Business Start-Ups or Disguised Unemployment? Evidence on the Determinants of Self-Employment from Urban China

This study provides evidence on the determinants of self-employment for urban local registration residents in China. Using CHIP2007, the employment status is divided into four categories: selfemployed employers, own-account workers, employees, and the unemployed. Several major conclusions emerge. First, compared with the employee, holding other factors (e.g., human capital) constant, the wage premium associated with the self-employed employer is higher, while the wage premium associated with own-account workers is lower. Second, the influence of the wage premium on the self-employed employer is negatively significant, and the influence of the wage premium on the own-account workers is insignificant. These results reveal that compared with employees, being a self-employed or own-account worker is seemingly not a better choice for employment in urban China; being self-employed is similar to disguised unemployment. Third, considering the influence of all the factors: the wage level categories (wage levels in the public and private sectors), the entry period categories (the SOE reform period and the recent period), the age categories (aged 50 and over, and aged below 50), and the regional categories (the East, the Central, and the West regions), robust checks were conducted. In the own-account workers group, the business creation hypothesis is nearly rejected again; in fact, it is only supported for workers who entered the self-employment sector in the SOE reform period (entered early into the selfemployment sector group), and workers aged over 50.


Introduction
The self-employed sector is a representative informal sector of the employment market as noted in previous studies 1 . Schumpeter described the sector as the "prime mover of economic growth" (Schumpeter, 1943). De Soto (1989) described the emergence of self-employed workers as "the foundation of development." Transition economists believe the rise of self-employment to be a sign of the growing importance of markets relative to the state (Hanley, 2000;Dimova & Gang, 2007). Along with the economic transition in urban China, the number of self-employed workers 2 increased from 150 thousands in 1978 to 21.36 million in 2000, before further increasing to 52.27 million in 2011 (NBS, 2012).
Why was there a large change in the size of the self-employed sector in urban China during the economic transition period? There are two primary possible explanations. The first is the great increase in the number of migrants along with economic development and the deregulation of the registration system after the 1980s. The second explanation is that along with the transition from a planned to a market economy, the government enforced ownership reform of State-Owned Enterprises (SOEs) from the 1980s onward. A section of employees with urban registration in the SOEs became laid-off workers and some of them re-employed in the informal sector (e.g., by becoming self-employed workers). In addition, a section of workers voluntarily left SOEs to become owners in the private sector. This paper discusses the determining factors for the self-employed workers in urban China.
There are two hypotheses about self-employment discussed in previous studies. One is the "disguised unemployment hypothesis," which is indicated in the dualism theory of Todarro (1969) and Harris and Todarro (1970) 3 . Migrants to the informal sector can be explained by this hypothesis, as in the case of workers in SOEs who lost their jobs because of SOE reconstruction. Most laid-off workers were not re-employed by SOEs, so many entered the informal sector (e.g., to become self-employed workers) in order to make a living (Knight & Song, 1999). Considering the above, self-employment may result from forced recourse to the informal sector, in which the individual's activities and wage slightly differ from what they would be if the individual were unemployed. It is thought that self-employed workers barely make a living from working, receiving lower wages and working longer hours than those in the formal sector. Conversely, self-employed workers may also be successful business owners who create new business opportunities and many innovative new products ("business creation hypothesis"). For example, along with ownership reform beginning in the 1980s, some workers left SOEs to become owners of private firms and started new businesses. A number of these new entrepreneurs (employers) were communist party members or cadres in the public sector in the past, and it has been pointed out that such social capital positively affects the premium that may be associated with self-employment (Wu, 2006) As a result, a high percentage of self-employed workers may reflect an environment that encourages risktaking, business creation, and market development ("business creation hypothesis"), or it may be a result of the lack of jobs in the formal sector in which wages are set just above the market-clearing level ("disguised unemployment hypothesis").
Which hypothesis can explain self-employment in urban China? In this paper, we provide some evidence to answer this question 5 . In the previous empirical studies on this issue, although Earle and Sakova (2000), Hanley (2000), Dimova and Gang (2007) utilized microdata of Central and Eastern European economic transition countries to test these two hypotheses, there has not been an empirical study for China. One of the purposes of this paper is to resolve this dearth in empirical data. This paper is structured as follows. Section 2 reviews the literature, and Section 3 describes estimate methods, including introduction to the data and models. Section 4 states descriptive statistics and estimated results, and Section 5 presents the main conclusions.

