Social benefits

Размещено на http://www.allbest.ru/

Table of contents

1. Theory

1.1 Introduction

1.2 Literature Review

1.3 Hypothesis

2. Data and Variables

2.1 Data and Variables: Variables

2.2 Data and Variables: Data

3. Model

3.1 Model: Variables Testing

3.2 Model: Correlation Analysis

3.3 Model Specification

3.4 Model Specification: Developed Countries

4. Analysis

4.1 Limitations

4.2 Limitation check

Conclusion

References

Applications

Tests

Corrections

Models

Correlations

1. Theory

1.1 Theory: Introduction

This research paper examines the dependence between social benefits size and number of self-employed people. On the one hand relationship between social benefits size and number of self-employed people seems negative, as higher social benefits reduce the incentive to start a firm. Indeed, higher social security expressed in social benefits in cash and in kind, may create an unfavourable environment for taking risk and becoming self-employed. On the other hand, higher social benefits mean that recipients have more security against negative outcome of business development and have more means to maintain a decent life during the business development. This assumption seems reasonable as well, as the fear of failure and the absence of “a safety cushion” prevents many businessmen from starting up a firm.

Earlier researches mainly focused on exploring the relationship between unemployment benefits and unemployment rate, while there were only few papers studying the social benefits and business development relationship. Those few that did focus on the business topic proved negative, if any, dependence of business formation on social benefits. Moreover the authors never distinguished between social benefits in cash and in kind, which in my opinion may have different effects on self-emplyment. Finally, the previous studies used old data, selecting a small sample size of developed countries that is very likely to result in selection bias.

In this work I expect to question the consensus view of the negative effect of social benefits on business formation. I hope to combine and challenge the existing views on this aspect of government social policy and to achieve the opposite results to prove that social benefits in kind and in cash may deliver different effects on self employment. I expect to achieve positive dependence between social benefits and self-employment in cash or in kind terms.

I am developing a model that allows me to examine this interdependence, using panel data for 14 years for 32 countries.

1.2 Theory: Literature review

There are only a few studies of social benefits effect on business formation. This relationship should be considered in a framework of many other factors, like innovation, unemployment, financing availability etc. Therefore, we should first review the previous social benefits and business frameworks and examine possible explanations provided by early research works. I divided this literature review into three parts: first I examine the equilibrium unemployment framework, than I focus on the few studies, researching the relationship between social benefits and business formation and finally I examine works, proposing other factors significance for business formation.

Mortenson Pissarides Matching Model (1994), built on the Diamond (1971) model, is the cornerstone of any equilibrium labour market theory. The authors assume an infinite time horizon, constant returns to scale and risk neutral participants. They suggest that the number of workers hired depends on the number of workers looking for a job and the number of vacancies available. Thus the authors state that more of each input results in higher match rate between employers and employees. The main goal of Mortenson Pissarides model is examination of endogenous job creation and destruction in opposition to the cyclical unemployment framework. It is essential to note, that in this model production requires a combination of a job spot and a worker. Each unit of labour is productive and is compensated by a fixed wage, so that a firm hires until profit per worker converges to zero, i. e. the value of an empty working place is the same as the value of a filled one. The employees agree to take a job if the revenue they get during working is greater than the reservation wage. If unemployed, they receive unemployment benefits, transfers etc. Mortenson and Pissarides additionally input a job distraction argument, based on the Beveridge Curve, which reflects the dependence between the number of jobs created and number of jobs destructed. The curve has a negative slope, reflecting the negative relationship between them. Clearly, the allocation of workers to jobs is inefficient and workers share a surplus fraction created during the work with employees. The surplus allocation in the Matching Model is determined by bargaining, which is possible because o the sunk costs of hiring existence.

Mortenson and Pissarides conclude that in the context of their model, job creation effect moves in the opposite direction from job destruction. Moreover, job creation has a lower magnitude of changes than job destruction. And finally job destruction moves unevenly during economic booms and recessions: it increases faster during the recessions than it decreases during booms. In the context of my work social benefits increase acts like a recession reducing the labour supply and thus enforcing jobs destruction. Job destruction process may create a desire of unemployed to start their own business, given higher social security in case of failure and the need to fulfil a reservation wage requirement.

