Effect of the Utilization of Non-Reciprocal Trade Preferences offered by the QUAD on Economic Complexity in Bene ciary Countries


 This article aims to contribute to the nascent literature on the effect of non-reciprocal trade preferences (NRTPs) on industrialization in beneficiary countries. In so doing, it complements the few existing works on the effect of NRTPs on export product diversification by investigating the effect of NRTPs (both the Generalized System of Preferences- GSP programs- and other non-reciprocal trade preferences) offered by the QUAD countries on the level of economic complexity in beneficiary countries. The analysis has relied on 110 beneficiary countries of these NRTPs over the period 2002–2018, and made primarily use of the two-step system Generalized Methods of Moments estimator. The findings are quite interesting. First, beneficiary countries tend to use GSP programs (rather than other trade preferences) to achieve greater economic complexity, and the positive effect of the utilization of GSP programs on economic complexity is higher for high income beneficiary countries than relatively less advanced beneficiary countries. Second, both GSP programs and other non-reciprocal trade preferences are strongly complementary in promoting economic complexity in beneficiary countries, in particular if their usage reach high levels. Third, the utilization of NRTPs enhances economic complexity in countries that receive high foreign direct investment flows. Finally, development aid flows are strongly complementary with the utilization of NRTPs in fostering economic complexity in beneficiary countries, especially for high amounts of development aid. This suggests the need for preference-granting countries (that are also suppliers of development aid) to offer both generous NRTPs and higher development aid flows if those NRTPs are to be effective in expanding the manufacturing base in the beneficiary countries.Jel Classification: F13; F14; O14.


Introduction
For more than half a century, developing countries have been enjoying non-reciprocal trade preferences (henceforth referred to as "NRTPs") offered by industrialized countries. The second conference of the United "Quadrilaterals" (also referred to as QUAD countries) on the level of economic complexity of bene ciary countries. The QUAD countries is comprised of Canada, European Union (EU), Japan and the United States of America (US).
As of today, NRTPs covers not only GSP schemes, but also non-reciprocal trade concessions provided by developing countries to LDCs (see WTO, 1999; and any other any other NRTP regimes authorised through a Waiver under the WTO[13] Agreement. In addition to their GSP schemes, a number of developed countries, including QUAD countries provide several NRTPs to selected developing countries, under a special WTO Waiver. For example, the EU currently offers a non-reciprocal preferential treatment to products originating from the Western Balkans. The US offers the African Growth and Opportunity Act (AGOA) to eligible countries in Sub-Saharan African (SSA); the "Caribbean Basin Economic Recovery Act" to Caribbean countries, and particularly the "Hemispheric Opportunity through Partnership Encouragement initiative" to Haiti; some non-reciprocal trade concessions trade preferences to Nepal; and the "Former Trust Territory of the Paci c Islands" to four countries, including Marshall Islands, the Federated States of Micronesia, the Northern Mariana Islands, and Palau. The WTO Preferential Trade Arrangements database contains all NRTPs, including both current ones and expired ones. It could be accessed online at: http://ptadb.wto.org/default.aspx The contribution of the present study to the extant literature on the bene ciary countries' export performance (in particular the export diversi cation) effects of NRTPs is three-fold.
First, it focuses on the effect of NRTPs on economic complexity rather than merely on the export diversi cation effect of NRTPs, as investigated by Gamberoni (2007), Persson and Wilhelmsson (2016) and Yannopoulos (1986).
Second, previous studies used either simple statistical indicators (e.g., Yannopoulos, 1986) or dummies that represent a country's eligibility for a given preference programme in a single year, to identify the export diversi cation effect of NRTPs (Gamberoni, 2007;Persson and Wilhelmsson, 2016). In contrast, the present analysis relies on the utilization rate of NRTPs to examine the economic complexity effect of NRTPs. Thus, rather than merely assessing the effect of the eligibility to NRTPs on countries' export product diversi cation (as done by Gamberoni, 2007;Persson and Wilhelmsson, 2016), the present work considers the effect of the utilization (rates) of NRTPs on economic complexity. Thus, the differences between the current analysis and the previous works lie not only on the dependent variable used (which is 'economic complexity' here, but 'export diversi cation' in previous analyses), but also on the fact that previous works considered the effect of eligibility to a NRTP, whereas the current work touches upon the effect of the utilization (rate) of NRTPs. In fact, 'eligibility' to a NRTP does not necessarily involve its utilization.
Third, and nally, in contrast with Gamberoni (2007) and Persson and Wilhelmsson (2016) who have used gravity models (and hence relied on a country-pair/year framework) in their analyses, the present analysis uses a country-year framework.
The focus of the present study on the NRTPs provided by QUAD countries is dictated by the fact that the most comprehensive existing dataset on the utilization rates of NRTPs by bene ciaries countries covers only the QUAD countries (as providers of these NRTPs). This dataset has been made recently available publicly by the UNCTAD [14].
The empirical analysis has used the two-step system Generalized Methods of Moments (GMM), and revealed several ndings. Take in isolation, GSP programs contribute to enhancing economic complexity, while other trade preferences by the QUAD countries rather adversely affect negatively economic complexity. However, both GSP programs and other trade preferences offered by the QUAD countries are strongly complementary in fostering economic complexity when they reach high utilization rates. These NRTPs also promote the development of complex products when bene ciary countries enjoy high FDI in ows. Finally, development aid plays a crucial role in the relationship between the utilization of NRTPs and economic complexity. Especially, development aid and the utilization of NRTPs are strongly complementary in enhancing economic complexity in countries that experience high amounts of development aid.
The remaining part of the analysis contains six sections. Section 2 provides a brief literature review on the effect of NRTPs on export earnings and export diversi cation. Section 3 discusses the effect of NRTPs on economic complexity. Section 4 deepens this discussion by analysing theoretically whether FDI in ows could help bene ciary countries of NRTPs make use of these NRTPs to develop complex products. Section 5 presents the estimation strategy. Section 6 interprets empirical results, and Section 7 concludes.
[2] The history of the GSP is provided by Cunha et al. (2005) and an overview on the legal and historical background of trade preferences could be found in Persson (2015).
[3] See Grossman and Sykes (2005) for a discussion on the legal and economic aspects, as well as on the European and US GSP scheme.
[5] See for example, Hartmann et al. (2017); Le Caous and Huarng (2020). It is worth highlighting that Kang-Kook and  have reached the opposite conclusion, whereby economic complexity does not reduce income inequality, but rather increases it.
[14] The dataset is accessible online at: https://gsp.unctad.org/home It is worth noting that while the WTO database on preferential trade arrangements also contains a wealth of information on NRTPs, the time coverage of the WTO dataset is lower than the one of the UNCTAD.

Brief Literature Review On The Effect Of Nrtps On Export Earnings And Export Diversi cation
The voluminous work on the export earnings (or export growth) effect of NRTPs has been inconclusive. In fact, some works [15] have reported a positive effect of NRTPs on bene ciary countries' exports to the markets of preference granting countries, and others [16] have established heterogenous outcomes across bene ciary countries, sectors and products. In the meantime, many other works have obtained a negative effect of NRTPs on bene ciaries' exports [17], and have, therefore, casted doubt on the effectiveness of NRTPs in terms of promoting exports of bene ciary countries, with some arguing that they should be replaced with reciprocal agreements (e.g., Admassu, 2020;Gil-Pareja et al., 2019;Özden and Reinhardt, 2005;Zappile, 2011). For example, a nding common to Gil-Pareja et al. (2019), Admassu (2020) and Zappile (2011) is that reciprocal trade agreements tend to be more effective in terms of promoting exports of bene ciary countries (e.g., African countries) than NRTPs. This situation is attributed, inter alia, to the uncertainty surrounding the expiration of these preferences, the erosion of preferential margins. This is, to some extent, con rmed by Hakobyan (2020) who has found that the 2011 expiration of the USA GSP has had adversely affected developing countries' exports to the USA market. Additionally, exports did not fully recover by 2012, thereby suggesting that the adverse export effect of the 2011 expiration of the USA GSP has been persistent over time. Herz and Wagner (2011) have obtained that while the GSP schemes have been associated, on average, with a 4% lower exports by developing countries, the impact of this scheme on developing countries' exports appeared nonetheless to be positive if the scheme existed for less than 10 years, and negative if the scheme existed for a long period (i.e., one or two decades). The negative effect could be attributed to the strict or complicated rules of origin, which in the long term, exert distortive effects on exports from developing countries, that ultimately export under most favoured nation (MFN) tariffs rather than under the non-reciprocal GSP programs [18]. Özden and Reinhardt (2005) have reported that countries that were removed from GSP tended to adopt more liberal trade policies than countries that remained eligible to this preference scheme.
As also noted above, the literature on the effect of NRTPs on export product diversi cation is very limited (e.g., Gamberoni, 2007;Persson and Wilhelmsson, 2016;Yannopoulos, 1986). Yannopoulos (1986) has explored the effect of NRTPs on export earnings, export product diversi cation and foreign direct investment in bene ciary countries. As far as the export product diversi cation effect of NRTPs is concerned, he has used the trade similarity index developed by Finger and Kreinin (1979) to compare the similarity of export patterns for countries over several years. He has observed a greater similarity of trade ows between 1963 to the year 1970 than from 1970 to 1979. He has then concluded that the preferences offered by the EU to Mediterranean countries have led to export diversi cation. However, as noted by Persson and Wilhelmsson (2016), relying on the trade similarity index of Finger and Kreinin (1979) to derive such a conclusion could be misleading. This is because the decrease in the values of the index may be due to a shift from a greater diversi cation in the beginning of the period to a greater specialization at the end of the period, as a result of NRTPs. Gamberoni (2007) has investigated the effect of the NRTPs offered by the EU over the period 1994-2005 on the bene ciary countries' extensive margin of trade. His analysis has covered 118 developing countries' exports of total product-level (HS 6-digit) to ten EU countries (EU12 minus Belgium and Luxembourg). The author has used dummies for GSP, GSP speci cally for LDCs, the drug regime within GSP11 and, nally for trade preferences in favour of ACP countries to identify where an effect of the trade preferences on the extensive margin lies. Using the Probit and Tobit econometric techniques, he has found that the GSP and the drug regime have contributed to export diversi cation at the extensive margins, but for the ACP countries, trade preferences have had an anti-export diversi cation effect. For LDCs, the effects are unstable and vary across speci cations. Persson and Wilhelmsson (2016) is the most recent work that has really tackled the issue of export product diversi cation effect of NRTPs. Their study has covered all preference schemes implemented by the EU during the period 1962-2007. Using the Poisson pseudo maximum likelihood (PPML) estimator, and dummies that represent a country's eligibility for a given preference programme in a single year, the authors have observed that while some trade preferences (e.g., European Union GSP and other preferences offered within that framework) are associated with export product diversi cation, preferences offered to Mediterranean countries have had no signi cant effects on the range of products that they exported (although with the exception of some very earlier versions of these programmes).
Few other studies (e.g., de Melo and Portugal-Pérez, 2008;Gradeva and Martínez-Zarzoso, 2016) are worth mentioning here, as they have examined the effect of NRTPs on manufacturing exports. This topic is closed in spirit to the studies that have examined the export diversi cation effect of NRTPs, as well as to the present study on the effect of NRTPs on economic complexity. de Melo and Portugal-Pérez (2008) have compared exports under the preferential market access offered by the EU (under the EBA initiative and the Cotonou's Economic Partnership Agreement) and the US (under the AGOA) for apparel exports to a set of African countries. While the offer of the preferential market was similar for the two preference granting countries, they differed in their product-speci c rules of origin. The product-speci c rules of origin of the EU imposed a "double transformation", whereby yarn should be woven into fabric and then made-up into apparel in the same country or in a country qualifying for cumulation. In contrast, a 'single transformation' rule is imposed for the production and export of apparel products to the US market under the AGOA preferential regime. Here, trading rms in lesser developed bene ciary African countries could use fabric from any origin and still meet the criteria for preferences. de Melo and Portugal-Pérez (2008) have established that thanks to the more lenient preferential rules of origin under the AGOA preferential regime of the US, the top seven bene ciaries of AGOA's special regime experienced an increase in the number of products exported, which has translated into an export volume by 300%. Gradeva and Martínez-Zarzoso (2016) have found empirically the EU's EBA initiative has led to an increase in ACP's exports of agricultural products and natural resources (which represent the main industries for developing countries -which might bene t the most from aid and trade preferences -and hence the main suppliers of export products). This positive export effect of the EBA initiative was particularly higher for ACP LDCs that received higher development aid ows. However, manufacturing exports were not signi cantly in uenced by the EU's EBA preferential regime. The authors have explained this nding by the erosion of preferences margins received by ACP LDCs due to MFN trade liberalization.
[18] The ndings by Gil-Pareja et al. (2014) line-up with those of Herz and Wagner (2011) concerning the impact of GSP schemes over a decade. In fact, Gil-Pareja et al. (2014) have uncovered that trade preferences have led to an expansion of bene ciaries' exports to preference-granting countries, with the cumulative impact ranging from 26% after 4 years to 88% after 8 years.

