The Generalized System of Preferences: Is it effective? And what if the beneficiary country loses its preferential treatment?

The Generalized System of Preferences: Is it effective?

And what if the beneficiary country loses its preferential

treatment?

MSc thesis

to obtain the degree of MSc in Economic Development and Globalization

at the

University of Groningen supervised by Dr T. Kohl and co-assessed by Dr X. Ye

by

Kasper Brueren - S3916847 - k.j.brueren@student.rug.nl

1 Abstract

The Generalized System of Preferences (GSP) was established to promote the exports of low-income countries to industrialized countries in order to support their economic growth and development. Using an extensive dataset, including data over a period from 1997 to 2016, evidence has been found that the GSP has positive effects on stimulating developing countries’ exports. This positive effect holds after including a set of country, year and product fixed effects. Furthermore, the analysis finds evidence that some countries experience difficulties to maintain high export levels after its GSP eligibility expires. The exports of several other countries are, however, not negatively affected by the GSP expiration.

2 List of abbreviations

AGOA African Growth and Opportunity Act

ATPDEA Andean Trade Promotion and Drug Eradication Act CBI Caribbean Basin Initiative

CNL Competitive Need Limitation

EBA Everything But Arms

ELG Export-Led Growth

EU European Union

FTA Free Trade Agreement

GATT General Agreement on Tariffs and Trade GAO General Accounting Office

GSP Generalized System of Preferences HTS Harmonized Tariff Schedule

HTSUS Harmonized Tariff Schedule of the United States LDB Least Developed Beneficiary

LDBDCs Least Developed Beneficiary Developing Country LDC Least Developed Country

MFN Most Favoured Nation

NAFTA North American Free Trade Agreement PTA Preferential Trade Agreement

PTPA United States-Peru Free Trade Agreement PTADB Preferential Trade Agreement Database

ROO Rules Of Origin

SSA Sub Saharan Africa

TPP Trans-Pacific Partnership

3

1 - Introduction

Usually, if trade is conducted between two countries, tariffs are imposed to protect local producers of the importing country. The imported product will become costlier when it enters a foreign country. For developing countries, it is then difficult to access developed markets. To facilitate this, the Generalized System of Preferences (GSP) was introduced in the 1970s. It is a system that is developed to stimulate trade between developed countries (donor countries) and developing countries (beneficiary countries). Under the system, the donor country agrees to not impose tariff barriers with the purpose to stimulate exports and economic growth for the beneficiary country. At the conference that launched the GSP, the United Nations Conference on Trade and Development (UNCTAD), it was stated that the objectives, in favour of the developing countries, were to ‘’increase their export earnings, promote their industrialization and to accelerate their rates of economic growth.‘’ Despite the longevity of the system and the potential importance for beneficiary countries, the empirical evidence on the effectiveness of the GSP program is divided. Some researchers find positive effects of GSP on exports of beneficiaries. Others discover a negative effect. Furthermore, criticism has been expressed about the Competitive Needs Limit (CNL). This is a specific restriction of the GSP, which excludes beneficiary products or countries from the tariff exemption, once a certain threshold is reached. Exporters of such products are designated as super competitive and deemed to be able to export without the preferential treatment, but are they?

This study aims to address two questions related to the effectiveness of GSP. First, does GSP eligibility genuinely increase trade? And second, what will happen to its exports when the beneficiary country has reached a threshold level and it will be excluded from the GSP? Has the country become competitive enough to continue exporting without favourable preferences? To examine these relevant issues, an extensive dataset was constructed, mainly by using product-specific import data from the United States International Trade Commission (USITC). This product-specific data was merged with data on preferential and free trade programs and the final dataset allows us to examine the effectiveness per program. As data on program composition per year is included, this data is also appropriate to test the impact of GSP expirations. A quantitative analysis will be executed to capture the effects of GSP expiration on exports of beneficiary countries. This econometric analysis will contribute to capturing the long-run effects of several preferential treatment programs, but the focus is on GSP. Whether exports fall significantly after expiration will be of both theoretical and practical importance to policymakers, national governments and everyone who has an interest in the effectiveness of preferential-and-trade agreements.

4

The contribution to the existing literature will be twofold: firstly, the effectiveness of the GSP program will be investigated to bring more consensus to the currently divided literature. This will be done by using an elaborate and recent dataset that includes data until 2016. Even though the effectiveness of GSP has frequently been investigated (Kohl, 2017; Gil-Pareja, 2014, Rose, 2004; Reynolds, 2009; Herz & Wagner, 2011), the model developed and tested in this work does make modifications. To the best of my knowledge, no research has been conducted with such a product-specific and recent dataset which contains information on several preferential programs. Furthermore, the continuous turnover of each program’s composition will be taken into account, which is critical when evaluating the effectiveness of GSP. By including more trade programs, this work generates information that goes beyond the GSP program. Secondly, it will be examined what GSP graduation does to a country’s exports. Even though the literature on GSP, in general, is voluminous, the literature on GSP expiration and its implications is largely descriptive and relatively scant. This quantitative work adds to the limited, yet important literature on the effects of GSP expiration. Hakobyan (2017) extensively discusses CNL episodes (which are essentially GSP expirations) but her work uses CNL exclusions until 2008, and she investigates individual commodities that have reached the CNL. This work attempts to take a broader macro-level perspective with regards to country GSP expiration effects and this has not been done before, certainly not with data as recent as the data used here.

To preview the main results, significant evidence was found that the US GSP program does increase beneficiary exports to the US market. This result was found after developing and testing a model in which the impact of several trade programs on US imports is measured. After including sets of country, year and product fixed effects, this result remained significantly positive. The second result regarding country graduations implies that some countries are negatively affected by the GSP expiration. The products that used to be exported to the US under the GSP in especially Malta, Ukraine and Uruguay struggled significantly after the GSP status was lost. Other countries did not experience decreased exports after GSP expiration. The exports dropped in approximately half of the investigated country observations, with the exports of the other half remaining stable or increasing. Understanding these findings will help beneficiary countries to evaluate the effectiveness of participating in programs such as GSP. The GSP expiration effects are not as negative as other researchers have claimed.

The remainder of this work is organized as follows. In section 2, the literature review will provide a theoretical background of the GSP and academic literature on its effectiveness is discussed. Furthermore, hypotheses will be developed. In section 3 the empirical model and the data collection process are described. Section 4 both provides and analyses the results. Finally, conclusions will be drawn, limitations are discussed, and directions for future research will be proposed.

2 - Literature review

5

gained popularity (Sapir & Lundberg, 1984). The question of whether trade is causally related to economic growth is of considerable relevance to development economists, politicians and many others. Hakobyan (2017) has identified ELG as one of the promising paths to promoting economic growth in developing countries. Likewise, studies by Balassa (1978) and more recently Awokuse (2007) have found support for the ELG hypothesis, which includes that export expansion is an important determinant of development. Dollar & Kraay (2004) argue that the statement ‘’openness to international trade boosts development’’ is one of the most widely held beliefs in economics by both the left and the right-wing. Although not completely unambiguous, most studies support the positive and significant effects of trade on output and economic development (Singh, 2010).

