Self-employment and the local business cycle

Abstract

The business cycle is likely to be of importance for self-employment rates. When the economy is growing, business opportunities open up and encourage the setup of new firms. In downturns, self-employment may be a way to avoid unemployment. The strength of these pull and push factors may depend on the amount of human capital a person has. The findings in this paper show that although the local business cycle is of minor importance for total self-employment rates in Sweden, there are heterogeneous effects across groups. People with higher human capital endowments are more likely to be pulled into self-employment, while those with lower human capital endowments are to a larger extent pushed into self-employment. This pattern is particularly strong for women. The study contributes to our knowledge of how individuals respond to business cycle changes as well as towards understanding why the association between the business cycle and self-employment rates differ across countries.

1 Introduction

The state of the economy may be of importance for self-employment rates. When the economy is growing, there are likely to be new business opportunities that encourage the setup of new firms. In downturns, self-employment may instead be a way to avoid unemployment. The business cycle may thus act to pull and push individuals into self-employment, which makes it unclear whether the self-employment rate is likely to be higher or lower in good times. Evidence in the literature on national business cycles also shows conflicting results; for example, Blanchflower (2000) and Koellinger and Thurik (2012) find that the correlation between the national business cycle and self-employment differs across countries.

Push and pull factors may be of importance for different groups depending on, for example, labor market status, education level, access to resources, or entrepreneurial skills. Several studies find that being unemployed increases the likelihood of becoming self-employed (see, e.g., Evans and Leighton 1990; von Greiff 2009), which suggests that people with weaker attachment to the labor market resort to self-employment to a greater extent. On the other hand, people with more resources in the form of human or financial capital may be more able to exploit business opportunities when they arise.

This paper studies the impact of the local business cycle on self-employment rates in Sweden among the working-age population between 1996 and 2007. More specifically, it looks at whether responses to the local business cycle differ depending on the amount of human capital the individual possesses. The hypothesis is that individuals with higher human capital endowments are less likely to become self-employed in a recession since their position on the labor market is stronger. Moreover, they may also have the skills necessary to exploit the business start-up opportunities that arise when the economy is booming. Self-employment among individuals with higher education should thus be more pro-cyclical compared to the self-employment rates among individuals with lower educational attainment.

This is analyzed by studying how the probability of being self-employed is affected by the local business cycles measured as vacancies in the local labor market. The results show that on average, the local business cycle measured as the local vacancy rate seems to be of minor importance for self-employment rates. For men, there is no effect on average, and for women, there is a small positive effect. The average effect hides considerable heterogeneity across groups with different educational attainment. People in a local labor market with a higher education level are more likely to be self-employed when the vacancy rate is high relative to people with a lower education level. Furthermore, studies of individuals who stay in the same region during the whole period show that highly educated individuals are more likely to switch to (from) self-employment in good (bad) times. The findings are consistent with the conclusion that individuals with higher human capital endowments are to a larger extent pulled into self-employment, whereas individuals with lower human capital endowments are to a larger extent pushed into self-employment. The pattern looks the same for both men and women but is much stronger for women.

This study adds to the relatively small volume of the literature on the effects of local labor market conditions on self-employment rates. It also provides evidence on how human capital endowments interlink with the local business cycle when people are deciding on whether to become self-employed or not. The study largely confirms the result found in Moore and Mueller (2002) and Ritsilä and Tervo (2002) that local labor market conditions have on average no impact on self-employment rates. Few previous studies have investigated whether this could be because different groups respond differently to the business cycle. One exception is Tervo (2006) who finds that the response to local unemployment differs depending on family background: High local unemployment pushes individuals from self-employed families into self-employment, while it lowers the probability to become self-employed for individuals from families where the parents were wage earners. Tervo points to the importance of entrepreneurial knowledge and contacts with other self-employed individuals for how people respond to the local business cycle. This study shows that human capital in the form of education may be important.