The reservation wage argument was a part of the first and most commonly used theory, search model. Search model is concerned with the unemployment benefits size being negatively related to the employment rates. This theory was developed by many economists, including (Steven Lippman and John McCall 1979). They propose that each worker picks wages sequentially once per period, aiming to maximize his utility through earnings and leisure. The best decision in this case is to accept any job offer, which proposes the wage higher than the reservation wage. Overall, the essence of the search theory in the context of unemployment benefits size is: the increase in unemployment benefits results in lower opportunity cost of leisure and thus lower employment in the economy. This conclusion applies both to self-employment and usual employment.

More specifically, it addresses the duration of job search, fluctuations in wage rate and the type of employee utility function. Unemployment insurance in unemployment benefits may be raised in order to increase participation rate, but in the “standard” model framework it is assumed to have only short term effects on the length of search unemployment. Further, it affects the labor demand side: higher unemployment benefits stimulate higher wage rates, which results in higher production costs for firms. Higher production costs encourage lower labor demand, reducing employment. Finally, the workers' utility functions depend on leisure and income and thus higher unemployment benefits may heavily reduce the incentive to give up leisure by increasing its opportunity costs. Overall the model predicts that the reservation wage (one for which a person chooses to become employed) increases when the unemployment benefits size increases, reducing the number of people willing to be employed.

“Standard” approach however faced a severe opposition from a number of economists. For example, the standard model assumptions include that once accepted the job will last forever and thus the unemployment benefit is an input in the decision function of whether to accept the job. Empirically, however, it contradicts the core purpose of unemployment benefits increase - the insurance against becoming unemployed according to Hey and Mavromaras (1981). Moreover, the standard model was called “partial-partial” by Rothschild (1973), who claimed that the assumption of standard model about wages being exogenously given is unrealistic.

To draw an intermediate conclusion, there were two powerful unemployment benefits theories, which had different explanations but the same result: unemployment benefits increase reduces employment. For my analysis I will focus on equilibrium theory of Mortenson and Pissarides, as I aim to explore the equilibrium relationship between social benefits and self-employment in equilibrium. Steven Lippman and John McCall (1979) standard search mode will, however, be used to challenge the Mortenson and Pissarides conclusions and to show different views on the subject.

The few works, exploring the relationship between social security and entrepreneurship activities have proven negative relationship between them. For example, Parker and Robson (2004) focused on business development as on a type of employment rather than a specific activity. They tested twelve OECD countries data for late 20th century, hypnotizing that there is a negative effect of higher social benefits on self-employment. The authors differentiated between different circumstances of starting a business and various perceptions of social benefits effects. First of all, a necessity motive may create a powerful incentive to earn one's leaving. To fight the scarcity of food and shelter one may choose to join the labour force rather than to start a risky business (become self-employed). If true, this hypothesis proves that higher social benefits reduce the incentive to become self-employed and thus have a negative correlation with entrepreneurship. Secondly, higher unemployment benefits increase unemployment, decreasing the purchasing power of individuals. This is likely to cause a lack of demand for products of almost any business, which implies negative correlation between social benefits and entrepreneurship. This perception is unique because the unemployment is viewed to have effect on the demand side rather than on supply side. Usually workers are perceived as a means of production, not as consumers.

Parker and Robson examined the effects of payroll tax, usual tax and social benefits on business formation among countries. The results obtained by this study concluded that there is a strong negative relationship between social benefits and self-employment, while there is a strong positive relationship between tax rate and self-employment. The first outcome is doubtable, as the explanations provided by Parker and Robson above may be challenged. Intuitively the second outcome is not unexpected, as lower taxes leave higher proportion of profits to business owners. The research result supported an idea that government policy has a great influence on entrepreneurship activity: “because tax and benefit variables are under direct government control, governments may have considerable influence on the extent of self-employment within their economies.” Simon C. Parker, Martin T. Robson, 2004, “Explaining International Variations in Self-Employment: Evidence from a Panel of OECD Countries”, Southern Economics Journal, 71 (2). Pp. 287-301, Page 298 This work, however, has a number of drawbacks -- it takes a small sample size of only 12 developed countries and accounts for a short 20^th century time period, which may be inapplicable to present time.

