Dragnet-Controls and Government Ideology

In many European countries established leftwing and rightwing parties have been threatened by policy convergence and, in turn, new populist parties that offer more polarized platforms. I investigate whether government ideology influences dragnet-controls - controls of persons conducted by the police without having any suspicion that the controlled person committed a crime. They took effect after the Schengen Agreement in 1995. I propose dragnet-controls to be a prime example of ideology-induced policies implemented by established parties. Dragnet- controls confine citizens’ liberties, but may help to detect criminals. Using data from the 16 German states over the period 1995-2017, I show that rightwing governments have been active in implementing dragnet-controls and leftwing governments have denied dragnet-controls since the mid 1990s. When evaluating how government/party ideology influences individual policies, previous econometric studies ignored initiatives in parliament. My study also considers initiatives in parliament. I conjecture that, since 2015, the rightwing CDU has used initiatives to introduce/extend dragnet-controls to also deal with the upcoming threat of rightwing populism. Future research should examine policy-differences between the established parties regarding more fine-grained policy measures also in other European countries.


Introduction
Exporter characteristics and dynamics vary significantly across countries. The average Chilean exporter is almost three times as large as the average Moroccan exporter, and Chile has one-third more exporters than Morocco. There are twice as many exporters in Bulgaria as in Bangladesh, but a Bulgarian firm exports on average half as much as a Bangladeshi firm. There is also wide variation in exporter dynamics across countries. Firms from Cameroon and Malawi have high entry rates into exporting but only 25 percent of entrants survive after the first year. In contrast, firms from Brazil and South Africa have lower entry rates, but among entrants, about half survive for more than one year. This paper shows that these measures of exporter behavior vary in systematic ways with country size and stage of development and considers potential explanations.
Our first goal is to introduce the Exporter Dynamics Database and use it to characterize export behavior at the firm level across countries. We use exporter-level customs information from 38 developing and 7 developed countries as input to build the "Exporter Dynamics Database" (henceforth referred to as the Database). The Database contains the number of exporters, average size of an exporter, exporter concentration, rates of entry and exit and entrant survival among exporters for each country at various levels of disaggregation. Using these measures, we examine how exporter behavior varies with country size and stage of development. Many previous studies use data from individual countries or a small group of countries in a region to study how firms export. 1 However, their focus has been on the size distribution of exporters and the process of resource reallocation in response to changes in trade costs within a country. In the absence of 3 cross-country variation, such studies were unable to examine how exporter behavior changes as countries get larger or richer. We seek to fill this gap in the literature.
Our second goal is to examine predictions from trade theory with heterogeneous firms and models with allocative inefficiencies for consistency with the new evidence on exporter behavior across countries. Systematic variation in export-sector characteristics across countries of different sizes or income levels could be a result of differences in resource endowments or the process of resource allocation, among other factors. To explore how differences in resource endowments affect exporter behavior, we stick to the most basic model, the standard heterogeneous firm trade model of Melitz (2003). The model draws a direct link between firm size and inherent productivity, so the biggest firm will always also be the most productive. A stark prediction comes out of this model: larger countries should export more because they have more firms (the extensive margin), as opposed to larger firms (the intensive margin). This theory has little to say about stage of development or firm dynamics.
To explore how variation in allocative efficiency across countries affects exporter behavior, we turn to the growing literature on efficiency gains from within-sector resource reallocation across firms. These studies attribute variation in firm size not only to inherent productivity differences (like Melitz 2003) but also allow a broad set of distortions to resource allocation between firms to affect the size distribution. 2 In support, empirical studies of developing countries find a relatively wide variation in firm performance and weak correlations between size and productivity within narrowly defined sectors in a country, suggesting that misallocation of resources between firms is indeed a major impediment to growth. 3 An implication from this 2 The distortions can be associated with several types of institutions and policies, such as labor market regulations, regulations on business entry, access to finance, taxation etc. 3 Syverson (2004) finds that a U.S. firm in the 90 th percentile in a four-digit SIC industry (443 industries) is on average twice as productive as a firm at the 10 th percentile in the same industry. In developing countries, productivity differences tend to be even 4 literature is that countries with fewer distortions (more developed countries) should have more exporters and higher survival rates of entrants because the most productive firms have the opportunity to grow, begin exporting and expand into foreign markets. But importantly, the implications for average exporter size and exporter concentration at the top of the distribution depend on which firms are most constrained. If only the most productive firms are able to overcome regulatory hurdles in economies with allocative inefficiencies and mid-productivity firms are held back-the so-called "missing middle" of the size distribution-then exporters should be relatively larger in developing countries and very concentrated at the top of the distribution (Tybout 2000, 2014and Alfaro, Charlton, and Kanczuk 2009. In contrast, if the highproductivity firms are the most constrained and fail to invest (Hsieh and Klenow 2009, Hsieh and Olken 2014, and Bento and Restuccia 2014, then exporters should be relatively small in more distorted economies with less concentration at the top of the distribution, which we refer to as the "truncated top" of the size distribution.
We examine whether the evidence from the Database is consistent with these predictions.
We find that export-sector characteristics are correlated with both country size and stage of development in a systematic way. In particular, larger economies and more developed economies have a greater number of exporters, a larger average exporter size, and a greater concentration of exports in the top 5 percent of exporting firms. The extensive margin (more firms) explains about two-thirds of the increase in exports of larger countries, while the intensive margin (larger firms) explains the remaining third. In contrast, the intensive margin is relatively more important as an explanation for why richer countries export more. These results are robust to different levels of larger with Hsieh and Klenow (2009) finding a ratio for the 90:10 percentile of 5:1 for China. Bartelsman et. al (2013) estimate the covariance between productivity and size in narrowly defined industries, in a group European countries and the US. They find the covariance between size and productivity was near zero (or negative) at transition in Eastern Europe and has since increased, i.e. allocative efficiency has improved sharply. 5 aggregation of the data, to controls for various types of fixed effects, to the prevalence of zeros in trade, and to proxies for the different importance of trade intermediaries across sectors.
The results on exporter dynamics reveal a robust and significant relationship with stage of development but not country size. We find that gross rates of entry into and exit from exporting are decreasing in stage of development, but net entry is unrelated to stage of development. There is also evidence that survival rates of entrants into export markets increase as countries develop.
These results are robust to different levels of aggregation of the data and to controls for various types of fixed effects.
These stylized facts imply that as countries develop and exports grow, the export expansion happens through both the extensive and the intensive margins, and more resources flow to the largest firms. Consistent with the standard model of trade with heterogeneous firms, we find a relatively large role for the extensive margin in explaining why larger countries export larger volumes. Consistent with resource allocation improving with stage of development, we find that more developed countries have both more exporters and more resilient exporters. Overall, the positive correlations between average exporter size and stage of development and between concentration of exports at the top of the size distribution and stage of development are consistent with models where firm growth is constrained in developing countries, especially among the highproductivity firms, the "truncated top".
The rest of the paper is structured as follows. Section 2 describes briefly the Database and the variability in outcomes across countries. Section 3 discusses how the theory can help understand this variability across countries. Section 4 presents the new stylized facts on the relationship between export-sector characteristics and dynamics on the one hand, and country size and stage of development on the other hand. Section 5 discusses the results in light of trade theory 6 and allocative efficiency and examines how exporter behavior changes with export growth over time for robustness. Finally, Section 6 concludes.