Discussion On The Effect Of Nrtps On Economic Complexity
At the outset, in light of the literature review provided in Section 2 concerning the export earning effects of NRTPs, we could argue that if NRTPs affect negatively bene ciary countries' exports, then one should not even expect these trade preferences to improve the economic sophistication of bene ciary countries. Conversely, if NRTPs are, as expected, positively associated with bene ciaries' export performance, one could eventually expect these trade preferences to result in greater economic complexity, in particular if the products exported to preference granting markets are increasingly sophisticated. This rst discussion shows that the precise direction concerning the effect of NRTPs on economic complexity is a priori unknown.
The uncertainty about the theoretical effect of NRTPs on economic complexity also nds some roots on the mixed conclusions reached by the few existing works on the effect of NRTPs on export product diversi cation, which as noted above, represents one major aspect of economic complexity (e.g., Gamberoni, 2007;Persson and Wilhelmsson, 2016;Yannopoulos, 1986).
From a theoretical perspective, the reduction of trade costs in bene ciary countries brought about by preferential access to foreign markets could make exports pro table and lead rms in the bene ciary countries to start exporting products that were not initially traded. In this case, NRTPs could promote export diversi cation (Persson and Wilhelmsson, 2016). However, greater export product diversi cation does not necessarily mean greater economic complexity, as the latter entails both exporting a high number of products, which are not only sophisticated, but also exported by few other countries. Thus, the pro tability of exports induced by preferential access to foreign markets might not necessarily result in greater economic complexity. On the other side, it is possible that trade preferences could adversely affect export product diversi cation, and exert a reduced impact on economic complexity. This can be the case when NRTPs cover products that do not have differences in preference margins, insofar as NRTPs rarely covers all products, including all those of export interest to the bene ciary countries (Persson and Wilhelmsson, 2016).
The trade creation effect of NRTPs in bene ciary countries would depend on these countries' sectors and products of comparative advantage. Hence, countries that are at early stages of industrialization [19] (and whose comparative advantage, therefore, lies on the production of primary commodities and low-skilled/labourintensive manufactures) are likely to experience an export expansion, at best, by exporting goods relatively intensive in the unskilled or low (or semi-skilled) labour inputs, and which are produced using standardized or unsophisticated technology (Yannopoulos, 1986). For these countries, while the utilization of NRTPs can lead to export product diversi cation, including towards light manufactured products, they may not signi cantly help to improve the level of their economic complexity of bene ciary countries.
This short theoretical discussion already shows the di culty of predicting theoretically the direction in which the utilization of NRTPs will affect economic complexity in bene ciary countries. Pending the additional theoretical discussion provided below, we can argue that only the empirical analysis would provide a clearer guidance on the direction of this effect.
Besides, several factors could constrain the ability of bene ciary countries to harness the opportunities provided by the NRTPs so as to produce complex products (Low et al., 2009;Persson, 2015b). Such factors could either limit the positive effect of the utilization of NRTPs on economic complexity, or prevent the utilization of NRTPs from generating the exportation of complex products. These factors include the preference erosion resulting from greater multilateral trade liberalization[20] (in particular since the creation of the WTO) and the end of multi-bre agreement as well as other elements that can be found both on the bene ciary countries' side and on the preference granting countries' side (e.g., Low et al. 2009: page 219). On the side of bene ciary countries, factors include for example, the limited supply response capacity (e.g., Gradeva and Martínez-Zarzoso 2016;Low et al. 2009) and 'domestic' trade policies[21] (e.g., Collier and Venables, 2007). On the preference granting countries' side, factors can include the exclusion of products of potential export interest to the bene ciary countries, the exclusion of countries on a variety of economic and other grounds, stringent (restrictive) rules of origin[22] that require higher-than-existing levels of manufacturing activity in preference-receiving countries, and administrative and other compliance costs that should be borne by a bene ciary country when it claims a given NRTP (e.g., Brenton and Özden 2009;Gitli, 1995;Gradeva and Martínez-Zarzoso 2016;Persson, 2015b;WTO, 2019).
The discussion below takes up many of these factors and discusses the extent to which they could limit bene ciary countries' ability to develop and export complex products.
At the same time, the literature has established that higher FDI in ows can contribute to fostering economic complexity in host countries of multinational enterprises (MNEs) (e.g., Eck and Huber, 2016;Javorcik et al., 2018;Li et al., 2021). As NRTPs can attract higher FDI in ows in bene ciary countries (e.g., Yannopoulos, 1986Yannopoulos, , 1987, one could expect MNEs to help make e cient use of the NRTPs and export complex products under those NRTPs. We will discuss this issue later in the analysis.