Anyway, Parker and Robson model is the one I will rely on the most in my research, as the conclusions that he makes are unexpected and new. In my work I will try to account for the effect of the corporate tax rate and social benefits. I expect to achieve different results though, as I am using additional factors: crisis, perceived opportunities, cost of borrowing etc. And account for social benefits in kind and in cash separately.

Hessels et al (2007) conducted a more detailed analysis on the dependence between self-emloyment and the level of social benefits. The authors tested the early stage entrepreneurship dependence on social security, using OECD data. This article, like Parker and Robson (2004), differentiates between the social entitlements (taxes) of the employer and the social receivables (social benefits) of the employee. The authors proved that the former (social entitlements) has large negative results on entrepreneurial activity. The letter (social benefits) was considered negligible with respect to participation in self-employment (establishing a business). According to their article, “being aware of these limitations, one possible conclusion that could be drawn from our empirical results is that social security benefits do not necessarily (or only) influence the choice between entrepreneurship and wage-employment, but rather (or also) the decision to participate on the labour market in general.” Jolanda Hessels, Andrй van Stel, Peter Brouwer, Sander Wennekers, 2007, “Social security arrangements and early-stage entrepreneurial activity”, Scientific Analysis of Entrepreneurship and SMEs, page 26 The paper thus indicates no proof of social benefits support to self-employment, but finds strong negative relationship between the tax rate and business formation.

Hypothesis 2:

To sum up, I expect social benefits in kind and in cash to have a positive effect on self-employment. At the same time I suppose that social benefits in kind have a stronger positive effect on self-employment than social benefits in cash.

2. Data and Variables

2.1 Data and Variables: Variables

To analyze the social benefits effect on business development I used four data sources: OECD statistics, GEM (Global Entrepreneurship Monitor) data, World Statistics, Trading Economics. I used panel data for 32 countries from 2000 to 2013. I determined the following variables: unemployment - unemployment as a % of population, self-employment - as a % of population, social benefits in cash and social benefits in kind - both as a % of GDP per capita, ST rates - represent short term interest rates, corporate tax rate in %, research and development - as a % of GDP, low-income - and a dummy variable: crisis - 2008-2009 crisis.

I have chosen panel data for the analysis as it provides a larger data set. Moreover the dynamics of each variable is explained better and the output has a higher quality.

1) Self-employment is a dependent variable, which represents the % of population, running a business. This variable was rarely analyzed in the context of social benefits effect. However, there is a relationship that is essential to study in order to determine the government policy effect on business formation (self-employment). Indeed, in regression - “selfempl”. Source: OECD Statistics

2) Social benefits in cash and in kind are the main explanatory variables, represented as a % of GDP per capita. Having assessed a number of works on this topic I still suppose that these variables will prove to have a positive effect on self-employment, acting as a “safety cushion” against a business failure. In regression - “socbeninkind” and “socbenincash”. Source: OECD Statistics

3) Unemployment variable, presented as a % of population is considered, as many works, including Morris Altman (2004), Mortenson Pissarides (1994) and many others researched the relationship between the social benefits amount and unemployment. Thus, if we don't test the significance of unemployment, social benefits coefficients may reflect a partial effect of unemployment. Moreover intuitively before building a business people are usually considered unemployed for a while. Therefore I expect positive relationship between the two variables, if any. In regression - “unemp”. Source: OECD Statistics

4) Short term interest rates are included in the regression, as a proxy for financing costs, reflected by the base rate. Most of entrepreneurs require debt financing, which is determined by the base rate plus a premium. According to Kerr and Nanda (2009), financing is one of the main constraints to building a business. Therefore I expect negative relationship between the interest rate size and self-employment. In regression - “strate”. Source: World statistics, Trading economics

5) Research and development spending force innovation, demanded by almost any business. Indeed, Rampini (2004) and Koellinger and Thurik (2009) have studied innovation as an important factor of business development and have proven that it stimulates progress and makes business activity more effective. Thus I expect to see positive relationship between self-employment and research and development. The data I use for this variable expends only to 2012. In regression - “randd”. Source: OECD Statistics

