 [Seite 63↓] 
The main goal of the International Organization for Standardization (ISO) is to harmonize standards around the world, which, as it is widely claimed, promotes trade and therefore global welfare. A prominent example of the work done by ISO is the ISO 9000 family of standards, often referred to as generic quality management standards. The vision of the developers is that
(…) through its worldwide acceptance and use, the ISO 9000 family of standards will provide an effective means for improving the performance of individual organizations and providing confidence to people and organizations that products (goods and services) will meet their expectations thereby enhancing trade, global prosperity and individual wellbeing.^{33} 
Critics of ISO 9000 claim, however, that it is merely a barrier to market entry and a tariff on international trade. There are valid arguments on both sides. On the one hand, ISO 9000 might be a common language, which lowers informational asymmetry between firms and allows them to organize trade more efficiently. Indeed, the standards emphasize clear and open communication with customers, as well as with suppliers.^{34} Furthermore, they provide a tool facilitating screening and performance evaluation. Learning this common language offers [Seite 64↓]an alternative for establishing vertical relations based on longterm relationship and brand reputation. On the other hand, ISO 9000 has been used as a standard against which to assess performance in government procurements and in setting of minimum quality requirements for imports. This raises a concern that the standard is mainly a tool for protecting domestic markets.
This paper empirically investigates the impact of ISO 9000 on international trade. We estimate a gravity equation for bilateral trade incorporating ISO 9000 adoptions in each country as a factor affecting bilateral trade barriers. As it has been pointed out in the literature, the causality might go both ways. International trade might benefit from standards’ harmonization, as trade barriers decrease, and standardization process might in turn be determined by intensity of foreign trade, which indicates openness of an economy ( see e.g. Casella, 1996; Moenius, 2000). As a consequence, empirical models of international trade using standardization as an explanatory variable may suffer from endogeneity bias. Formal tests, however, do not reject strict exogeneity of ISO 9000 adoptions in our gravity equation. Nevertheless, we estimate ISO 9000 diffusion equation to obtain additional insights into the role of ISO 9000 for international trade.
The empirical literature investigating the impact of common standards on trade is scarce. In particular, we are not aware of any study that investigates the performance of ISO 9000 in this context. Few existing empirical analyses of ISO 9000 focus on managers’ motivation to seek the certification and on market reaction to it. Examples include Anderson et. al. (1999), who find that after controlling for regulatory and customer pressures, providing credible signals of quality assurance to external parties motivates the adoption decision. Further, Docking and Dowen (1999) examine the reaction of the firms’ stock price to the announcement of ISO 9000 registration. They find that, for the smaller firms, investors react positively to the announcement and that there was no significant reaction for the larger firms.
The work by Moenius (2000), who looks at the impact of countryspecific and bilaterally shared product and process standards on international trade, is probably closest related to our work. He finds that both countryspecific and shared standards are favorable to trade. Similarly, Swann et al. (1996) report that both international and countryspecific product standards promote imports into the U.K. Further, Blind (2001) analyses Switzerland’s trade of measurement and testing products with Germany, France, and the U.K. He finds that the stock of both national and international standards in this sector has a positive impact on [Seite 65↓]the trade flows. In turn, Blind (2002) investigates factors responsible for intensity of standardization in 20 industrial sectors of seven countries. He reports empirical evidence on the positive relation between the stock of national and international standards in a sector and the ratio of exports to total production in that sector.
To a great extent, our work relates also to the strand of literature considering the role of networks in reducing information costs associated with international trade (e.g. Rauch, 1999; Rauch and Trindade, 2002). In our view, the role of ISO 9000 for international trade very much overlaps with the role of ethnic Chinese networks for trade studied by Rauch and Trindade (2002). To the extent that ISO 9000 lowers information and search costs, it also relates to the role of Internet for trade studied by Freund and Weinhold (2004). Neither of these studies, however, discusses the potential endogeneity of networks’ formation.
The rest of the paper is organized as follows. Section 4.2 describes the ISO 9000 family of standards and its role in international trade more in detail. Section 4.3 provides theoretical models of bilateral trade flows and standard diffusion, which guide our empirical analysis. Data, empirical implementation of the theoretical models, and discussion of the results are presented in section 4.4. Section 4.5 concludes.
The ISO 9000 family of standards is often referred to as generic quality management standards. They are generic in the sense that they can be implemented by any organization regardless of its size, the sector of its activity, and its managerial or national culture. Quality management reflects what the organization does to enhance customer satisfaction by meeting his/her requirements and expectations.^{35} Compliance with ISO 9000 indicates consistent use of documents and standardized procedures to produce a good or service, for which the customer contracts. In other words, ISO 9000 certifies that the firm’s products conform to the specification.
The history of ISO 9000 started in 1987 with publication of the ISO 9000 Quality Assurance Standards by a Technical Committee (TC 176) of the International Organization [Seite 66↓]for Standardization (ISO). By the end of 2001, the number of ISO 9000 certificates exceeded half a million in 161 countries around the world, contributing to its reputation as an international reference for quality requirements in businesstobusiness dealings.^{36} We treat ISO 9000 as a uniform standard although it consists of a series of nested standards, which evolved over time. Originally, the core members of the family, with which firms could actually be certified, were ISO 9001, ISO 9002, and ISO 9003. They differed in terms of the quality system elements they covered. The nested nature of these standards allowed firms to accommodate differences in the scope of their operations.^{37}The 2000 edition of the ISO 9000 family replaced these three standards with a single one labeled ISO 9001: 2000. As supplementary standards, the 2000 edition included ISO 9000: 2000, which describes fundamentals and specifies vocabulary for a quality management system, and ISO 9004: 2000, which provides guidelines for performance improvements. Both of them were developed on the basis of previous standards, which they replaced. Given that the core members of the ISO 9000 family were finally replaced by a single one, our simplifying assumption treating ISO 9000 as a uniform standard seems justified.
ISO 9000 adoption is a sovereign decision of each firm, however, they can seek certification only in their home countries. Each country has one governmentdesigned accrediting agency, which certifies the competency of third party registrars to conduct ISO 9000 quality audits. The registrars are also charged with issuing of certificates.^{38} In general, motivations behind the implementation of ISO 9000 could be divided into three main categories: i) compliance with government regulations, ii) entering new form of vertical relations due to use of a common language, and iii) internal efficiency gains. In fact, all factors influencing managers to seek ISO 9000 certification that Anderson et al. (1999) identify after a comprehensive review of practitioners journals fall into on of the three categories.
The first category stems from the fact that ISO 9000 has been used as a standard against which to assess performance in government procurements and in setting of minimum quality requirements for products that affect public safety. The Single Market Initiative of 1992 initiated by the European Community involves the most noticeable example of such regulations. The public safety argument obliged firms to attain a uniquely designed EC Mark [Seite 67↓]in order to get access to certain markets.^{39} ISO 9000 was selected as a means to attain the mark in most of the cases. This raises the concern that ISO 9000 can be a barrier to market entry and a tariff on international trade. We relate our empirical findings to potential effects of the regulations.
