Over the past years, numerous empirical studies have analyzed how economic freedom affects levels and growth rates of GDP per capita (for a survey, see De Haan, Lundstrom, and Sturm 2006). In most cases, they found that economic freedom, in particular an increase in economic freedom, has a significant positive effect on both. However, so far it has not been analyzed whether and to what extent economic freedom affects unemployment. This paper is the first to tackle this issue. Section 1 describes the data used in this paper. Sections 2 and 3 present and discuss the regression results. Section 4 concludes.
This paper applies the Economic Freedom of the World (EFW) index that has been developed by a group of North American economists under the auspices of the Canadian Fraser Institute with the aid of a worldwide network of further economists and institutes (Gwartney and Lawson 2005). The index has often been assessed to be a good measure of economic freedom and thus has been widely used in the empirical literature. It subdivides economic freedom into five areas: (i) size of government, (ii) legal structure and security of property rights, (iii) access to sound money, (iv) freedom to trade internationally, and (v) regulation of credit, labor, and business (for variable definitions and sources, see Appendix A). The two main authors of the index have explained their concept and measurement of economic freedom in each report (e.g., Gwartney and Lawson 2005) and elsewhere (e.g., Gwartney and Lawson 2003).
Economic freedom is complex and multidimensional. A major advantage of the EFW index is that it reflects all aspects of economic freedom and adequately groups them into five different areas. To provide information on the overall linkage between economic freedom and unemployment, we use the EFW summary index. Additionally, we use the five area indices. This allows us to analyze which economic freedoms have the strongest effects. As the correlation matrix indicates, there is substantial correlation among several of the area variables (Appendix B). Therefore, we estimate specifications that include these measures one at a time. However, the correlation matrix also indicates that the area variables are not highly correlated with each other. Therefore, we also present specifications in which we include them jointly. Estimating the effects of the area variables both separately and jointly allows us to check the robustness of our results for these variables.
The EFW index not only has the advantage that it covers all aspects of economic freedom and adequately groups them into five different areas. It also has the advantage that it is based largely on objective data. These data and the ratings derived on the basis thereof may be verified by anybody. Aspects of economic freedom that cannot be measured objectively (mainly in the areas legal structure and property rights and regulations) are measured using subjective data from the Global Competitiveness Reports of the World Economic Forum and the International Country Risk Guides of the PRS Group. The data in the Global Competitiveness Reports are based on annual surveys among approximately 4000 to 5000 senior business executives. The data in the International Country Risk Guides are computed from an in-house panel of experts of the PRS Group. In both cases, the data are compiled professionally. For example, the selection of respondents for the World Economic Forum's surveys is largely representative of the structure of each economy, and the respondents have excellent knowledge of their country's economy. The International Country Risk Guides are based on the assessments of experienced experts. (1)
For each variable included in the EFW index (a total of 38), the relevant data are translated into ratings with the aid of a formula that is published in the reports of the Fraser Institute. The rating scale ranges from 0 to 10, with 0 representing the lowest and 10 the highest degree of economic freedom. The area ratings for each of the five areas are calculated as the arithmetic means of the ratings for their respective components. In turn, the summary rating is the average of the five area ratings.
Applying the most recent methodology, which was further developed over the years, the EFW index was calculated for every fifth year from 1970 on plus for the years 2001, 2002, and 2003. As no unemployment data compiled according to a uniform methodology are available for years prior to 1980 (see the following discussion), the following regressions are based on data for the years 1980, 1985, 1990, 1995, 2000, 2001, 2002, and 2003. This covers almost a quarter of a century up to the most recent past. However, a drawback is that only a maximum of eight observations is available for each country. Certain countries, mainly transition countries and some developing countries, were included only in the ratings from the late 1980s or early 1990s.
The country group consists of no fewer than 87 countries (Appendix C). Whereas almost all previous cross-country labor market studies only covered industrial countries, this paper also includes a large number of developing and transition countries. The country sample thus is unusually large for a labor market study. However, three qualifications should be noted in this respect. First, as mentioned previously, unemployment data are available only from 1980.
Second, the EFW index even covers 127 countries, 40 of which could not be taken into account, as no labor market data are available for these primarily African countries. Third, for many of the developing and transition countries included in the sample, unemployment data are even more scant than EFW data, particularly for the 1980s. This is the main reason why the number of observations is always markedly below 696 (87 countries, eight years) in our main regressions (section 2).
To measure the effects on unemployment, we use not only the overall unemployment rate but also unemployment rates relating to female and young workers (for definitions and sources, see Appendix A). This enables us to determine not only to what extent economic freedom affects unemployment among the total labor force but also to what extent it affects two important demographic groups that often have above-average unemployment rates.
Almost all the unemployment data used in this paper come from the latest edition of the International Labour Office's (ILO's) Key Indicators of the Labour Market (International Labour Office 2005). All variables are exclusively based on labor force survey data. Thus, the data do not refer to registered unemployment. Instead, they are based on an international standard that defines the unemployed as all persons above a specific age who, during the reference period, were without work, currently available for work, and seeking work. Although national coverage of unemployment can vary with regard to factors such as age limits and criteria for seeking work, in the latest edition of its Key Indicators of the Labour Market, the ILO has undertaken great efforts to produce series that are comparable across countries. With regard to age limits, for example, most national series presented in this publication refer to the age-group 15 years and older. Furthermore, for the latest edition of its Key Indicators of the Labour Market, the ILO has "cleaned" the national time series to eliminate breaks in series. Thus, the data are comparable over time. Although the ILO's unemployment data are not completely harmonized across countries, they are comparable to the greatest extent currently possible.
To control for business cycle conditions, we normalized each country's GDP growth rate for its average growth rate over the previous 10 years. (2) In addition, we use year dummies to control for year-specific effects. We also control for the percentage share of children in the population. This share varies widely across countries, especially between developing and industrial countries. Large variations in this share are likely to affect the labor market.
Furthermore, we use two variables to control for the impact of geographical conditions. In a series of papers, Sachs and coauthors have demonstrated that levels and growth rates of GDP per capita are strongly correlated with key geographical variables (e.g., Gallup, Sachs, and Mellinger 1999; Sachs 2001). Our first geographical control is the share of land area in geographical tropics. Sachs and coauthors argue that tropical climates hinder production and development. One may thus hypothesize that they may also increase unemployment. Our second geographical control is the mean distance to the nearest ice-free coastline. A long distance is likely to increase transport costs for international trade, thus possibly increasing unemployment.
We also control for ethnic fractionalization. In ethnically heterogeneous societies, the group that comes to power may implement policies that lower output and increase unemployment (e.g., policies that prohibit the growth of industries dominated by the losing groups). Indeed, previous studies have found ethnic fractionalization to be inversely related to GDP per capita growth (e.g., Easterly and Levine 1997; Alesina et al. 2003). Finally, we employ a control variable for interstate and internal wars because they may severely disrupt the labor markets of the countries in which they take place.
In our main regressions, the coefficients are estimated using the random effects, feasible generalized least squares (FGLS) procedure that incorporates time-invariant country effects (Swamy-Arora method). This enables us to exploit both the cross-country and the time-series variation included in the sample while simultaneously controlling for unobserved country effects. (3) Allowing for cross-country differences in unemployment that reflect the influence of omitted variables is highly desirable, but the random effects method for...