6) Higher education is another important factor, considered by many previous business development works, for example by Wadhwa, Freeman and Rissing (2008). Intuitively, higher education delivers better opportunities for starting a business due to larger knowledge, wider network and more experience. Consequently I suppose that this variable will deliver a positive effect on self-employment. I will use data as a % for people aged between 24 and 60 for 2005 to 2013. In regression - “Education”. Source: World Statistics

7) Corporate taxes have been examined along with social benefits in both Parker and Robson (2004) and Hessels and authors (2007) and proved to be significant determinants of self-employment. Low taxes allow higher reinvestment rates and better opportunities to develop a business. Therefore I expect negative relationship between the tax rate and business development. In regression - “corptax”. Source: Trading Economics

8) Low income is the new variable I propose. It is calculated as wage level divided the lowest pay. The variable reflects the percentage of workers, earning less than 2/3 of median earnings within a country. The effect of this variable is likely to be negative on self-employment in less developed countries, as it is more difficult to obtain financing there and there are such constraints as corrupting, government monopoly etc. However, I expect this variable to have a positive effect on self-employment in my analysis, as I use OECD countries for analysis, most of which don't face problems listed. Therefore low income will create a higher motivation for workers to leave their low-pay jobs for starting a business. In other words, opportunity cost of failure is lower. In regression - “lowincome”. Source: OECD Statistics

9) Perceived opportunities are used to account for the level of country democratic development and the conditions for self-employment. Countries have different internal constraints, like corruption, difficult legal processes, monopolistic market etc. To differentiate countries I add this variable as the results of a survey, which asked citizens to estimate facilities for starting a business in their home country. In regression - “opps”. Source: OECD Statistics

10) I also account for crisis, using a dummy variable.

The summary of all variables is represented below:

Table 1

2.2 Data and variables: Data

There are three types of panel data analysis to apply: pooled regression, fixed and random effects regressions. Pooled regression assumes similar behaviour of all the countries and any endogeneity and deviation is excluded from the analysis. Fixed effects model solves the endogeneity of each country problem by selecting a separate intersection for each of the countries, invariant in time. Random effect model provides users with more statistically significant estimates and has fewer bounds than pooled one.

Let us first estimate the each of the regressions, determine the appropriate one and then test for heteroscedasticity, autocorrelation and stationary processes.

The data summary is provided in Table 2.

Table 2

* the data has a different time horizon

3. Model

3.1 Model: Variables testing

In order to test my hypothesis about the positive relationship between self-employment and social benefits, I analyzed the data described above. My analysis consisted of testing for stationary process, autocorrelation and heteroscedasticity - main problems that may cause any bias in the work - and comparing fixed-, random- and pooled effects for the different regression specifications.

First of all I have to test variables for stationary process. It occurs when the panel series estimates' variance and mean are independent of time. Otherwise, this dependence will be a part of explanation of the regression results, which will bias the actual explanatory variables effects on the dependent variable. To test for stationary I will use Dickey Fuller test in Stata. It is important to notice that I use percentages in the right and left sides and therefore even small changes with respect to time will be counted as an indicator of non-stationary process. Moreover I use a relatively small time period - 14 years, which may detect non-stationary process caused by the market microstructure.

The variables I will test are: self-employment, unemployment, education and perceived opportunities. I exclude social benefits in cash and in kind, low income and research and development from testing, as intuitively it is impossible for them to grow or decrease infinitely in time and they are expressed as a share of GDP, which is bounded by 100%. I also don't test short-term interest rates, as they depend on business cycle and government policy rather than on time. Finally, I don't test crisis as it is a dummy.

Self-employment variable proved to be non-stationary (there are small percentage points increases in value) - Test 1. Dickey Fuller test indicates strong difference non-stationary process with p-value of 47%. I correct it by differencing and get p-value of zero, indicating stationary process (Correction 1).

The test indicates that unemployment is likely to be non-stationary (Test 2) as it has p-value of 14%. Intuitively unemployment cannot increase forever and thus such a p-value may be applicable to my analysis, as empirically it is hardly possible.

Other variables are highly unlikely to reflect non-stationary processes signs and they don't: education has p-value 0% (Test 3), perceived opportunities have a 0% p-value as well (Test 4).

Overall only self-employment variable is non-stationary, which is corrected by first order differencing. There is no sign of cointegration, because only one variable is non-stationary.