The microeconomic approach applied in the paper also contributes toward an understanding of the inconsistent results found in cross-country studies of the relationship between the national business cycle and self-employment rates. My results show that the strength of pull and push factors differs depending on individual characteristics suggesting that both institutions and labor force characteristics may affect the relationship between the business cycle and self-employment rates. Thus, the effect of the business cycle is likely to differ across countries.

The second section in this paper discusses theoretical considerations and earlier literature. The third section describes the empirical model, while information on data and variables is presented in section four. Section five provides cross-sectional evidence on the correlation between local labor market conditions and self-employment, before turning to the main results from panel regressions and robustness checks. I also present some results on the choice of organizational form and the business cycle. The last section concludes.

2 Local business cycle and self-employment: theory and literature review

The motives for being self-employed are many. The underlying model guiding this study, as well as most other studies of self-employment, is as follows: A person becomes self-employed if the net present benefits of becoming self-employed exceed the net present value of the costs involved. The opportunity cost consists of the net market wage he or she can earn as wage employed and the unemployment benefits that the person will receive if he or she is unemployed.¹ Pull factors are then positive in the sense that they increase the value of being self-employed or lower the direct cost of starting a firm. Push factors on the other hand lower the opportunity cost thereby making self-employment a more attractive option.

The literature points to several factors that may pull individuals into self-employment, such as access to capital or entrepreneurial knowledge, flexible working hours, the desire to be one’s own boss, or innovation-related motives.² The main push factor studied in the literature is unemployment. Since unemployment lowers the value of not being self-employed, self-employment may be a way to escape unemployment. Several studies also confirm that the unemployed are more likely to become self-employed (see, e.g., Carrasco 1999; Evans and Leighton 1990; von Greiff 2009).

Other factors discussed in the literature are discrimination and culture. Discrimination in wage employment has been suggested to explain high self-employment rates among some ethnic groups since it makes self-employment more attractive.³ Although many of the factors described above, such as flexible working hours or the desire to be one’s own boss, do not vary over the business cycle, some factors do. In particular, business opportunities are more likely to open up in good times, thereby pulling people into self-employment. In downturns, when employment prospects worsen, people may be pushed into self-employment.

Evidence for the impact of the national business cycle on self-employment is mixed. Evans and Leighton (1990) find support for the push theory that there is a positive relationship between self-employment and unemployment rates on the national level in the USA. In a study of GDP growth on the transition between self-employment, wage employment, and unemployment in Germany, Constant and Zimmermann (2004) find a pattern consistent with the conclusion that self-employment is used to escape unemployment. On the contrary, a study by Robson (1998), based on various sources of aggregate data in the UK, finds no evidence of a recession push. In a panel study of OECD countries, Blanchflower (2000) finds that the relationship between self-employment and unemployment rates varies across countries. High unemployment rates seem to increase self-employment in countries such as the UK, Germany, and Sweden, and decrease self-employment in Austria, while there is no correlation in France. Koellinger and Thurik (2012) also find that the relationship between the self-employment rate and the business cycle differs across countries. Using data from 22 OECD countries, they find that changes in the self-employment rate precede the global cycle but lag behind the national unemployment rates. A closer study of each country shows that the relationship with the unemployment rate is only visible for 7 out of the 22 countries.

There are several reasons as to why results could differ across countries. First, different patterns across countries may reflect institutional differences and the way the labor market works. For example, rigid wages may mean that individuals with lower productivity are pushed into self-employment when demand is low and tax systems may be more or less favorable to the risk-taking linked to new business opportunities. Second, studies on country-level data cannot control for policy changes on the national level which may affect self-employment rates. Correlations between the business cycle and self-employment may be blurred by national policy changes that affect self-employment.

By studying correlations with the regional business cycle, the researcher can control for changes in national policy. The literature on effects of regional business cycles on self-employment shows that the local business cycle has limited impact. Using Canadian data, Moore and Mueller (2002) find that although unemployment increases the likelihood of becoming self-employed, self-employment decisions are independent of the situation on the labor market as measured by the unemployment rate. Ritsilä and Tervo (2002) find that self-employment rates increase when national GDP increases, but find no effects of regional labor market conditions in Finland.