The Exporter Dynamics Database
The Exporter Dynamics Database contains aggregated measures on export-sector characteristics and dynamics presented at the following levels: a) country-year, b) country-product (HS 2-digit, HS 4-digit, or HS 6-digit)-year and c) country-destination-year, and d) country-HS 2-digit productdestination-year (for a restricted sample of countries). The data are primarily for the period between 2003 and 2010. The measures are constructed using exporter-level customs data as input, covering the universe of annual exporter transactions for 38 developing countries and 7 developed countries. 4 The measures at the country-sector-destination-year level are available for 34 developing countries for which we have access to the raw exporter-level data and for 3 developed countries. 5 The Database includes measures on the basic characteristics of the export sector in each country (e.g., number of exporters, average exporter size and exporter growth rates); concentration/diversification (e.g., Herfindahl index, share of top exporters, number of products and destinations per exporter); firm, product and market dynamics (e.g., entry, exit, first-year and 4 The stylized facts presented in this paper use only the aggregated measures contained in the Database. The exporter-level customs datasets used as inputs to construct the Database are described to explain how the measures in the Database were built, but are not utilized in this paper for any other purpose. These exporter-level datasets are available upon request for a selected group of countries, determined by the willingness to share data by the data providers. Given the confidential nature of the exporter-level (firm-level) data used as input, we structured the Database to provide as much detail as possible on the underlying exporter structure behind an export flow, without revealing any information that could be traceable to an individual firm, as per what we agreed with all of our data providers. While the data underrepresent industrial countries and Asia, the patterns we find are robust and available statistics from other studies, such as for the US and the EU, are consistent with the patterns observed here. We aim over time to expand coverage of the Database. 5 Table 1 marks with an asterisk the developing countries for which we have access to the raw exporter-level data. The measures at the country-year-HS2-digit product-destination level are currently only available upon request to the authors. They will be made publicly available with the second release of the expanded Database. Measures at the country-year-HS2-digit product-destination level for Norway, Portugal, and Spain were made available to the authors by Andreas Moxnes, Joana Silva, Asier Minondo and Francisco Requena, the researchers working with data for those countries.
second-year survival rates of entrants); and unit prices (e.g., per exporter of a given product). The Database is publicly available at http://data.worldbank.org/data-catalog/exporter-dynamicsdatabase. Details on how the countries' customs data were cleaned and harmonized and how the Database was constructed are provided in Appendix 1. Further details on all the measures in the Database for each type of aggregation level are provided in Cebeci, Fernandes, Freund, and Pierola (2012) (2015) use the cross-country exporter-level data and show that exports are dominated by a small group of very large exporters (so-called 'export superstars') considered as the top exporter or the top 5 exporters, who account for on average 15 and 30 percent of exports, respectively. Their focus is on how this group of large firms defines the sectoral pattern of exports. Focusing on data at the country-industry or product level, Easterly and Reshef (2014) show that both African and non-developing countries concentration is smaller than for the developed countries in the sample, such as Norway, Sweden, and New Zealand. Similarly, Bernard, Jensen, and Schott (2009) and Mayer and Ottaviano (2008) show that the United States (U.S.) and other European countries exhibit high concentration in the top firms.
The measures of exporter dynamics also exhibit important variability across countries.
Entry rates-defined as the number of exporters in year t not present in year t-1, over the total number of exporters in year t-range from 22 percent in Brazil to more than 50 percent in Malawi, Yemen, the Lao People's Democratic Republic and Tanzania. Exit rates-defined as the number of exporters in year t-1 not present in year t, over the total number of exporters in year t-1-range from 22 percent in Bangladesh to 61 percent in Malawi. 9 First-year survival rates of entrantsdefined as the number of exporters in year t and t+1 not present in year t-1, over the total number of exporters in year t-vary between 23 percent in Cameroon and 61 percent in Bangladesh. The magnitude of the survival rates in Table 1 suggests an extremely high attrition rate of new entrants after just one year in export markets, particularly in Africa. Table 2 presents the same set of measures but for groups of HS 2-digit productshenceforth designated as sectors -where the measures for each group of sectors are obtained as averages across all countries that export that particular sector, again focusing on the period 2006-2008. Table 2 shows a less pronounced variation in the number of exporters across sectors though African countries exhibit very fat-tailed distributions of export value across products where export big hits are uncommon but are much more common than would be expected in a normal distribution whereas Hanson, Lind, and Muendler (2014) document a strong empirical regularity across countries in their patterns of comparative advantage which is hyper-specialization in exporting, the fact that in any given year, exports in the typical country tend to be highly concentrated in a small number of industries. 9 These rates of entry into and exit from export markets are substantially larger than the rates of entry into and exit from manufacturing activity reported by Bartelsman, Haltiwanger, and Scarpetta (2013). Focusing on census/registries of manufacturing firms (excluding 1-employee businesses) they document turnover of 10-20 percent in industrial countries and somewhat higher in developing countries. 9 some important sectoral differences in other measures remain. The significance of these differences will be accounted for in Section 4.

Relationship with Country Size and Stage of Development
In this section, we explore the predictions about how export-sector characteristics and dynamics are likely to vary with country size and stage of development. We draw on the growing theoretical trade literature on heterogeneous firms and recent empirical evidence on how the allocation of resources differs across countries.