Models
Firstly, to explicate the determinants of the self-employed in urban China, the employment status probability function is estimated using a multinomial logit model, which is represents in Equation (1). The explained variable takes on one value for four categories of employment status (self-employed employer, own-account worker, employee, and the unemployed). Here, referring Earle and Sakova (2000), Hanley (2000) and Dimova and Gang (2007), we defined own-account workers are those who work in small firms (or unit) which only him (herself) or no-paid family workers work in, self-employed employees are those who work in small firms with workers less than eight and they are the owners of these small firms. The reference category is the employee group.

( )
Pr i Y n = indicates probability of one kind of employment status, X are factors affecting the employment status probability, β are the estimated coefficients, and α is a constant.
Then we used two kinds of methods to test the "disguised unemployment hypothesis" and "business creation hypothesis". The one is a comparison of average wage levels of self-employed employer group, own-account worker group and employee group (Hanley, 2000). For example, holding the other factors (such as human capitals) constant, if the average wage level of own-account worker group is lower greatly than employee group, it shows that own-account workers are nearly the disguised unemployed, and labor market is segmented. In order to gain these imputed wage, wage functions by different employment status groups are estimated. Here, Maddala (1983)  In Equation (2.1), i denotes workers, and logWage indicates the dependent variable (as the logarithm of wage rate). Emp is an index indicating employment status (self-employed employer, own-account worker and employee), H are factors affecting earnings. Emp γ and H γ are the estimated coefficients. Further, α is a constant and u is the error term.
Considering the selection bias problem in Equation (2.1), the selection bias corrected wage function model is proposed (Maddala, 1983). Equation (2.2) expresses the probability of employment status using the probit regression model. For example, the probability to become a self-employed employer is expressed as and the other probability (such as employee, own-account worker, the unemployed) is expressed as The other test is the estimation of the effects of earning premiums on the probability of employment status (Earle & Sakova, 2000). We utilize a multinomial logit model shown in Equation (3). In Equation (3), wage premiums (WP) are added as new variables, the other variables are similar with Equation (1). It is thought that higher the wage premium, higher the probability to choice the employment status. For example, when the estimated results of wage premium ("Wer/Wee") is positive significantly on the probability to become a selfemployed employer, it is shown that the self-employment is a new business to gain more income and create more values (such as create new jobs for others, and new goods), so the "business creation hypothesis" is supported. While, when the estimated results of wage premium ("Wer/Wee") is negative significantly (or insignificantly) on the probability of self-employed employer, it is shown that although becoming a self-employed employer can't gain more, he (she) has to choice to become a self-employed employer, it indicates that the entry to the informal sector may be an involuntary behavior, and "disguised unemployment hypothesis" is supported.