The major benefit of trade is that it allows countries to specialize which results in a more efficient allocation of resources. This reasoning is an old economic theory developed by Ricardo (1817), but recent studies by for instance Anderson & Yotov (2016) support the positive relationship between trade and more efficient allocation. The aforementioned benefits of trade oblige to dig deeper into what factors influence trade. Trade is determined by factors such as geographical location, comparative advantage, political policy, infrastructure and free trade agreements (FTAs). In trade research, the gravity model is a widely used concept and the aforementioned factors are typically among the included variables of this gravity model. Given the positive association between trade and development, several instruments which aim at stimulating trade have been constructed. The General Agreement on Tariffs and Trade (GATT) and its successor, the World Trade Organisation (WTO), have been developed to reduce tariffs and barriers to trade, with the purpose to increase international trade. One of the core principles of this multilateral trading system is the most favoured nation principle (MFN), which ensures equal treatment among all WTO members: grant a lower tariff to country X, and this lower tariff should be extended to all WTO-members (WTO, 2019). The most recent study on the effectiveness of GATT/WTO membership on trade by Larch et al. (2019), finds that this program, as intended, increases trade significantly. On average, joining GATT/WTO has increased trade between members by 171% and trade between member and non-member countries by 88%. The finding that the WTO increases trade is supported by both Subramanian & Wei (2007) and Kohl (2017), who invalidate an earlier paper by Rose (2004), which found that the WTO did not increase trade.

6

(CBI) which has the same purpose for the Caribbean area. PTAs as AGOA have received support by for instance Frazer & van Biesebroeck (2010), who found that the AGOA has had significantly positive effects on its beneficiaries’ agricultural and manufactured imports to the US. After controlling for product specific baseline levels of imports and for country-specific and product-country-specific import trends in the post-AGOA period, it was concluded that AGOA-GSP products imported by the US increased by an average of 13%, as a result of these preferential programs.

The effectiveness of these PTAs is, nevertheless, not undisputed. The AGOA has received criticism because the products imported by the US are not sufficiently diversified. In 2007, 80% of the imports under AGOA were petroleum products. The oil industry is extremely capital intensive and does not provide comprehensive employment opportunities for the African population. Furthermore, 87% of the US imports came from three countries; Nigeria, Gabon and South Africa. Two of those are major oil-producing countries (Guseh & Oritsejafor, 2009). These PTAs are not unambiguous and smaller in size than the most famous PTA: the Generalized System of Preferences (GSP), under which developed countries grant preferential tariffs to imports from a wide range of developing countries (Baldwin & Murray, 1977).

2.1 US GSP eligibility

Under the US GSP scheme, products that are eligible for GSP treatment enter free of duty. The program was proposed by developing countries because equal treatment through the GATT/WTO was not considered good enough to offer development opportunities. The argument was that poorer countries were in need of protection to help promote their industries and this is how the GSP system emerged. The United States Trade Representatives (USTR) regard the GSP system as mutually beneficial for both parties of the agreement. The system provides trade opportunities to the poorest countries to grow their economies and climb out of poverty. At the same time, the US benefits because moving GSP imports from the docks to the consumers creates numerous jobs. Furthermore, the program promotes competitiveness as intermediate inputs can be acquired at lower prices. Lastly, GSP supports progress by beneficiary countries with respect to governance standards as human rights. If the beneficiary country violates human rights, its GSP eligibility can be terminated (USTR, 2019). Bangladesh has recently been stripped of its GSP status for the lack of improvements in the legal infrastructure that protects human rights (UNCTAD, 2018). In the eighties, GSP privileges for Nicaragua, Paraguay and Chile had also been suspended for reasons of workers’ rights (UNCTAD, 2016). The existence of the GSP program may encourage developing countries to realise improvements in these fundamental rights. To qualify for the preferred treatment, requirements with regards to country eligibility, product eligibility and rules of origin (ROO) must be met (UNCTAD, 2016).

2.1.1 Country eligibility

7

treatment to the poorest beneficiaries. For the US this is the African Growth and Opportunity Act (AGOA), and the EU has both the GSP+ and an Everything But Arms (EBA) program. The EBA is an initiative under which all restrictions on imports to the EU from the least developed countries (LDCs) are removed, with the exception of armaments. It happens on a non-reciprocal basis, aiming to further improve access of LDCs access to the EU market (Yu & Jensen, 2005). The essence of the US and EU GSP systems is the same, but some differences exist. Whereas the US offers tariff cuts all the way to zero, the EU offers smaller tariff rate cuts which apply to a broader range of products (Keynes & Bown, 2018). By reason of literature and data availability, the focus for this work will be on US GSP.

The US GSP currently has 119 beneficiaries, of which 44 are least developed beneficiaries (LDB). The beneficiaries of the US GSP program are listed in table 1.

Table 1 GSP eligible countries in December 2019

Afghanistan*, Albania, Algeria, Angola*, Anguilla, Argentina, Armenia, Azerbaijan, Belize, Benin*, Bhutan*, Bolivia, Bosnia-Hercegov, Botswana, Br Virgin Is, Brazil, Br Indian O Ter, Burkina Faso*, Burma (Myanmar)*, Burundi*, Cape Verde, Cambodia*, Cameroon, Cen African Rep*, Chad*, Christmas Is, Cocos Is, Comoros*, Congo (DROC)*, Congo (ROC), Cook Is, Côte d'Ivoire, , Djibouti*, Dominica Is, Ecuador, Egypt, Eritrea, , Ethiopia*, Falkland Is, Fiji, Gabon, Gambia*, Georgia, Ghana, Grenada Is, Guinea*, Guinea-Bissau*, Guyana, Haiti*, Heard & McDn Is, Indonesia, Iraq, Jamaica, Jordan, Kazakhstan, Kenya, Kiribati*, Kosovo, Kyrgyzstan, Lebanon, Lesotho*, Liberia*, Macedonia, Madagascar*, Malawi*, Maldive Is, Mali*, Mauritania*, Mauritius, Moldova, Mongolia, Montenegro, Montserrat Is, Mozambique*, Namibia, Nepal*, Niger*, Nigeria, Niue, Norfolk Is, Pakistan, Palestine, Papua New Guin, Paraguay, Philippines, Pitcairn Is, Rwanda*, Samoa*, Sao Tome & Prin*, Senegal*, Serbia, Sierra Leone*, Solomon Is*, Somalia*, South Africa, South Sudan*, Sri Lanka, St Helena, St Lucia Is, St Vinc & Gren, Suriname, Swaziland, Tanzania*, Thailand, Timor-Leste*, Togo*, Tokelau, Tonga, Tunisia, Tuvalu*, Uganda*, Ukraine, Uzbekistan, Vanuatu*, Wallis and Futuna, Western Sahara, Yemen*, Zambia*, Zimbabwe.

*indicates Least Developed Country (LDC)

Source: Preferential Trade Arrangement DataBase (PTADB)

8 2.1.2 Product eligibility

Articles eligible for duty-free treatment are defined in the Harmonized Tariff Schedule of the United States (HTSUS). Most products eligible for GSP are dutiable manufactures and semi-manufactures, fishery and primary industrial products (UNCTAD, 2016). About 5000 tariff items are eligible for GSP benefits. Approximately 3500 of these are available to all GSP eligible countries and the rest is solely available to Least Developed Beneficiary Developing Countries (LDBDCs). The total U.S. imports under GSP amounted to $23.6 billion in 2018. The most benefitting top five GSP products had been gold necklaces ($428 million), ferrochromium ($365 million), rubber gloves ($292 million), non-alcoholic beverages ($276 million) and beverage preparations such as herbal tea ($257 million) (USTR, 2019). Certain articles are prohibited from receiving GSP benefits for varying reasons. Most textiles, watches, footwear, work gloves, leather apparel, steel, glass and electronic articles are excluded from GSP benefits (USTR, 2018).