Export-Sector Characteristics
The standard heterogeneous firm model of trade (Melitz 2003) assumes firms are differentiated by the variety they produce and by their productivity, and that there is a fixed entry cost into exporting.
A firm will enter the export market if its net export profits cover the fixed exporting cost. Because of the entry costs, there is a cutoff productivity level, such that firms at or above it will export while those below it will produce only for the domestic market. Thus, firms that export are more productive than firms that produce only for the domestic market, and export quantity is increasing in productivity. The basic model assumes a common productivity distribution across countries and a continuum of atomistic firms. This implies that greater exports of larger countries should be driven by a greater number of exporting firms, the extensive margin, and not by larger firms. The standard heterogeneous firm model of trade has little to say about stage of development.
We use recent empirical literature on allocative efficiency between firms to inform us about how stage of development and export-sector characteristics relate. This literature suggests that countries with fewer distortions (richer countries) should have more exporters but the implications for average exporter size and concentration at the top of the size distribution vary, depending on which firms are more constrained. If taxes and regulations (e.g., on the labor market and on business entry) become heavy as firms grow larger, only the most productive can overcome these costs and there would be a "missing middle": many small firms and a few large firms but few medium-sized firms. Consistent with a "missing middle", Tybout (2000) shows a U-shaped firmsize histogram for a number of developing countries, when firms are allocated to relatively wide size bins. 10 Tybout (2014) shows that the "missing middle" does not necessarily imply a bimodal firm-size distribution. Assuming that the undistorted firm-size distribution should be an exact Pareto distribution, he compares the actual distribution with an estimated country-specific Pareto distribution. Using data for a number of Asian countries, the U.S., and Mexico, he finds that the actual firm-size distribution is wider than the estimated Pareto distribution, as represented in Panel A of Figure 1, particularly in developing countries. His results, however, are dependent on allowing the Pareto dispersion parameter to vary across countries, and he estimates a lower Pareto index (a more skewed firm-size distribution) in higher-income countries. In related work, Alfaro et al. (2009) use Dun and Bradstreet's WorldBase data for 79 countries and find that establishment size is decreasing in GDP per capita, and interpret this as being consistent with too few mediumsized firms in lower-income countries. 11 In contrast, Hsieh and Klenow (2009) calculate a hypothetical efficient firm-size distribution, assuming the marginal products of capital and labor are equalized across all firms in an industry and compare it to the actual firm-size distribution. They find that in China, India, and the U.S. the efficient firm-size distributions were more dispersed than the actual distributions.
Panel B of Figure 1 illustrates their general finding. In contrast to the work of Tybout (2014), there are not enough very large or very small firms. They also find that the distortion is larger for the developing countries: the observed U.S. firm-size distribution is closer to the hypothetical efficient firm-size distribution than either the observed Indian or Chinese firm-size distribution.
Hsieh and Olken (2014), using data from India, Indonesia and Mexico, find that the fraction of "missing" firms is increasing in firm size. There are more missing large firms than missing medium-sized firms, and this helps explain why less developed countries have lower income. Bento and Restuccia (2014) use comparable firm-census data for 124 countries with size measured by employment and show that the average firm size is increasing in level of development. They develop a model to explain this fact, using distortions correlated with productivity to produce too little investment, especially among firms that should grow large and such distortions can explain productivity differentials between higher-and lower-income countries. This body of research points to a "truncated top" of the firm-size distribution.
A wealth of research shows that small firms do not export because of high fixed entry costs. 12 In Figure 1, it is therefore only the right half of the firm-size distribution that is relevant for exporters. A "missing middle" in less efficient (less developed) economies would suggest that exporters are on average relatively large and their size distribution is more concentrated at the top.
A "truncated top", with firms missing at the top of the size distribution in less efficient economies, 12 would imply that exporters are on average smaller and the concentration in the top 5 percent of firms is lower. 13 Hypothesis on export-sector characteristics: The standard heterogeneous firm trade model implies that larger countries should export more primarily through the extensive margin.
Allocative efficiency implies that less developed countries have fewer exporters. If distortions in less developed countries hinder large firms from growing relatively more, a "truncated top" (Hsieh and Klenow 2009, Hsieh and Olken 2014, and Bento and Restuccia 2014, then less developed countries will have smaller exporters on average and lower export concentration at the top. If distortions primarily limit medium-sized firms from growing, a "missing middle", then the reverse is true (Tybout 2000, 2014and Alfaro et al. 2009).

Exporter Dynamics
Allocative efficiency has important implications for exporter dynamics. In a less distorted environment, where the most productive firms grow most rapidly, the survival rates of new exporters should be relatively high. In contrast, in an economy where regulatory burdens and distortions in access to finance or information uncertainties allow less productive firms to grow, survival rates of entrants in competitive foreign markets will be relatively low. Along these lines, Araujo, Mion, and Ornelas (2012) integrate work on institutions and heterogeneous firm models of trade, and predict greater exporter survival rates in countries with better institutions (which tend to be at a higher stage of development), where uncertainty is lower.
Allocative efficiency's implications for exporter entry depend on what type of distortions tend to block efficient firms from expanding and/or allow inefficient firms to remain in the export market. Entry should be lower in developing countries if the sunk costs associated with entering foreign markets (such as bureaucratic hurdles, wait times at customs, and corruption) are exceptionally high. However, uncertainty about export success at the firm level will enhance entry and exit, all else equal. With idiosyncratic uncertainty, several models of exporter dynamics show that some firms will enter the export market only to immediately exit (Albornoz et al. 2012, Segura-Cayuela and Vilarrubio 2008and Freund and Pierola 2010. The intuition is that if export profitability is uncertain at the firm level then firms may enter foreign markets to learn their potential. Since a firm can exit, there is a one-period loss if the firm gets a poor draw, but there is a stream of future profits if the firm gets a good draw. For a given entry cost, the greater the uncertainty, the higher the "option value of entry" will be-implying more exporter entry and exit. 14 Hypothesis on exporter dynamics: Allocative efficiency implies that entrant survival rates should be higher in more developed countries. If uncertainty is the critical distortion affecting export decisions in less developed countries, then there should be relatively higher exporter entry and exit. If sunk entry costs are higher in less developed countries and matter more than uncertainty for export decisions, then there should be relatively less exporter entry (and exit consequently).