Data
The 2007 Chinese Household Income Project Surveys data (CHIP2007) are used for the analysis. The survey was conducted by NBS (National Bureau of Statistics) and Beijing Normal University in December 2008 including respective information about employment status and wages of the urban registration residents. The survey selected the represented districts in China, including the East region (Liaoning, Jiangsu, Guangdong), the West region (Sichuan, Yunnan, and Gansu), the Central region (Beijing, Shanxi, Henan, Hubei, Anhui). The sampling method is stratified random sampling method based on the national samples of the NBS. From the above, the dependent variable is an employment status category variable. The independent variables are as follows ( Table 1 shows sample statistical descriptions by employment status).
First, we utilized some variables used in previous studies 7 . These include individual variables likely to affect employment status choice, such as schooling years, tenure years, health status (very good, good, normal, bad) dummy variables, which are the index of human capital, female (female is a binary variable coded 1 if the respondent is a female and 0 otherwise), and Han race (Han race is a binary variable code 1 if the respondent is Han race, and 0 minority). In addition, it is thought the risk aversion preferences vary with these individual attributes, and the risk aversion preference becomes more likely with increasing age.
Some previous studies indicate that family factors, such as child, marriage status, parent education, and parent occupations, can affect the choice to become a self-employed worker, particularly for female workers. We used a marriage dummy, number of children, parent education (a senior high school and over dummy), and parent occupation (manager dummy) to control the influence of these factors.
As indicated in the liquidity constraint hypothesis, financial factors may affect the choice to become selfemployed, and here, we use living together with parents and household income as the financing constraint index.
It is pointed out that social capital also affects self-employment. It is thought that with higher social capital, there is a greater possibility for settling financial constraint problems and thus a greater chance for success with self-employment. We use two variables-social relations numbers and frequency of contact to relations-as the index of social capital, using the following questionnaire items: "How many persons do you contact?" and "How frequently do you contact your relations?" We also consider the influences of some special factors in the Chinese labor market. For example, in urban China, the change from rural registration to urban registration is very difficult except under special conditions, such as workers with higher levels of schooling or with higher skill levels, enlistment in the army, and purchase of a commercial house (investment in housing) in the urban area. It is thought that registration change may influence the choice of employment status, so we add a registration change dummy (a binary variable coded 1 if the worker experienced a registration change and 0 otherwise). In addition, because there are regional disparities in China, it is thought that labor demands vary by region, so we add three regional dummies (West, East, and Central regions). In order to test the hypothesis, wage 8 premiums (wage differentials between employment status groups) are calculated-the wage ratio of own-account worker to employee (Woa/Wee) and wage ratio of self-employed employer to employee (Wer/Wee) 9 . As the distribution of this variable is skewed, its natural logarithmic forms are used.
We also distinguish employee wages by private and public sector and perform robust checks to test the hypotheses. Reduced earning function estimation results are utilized to calculate these imputed wages and wage differentials.
This paper focuses on self-employed employers, own-account workers, employees, and the unemployed. Considering that the retirement system is structured within the public sector, in order to diminish the effect of that system on analysis results, the analytic objects are limited to groups between the ages of 16 and 60. The samples utilized in the following empirical studies comprise 10,806 urban residents and 6,267 migrants.

Distributions of Employment Status in Urban China
The distributions of employment status of urban registration residents in urban China are shown in Table 2, the divisions are as follows: 1.4 percentage self-employed employers, 2.71 percentage own-account workers, 65.84 percentage employees, and 30.02 percentage the unemployed.

Wages, Working Hours and Household Income by Employment Status Groups
Mean values and standard deviations of wages, working hours, and household income by employment status group are shown in Table 3, Kernel density estimation results are expressed in Figures 1-3.      Considering average wages for each group, compared with employees group, wages are higher for selfemployed employers group (er/ee 1.70) and lower for own-account workers group (oa/ee 0.84). There are working hour disparities among employment status groups. Concretely, compared with employees group, the working hours are longer for both own-account workers group (oa/ee 1.45) and self-employed employers group (er/ee 1.31). However, the household income differentials by employment status group are smaller.
Kernel density estimation results show that compared with employees and own-account workers groups, wage and household income are higher for self-employed employers group. In addition, compared with employees, most of self-employed workers (either self-employed employers group or own-account workers group) are working in longer time.
Although these tabulated calculations indicate the existence of wage, work hours, and household income differentials by employment status group, it is not clear as to what determines the choice of employment status and which hypothesis explains self-employment. These questions will be answered using the econometric analysis results discussed in the following section. Table 4 indicates the estimated results of the determinants of entry into the self-employment sector, except for the wage premium. The main findings are as follows.