Issue of preference erosion and product coverage of NRTPs
Preference erosion has become an issue of serious concern since the inception of the WTO. It has resulted in lower export opportunities for bene ciary countries of NRTPs, in particular for the poorest among them, i.e., LDCs (e.g., Klasen et al., 2021) and can limit the capacity of these countries to diversify their export product baskets towards sophisticated goods. Low et al. (2009) have estimated that MFN tariffs cut on non-agricultural products by the QUAD economies plus Australia would lead to a signi cant erosion of preferences for LDCs as well as for a set of countries [23]. More importantly, the erosion of preferences would concern essentially clothing, especially for the LDCs (except Madagascar), and textiles, sh and sh products, leather and leather products, electrical machinery, and wood and wood products for the top ve affected countries. These would undermine the capacity of these countries to develop infant industries and engage on a sustainable industrialization path. In terms of trade solutions to preference erosion, the authors have then concluded that improving the utilization rates of NRTPs may or may not have a decisive effect (in terms of export) in most of the affected countries.
A high utilization rate of NRTPs, in particular across many sectors and products covered by the provisions of NRTPs indicates that the NRTPs are being effective in generating high export earnings. Bene ciary countries with such high utilization rates of NRTPs could then be exposed to loss of preferential margins (i.e., to trade preference erosion) and as a consequence, to a potential trade diversion, if MFN tariffs on products exported under the NRTP fall (Inama, 2003;Reynolds, 2009). In such countries, rms producing for the international trade market would be facing strong competition from foreign rms (foreign rms exporting under MFN tariffs to preference granting countries). Additionally, the limited trade capacity[24] of these countries may hinder their ability to sustain the competition in the international trade market. This suggests that the erosion of trade preferences could limit the capacity of bene ciary countries to produce sophisticated/complex products (under the NRTPs) if the erosion of trade preferences were not accompanied by the supply of higher development aid ows, including Aid for Trade (AfT) ows (by donor-countries, including preference-granting countries). Such aid in ows could help bene ciary countries strengthen their trade capacity, maintain a good level of competitiveness of export to the markets of preference-granting countries, even in the context of signi cant preferential margins losses (see also Inma, 2003). Gradeva and Martínez-Zarzoso (2016) have shown empirically that while the EU GSP scheme under the 'Everything But Arms' (EBA) Initiative has not provided an additional positive effect on the export performance of the African, Caribbean and Paci c countries (ACP) LDCs, the effect has become signi cantly positive once such preferences are accompanied by higher development aid ows [25]. The authors have then concluded that the EU should support a development strategy that entails the provision of both types of assistance, i.e., development aid and non-reciprocal trade preferences.
Incidentally, if a NRTP authorizes exports of products that are not exported by non-bene ciary countries, then there would be a strong export stimulating trade diversion effect of the NRTP, and this may eventually encourage the bene ciary country to produce and export high-value added (in particular complex) products. In this case, the offer of an NRTP could be associated with greater economic complexity.
Conversely, if a NRTP covers essentially products that are exported by many non-bene ciaries countries (that is, other countries that could not claim the NRTP or those that have lower preference margins), then there could be a limited scope for export stimulating trade diversion effect of the NRTP. This could be the case if the export structure of those non-bene ciary countries is similar to (or overlap with) the export structure of bene ciary countries, and if in addition, the offer of the NRTP has been followed by MFN tariff liberalization (at the multilateral level) of the products authorized under the NRTP. Nevertheless, the trade diversion effect of trade preferences erosion might be less important for bene ciary countries that have many sectors with minimal utilization rates of NRTPs, or for those that have enjoyed higher development aid ows to compensate the loss of preferential margins (see also Inama, 2003). In such countries, the utilization rates of NRTPs can result in a greater economic complexity if they end-up exporting a high number of increasingly complex products under NRTPs.
Notwithstanding this, low utilization rates of NRTPs for a bene ciary country means that its exports mostly takes place under MFN tariffs. This country might, therefore, not lose from the reduction of the preference margins induced by the MFN tariffs liberalization (e.g., Inama, 2003). It may eventually draw bene ts from the MFN tariffs liberalization if its current export product structure is comprised of goods facing MFN tariff peaks that are excluded from preferential coverage (e.g., Inama, 2003). Hence, such countries would not likely experience a diversi cation of the products exported (let alone the export of sophisticated products) under the NRTPs, as their trade takes place essentially under MFN tariffs. This signi es that low utilization rates of NRTPs might not be associated with higher economic complexity. An improvement in economic complexity might eventually be achieved for products exported under MFN tariffs or for products that face MFN tariff peaks and are excluded from preferential coverage.

Issue of supply-side capacity
On the supply-side capacity front, Collier and Venables (2007) have stressed the need for bene ciary countries of NRTPs to develop skills and infrastructure that are near the threshold of global manufacturing competitiveness if they were to enhance their manufacturing exports under the trade preferences. Iimi (2007) has pointed out the key role of public infrastructure (e.g., paved roads) in reducing production costs and thus fostering external competitiveness and market shares for several African countries that enjoy preferential treatment for exporting to the EU beef market.
As also noted above, Gradeva and Martínez-Zarzoso (2016) has underlined the need for donors, including the EU to support a development strategy based on the provision of both development assistance and NRTPs to bene ciary countries. Prowse (2010) has underlined the importance of development aid, including AfT ows for supporting reform of NRTPs programs. According to the author, AfT ows could serve, inter alia, to identify the supply side and policy constraints in bene ciary countries, while also addressing them. They could also be used to support a process of graduation and adjustment to preference erosion. More generally, there is a wealth of studies that have underlined the role Aid for Trade (AfT) ows in enhancing the trade capacity and encouraging exports by recipient-countries (see for example a literature review in Cadot et al. 2014a; OECD-WTO, 2017; Lammersen and Roberts, 2015;Velde te et al. 2013). In addition, recent studies (Gnangnon, 2019a, Kim, 2019 have established that higher AfT ows could promote export product diversi cation in recipient-countries. Gnangnon (2021b) has obtained evidence that higher AfT ows are associated with greater economic complexity, with less advanced economies (such as poorest countries) enjoying a higher positive effect than relatively advanced countries among AfT recipients. It is also worth noting that NonAfT ows (i.e., other ows of total development aid than AfT ows) can also in uence the ability of bene ciary countries to export sophisticated products under NRTPs, inter alia, through their eventual positive effect on human capital (e.g., Dreher et al. 2008;Kotsadam et al. 2018) and their positive effect on the institutional and governance quality (e.g., Dijkstra, 2018;Dzhumashev and Hailemariam, 2021;Jones and Tarp, 2016). These positive effects might be mitigated or outweighed by the eventual negative effect of NonAfT ows on the real exchange rate (e.g., Gnangnon, 2021c) in the bene ciary countries.
All these suggest that by fostering the trade capacity of bene ciary countries of NRTPs, higher development aid ows may contribute either to enhancing the positive effect of the utilization of these trade preferences (if there is any positive effect at all) on economic complexity, or eventually help to turn a negative effect of the usage of these preferences on economic complexity into a positive one. We will test later this hypothesis (Hypothesis 1).

Issue of preferential rules of origin
As noted above, preferential rules of origin may represent a major constraint to the usage of NRTPs to export complex products. Preferential rules of origin aim not only to prevent trade de ection [26], but also to provide incentives to rms to add signi cant value to their products or to source more of their intermediate inputs domestically, with a view to broadening and diversifying the manufacturing base of bene ciary countries.
Restrictive preferential rules of origin can constrain the bene ts of tariff preference margins (e.g., Cadot et al., 2014b) and more generally, deter[27] exports, and undermines the utilization of NRTPs (Hakobyan, 2015;Inama, 2003;Sytsma, 2021;WTO, 2019). Building on the case of Vietnam, Doan and Xing (2018) have demonstrated that lenient rules of origin improve countries' export e ciency, de ned as the ratio of actual exports to the maximum possible volume. Thus, stringent preferential rules of origin could alter signi cantly the ability of rms in bene ciary countries to produce and export high value-added products. This could be the case if preferential rules of origin were designed by preference granting countries to protect their domestic industries from import competition and restrict the possibility for bene ciary countries to source inputs from third countries (e.g., Brenton and Özden 2009;Cadot and de Melo, 2008;UNCTAD, 2004). According to Collier and Venables (2007), trade preferences can have a catalytic role in stimulating African bene ciaries' manufacturing supply response if, among others, they allow for the importation of complementary inputs. Restrictive (preferential) rules of origin could bene t to larger rms at the expense of smaller rms (e.g., Hayakawa[28] et al. 2009;Ulloa and Wagner-Brizzi[29], 2013) in particular if they are captured by special interests (Portugal-Perez, 2011).
A note by the WTO (WTO, 2019) has explained why LDCs experienced low rates of preference utilization for several agricultural exports under many preferential schemes, while agricultural products are usually very simple products, and often subject to simple rules of origin. The note has pointed out that a variety of factors matter for the utilization rates of trade preferences. These include the choice and design of origin criteria; the di culties in complying with other origin requirements (e.g., proof of origin; direct consignment); and the deliberate choice by the trading rms to refrain from claiming duty-free treatment under an NRTP, notably if there are other available NRTPs, the preferential tariff margin under the NRTP concerned is low, or if there is an insu cient knowledge or lack of knowledge about the existence of trade preferences.
Besides, stringent preferential rules of origin can discourage e ciency-seeking FDI in ows (UNCTAD, 2002), and hamper the possibility of producing complex export products (see the discussion below on the effect of FDI in ows on economic complexity).
Against this backdrop, we could expect stringent rules of origin to result in a low utilization rate of NRTPs, and consequently to impede the ability of bene ciary countries of producing and exporting complex products.
Summing-up the discussion under Section 3, it would be di cult to anticipate the direction of the effect of the utilization of NRTPs on economic complexity. Only the empirical analysis would provide a clearer guidance on the direction of this effect.
[19] Nowadays, this is the case of many developing countries, and few least developed countries (such as Bangladesh and Lesotho) whose exports are still dependent on export of light manufacturing products, such as textiles and apparel (the export product structure of LDCs is highly dependent on primary commodities (WTO, 2020). It is important to recall here that LDCs are in the world, the poorest and most vulnerable countries (to external and environmental shocks. Further information on this group of countries could be found online at: http://unohrlls.org/about-ldcs/criteria-for-ldcs/ [20] Since the inception of the WTO, the world has undergone a signi cant liberalization of most-favoured-nation (MFN) tariff rates, which has led to the reduction of the preference margins granted under NRTPs.
[21] Inconsistent trade policies could adversely affect rms' production and exports, and hence their ability to take advantage of the NRTPs, let alone to export sophisticated. For example, Collier and Venables (2007) have recommended that one policy, as part of the catalytic action for exporting a manufacturing 'task' in African countries bene ciaries of NRTPs, is to import without restrictions, all the complementary upstream tasks.
[22] Rules of origin specify the minimum level of local transformation (i.e., the amount of local content or processing) required to make manufactured products eligible for preferential tariff treatment (e.g., Cadot et al., 2006;Sytsma, 2021).
[23] The 10 largest losers from the preferences erosion among NonLDC developing countries were the Dominican Republic, Honduras, Kenya, Mauritius, St. Lucia, El Salvador, Guatemala, Namibia, Nicaragua, and Swaziland.
[24] Many bene ciaries countries request NRTPs because they do not usually have the requisite level of trade capacity to sustain international competition under MFN tariffs. Hence, NRTPs allow them to progressively develop their infant industries so as to become competitive in the international trade market when NRTPs will be withdrawn (e.g., Hoekman and Özden, 2005).
[25] It is worth noting that the positive effect of the EU's GSP scheme on export performance of ACP LDCs (found by Gradeva and Martínez-Zarzoso, 2016 -see the results on page 25) concerned essentially the agricultural export performance, but not manufacturing export performance, as for the latter, there was no signi cant joint effect of the EU's GSP scheme and development aid.
[26] Trade de ection occurs when to avoid paying duties, non-bene ciary countries of a NRTP that wish to export their products to the preference granting country of the NRTP would re-direct these products through the bene ciary country of the NRTP.
[28] According to Hayakawa et al. (2009), rules of origin are not binding on larger a liates and rms that have the most widely spread intermediate sourcing.
[29] Ulloa and Wagner-Brizzi (2013) have shown theoretically that small rms can be made worse off by the mere existence of trade preferences because large rms expand to take advantage of them and bid up the price of domestic factors of production.