The models make no predictions about exporter dynamics and country size.
One remark in order at this point is that, although we focus on differences in resource endowments and resource allocation across countries, we are aware that other factors may be at work. For example, differences in the underlying productivity distribution of firms or in market access across countries may also help to explain our stylized facts. In our empirical work, we control for potential variation in productivity distributions across sectors, or any other differences that may affect the size distribution of firms across sectors, by focusing on exports within a sector; and, similarly, for market access by focusing on exports within a destination or sector-destination, as well as controlling for distance and other bilateral trade costs. Still, to the extent the underlying productivity distribution varies across countries within a sector-destination, this would influence our results. We discuss this possibility in more detail in Section 5.

Relationship with Country Size and Stage of Development
In this section, we evaluate the empirical evidence from the Database on how export-sector characteristics and dynamics vary with country size and stage of development. We first show the importance of both country and sector characteristics in explaining exporter behavior. We then present two stylized facts on how export-sector characteristics and dynamics change as countries get larger or richer, controlling for sectoral variation in exports and variation across destinations, as well as sector-destination variation.
on GDP and on GDP per capita in constant 2005 Purchasing Power Parity (PPP) USD. 17 We also use data on determinants of bilateral trade -distance, contiguity, common language and common colonizer -from CEPII described in Mayer and Zignago (2011). Using the yearly data from 2004 to 2008, we next explore these relationships in more detail, controlling for sectoral variation in exports, variation within destinations, and sector-destination variation, in addition to controlling for year fixed effects in all specifications. The standard errors reported in all tables are clustered by country (the most conservative type of clustering possible). 19 Table 4 reports the results from regressing log total exports, log number of exporters, log average exporter size, and share of exports accounted for by the top 5 percent of firms on log GDP per capita and log GDP. Columns (1)-(4) report the results using the country-sector-year level data for years 2004 to 2008 and controlling for year and sector fixed effects. Accounting for sector fixed effects is key as some sectors exhibit much larger average exporter sizes than others, for example due to their higher capital intensity or degree of returns to scale, as shown by the importance of sector effects in the ANOVA (Table 3). 20 The results indicate that, within sectors, larger countries and richer countries have more exporters, larger-sized exporters (on average), as well as higher concentration in the top 5 percent of firms. 21 The economic magnitude of these effects is such that if GDP per capita increased from the value for the country in the 25 th percentile (Cameroon) to the value for the country in the 75 th percentile (Lebanon) the number of exporters would increase by 80 percent, average exports per firm would increase by 71 percent, and the share of the top 5% exporters would increase by 8 percentage points. 22

Export-Sector Characteristics and Country Characteristics
Columns (5)-(8) in Table 4 use the country-destination-year level data for years 2004 to 2008 and results are qualitatively similar. In addition to destination fixed effects these specifications control for traditional determinants of bilateral trade in gravity equations: distance, contiguity, common language, and common colonizer. The estimates show that controlling for year and market-specific characteristics, such as trade costs, more developed countries and larger countries have more exporters that on average export larger volumes and have higher export concentration at the top of the size distribution. 23 20 The results remain similar if we use country-HS 4-digit-year level data and control for 4-digit sector fixed effects. 21 Total exports are equal to the product between the number of exporters and average exporter size. Since our regressions use the log of each of those variables as dependent variable and are estimated linearly, the coefficient on GDP (or GDP per capita) in the total exports regression is exactly equal to the sum of the coefficients on GDP (or GDP per capita) in the number of exporters regression and in the average exporter size regression, provided that the samples used in the three regressions are similar. Therefore in Tables 4, 5, 6 and Appendix Table A2.2 we restrict the sample in the number of exporters' regression to be the same as that used in the total exports or average exporter size regressions, that is, it excludes observations with the number of exporters equal to 1. However, note that in unreported results where we estimate number of exporter regressions including those observations the effects of GDP per capita and GDP are stronger in magnitude and significance that those shown. 22 These magnitudes are obtained as the product of the difference in logs GDP per capita, 1.704, and the corresponding coefficient in columns (2)-(4) of Table 4: 0.471, 0.419, and 0.047. 23 As robustness check to the result on average exporter size, in Appendix Table A2.1, we re-estimate the specifications for average exporter size using data on incumbent exporters only, in order to exclude possible biases due to the use of "partial year" data as discussed in Bernard et al. (2013). In their analysis, they show that the entry and exit of firms into and from export markets at different points in a calendar year means that the use of calendar annual data on export values to measure the average size of exporter entrants or exiters may introduce a bias -mainly their size being under-estimated during the first year -which they refer to as "partial year" effects. Our consideration of the average size of incumbent exporters addresses this concern to the extent that bias should not permeate the calculation of average size for exporters that have not interrupted their exporting activity. We find that the results from Table 4 remain qualitatively similar under this alternative sample.
Note that the coefficients in columns (1)-(3) as well as columns (5)-(7) imply that the number of exporters, the extensive margin, explains about two-thirds of the variation in total exports across countries of different size, and about 60 percent of the variation in total exports as a result of stage of development.
Columns (9)-(12) in Table 4 present results using the country-sector-destination-year level data for years 2004 to 2008 for the sample of 34 developing countries and 3 developed countries. 24 This allows us to control for interacted sector-destination fixed effects in the same regression (as well as for bilateral gravity determinants), but has the disadvantage of less variation in terms of stage of development, with only 3 developed countries included. The results confirm that within a given sector-destination, larger countries and richer countries have more exporters, larger exporters, and a greater share of exports controlled by top 5 percent of exporters. Moreover, the findings are maintained in specifications that include more stringent sector-destination-year fixed effects to account for the possibility of temporal shifts affecting sectors and destination markets and influencing exporter behavior due to trade networks and relationships, as seen in Appendix Table A2.2 (Panel A). Of interest, in the country-sector-destination regressions, the contribution of the intensive margin rises when we control for sector and destination, especially with respect to GDP per capita: the intensive margin now accounts for more than half of the variation in exports with GDP per capita.
It is worth noting that the coefficients on trade costs shown in columns (5)-(12) of Table 4 indicate that they have opposite effects on export-sector characteristics, relative to country size and stage of development. Higher trade costs (proxied by higher distance, absence of either common language or border) reduce exports through both the extensive and intensive margins and lead to lower exporter concentration. These results are similar to results obtained using data on U.S. exports (Lawless 2010).
One potential concern about the average exporter size results is that they may pick up the activity of intermediaries, individual firms that intermediate trade for many producers. An intermediary will be observed in our data as a single firm, but it actually consolidates exports for many firms, and there is no way to distinguish these firms from direct exporters. Ahn, Khandelwal, and Wei (2011) offer evidence on the importance of these firms in trade. They find that a significant share of exports in China went through intermediaries in the early 2000s. However, they also find that as trade expands, the importance of intermediaries declines-their share fell from 35 to 22 percent between 2000 and 2005 as China's exports surged. In part, this is because as firms expand their exports they switch from exporting through intermediaries to direct exporting. This suggests that the average exporter size variable is more likely to overstate actual exporter size and the number of exporters is more likely to understate exporter count in the less developed or small countries as compared with richer or larger countries. To examine explicitly the importance of intermediaries we estimate specifications where GDP per capita and GDP enter by themselves and interacted with a dummy variable identifying sectors with a larger presence of export intermediaries. 25 The corresponding estimates shown in Table 5 reveal significantly larger numbers of exporters, larger average exporter size and higher concentration in richer countries and in larger countries as in Table 4. 26 In sectors with more intermediaries, the overall effect of income per capita on the number of exporters and on average exporter size is smaller but remains positive and significant overall.
Finally, a potential concern about the results in Table 4 is that our measures are defined only for cells with a positive export flow but trade zeros can be important. By expanding our original data to have zeros in cells with no export flows, we find that the prevalence of trade zeros is negligible in the country-sector-year data (5.9% on average across countries) but is very large in the country-destination-year data (47% on average across countries) and the country-sectordestination-year data (91% on average across countries). The trade zeros in our data are not random but rather are more prevalent both in smaller countries and in less developed countries. Hence our findings could be influenced by the relationship between zeros, country size and stage of development. To investigate this possibility we use the non-linear Poisson pseudo-maximumlikelihood (PPML) estimator introduced by Silva and Tenreyro (2006) to estimate our specifications in Table 4 using the expanded data. 27 The corresponding estimates, shown in Table   6, reveal that our main results on stage of development are robust to the inclusion of zeros. 28 With the exception of average exporter size, the results on country size are robust in the country-sector and country-destination regressions, and all results remain strong in the country-sector-destination regressions.
Stylized Fact on export-sector characteristics: Larger economies and more developed economies have more exporters, a larger average exporter size, and more concentrated export sectors among firms, after controlling for the sectoral distribution of exports and for export destinations. The extensive margin explains about two thirds of the variation in exports across countries of different size. The extensive margin is relatively less important in explaining why richer countries export more.