What Determines a Worker's Entry into the Self-Employment Sector?
First, self-employed employer increases with increase in the household income. This can be explained by the existence of financial constraints for self-employed employers in these two groups. This result provides evidence that in trying to create new jobs or businesses, policies (such as financial support policies for small enterprise) to resolve financial constraint problems are important in a transition economy (and elsewhere).
Second, the Hukou (registration) system influences entry into the self-employment sector. Concretely, the experience of change from the rural registration to the urban registration increases the probability of becoming a self-employed employer or an own-account worker. It is indicated that compared with the local urban registration residents group, urban residents through migration mostly work in the informal sector. This result is consistent with previous studies on Mexican immigrants and self-employment (Waldinger, 1986;Borjas, 1986) 10 .
Third, considering social capital, compared with the workers who contact relations once every week, those who contact relations once every month are more likely to become own-account worker.
Finally, compared with workers in the West and Central regions, the probability of becoming a self-employed worker is lower for those in the East region. This may be because compared with the West and the Central regions, the level of economic development is higher in the East region, so labor demands are also relatively higher in that region.
The results using subsamples show a standard shape of the wage function for employees, particularly in terms of human capital variables and gender, whereas the effects of these factors on the wages of self-employed employers and own-account workers are small. In addition, estimated results show the existence of wage differentials between employment status groups. For example, compared with the employee and holding other factors (e.g., human capital) constant, wages are higher for the self-employed employer, whereas wages are lower for the own-account self-employed worker than that for the employee. Holding other factors constant, a worker can gain more economic benefits by becoming a self-employed employer but gains less by becoming an own-account employer. Compared with the employee, the economic benefit for the self-employed employer is better, but it is worse for the own-account worker. Based on these estimated results, the disguised unemployment and business creation hypotheses are not clearly supported.

Hypothesis Test: Business Start-Up or Disguised Unemployment?
In order to directly test these hypotheses using the imputed wages calculated based upon the results shown in Table 5, the reduced multinomial logit analysis is estimated. These estimated results are represented in Table 6. Source: Calculated using CHIP2007. Note: 1) * , ** , *** : statistical significant in 10%, 5%, 1% level.

X. Ma
2) The specification of strauctural MNL is similar to that shown in Table 4, but dependent variable has only three categories (omitting the unemployed group) and the imputed wage differentials log(Woa/Wee) and log(Wer/Wee) are added to the regressors. All other independent variables shown in Table 4 are also included here, but not be shown.
First (Panel A), the results show that the wage premium (logWoa/Wee, logWer/Wee) does not statistically affect the probability of employment status. Based on the individual utility maximum rule (e.g., to gain the highest income), workers possibly chooses to become a self-employed employer or an own-account worker when their associated wage levels are higher than those for employees. There is no significant positive relation between the wage premium and the probability of being self-employed, indicating that the choice to enter the self-employment sector (as either self-employed employers or own-account workers) does not result from perceived economic gains and benefits. These results support the disguised unemployment hypothesis, whereas the business creation hypothesis is rejected.
Second, as shown in Panel B, considering the labor market in urban China is segmented by the public and private sectors 11 , and the wage level in the informal sector is close to that in the private sector, we analyze the estimated results of the effects of wage premiums between the private sector and other sectors in the following discussion. The influences of wage premiums (logWer/Weepri) on the probability of becoming a self-employed employer are negatively significant (−0.4643). These results are consistent with the above, and the disguised unemployment hypothesis is supported, whereas the business creation hypothesis is rejected. In addition, the influence of the wage premium (logWoa/Weepri) on the probability of becoming an own-account worker is insignificant, so the business creation hypothesis is rejected in these two groups again.
Third, as shown in Panel C, as described above, in the 1990s, the Chinese government enforced ownership reform in SOEs. As a result, a section of workers were displaced, and some of them entered into informal sectors. In contrast, some workers voluntarily left SOEs to become self-employed employers. If the effects of those who left voluntarily is greater than those who were displaced, the wage premium should positively affect the choice to become a self-employed employer or an own-account worker. We considered the subsample of those with more than 10 years tenure as the group who entered the self-employed sector before 1997 (named the "SOE reform period") and the subsample with less than 5 years tenure as the group who entered into the self-employed sector after 2003 (named the "recent period"). The influence of wage premiums (logWer/Wee) on the probability of becoming a self-employed employer is negatively significant in both the SOE reform period and the recent period (the SOE reform period −1.1126, and the recent period −1.5110). These results indicate that the choice to become self-employed does not result from the perceived economic gains and benefits; it is more likely an involuntarily choice or behavior. In other words, a worker is pushed into the self-employed sector to make a living. Based on these results, in the self-employed worker group, the business creation hypothesis is also rejected in both periods. However, the influence of wage premiums (logWoa/Wee) on the probability of becoming an own-account worker is positively significant in the SOE reform period (1.0779). This indicates that the choice to become an own-account worker results from perceived economic gains and benefits. Therefore, this is perhaps a voluntarily behavior among the own-account workers group, the business creation hypothesis is supported in the SOE reform period.
Fourth, as shown in Panel D, as indicated in previous studies, risk aversion preferences become stronger with increasing age. The choice of becoming a worker in the self-employment sector is different by age groups. Two subsamples were utilized to test the hypothesis again-the age 50 and over group and the under age 50 group. In the group aged 50 and over, the influences of wage premiums (logWoa/Wee) on the probability of becoming an own-account worker is positively significant (0.7143) at the 10% level. Compared with the younger workers, middle-aged and elderly workers are more likely to become own-account workers based on perceived economic gains and benefits. The business creation hypothesis is rejected in the younger group while it is more likely supported in the middle-aged and elderly groups.
Fifth, as shown in Panel E, the influence of wage premiums (logWer/Wee) on the probability of becoming a self-employed employer is negatively significant in the East region (−0.8592), while the influences of wage premiums on the probability of becoming an own-account workers are insignificant in both the Central/West regions. The business creation hypothesis is rejected again in each region.