To test for autocorrelation I estimate the basic model with self-employment first difference as a dependent variable and all the others as explanatory ones (Model 1).

I used panel data for my analysis and thus I should estimate and compare three models: fixed effects, random effects and pooled model. Pooled model assumes that there is no heterogeneity between countries and treats them as the same, different only in the variables listed. Fixed effects model assumes that each country in a panel has its own intercept coefficient, constant in time. This allows the model to reflect limited heterogeneity, independent of time. Random effects model assumes that the intercept has zero correlation with repressors and thus it may be made a part of the disturbance term, which will not be correlated with the actual regression coefficients.

In the first regression I estimated all the variables I had using the three models.

The pooled model (Model 1) indicates that social benefits in kind are significant both on 5% and on 1% significance level and are positively related to self-employment. However, social benefits in cash are strongly insignificant. I suppose that these results may be caused by incorrect specification and multicollinearity, as I included all the variables in the regression.

Fixed effects model (Model 2) results in significant coefficients of social benefits in cash and social benefits in kind. This model thus seems to be a better fit than the previous one. This conclusion is supported by p-value of F-test less than 0%. Random effects model (Model 3) has only two significant coefficients: social benefits in kind and unemployment.

To get unbiased estimates, I had to check for heteroscedasticity and autocorrelation. I used LR test with null hypothesis of homoscedasticity to detect heteroscedasticity. I got (Test 5) a p-value of 0%, reflecting heteroscedasticity presence. I use Wooldridge test to check whether autocorrelation is present (Test 6). The test shows a p-value of 78.6%, which indicates no autocorrelation. To correct heteroscedasticity I applied robust standard errors.

The corrected pooled model reflects significant social benefits in kind, corporate taxes and low income (Correction 2). Fixed effects model indicates no significance in social benefits (Correction 3), as well as random effects model (Correction 4). These results are preliminary and are likely to be insignificant because of the wrong specification and multicolliniarity. However examining the relationship between them in such a state is crucial for further model specification corrections.

To find out, which model is more applicable to this situation, I compare fixed, random effects and pooled regression. Fixed effects model has an F-statistics p-value of 0%, which means that corrected pooled model of this specification in worse than corrected fixed effects model. Random effects model proves to be more effective than pooled effects model, as Breusch and Pagan Lagrangian multiplier test has a p-value of 1.8% (Test 7). Hausman test for the difference between fixed and random effects model. It reflects a p-value of 3.4%, which means that fixed effects model is better for the initial specification.

Overall I cannot draw any conclusions yet, as the model is not correctly specified. In current state fixed effect model is the best one and reflects positive relationship between social benefits in cash and in kind and self-employment, which approves my hypothesis.

3.2 Model: Correlations analysis

To achieve reasonable results I have to specify the model correctly. To do that I will first estimate the correlations between the coefficients to detect multicollinearity. Secondly I will examine the relationship between the dependent variable and each explanatory variable and its three lags. After that I will estimate a model and get rid of the insignificant coefficients and estimate the final version to apply. social benefit business availability

The correlation matrix (Correlation 1) indicates that self-employment has the highest positive dependence on social benefits in kind and the lowest negative - on crisis. Short term interest rates, corporate tax rate, unemployment and crisis are negatively related to self-employment, which is intuitively correct. Higher short term rates increase the cost of financing a business Kerr and Nanda (2009), higher corporate taxes reduce the expected value of starting a business Parker and Robson (2004), higher unemployment reduces self-employment Jolanda Hessels at el (2008) and it is difficult to do a business during crisis, due to lower demand etc. To detect multicollinearity I examine correlations between explanatory variables. There is a 46% correlation between social benefits in cash and short term interest rates and 40% between social benefits social benefits in kind. This is likely to cause insignificance of social benefits in cash in the previous model. 46% does not reflect definite multicollinearity, but arises doubts about the need to put short-term interest rates in the model. Another variable that is likely to cause bias is unemployment, which has a 40% correlation with social benefits in cash. Moreover, social benefits in cash and in kind have a 52% correlation, which means that multicollinearity is present. That is why I suppose that one way to solve it is to estimate two regressions separately for self-employment dependence on social benefits in cash and in kind. Finally, research and development and education have 54% correlation, which means that only one of them should be present in the regression to reflect reasonable results.

