Informal Self-Employment in Kazakhstan

We use data from the Kazakhstan Labor Force Survey (KLFS) for the period 2006-2011 to examine causal factors that determine informal self-employment. In addition, the paper expands the analysis to consider gender differences in informality and examines the response of informality propensities to the recent global crisis. Our decomposition analysis shows that education, work hours and tenure in self-employment are particularly important determinants of business formalisation.


INTRODUCTION
Research has long focused on the informal sector, and it is now well understood in economics that this sector affects macroeconomic stability (Fiess et al. 2010), poverty alleviation (World Bank, 2013) and even job satisfaction (Perry et al. 2007). In recent years, research has shown a great deal of interest in women's informal employment in developing countries that arises from the reliance of millions of women on the informal sector as a major source of employment and income (ILO, 2012) and from its potential impact on women's economic empowerment, gender equality and family well-being (Kantor, 2001;Chen, 2012).
International evidence has consistently revealed gender differences in informality rates, observed in both developed and developing countries, and interest in explaining these differences has increased. The International Labour Organization (ILO) argues that the informal sector is generally 'a larger source of employment for women than for men in the developing world' (ILO, 2002, p. 7). Hazans (2011) reported that female workers are more likely to be informal in Western Europe and parts of Eastern Europe, as well as in the southern peripheries of the European Union. More recently, the World Bank's World Development Report asserted that women worldwide are overrepresented in the informal sector (World Bank, 2012, p. 200). 1 Although many empirical studies have examined informal employment in transitional economies (see, for example, Rutkowski, 2006;Lehmann and Pignatti, 2007;Lehmann et al. 2012), relatively little is known about informal self-employment in these countries, especially in relation to gender.
One such interesting case is Kazakhstan, where forced self-employment observed throughout the 1990s has persisted well into the 2000s and around half of all workers in the informal sector are self-employed, mainly as own-account workers (Verme, 2001;Allen et al. 2007;Rutkowski, 2011). There was also a sharp rise in the number of own-account workers during the crisis of the late 2000s 2 , suggesting that self-employment correlates positively with economic downturns, and those working in the informal sector in Kazakhstan, mostly women, were the most vulnerable to a descent into poverty (Gavrilovic et al. 2009).
In this study, our objective is to identify causal factors that shape the informality decision amongst the self-employed workers. It is also important to evaluate the extent of informality changes in recessionary times. We perform this analysis by estimating the choice between formal and informal self-employment using data from the Kazakhstan Labor Force Survey (KLFS) for 2006, 2009 and 2011. Our decomposition analysis provides a nuanced perspective on what matters for the transition to formalisation. The paper proceeds as follows. In Section 2, we provide descriptive analysis of issues related to the persistence of informal self-employment.
Section 3 introduces the methodology. Section 4 describes the data set and defines the variables used in the analysis. We present empirical results in Section 5 and Section 6 concludes.

Macro-economy and self-employment
As can be seen in Table 1, real GDP growth averaged 10% during the 2001-2007 commodity boom, but fell sharply through the period of global crisis beginning in 2008. Unemployment rates fell throughout the recovery period despite the recession, and non-agricultural selfemployment rose during 2001-2011. The rise in female self-employment rates was moderate, but the gender gap in self-employment had widened during the 2000s. Table 1 also reveals that the share of registered businesses decreased during the crisis and the negative trend for females continued when growth resumed in 2010 (col. 5), but then rose sharply in the following year. 3 It is also useful to compare Kazakhstan's self-employment rates with those of other Asian economies, where the share of self-employed workers often exceeds half of the workforce (e.g. Indonesia and Thailand). It appears, as Figure 1 illustrates, that the Kazakh experience is not markedly different: firstly, we find that more than half the self-employed women work in subsistence farming, as in many developing countries; secondly, we observe that the level of subsistence farming in Kazakhstan has a negative relationship with per capita income, but the gender gap continues to persist; and finally, we also find that self-employed men tend to work on their own account. Overall, own-account self-employment among men has moderately expanded, whilst the female trend initially lagged, remained constant in the late 2000s but fell slightly in 2011. Evidence also reveals that women in Kazakhstan are forced into self-employment, as in many regions of the developing world (e.g. IFC, 2011, p. 43). According to Allen et al. (2007, p. 15), roughly 40% of women start a business out of necessity. Duban (2012) also reported that forced self-employment more often explains women's self-employment in Kazakhstan, raising concerns about a causal link between self-employment and the inclination towards informality.