Conclusions
This paper provides evidence on the determinants of self-employment for urban registration residents and migrants in urban China. Using CHIP2007, the employment status is divided into four categories: self-employed employers, own-account workers, employees, and the unemployed. Several major conclusions emerge.
First, compared with the employee, holding other factors (e.g., human capital) constant, the wage premium associated with the self-employed employer is higher, while the wage premium associated with own-account workers is lower.
Second, the effects of the sample-selection-bias-corrected wage premium for self-employed employers are insignificant. The business creation hypothesis is rejected, while the disguised unemployment hypothesis is supported, showing that the self-employed workers possibly works in sectors with lower economic benefits.
Third, considering the influence of all the factors: the wage level categories (wage levels in the public and private sectors), the entry period categories (the SOE reform period and the recent period), the age categories (aged 50 and over, and aged below 50), and the regional categories (the East, the Central, and the West regions), robust checks were conducted. In the own-account workers group, the business creation hypothesis is nearly rejected again; in fact, it is only supported for workers who entered the self-employment sector in the SOE reform period (entered early into the self-employment sector group), and workers aged over 50. These estimated results revealed that compared with employees, self-employed employers and own-account workers do not gain more, and there seemingly are no better choices in urban China. The one of reasons is that an employer has to face business risks and financial constraints. If the self-employed employer (e.g., the owner of a small private firm) cannot settle the financial constraint problem through the formal financial market (e.g., by getting a loan from a government bank), business continuity will become difficult. Financial constraint problems already exist in China. It is known that the public banks do not like to lend to small private firms, so most small firms gain financial support through informal financial markets (e.g. inter-household risk sharing and illegal loans). The estimated results in this paper showed that the effect of household income on the selfemployed employer group is greater than that for the other groups. In order to promote more new business for greater economic growth in the future, the Chinese government should establish and implement financial support policies for small firms.
Finally, although we conducted an empirical study to reveal the determinants of self-employment and used the hypothesis tests discussed in this paper, there are two points worthy of attention. First, because we utilized one period of cross-sectional data, there might be heterogeneity and endogeneity problems, and a study using panel data should be conducted in the future. Second, this paper is a static analysis for self-employment. It is thought that empirical studies on dynamic changes in self-employment (the transition into and exit from selfemployment) are also important issues (Le, 1999) 12 , so dynamic analysis should be taken in the future studies.