The next step of my correlation analysis conducted to estimate a correct specification is checking the lags of each variable to determine the best fit for the model.

1) Social benefits in cash (Correlation 2). The lagged correlation analysis shows that the first lag is a better fit for the model than the second one, as the correlation between it and the self-employment is higher. Intuitively this specification can be accepted as the business formation is a long-term process. Once a potential businessman has observed the change in social benefits, he is likely to spend some time on finding a team, arranging the legal structure, obtaining financing etc. Indeed, it is more relevant for the model to use the first lag of social benefits in cash.

2) Social benefits in kind (Correlation 3). Unlike social benefits in cash, social benefits in kind have almost the same correlation with self-employment as the lag of social benefits in kind. Therefore I can use either the present social benefits in kind value or the lagged. The smaller difference between the present and lagged value of social benefits in kind than in social benefits is not surprising.

3) Short-term interest rates (Correlation 4). Short-term interest rates have a double effect on self-employment: on the one hand higher rates increase the cost of business financing and imply negative correlation, but on the other hand higher interest rates imply that the economy is in a good state (government usually decreases interest rates during crisis) which implies positive influence on self-employment. The correlation matrix indicates that present value and first lag value have a negative effect on self-employment and thus financing costs are more significant than the effect of the economic cycle. The sign, however changes in the second lag to positive, as financing costs two year ago have very low, if any, effect on self-employment. Thus for my regression I will choose the first lag of short term interest rates, as the businessman firs observes the financing costs and then makes the decision and arrangement, so it is intuitively correct to assume that it takes him a year.

4) Unemployment (Correlation 5). Unemployment variable has a very unusual correlation matrix: in the present year it is negatively correlated with self-employment, while in the consequent ones - positive. Moreover, the present value and the first lag of unemployment have very low correlation with self-employment, while the second lag is almost twice more significant for self-employment determination.

5) Education (Correlation 6). To check the self-employment dependence on education I take four lags, as education may need more time to adjust and reflect on the business activity. The correlation matrix indicates that all lags have almost the same correlation with self-employment, increasing in time insignificantly up to the third lag. The third lag thus is the best fit, but the other ones may also be used. Intuitively it is reasonable to assume that most businessmen choose to work after getting higher education to get some real life experience and to save some money to start up a business. That is why the third lag use is justified.

6) Research and development (Correlation 7). Research and development do not normally deliver progress in the first year of investment, as each innovation should be realized over a certain time period to reach the public. Therefore the highest correlation between the fifth lag and the self-employment is justified. Indeed, importance of innovation for business development, outlined by Koellinger and Thurik (2009) is supported by my data. When a government invests in research and development, the product first is not available to the public. However in 5 years it increases the volume of business formation, as high technologies leave room for more start-ups.

7) Corporate tax rate (Correlation 8). Correlation between self-employment and corporate tax rate is small negative. There is no significant difference in the correlation of lags and self-employment, therefore we can use either the present value or the first lag. Overall I don't think that this variable will prove to be significant

8) Perceived opportunities (Correlation 9). Perceived opportunities have the highest correlation with self-employment without lags. This is intuitively correct, because the perception of ability to start a business arises shortly before starting a business.

9) Low income (Correlation 10). The second lag is negative and the most significant. Indeed, the higher is inequality, the lower is self-employment.

10) Crisis. Crisis is a dummy variable and should not be lagged intuitively, so we don't consider its correlation matrix.

Overall, having examined the correlation between variables and their lags, I did not find any unexpected results. I will take the first lag of social benefits in cash, first lag of short-term interest rates, second lag of unemployment, third lag of education, fifth lag of research and development, second lag of low income, and present values of social benefits in kind, corporate tax rate, perceived opportunities and crisis. After this detailed correlation analysis I estimate another correlation matrix with the best lags chosen (Correlation 11).

The new correlation matrix significantly improves the initial one: social benefits in kind now have only 28% correlation with short term interest rates, in cash - 29%, social benefits in cash and unemployment have 25% correlation. However, social benefits in cash and in kind are still very interdependent (49% correlation) and education and research and development now have 60% correlation, which is a definite sign of multicollinearity. Keeping all these results in mind, it is now possible to estimate a better model specification.