One explanation for the absence of effects of the local business cycle is that aggregate numbers conceal different responses across groups. Groups that are more vulnerable to unemployment may in recessions be more likely to be pushed into self-employment. Arising business opportunities are more likely to be exploited by groups that are less financially constrained and have greater local network access and more suitable entrepreneurial skills. Human capital is a factor that should be important, both for vulnerability to unemployment and the ability to exploit business opportunities.

3 Empirical specification

This study uses education level as a proxy for a person’s human capital endowment. The categories are whether the individual has (1) only compulsory schooling (<10 years of education), (2) high school and some further education (10–14-year of education), or (3) at least 3 years of higher education (>14 years of education).

How do self-employment rates correlate with the business cycle? And to what extent do they vary with education level? I examine these questions by estimating the empirical model below using the local vacancy rate as proxy for the local business cycle.

$${\text{Self}}\;{\text{employed}}_{irt} = \alpha + \beta_{0} LM_{rt} + \beta_{1} High \, school_{it} + \beta_{2} University_{it} + \beta_{3} High \, school_{it} \times {\text{LM}}_{rt} + \beta_{4} \,{\text{University}}_{it} \times {\text{LM}}_{rt} + \beta_{4} \,{\text{University}}_{it} \times {\text{LM}}_{rt} + \beta_{7} \,{\text{age}}_{it} + \lambda_{t} + {\text{region}}_{r} /{\text{ind}}_{i} + \varepsilon_{irt}$$

where the outcome self-employed is an indicator variable taking value 1 if an individual i is self-employed in year t. Our variable of interest is LM (local labor market), which will be measured with local vacancy rate in region r in year t. High school and University are dummies taking value 1 if the individual has as maximum a high school diploma, or at least three years of university studies, respectively. These variables are interacted with LM to study heterogeneous responses to the status of the local labor market. I include age, age², and year fixed effects or group-specific year fixed effects. To study changes over time, regional or individual fixed effects will be included. ε is the usual error term. Since there may be common shocks to the local labor market, standard errors will be clustered on the regional level. As a robustness check, the model will also be estimated using the logit model.⁴

The interpretation of the estimates differs somewhat depending on the set of fixed effects included. Regional fixed effects will remove permanent differences across regions that are correlated with self-employment. A change on the local labor market can be correlated with self-employment because (1) people living in the region change to/from self-employment; (2) people change region, and this systematically correlates with local labor market conditions; or (3) there is a systematic correlation between local labor market conditions and the occupational choice of people entering or exiting the sample. In specifications using individual fixed effects, the sample will be restricted to individuals who live in the same region during the whole period. In that case, the estimated effects will show how local labor market conditions are correlated with the individual’s choice to change to and from self-employment.

The reason to focus on people who stay in the same region in the fixed-effects specification is that then all other labor market factors are the same. People who move are not only exposed to a different labor market situation; the whole economic environment is different. However, labor market conditions may of course be the reason for moving to another labor market, and people who move may be different from people who stay. If people who stay in a region are more or less likely to become self-employed depending on the local business cycle, than people who move the results are not representative for the population as a whole. This should be kept in mind when interpreting results from the individual fixed-effects model.

In this model, individuals are categorized as self-employed or not. This means that individuals who have income from both self-employment and employment have to be referred to only one of the categories. How this is done will be explained in the next section. Another strategy could be to estimate a model where individuals are allowed to choose between three states: only self-employment, self-employed and employed, or not self-employed. Using this strategy would mean using more of the information in the data. I will therefore also estimate a multinominal logit model where individuals are categorized into these three states.⁵

The literature has identified several other factors that correlate with self-employment, such as marital status, number and age of children, and access to financial capital, which are not included in the empirical model (see, e.g., Hammarstedt 2009; Lindh and Ohlsson 1996). For some of these factors, such as number of children, it is not clear how occupational status may be correlated with local labor market conditions and they are therefore unlikely to be important confounders. For other variables, such as access to financial capital, there may be some correlation with the local business cycle. Perhaps, access to capital is better when prospects on the market are good. However, the data do not include any measures of access to financial capital and can therefore not be controlled for in the empirical study. Since this is likely to be correlated with education level, it implies that effects of education could be partly dependent on access to financial capital.