2.1.3 Rules of origin (ROO)

Apart from the country and product requirements, rules with regards to origin must be met. The ROO have been implemented to ensure that minimum levels of local content in products are realised. As the eligible article enters the US, at least 35 per cent of the value (materials plus processing costs) of the article must have been contributed by the beneficiary country (UNCTAD, 2018). The products imported under these preferential schemes should not be merely transhipped to US markets from non-eligible countries via eligible countries without the addition of local value. The ROO requirements help the beneficiary country to, as intended, actually reap the benefits of the program (Lippoldt, 2006) and it combats misuse of the program by non-eligibles.

The USTR are positive about the benefits of the GSP for both parties of the program. The latest GSP newsletter, by UNCTAD (2018) provides an evaluation of the GSP schemes of several GSP granting countries. It was stated that US GSP has increased trade numbers. Every year, millions of products enter the US duty-free through the GSP program. The theoretical details and requirements of the program have now been discussed, mostly by using government reports. Besides this theoretical information, an increasing body of economic research investigates the question of whether the GSP is effective in increasing beneficiary exports. The following section discusses this academic literature.

2.2 Academic GSP support

9

with the conclusions of Rose (2004). He finds that the GSP has a strong effect and he claims that GSP eligibility is associated with doubling trade numbers. To reach this conclusion a standard ‘’gravity’’ model of bilateral merchandise trade was developed. In trade research, this gravity model is widely used. It typically includes factors as economic sizes and distance between countries to predict bilateral trade flows. To test the model, Rose used a large panel dataset, including observations for 175 countries over 50 years.

Snyder (2011) states that it has been acknowledged that nonreciprocal preferences as GSP, have had a positive effect on development and economic growth of the beneficiary countries. He refers to Collier and Venables (2007), who examined the employment creation in developing countries as a result of GSP. The results indicated that GSP’s impact on the development of beneficiaries’ nascent industries is positive, but with varying effects per industry: whereas oil products result in low developmental benefits, agricultural products and processed foods could carry significant benefits for developing countries. An empirical study by Seyoum (2006) investigates the relation between US GSP trade preferences and beneficiary country exports. A large dataset covers US imports from 120 developing countries and the results indicate that the GSP has a significantly positive effect on developing country exports to the US for both country and product groups. Likewise, using a dataset covering 181 countries that contains observations for the period 1948-2007, Kohl (2017) concludes that GSP schemes are trade-promoting. This is in accordance with Frazer and Van Biesebroek (2010). After applying heavy econometric tools, significantly positive effects of GSP eligibility on developing countries’ exports were found. The positive effects are larger for the schemes designed for the least developed countries, such as the GSP+, AGOA and EBA. This is, in a way, a desirable finding: the poorest countries should be able to benefit the most.

Many researchers acknowledge the positive relation between GSP and trade, and beneficiary countries have also been positive about the program. The US General Accounting Office (GAO) visited several beneficiary countries. Brazil, The Dominican Republic, Hungary, Malaysia, Thailand and Turkey all expressed enthusiasm about the program. It was said the program lends support to realize economic development, even though the level of development attributable to GSP could not exactly be determined (US GAO, 1994). Two of the most successful stories of the program are India and Thailand, which expanded their jewellery industries (partially as a result of the favourable tariffs). Increases in standards of living were said to be realized in the benefitting industries of these countries: initially, factory workers walked to work. Later on, workers used their own mopeds to go to work (Keynes & Bown, 2018). Clearly, this development should not be fully attributed to the GSP, but it is reasonable to suggest the program has played a role.

Whether the GSP system is effective in increasing trade between beneficiary and donor countries is a critical question when evaluating the GSP system and accordingly, the first hypothesis that will be investigated in the remainder of this work is as follows.

10

2.3 Academic GSP criticism

The literature that has been discussed so far tends to conclude that the GSP system is a program that increases the beneficiary’s exports and offers a development opportunity. However, conclusions cannot be drawn too early. Many economists have been sceptical about the benefits of the GSP. One important restriction of the GSP is, as Hakobyan (2019) discusses, that it is not a permanent program; it has frequent expirations and it needs to be periodically renewed by the congress. Even though the duties paid during periods of expiration are reimbursed, the (temporary) expiration of GSP has negative effects on the exports of developing countries. Perhaps the poor countries lack access to trade finance and cannot borrow the necessary money to cover the suddenly higher costs resulting from the expiration. Hakobyan conducts a triple difference-in-differences (product, country, expiration) approach to investigate the numerical effect of the expiration on exports. She finds that the expiration of GSP had a significantly negative impact on the level of exports to the US. On average, exports dropped by 3% in 2011, with exports in several sectors dropping to 9%.

Apart from temporary expiration, beneficiary products can reach the Competitive Needs Limit (CNL). Products that exceed this limit will no longer be eligible to receive the preferred treatment. The CNL restriction has two objectives. Firstly, to exclude country-product pairs that are identified as adequately competitive and no longer in need of the tariff exemption. The second objective of CNLs is to increase market access opportunities for other GSP beneficiaries, which are not competitive enough yet. Terminating the GSP eligibility of the ‘’super competitive’’ should stimulate exports from other countries that receive preferential treatment (Hakobyan, 2017). Hakobyan’s analysis, however, contradicts this second objective. She shows that lost market shares of the affected countries are mainly captured by richer, non-GSP countries and not, as aspired with the CNL restriction, by other non-GSP eligible countries. This finding is not supportive of the initial purpose of the CNL.

The CNL is triggered on a product if US imports from a country account for half or more of the value of total US imports of that product, or when a certain dollar value is exceeded ($190 million in 2019). When the imports of GSP eligible products from beneficiary countries exceed the CNL, automatic termination before 1 July of the following calendar year will be applied, unless a CNL waiver is granted. Products from a beneficiary are identified as ‘’sufficiently competitive’’ when the limits are reached. CNLs do not apply to least-developed beneficiary developing countries (LDBDCs) and beneficiaries of the AGOA. Besides product-specific limits, entire countries can ‘’graduate’’ and lose GSP eligibility on the basis of factors related to national income or competitiveness. A country loses its preferential treatment when it reaches the ‘’high-income country’’ status. Numerous countries have graduated from GSP: some are the Asian tigers (The Republic of Korea, Singapore, Hong Kong and Taiwan) in 1989, Israel in 1995 and The Netherlands Antilles in 1998 (UNCTAD, 2016).

11

GSP benefits for developing countries. The uncertainty about future tariffs restricts the economic development benefits of this trade-instead-of-aid program. Research has indicated that companies tend to underinvest in exports when uncertainty about future tariffs is present (Bown, 2018). Furthermore, economist Hakobyan has found that developing countries do regularly not take advantage of GSP because of its considerable compliance costs. These restrictions might have harmful effects on the economic structure and trading patterns of GSP eligible countries in the long run (Panagariya, 2003).

In 1994, the US General Accounting Office (US GAO) reported on the detrimental impact of CNLs. Interviews with industry representatives in six GSP beneficiary countries were conducted and many concerns were expressed about the unpredictable nature of CNL exclusions. After exclusion, a country does not know when, and even if, the preferential tariffs will be reinstated through a waiver (US GAO, 1994). The first researchers to write on the impact of GSP expirations (CNLs) were MacPhee & Rosenbaum (1989) and Devault (1996). MacPhee & Rosenbaum (1989) investigate 816 CNL exclusions between 1976 and 1983. Descriptive statistics on trade flows before and after the exclusion show that the exclusion has a negative effect on the market shares of affected countries. Devault (1996) examines the effect on import values using 45 cases of CNL exclusions between 1988 and 1993. His results are in harmony with results by MacPhee & Rosenbaum. Import shares of affected countries seem to decline a year or two after the CNL exclusion. In some cases, CNLs reduce import shares by more than half. He concludes that the CNLs contradict the original intent of the GSP. Despite the fact that their findings are primarily descriptive, insightful early findings on the impact of CNLs had been found.

