countries and advice from country experts (OECD 2020).
Table 2 provides an overview of the main characteristics of the variables employed in our analysis.
Table 2 Summary statistics of main variables Lastly, migrants’ propensity to become self-employed may depend on the national socio-economic context and institutional configuration and the opportunities and constraints they create. In order to account for this potential variation in self-employment rates, and keeping in mind the limited number of observations in our dataset, we turn to the Varieties of Capitalism (VoC) literature (Ulceluse 2016; Bechter et al. 2012; Hall and Soskice 2001), as one of the most influential explanations for variation in economic outcomes across countries. The VoC literature emphasizes how institutions relating to finance, employment, welfare, industrial relations, and education and training evolve differently in each country and how the interaction between them translates into different models of capitalism (Dilli et al. 2018). These models, which are fairly stable over time, can help explain quantitative and qualitative variation in the supply of migrant labour over time and across space, and migrants’ subsequent economic outcomes (Devitt 2011). We thus include in our model a dummy variable that accounts for the four types of capitalist regimes in our sample, namely, Liberal (Ireland, UK), Continental (Austria, Belgium, Germany, Luxembourg, Netherlands), Nordic (Denmark, Finland and Sweden), and Southern (France, Greece, Italy, Portugal and Spain).
Empirical model
To assess the effect of the removal of TAs on the self-employment rates of EU2 and EU8 nationals, we estimate the following model:
$${Y}_{it}={\beta }_{0}+{\beta }_{1}{X}_{ i t}+{\beta }_{1}{Z}_{ i t}+{\varepsilon }_{it}$$
(1)
where \({Y}_{it}\) is the dependent variable, either self-employment rates for EU2 or self-employment rates for EU8 migrants, \(X\) represents the independent variable capturing the post-transitional arrangements period, \({\beta }_{1}\) is its slope, \(t\) refers to the time unit — years, \(i\) to the cross-national units — countries, while \(\varepsilon\) is the error term. \(Z\) represents a vector of the control variables described above.
Using the EU LFS and other sources of data, we constructed a longitudinal dataset of the EU15 countries for the period of 2004–2019 on which we estimate Eq. 1 using the fixed-effects and random-effect panel estimators. In order to decide on the appropriate model for our data, we conducted a series of specification tests. We begin with a Hausman (1978) test, which assesses whether the errors (ui) are correlated with the regressors, with the null hypothesis being that they are not. The tests suggest that the random effects estimator is consistent both in case of the EU2 model (p = 0.993) and in case of the EU8 model (p = 0.303). The Wooldridge test for serial correlation indicates that the residuals are autocorrelated in both the EU2 (p = 0.028) and EU8 (p = 0.022) models. Lastly, a Breusch-Pagan Lagrange multiplier test for contemporaneous correlation suggests that residuals are correlated across countries in the same cross-section (p = 0.000 for both models). Based on these results, we proceed to employ a fixed effects model with robust standard errors clustered at the country level, which corrects for these deviations and allows for a better inference using time series cross-sectional data. In order to control for the influence of aggregate time series trends, we employ time fixed effects across all our models.