Informality
Numerous common criteria define informal employment and, at the practical level, the results vary by definition (Henley et al. 2009;Kanbur, 2011). However, for our purpose, informality is measured as lack of compliance (i.e. registration) among the non-agricultural own-account workers. These self-employed workers may work alone (e.g. self-employed professionals or labourers) or hire occasional employees (e.g. owners of microenterprises), including family workers, outside the purview of the state regulation.
We now present the percentage of male and female own-account workers operating informally. Table 2 reports a negative trend in informality rates during the recent financial crisis, after which they remained below the pre-crisis levels. The disaggregated data also exhibits little 6 variation between male and female informality rates, except for 2011. The data also reveals variations in the characteristics of men and women with respect to education in that year. In the pre-crisis period, roughly 66% (61%) of men (women) in informal self-employment reported vocational education (or less) as opposed to 46% (47%) of men (women) in formal selfemployment. Overall, we find differing dynamics in the distribution of informality at technical and degree level of qualifications. That is, over the period 2006-2011, the percentage of women with technical education in informal self-employment rose (28% to 37%) whilst it was relatively stable in the formal sector. 4 In explaining the patterns of informal self-employment, we also need to examine ethnic variations. As

Government policy
In December 2012, the government announced its intention to double the SME sector's contribution to GDP by 2030, embedded in the 'Kazakhstan 2050' long-term development strategy. 5 However, we argue that for the initiative to be effective, the government must implement policies with a major impact on informality levels, such as removing exclusionary factors related to the institutional environment, preventing entry into the formal sector. 6 The government can also achieve effective reduction of informality by addressing internal heterogeneity of the informal sector, and therefore designing a broader set of measures that extend beyond labour market rules and regulations. 5 The number of small and medium-sized enterprises (SMEs) rose sharply during the recovery period, but their economic importance, measured as contribution to GDP, fell from 17.8% to 17.5% over the period (KAS, 2013. We must also note that the economically active SMEs as a percentage of registered businesses averaged less than 70% between 2007 and 2009 and fell to 53% in 2010 (KAS, 2012, Small and Medium Enterprises in the Republic of Kazakhstan, 2007. 6 See, for example, Perry et al. (2007) for the discussion of exclusionary factors and motivations for participation in informal self-employment.

8
A number of factors can reduce or discourage entry into formal self-employment. Access to finance may act as a barrier overall, but especially in meeting the costs of formal sector registration. Rutkowksi (2011) found that insufficient access to credit markets was of particular concern to small firms in Kazakhstan, similar to what was reported for many developing countries in Central and East Asia, where women work largely in small and low-growth firms and thus find themselves prevented by financial constraints from exploiting growth opportunities (IFC, 2011, Figure 1.5). In addition, Duban (2012) found that female shuttle traders in Kazakhstan encounter borrowing constraints that impede their entry into formal selfemployment.
Against this background, however, we find no evidence suggesting that formal SMEs, especially those woman-owned enterprises, have greater access to SME assistance programmes, delivered by the Damu Fund (Entrepreneurship Development Fund). Table 4 shows that targeted loans totalled US$ 3.93bn in the two years between 2009 and 2011, but men were the primary loan recipients, and self-employed women were excluded from leasing finance and access to the regional funding systems. In any of the financial support programmes with women borrowers, the percentage of women loan recipients does not exceed 30%, except for the women-only lending programme (row 4). We must also emphasise that these loans covered only 547 women who borrowed, on average, around US$ 30000. 7 At the start-up level, Gurevich (2010)  The general conclusion that emerges from this section is that targeted policies are needed within government's strategy to escape the middle-income trap and expand the SME sector. For example, we find that government interventions in credit markets were not broad-based (e.g. favouring large firms), making it difficult for even registered businesses to obtain loans, and that there is great urgency for financing strategies for microenterprises. 8 A second important finding is that minority women have markedly higher informality rates, especially during a recession.
Thirdly, women from all ethnic groups may be further disadvantaged by non-financial barriers.
Therefore, we argue that cultural environment impinging on women in different ethnic groups represents another area of concern that government has not yet addressed. In particular, informal institutions often shape gendered norms and values, and these gender-specific perceptions strongly affect entrepreneurship decisions (Terrell and Troilo, 2010). In Kazakhstan, multiple barriers associated with the resurgence of traditional and religious customs and social norms (e.g. restrictions on mobility and domestic responsibilities) might play a part in imposing restrictions on women's self-employment activities outside of the informal sector.
8 According to KAS (Men and Women in Kazakhstan 2013, Table 8.6), this trend persisted over the period 2010-2012. In addition, Kalyuzhnova and Nygaard (2011) cast doubt on the ability of the government to identify the 'right' places to direct resources from the oil and gas revenues through the state financial vehicles.