More recent research shows that early concerns regarding CNLs are still problematic. Blanchard and Hakobyan (2015) are critical about the idea that countries are assumed to be competitive enough to continue exporting at a similar rate when graduation from GSP is realised. Strikingly, the paper discusses the island Nauru, which was removed from GSP in 1988 because it had reached a high-income status. Soon after the removal, its GDP per capita levels decreased and Nauru returned to the class of middle-income countries. Interestingly and related, it is claimed that ‘’when a developing country loses GSP access, its export in affected industries fall by an average of 19 per cent in the year of exclusion, an additional 20 per cent in the first year, and are still 60 per cent below pre-exclusion levels three years later.’’ Hakobyan (2017) questions whether exporters subject to CNLs are truly competitive enough to continue exporting to the US absent the tariff exemptions under GSP. Similar concerns are raised by Keynes & Bown (2018), who spoke with an American importer who imported duty-free products from Thailand through the GSP program. He indicated that he switched his sourcing to (the still eligible) Cambodia after Thailand's’ GSP eligibility expired.

12

Hypothesis 2: Exporters can maintain similar export levels after the preferential treatment of GSP is lost

Many researchers are suspicious, to say the least, about the well-intended GSP purpose. Some critics even compare nonreciprocal preferences like GSP to a ‘’Faustian bargain’’ (World Bank, 1987; Herz & Wagner, 2011). Apart from the criticism with regards to the CNL, more researchers raised questions about the effectiveness of GSP. Reynolds (2009) claims that GSP can carry negative results because non-GSP related measures will indirectly have an impact on GSP success. If the US tariffs applicable to all countries decrease, GSP eligible countries will lose because their favourable treatment becomes relatively less beneficial. He finds that a 1% reduction in the US tariff that is currently imported duty-free from developing countries will decrease imports of that product from lower-middle-income countries by an average of 2,6%. Furthermore, as Keynes and Bown discuss in their podcast about the economics of trade policy (2018), conflicting interests may cause problems. It is explained that individual senators and congress members can influence the effectiveness of GSP by either delaying the process or even by suspending the program. In 2010, Jeff Sessions, senator of Alabama, removed sleeping bags from the program, because a company in his state did not want to compete with sleeping bags from Bangladesh. This suspended the preferential tariffs for Bangladeshi exporters of the product for nearly a year.

Baldwin and Murray (1977) investigated how the trade benefits of the GSP were distributed among developing countries. The major finding indicates that the regional distribution of the GSP benefits is very unequally distributed. Whereas Asia and Oceania gain, the effects in Africa are minimal. A recent newsletter (UNCTAD, 2018), shows that Baldwin and Murray’s findings in 1977 are still a relevant problem today. The newsletter criticized the distribution of the benefits. It was stated that GSP should ‘’refocus benefits on those countries most in need while reducing benefits provided to countries that have become globally competitive.’’ Furthermore, goods for which developing countries have a competitive advantage (textiles and clothes for instance), are often considered ineligible for the GSP. This makes sense for developed countries, but it does not for developing countries (Panagariya, 2003). Mattoo et al. (2003) even argue that the tariff preferences are driven more by the interests of US multinationals, and not by the desire to contribute to realising development in poorer countries. They accuse GSP granting countries of opportunistic behaviour and claim that GSP is not an instrument to promote exports of developing countries, but rather a way to improve the trade position of industrialized countries. The criticism is especially aimed at the design of the rules of origin criteria. Intermediate inputs imported from the GSP granting country count as local value-added, as required by the requirements of origin. Therefore, the beneficiary is likely to import intermediate goods from the donor country, and this will subsequently lead to double benefits for GSP granting countries: increased exports and cheaper imports.

13

treatment goes against the most favoured nation (MFN) principle of the WTO and this is considered unfair by some non-benefiting, middle income, countries.

One of the most critical studies on GSP was conducted by Herz and Wagner (2011), who conclude that GSP fosters developing countries’ exports in the short run, but that the program hampers exports in the long run. The long-run negative effects are not limited to the beneficiary countries; GSP granting countries promote their own short term exports, whereas their long-run exports decrease. The hard conclusion drawn by Herz and Wagner is that GSP is not a solid instrument to promote sustainable economic growth and development opportunities for poor countries. The argument is that GSP eligibility leads to distortions in the economic structure of the developing countries. It is a temporary preferred treatment, and it is not structurally beneficial. According to them, engaging in GATT/WTO, under reciprocal terms, is a better approach for developing countries. They suggest the short-sighted view could result from political-economical determinants, where interest groups profit while the country as a whole loses. Herz & Wagner (2006) are critical of the structure of the GSP. They claim the industrial GSP granting countries engage in opportunistic behaviour: as long as the concerned products are relatively unimportant, GSP is granted. Yet, as soon as the products become relevant, the benefits are restricted. Each GSP granting country strategically determines which product is ‘’too sensitive’’ to be eligible for GSP. Japan, for instance, excludes rice (Keynes & Bown, 2018). As sceptical as Herz and Wagner are Özden and Reinhardt (2005), who find that developing countries could better fully integrate into the reciprocity-based world trade agreements, rather than continuing GSP-style preferences.

Some researchers have also questioned whether the GSP (and other preferential treatment programs) should actually be interpreted as a non-reciprocal. Typically, GSP granting countries implement side conditions mainly related to human rights, intellectual property rights and labour conditions. Grossman and Sykes (2005) conclude that these side conditions actually bear a resemblance of reciprocity into GSP. It could be interpreted as if GSP donors use the program to disseminate their ideology and impose it on beneficiary countries. These concerns are shared by the former Indonesian trade minister ad professor Mari Pangestu. In the podcast TradeTalks, hosted by Keynes and Bown (2019), she mentions that the US demands being made to Indonesia are heavy: if a beneficiary country starts benefitting significantly from the program, and the US trade deficit rises, more and more requests will be made by the US. With the threat to terminate GSP, the US can make countries obey their rules and regulations. Indonesia is a major beneficiary of the GSP, with 10% of its exports to the US falling under the preferred GSP tariffs. A possible termination would carry negative effects, but exclusion seems realistic.

14

3 - Research design - Empirical Model and Data

The two main methods to investigate the effect of GSP (or trade agreements in general) on trade flows are illustrated and explained below.

Figure 1 Figure 2

The methods in both figures allow to investigate GSP’s effectiveness, but to justify which one was chosen, it is important to briefly discuss each. As illustrated, the method in figure 1 uses country data of, in this case, GSP and trade. This data is not specified at the product level, but rather measures worldwide trade flows and investigates the effect of GSP. It includes all GSP programs, while the method illustrated in figure 2 focuses on the GSP program of the US. Whereas the dependent variable in method 1 would be total world (bilateral) trade, the dependent variable in method 2 would be (unilateral) US imports. Appropriate data to work with method 1 is available at the CEPII database. Method 2 uses detailed data on the product level. For each product from all countries it is investigated whether the US grants GSP, and how this affects its trade volume. It examines the trade flows between the US and other countries in the world and the impact of the preferred treatment. In the second method, countries are connected by arrows to the US, but not to each other. This method can be interpreted as less comprehensive, as it examines fewer country pairs and fewer GSP programs. At the same time, however, it can be perceived as more extensive, as it works with more detailed data at the product level.