4 Data and variables

The empirical analysis is based on register data from Statistics Sweden and data from the public employment services (PES). The data set spans 1996–2007 and includes information on employment, income, and education attainment, as well as age and sex for all individuals living in Sweden from the age of 16. In this paper, the population is restricted to men and women aged 30–60. The age restriction focuses the study on individuals who are likely to have finished their studies and who do not choose self-employment as a way to partly retire. As is common in studies of self-employment, individuals in the agricultural sector are excluded.

As discussed in Sect. 2, the theoretical model of self-employment predicts that a person becomes self-employed if the net present benefits of becoming self-employed exceed the net present value of the costs involved. The opportunity costs are the earnings he or she can get as wage employed and the unemployment benefits available if he or she is unemployed. I therefore study here how the occupational choice between self-employment and wage employment/unemployment correlates with the business cycle on a sample that includes all individuals in Sweden aged 30–60.⁶

Information on employment status comes from November and is based on information from the tax authorities. The definition of self-employment is the one used by Statistics Sweden based on source of income and whether the self-employed is actively running the firm. An individual is defined as self-employed if he or she in November only had income from his or her own firm. Individuals with income from both their own firm and wage employment are categorized as self-employed if the income from self-employment multiplied by 1.6 is higher than the income from wage employment. The income from self-employment is weighted up with 1.6 because earnings from self-employment are on average lower in comparison with the time spent working.⁷ In addition, to be categorized as self-employed, the person should have indicated to the tax authorities that they are running the firm actively. The self-employed with both incorporated (firms with less than 11 stockholders) and unincorporated businesses are included. The organizational forms differ in that incorporated firms required an initial capital investment during this period of 100,000 SEK. Another difference is that an incorporated firm has limited liability, implying that the business owner does not have personal responsibility for the financial health of the firm, which is the case in an unincorporated firm. In 2004, Statistics Sweden made some changes to the classification of self-employment, resulting in an increase in the number of self-employed, especially self-employed with incorporated business. To account for this change, an indicator for the years 2004–2007 is interacted with the education dummies and the age variables.

The factor 1.6 has been calculated from the earnings statistics and is the ratio between the average earnings for full-time employed and full-time self-employed in the age group 20–64. A potential problem with deriving the weight this way is that self-employed might earn a different wage on average if employed and an employed person would on average earn a different income from self-employment if self-employed. As a robustness check, I derive another measure of self-employment using data on earnings from self-employment and employment for people who have income from both. A detailed description of the measure and results is available in the online appendix.

The data also include information on education level from the Statistics Sweden education register. Education categories are constructed using these data.

The indicator used to measure the local business cycle is the local vacancy rate. The vacancy rate is calculated as the number of new vacancies remaining more than 10 days after being posted by the PES divided by the population aged 18–64. Many studies have used the unemployment rate as a proxy for the business cycle. The problem with using the local unemployment rate is that unemployed are also part of the dependent variable, which creates a mechanical dependence between the two. The vacancy rate is not directly affected by someone moving from unemployment to self-employment. However, there may be an indirect effect since the number of vacancies will be affected if someone takes a job or chooses to become self-employed instead of filling the vacancy. A potential weakness of this measure is that not all vacancies are reported to PES since PES has a market share of about 30–40 % of all recruitments. According to PES, this market share tends to rise when the demand for labor is high and decreases somewhat when demand is low. This measure is, however, regarded as a good indicator of the demand for labor in the economy.⁸ Moreover, what is important for this study is that there are no geographical differences in reporting of vacancies that change over time. There is no indication that the reporting differs across regions.