Do Fdi In ows Matter For The Effect Of Nrtps On Economic Complexity?
As stated above, FDI in ows may help bene ciary countries harness the opportunities offered by NRTPs, and export complex products under these NRTPs. This section seeks to address theoretically this question by discussing how FDI in ows can matter for the effect of the utilization of NRTPs on economic complexity.
On the one hand, the literature on the macroeconomic determinants of economic complexity has established that MNEs could be instrumental in enhancing economic complexity in the host countries through setting-up plants in these countries (i.e., through FDI in ows). MNEs can help foster economic complexity in the host countries by helping improve the innovation performance of host countries through their cost discovery activities, their supply of better inputs to local producers (e.g., Eck and Huber, 2016;Javorcik et al., 2018) and the transfer of knowledge to domestic rms (e.g., Javorcik, 2004;Havranek and Irsova, 2011). On the empirical front, Eck and Huber (2016) have obtained that spillovers from multinationals to local Indian rms have encouraged the manufacturing of sophisticated products in India. Javorcik et al. (2018) have shown that foreign a liates in Turkey have allowed Turkey rms to introduce increasingly complex products. Likewise, Xu andLu (2009) andHausmann (2016) have shown that FDI in ows promote the manufacture of sophisticated products in the host countries. Li et al. (2021) have uncovered that the liberalization of FDI in 2002 on different industries in China, has had a positive effect on export sophistication. Meanwhile, Antonietti and Franco (2021) have demonstrated empirically that FDI in ows (Granger) cause economic complexity, but that economic complexity does not (Granger) cause FDI in ows in developing countries.
On the other hand, the study by Yannopoulos (1986), which is one of the existing rare studies on the effect of NRTPs on FDI in ows, has noted that the offer of NRTPs to a bene ciary country could make it particularly attractive to a foreign investor (notably a MNE) if this country had lower labour costs and other related locational advantages. If a MNE (be it from the preference granting country or from third countries that have export interests in the bene ciary country) has the ownership speci c advantages to compete effectively with local rms in the host country, then the offer of a NRTP would lay the groundwork for international production within the bene ciary country (this is the "tariff factory effect of NRTPs", Johnson, 1967). More generally, the type of FDI (branch plant development, joint ventures, subcontracting arrangements) in which the MNE would engage in the bene ciary country of a NRTP would depend on a variety of factors, including the preference margin associated with the NRTP, the production capabilities of local rms (i.e., in the bene ciary country), the availability of complementary inputs in the bene ciary country, the technology of production (i.e., the extent to which unskilled labour intensive processes can be packaged and relocated elsewhere), and the market structure within which a given MNE operates (see Yannopoulos, 1986: p19). Yannopoulos (1987: p98) has argued that FDI ows (either from the preference granting country or from third countries) to a bene ciary country would be larger and contribute signi cantly to the bene ciary country's export expansion under the NRTP, the greater is the marketing intensity of production in the export sectors whose pro tability has increased thanks to the NRTP. This is because foreign rms may have a competitive advantage over local rms in producing exportable goods under the NRTP. This includes even standardized goods such a labour-skilled and fair technology intensive goods such as clothing may require superior knowledge and marketing skills. MNEs that are established in the bene ciary country to take advantage of the NRTP would also contribute to the expansion of this country if the additional export production driven by the NRTP required the utilization of specialized informational assets that are transferable through intra-rm mechanisms. Additionally, MNEs (including large ones) often experience lower costs of scanning international markets for inputs than local rms in the bene ciary country. Therefore, the greater is the share of inputs to be purchased from outside the bene ciary to produce exportable goods under the NRTP, the higher is the contribution of MNEs to the bene ciary country's export expansion under the NRTP.
Building on these arguments, Gnangnon and Iyer (2021) have obtained empirically that FDI in ows exert a positive effect on the utilization rate of both GSP programs and other NRTPs offered by the QUAD countries, although the positive effect of FDI in ows on GSP programs is higher than their positive effect on other NRTPs.
Against this backdrop, we postulate that in light of the competitive advantages of MNEs over local (indigenous) rms in the bene ciary countries of NRTPs to expand exports under the NRTPs, MNEs (through their FDI in the bene ciary countries) could help enhance the utilization of those NRTPs, and contribute to fostering economic complexity through its effect on innovation, the supply of better inputs to local producers (e.g., Eck and Huber, 2016; Javorcik et al., 2018) and the transfer of knowledge to domestic rms (e.g., Javorcik, 2004;Havranek and Irsova, 2011). We also expect the positive effect of the utilization rate of NRTPs on economic complexity to be greater (in magnitude) as FDI in ows become larger (Hypothesis 2).

Estimation Strategy
This section presents the baseline model speci cation used in the empirical analysis (sub-section 5.1); it provides an analysis of data concerning our main variables of interest (sub-section 5.2) and discusses the econometric approach employed in the empirical exercise (sub-section 5.3).

Benchmark speci cation
To investigate the effect of the utilization of NRTPs on economic complexity, we build on previous works on the macroeconomic determinants of economic complexity[30], in particular Chu (2020) (2021b); Saadi (2020) and Sweet and Maggio (2015). According to the in uential paper by Hausman et al. (2007), a country's relative costs and the patterns of specialization are determined by a number of this country's fundamentals, including its endowments of physical and human capital, labor, natural resources and the overall quality of its institutions.
In the present analysis, the geographic potential of a country and the size of the labour force are measured here by the population density (e.g., Lapatinas, 2009). The quality of the labor force is represented by the human capital (e.g., Lapatinas, 2009). Hausman et al (2007) and other studies such as Lapatinas and Litina (2019) and Sweet and Maggio (2015) have reported a positive and signi cant effect of the real per capita income (which represents a proxy for countries' development level) on economic sophistication. Hausman et al (2007)  The literature has also established that nancial development can be positively associated with economic complexity Su, 2021a, 2021b). A good institutional quality is conducive to greater economic complexity through its effect on human capital and incentives for innovation (e.g., Lapatinas and Litina, 2019;Trung, 2021). Finally, terms of trade improvements would lead to greater economic complexity if the revenues extracted were invested in innovative activities that would help to manufacture higher value-added products (e.g., Gnangnon, 2021b). As also discussed above, FDI in ows can have a signi cant in uence on economic complexity, including in developing countries.
Against this background, we postulate a benchmark model speci cation that includes, in addition to our indicators of utilization rate of NRTPs, a set of control variables derived from those previous works, and that matter for the effect of the utilization rate of NRTPs on economic complexity. Control variables include the real per capita income, FDI in ows, trade policy, human capital, nancial development, institutional quality, dependence on natural resource rents, the population density, and terms of trade.
The benchmark (baseline) model takes the following form: A country and time-period are respectively represented by the subscripts i and t. The panel dataset utilized to estimate model (1) (and its different variants described below) is unbalanced, and contains 110 bene ciary countries of NRTPs over the period 2002-2018. It has been constructed based on data available. To avoid modelling business cycles, we have used averages of variables over non-overlapping sub-periods of 3-year. These sub-periods are: 2000-2002; 2003-2005; 2006-2008; 2009-2011; 2012-2014 and 2015-2018. The parameters to are to be estimated. are time invariant speci c effects of a country, and re ect global shocks (captured by time dummies) that affect the economic complexity paths of all countries taken together. is an error-term.
The dependent variable "ECI" is the index of economic complexity. It re ects the diversity and sophistication of a country's export structure, and hence indicates the diversity and ubiquity of that country's export structure. It has been estimated using data connecting countries to the products they export, and applying the methodology described in Hausman et al. (2009). Higher values of this index re ect greater economic complexity.
Following previous works (e.g., Chu, 2020; Gala et al. 2018;Gnangnon, 2021b;Laverde-Rojas, 2019;Nguyen and Su, 2021b;Saadi, 2020;Sweet and Maggio, 2015), we have introduced the one-period lag of the economic complexity variable as a regressor in model (1). This helps not only to capture the state dependence path of economic complexity (i.e., its persistence over time), but also to control for the omission of some variables. In particular, the lag of the dependent variable can help to control for the absence from the benchmark model of indicators of the utilization rates of NRTPs offered by other preference granting countries than the QUAD countries.
The variable "URGSP" represents the utilization rate of GSP programs provided by the QUAD countries to developing countries. It measures the extent to which imports that are eligible for GSP programs are actually imported under these preferences. It has been computed using a formula adopted by both the WTO and the UNCTAD (e.g., WTO, 2016). The formula goes as follows: URGSP = 100*(GSP received imports)/(GSP covered imports), where "GSP received imports" refers to the value of imports that received GSP treatment, and "GSP covered imports" represents the value of imports classi ed in tariff lines that are dutiable and covered by the GSP scheme of the preference-granting country. Values of the computed indicator "URGSP" range initially from 0 to 100, with higher values indicating a greater utilization rate of GSP programs. However, for ease of interpretation of empirical results, this indicator has been re-scaled (i.e., divided by 100) so that its values range now between 0 and 1.
The indicator "UROTP" is the utilization rate of the other NRTPs (henceforth, referred to as "other trade preferences") offered by the QUAD countries to developing countries. For the US, the set of other trade preferences covers the African Growth and Opportunity Act (AGOA) and the Caribbean Basin Initiative. In the case of the EU, it encompasses preferences under the Economic Partnership Agreements entered with selected Africa Sub-Saharan countries. It has been computed as follows: UROTP = 100*(Other-preferential imports)/(Other Preferential covered imports). "Other-preferential imports" refers to the value of imports that bene tted from NRTPs other than GSP programs. "Other-preferential covered imports" refers to the value of imports classi ed in tariff lines that are dutiable and covered by the other-preferential schemes. The values of the computed indicator "UROTP" range initially from 0 to 100, with higher values indicating a greater utilization rate of GSP programs. However, to facilitate interpretation of empirical results, the variable has been re-scaled (i.e., divided by 100) so that its values range now between 0 and 1.
The two indicators of the utilization of NRTPs are simultaneously introduced in model (1) because of the potential overlap (in terms of product coverage) between the different NRTPs that a country enjoys (e.g., Hakobyan, 2015;Keck and Lendle, 2012). Hakobyan (2015) has argued, and provided empirical support to the fact that the utilization rate of the US GSP scheme falls with the availability of other trade preference programs to bene ciary countries. Gnangnon and Iyer (2021) have also provided strong empirical support to this hypothesis.
The variable "GDPC" is the real per capita income, used as a proxy for the economic development level. "POPD" is the population density. Both "GDPC" and "POPD" have been logged (using the natural logarithm) in order to reduce their skewed distribution. "FDI", "TP", "HUM" and "FINDEV" are respectively the share of FDI in ows in GDP, trade policy, human capital, and nancial development. The variables "INST", "RENT", and "TERMS" are respectively the indicators of institutional and governance quality, the share of natural rents in GDP (as a proxy for dependence on natural resources), and terms of trade. Note that values of the variables "FDI", "TP", "HUM", INST", "RENT" and TERMS" originally range between 0 and 100. As we did for "URGSP" and "UROTP", these variables have been re-scaled (i.e., divided by 100) for ease of results' interpretation. All variables introduced in model (1) have been described in Appendix 1, and their descriptive statistics (over the full sample) are presented in Appendix 2a. The list of the 110 countries of the full sample is provided in Appendix 3a.