Exporter Dynamics and Country Characteristics
We next turn to exporter entry, exit and survival. Figure 3 shows how exporter entry, exit, and survival vary with country size and stage of development, using country-year level data averaged for the period 2006-2008. With respect to entry and exit, there is a negative relationship indicating that richer countries and larger countries have lower entry and exit rates. The relationship with entrant survival is less transparent.
Columns (1)-(5) of Table 7 show the results from regressions of exporter entry, exit, firstyear survival of new exporters, net entry, and turnover on log GDP per capita and log GDP based on the country-sector-year level data for years 2004 to 2008, controlling for sector and year fixed effects. 29 It is crucial to account for sector fixed effects as some sectors have much lower turnover than others, as highlighted by the relative importance of sector effects in the ANOVA (Table 3).
The results show that controlling for the sectoral composition of trade, stage of development has important effects on exporter dynamics. In particular, richer countries have lower turnover and higher rates of entrant survival, leaving net entry roughly unchanged. The economic magnitudes of the results are sizable-increasing GDP per capita from the value for the country in the 25 th percentile (Cameroon) to the value for the country in 75 th percentile (Lebanon) would lead to a decline in entry and exit rates by 11 percentage points, and to an increase in the first-year survival rate of entrants by 6 percentage points. 30 In columns (6)-(10) of Table 7 we report results using country-destination-year level data for years 2004 to 2008 and controlling for destination and year fixed effects, along with bilateral gravity variables. The results on the effect of stage of development are qualitatively similar to the results using country-sector-year level data, though the magnitudes are smaller. In this case, there is evidence that country size also matters, with larger countries having lower entry and exit and higher entrant survival rates.
Columns (11)-(15) of Table 7  These results show that exporter dynamics change as countries get richer. In less developed countries, turnover is largely a process of entry and exit where many firms enter into export markets and exit almost immediately. In richer countries, fewer but more resilient exporters enter in any given year.
Stylized Fact on exporter dynamics: Exporter turnover is significantly lower in more developed countries, exporter entrant survival is higher, while net entry is invariant with respect to income per capita, controlling for the sectoral distribution of exports and for export destinations.

Discussion of the Results
The standard heterogeneous firm trade model predicts that holding the distribution of firm productivity constant across countries, the extensive margin should explain why larger countries export more. As theory predicts, we do find that the extensive margin dominates the intensive margin in explaining cross-sectional variation in exports, though there is still a significant role for the intensive margin. Specifically, the extensive margin explains about two thirds of the effect of country size on total exports. The importance of the intensive margin in explaining why larger countries export more suggests that the firm productivity distribution may vary in a systematic way with country size and warrants further study. 31 The intensive margin is relatively more important in explaining the variation in exports owing to stage of development. Specifically, within a sector-destination, average exporter size explains more than half of the effect of GDP per capita on total exports. We argue that the relationship between stage of development and both the intensive margin and concentration at the top of the distribution is consistent with better resource allocation across firms in more developed countries along the lines suggested by Bento and Restuccia (2014), where the most productive 31 Consistent with our finding, Spearot (2013) adapts a heterogeneous firm model, allowing the Pareto shape parameter of the firm productivity distribution to vary across countries. He shows that such an accommodation can explain variation in the average size of exporters across countries. He further shows that the response of export flows across countries of varying sizes to a trade shock can be used to estimate the Pareto shape parameters across countries. Using this methodology, he finds that larger and richer countries have more skewed firm productivity distributions (hence exporter size distributions), a result that could explain our findings on the larger average size in larger and richer countries. His model, however, does not explain why the Pareto shape parameter is more skewed in larger or richer countries. Alternatively, Head, Mayer, and Thoenig (2014) show that a lognormal distribution fits the observed exporter size distribution of French and Chinese firms better than a Pareto. Such a distribution could allow for a link between the intensive margin and country size, within a heterogeneous firm framework. firms fail to invest enough. 32 Our finding in Tables 4 and 6 that as trade costs (as measured by the gravity variables) fall, average size and concentration at the top of the exporter size distribution rise, is also consistent with the allocative efficiency story. To the extent that higher trade costs impede competition, allowing a less efficient allocation of resources to persist, we would expect the largest and most productive firms to perform especially well when the distortions are reduced.
In Section 5.1, we turn to time-series evidence as a further robustness test of the results.
The results on exporter dynamics also speak to a more efficient allocation of resources in more developed countries. Better firms grow and these high-productivity firms are less likely to exit immediately following entry into the export market. The relatively higher entry and exit in developing countries suggests that the distortions relating to uncertainty about export profitability are greater than entry cost distortions. Effectively, there is a lot of trial and error in exporting, especially in less developed countries where exit goes up precisely because there are more entrants.
Indeed, this is consistent with high correlations between entry and exit at the country level and country-sector level. 33

