본 논문은 기업가비율이 실업률, 소득수준과 어떠한 장기적 관계에 있는지, 그리고 어떠한 설명변수들이 기업가비율에 영향을 미치는지 분석한다. 이를 위해 1980년부터 2008년까지 OECD 23개 국가를 패널 분석한다. 잘못된 추정과 편의의 위험을 피하기 위해 분석대상 변수들의 단위근 검정을 통해 수준변수들이 비정상적인지를 검정하고, 비정상적 변수들 간에 공적분관계가 존재하는지를 검정한다. 소득수준, 개방도, 규제정도, 실업률, 금융접근성, 노조 조직률 그리고 제조업비율이 기업가비율과 장기균형관계에 있음을 확인하였다. 기업가비율의 추세는 국가별로 큰 편차를 보이기 때문에, 비슷한 패턴을 보이는 국가그룹으로 분류하여 DOLS추정하였다. 증가추세의 국가그룹에서 소득수준, 개방도, 실업률, 제조업비율, 금융접근성이 유의한 양의 영향을, 인적자본수준과 노조 조직률은 유의한 음의 영향을 주고 있다. 이러한 추정결과는 기업가비율이 감소패턴을 보이는 국가그룹과 상이하다. 감소패턴의 국가그룹에서는 노조 조직률을 제외한 나머지 모든 변수에서 상반된 추정 계수값을 보인다. 즉, 어떤 그룹에 속하는지에 따라 기업가비율에 영향을 미치는 요소의 영향의 방향이 달라진다. 기업가비율이 다른 설명변수들과 반대의 인과관계를 갖는지를 테스트하기 위해 오차수정모형(error correction model)을 추정하였다. 기업가비율은 소득수준 및 실업율과의 관계에서 양방향 인과관계가 성립하며, (기업가비율 ⇄ 소득수준), (기업가비율 ⇄ 실업률)의 충격반응모형을 통해, 충격에도 불구하고 장기균형수준에 수렴함을 보인다.
What kind of explanatory variables affect business ownership rate? How is the business ownership rate related to income per capita and what are the direction of causality between them?This paper examines the long-term relationship between business ownership rate and a number of variables for a panel of 23 OECD developed countries over the period from 1980 to 2008. To do this we first investigate the stationarity of variables concerned, then test cointegration among the variables by employing the Westerlund’s (2007) error correction panel method. Commonly used econometrics methods are the so-called first generation panel unit root tests, such as Hadri and Levin A, Lin CF, Chu(=LLC). Because these first-generation tests assume cross-sectional independence, exhibit severe size distortions in the presence of cross-sectional dependence, we also use second-generation panel unit root tests such as CIPS(=cross-sectionally augmented IPS) to allow for cross-sectional dependence. For all level variables, both Hadri and LLC test do not reject the null hypothesis of a unit root at from the 1 % to 5% level. With regard to the first differences of the series, the test statistics are greater than the 1% or 5% critical absolute value and thus the null hypothesis is rejected. The CIPS test results also offer identical conclusions,To test if the null hypothesis of no cointegration can be rejected, Westerlund (2007) has developed two group-mean tests and two analogous panel tests. In the two group-mean based tests, the alternative hypothesis is there is cointegration at least in one cross section unit, which is the same in many traditional panel cointegration tests. Results using two pairs (no intercept, no time trend), (intercept, time trend) and one-period lead and lag values indicate that all four tests reject the null of no cointegration at the 1% or 5% level. Finally we bootstrapped robust critical values for the test statistics. Overall, those results show that there exists a long run cointegrating relationship among the variables in equation.
Having confirmed the stationarity of variables and the existence of a long run relationship between the variables in equation, DOLS(Dynamic OLS) were utilized to estimate the respective parameters. Empirical results show that the sign of estimated parameter coefficients are different between country groups. While income per capita, economic openness, level of human capital, and credit accessibility significantly positive affect business ownership rate in riser country group, the signs of those coefficients in faller group are almost negative. This implies that the parameters of determinants are heterogenous between two groups of countries even though all of these countries are the OECD members. The above interpretation of the estimation results is based on the assumption that long-run causality runs from explanatory variables to business ownership rate. However, while cointegration implies causality in at least one direction, it says nothing about the direction of the causal relationship between the variables, as discussed above. Causality may run in either direction, from income per capita to business ownership rate or from business ownership rate to income per capita, or in both directions. To test the direction of long-run causality, we follow common practice in the applied panel cointegration methods and employ two-step procedure. And same procedure applies to causal relationship between unemployment rate and business ownership rate.
In the first step, we use the DOLS estimate of the long-run relationship to derive the disequilibrium term. In the second step, we estimate the error correction model. From this we conclude that long-run causality is bidirectional, implying that increasing business ownership rate is both a consequence and cause of increasing level of income per capital or decreasing unemployment rate. To check the robustness of this conclusion, we perform a panel Granger causality test based on a levels VAR regression with fixed effects. the null hypothesis of no Granger causality from business ownership rate to income per capita or unemployment rate is not rejected. This confirms our result that increasing business ownership rate leads to increasing income level or decreasing unemployment rate. We employ generalized impulse response functions based on a one or two-lag panel vector error correction model over a 28-year horizon. A one-standard-deviation shock in business ownership rate results in a gradual and permanent decrease in income level or unemployment rate and reaches its full impact after 15 years. And we also shows that income or unemployment rate gradually and permanently decreases in response to a one-standard-deviation shock in business ownership rate and that the full impact is reached after 15 years. The impulse response functions are thus consistent with the Granger causality tests.
The main contribution of this paper is that it is the first attempt to apply panel cointegration technique to examine the long-run relationship between business ownership rate and number of concerned variables such as income per capita, unemployment rate, economic openness, human capital level, union membership rate, credit accessibility and manufacturing sector ratio. And also we shed light on the direction of causality between income level, unemployment rate and business ownership rate.
While structural breaks in OECD countries such as collapse of Berlin wall, formation of Euro-zone, political regime shifts, financial or trade liberalization, financial or fiscal crisis and so on may affect the above panel test statistics, we do not take into account structural breaks, which may limit our conclusions and implications.