METHODOLOGY
In the second half of the 20th century, economists developed a number of theoretical models that may serve in analysing worker participation in informal employment. These models can be distinguished by the underlying causes to which they attribute reasons for selecting informal sector employment. The traditional view argues that workers enter the informal sector because they do not have alternative sources of income (e.g. Fields, 1975). This is because "unemployment in the city is a distinct possibility" for workers excluded from formal sector employment (Fields, 1990, p. 50). This view effectively considers the informal sector a stepping stone (the "staging area hypothesis"), which rural migrants enter to earn income to finance their job search in the formal sector. Another strand challenges the segmentation view (Maloney, 1999) arguing that majority of workers in the informal sector have voluntarily chosen that sector and that the traditional dualistic view can become 'more relevant in the presence of deep recession and large labor distortions' (Maloney, 2004(Maloney, , p. 1173. In seeking to explain the process of sector choice, it is natural to follow Hart's (1972Hart's ( , 1973 proposition that the informal (undocumented) sector is not intrinsically bad. Relying on the rational choice argument we can assume that workers may freely choose informal activity and that the decision to become an informal worker depends on the risk-adjusted relative awards.
Indeed, informality in self-employment offers benefits -such as tax evasion -as well as the measurable uncertainty associated with the risk of detection. Therefore, a decision on whether or not to engage in informal self-employment activities can be seen as the outcome of random utility maximization based on the individual's perception of whether the utility stream from unregistered self-employment exceeds that of a legally registered activity. 9 For example, Arabsheibani and Staneva (2012) have shown that in Tajikistan the average post-tax earnings in the informal sector are higher than in the formal sector. Additionally, Gimpelson and Kapeliushnikov (2013) have shown that informality carries an earnings premium among Russia's self-employed workers. This situation obviously creates incentives to choose informal selfemployment. Overall, we can treat the utility function as a black box, but in principle it reflects expected benefits (e.g. pension contribution evasion) as well as expected costs (e.g. maintained book of accounts). We also assume that switching between the two (legal and illegal) states is 9 It is, of course, possible that some workers may be displaced involuntary.

11
costly and that individuals demonstrate heterogeneous aversion to risk. For example, more riskaverse (e.g. more educated) individuals operate registered businesses.
For empirical specification, we approximate the utility function with the following 'latent' Eqn. (1): X is a set of explanatory variables,  is the corresponding vector of coefficients is the error term.
Since we do not observe * i I , the decision of whether to register is indicated by the binary . However, if selection into self-employment is not random, then the relationship between the self-employment decision (selection equation) and informality (outcome) can be formed through observable and unobservable characteristics. And if these characteristics are correlated, this will generate an incorrect conclusion regarding the impact of the observable characteristics on the choice of informality. 10 Thus, we apply the equivalent of Heckman's of selection model (the bivariate probit model with sample selection) to correct for the possibility of sample selection bias. 11 The selection equation is specified as follows: where i S is the binary choice variable indicating the endogenous selection process that determines the decision to enter self-employment. i Z is a vector of the observed characteristics,  is the corresponding vector of coefficients and ) 1 , The decision to enter self-employment is indicated by: From Eqs. (1) and (2), it is clear that Z and X have a bivariate normal distribution with 12 zero means and correlation  ) 0 (   and that three types of observations exist, with the following probabilities: The model can be estimated by fitting maximum-likelihood probit models with sample selection. The correlation between the two residuals in the selection and outcome (informality) equations in the MLE estimation is not directly estimated. Instead, the inverse of the hyperbolic tangent is estimated as follows: Conditions of the Heckman double probit model require at least one variable to be included in i Z that does not also appear in i X . Identification restrictions are required to achieve efficiency, and therefore we need a variable that we think affects selection into the sector, but not the informality choice. However, few candidates usually exist for the inclusion of additional variables in i Z .
To decompose the predicted changes in informality, we use the non-linear decomposition technique proposed by Gomulka and Stern (1990) and Even and Macpherson (1993) for binary outcomes, in which counterfactual conditional expectations are computed and averaged across observations. The decomposition is expressed in terms of probabilities. Specifically, the univariate (marginal) predicted probability of success (I = 1) is estimated as the sum of probabilities in Eqs. (6) and (7) as follows: Using the second-stage probit coefficients, the average predicted probabilities of informality for an individual in group j ) , ( f m j  , male or female, and time t ) 1 , 0 (  t is expressed as: and the predicted change in informality rates between two periods (0 and 1) is then expressed as: Using the baseline structure for period 0 as the reference, we can decompose a change in the predicted informality rates into explained and unexplained portions of the gap as follows: where the change in endowments explains the difference in informality rates between the two periods in the explained component, attributed to the change in characteristics over time for a single group, whilst the unexplained component is caused by the change in the underlying structures determining informality between the two periods. 12 Given that the explained component is a sum over the individual contributions, the contribution to the explained component (Eqn. 12) made by the th r regressor is equal to: where the weighted contributions of the th r individual predictor is determined by the group difference in means and evaluated by using coefficient estimates from the Heckman probit model. Note that Jones (1983, p. 130) demonstrated that the unexplained portion of the gap "cannot be uniquely determined because the value for the difference in intercepts depends on measurement decisions." That is, the decomposition arbitrarily depends on the choice of the omitted group and the elements of the detailed decomposition must rely upon arbitrary normalisations (Fortin et al. 2010, pp. 40-42). Therefore, we do not attempt to estimate the separate contributions made by individual characteristics to the unexplained change in informality rates. 12 The decomposition cannot be computed by plugging in the estimated ˆ and the mean values of X, as in the standard Oaxaca-Blinder technique. Counterfactual conditional expectations must instead be computed and averaged across observations. See Fairlie (2005) and Jann (2008) for the detailed discussion of the decomposition method for nonlinear response models.