Data analysis
This section presents a brief graphical analysis of our variables of interest, namely the utilization of NRTPs and the economic complexity. To get rst sense of the relationship between the utilization rates of NRTPs and economic complexity, we use the dataset of 110 bene ciary countries of NRTPs over nonoverlapping sub-periods of 3 years, to show in Figure 1 (over the full sample) how the variables "URGSP" and "UROTP" on the one hand, and "ECI" on the other hand, have evolved over time. Using the same dataset, Figure 2 presents over the same full sample, the cross plot between "URGSP" and "ECI" on the one hand, and between "UROTP" and "ECI", on the one hand.  Figure 1 shows that, on average, over the full sample, the utilization rates of NRTPs tend to move in the same direction with the level of economic complexity.
[Insert Figure 2, here] However, Figure 2 presents different patterns of these variables. It reveals that the utilization rate of GSP programs and the utilization rate of other trade preferences are all negatively correlated with economic complexity in bene ciary countries of these preferences. The empirical analysis, which would help uncover a causal effect of the utilization rates of NRTPs on economic complexity, might con rm or invalidate what we observe in Figure 2.

Econometric approach
We start the empirical analysis by estimating model (1) using the Pooled Ordinary Least Squares, the Within Fixed Effects and Random Effects estimators. The results of these regressions are presented in Table 1. However, these outcomes are likely biased for several reasons. First, the lagged dependent variable in model (1) is correlated with the countries' xed effects in the error term, and lead to a biased coe cient of this variable, in particular for dynamic panel datasets with a small time-period and a large cross-section dimensions (this bias is referred to as the 'Nickell bias', see Nickell, 1981). For example, if the lagged dependent variable is correlated with the xed effects in the error term, Pooled Ordinary Least Squares estimator and the within xed effects yield biased estimates of the lagged dependent variable, that move in opposite directions. The Pooled Ordinary Least Squares estimator leads to an upward bias of the estimate associated with the dependent variable by attributing to it the predictive power that belongs to the countries' xed effects. At the same time, the within xed effects estimator generates a downward bias of the coe cient of the lagged dependent variable due to due to the leading negative correlations between the within-transformed lagged dependent variable term and the withintransformed error term (Bond et al., 2001). In addition, the potential endogeneity of many regressors in model (1) due to the reverse causality problem (i.e., the feedback effect from the dependent variable to the regressor concerned) can lead to biased coe cients of these variables in the regressions that use the three estimators mentioned above. The reverse causality problem concerns the variables "URGSP", "UROTP", "GDPC", "FDI", "TP", "HUM", "FINDEV", "INST" and "RENT". The rationale for the reverse causality issue for the indicators of the utilization rates of preferences is as follows. We expected theoretically that the utilization rates of NRTPs would affect bene ciary countries' level of economic complexity. However, it could also be envisaged that as preference-granting countries de ne the eligibility criteria for NRTPs, they may select countries with low levels of economic complexity to enjoy the trade preference. This is consistent with the spirit of the provision of NRTPs to developing countries (see the Resolution 21(II) of the second conference of the UNCTAD). Therefore, the level of economic complexity may determine the eligibility to NRTPs, and consequently the utilization of these preferences.
As noted above, economic complexity signi cantly affects the level of economic complexity, which we have proxied by the real per capita income; hence, the reverse causality from the economic complexity variable to the real per capita income variable. Countries that improve their level of economic complexity can attract FDI in ows (e.g., Gómez-Zaldívar et al. 2021) because MNEs may choose to set up plants in countries that manufacture complex products, given the speci c advantages of these countries in terms of productive knowledge and exclusive capabilities. Gómez-Zaldívar et al. (2021) and Sadeghi et al. (2020) have found evidence that countries with greater economic complexity attract higher FDI ows, while Antonietti and Franco (2021) have uncovered that economic complexity does not (Granger) cause FDI ows to developing countries. In spite of the unconclusive outcomes of the limited existing literature on the effect of economic complexity on FDI in ows, we do consider the variable "FDI" as endogenous in the present analysis. Incidentally, countries with a low level of economic complexity may wish to liberalize their trade policies, especially by reducing trade barriers, including tariffs on intermediate inputs. As a result, the trade policy variable could be considered as endogenous. As also noted above, economic complexity affects human development[31] (e.g., Le Caous and Huarng, 2020), including health outcomes (e.g., . These make the human capital variable an endogenous variable de facto. In light of the potential positive effect of economic diversi cation, and in particular of export product diversi cation on nancial development (e.g., Gnangnon, 2019b;Hattendorff, 2014;Ramcharan, 2006), we expect a feedback effect from economic complexity to nancial development (insofar as export diversi cation is a major aspect of economic complexity). This renders the nancial development variable potentially endogenous.
As the institutional quality is a fundamental determinant of economic sophistication (Hausman et al., 2007), one could expect countries with lower levels of economic complexity to endeavour to improve their institutional quality so as to develop and export complex products. Finally, Nguyen et al. (2020b) have found that economic complexity has reduced impact on natural resource rents, although the effect varies across groups of countries and across the type of natural rents. This indicates the possibility of the reverse causality from the dependent variable to the natural resource rents variable, which makes this variable endogenous in the regressions. The variables representing the population density and the terms of trade have been treated as exogenous in the analysis.
Despite the risk of obtaining biased estimates when using the Pooled Ordinary Least Squares, the Within Fixed Effects and Random Effects estimators, we have presented these outcomes with a view to comparing them with those of an alternative estimator, i.e., the two-step system Generalized Methods of Moments (GMM), which is suitable for panel data with limited time horizon (small T) and a large cross-section dimension (large N).
As a matter of fact, and in light of the di culties of nding valid external instruments to address the endogeneity issues raised above, we opt for using the two-step system GMM estimator[32] developed by Arellano and Bover (1995) and Blundell and Bond (1998). This estimator helps to correct for the unobserved country heterogeneity, measurement errors, potential endogeneity concerns highlighted above, and the omitted variable bias (here for example, the absence in the baseline speci cation of the utilization rates of NRTPs provided by other preference-granting countries than the QUAD countries). While some authors have utilized the difference GMM estimator[33] proposed by Arellano and Bond (1991) to address the endogeneity issues raised above, this estimator suffers from sample bias (and generates weak instruments) when the time dimension of the panel is small, and time series (including the dependent variable) show a high degree of persistence (e.g., Alonso-Borrego and Arellano, 1999;Bond et al., 2001;Bond, 2002). The two-step system GMM estimator helps to reduce the imprecision and potential bias arising from the use of the difference GMM estimator by allowing the use of lagged levels and lagged differences of variables as instruments. In so doing, it improves the consistency and e ciency of the estimates. It estimates a system of equations that combines an equation in differences and an equation in levels. Lagged rst differences are used as instruments for the levels equation, and lagged levels are used as instruments for the rst-difference equation.
As the pooled ordinary least squares estimator yields upward biased coe cient of the lagged dependent variable, and the within xed effects estimator yields a downward bias of the coe cient of this variable, then the coe cient of this variable obtained by the two-step system GMM estimator should lie between the estimate generated by the within xed effects estimator and the one generated by the pooled ordinary least squares estimator (e.g., Bond et al., 2001).
To evaluate the correctness of our different model speci cations described below (that build on the baseline model (1)), we use three criteria: the Arellano-Bond test of rst-order serial correlation in the rst-differenced error term (denoted AR(1)), the Arellano-Bond test of the absence of second-order autocorrelation in the rstdifferenced error term (denoted AR(2)), and the Sargan/Hansen test of over-identifying restrictions (OID), which helps to test the joint validity of the instruments used in the regressions. For our model to be correctly speci ed, we should reject the null hypothesis of absence of rst-order serial correlation in the rst-differenced error term (for the AR(1) test[34]), but we should fail to reject the null hypothesis of the AR(2) test[35] (i.e., absence of second-order correlation in the rst-differenced error term), and the null hypothesis of the OID test [36], which is the validity of the internal instruments used in the regressions.
However, the e ciency gains arising from using the two-step system GMM estimator (at the expense of the difference GMM estimator) comes with a cost, which is the exponential increase in the number of instruments as the number of periods rises. This proliferation of instruments can raise other problems, such as a matrix with high estimated variances, an over-adjustment of endogenous variables, and a weakening of the power of the various tests that help to assess the validity of the two-step system GMM technique. To overcome the problem of instruments proliferation, we follow the rule whereby the number of instruments should not exceed the number of countries (e.g., Roodman, 2009), and limit the number of lags used to generate internal instruments in our regressions. Especially, we use a maximum of two lags of the dependent variable as instruments, and of endogenous variables as instruments.
Several speci cations of model (1) have been performed using the two-step system GMM. Table 2 reports the outcomes from the estimation of the baseline model (1). Table report estimates that allow assessing how the effect of the utilization rates of NRTPs varies across countries in the full sample. These estimates have been obtained by estimating a speci cation of model (1) that contains the interaction between the utilization rates of GSP programs and the real per capita income variable, and the interaction between the utilization rates of other trade preferences programs with the real per capita income variable. Both interaction variables are included simultaneously in the model speci cation.