The second category of motivations is the focus of this study. As noted by Bénézech et al. (2001, p. 1396), “the ISO 9000 series can be viewed as a code, a language used by firms to extend their industrial relationship”. Thanks to standardized documentation flow and organizational procedures within certified firm, ISO 9000 provides a screening device that allows other firms to observe and to evaluate its performance. This naturally lowers informational asymmetries between firms. Consequently, ISO 9000 proxies for conformance of the firm’s product to the specification, for which the customer contracts. This leads to lower transaction and search costs in vertical relations between firms. To realize the benefits of ISO 9000, however, both contracting parties should have adopted (i.e. learned) it in the first place. This is why the common language analogy is appropriate.^{40} Learning this common language could be viewed as an alternative for establishing vertical relations based on longterm relationship and brand reputation. This explains the potential of ISO 9000 for reducing barriers to market entry and nontariff barriers to trade.
Finally, firms seek ISO 9000 certification to realize efficiency gains. The discipline of documentation and organizational procedures could reduce waste, lower costs, and improve productivity. For example, relying on a survey of ISO 9000 certified firms in the U.S. (sales from $100 million to $500 million), Anderson et al. (1999) report average annual savings of $200 000 due to the certification.
The same authors further report that obtaining ISO 9000 certification at a manufacturing site in the U.S. takes from 9 to 28 months and approximately 3540% of all sites fail the first audit. The costs of the standard adoption and certification are substantial. A medium size manufacturing facility employing 100 people can expect to spend $50 000. For larger firms (sales from $100 million to $500 million), the average cost that the authors report is $300 000. We relate to these efficiency gains, costs, and timing of adoption modeling ISO 9000 diffusion in section 4.3.2.
 [Seite 68↓] 
The standard empirical framework used to predict international trade flows is the gravity equation. In a simple form, which explains its name, the equation reads
where V_{ij} is value of exports from country i to country j, Y_{i} and Y_{j} are their economic masses often measured by GDP or GNP, D_{ij} is a measure of the distance between them and A is a constant of proportionality.^{41} Attractiveness of the gravity equation originally stems from its empirical explanatory power. Recent developments, however, show that the equation can also be theoretically motivated, in particular, in the context of classical HeckscherOhlin framework, as presented by Deardorff (1998).
The following discussion is going to selectively overview the theoretical foundations of the gravity equation in order to facilitate selection of the explanatory variables, as well as to help the interpretation of our empirical results.
Suppose first that preferences of consumers in every country are identical and homothetic. Suppose further that trade is balanced – so that each country’s expenditures equal its income – and frictionless, i.e. there are no transport costs and no other impediments to trade. Then, as shown by Deardorff (1998), the value of i's total exports to j is
where Y^{w} stands for world income. Thus, an even simpler gravity equation than (4.1) emerges with constant of proportionality A = 1/Y^{w}. Deardorff (1998) further shows that if preferences are not identical and /or not homothetic, then bilateral trade flows are on average given by the simple frictionless gravity equation (4.2) under some additional regularity conditions. He [Seite 69↓]argues also that the tendency of highincome consumers to consume larger budget share of capitalintensive goods will lead highincome capitalabundant countries to trade more than average with each other and less than average with lowincome laborabundant countries. This would motivate inclusion of per capita income in the gravity equation, as it is often the case in empirical literature.
Now, we turn to the more realistic case where trade is impeded. That is, we allow for transportation costs, tariffs and other nontariff barriers to trade that increase the price of domestically produced goods in foreign markets. A particularly elegant gravity equation for this case is derived by Anderson and Wincoop (2003). In their model goods are differentiated by the place of origin and each region is specialized in the production of only one good. Moreover, preferences in each region are identical and homothetic, approximated by a constant elasticity of substitution (CES) utility function. Their gravity equation then reads
where σ is the elasticity of substitution between all goods in the CES utility function. The trade barriers between i and j are captured by the trade cost factor t_{ij} such that the price of region i goods for region j consumers equals p_{i}t_{ij} and p_{i} denotes the exporter’s supply price, net of trade costs.^{42} ^{43}P_{i} and P_{j} are the consumer price indices of i and j. The key implication of the model is that trade between regions is determined by relative trade barriers, as the price indices {P_{i}} depend on all trade cost factors {t_{ij}}. Unfortunately, empirical implementation of the model is troublesome. As discussed by the authors, the price indices {P_{i}} are not observable, since they cannot be interpreted as consumer price levels.^{44} Moreover, the authors claim uniqueness of their implicit solution for the price indices only for the case with symmetric trade cost factors.
Given the difficulties of the structural approach, we are going to model the impact of ISO 9000 adoptions on trade in a semistructural way. As we have already argued, ISO 9000 [Seite 70↓]could be understood as a common language, adoption of which allows firms to lower the transaction and search costs. We will assume that, by lowering these costs, adoptions of the standard in both country i and country j will decrease their bilateral trade barriers. Section 4.4.2 on empirical implementation of the model presents that more in detail.
The main hypothesis, which drives our adoption process, is that ISO 9000 exhibits significant network effects. Network effects naturally arise in this context, as in the context of any other language. The number of ISO 9000 adopters determines the size of the pool of potentially more efficient business contacts, hence the value of the standard for each adopter. Since we are interested in the link between ISO 9000 and international trade, the relevance of foreign adoptions of the standard for the home country adopters is of crucial importance. In particular, we would like to test whether economic distance between countries in terms of trade related factors matters for the relevance of foreign adoptions. Finding such relationship would give us additional insight about the role of ISO 9000 in international trade. The model developed in the second chapter of this thesis facilitates such analysis. It allows us to derive structural countryspecific diffusion equations from adoption decisions of individual firms in each country. The model in chapter 2, however, is formulated to study the adoptions of competing network goods by consumers in one country or region. With ISO 9000, we face the situation where a single network good is supposed to be adopted by consumers in various countries. Therefore, we need to slightly reformulate the model.
We assume that in each country i = 1, 2, …, I there is an infinite number of heterogeneous firms, which instantaneously decide whether to adopt ISO 9000 or not. The adoption decision is influenced by the firmspecific intrinsic valuation of the standard, denoted by v_{i}, which corresponds to the efficiency gains that the firm realizes after the adoption. Another factor influencing the adoption decision is the network size of certified firms at time t denoted by x_{i}(t). Network effects arise in the adoption of ISO 9000 due to potentially more efficient contracting among the certified firms. The efficiency gains and network effects together shape the net instantaneous benefits of ISO 9000 adoption, which for simplicity takes the following functional form
[Seite 71↓]
(4.4)
where c and d are parameters that capture the extent of network effects. The costs of ISO 9000 adoption consist of a sunk investment in the reorganization of the firm q _{i} and an instantaneous audit fee p _{i}, both of which are assumed to be constant over time.
Firms maximize the net benefits adopting ISO 9000 when the present value of the stream of future benefits exceeds the present value of the costs. As shown in appendix 4.6.1, we can calculate the netofcost intrinsic valuation of indifferent firm in country i at time t, denoted as , from the following first order condition
The lag of network size δ in (4.5) is crucial for the dynamic properties of the model. It can be motivated by excessive optimism of the firms regarding time needed to implement the standard in their own sites. In fact, as we point out in section 4.2, the adoption process takes time and large share of firms fails the first audit.
Assume that the netofcost intrinsic valuations in each country – – are uniformly distributed over an interval (∞, a_{i}] with some density b_{i}. Integrating over all firms with intrinsic valuation higher than , we obtain the equilibrium number of adopters in each country