For the US, this product-specific data is available at the United States International Trade Commission (USITC). As method 1 has frequently been used in the existing literature through the standard gravity model, see for instance Gil-Pareja et al. (2014), the decision was made to develop and work with method 2. Furthermore, to test GSP expiration effects, detailed product data is more useful. Accordingly, to test the constructed hypotheses, method 2 is better applicable. The analysis examines the unilateral import flows of the US from other countries, and it does not include trade flows between for instance South Africa and Sri Lanka, which would have been the case if the method in figure 1 had been adopted.

3.1 Empirical strategy

15 𝐿𝑛𝐼𝑚𝑝𝑜𝑟𝑡_𝑐𝑝𝑦= α + β₁𝐺𝑆𝑃_𝑐𝑝𝑦+ β₂𝐶𝐵𝐼_𝑐𝑝𝑦+ β₃𝐴𝐺𝑂𝐴_𝑐𝑝𝑦+ β₄𝐴𝑇𝑃𝐷𝐸𝐴_𝑐𝑝𝑦+ β₅𝑊𝑇𝑂_𝑐𝑝𝑦 + β6𝑁𝐴𝐹𝑇𝐴𝑐𝑝𝑦+ β7𝑇𝑃𝑃𝑐𝑝𝑦+ β8𝐺𝑠𝑝𝐶𝑏𝑖𝑐𝑝𝑦 + β9𝐺𝑠𝑝𝐴𝑔𝑜𝑎𝑐𝑝𝑦 + β₁₀𝑊𝑡𝑜𝐺𝑠𝑝_𝑐𝑝𝑦+ β₁₁𝑊𝑡𝑜𝐶𝑏𝑖_𝑐𝑝𝑦+ β₁₂𝑊𝑡𝑜𝐴𝑔𝑜𝑎_𝑐𝑝𝑦+ 𝐹_𝑐 + 𝐹_𝑦+ 𝐹_𝑝 + ε_𝑐𝑝𝑦 (1) where 𝑐 denotes country, 𝑝 denotes product and 𝑦 stands for year. Αlpha is a common intercept, and the standard error term ε comprises all factors affecting the dependent variable other than the included independent variables. It is impossible to capture all relevant variables and some variables are omitted or unpredictable due to random human behaviour or other unexpectancies. The dependent variable in this model, 𝐿𝑛𝐼𝑚𝑝𝑜𝑟𝑡_𝑐𝑝𝑦, refers to US log imports from country 𝑐, for product 𝑝 in year 𝑦. The natural log of US imports is taken as this allows interpreting the results in percentage change instead of unit change. Besides, the logarithmic transformation normalizes the dataset. The other variables are dummy variables and defined as follows: 𝐺𝑆𝑃_𝑐𝑝𝑦 is 1 if GSP is granted to the country for the product in that specific year. If not, the value is 0. The identical intuition applies to the other non-interaction variables. To find the full name of the agreements used in the other variables instead of the abbreviation, consult table 2. 𝐺𝑠𝑝𝐶𝑏𝑖_𝑐𝑝𝑦 is an interaction variable which takes the value of 1 if the country-product-year combination is valid for both GSP and CBI, and zero otherwise. The other interaction variables have a similar meaning. The main variable of interest in this specific model is the GSP-dummy. Testing this model will generate relevant information about the impact of various (preferential and free trade) agreements on exports to the US. The general expectation is that countries that are awarded preferential tariff treatment should experience increased export numbers (to the US) as a result of the lower tariffs when compared to competing non-GSP countries (Reynolds, 2009). The second expectation is that the GSP program will statistically be the most beneficial preferential trade program when compared to CBI or AGOA since the current literature generally agrees on the positive influence of GSP.

16

Apart from these single fixed effects, interactive fixed effects are included to allow for heterogeneity in the level of exports of any product from any country in a year. Consider a country that either gained or lost favourable tariffs through GSP at a time when the economic conditions were either deteriorating or improving, resulting in decreasing or increasing exports to the US. These interactive fixed effects are useful as they accurately attribute country- or product-level trends in exports to the impact of GSP designation or expiration. Not including fixed effects would have resulted in biased outcomes. Think about the following imaginative example. Country X loses GSP access in 2009 at the time of the financial crisis and experiences lower exports to the US after the expiration. This country X is hit harder by the financial crisis than other countries. Not including the fixed effects would attribute the entire loss in exports to the GSP termination. At the same time, however, worldwide trade was hit by the financial crisis and these sorts of shocks are incorporated by the interactive fixed effects in the analysis. The empirical specification includes a set of country+product, country+year and product+year fixed effects.

3.2 Data collection process

To test the hypotheses, which aim to investigate the effectiveness of GSP and the consequences of its expiration, a large dataset1 was constructed using trade and tariff data at the US Harmonized Tariff Schedule (HTS) 8-digit level, extracted from the USITC data web. The USITC records product-specific tariffs of all US imports from all countries. The USITC data was used since alternative sources (such as the United Nations and World Bank) do not disaggregate imports by import programs as CBI, AGOA etc. The data used in the model are import statistics. The US carefully track its imports per country, even disaggregated per import (preferential) program. It would have been impossible to conduct this analysis from the perspective of, and with data of the developing countries. Data on developing countries is often either not reliable or not provided at all. The essence, nevertheless, is the same; investigating US imports from developing countries or exports of developing countries to the US makes no difference.

US import trade data from all countries was collected at the USITC for the years 1997-2016. Appending all these years and countries into one file resulted in a dataset involving 3.775.027 observations and 4 variables: country, year, product type (by HS8 number) and US imports. The second dataset includes the tariff rates per product and whether the product is eligible for certain programs as CBI, GSP etc. This dataset contains all HS8 products imported by the US from 1997 to 2016, and the tariffs that applied. Furthermore, the countries that are exempted from preferential tariffs for a specific product are listed. In 2010, for example, Argentina was excluded from GSP benefits for milk protein concentrates (HTS8: 04049010). Argentina, Brazil and India were the countries excluded most frequently. All country GSP exclusions per product are incorporated in the data. The process of ‘’cleaning’’ the data was time-consuming, yet important to find valid results.

17

In the data, I also take into account that the member composition of preferential trade agreements (PTAs) and free trade agreements (FTAs) continually changes. In 1997 and 1998, for instance, Gabon and Mongolia were not eligible for the GSP. In 1999, both countries were designated for the GSP. The same holds for Eritrea and Nigeria, who were designated for the preferential treatment in 2000. Similarly, the WTO has been expanding. Currently, the WTO consists of 164 countries and the most recent joiners were Afghanistan and Liberia in 2016 (WTO, 2019). It is important to account for the adjustments in all program compositions because the composition of the programs in 1997 is very different from the compositions in 2016. The US GSP had 148 beneficiary countries in 1997 and 119 in 2019. Besides, this analysis allows specifying whether a country is included in several agreements at the same time. Whether a country is a GSP beneficiary or both a GSP beneficiary and part of the CBI program, makes a difference, that should, and will be incorporated in the analysis. Table 2 lists all PTAs and FTAs included in the empirical analysis.

Table 2 Programs included in the analysis

Program Abbreviation PTA or FTA # of members*

Generalized System of Preferences GSP PTA 119

Caribbean Basin Initiative CBI PTA 16

African Growth and Opportunity Act

AGOA PTA 39

Andean Trade Promotion and Drug Eradication Act

ATPDEA PTA 0

World Trade Organization WTO FTA 164

North American Free Trade Agreement

NAFTA FTA 3

Trans-Pacific Partnership TPP FTA 12

*in December 2019

Sources: Preferential Trade Arrangements Database (PTADB), WTO.org, USITC.gov

18

The decision to work with data from 1997 until 2016 was made since the USITC data web does not contain data on preferential programs dating back further than 1997. All countries that the US imported from during the period of research are included in the analysis and depicted in table 3. Product data was collected at the HS 8-digit level since this provides detailed and specific information. Besides, one of the most respected researchers in this topic area, Hakobyan, also uses product-specific data at the 8-digit level in her work on the impact of CNLs (2015).