Column [2] of the same
We estimate a third speci cation of model (1) in which we introduce the interaction between the two indicators of the utilization rates of NRTPs. The objective pursued in estimating this speci cation of model (1) is to investigate the extent to which the simultaneous use of both GSP programs and other trade preferences by bene ciary countries affects their level of economic complexity. The outcomes of this estimation are presented in column [3] of Table 2.
Next, we wish to test hypothesis 1 whereby development aid ows, in particular AfT ows can contribute to enhancing bene ciary countries' trade capacity, including their ability to manufacture and export high valueadded goods under the NRTPs regimes. To recall, we expected, in light of the ndings by Gradeva and Martínez-Zarzoso (2016), that the utilization rates of NRTPs would enhance economic complexity in countries that receive higher development aid in ows. To test this hypothesis, we focus on development aid recipient countries among bene ciary countries of NRTPs in our full sample (see the list of these countries in Appendix 3b), and use several development aid variables, i.e., total development aid ows, and some of its major components. Each of these aid variables are used to rst estimate several other variants of model (1), i.e., model (1) in which we introduce each aid variable, with the view to checking whether they alter the coe cients of the two variables measuring the utilization rates of NRTPs. The outcomes of the estimation of these different speci cations of model (1) are provided in Table 3. Second, we estimate different other speci cations of model (1), i.e., model (1) that incorporates a development aid variable along with its interaction with each of the indicators of utilization rate of NRTPs. The results of these different estimations are shown in Table 4. Development aid considered here (see Appendix 1 for more details on the aid variables used) are: the total the real gross disbursements of total ODA expressed in constant prices 2018, $US ("ODA"), or alternatively each of its two components, namely the total real gross disbursements of AfT ("AfTTOT") and the real gross disbursements of other development aid ows ("NonAfTTOT"), which represents the difference between total development aid and AfT ows. Given the importance of AfT for the integration of developing countries into the global trading system, we additionally look at which components (among the three main categories of total AfT ows) in uence the most the effect of the utilization of NRTPs on economic complexity. The components of total AfT ows considered are the real gross disbursements of AfT dedicated to the build-up of economic infrastructure ("AfTINFRA"), the real gross disbursements of AfT allocated to the strengthening of productive capacities ("AfTPROD"), and the real gross disbursements of AfT allocated for trade policy and regulation ("AfTPOL"). Each of these components is introduced once in the speci cation of model (1). Appendix 2b displays the list of the aid-recipient countries, and Appendix 2b contains descriptive statistics on all variables for aid recipient countries. Table 3 contains regressions' outcomes that help test hypothesis 2, i.e., to assess whether and if so, the extent to which FDI in ows in uence the effect of the utilization rates of NRTPs on economic complexity. These outcomes are obtained by estimating a second speci cation of model (1) that includes the interaction variables between each indicator of the utilization rate of NRTPs and the variable capturing FDI in ows. Laverde-Rojas (2019); Nguyen and Su (2021b); Saadi (2020) and Sweet and Maggio (2015).

Finally, column [2] of
[32] The difference GMM estimator wipes out countries' xed effects and uses lags of variables as instruments of endogenous variables.
[33] The p-value of the AR(1) test should be lower than 0.10 at the 10% level of statistical signi cance.
[34] The p-value related to the AR(2) test should be higher than 0.10 at the 10% level of statistical signi cance.
[35] The p-value associated with the Sargan test of over-identifying restrictions should be higher than 0.10 at the 10% level of statistical signi cance.
[36] There is a growing work on the macro-determinants of economic complexity. Existing studies include for example Chu (2020); Gala et al. (2018); Gnangnon (2021b);Hausmann et al. (2007) Sweet and Maggio (2015) and Trung (2021). See for example, Nguyen et al. (2020a) for a literature review on current studies on economic complexity.