The final step in setting up the model is to define the relevant network for ISO 9000 adopters in each country. We are not going to differentiate the firms in a given country in terms of their cooperation prospects with each other. Instead, we want to emphasize the difference between the foreign and the domestic firms. Therefore, we define the network size of ISO 9000 adopters as
 [Seite 72↓] 
where w_{ij} reflects the relative importance of country j adopters for country i adopters. Since we expect that foreign markets are relevant for the adoption decisions of domestic firms, we are going to relate w_{ij} to bilateral trade between i and j in the next section. The general idea is that the intensity of trade indicates closeness of the economies, hence the relevance of foreign firms as business partners for the domestic firms.
Finally, substituting (4.7) in (4.6), we arrive at the following adoption equation
which guides the empirical analysis in section 4.4
Data on ISO 9000 adoptions comes from ISO (2002). Bilateral exports are taken from the UN Commodity Trade Statistics Database (Comtrade) and GDP and population figures come from the International Financial Statistics published by the IMF. The data ranges over 19952001 and covers 101 countries. Table 4.1 lists the countries included in this study together with the number of firms that adopted ISO 9000 standard till the end of 2001 in each of the countries. Summary statistics of the variables used in the estimations are given in table 4.2.
A distinctive feature of this study is its particularly wide coverage of countries. In fact, our sample covers approximately 80% of the world trade and 99% of the world ISO 9000 adoptions. A primary reason for this is that the adoptions of ISO 9000 in each country depend on the world diffusion of the standard, as predicted by equation (4.8). Therefore, the smaller the coverage of the sample is, the more sever is the concern about omitted variable bias in the estimates. Additionally, inclusion of the less developed countries, for which the trade barriers aspect of ISO 9000 is potentially more severe, is important for the generality of our results. In [Seite 73↓]the context of gravity equation for trade, the wide coverage of countries might seem problematic. The theories of trade in imperfect substitutes, which were the first to justify the gravity model, were thought to apply only to the industrialized countries. However, as found by Hummels and Levinsohn (1995), the model works equally well for the larger set of countries.^{45}
In our empirical implementation of gravity equation, we apply panel data techniques, which have the advantage over crosssection estimations, that they can capture all time invariant trade determinants by means of countrypair specific effects. This is very useful, since the trade barriers in gravity models are usually difficult to quantify, as they might consist of tariff barriers, transportation costs, information costs, etc., some of which are not even observable. Among many variables proposed by researchers to approximate the trade barriers, there are geographical distance, linguistic and colonial ties, membership in trade agreements and monetary unions, and common border. To the extent that these variables are time invariant, which is very likely given the relatively short time span of our data, the countrypair specific effects will account for them. A particular ingredient of the trade barriers, we consider, are search and transaction costs. According to the main hypothesis of this paper, they can be lowered for firms that invested in learning of the common language – ISO 9000. Following other authors we assume that the trade barriers can be approximated by a loglinear function. Then, our specification of the trade barriers in (4.1) reads^{46}
According to (4.9) the lack of ISO 9000 awarded firms in both countries i and j simplifies the measure of the bilateral trade barriers to a function of countrypair specific effect. Since, we do not impose symmetry on the trade barriers, η_{ij} is in fact importerexporter specific. This asymmetry is going to be important later on, because it allows us to distinguish the distance to [Seite 74↓]foreign customer (importer) from the distance to foreign supplier (exporter) from the domestic firm viewpoint. Unilateral adoptions of ISO 9000 affect average trade barriers only marginally.^{47} Bilateral adoptions multiply this effect, as follows from our interpretation of ISO 9000 as a common language. For the sake of generality, we do not restrict the parameters δ_{1} and δ_{2} to be equal, although this would fit our commonlanguage hypothesis.
Then, after substituting the measures of economic masses and the constant of proportionality in (4.1) with specific functions of the observables, the gravity equation that we estimate becomes
where X_{ijt} denotes exports from country i to country j in year t.^{48} We allow the parameters on countries’ GDP – β_{1} and β_{2} – to be different from one and from each other. We also include POP variable, which measures the countries’ population in millions. The reason for that is to capture the tendency that reach countries trade more than average with each other, as explained in section 4.3.1.^{49} Finally, λ_{t} stands for time effects, which are meant to capture changes in the world income, and ε_{ijt} is a usual i.i.d. error term.
Now, we turn to the empirical implementation of the ISO 9000 adoption equation (4.8). As already mentioned, we are mainly interested in the relevance of international markets for the adoption decisions of domestic firms. Our assumption, which follows from the commonlanguage hypothesis, is that ISO adoptions in close economies reinforce each other relatively more than ISO adoptions in distant economies. There are two natural candidates for measuring this economic closeness/distance; intensity of bilateral trade and bilateral trade barriers. In the context of gravity equation (4.10), the difference between the two is that intensity of bilateral trade depends on the countries’ economic masses and some [Seite 75↓]unobserved factors captured by the error term on top of the bilateral trade barriers. We assume that the measure of economic distance relevant for the ISO 9000 adoptions are the bilateral trade barriers. In other words, we assume that the relevance of a foreign ISO adoption for domestic firms depends neither on the size of the foreign economy in which the adoption took place, nor on the other unobserved factors captured by the error term in (4.10). Our specification of the economic closeness then reads
where w_{1} and w_{2} are some constant weights and {η_{ij}} are the importerexporter specific effects in (4.10). Since η_{ij} reflects the barriers for exports of i to j and η_{ji} reflect the barriers for imports of i from j, specification (4.11) distinguishes the distance to foreign customers – weighted by w_{1} – from the distance to foreign suppliers – weighted by w_{2}. In fact, estimation of the parameters w_{1} and w_{2} drives our interest in the ISO 9000 adoption equation. Positive values of w_{1} and/or w_{2} would suggest that the firms’ adoption decisions are indeed affected by the number of potential foreign customers and/or suppliers certified with ISO 9000.
We also need to relate countryspecific parameters of the types’ distribution a_{i} and b_{i} to some observables in order to avoid estimation of excessive number of parameters in (4.8). One could expect that b_{i}, which reflects the number of firms with a given efficiency gains prospects from ISO 9000 adoption, depends on the total number of firms in country i, which in turn positively correlates with the GDP of that country. Similarly, one could argue that a_{i}, the maximum realizable efficiency gains across firms in country i, depends on the country’s GDP. The rationale is that the efficiency gains due to ISO 9000 adoptions increase with the firm size – so a_{i} most probably reflects the efficiency gains of the largest firms in the country – and the world largest firms are located in the richest countries. We are going to relate a_{i} and b_{i} to the country’s GDP one at a time to keep the model as simple as possible. This leads us to the following specifications
and
 [Seite 76↓] 
Then, after applying specifications (4.11) and (4.12) and substituting theoretical values in the equation (4.8) with observables, the ISO 9000 adoption equation that we estimate becomes
where the variables ΣISO^{C} _{i(t1)} and ΣISO^{S} _{i(t1)} are the indices of foreign customers’ adoptions and foreign suppliers’ adoptions respectively.^{50} The indices make use of the asymmetry in the importerexporter specific effects , which reflect the asymmetry in bilateral trade barriers. They are defined as and , where are the estimates of. ψ_{it} is an i.i.d. error term, which capture the influence of unobserved factors. Alternatively, applying specification (4.12’) instead of (4.12), we obtain
Note that ω_{1} replaced w_{1} and ω_{2} replaced w_{2} in both equations (4.13) and (4.13’). This is because the importerexporter specific effects cannot be identified separately from the constant α in equation (4.10) without additional assumptions. To obtain we will arbitrarily assume that, i.e. we will normalize the smallest netofISO9000 measure of trade barriers in our sample to 1.^{51} In case of any such normalization, the relation between w_{k} and ω_{k} can be shown to be, where Δ is some unknown [Seite 77↓]constant. This means that we are still able to make some inference about w_{k} having estimated ω_{k}. In particular, they have the same signs and the ratio of ω_{1} and ω_{2} equals the ratio of w_{1} and w_{2}.