Table 3 Countries included in the analysis

Afghanistan, Albania, Algeria, Andorra, Angola, Anguilla, Antigua Barbuda, Argentina, Armenia, Aruba, Australia, Austria, Azerbaijan, Bahamas, Bahrain, Bangladesh, Barbados, Belarus, Belgium, Belize, Benin, Bermuda, Bhutan, Bolivia, Bosnia-Hercegov, Botswana, Br Virgin Is, Br Indian O Ter, Brazil, Brunei, Bulgaria, Burkina Faso, Burma (Myanmar), Burundi, Cambodia, Cameroon, Canada, Cape Verde, Cayman Is, Cen African Rep, Chad, Chile, China, Christmas Is, Cocos Is, Colombia, Comoros, Congo (DROC), Congo (ROC), Cook Is, Costa Rica, Cote d`Ivoire, Croatia, Cuba, Curacao, Cyprus, Czech Republic, Denmark, Djibouti, Dominica Is, Dominican Rep, Ecuador, Egypt, El Salvador, Eq Guinea, Eritrea, Estonia, Ethiopia, F St Micronesia, Falkland Is, Faroe Islands, Fiji, Finland, Fr Polynesia, Fr S & Ant land, France, French Guiana, Gabon, Gambia, Gaza Strip, Georgia, Germany, Ghana, Gibraltar, Greece, Greenland, Grenada Is, Guadeloupe, Guatemala, Guinea, Guinea-Bissau, Guyana, Haiti, Heard & McDn Is, Honduras, Hong Kong, Hungary, Iceland, India, Indonesia, Iran, Iraq, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kenya, Kiribati, Korea, Kosovo, Kuwait, Kyrgyzstan, Laos, Latvia, Lebanon, Lesotho, Liberia, Libya, Liechtenstein, Lithuania, Luxembourg, Macao, Macedonia, Madagascar, Malawi, Malaysia, Maldive Is, Mali, Malta, Marshall Is, Martinique, Mauritania, Mauritius, Mayotte, Mexico, Moldova, Monaco, Mongolia, Montenegro, Montserrat Is, Morocco, Mozambique, Namibia, Nauru, Nepal, Netherlands, Netherlands Ant, New Caledonia, New Zealand, Nicaragua, Niger, Nigeria, Niue, Norfolk Is, North Korea, Norway, Oman, Pakistan, Palau, Panama, Papua New Guin, Paraguay, Peru, Philippines, Pitcairn Is, Poland, Portugal, Qatar, Reunion Romania, Russia, Rwanda, Samoa, San Marino, Sao Tome & Prin, Saudi Arabia, Senegal, Serbia, Serbia pre-2009, Serbia/Monteneg, Seychelles, Sierra Leone, Singapore, Sint Maarten, Slovak Republic, Slovenia, Solomon Is, Somalia, South Africa, South Sudan, Spain, Sri Lanka, St Helena, St Kitts-Nevis, St Lucia Is, St Pierre & Miq, St Vinc & Gren, Sudan, Sudan pre-2012, Suriname, SvalbardMay Is, Swaziland, Sweden, Switzerland, Syria, Taiwan, Tajikistan, Tanzania, Thailand, Timor-Leste, Togo, Tokelau Is, Tonga, Trin & Tobago, Tunisia, Turkey, Turkmenistan, Turks & Caic Is, Tuvalu, Uganda, Ukraine, United Arab Em, United Kingdom, Uruguay, Uzbekistan, Vanuatu, Vatican City, Venezuela, Vietnam, Wallis & Futuna, West Bank, Western Sahara, Yemen, Zambia, Zimbabwe

Notes:

(i) All US imports at the 8hts level are included in the analysis for all these countries.

(ii) Abbreviations are used for some country names. The country names are kept consistent

19

The three different datasets (the first including trade data, the second consisting of tariff data and the third including program composition per year) are merged to construct the final panel dataset. This is the dataset that is used to test the model and it contains 27 variables and 3.577.251 observations. The dataset includes detailed information on preference eligibility per product and program. The disadvantage of a large dataset is that large samples can be cumbersome to work with. The advantage is that a large sample improves the accuracy of the results and mitigates the risk of statistical biases. This advantage outweighs the disadvantage and accordingly, it was decided to work with this large dataset.

3.3 Method hypothesis 2

To provide more clarity about the second hypothesis regarding the GSP expiration, the intuitive idea will be explained here by investigating some recent country graduations. More specifically, the effect of expiration on industries that exported under the GSP program is examined. Initially, data in nominal values per country (from USITC) on all products that have been exported under the GSP was extracted. Usually, the exports under GSP of one product type are not the complete exports. This product will also be exported under normal conditions. For instance, two of the products Equatorial Guinea exports to the USA are petroleum oils and oils from bituminous minerals, crude (HS 27090020). In 2006, 80% of the total exports of this product had been exported through GSP. So, when a product receives GSP, it does not mean all exported products in the product category receive the preferential status. Therefore, to construct the graphs, and to execute statistical tests, firstly the GSP eligible products are identified, but then the analysis will continue with the numbers of total exports of the specific products. This is necessary because results would not reflect reality if the analysis had been conducted with mere export numbers under GSP. The exports under GSP, after all, fall to zero after expiration.

20

Graph 1 Data used to make this graph extracted from USITC

Apart from two recent GSP graduations, it is interesting to briefly discuss the case of Ukraine. This country received the preferential GSP treatment for a lengthy period but was suspended from its benefits in 2001 due to a violation of intellectual property rights. In 2006, the GSP was reinstated by President Bush (GSP handbook, 2016). Even though the Ukrainian case is not explicitly similar to the graduations that are discussed earlier, it will be thought-provoking to find out how the Ukrainian industries were, or where not, affected by the GSP suspension in the period of 2002-2005. The idea, again, is that Ukraine should be able to maintain comparable export numbers in all sub-periods. The periods are 1997-2001 when Ukraine received GSP, 2002-2005 when the GSP was terminated, and 2006-2011 when GSP was reinstated. Data on 582 HTS8 products was collected to construct graph 2, which depicts the upward trend of Ukrainian exports to the USA. This upward trend was temporarily interrupted in the period when GSP was not granted. It is tempting to conclude that Ukraine was substantially harmed by the GSP suspension. Section 4 will present more formal results to investigate whether the exports in GSP and non-GSP periods differ significantly. The appendix lists all industry numbers at the HTS 8-digit level used to create the presented graphs.

21

It is important to keep in mind that graduation from and suspension of GSP are not identical. Whereas suspension is imposed when a beneficiary violates workers’ rights or intellectual property rights, graduation occurs when a beneficiary country is assumed to be developed enough to be competitive without the preferential treatment. Both forms of GSP termination are analyzed. Croatia and Equatorial Guinea graduated in 2011, and Ukraine’s GSP was temporarily suspended. The logical expectation is that the termination impact for Ukraine would be more extreme than for the two graduated countries. How countries react to GSP exclusion will be insightful knowledge for national governments, since it provides information on how beneficial it is to engage in this preferential program which sooner or later expires. Besides, it is an evaluation of the program in the sense that it tests whether the statement that countries are competitive enough after graduation is justified.