Average Exporter Size and Concentration and Allocative Efficiency
While the results are consistent with high-productivity firms failing to invest enough, resulting in too few large firms and lower exports at an earlier stage of development, they could alternatively reflect systematic differences in the underlying productivity distributions and not be a result of resource allocation. To explore how changes in the allocation of resources affects the number of 32 Consistent with better resource allocation in richer countries, we also find in unreported results evidence of more diversification at the firm level as countries get richer. This is consistent with models of multiproduct firms, such as Bernard, Redding and Schott (2011), where only higher-ability firms are able to generate variable profits to cover fixed costs of entry for more products and thus supply a wider range of products to each market. 33 For example, using the data for years 2004 to 2008, the simple within-country correlation between entry and exit is 0.77 and the within-country-sector correlation between entry and exit is 0.76. exporters, average exporter size, and concentration, we consider times-series evidence. We use country-sector-destination-year level data for years 2004 to 2008 for the sample of 34 developing countries and 3 developed countries. Assuming the productivity distribution is constant within country-sectors over time, focusing on within-country-sector variation allows us to rule out differences in the productivity distribution as the cause. The disadvantage of panel data is that the effect of independent variables that have little within-country-sector variation, like country size and stage of development, are harder to isolate with precision. For this reason, we focus on export growth, which varies across countries, sectors, and destinations.
One problem with using export growth is that its underlying cause is important. For example, exports could grow because of demand or supply shocks, which may affect exporter behavior but be unrelated to the resource allocation across firms in the country. To examine export growth that is likely to be a result of changes in resource allocation, we control for country-sectoryear fixed effects, which capture productivity shocks (as well as other country-sector time-varying factors in the exporting country-sector). We also include destination-sector-year fixed effects, which capture demand shocks (or other destination-sector-year time-varying factors) in the importing country-sector in our most stringent specification. This allows us to decompose export growth in a sector that is specific to the bilateral relationship, and hence very likely related to changes in trade costs, which we know affect resource allocation. For example, we are decomposing Peru's export growth in apparel to the US that did not result from an increase in overall US demand for apparel or from Peruvian export growth in apparel.
The null hypothesis that we test is that export growth in a country-destination-sector is uncorrelated with average exporter size and exporter concentration, controlling for country-sectoryear shocks and destination-sector-year shocks. A positive correlation between export growth and both variables shows that exporters become on average larger and the firm-size distribution becomes heavier at the top as exports grow.
To the extent that better resource allocation promotes export growth, we would expect average exporter size to increase and exporter concentration to intensify as exports grow. If differences in average exporter size and concentration are more closely related to the underlying firm productivity distribution than to allocative efficiency then there need not be any correlation between export growth and average exporter size and concentration, especially after controlling for country-sector-year effects and destination-sector-year fixed effects. Table 8 reports results using the country-sector-destination-year level data with countrysector-destination (panel) fixed effects and country-sector-year and destination-sector-year fixed effects. Nearly 90 percent of the time-series variation in exports in comes from larger exporters.
The estimated within-effects show that, as exports grow, the average size of exporters and the share of exports by the top 5 percent increase significantly. To the extent that export growth reflects improved resource allocation over time, these results suggest that export growth is associated with more resources being allocated to the largest (and presumably most productive) exporters. 34

Conclusion
While the literature on exporter dynamics is rapidly growing, there has not been a comparison across a large number of countries of different sizes and at differing stages of development. The Exporter Dynamics Database fills this gap and establishes two novel stylized facts on how exportsector characteristics and dynamics vary with country characteristics.
The data point to systematic ways in which the micro structure of exports changes as countries develop. Larger countries and richer countries have both more and larger exporters on average, with greater concentration in the top 5 percent of firms. Together with results from previous studies showing that exporters are more productive than non-exporters and larger exporters are more productive than smaller exporters, our results are consistent with allocative efficiency in export markets improving as countries develop, with more resources flowing to the largest and most productive firms.
Of interest is how reallocation happens. In particular, exporter dynamics are also closely linked with stage of development, with richer countries experiencing higher entrant survival rates and lower entry and exit rates. This is consistent with allocative efficiency improving as countries develop: only the more productive firms grow and enter the export market in more developed countries, and these relatively good firms are less likely to exit. The results suggest that the distortion that affects exporter dynamics across countries is more closely related to variation in the idiosyncratic uncertainty about the profitability of exporting than to the costs of entry into exporting.
Overall, we hope that the measures in the Database will allow the examination of several interesting cross-country questions, cross-country cross-sector questions, and within-country questions. The measures can also be used as controls in estimations that require exporter characteristics at the country-sector level. In particular, as the measures in the Database offer the first opportunity to study exporter behavior on a global basis, some of the facts described above open the door to questions such as: Does trade promote growth via firm size or firm count? How 28 can countries attract more large multi-product firms? What determines exporter survival? How does comparative advantage relate to the typical exporter characteristics in an industry? These and many other questions will hopefully be addressed in future research to be conducted using the Database.

Country-Destination Regressions
Notes: Robust standard errors in parentheses, clustered at the country level. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively. The regressions in columns (1) Table 1) as well as Norway, Portugal, and Spain. Kuwait is excluded from all regressions due to lack of data for the 2004-2008 period. The estimating sample for the regressions in columns (1)-(3), (5)-(7) and (9)-(11) excludes observations whose corresponding number of exporters is equal to 1 while that for the regressions in columns (4), (8) and (12) excludes observations whose corresponding number of exporters is smaller than 20.