The assumption, which gives rise to the specification (4.11), that the bilateral trade barriers are the relevant measure of economic distance in the context of ISO 9000 diffusion, has an important implication for the econometric treatment of our model. It implies that ISO 9000 adoptions and bilateral trade are not interdependent. So, we avoid the necessity of estimation simultaneousequation model. An indirect way to test this assumption is to test the exogeneity of ISO variables in (4.10). If they would indeed depend on the bilateral trade itself rather than on bilateral trade barriers, the test should fail.
First, we estimate (4.10) by fixed effects (FE). Typically, researchers report FE estimation results along with randomeffects (RE) results. The advantage of FE estimation over RE estimation is that consistency in the former does not rely on orthogonality between the countrypair specific effects η_{ij} and all the other explanatory variables.^{52} We skip the RE estimation, since we expect the adoptions of ISO in country i to depend on the economic distance to each trade partner indicated by η_{ij}. The FE estimation results are presented in table 4.3.
The first three columns (1) – (3) contain results of the regressions, in which we exclude some of the explanatory variables and column (4) corresponds to the regression with the full set of covariates. We see that estimated coefficients vary only marginally across these different specifications. In general, we find the coefficient on countries’ own income β_{1} and the coefficient on partner’s income β_{2} to lie about 0,3 and 0,7 respectively. Moreover, the coefficients on population γ_{1} and γ_{2} tend to be of the same magnitude with reversed sign as income coefficients. These estimates imply that, in contrast to theoretical prediction and the results from crosssectional studies (see e.g. Frankel et al., 1997, table 4.2 and table B6.6), trade is almost entirely driven by countries’ income per capita. Glick and Rose (2002), who also apply FE estimation, report findings, which are similar to ours (see table 4 in there).
 [Seite 78↓] 
The next four regressions (5) – (8) in table 4.3 augment the first four regressions by inclusion of the leading explanatory variables, as suggested by Wooldridge (2002, p. 285), in order to test the strict exogeneity assumption. The null hypothesis of the Wald test reported in table 4.3 is that coefficients on the leading explanatory variables equal zero. We see that the Wald rest rejects the null – and thereby strict exogeneity of the explanatory variables – in all four cases at very high significance. Without strict exogeneity, FE estimator becomes inconsistent, so the estimates of the gravity equation coefficients in table 4.3 can be misleading.
We believe that the reason why strict exogeneity failed in the FE estimations is that macroeconomic indicators like exports and GDP tend to be trending variables. This is also true for population and ISO adoptions variables. As it is well known from the time series literature, simple OLS is likely to deliver spurious correlations between trending variables. The most straightforward remedy for that is to use first differencing (FD) estimation.
Table 4.4 reports the results of the same exercise, as in table 4.3, using FD estimation. Again, we see that the estimated coefficients of the gravity equation are stable across different specifications. However, the estimates changed in comparison to the FE results. The most noticeable change concerns population variables. The coefficient on country’s own population γ_{1} changed sign to positive and the coefficient on partner’s population γ_{2} decreased in magnitude. Both coefficients are now statistically insignificant, which is in contrast to the findings of crosssectional studies. The reason for this discrepancy might be that the intended effect of countries’ income per capita on trade is already capture by the countrypair specific effects η_{ij}. This is in line with the Anderson and Marcouiller’s (2002) alternative explanation of why capitalabundant countries trade disproportionately with each other. They argue that the reason are strong institutions to support trade security, which can be plausibly treated as constant in our data.
Given the interpretation, that countryspecific effects absorb the impact of per capita income, the coefficient on partner’s GDP in table 4.4, which equals roughly 0,77, corresponds well to the findings of crosssectional studies. The coefficient on own GDP is however about 10 times smaller and only marginally significant.^{53} The most important indicator, which gives us some confidence about plausibility of our results, is the Wald test. It does not reject strict [Seite 79↓]exogeneity of the covariates in first differences, at least in the regressions without population variables (column 13 and 15 in table 4.4).
Finally, we turn to the impact of ISO9000adoptions’ variables, which are the focus of our study. They FD estimates are lower than the FE estimates, but still significant at 5% level. The coefficients on own and partner’s adoptions equal 0,027 and –0,026 respectively. This means that 10% increase in the number of firms awarded with ISO 9000 certificate in a country leads on average to 0,27% increase in bilateral exports and 0,26% decrease in bilateral imports of that country. These results provide an empirical evidence for the role ISO 9000 plays in international trade, although they are not fully in line with our expectations. The common language hypothesis, as we stated it, suggests that bilateral trade flows should rise with both, exporter’s and importer’s, adoptions, since they both contribute to the number of potentially more efficient business links. However, this line of argument does not take into account the possibility of substitution between suppliers, i.e. exporters in this case. Freund and Weinhold (2004) argue along the same lines interpreting their findings on impact of the Internet on bilateral trade. Also, Anderson and Wincoop (2003) point to the fact that bilateral trade flows depend on trade barriers between all trading parties.^{54} In our specification of the gravity model, concentration of ISO 9000 adoptions in few countries around the world could explain falling average bilateral imports with the number of adoptions. In fact, the OECD members (30 out of 101 countries in our sample) account for 75% of ISO 9000 adoptions in table 4.1.
To explore further the substitution effect of ISO 9000 on international trade we restrict our sample to the OECD countries and repeat the FD estimations. The results are reported in table 4.5. Again, the population variables proved to be insignificant and cause endogeneity problems. Coefficients on own and partner’s GDP are higher than for the whole sample roughly by 0,1 and 0,2 respectively and highly significant. Coefficients on own and partner’s ISO 9000 adoptions are also significant and amount to 0,063 and 0,036 respectively. This supports the substitution effects hypothesis. Bilateral exports between ISO 9000 abundant countries indeed increase with both domestic and foreign adoptions. These results, however, should be treated with caution, because the Wald test rejects strict exogeneity even in the regressions without population variables (columns 21 and 23 in table 4.5). Now, the GDP variables are responsible for the rejection. There are good reasons to believe that the [Seite 80↓]endogeneity of GDP in these regressions is merely a statistical phenomenon and that it does not significantly bias the results. First, the inclusion of variables in levels does not significantly change the coefficients on variables in differences. Second, Hummels and Levinsohn (1995) report that correcting for the endogeneity of GDP with instrumental variables makes very little difference.
As we mentioned in section 4.2, critics of ISO 9000 reasonably argue that the standard is actually a barrier to trade, since it has been used as a tool for introducing import restrictions. In fact, the positive effect of domestic ISO 9000 adoptions on exports, that we found, could be explained by increasing access to the regulated markets. However, the impact of domestic adoptions on imports, that we found, cannot be explained by the trade barrier hypothesis. In particular, the finding that imports increase with domestic ISO 9000 adoptions within OECD countries cannot be supported by this hypothesis. The reversed relation in the full sample might be due to the use of ISO 9000 in introducing import restrictions, however, under additional assumption that the restrictions increases with domestic ISO 9000 adoptions.^{55}
Estimation results of the ISO 9000 diffusion equation provide us additional insights into the link between the standards and international trade. To estimate (4.13) and (4.13’) we follow the general method of moments (GMM) approach for linear dynamic panel data models, which was proposed by Arellano and Bond (1991). By doing so, we allow for additional unobserved heterogeneity on top of (4.13) and (4.13’). This unobserved heterogeneity accounts for institutional factors, like the national accrediting agencies, which could either spur or hamper the diffusion process.