As previously mentioned, India’s preferential status has recently been terminated. Given that India extensively used the GSP program to export to the US market under favourable terms, it would be interesting to investigate how the Indian industries are affected by the recent program expiration. Due to the recency of the termination, this has not been possible at the moment of writing, but it is an interesting topic for future research. The expectation is that the withdrawal will hurt small and medium-sized businesses in sectors like jewellery, auto parts, pharmaceuticals and chemicals. It could be difficult for businesses in these industries to survive price-based competition with suppliers from other developing countries who remain GSP beneficiaries (Pandit, 2019). Results of the aforementioned exclusions, which will be presented in the following section, could give an indication of what India can expect. Country size and the extent to which GSP is utilized among these countries is, however, different, so the consequences of the GSP expiration could also differ.

22

4 - Analysis and discussion of results

4.1 Hypothesis 1

To discuss the first hypothesis, which aims to investigate whether GSP eligibility is associated with increased trade numbers, a fixed effects regression was executed. In this linear regression, the natural log of US imports is the dependent variable and GSP is the independent variable. Table 4 reports the results of this simplified regression.

Table 4 The impact of GSP eligibility on the natural log of US imports

(1) (2) (3) (4) (5) No FE Country FE Year FE Country + Year FE Country * Year FE GSP 0.254*** 0.197*** 0.251*** 0.198*** 0.201*** (79.70) (65.45) (78.45) (65.69) (22.48) Constant 11.43*** 11.45*** 11.43*** 11.45*** 11.45*** (5963.42) (6347.49) (5955.30) (6337.75) (1953.09) N 3577251 3577251 3577251 3577251 3577205 R2 0.002 0.119 0.002 0.119 0.124 t statistics in parentheses * _{p < 0.05,} ** _{p < 0.01,} *** _{p < 0.001}

23

Table 5 The effect of various agreements on the dependent variable log US imports

(1) (2) (3) (4) (5) (6) (7)

No FE Ctry FE Y FE Ctry and Y FE Ctry and Prod FE Y and Prod FE Ctry Prod Y FE GSP -0.980*** 0.472*** -0.928*** 0.461*** 0.523*** -1.212*** 0.509*** (-71.38) (21.01) (-67.49) (20.46) (27.31) (-98.58) (26.55) CBI -1.341*** 0.0963 -1.266*** 0.0523 -0.0377 -2.028*** -0.0723 (-45.00) (0.34) (-42.47) (0.18) (-0.16) (-75.94) (-0.30) AGOA -2.016*** _{0.196 -1.860}*** _{0.250 0.167 -2.351}*** _0.217 (-11.17) (0.80) (-10.31) (1.02) (0.80) (-14.61) (1.04) ATPDEA -0.608*** _0.139*** _-0.629*** _0.143*** _0.150*** _-0.675*** _0.142*** (-51.61) (5.88) (-53.36) (6.03) (7.47) (-63.72) (7.03) WTO 0.233*** 0.326*** 0.323*** 0.425*** 0.406*** 0.455*** 0.483*** (26.15) (27.91) (35.70) (35.08) (40.64) (56.16) (46.84) NAFTA 1.409*** 0 1.386*** 0 0 1.510*** 0 (231.94) (.) (227.74) (.) (.) (274.24) (.) TPP -0.0820*** -0.0626*** 0.0901*** 0.00420 -0.0732*** 0.0414** -0.0143 (-5.32) (-4.18) (5.35) (0.26) (-5.74) (2.76) (-1.02) GspCbi 0.471*** -0.0325 0.414*** -0.0201 0.0849* 0.763*** 0.0898* (14.41) (-0.76) (12.67) (-0.47) (2.31) (26.12) (2.45) GspAgoa 1.320*** -0.334 1.237*** -0.310 -0.451* 1.726*** -0.451* (7.57) (-1.61) (7.10) (-1.49) (-2.55) (11.11) (-2.55) WtoGsp 0.453*** -0.619*** 0.384*** -0.666*** -0.676*** 0.655*** -0.713*** (31.77) (-27.71) (26.91) (-29.75) (-35.51) (51.27) (-37.36) WtoCbi 0.399*** 0.302 0.328*** 0.299 0.354 0.452*** 0.345 (10.64) (1.05) (8.75) (1.04) (1.44) (13.50) (1.41) WtoAgoa -0.140** _{0.00748 -0.183}*** _{-0.00746 0.218}* _-0.530*** _0.202* (-2.91) (0.07) (-3.81) (-0.07) (2.41) (-12.35) (2.24) Constant 11.40*** _11.23*** _11.32*** _11.15*** _11.15*** _11.22*** _11.09*** (1308.83) (974.35) (1284.10) (944.28) (1133.22) (1420.76) (1103.25) N 3577251 3577251 3577251 3577251 3577184 3577184 3577184 R2 0.036 0.118 0.038 0.119 0.366 0.241 0.367 t statistics in parentheses * _{p < 0.05,} ** _{p < 0.01,} *** _{p < 0.001}

24

Column 1 reports the results of the equation without fixed effects. Columns 2 to 7 include various sets of country, year and product fixed effects. The estimated coefficients suggest that several agreements positively and significantly boost exports to the US. For all specifications where country fixed effects are added, the GSP coefficient is significantly positive. The inclusion of country fixed effects is critical as it controls for the average differences across countries in any observable or unobservable predictors, such as differences in quality of institutions or geographical location. It soaks up across-group action. The fact that Chad, for instance, is landlocked is a characteristic that does not change over time but should be accounted for since it might impact Chad’s exports to the US. A lot of country characteristics remain constant during the study period (1997-2016) and the country fixed effects attempt to control for these characteristics. Therefore, it is predictable that the PTA coefficients (GSP, CBI, AGOA and ATDPEA) are negative in column 1 when no fixed effects are applied. The countries that receive this preferred treatment are expected to be relatively poor and subsequently export lower volumes of goods to the US. The eligible countries typically export products of lower quality at a lower price. Furthermore, weak institutions and a bad state of the infrastructure have negative effects on exports. To correct for this, the fixed effects have been included. After including fixed effects, the results for the four mentioned PTAs change tremendously.

Column 7 indicates that the positive effect of GSP on exports remains after including country, product and year fixed effects. These results are supportive of hypothesis 1 and therefore the hypothesis that the preferential treatment received from GSP eligibility increases exports of beneficiary countries cannot be rejected. The columns where year fixed effects are included reveal opposing results. Including just year fixed effects is not enough. The low R2 for the year fixed effects specification (0.038) implies that the year fixed effects specification is not the most important when explaining the impact of the programs on US imports. Table 5 also provides results on GSP interactive variables. GspCbi is a dummy variable which is 1 if the country and the product are both eligible for the GSP and the CBI in a specific year. Even after including country-year-product fixed effects, the coefficient is positive.

25

with a 23.3% increase in exports to the US. The results for the TPP are less convincing and this might have been the motivation for the US to withdraw from it.