Although there may be a mechanical connection between unemployment rates and the probability of being self-employed, I will also present results using data on local unemployment from the Public Employment Agency. Since data on unemployment are available from earlier years, this part of the study will be conducted on data from 1993 to 2007.

Local labor markets are constructed by Statistics Sweden based on commuting patterns and derived in two steps. First, Statistics Sweden identifies local labor market centers. These centers are municipalities where at least 80 % of the population work in the municipality and no more than 7.5 % of the population commute to another specific local labor market center. In the next step, other municipalities are connected to the local labor markets that most people in the municipality are commuting to. Although the definition of local labor markets makes sure there is limited mobility across local labor markets, this study shares the problem with similar studies that some workers commute or move to other local labor markets. It is, however, difficult to know how this will affect the interpretation of the results. This study uses the local labor market categorization in 2000 when there were 90 local labor markets in Sweden. Variable definitions and data sources are summarized in the “Appendix.”

Table 1 displays summary statistics for the total sample and individuals that are self-employed. Summary statistics for men (presented in the upper panel) and women (lower panel) show that 9.7 % of the population of men and 4.2 % of women are self-employed, and 42 % of the self-employed men have an incorporated firm, while the share for women is 30 %. The background characteristics look rather similar for men and women: Individuals with university education are less likely to be self-employed, and individuals with only compulsory schooling are more likely to be self-employed. Finally, the self-employed are on average almost two-year older than the average population age.

Table 1 Summary statistics

The vacancy rate is on average 0.67 % over the period. An inspection of the average vacancy rate for the different groups shows that self-employment is slightly more common in local labor markets where the vacancy rate is high. There is considerable variation across local labor markets and, more importantly for this study, within labor markets. For example, in labor market Stockholm, the average yearly change is 25 % age points, and the increase from the lowest vacancy rate (in year 2003) to the highest (in year 2007) is 145 %.

5 Results

5.1 Long-term differences across regions

Before turning to the panel results let us take a look at the correlation between self-employment and local labor market conditions in a cross section. This relationship is likely to reflect long-term differences across regions, both in labor market conditions and other permanent differences, such as industry structure and geographical components. Table 2 displays the results from regressions on the pooled data controlling for year fixed effects. The first two columns show the result for women and columns 3 and 4 the results for men.

Table 2 Cross-sectional study of local labor market conditions and self-employment

The estimates presented in the first column show that a woman is more likely to be self-employed if she resides in a labor market region with a high vacancy rate: A standard deviation (a 40 %) increase in the local vacancy rate increases self-employment by about 0.5 % age points or 11.0 %. The results presented in column 3 imply that for men, the size of the effect is 0.8 % age points or 8.3 %.

Next, I study whether there are differences depending on the individual’s education level. The reference group, here and throughout the paper, is people with only compulsory schooling. According to the results presented in columns 2 and 4, individuals with a higher education level are even more likely to be self-employed in regions with high vacancy rates compared to individuals with a lower education level. The difference is statistically significant for women, but for men there is no statistically significant difference between individuals with a high and low education level. The estimated effect of a one standard deviation increase in local vacancy rate for women (men) with university-level studies is 0.6 (0.8) % age points or 15.2 (7.7) %. The results thus show that in regions with high vacancy rates, a larger share of the workforce chooses to be self-employed. This pattern is especially pronounced for women with high education.⁹

The estimated effects of the different individual characteristics are similar to other studies on Swedish data: Self-employment increases with age, although at a diminishing rate, and individuals with higher education are less likely to be self-employed (see, e.g., Hammarstedt 2006). Overall, these results suggest that people are to a larger extent pulled rather than pushed into self-employment. This pattern is, as predicted, stronger for people with higher human capital endowments.