DATA
The KLFS is a longitudinal (rotating) household-based survey conducted quarterly with a sample size of 21,000 households, 75% of which is held over for the next wave, with the rest dropped and replaced by new households. In 2011, however, every household in the sample was replaced To focus on working age adults, we have restricted our sample to working people by excluding students, children under 16 years of age, pensioners and the unemployed. We also exclude other groups of self-employed workers: unpaid family members, subsistence farmers and members of worker co-operatives. The definition of informality relies on the enterprise-based criterion that considers own-account enterprises informal if they failed to register. We also 15 exclude own-account and paid workers engaged in agricultural activities. In the selection equation, paid workers and employers form the reference category (salaried workers).
Estimates that rely on a functional form for identification are usually unstable and stronger identification restrictions are required to achieve efficiency. Therefore, we use previous self-employment experience as an exclusion restriction to identify the model. We argue that this variable affects the self-employment choice but is not related to the probability of informality. In Appendix Tables A.1 and A.2, we describe the main variables derived from the survey and report the descriptive statistics for the variables used in our empirical analysis.

Selection estimates
In the first step of our analysis we measure the probability of sector choice (self-employment or salaried work) against the selected independent variables. The base outcome offers greater job security than self-employed work because of the regulating provisions and additional benefits. Table 5 reports the estimates. We find that education is negatively associated with selfemployment propensity in every year for both women and men. Several explanations may account for this phenomenon. Education may correlate with tastes for leisure and subsequently favour under-employment (e.g. government work). It could also lower the search costs for formal employment, relative to those for self-employment opportunities, by satisfying job requirements.
Another possible reason for the negative effect of education on self-employment is that education may correlate positively with risk aversion. Finally, this result could be explained by the employers setting strong criteria for applicants who apply for salaried work.
We also find that individuals with previous self-employment experience (attempted to start a business) have a higher tendency to choose the self-employed sector. Spatial variations in self-employment propensities suggest that such propensities are negative in urban areas (for men) and locations with higher-than-average levels of income (e.g. Almaty and Astana). These findings may be explained by the housing bubble, which had over-stimulated the economy in urban locations and, in turn, led rural-urban migrants to anticipate more permanent paid employment than was actually available. It is also plausible to assume that inputs (e.g. rent) are 16 more expensive in urban locations, as asserted by Parker (2004, pp. 99-102). The estimated effect of the exclusion restriction (Experience) is positive and significant.
Noticeable differences exist in the probability of entering self-employment for ethnic minorities. Compared to other women, the European and Kazakh women are much more likely to enter salaried employment than are other minorities, largely Uzbek and other central Asians.
This finding may relate to customs and religious beliefs ('pull' factors). The higher propensity to be self-employed for ethnic minorities may be also attributed to institutional factors and discrimination in the paid employment sector ('push' factors). reports that household characteristics and differences in locality explain women's participation in self-employment. Aidis et al. (2007) analysed survey data from the Ukraine collected in the summer of 2002. The authors conclude that gendered norms and values (e.g. the resurgence of patriarchy), as well as institutional deficiencies, restrict women's self-employment opportunities.