The GSP is the most famous and, according to several researchers, the least controversial PTA. The results in the table support this. The findings (before and after applying fixed effects) are more positive for GSP than for CBI or AGOA. The interaction variables are not always statistically significant, but it can be concluded from the results that PTAs such as CBI and AGOA work better when used in combination with membership of GSP or WTO. This is a desirable result. As mentioned earlier, all CBI and AGOA beneficiaries are also GSP eligible. The CBI and AGOA programs are constructed to grant an even more favourable treatment to the least developed GSP beneficiaries in the Caribbean area (CBI) or Sub-Saharan-Africa (AGOA). Therefore, it never happens that a country is exclusively CBI or AGOA eligible, without being a GSP beneficiary. If a country is CBI or AGOA eligible, it is at the same time GSP eligible. Being aware of this, it can be concluded that the single CBI and AGOA variables are not so important to investigate. The interaction variables GspCbi and GspAgoa are more relevant here, and this is why these variables have been created. In most of the specifications, the coefficients of these two interaction variables are positive, especially for the CBI. The results generated in table 5 go further than just the GSP program. When the PTAs are interacted with the WTO, some estimates turn insignificant but most of the significant results are positive. The ATPDEA is the fourth and final PTA included in the empirical analysis. With Ecuador’s “graduation” in 2013, the program has not been in use since then. To generate more information about the effectiveness of PTAs other than GSP, it was decided to still include it in the analysis. The results, after including country fixed effects, indicate that this PTA was significantly successful in increasing beneficiary countries’ exports to the US. If no fixed effects are included, or when the country fixed effects are absent, the results are significantly negative.

4.2 Hypothesis 2

The aim of hypothesis 2 has been described and explained. Graphs 1 and 2 have shown interesting export patterns that need to be investigated in more detail. A clear difference in the trend line of export numbers was perceived between the GSP eligible period the post-GSP period. The following section will test for 25 countries whether the GSP and post-GSP differences are statistically significant by investigating export numbers per country in periods of six years: the three last years of GSP eligibility are compared to the first 3 years after GSP expiration. This will allow to draw more substantiated conclusions about the impact of GSP graduation on a country. These results are relevant for governments of developing countries that are evaluating the success of this preferential treatment. Some researchers are concerned that the drop in exports after expiration is so detrimental, that it obliterates all the good the preferred tariffs have brought.

26

and the “PostGSP” period. If this number is positive, it means the average exports had been higher in the period when the country was GSP eligible. If the difference is negative, it means that the country actually exported more to the US after GSP expired. The hypothesis tested is that the GSP average and PostGSP average are the same. This is, after all, the purpose of the program: after graduating, the countries are expected to be developed enough to compete without the preferential tariffs. The “P_value” provides information on whether this hypothesis is acceptable, or whether it should be rejected. A high p-value is evidence that the hypothesis that GSP and post-GSP exports are the same cannot be rejected. A low p-value provides evidence that the exports under GSP and exports after GSP termination are (significantly) different; either lower or higher.

Table 6 - T-test results to compare export numbers per country pre and post GSP expiration

Country Expiration year Obs GSP av PostGSP Dif St_Err T_value P_value

Ukraine 2002 582 479000 129000 350000 183000 1.9 .056 Malta 2002 82 673000 375000 298000 159000 1.9 .064 Slovenia 2003 393 354000 343000 11114.56 36233.09 .3 .759 Chile 2004 615 857000 1370000 -513000 212000 -2.4 .016 Czech Rep 2004 1003 482000 740000 -258000 64008.75 -4.05 0 Hungary 2004 620 892000 1030000 -139000 120000 -1.15 .247 Estonia 2005 136 249000 180000 69005.56 51148.04 1.35 .179 Poland 2005 1049 551000 724000 -174000 53455.23 -3.25 .001 Slovak Rep 2005 303 313000 373000 -60300 44438.06 -1.35 .176 Morocco 2006 264 157000 219000 -61500 46261.75 -1.35 .185 Guatemala 2007 242 1140000 1380000 -238000 152000 -1.55 .12 El Salvador 2007 119 537000 645000 -109000 81335.54 -1.35 .185 Honduras 2007 128 3580000 3970000 -383000 418000 -.9 .36 Dom Rep 2007 339 1870000 1980000 -119000 312000 -.4 .704 Bulgaria 2007 360 165000 119000 46384.21 29987.89 1.55 .123 Romania 2007 381 821000 707000 114000 138000 .85 .409 Peru 2009 555 929000 1030000 -98400 117000 -.85 .402 Croatia 2011 279 381000 203000 178000 168000 1.05 .29

Eq Guinea 2011 16 1.30e+08 2.73e+07 1.03e+08 1.03e+08 1 .334

Argentina 2013 707 811000 676000 135000 89144.27 1.5 .131 Colombia 2013 2037 318000 340000 -22100 82791.05 -.25 .79 Bangladesh 2014 171 209000 259000 -49900 31268.41 -1.6 .113 Russia 2015 486 1250000 1040000 215000 138000 1.55 .118 Uruguay 2017 160 600000 318000 281000 128000 2.2 .029 Venezuela 2017 135 712000 507000 205000 177000 1.15 .251 Notes:

(I) For all countries, the study period is six years. Three years before the expiration and three years after the expiration are included. This is not the case for Ukraine, where 3 subperiods are measured: 1997-2001, 2002-2005 (the period where GSP was withdrawn) and 2006-2010. This is not the case either for both Uruguay and Venezuela. Due to their recent graduation (2017), it was decided to measure 4 years. Two years pre and two years post GSP expiration. Lastly, for some countries, the expiration year has been excluded, since the exclusion was not abrupt. For the Czech Republic and the Dominican Republic, this transition year has been excluded. Finally, the study period of Eq Guinea covers 10 years.

(II) “GSP av” and “PostGSP” are the average exports per industry per year. Only the products that used to export under GSP, when the country was eligible, are included.

27

The table contains information on 25 country graduations or suspensions. 12 of these experienced lower exports to the US after the GSP eligibility was withdrawn. In Ukraine, Malta and Uruguay, the negative difference is statistically significant at the 10% level. It can be concluded that the industries in these three countries that used to export under the preferred GSP tariffs were, with regards to exports to the US, substantially harmed by the GSP withdrawal. Bulgaria (0.123), Argentina (0.131) and Russia (0.118) also have relatively low p values, so, although not significant at the 10% level, the GSP termination is likely to have had negative effects on their exports to the US. The findings for these countries are in line with the conclusions by Blanchard and Hakobyan (2015), who claim that “when a developing country loses GSP access, its export in affected industries fall by an average of 19 per cent in the year of exclusion, an additional 20 per cent in the first year, and are still 60 per cent below pre-exclusion levels three years later”.

The other 13 countries, however, show increased exports after the GSP terminates. In the cases of Chile, the Czech Republic and Poland, the export increases were statistically significant at the 5% significance level. In many of the 13 cases, a new agreement was signed after the GSP expired. Peru graduated from GSP in 2009 but experienced increased exports to the US afterwards. This has to do with the fact that the United States-Peru Free Trade Agreement (PTPA) entered into force on February 1, 2009, with the aim to promote Peruvian-American trade. It turned out to be successful in increasing exports (USITC, 2019). The same applies to the Czech Republic and Poland, who entered the European Union after GSP expired. As a result of their entrance, trading with the US became easier. It would not be correct to state that exports increased as a result of the GSP expiration. More factors play a role and these new programs should be taken into consideration when evaluating GSP’s effectiveness.

To come back to graphs 1 and 2, which were presented earlier, it can be concluded that these patterns are slightly extreme and not completely representative of all GSP expirations. The depicted drop in Ukrainian exports was statistically significant, but the drops for Equatorial Guinea and Croatia were not. The results indicate that the exports of suspended countries (Ukraine and Argentina) are, on average, hit harder than graduated countries. This finding could motivate GSP eligible countries to put effort to avoid GSP-suspension. How a country is affected by a GSP expiration has to do with many factors. In the sample, half of the countries were able to maintain or even increase exports to the US after GSP withdrawal. The other half, however, struggled to maintain similar export rates.