5.2 Local business cycle and self-employment

The cross-sectional evidence suggests that self-employment is positively related to the vacancy rate. The problem with cross-sectional studies is that they suffer from the problem that the effect of the vacancy rate may reflect permanent differences across regions. A panel study of different regions can solve some of these problems since it is then possible to control for both permanent differences across regions and changes on the national level that may influence the level of self-employment.

So does the local business cycle matter for self-employment rates? To investigate this question, I now proceed to the panel study. Tables 3 and 4 present the estimates from a panel study of the effect of changes in local labor market conditions on self-employment for women and men, respectively.

Table 3 Local business cycle and self-employment among women

Table 4 Local business cycle and self-employment among men

Starting with women, results presented in Table 3 column 1 show that on average, there is a positive effect of changes in the local vacancy rate on the probability to be self-employed. The effect is small: A standard deviation (a 40 %) increase in the vacancy rate increases self-employment with 2.1 %, suggesting that the average effects found in the cross-sectional study to a large extent reflect permanent differences across regions that correlate with the self-employment rate.

Are there heterogeneous effects depending on a person’s human capital endowment? As can be seen in the second column, women with higher education, compared to the reference group, are more likely to be self-employed when vacancy rates are high. The estimated differential effects across groups are similar to the cross-sectional evidence presented in Table 2 column 2. A standard deviation increase in the vacancy rate increases self-employment among highly educated women with 5.7 %. Not only are there statistically significant differences across groups, the total effect is statistically significant showing that self-employment among women with only compulsory schooling is countercyclical, whereas self-employment among women with higher education is pro-cyclical.

The estimated effects of local vacancies presented in the second column depend on several things. First, a change in labor market conditions can be correlated with a change in an individual’s choice between wage employment and self-employment. Second, the composition of people may change because individuals move to a new region or because some enter/exit the sample. To isolate the different mechanisms, I first exclude all individuals who move to another region during the period. As can be seen in the third column, the results do not change much by restricting the sample to individuals who stay in the same labor market region. In the fourth column, the year fixed effects are replaced by educational group × year fixed effects to allow self-employment among different educational groups to be on different trends. The estimates remain unchanged; if anything, the estimated difference across educational groups is slightly larger.¹⁰

To isolate the effect of the local business cycle on individual behavior, the regional fixed effect is exchanged for an individual fixed effect. The results in columns 5 and 6 show a similar pattern to the previous results. However, the negative effect of the local labor market conditions on women with low education now disappears. This suggests that the effect was due to those women entering (and exiting) the sample being to a lesser (and larger) extent self-employed in times when local labor market conditions were good. Remember that people enter the sample aged 30 and exit aged 60. Since the negative effect does not arise from women in the sample changing to self-employment when the local vacancy rate is high, the effect is due to the self-employment pattern of women entering and exiting the sample. Self-employment among women with higher education is pro-cyclical also in this specification.

The results for men are presented in Table 4. The results are similar to the results found for women, but much weaker. The results in the first column show that on average, there is no effect of local labor market conditions. Including interactions with education level, the results presented in column 2 show that self-employment among men with higher education is pro-cyclical (the total effect is statistically significant), yet there is no statistically significant difference between men with high and low education level. Restricting the sample to men who live in the region, the whole period strengthens the result (see column 3).¹¹ In order to study whether men change occupational status depending on local labor market conditions, the regional fixed effect is replaced by an individual fixed effect. The results presented in the two last columns show that men with higher education are more likely to switch to (switch from) self-employment when local labor markets are good (bad). Results presented in columns 4 and 6 show that the inclusion of education group × year fixed effects weakens the estimated differences between educational groups. Calculating the size of the effect using the estimates in column 6 suggests that one standard deviation increase in the vacancy rate increases self-employment among highly educated men with 2 %.

These results are robust to using an alternative definition of self-employment and estimating the relationship in a logit model. A detailed description of the robustness tests and the results can be found in the online appendix.

To summarize, local labor market conditions have no (for men) or weak (for women) effect on self-employment on average. However, there are differences across groups to support the hypothesis that people with higher human capital endowments are to larger extent pulled rather than pushed into self-employment.