To obtain linear models we multiply out the terms in the structural equations (4.13) and (4.13’). By multiplying out the terms, the parameters ω_{1} and ω_{2} – the transformed weights in our measure of economic closeness (4.11) – become overidentified. This allows us to test the underlying structure of the empirical diffusion equations. The estimated coefficients of the linearized equations (4.13) and (4.13’) are reported in table 4.6, in columns (25) and (25’) respectively.^{56} Except for the lagged dependent variable, all the regressors are [Seite 81↓]treated as strictly exogenous. We see that both regressions perform reasonably well in statistical sense, as indicated by the Sargan and the ArellanoBond test statistics at the bottom of the table. However, the underlying economic structure of both regressions (25) and (25’) is rejected by the data. According to the structural equation (4.13), the ratio of coefficients on ΣISO^{S} _{i(t1)} and ISO_{i(t1)} in the regression (25) should be equal to the ratio of coefficients on GDP_{it}ΣISO^{S} _{i(t1)} and GDP_{it}ISO_{i(t1)}, because both ratios identify the same parameter ω_{2}. Yet, the ratios have opposite signs. Similarly, the equation (4.13’) predicts that the ratio of coefficients on (ΣISO^{C} _{i(t‑1)})^{2} and ISO^{2} _{i(t1)} in the regression (25’) should be equal to the squared ratio of coefficients on ΣISO^{C} _{i(t1)} and ISO_{i(t1)}, since they identify (ω_{2})^{2}. But, the first ratio is negative.^{57} Therefore, we are going to treat the results as coming form reduced form approach and limit the discussion to pointing out major correlation patterns.
The results of the regressions (25) and (25’) suggest in general that ISO 9000 adoptions in each country are positively related to the market size measured by the country’s GDP. The adoptions exhibit also significant inertia, as indicated by the coefficients on the lagged dependent variable. Since we are mostly interested in the foreign trade considerations in the adoption decisions of firms, the crucial result in table 4.6 are the coefficients on the indices of foreign customers’ adoptions and foreign suppliers’ adoptions. We find that domestic ISO 9000 adoptions are positively related to foreign customers’ adoptions, as expected. In contrast, foreign suppliers’ adoptions are not or weakly negatively correlated with domestic adoptions depending on the specification. In other words, the diffusion of ISO 9000 seems to proceed from customers to suppliers and not the other way around, at least in the international context. This asymmetry does not necessarily contradict our common language hypothesis. It might reflect the fact that business customers are able to benefit from ISO 9000 without actually being certified, as someone, who understands a language without having passed exams that certify that. At the same time, certified suppliers might not care about certification of the customers, if their relations are already established.
To check the robustness of these findings, we perform additional two regressions (26) and (26’), in which the indices of foreign ISO 9000 adoptions in (25) and (25’) are replaced with simple unweighted sum of foreign adoptions. The coefficients on the unweighted foreign adoptions are much less significant than those on the indices of foreign adoptions in the previous regressions. In other words, our indices are much better predictors of domestic [Seite 82↓]adoptions than is the unweighted sum. This gives us some confidence in the effects of foreign trade considerations on domestic adoptions we found in (25) and (25’).
In this paper, we empirically assess the link between the ISO 9000 family of standards and international trade. According to the vision of its developers, ISO 9000 should provide confidence to people and organizations that products will meet their expectations, thereby enhancing trade and global welfare. In contrast, its critics claim that it is merely a barrier to market entry and a tariff on international trade.
Our modeling strategy is to look at the impact of ISO 9000 adoptions on bilateral trade flows between countries. We estimate a gravity equation for bilateral exports using data on 101 countries over 19952001. To obtain additional evidence, we estimate an international ISO 9000 diffusion equation and test whether the number of certified foreign trade partners plays a role in the domestic firms’ adoption decisions.
Using the full sample, we find that domestic ISO 9000 adoptions spur bilateral exports and hamper bilateral imports. In the sample restricted to the OECD countries, however, the domestic adoptions are positively related to both bilateral exports and imports. Additionally, we find empirical evidence on the positive effect of foreign customers’ adoptions of ISO 9000 on domestic adoptions.
In general, these findings suggest that the ISO 9000 standards have indeed significant positive impact on international trade. They are consistent with the common language hypothesis, which states that ISO 9000 lowers informational asymmetry between firms and allows them to organize vertical relations more efficiently. The negative effect of ISO 9000 adoptions on bilateral imports in the full sample could be explained by the substitution effect. Since domestic certified firms are likely to choose other certified firms as suppliers, the average bilateral imports from all trade partners will fall with domestic adoptions if the concentration of adoptions around the world is high enough.
The hypothesis that ISO 9000, as a tool for introducing import restrictions, is a barrier to international trade is not able to explain our empirical findings, although we cannot reject [Seite 83↓]it. The substitution effect we pointed out might be, however, another way, in which ISO 9000 constitutes an effective trade barrier. If the sluggish diffusion of ISO 9000 in the less developed countries continues, the benefits of ISO 9000 envisioned by its developers will remain in the developed countries’ domain.
 [Seite 84↓] 
We assume that once the standard is adopted, it yields the infinite stream of future benefits. The present value of the stream of benefits can be written as
where ρ is a common instantaneous discount rate. The cost of adoption consists of the sunk investment in reorganization of the firm q _{i} and the instantaneous audit fee p _{i}, both of which are assumed to be constant over time. We write the present value of the cost as
The optimal timing of adoption involves solving the following maximization problem
Assuming that problem (4.16) is concave, we obtain the intrinsic valuation of indifferent firm v _{i,t} ^{*} as the solution to
which simplifies to
Now, substituting benefit function (4.4) in (4.18) we finally obtain
where denotes the netofcost intrinsic valuation of indifferent firm. Further, we modify solution (4.19) introducing a lag δ into the network size. The lag is crucial for identification in this model, as explained in chapter 2. In the context of ISO 9000 adoption [Seite 85↓]decision, the lag can be motivated by an excessive optimism of firms regarding the time needed to implement the standard. The fact that firms obtain certification later than implied by (4.19) leads to the modified first order condition
Note that deriving (4.19) we assumed that the firms have perfect knowledge about the future diffusion of network. This assumption carries over to (4.20). In other words, the firms know that the others are overoptimistic.
 [Seite 86↓] 
Country/Region 
# ISO 9000 certifications Dec. 2001 
Country/Region 
# ISO 9000 certifications Dec. 2001 
Country/Region 
# ISO 9000 certifications Dec. 2001 
Argentina 
2324 
Guatemala 
18 
Panama 
33 
Australia 
26750 
Honduras 
11 
Paraguay 
46 
Austria 
4000 
Hong Kong 
3814 
Peru 
200 
Bahrain 
59 
Hungary 
6362 
Philippines 
961 
Bangladesh 
38 
Iceland 
30 
Poland 
2622 
Barbados 
11 
India 
5554 
Portugal 
2474 
Belarus 
78 
Indonesia 
1395 
Qatar 
52 
Belgium 
4670 
Iran 
618 
Romania 
1670 
Belize 
4 
Ireland 
3700 
Russian Federation 
1517 
Bolivia 
42 
Israel 
6447 
Senegal 
8 
Botswana 
5 
Italy 
48109 
Seychelles 
6 
Brazil 
9489 
Japan 
27385 
Singapore 
3513 
Bulgaria 
469 
Jordan 
402 
Slovak Republic 
827 
Cameroon 
8 
Kazakhstan 
41 
Slovenia 
1026 
Canada 
11635 
Korea 
17676 
South Africa 
2263 
Chile 
229 
Latvia 
67 
Spain 
17749 
China 
63188 
Lithuania 
202 
Sri Lanka 
155 
Colombia 
1117 
Luxembourg 
108 
St. Lucia 
3 
Costa Rica 
60 
Macao 
39 
Sudan 
3 
Croatia 
415 
Malawi 
1 
Swaziland 
11 
Cyprus 
334 
Malaysia 
3195 
Sweden 
4652 
Czech Republic 
5627 
Malta 
207 
Switzerland 
8605 
Denmark 
2163 
Mauritius 
175 
Thailand 
3870 
Dominica 
3 
Mexico 
2233 
Trinidad and Tobago 
29 
Dominican Republic 
25 
Moldova 
7 
Tunisia 
302 
Ecuador 
33 
Mongolia 
2 
Turkey 
2949 
Egypt 
546 
Morocco 
158 
Uganda 
60 
El Salvador 
17 
Namibia 
24 
United Kingdom 
66760 
Estonia 
202 
Netherlands 
12745 
United States 
37026 
Finland 
1870 
New Zealand 
2069 
Uruguay 
241 
France 
20919 
Nicaragua 
5 
Venezuela 
373 
Germany 
41629 
Norway 
1703 
Zambia 
10 
Greece 
2325 
Oman 
67 
Zimbabwe 
134 
Grenada 
3 
Pakistan 
539 
Source: ISO(2002)