The Role of Intermediaries
Notes: Robust standard errors in parentheses, clustered at the country level. *** and ** indicate significance at 1% and 5% confidence levels, respectively. The regressions in columns (1)-(4) are based on measures at the country-sector-year level for years 2004-2008 (Brazil and New Zealand are excluded due to the lack of measures at that level). The regressions in columns (5)-(8) are based on measures at the country-sector-destination-year level for years 2004-2008 for the 34 developing countries for which we have access to the raw exporter-level data (the list is provided in Table 1) as well as Norway, Portugal, and Spain. Kuwait is excluded from all regressions due to lack of data for any year in the 2004-2008 period. The estimating sample for the regressions in columns (1)-(3) and (5)-(7) excludes observations whose corresponding number of exporters is equal to 1 while that for the regressions in columns (4) and (8) excludes observations whose corresponding number of exporters is smaller than 20. The sectors with more intermediaries are food (HS01-24), apparel (HS41-43 and HS60-67), textiles (HS50-59), metals (HS72-83), machinery (HS84), wood (HS44-46), and miscellaneous (HS90-99).

Table 6: Export-Sector Characteristics and Country Characteristics -Accounting for Zeros in Export Flows
Notes: Robust standard errors in parentheses, clustered at the country level. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively. The estimation method PPML used in columns (1)-(6) is the Poisson Pseudo-Maximum Likelihood proposed by Silva and Tenreyro (2006) while that used in columns (7) Table 1) as well as Norway, Portugal, and Spain. Kuwait is excluded from all regressions due to lack of data for the 2004-2008 period. The number of observations in columns (7)-(9) follow the modification by Guimaraes and Portugal (2010) that drops from the estimating sample all observations that belong to a country-sector-destination-year cell where all values of the dependent variable are 0 and the sample in column (8) is forced to be the same as that used in columns (7) and (9).  Table 7.

Exporter Dynamics and Country Characteristics -Country-Sector-Destination Regressions
Notes: Robust standard errors in parentheses, clustered at the country level. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively. The regressions in columns (1) (1)-(2) excludes observations whose corresponding number of exporters is equal to 1 while that for the regression in columns (3) excludes observations whose corresponding number of exporters is smaller than 20.