5.3 Alternative measure of local business cycle: the unemployment rate

Table 5 shows the results using local-level unemployment as a proxy for the business cycle. As discussed earlier, this measure has previously been used in the literature but may suffer from larger endogeneity problems than the vacancy rate. The top panel displays the results for women and the bottom panel for men. As can be seen in the first column, there is on average no correlation between self-employment and the local unemployment rate in any of the samples. However, there seem to be some differences depending on the individual’s education level. The results in column 2 show that men and women with a higher education level are less likely to be self-employed when the local unemployment rate is high. The size of the effect is somewhat larger compared to the results when using the vacancy rate: A standard deviation (35 %) increase in local unemployment reduces self-employment among women with a university degree with 0.7 % age points or 17 % compared to women with only compulsory schooling. For men, the decrease is 5.6 %.¹²

Table 5 Unemployment

Column 3 shows that the inclusion of group-specific time effects, thus allowing for different trends for different educational groups, does not the affect the result for women but does reduce the level of statistical significance on the estimated effects for men. The two last columns display the results for the fixed-effects model. In contrast to earlier results, including group-specific time effects has a major impact on the result for both men and women. The results in column 4 show that individuals with more education are less likely to switch to (from) self-employment when the local unemployment rate is high (low). The results in column 5 show no such effects. Instead, it seems like a higher local unemployment increases self-employment among individuals with a medium level of education compared to other groups. How can this result be interpreted? Comparing the results in columns 3 and 5 with the upper panel suggests that the negative interaction term for highly educated is due to a composition effect: Women who enter the sample (30-year olds) are to a larger extent self-employed than women who leave the sample (60-year olds) when the unemployment rate is high. However, as can be seen when comparing columns 4 and 5, when controlling for group-specific time trends, there is no evidence that highly educated women switch to self-employment when unemployment rates are low.

5.4 Being both self-employed and employed

As discussed in Sect. 3, categorizing people as self-employed or not can be too restricting because people may choose to be both self-employed and employed. One way to take this into account is to define three states: (1) only self-employed, (2) self-employed and employed, and (3) not self-employed. This model is estimated using multinominal logit. Table 6 displays the results for women. Column 1 show the odds ratios for a person in a local labor market to have income from both self-employment and employment, and column 2 show the odds ratio for a person to be self-employed.¹³ The baseline category is to be not self-employed. We can see how a person’s education level is correlated with the probability of being in different states of self-employment by inspecting the estimate effects of High school and University. Interestingly, it seems as people with a high education level are more likely to have income from both self-employment and employment and are less likely to be only self-employed. The chance that women with higher education have income from both self-employment and employment is more than twice the chance that women with compulsory schooling have income from both sources. In contrast, women with only compulsory schooling are twice as likely to be only self-employed.

Table 6 Partly and only self-employed: women

Turning to the connection with the business cycle, the results show that the highly educated women are more likely to be self-employed, both partly and fully, when the vacancy rate is high. To facilitate the interpretation of the coefficients, I have standardized the vacancy and unemployment rates to zero with standard deviation one. The odds ratios for a highly educated are 1.076 for having income from both self-employment and employment and 1.242 for only being self-employed. This means that the odds on becoming only self-employed when the vacancy rate increases with a standard deviation is 24 % higher for highly educated women than for women with low education.

The next two columns show that the results are very similar when using the local unemployment rate as a measure for the local business cycle. The results show that people with higher education are less likely to have income from both self-employment and employment in labor markets when the unemployment rate is high (column 3) as well as having income only from self-employment (column 4).

Table 7 presents the corresponding results for men. As for women, men with a university degree are more likely to have income from both self-employment and employment and less likely to be only self-employed. Moreover, the results confirm previous findings that people with higher education are more likely to be self-employed when the local labor market is good. In line with previous results, the effects are weaker for men.