Also, non-pecuniary motivation such as flexibility (Burke et al. 2002), educational choices (Leoni and Falk, 2008), differences in human capital and labour market experience (Georgellis et al. 2005) often explain the determinants of women's self-employment decision.     Comparing the "uncorrected" regression results with the Heckman estimates, we find that the probit regression coefficients are different in statistical significance (Kazakh and Professional) and lower in magnitudes. The coefficients presented in Table 6, however, are not easily interpreted in regard to the effect of recession, and thus it is useful to examine the change in informality propensities over time (pre-and post-crisis). Table 7 reports the estimated marginal effects. We find that working fewer hours increases informality probabilities. This result may be explained by the relative cost of compliance for part-time workers, but the estimated marginal effect for the post-crisis period is lower for both women and men. That is, the informality profiles demonstrate an upward trend as age rises, suggesting that age effects on informality are stronger for older workers in the crisis period. 13 Qualified professionals are more likely to be formal than non-professionals, although the marginal effect falls from 12 percentage points in 2006 to 6 percentage points in 2009 for men. 14 13 We do not find a U-shaped relationship between informality and age. Estimates can be provided by the authors on request. 14 The university enrolment trend that has been recently observed might undermine the process of formalisation.   Notes: Marginal effects (at mean) were computed in Stata using the "margeff" command (Bartus, 2005). Robust standard errors are in parentheses. Significance levels: c p<0.05, b p<0.01, a p<0.001. Table 8 reports decomposition results of the change in the probability of informality before and after the crisis for each gender group. The estimates reveal a sharp decrease in the probability of informality, especially for women. We find that the gap in mean informality probabilities in the pre-and post-crisis periods decreases from 36 (33) percentage points to 20 (13) percentage points for men (women) between the two periods. Differences in characteristics explain roughly 25% (24%) of the change for men (women), whilst the unexplained decline accounts for a significant portion of the observed change between the two periods.

Decomposition estimates
Notes: Probabilities evaluated at the mean value of variables. The estimated standard errors of the predictions, based on Stata's delta method command, are in parentheses (row 1). The explained part of the predicted change in informality rates between 2006 (period 0, pre-crisis) and 2011 (period 1, post-crisis) is attributed to the change in informality that occurs only if the composition value (∆X) changes from period 0 to period 1.
Generally, our analysis of individual contributions indicates that a relatively high share of the explained change results largely from an increase in working hours and duration in selfemployment (survival) after the crisis. The contribution made by education in the explained

CONCLUSION
This paper analysed the determinants of entry into informal self-employment in the Kazakh labour market. The results offer an additional piece of evidence of gender differences in informality propensities in a transitional setting. Overall, our empirical findings indicate that work hours and years in current business affect the propensity to informal self-employment.
Moreover, the probability of informality also diminishes with education. Those independent workers representing minority ethnicity groups have a greater tendency to operate informally.
The decomposition reveals that the unexplained component explains much of the decline in informality. We assert that the large unexplained decline can result from changes in the shape of structural effects such as changes in taxation that make one sector relatively more attractive than the other, employers' hiring and firing behaviour, changes in labour regulation and a change in preferences (taste) for the informal sector. Another possible explanation for this finding is that the government introduced post-crisis reforms designed to ease the administrative burden on business. For example, the government reduced the number of licensed activities and eased registration procedures in the late 2000s (OECD, 2012).
In general, the analysis suggests that reducing informality may require coordinating policies rather than pursuing a few narrow, ostensibly distinct, policies. For example, we believe that the government's policy of increasing the size of the SME sector should comprise policy instruments that tackle informality among the self-employed. The finding that human capital makes a key contribution to explaining of the change in informality suggests that the government should target informality through skills training and education. These policy instruments are especially important if, in an economic slowdown, the ratio of educational cost to income increases so much for some groups that it may discourage low-income women in particular from attending college or university, or low-skilled workers from acquiring training that may later increase their probability of self-employment and informality.