Variable 
Description 
Mean 
Std. dev. 
Min 
Max 
X_{ij} ^{a} 
Exports from country i to country j in billion current US$ 
0,50 
4,12 
0 
239,95 
GDP_{i} ^{b} 
GDP of country i in trillion current US$ 
0,28 
1,03 
0,0002 
10,08 
POP_{i} ^{b} 
Population of country i in millions 
49,06 
160,05 
0,07 
1285,23 
ISO_{i} ^{c} 
Number of country i's ISO 9000 adopters in thousands 
2,93 
8,33 
0 
66,76 
ΣISO_{i} ^{d} 
The sum of ISO 9000 adopters in all countries except country i in thousands 
287,20 
124,37 
74,46 
505,54 
ΣISO^{C} _{i} ^{d} 
Index of ISO 9000 adoptions by all foreign business customers of firms in country i 
3,49 
7,38 
0,002 
60,94 
ΣISO^{S} _{i} ^{d} 
Index of ISO 9000 adoptions by all foreign suppliers of firms in country i 
10,78 
10,98 
0,45 
103,56 
Note: All variables are on yearly basis.
Sources: a UN Comtrade databank; b IMF International Financial Statistics; c ISO(2002); d own calculations
based on all previous sources

Dependent variable: ln Exports_{ijt} 

Independent variables^{a} 
(1) 
(2) 
(3) 
(4) 
(5) 
(6) 
(7) 
(8) 
ln GDP_{it} 
0,300^{***} (8,51) 
0,303^{***} (8,56) 
0,296^{***} (8,46) 
0,299^{***} (8,52) 
0,117^{**} (2,38) 
0,118^{**} (2,38) 
0,097^{**} (2,01) 
0,091^{*} (1,87) 
ln GDP_{jt} 
0,697^{***} (20,21) 
0,711^{***} (20,45) 
0,722^{***} (21,16) 
0,736^{***} (21,41) 
0,736^{***} (14,60) 
0,746^{***} (14,70) 
0,758^{***} (15,41) 
0,773^{***} (15,57) 
ln POP_{it} 
0,230 (1,13) 
0,295 (1,43) 
1,072 (0,82) 
2,948^{**} (2,01) 

ln POP_{jt} 
0,651^{***} (3,28) 
0,704^{***} (3,49) 
1,263 (1,06) 
0,541 (0,37) 

ln ISO_{it} 
0,043^{***} (5,18) 
0,043^{***} (5,16) 
0,021^{*} (1,69) 
0,023^{*} (1,88) 