Table 7 Partly and only self-employed: men

5.5 Differences across organizational form

This section will look closer at how the situation on the local labor market correlates with whether the self-employed choose to run an incorporated or unincorporated firm. As can be seen in all specifications in Table 8, men and women with university education are more likely to have an incorporated firm if self-employed. Potential explanations to this pattern are that they may find it easier to make the initial capital investment of 100,000 SEK, or find the form more suitable or profitable for the type of service or good they are producing.

Table 8 Local business cycle and organizational form of self-employment

Next, we turn to the correlations between organizational form and the local business cycle. The results in columns 1 and 4 show on average no effect of the local vacancy rate on organizational form. Turning to the model that allows for heterogeneous responses across educational groups, we see an interesting pattern. The results in column 2 show that having an incorporated firm if self-employed is less common among highly educated women, compared to women with less education, when the local vacancy rate is high. However, when studying changes in self-employed women’s behavior, the results (presented in column 3) show that highly educated women are more likely to switch to an incorporated firm when the vacancy rate is high. These opposite results in the two specification suggest that highly educated women who enter (exit) self-employment in good (bad) times are more likely to run (have run) an unincorporated firm.

For men, the results from both the specification with a region fixed effect (column 5) and the specification with an individual fixed effect (column 6) show that men with higher education are more likely to run an incorporated firm when the vacancy rate is high.

How should these results be interpreted? It is not clear how the business cycle should affect the choice to have an incorporated or an unincorporated firm. One explanation for highly educated individuals switching to incorporated firms when the vacancy rate is high could be that it is easier to access financial capital in good times, especially if you have higher education.

6 Concluding discussion

This paper has investigated whether self-employment rates are correlated with the local business cycle. When the economy is growing, business opportunities open up to encourage the setup of new firms. In downturns, self-employment may be a way to avoid unemployment. The importance of these pull and push factors may be different depending on the amount of human capital a person has. The findings in the paper support this hypothesis: Individuals with a higher education level are more likely to be self-employed in regions where the vacancy rates are high compared to individuals with lower education. A part of this regional variation can be explained by regional fixed effects reflecting long-lasting differences across regions. However, when including regional fixed effects, there is still a correlation between self-employment and the local vacancy rate, which suggests that people with higher education are more likely to be pulled into self-employment when business opportunities arise compared to people with compulsory schooling only.

A fixed individual effect model is employed to study how individuals switch from and to self-employment, and it reveals that individuals with a higher education level are more likely to change to self-employment in good times compared to others. This pattern is especially strong for women even if the size of the effect is modest: A standard deviation increase in the vacancy rate increases self-employment among the highly educated by about 6 %, and among women with only compulsory schooling, there is no or even a negative correlation in some specifications between self-employment and the business cycle.

One reason for why previous studies have failed to find a relationship between self-employment and the regional business cycle could be because different groups are affected differently. Thus, studying aggregate numbers may conceal heterogeneous responses to changes in the economic environment. The micro-level approach chosen in this study may also contribute toward understanding macroeconomic phenomena. As discussed earlier, cross-country studies find no consistent relationship between the self-employment rate and the national business cycles. The results in this study suggest that this may be because the strength of the push and pull factors differs across countries. At national level, it is clear that institutions and the functioning of the labor market are important for how self-employment rates respond to changes in the business cycle. For example, push factors may be strong if wages do not adjust in downturns, and unemployment benefits are low since more people may resort to self-employment to be able to support themselves in downturns. On the other hand, pull factors may be more important if rules and tax systems are favorable to risk-taking and promote self-employment when there are many business opportunities. It is thus an open question whether we would see the same pattern in other countries with different labor market institutions.

This paper also provides some evidence on the choice of organizational form by those who are self-employed. Highly educated men and women are more likely to switch to an incorporated firm when local labor market conditions are good. Since starting an incorporated firm requires an initial investment, this result may indicate that access to financial capital is important for the choice of organizational form. These results pose interesting questions about the effect of the business environment and are left for future research.

Appendix

See Table 9.

Table 9 Detailed variable description