APPENDIX Appendix 1: Constructing the Exporter Dynamics Database Using Customs Data at the Exporter-Level
The measures included in the Database are computed using customs data from 45 countries at the exporter-product-destination-year level covering the universe of export transactions. 35 For 11 countries (Brazil, the Arab Republic of Egypt, Estonia, Lao PDR, New Zealand, Norway, Portugal, Spain, Sweden, and Turkey), we do not have access to the raw data and the statistics were calculated for us by researchers or institutions with access to the raw data. Pooling across the raw datasets for the remaining 34 countries, we obtain a panel with 15 million unique observations at the country-firm-product-destination-year level which is the raw dataset used to construct the Database. Some of the variables in this cross-country raw dataset were subjected to a series of cleaning procedures or had peculiarities that we briefly describe below: firm codes, country of destination, product and export values. 36 Regarding firm codes, exporters are uniquely identified by their actual names, their tax identification number or artificial unique codes randomly created by our data providers. In the case of Albania, Burkina Faso, Cambodia, Cameroon, Mexico, and Uganda their firm coding systems underwent changes in 2007 and in the case of Yemen in 2008. Hence, it is not possible to calculate exporter dynamics measures in the years when those changes occur.
Regarding the country of destination variable, two cleaning operations are applied. One relates to the use of special trade regimes by Bulgaria, Costa Rica, Colombia, Ecuador, Egypt, Jordan, Kuwait, Lebanon, Morocco, Peru, Turkey, Yemen, which implies that customs agencies record the sales from inland to their own free zones/customs warehouses as exports resulting in a larger set of potential destinations. 37 Since there is a lack of uniformity across countries in the definition of the special trade regime we drop the observations related to sales to free zones from our dataset. 38 The other relates to the changes in names that some statistical territories have undergone over time due to spatial divisions. When territories split, we keep them as a single destination since disregarding these changes in names would bias some calculations, especially those related to destination diversification and dynamics. In the end, each country has a potential set of 246 destinations as of end-2011.
Regarding products, although most customs agencies record export transactions at 8-digit of the Harmonized System (HS) or higher, we aggregate the information to the HS 6-digit, since this is the most detailed level comparable internationally. Several cleaning operations are applied to the product variable given that our sample period spans three different HS classifications (HS1996, HS2002 and HS2007). First, we combine all the codes existing under the three HS classifications into an aggregated list of 6065 unique HS 6-digit codes and eliminate the 35 The providers of the raw datasets for each country were mostly governmental agencies, mainly customs offices. Appendix 1 in Cebeci et al. (2012) provides a complete list of the countries included in the Database, the periods for which data is available, and the data sources. 36 Cebeci et al. (2012) describe the cleaning procedures in further detail. Note that the cross-country raw dataset at country-firmproduct-destination-year level includes also quantities exported. 37 The "special trade regime" considers transactions where the goods are sold from the domestic territory only to both third countries and free zones/customs warehouses of the origin country as exports. In contrast, the "General trade regime" considers transactions where goods are sold from any national territory (including free zones) to third countries only as exports (see p. 32 of United Nations, 2008). 38 See p. 34 of United Nations (1998). This operation had a minor effect on total export volumes as sales to free zones/customs warehouses are negligible in most cases. The export volumes of Turkey, the country most affected by this operation, drop by 2-3 percent.
observations with a code that is not in that list. 39 Second, through a process of "consolidation" among the three HS classifications, we account for the modifications introduced in each of the HS revisions: a) two different codes with low trade volume were converted into a single code and b) an existing code with an increasing trade volume was split into various codes. Ignoring these modifications would create problems for the tracking of trade volumes for certain products over time. 40 The basic principle of consolidation is to identify the HS codes related to each other (e.g., codes that were split or merged with the modifications introduced by the HS2002 or the HS2007) and to replace them with a single code for the entire period. As a result of this consolidation. The final number of unique potential HS 6-digit codes in the consolidated classification is 4961. 41 In our cross-country raw dataset 4767 of those 4961 HS 6-digit codes are present. In terms of the array of products included in the cross-country raw dataset, we eliminate observations in HS Chapter 27 (hydrocarbons such as oil, petroleum, natural gas, coal etc.) given that we do not have exporter-level data on that chapter for important oil exporting countries such as Burkina Faso, Cameroon,Iran,Kuwait,and Yemen. 42 Regarding export values in the cross-country raw dataset, they are measured in US Dollars (USD) and they are Free on Board (FOB) figures, except for El Salvador and Senegal, whose export values represent Cost, Insurance and Freight (CIF) figures. 43 This difference should be taken into account for cross-country comparisons of measures related for example to the size of exporters.
In order to have a sense of the reliability of the raw data in each of the countries included in the Database, we apply two filters. For the first filter we compare the total values exported (excluding HS 27) calculated from the cross-country raw dataset with the total values exported from the United Nations' COMTRADE database (excluding HS 27) for every country and year. 44 This comparison yields quite different results for different countries. On the one hand, for Albania, Brazil, Cameroon, Chile, Colombia, Costa Rica, Ecuador, Guatemala, Kenya, Lebanon Mexico, Morocco, Pakistan, Peru, South Africa, Tanzania and Turkey, the ratio of total values exported in the raw dataset to total values exported in COMTRADE is about 100 percent. While for countries such as Mali and Yemen, the ratios indicate that total values exported in the raw dataset are as low as half of the total values exported in COMTRADE, for others such as Mauritius total values exported in the raw dataset are on average 30 percent above total values exported in 39 The number of HS categories included in the original classifications HS 1996, HS 2002, and HS 2007 Most of this elimination results from eliminating observations with a product code belonging to HS Chapter 99 as this is reserved for national use and the HS 6-digit codes under this chapter differ across countries. This elimination accounts for about 0.5 percent of total exports in our cross-country raw dataset. 40 See Cebeci (2012) for the methodology used in the consolidation. The paper along with a list of consolidated codes and concordances are available at http://econ.worldbank.org/exporter-dynamics-database. A similar process was used by Schott and Pierce (2012) to concord 10-digit United States Harmonized System codes between 1989 and 2007 and by Wagner and Zahler (2011) to homologate among 6-digit HS1992, HS1996, and HS2002 classifications. 41 The number of HS 6-digit codes not affected by the consolidation process is 4559, obtained as the total number of codes 6065 minus the 1104 codes that disappeared and minus the 402 codes whose content changed. 42 Possible reasons for this lack of data are confidentiality reasons or the fact that goods exported through pipeline are not recorded at customs but instead are recorded by other government/private institutions in the countries. 43 For countries for which export values were provided in local currency, they were converted to USD using the exchange rate series PA.NUS.FCRF taken from the World Development Indicators (whose original source is the IMF's International Financial Statistics) which is based on an annual average of daily exchange rates determined by national authorities or in the legally sanctioned exchange market. 44 Given our understanding about which transactions are included in the raw files, for this comparison we consider gross export figures in COMTRADE (which include re-exports) for Albania, Kenya, Mauritius, Tanzania and Senegal and net export figures (gross exports minus re-exports) for Jordan, Uganda and Yemen. For all other countries, it does not matter which COMTRADE figure we choose as they report either insignificant or no re-exports. COMTRADE. 45 For the second filter, we focus on the countries and years that would have been left out because of a unfavorable match with COMTRADE (below 60 percent) and we keep those countries (and years) where we observe internal consistency within the export totals calculated from the corresponding exporter-level raw dataset over time. 46 As a result of this second filter, we keep in the raw dataset the information for Macedonia, Jordan, Mali, Mauritius, and several years of El Salvador's data which would be left out due to an unfavorable match with COMTRADE data. 47 To protect the confidentiality of the firms in the raw datasets at the exporter-productdestination-year level, some of the measures in the files with product, destination, or HS 2-digit product-destination disaggregation are missing when the underlying country-product-year, country-destination-year, or country-product-destination-year cell includes a single firm whose individual information cannot be revealed. 48 For the measures of firm dynamics in the country-year dataset, of firm-sector dynamics in the country-sector-year dataset, of firm-destination market dynamics in the country-destination-year dataset, and of firm-sector-destination market dynamics in the country-sector-destination-year dataset that we use in the paper we consider the following definitions:  Exporter t : any firm that exports in year t;  Entrant t : a firm that does not export in year t-1 but exports in year t;  Exiter t : a firm that exports in year t-1 but does not export in year t;  Incumbent t : a firm that exports in both years t-1 and t;  Survivor t : a firm that does not export in year t-1 but exports in both years t and t+1.
The measures of dynamics are defined as follows:  Firm Entry Ratet = Number of Entrantst / Number of Exporterst ;  Firm Exit Ratet = Number of Exiterst/ Number of Exporterst-1 ;  Firm Survival Ratet = N. Survivorst / N. Entrantst . 45 The Appendix in Cebeci et al. (2012) provides detailed results on these comparisons. 46 Although both sources of trade data, COMTRADE as well as the exporter-level raw datasets that we use, originate from customs authorities, we are aware of potential difficulties in the processing of the exporter-level raw datasets that may justify the differences sometimes observed between the export totals obtained from the exporter-level raw datasets and the export totals available in COMTRADE. One of the potential reasons for these difficulties relates to the manual registration of export transactions that still takes place in some countries and may result in under-recording of export transactions for some countries and years. 47 For countries such as Macedonia for which the match with export totals in COMTRADE in the first few sample years is relatively poor, the match improves over time and this is likely due to a shift from manual to digital registration of export transactions. 48 For such cells the number of exporters shows a value of 1 and most measures except those based on exiter firms in Appendix Table 1 are missing. The developed countries in our sample applied their own confidentiality rules (often stricter) as detailed in Cebeci et al. (2012). A file with the percentage of total exports corresponding to the hidden values by country and year, for the developing countries for which we have the raw exporter-level data is available from the authors upon request.

Appendix 2 Table A2.1 Average Exporter Size and Country Characteristics -Sample of Continuing Exporters
Notes: Robust standard errors in parentheses, clustered at the country level. *** and ** indicate significance at 1% and 5% confidence levels, respectively. The regression in column (1) is based on measures at the country-sector-year level for years 2004-2008 (Brazil and New Zealand are excluded due to the lack of measures at that level); that in column (2) is based on measures at the country-destination-year level for years 2004-2008 (Brazil and New Zealand are excluded due to the lack of measures at that level), and those in column (3) are based on measures at the country-sector-destination-year level for years 2004-2008 for the 34 developing countries for which we have access to the raw exporter-level data (the list is provided in Table 1) as well as Norway, Portugal, and Spain. Kuwait is excluded from all regressions due to lack of data in the 2004-2008 period.  Table A2.