ln ISO_{jt} 
0,042^{***} (5,09) 
0,043^{***} (5,25) 
0,059^{***} (5,37) 
0,059^{***} (5,33) 

const ^{b} 
8,598^{***} (7,00) 
6,903^{***} (5,08) 
9,098^{***} (7,47) 
7,094^{***} (5,22) 
12,165^{***} (7,71) 
11,332^{***} (6,53) 
12,325^{***} (7,84) 
10,815^{***} (6,14) 
ln GDP_{i(t+1)} 
0,316^{***} (6,22) 
0,316^{***} (6,21) 
0,326^{***} (6,48) 
0,326^{***} (6,49) 

ln GDP_{j(t+1)} 
0,027 (0,53) 
0,023 (0,46) 
0,031 (0,63) 
0,030 (0,60) 

ln POP_{i(t+1)} 
1,041 (0,77) 
2,906^{*} (1,92) 

ln POP_{j(t+1)} 
0,731 (0,60) 
0,165 (0,11) 

ln ISO_{i(t+1)} 
0,033^{**} (2,57) 
0,030^{**} (2,31) 

ln ISO_{j(t+1)} 
0,000 (0,01) 
0,001 (0,10) 

Wald test (χ^{2}) 
 
 
 
 
39,02^{***} 
40,00^{***} 
48,24^{***} 
51,06^{***} 
Number of:  
observations 
46909 
46909 
45467 
45467 
39023 
39023 
37459 
37459 
groups 
8803 
8803 
8724 
8724 
8616 
8616 
8426 
8426 
Observations per group:  
min 
1 
1 
1 
1 
1 
1 
1 
1 
avg 
5,3 
5,3 
5,2 
5,2 
4,5 
4,5 
4,4 
4,4 
max 
7 
7 
7 
7 
6 
6 
6 
6 
R^{2}:  
within 
0,026 
0,027 
0,030 
0,030 
0,025 
0,025 
0,028 
0,029 
between 
0,445 
0,211 
0,486 
0,167 
0,517 
0,496 
0,559 
0,352 
overall 
0,417 
0,197 
0,461 
0,155 
0,482 
0,469 
0,533 
0,334 
a Yeardummies’ coefficients suppressed.
b The constant term is defined here as the average of importerexporterspecific effects.
*** denotes significance at 1% level, ** at 5% level, * at 10% level; tstatistics in parentheses.

Dependent variable: Δln Exports_{ijt} 

Independent variables^{a} 
(9) 
(10) 
(11) 
(12) 
(13) 
(14) 
(15) 
(16) 
Δln GDP_{it} 
0,076 (1,56) 
0,074 (1,53) 
0,073 (1,52) 
0,072 (1,50) 
0,072 (1,48) 
0,071 (1,46) 
0,071 (1,47) 
0,077 (1,57) 
Δln GDP_{jt} 
0,760^{***} (15,34) 
0,764^{***} (15,39) 
0,776^{***} (15,96) 
0,778^{***} (15,98) 
0,758^{***} (15,25) 
0,762^{***} (15,31) 
0,776^{***} (15,74) 
0,784^{***} (15,85) 
Δln POP_{it} 
0,338 (0,69) 
0,304 (0,61) 
0,062 (0,12) 
0,046 (0,09) 

Δln POP_{jt} 
0,539 (1,14) 
0,343 (0,71) 
0,770 (1,55) 
0,561 (1,10) 

Δln ISO_{it} 
0,027^{**} (2,27) 
0,027^{**} (2,27) 
0,025^{**} (2,09) 
0,018 (1,48) 

Δln ISO_{jt} 
0,026^{**} (2,32) 
0,026^{**} (2,31) 
0,026^{**} (2,36) 
0,031^{***} (2,70) 

ln GDP_{it} 
0,004 (1,39) 
0,011^{***} (2,72) 
0,005 (0,97) 
0,020^{**} (2,54) 

ln GDP_{jt} 
0,002 (0,93) 
0,007^{*} (1,67) 
0,004 (0,79) 
0,014^{*} (1,79) 

ln POP_{it} 
0,011^{**} (2,45) 
0,012^{***} (2,62) 

ln POP_{jt} 
0,005 (1,20) 
0,009^{*} (1,85) 

ln ISO_{it} 
0,001 (0,28) 
0,006 (1,25) 

ln ISO_{jt} 
0,001 (0,30) 
0,004 (0,84) 

Wald test (χ^{2}) 
 
 
 
 
2,46 
10,04^{**} 
3,32 
13,63^{**} 
Number of observations 
36583 
36583 
35250 
35250 
36583 
36583 
35250 
35250 
R^{2} 
0,013 
0,013 
0,014 
0,014 
0,013 
0,013 
0,014 
0,014 
a Yeardummies’ coefficients suppressed.
*** denotes significance at 1% level, ** at 5% level, * at 10% level; tstatistics in parentheses.

Dependent variable: Δln Exports_{ijt} 

Independent variables^{a} 
(17) 
(18) 
(19) 
(20) 
(21) 
(22) 
(23) 
(24) 
Δln GDP_{it} 
0,322^{***} (5,47) 
0,316^{***} (5,33) 
0,311^{***} (5,28) 
0,306^{***} (5,16) 
0,319^{***} (5,42) 
0,259^{***} (4,34) 
0,308^{***} (5,22) 
0,261^{***} (4,36) 
Δln GDP_{jt} 
0,879^{***} (15,19) 
0,883^{***} (15,10) 
0,871^{***} (15,03) 
0,874^{***} (14,95) 
0,879^{***} (15,22) 
0,874^{***} (14,88) 
0,859^{***} (14,70) 
0,863^{***} (14,60) 
Δln POP_{it} 
0,461 (0,73) 
0,390 (0,62) 
0,696 (1,10) 
0,698 (1,08) 

Δln POP_{jt} 
0,272 (0,42) 
0,299 (0,47) 
0,319 (0,50) 
0,561 (0,86) 

Δln ISO_{it} 
0,063^{***} (3,72) 
0,063^{***} (3,70) 
0,060^{***} (3,55) 
0,018 (0,97) 

Δln ISO_{jt} 
0,036^{**} (2,07) 
0,036^{**} (2,08) 
0,036^{**} (2,09) 
0,035^{*} (1,85) 

ln GDP_{it} 
0,011^{***} (3,69) 
0,044^{***} (7,54) 
0,011^{*} (1,92) 
0,044^{***} (5,42) 

ln GDP_{jt} 
0,008^{**} (2,56) 
0,013^{**} (2,23) 
0,001 (0,10) 
0,001 (0,14) 

ln POP_{it} 
0,041^{***} (6,58) 
0,038^{***} (5,64) 

ln POP_{jt} 
0,007 (1,08) 
0,001 (0,21) 

ln ISO_{it} 
0,000 (0,07) 
0,002 (0,36) 

ln ISO_{jt} 
0,008 (1,46) 
0,009 (1,60) 

Wald test (χ^{2}) 
 
 
 
 
19,48^{***} 
63,64^{***} 
20,20^{***} 
52,62^{***} 
Number of observations 
4559 
4559 
4559 
4559 
4559 
4559 
4559 
4559 
R^{2} 
0,085 
0,085 
0,089 
0,089 
0,089 
0,098 
0,093 
0,099 
a Yeardummies’ coefficients suppressed.
*** denotes significance at 1% level, ** at 5% level, * at 10% level; tstatistics in parentheses.