Empirical Results
This section analyses the empirical outcomes reported in Tables 1 to 5. Unless otherwise speci ed, we interpret the estimations' outcomes at the 5% level, i.e., estimated coe cients are considered as statistically signi cant only at the 5% level.
Across all ve Tables, the coe cient of "ECI" is positive and signi cant at the 1% level, and this con rms the nding of previous studies that economic complexity exhibits a state dependence path. Interestingly, and as expected, the coe cients of the lagged dependent variable arising from the use of the two-step system GMM approach are all comprised between the estimate of this variable obtained from the pooled ordinary least squares estimator and the coe cient of the same variable based on the within xed estimator (see Tables 1 to   5).
[Insert Table 1, here] We obtain from column [1] of Table 1 that at least at the 5% level, the utilization rate of both GSP programs and other trade preferences are negatively and signi cantly associated with economic complexity. Similar results (including in terms of magnitude of the effects) are obtained from the regression based on the random effects estimator (see column [3] of Table 1). This nding aligns with the negative correlation pattern observed in Figure  2 between the indicators of the utilization rate of NRTPs and economic complexity. In terms of the magnitude of the effects (building on results in column [1]), we obtain that 1-point increase in the utilization rate of GSP programs is associated with a 0.086-point increase in the index of economic complexity, and a 1-point increase in the utilization rate of other trade preferences is associated with a 0.08-point rise in the index of economic complexity. Results in column [2] of Turning now to Tables 2 to 5, we observe that all model speci cations whose results [37] are reported in these Tables meet the criteria provided above to ensure the validity of the two-step system GMM estimator (see the bottom of all Tables).
[Insert Table 2, here] Estimates presented in Table 2 show that the utilization rates of GSP programs and of other trade preferences are robustly and negatively associated with economic complexity in bene ciary countries (the coe cients of "URGSP" and "UROTP" are negative and signi cant at the 1% level -and have similar magnitudes). We conclude that the utilization rates of these two groups of NRTPs have a reduced impact on economic complexity. In terms of magnitude, we obtain that a 1-point increase in the utilization rate of GSP programs is associated with a 0.233-point fall in the index of economic complexity. Similarly, a 1-point increase in the utilization rate of other trade preferences generates a 0.22-point fall in the index of economic complexity. The magnitudes of these effects are far higher than the ones obtained in Table 1, notably from results based on the pooled ordinary least squares estimator (the coe cient was -0.086) and results obtained by using the random effects estimator (the coe cient was -0.077).
These negative effects of the utilization of GSP programs and other trade preferences (taken separately) on economic complexity can be interpreted in various ways. They may re ect differentiated effects across countries in the full sample (we will check this later by analysing the results in column [2] of Table 2). These results may also suggest that taken separately, the utilization of GSP programs and of other trade preferences might not be su cient to enhance economic complexity in bene ciary countries. This outcome may also re ect (as hypothesized in section 2) that the effect of the utilization of NRTPs on economic complexity in bene ciary countries may be dependent on FDI in ows. Yet data is not available on MNEs engaged in exporting activities in the bene ciary countries, with a view to taking advantage of the NRTPs. But the use of the indicator of the net (total) FDI in ows as a share of GDP could provide an insight into the extent to which FDI in ows matter for the effect of the usage of NRTPs on economic complexity (see results in Table 5). Finally, these negative effects of the utilization of NRTPs on economic complexity may re ect an insu cient trade capacity of bene ciary countries, a negative effect of the erosion of preference margins, or the absence of lenient preferential rules of origin. As noted in the previous section, results reported in Table 5 would help examine how development aid (including both total development aid, as well as its major components) in uences the effect of the utilization rates of NRTPs on economic complexity, inter alia, through its trade capacity enhancing effect, human development institutional quality and eventually real exchange rate effects.
Regarding control variables in column [1], we nd that the rise in the real per capita income, higher FDI in ows, greater trade policy liberalization, a higher accumulation of human capital, an improvement in the level of nancial development (i.e., an increase in the volume of credit allocated to the private sector), and an increase in the population density are robustly associated with greater economic complexity. On the other side, terms of trade improvements exert no signi cant effect on economic complexity, but an increase in the dependence on natural resources hinders economic sophistication. Finally, we obtain a negative and signi cant effect of the institutional and governance quality on economic complexity. This surprising outcome contrasts with the ndings of the literature (e.g., Lapatinas and Litina, 2019;Trung, 2021). We suspected that this outcome may re ect an interplay between the utilization rates of NRTPs and the institutional and governance quality in affecting economic complexity. To check this, we introduced in model (1) both the interaction between the variables "URGSP" and "FDI", and the interaction between "UROTP" and "FDI". The resulting model speci cation is estimated using the two-step system GMM approach. The outcomes[38] of this estimation indicate that the utilization of GSP programs in uences positively economic complexity as the quality of institutions and governance improves. However, the utilization of other trade preferences exerts a higher negative effect on economic complexity as the quality of institutions and governance improves. These suggest that when their quality of institutions and governance improves, bene ciary countries tend to rely on GSP programs (rather than other trade preferences) to enhance economic complexity.
The outcomes of control variables in columns [2] and [3] of Table 2 are quite similar to those in column [1] of the same Table. As indicated in the previous section, the outcomes in column [2] of this Table help to examine how the utilization rate of NRTPs affects economic complexity for varying levels of real per capita income. First, we nd that the coe cient of the variable measuring the utilization rate of GSP programs is negative and signi cant at the 1% level, but the interaction term of the variable ["URGSP*[Log(GDPC)]") is positive and signi cant at the 1% level. These two outcomes suggest that there may be a level of the real per capita income beyond which the effect of the utilization rate of GSP programs on economic complexity becomes positive. This level of the real per capita income is US$ 2.80 [= exponential (0.378/0.368)], and appears to be far lower than the minimum value of the real per capita income (US$ 316.13) in the full sample. We infer that the effect of the utilization rate of GSP programs on economic complexity is always positive, and becomes higher as the real per capita income improves. In other words, advanced bene ciary countries of GSP programs experience a higher positive effect of these trade preferences on economic complexity than relatively less advanced bene ciary countries. Second, we obtain from column [2] of Table 2 that the coe cient of "UROTP" is not signi cant (not even at the 10% level), while the coe cient of the interaction variable ("UROTP*[Log(GDPC)]") is negative and signi cant at the 5% level. The combination of these two outcomes suggests that the utilization of other trade preferences always affects negatively economic complexity, and higher income bene ciary countries tend to experience a higher negative effect of the utilization rate of other trade preferences on economic complexity than relatively less advanced countries. Based on all these outcomes, we conclude that bene ciary countries of NRTPs rely on GSP programs (at the expense of other trade preferences) to enhance their economic complexity.
The outcomes reported in column [3] of Table 2 show that coe cients of the two indicators of utilization rate of NRTPs are negative and signi cant at the 1% level, while the interaction term of the variable ["URGSP*UROTP"] is positive and signi cant. We conclude that both GSP programs and other trade preferences are strongly complementary in enhancing economic complexity in bene ciary countries that use simultaneously the two blocks of NRTPs. Especially, the utilization of GSP programs promotes economic sophistication when the utilization rate of other trade preferences exceeds the value of 0.759 (or 75.9%) (= 0.435/0.573). On the other hand, the utilization of other trade preferences fosters economic complexity when the utilization rate of GSP programs exceeds the value of 0.677 (or 67.7%) (= 0.388/0.573). It might be useful to have an idea on the slate of bene ciary countries that experienced a joint positive effect of the two blocks of NRTPs on economic complexity. We present in Appendix 3c the list of bene ciary countries (for the year 2018) whose joint utilization of both GSP programs and other trade preferences contributed to fostering economic complexity. The list of countries concerned in provided in the ascending order in terms of the utilization rate of the relevant NRTP. As it could be noted from this Appendix, in 2018, there were 25 bene ciary countries for whom GSP programs contribute to enhancing economic complexity, and 27 bene ciary countries for whom other trade preferences foster economic sophistication.
As the outcomes provided in column [3] of Table 2 represent 'average' effects across countries in the full sample, it might also be worth considering how the effect of the utilization of GSP programs affects economic complexity as the utilization rate of other trade preferences changes, and vice-versa. In this regard, Figure 3 shows, at the 95 per cent con dence intervals, the development of the marginal impact of the utilization of GSP programs on economic complexity for varying rates of utilization of other trade preferences.
[Insert Figure 3, here] The statistically signi cant marginal impacts (at the 95 per cent con dence intervals) are those encompassing only the upper and lower bounds of the con dence interval that are either above or below the zero line. This marginal impact appears to be increasing as the level of utilization of other trade preferences rises, but it is positive only for very high rates of utilization of other trade preferences. Nevertheless, it is not always statistically signi cant at the 5% level. As a matter of fact, this marginal impact is not statistically signi cant when the rates of the utilization of other trade preferences range between 0.7 and 0.855. Thus, countries whose utilization rate of other trade preferences is comprised between 0.7 and 0.855 experience no signi cant effect of the utilization of GSP programs on economic complexity. In contrast, countries whose utilization rate of other trade preferences is lower than 0.7 (i.e., 70%) experience a negative effect of the utilization rate of GSP programs on economic complexity, and for these countries, the higher the utilization rate of other trade preferences, the lower is the magnitude of the negative effect of the usage of GSP programs on economic complexity. Finally, countries that experience a high degree of utilization of other trade preferences (at least of 0.855, i.e., 85.5%) enjoy a positive effect of the utilization of GSP programs on economic complexity, and for these countries, the higher the rate of usage of other trade preferences, the greater is the magnitude of the positive effect of the utilization of GSP programs on economic complexity. The take-home message of Figure 3 is that GSP programs and other trade preferences are strongly complementary in enhancing bene ciaries countries' level of economic complexity, and does so for very high utilization rates of other trade preferences (for a given level of the utilization of GSP programs).
[Insert Figure 4, here] Figure 4 displays, at the 95 per cent con dence intervals, the development of the marginal impact of the utilization of other trade preferences on economic complexity for varying utilization rates of GSP programs. It shows a pattern similar to the one observed in Figure 3, with the exception here that this marginal impact is statistically nil for utilization rates of GSP programs ranging between 0.583 and 0.816. In other words, there is a positive effect of the utilization of other trade preferences on economic complexity when the rate of utilization of GSP programs exceeds 0.816, (i.e., 81.6%), and the greater this rate, the higher is the magnitude of the positive effect of the utilization of other trade preferences on economic complexity. As a result, the strong complementarity between GSP programs and other trade preferences in strengthening economic complexity in bene ciary countries takes place when for a given level of the utilization rate of other trade preferences, the rate of utilizing GSP programs is very high, and exceeds 81.5%.
[Insert Table 3, here] We now consider estimates reported in Table 3. To recall, these estimates concern recipient countries of development aid among bene ciary countries of NRTPs, and aim to check whether the effect of the utilization rate of GSP programs and other trade preferences on economic complexity changes sign and/or statistical signi cance when we include development aid variables in model (1). We note from these Tables that while the coe cients of "URGSP" and "UROTP" remain negative and signi cant at the 1% across all six columns (as we include different aid variables in the regressions), the magnitude of these coe cients remain lower than the ones obtained in column [1] of Table 2. Across the six columns of Table 3, these coe cients range between -0.139 and -0.114 for "URGSP" and between -0.147 and -0.109 for "UROTP". At the same time, only the coe cients of the variables capturing respectively the total development aid and AfT for trade policy and regulation are statistically signi cant, respectively at the 1% level, and the 5% level. All other aid variables hold coe cients that are not signi cant at the 5% level. The negative and signi cant effects of total development aid, and of AfT related to trade policy and regulation on economic complexity, and the lack of signi cant coe cients of the other aid variables at the 5% level, in the context where all coe cients of the variables "URGSP" and "UROTP" are signi cant at the 1% level, suggest that there may exist interaction effects between the usage of NRTPs and aid variables in in uencing economic complexity. The estimates presented in Table 4 help to address this question.
[Insert Table 4, here] We nd from Table 4 that while the variables measuring the utilization of GSP programs and other trade preferences hold coe cients that are negative and signi cant at the 1% level in all six columns of the Table, the interaction terms of all interaction variables are positive and signi cant at the 1% level. These results reveal that development aid ows (including all its major components) that accrue to bene ciary countries of NRTPs matter signi cantly for the effect of the utilization of NRTPs (either GSP programs or other trade preferences) on economic complexity. It seems, in particular, that the utilization of each of these two blocks of NRTPs affects positively economic sophistication when development aid in ows exceed certain amounts (as otherwise, the effect is negative). We have computed these amounts (turning points [39] for each relevant aid variable) corresponding to each of the two blocks of NRTPs, and reported them at the bottom of Table 4. It could be observed that the values of turning points associated with the relevant aid variables are all lower than the maximum value of each relevant aid variable. As a result, the effect of the utilization of NRTPs (either GSP programs or other trade preferences) on economic complexity is positive only when development aid ows are substantial, and exceed the amounts reported in Table 5 respectively for the total development aid, and each of its main components, namely AfT ows (and its three components) and NonAfT ows. The higher the amounts of development aid in ows (as far as these amounts exceed the levels reported at the bottom of Table 4), the greater is the magnitude of the positive effect of the utilization of NRTPs (GSP programs or other trade preferences) on economic complexity. The degree of complementarity between each block of NRTPs and NonAfT ows in promoting economic complexity is well higher than the level of complementarity between each block of NRTPs and AfT ows in fostering economic complexity. Moreover, we observe that in terms of the degree of complementarity between the components of total AfT ows and the usage of NRTPs (GSP programs or other trade preferences), AfT interventions for enhancing productive capacities show the highest degree of complementarity, followed by AfT interventions for trade policy and regulation, and then by AfT interventions for economic infrastructure. This is not surprising, as AfT ows for economic infrastructure is more sector-neutral (e.g., Cirera and Winters, 2015) than AfT ows for productive capacities, and AfT for trade policy and regulation.
These ndings suggest the existence of a strong complementarity effect between development aid ows (total development aid and each of its components) and the usage of both GSP programs and other trade preferences on economic complexity, in particular when such aid ows exceed certain amounts. These ndings, which support hypothesis 1, align well with the one obtained by Gradeva and Martínez-Zarzoso (2016), and suggest that NRTPs are effective in enhancing economic complexity in bene ciary countries when they are accompanied by high amounts of development aid.
[Insert Table 5, here] Estimates presented in Table 5 allow exploring the extent to which FDI in ows matter for the effect of a block of NRTP (GSP programs or other trade preferences) on economic complexity. We observe that the coe cients of the two indicators of NRTPs are both negative and signi cant at the 1% level, while the interaction terms of the interaction variables "URGSP*FDI" and "UROTP*FDI" are both positive and signi cant at the 1% level. These outcomes suggest that FDI in ows do matter for the effect of each of group of NRTPs on economic sophistication. Especially, each of these two blocks of NRTPs exerts a positive effect on economic complexity, notably when the share of FDI in ows in GDP exceeds a certain level. More precisely, the usage of GSP programs is positively associated with economic complexity when the share of FDI in ows in GDP is higher than 0.23 (or 23%) (= 0.292/1.270) (values of the variable "FDI" range between -0.112 and 1.38 in the full sample -see Appendix 2a). Otherwise, the utilization of GSP programs in uences negatively economic complexity. The effect of the usage of other trade preferences on economic complexity is positive (negative) when the share of FDI in ows in GDP is higher (lower) than 0.2045 (i.e., 20.45%) (= 0.236/1.154). Summing-up, the utilization of GSP programs in uences positively (negatively) economic complexity in bene ciary countries that experience a share (%) of FDI in ows in GDP higher (lower) than 23%, and the greater this share, the higher is the magnitude of this positive effect. In the meantime, the usage of other trade preferences in uences positively (negatively) bene ciary countries' level of economic sophistication when these countries experience an in ow of FDI (% GDP) higher (lower) than 20.45%. For these countries, the magnitude of the positive effect of other trade preferences on economic complexity increases as these countries attract greater in ows of FDI (as long as the share of FDI in ows in percentage of GDP exceeds 20.45%). These ndings lend support to hypothesis 2 set out above.
[Insert Figure 5, here] To get a better insight on these impacts, we display in Figure 5, at the 95 per cent con dence intervals, how the marginal impact of the utilization of GSP programs on economic complexity evolves for different shares of FDI in GDP (it is worth recalling that the FDI-to-GDP ratio is not expressed in percentage). We also depict in Figure 6, at the 95 per cent con dence intervals, how the marginal impact of the utilization of other trade preferences on economic complexity evolves for different shares of FDI in GDP. Figure 5 shows that the marginal impact of the utilization of GSP programs on economic complexity increases as the share of FDI in GDP rises. It is negative for values of the FDI-to-GDP ratio lower than 0.161, and positive for values of the FDI-to-GDP ratio higher than 0.283. For values of the FDI-to-GDP ratio ranging between 0.161 and 0.283, the marginal impact of the utilization of GSP programs on economic complexity is statistically nil. The key message conveyed by Figure 5 is that the utilization of GSP programs contributes to improving countries' level of economic complexity when they experience higher FDI in ows, in particular when the FDI-to-GDP ratio is higher than 28.3%. For these countries, the greater the FDI-to-GDP ratio (especially when it is higher than 28.3%), the higher is the magnitude of the positive effect of the utilization of GSP programs on economic complexity. However, when the FDI-to-GDP ratio is lower than 28.3%, there is at best a non-signi cant effect of the utilization of GSP programs on economic complexity, as this effect becomes negative when the FDI-to-GDP ratio is lower than 16.1%.
[Insert Figure 6, here] Figure 6 reveals a pattern similar to the one in Figure 5. The marginal impact of the utilization of other trade preferences on economic complexity takes both positive and negative values. Additionally, it increases as the share of FDI in GDP rises, although it is not always statistically signi cant. It is not statistically signi cant for values of FDI-to-GDP ratio ranging between 0.161 and 0.243. Thus, countries whose share of FDI in GDP ranges between 0.161 and 0.243 experience no signi cant effect of the utilization of other trade preferences on economic complexity. In contrast, countries whose FDI-to-GDP ratio exceeds 0.243 (i.e., 24.3%) experience a positive effect of the utilization of other trade preferences on economic complexity, and the greater the FDI-to-GDP ratio, the higher is the magnitude of the positive effect of the utilization of other trade preferences on economic complexity. On the other hand, countries that receive FDI in ows-to-GDP ratio lower than 16.1% experience a negative effect of the utilization of other trade preferences on economic complexity. For these countries, the lower the FDI-to GDP ratio, the higher is the magnitude of the negative effect of the utilization of other trade preferences on economic complexity. Overall, Figure 6 also carries the message that the usage of other trade preferences helps to improve countries' level of economic complexity when these countries enjoy higher FDI in ows, notably when the FDI-to-GDP ratio is higher than 0.243 (i.e., 24.3%).
Finally, we note that results concerning control variables in Tables 3 to 5 are consistent with those in Table 2, although with few exceptions such as a negative effect of terms of trade improvements on economic complexity observed, for example, in Tables 5 and 6 (this outcome suggests that improvements in terms of trade do not provide incentives for countries to diversify their export product baskets towards sophisticated products).
[37] As expected, the p-values of the AR(1) test are all lower than 0.10, while the p-values related to the AR(2) test and the Sargan test are all higher than 0.10.
[38] These outcomes are not presented here, to save space, and could be obtained upon request.
[39] These turning points (amounts) of development aid are obtained by dividing the coe cient of the relevant NRTP (either "URGSP" or "UROTP") by the coe cient of the associated interaction variable, and taking the exponential of the value obtained. For example, the amount of total development aid above which the utilization of GSP programs exerts a positive effect on economic complexity is given by US$ 503.7 million [= (exponential