Dependent variable: ΔISO_{it} 

Independent variables^{a} 
(25) 
(25’) 
(26) 
(26’) 
Market size  
GDP_{it} 
6,261^{***} (10,90) 
1,0371^{***} (3,39) 
4,3072^{***} (7,29) 
0,4972 (1,51) 
Domestic adoptions  
ISO_{i(t1)} 
1,397^{***} (41,49) 
1,386^{***} (40,71) 
1,4737^{***} (56,65) 
1,251^{***} (47,22) 
ISO^{2} _{i(t1)} 
0,0129^{***} (15,15) 
0,0027^{***} (7,57) 

Foreign customers’ adoptions  
ΣISO^{C} _{i(t1)} 
0,2309^{***} (7,50) 
0,1697^{***} (5,43)  
(ΣISO^{C} _{i(t1)})^{2} 
0,0046^{***} (2,99)  
Foreign suppliers’ adoptions  
ΣISO^{S} _{i(t1)} 
0,0383^{***} (2,76) 
0,0277 (1,15)  
(ΣISO^{S} _{i(t1)})^{2} 
0,00006 (0,09)  
Unweighted foreign adoptions  
ΣISO_{i(t1)} 
0,0033^{**} (2,60) 
0,0041 (0,89) 

(ΣISO_{i(t1)})^{2} 
0,000003 (0,97) 

Interaction terms  
GDP_{it }ISO_{i(t1)} 
0,0362^{**} (2,09) 
0,1136^{***} (20,26)  
GDP_{it }ΣISO^{C} _{i(t1)} 
0,0020 (0,17)  
GDP_{it }ΣISO^{S} _{i(t1)} 
0,2369^{***} (10,27)  
GDP_{it }ΣISO_{i(t1)} 
0,0023^{**} (1,97)  
ISO_{i(t1)} ΣISO^{C} _{i(t1)} 
0,0101^{***} (4,19)  
ISO_{i(t1)} ΣISO^{S} _{i(t1)} 
0,0150^{***} (6,37)  
ΣISO^{C} _{i(t1)} ΣISO^{S} _{i(t1)} 
0,0041^{*} (1,69)  
ISO_{i(t1)} ΣISO_{i(t1)} 
0,0004^{***} (10,84) 

Sargan test (χ^{2})^{b} 
17,98 (14) 
16,78 (14) 
17,11 (14) 
21,03 (14) 
ArellanoBond m_{2} test^{c} 
1,45 
1,41 
1,47 
1,34 
Number of observations 
475 
475 
475 
475 
*** denotes significance at 1% level, ** at 5% level, * at 10% level; tstatistics in parentheses.
a All variables are in first differences.
^{b} Sargan test of overidentifying restrictions; degrees of freedom in parentheses.
c ArellanoBond test of secondorder serial correlation in residuals.
^{33} http://www.tc176.org/About176.asp
^{34} See ISO (2001).
^{35} See ISO (2002).
^{36} Ibid.
^{37} See Anderson et al. (1995, 1999) for details.
^{38} Ibid.
^{39} Ibid; depending on the product category, the EC Mark must have been attained till 19921995.
^{40} We give the common language hypothesis a slightly different spin than Bénézech et al. (2001). They concentrate rather on the role of ISO 9000 as a means to codify the knowledge within a firm.
^{41} Some empirical studies, in which gravity equation for trade is not theoretically motivated, often define Vij as bilateral trade flows rather than exports alone.
^{42} Note that the trade cost factor tij negatively affects the value of exports Vij in (4.3) only if the substitution between goods is elastic (σ > 1).
^{43} The theoretical gravity equation (4.3) is valid under the assumption that the trade cost factors are symmetric, i.e. .
^{44} While {Pi} are consumer price indices in the Anderson and Wincoop’s (2003) model, they should be interpreted much broader, as indicated by the authors. E.g., a home bias in preferences instead of trade costs might lead to exactly the same gravity equation for exports as (4.3).
^{45} See the discussion in Frankel et.al. (1997), p. 55.
^{46} As it is clear from the discussion in previous chapter, the measure of distance between two countries Dij could be broader interpreted as the measure of trade barriers between them. However, according to the model by Anderson and Wincoop (2003), this interpretation implicitly assumes that the elasticity of substitution between goods produced in different countries in bigger than one.
^{47} We would like to stress that specification (4.9) describes average bilateral trade barriers, since the trade barriers faced by ISO 9000 certified firms and non certified firms differ.
^{48} In equation (4.10), a small technical difficulty arises, because of zerovalued entries in bilateral exports Xijt. Following Frankel et al. (1997), we treat these observations as missing.
^{49} Actually the inclusion of lnPOP or ln(GDP/POP) in (4.6) is mathematically equivalent. What changes is only the interpretation of the parameters. See Frankel at al. (1997) for a simple exposition of this point.
^{50} Note that in contrast to (4.8), time in (4.13) is treated as discrete variable. As a consequence, δ in (4.8) becomes “one period” in (4.13).
^{51} The reason why we choose this particular normalization is merely that the order of magnitude of coefficients in (4.13) and (4.13’) is not too diverse. This makes exposition of the estimation results look nicer.
^{52} We follow the approach expressed by Wooldridge (2002, p. 251–252), that FE and RE correspond to the assumptions we are willing to impose on unobserved effects ηij in order to estimate the model rather than to their deterministic or stochastic nature.
^{53} Freund and Weinhold (2004) also report insignificant FD estimate of own GDP coefficient in their gravity equation for exports. Moreover they find the coefficient on partner’s GDP to lie just slightly above 0,10 (see table 3 in there).
^{54} See the discussion in section 4.3.1.
^{55} Since the panel data estimation techniques we use provide timeseries evidence, because crosssection differentiation is captured by countrypair specific effects, this assumption is much more heroic that it might seem.
^{56} In the regression (25) we additionally assumed that d = 0 in (4.13). Without this assumption the main coefficients of the model turned out to be insignificant and the ArellanoBond test indicated secondorder autocorrelation in the residuals. The likely reason for this is the multicollinearity of the explanatory variables.
^{57} We also carried out formal Waldtype tests, which confirm this intuitive argumentation.
© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.  
XDiML DTD Version 4.0  Zertifizierter Dokumentenserver der HumboldtUniversität zu Berlin  HTMLVersion erstellt am: 14.04.2005 