Conclusion
Non-reciprocal trade preferences (NRTPs) have been provided by major (old) industrialized countries to developing countries with a view to helping them increase their export earnings, promote industrialization and accelerate economic growth rates, as per the Resolution 21(II) adopted by the members of the UNCTAD at the second conference of UNCTAD in 1968.
A wealth of studies has investigated whether the goal of "increasing export earnings" has been achieved by NRTPs offered by preference granting countries. Bearing in mind that the concept of "industrialization" may have different meanings, few other works have looked at whether NRTPs have contributed to promoting industrialization in the bene ciary countries (which is another stated objective of the NRTPs). They did so by investigating the effect of NRTPs (notably the eligibility to NRTPs, including to GSP programs) on export product diversi cation in these bene ciary countries.
The current paper contributes to the strand of the literature that has assessed whether NRTPs have been effective in promoting industrialization in bene ciary countries. It has examined whether NRTPs (GSP programs and other trade preferences) provided by the QUAD countries have helped bene ciary countries improve their level of economic complexity, i.e., by both diversifying their export products base, and exporting increasingly complex (i.e., high value-added) products that are not exported by many other countries. In other words, the current paper has examined the effect of the utilization of NRTPs offered by the QUAD countries on the bene ciary countries' level of economic complexity.
It differs in several fronts from previous works that have tried to assess whether NRTPs have been effective in promoting industrialization in bene ciary countries. One of these differences is its focus on 'economic complexity' rather than on 'export product diversi cation' as the way to measure the level of 'industrialization' of bene ciary countries. Another difference lies on the fact that in contrast with previous works that used either statistical indicators or dummies (that capture the eligibility to a NRTP) to evaluate the effect of that NRTP on bene ciary countries' degree of export product diversi cation, the present analysis has made use of the indicators of the utilization of NRTPs. In fact, eligibility to a NRTPs does not necessarily involve the utilization of that NRTP.
The present analysis has relied a panel dataset of 110 bene ciary countries of GSP programs and other (nonreciprocal) trade preferences offered by the QUAD countries over the period 2002-2018, to examine the effect of these NRTPs on the bene ciary countries' level of economic complexity. It has revealed several ndings.
First, bene ciary countries of NRTPs rely more on GSP programs (and not on other trade preferences) to enhance economic complexity. In fact, the utilization of GSP programs exerts a positive effect on economic complexity, with high income countries enjoying a higher positive effect than relatively less advanced countries.
Conversely, the usage of other trade preferences hampers economic complexity, and this negative effect is higher in high income countries than in relatively less advanced countries.
Second, GSP programs are strongly complementary with other trade preferences in enhancing economic complexity in bene ciary countries, in particular when the utilization rates of these NRTPs are high.
Third, development aid ows matter signi cantly for the effect of the utilization of NRTPs on economic complexity. Developing countries that enjoy both NRTPs and high amounts of development aid (that help to enhance trade capacity) experience a positive effect of the utilization of those NRTPs on economic complexity, and the magnitude of this positive effect rises as the amounts of aid in ows increase.
Fourth, and nally the utilization of both GSP programs and other trade preferences exerts a positive effect on economic complexity in countries that enjoy high shares of FDI in ows in GDP, with the magnitude of this positive effect rising as the volume of FDI in ows increases.
These outcomes show that non-reciprocal trade preferences can help bene ciary countries foster their level of economic complexity if they are used in a complementarity way. Bene ciary countries can also export complex products if the NRTPs that they enjoy, contribute to driving in higher FDI ows, including MNEs that would engage in exporting activities in the bene ciary countries, with a view to taking advantage of the trade preferences. The implementation by bene ciary countries of policies and reforms that promote a businessfriendly environment would enhance the attractiveness of MNEs in the exporting sectors that bene t from the NRTPs. Providers of NRTPs can also contribute to the promotion of FDI in ows in the bene ciary countries of NRTPs. Kumura and Todo (2010) have shown the existence of a possible "vanguard effect" of aid on FDI, whereby aid provided by a given donor to a developing country could promote FDI from the same donor to that country. Hence, a country that is a provider of both development aid and NRTPs could provide incentives to its enterprises (i.e., those already engaged in/or aiming at engaging in international production) to set up plants in bene ciary countries with a view to taking advantage of the NRTPs and exporting to their home countries at preferential tariffs. In fact, the supply of development aid by a donor-country to a developing country could enable the latter to reduce the investment risks perceived 'subjectively' by rms of the donor-country. Moreover, such aid may help the donor-country to develop close relationships with the recipient country (compared to nonproviders of aid to developing countries), which would, in turn, enable it to have access to speci c information concerning the local business environment of the recipient country. Such an information could be transmitted to rms of the donor country and facilitate the location of these rms' activities in the recipient country of both aid and NRTP(s).
Finally, the offer of both NRTPs and high development aid ows to developing countries would help to ensure that those preferences would be effective in expanding the manufacturing base of recipient countries, including through the production and export of increasingly sophisticated products.

Figure 2
Cross plot between the utilization of non-reciprocal trade preferences and economic complexity_over the full sample. Source: Author Figure 3 Marginal Impact of "URGSP" on "ECI", for varying levels of "UROTP". Source: Author

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Appendix.docx