Aid for Trade and Trade in Services

Abstract Existing research generally finds weak positive effects of aid for trade (AfT) on aggregate merchandise trade of recipients once endogeneity in the AfT-trade relationship is accounted for. In this paper, we confirm weak findings for both aggregate merchandise and services trade of recipients, using GMM and IV estimations. Moreover, estimates lose statistical significance if non-AfT explanatory variables are treated as endogenous in estimation suggesting identification issues may not have been adequately addressed in extant work. We then examine an alternative proposition: that effects of AfT and different categories of AfT may be observed along the conditional distributions of exports and imports. Our findings confirm this hypothesis. AfT allocated to economic infrastructure, productive capacity building in services and trade policies and regulation is more effective for smaller trading economies, especially in services. We also observe considerable heterogeneity in the trade effects of AfT allocated to individual services sectors, indicating the importance of country-specific diagnostics in targeting AfT allocation.


Introduction
The launch of the WTO Aid for Trade (AfT) initiative reflected a recognition that negotiations to lower trade barriers would benefit lower income countries more fully if complemented with development assistance targeted at improving the supply side of the economy (Njinkeu & Cameron, 2008). The international development community has provided significant volumes of aid for trade since the early 2000s (OECD and WTO, 2017). Much of this assistance has been devoted to economic infrastructure, improving productive capacities of firms and efforts to lower trade costs through trade facilitation projects.
Most of the empirical literature is devoted to examining the effects of AfT on different dimensions of merchandise trade, generally finding statistically significant correlations between AfT and different dimensions of goods trade. Work that accounts for endogeneity in the aggregate AfT-trade relationship generally finds weaker effects of AfT on aggregate merchandise trade of recipient countries. More recently, attention has moved to also studying the services trade effects of AfT (Ferro, Portugal-Perez, & Wilson, 2014;Hoekman & Shingal, 2020;Martínez-Zarzoso, Nowak-Lehmann, & Rehwald, 2017). The focus on services emanates from the increasing role that services are playing in all sectors of the economy and in international trade. A wide range of producer services activities such as finance, information and communications, transport, logistics, and professional services are inputs into modern production processes. The availability and cost of services determine economic opportunities and the performance of manufacturing and agricultural sectors. Many services are high productivity activities (Young, 2014) that offer prospects for positive external effects by contributing to the performance of other sectors. Many services are also critical for achieving the sustainable development goals (Fiorini & Hoekman, 2018).
In this paper, we further explore the effects of AfT on trade in services and make two contributions. First, using GMM and IV estimation, we confirm the findings on the weak effect of AfT on aggregate merchandise and services trade of recipients, but show that adding controls for the possible endogeneity of non-AfT explanatory variables results in estimates of the AfT-trade relationship that are no longer statistically significant. Second, since the theoretical AfT-trade literature provides reasons why AfT may or may not be associated with positive trade effects, we examine an alternate proposition: that trade effects of AfT and its sub-types are more likely to be observed along the conditional distribution of exports and imports. We examine this hypothesis using quantile analysis, incorporating recent advancements in the estimation of non-additive fixed-effects IV quantile regressions (Powell, 2015). 1 Three stylised facts provide the economic intuition for the alternative proposition. First, the bulk of AfT is allocated to sectors classified as services according to the OECD, including transport and storage infrastructure and information and communications technology (ICT) services. Second, trade costs for services are higher than those for goods, and the rate of decline observed in services trade costs since the early 2000s has been much less than that for goods (Miroudot & Shepherd, 2016), in part reflecting ad valorem equivalents of policy restrictions on services trade that are significantly higher than average import tariffs on goods (Jafari & Tarr, 2017;WTO, 2019). Third, many developing economies have seen significant growth in trade in so-called commercial services. During the 2000s, the group of least developed countries (LDCs) taken together expanded their services exports more rapidly than the world average, suggesting services are an area of revealed comparative advantage. The LDC share of global trade in services rose from 0.4 per cent in 2005 to 0.8 per cent in 2015, with commercial services exports growing by 14 per cent over this period, more than twice the rate of other countries (WTO, 2016), faster than exports of merchandise.
Given that services trade costs are high as a result of policy barriers to trade, technology-related supply constraints, and weaknesses in services-related infrastructure and institutions in developing countries, we expect that if AfT is effective, it would have a greater marginal impact in expanding services trade of low-income countries than on trade in goods or on trade of countries that have better infrastructure and institutions. The quantile regression analysis in this paper does not reject this hypothesis. Our results suggest that the heterogeneity of trade in services matters for responses to AfT: the effects of AfT allocated to services are both larger and more precisely estimated at lower quantiles of the services export distribution, and has more limited and smaller effects on merchandise trade relative to those observed for services trade. We also find that AfT allocated to non-services activities is not associated with a statistically significant positive effect on either exports or imports of goods and services.
Our identification strategy exploits changes in the AfT-recipient status of some of our sample countries over the time period of analysis (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) and instruments AfT with the average AFT in the geographical neighbourhood of a recipient country following Uberti and Jackson (2020). In a departure from existing literature, we also treat non-AfT explanatory variables as endogenous in our estimation strategy as these variables are unlikely to be exogenous to goods and services trade. 2 This more complete treatment of endogeneity results in estimated effects of AfT on aggregate merchandise trade of recipients no longer being statistically significant, in contrast to the positive, if weak, effects of AfT on merchandise trade found in existing literature.
The plan of the paper is as follows. Section 2 briefly discusses insights and hypotheses emerging from previous analysis on the relationship between AfT and trade and reviews the related literature. Section 3 describes the AfT variables and the allocation of AfT across countries and sectors. Section

Related literature: conceptualising the AfT-trade relationship and previous research
In their survey of the AfT-trade literature, Suwa-Eisenmann and Verdier (2007) note that 'aid flows may affect trade flows, either because of the general effects they induce in the recipient country, or because aid is directly tied to trade, or because it reinforces bilateral economic and political links (or a combination of all three).' Considerable ambiguity persists on the major transmission channels for the trade-enhancing effects of AfT, 'not to speak of the relative effects on trade in opposite directions' (Hühne, Meyer, & Nunnenkamp, 2014).
From a macroeconomic perspective, aid supplements domestic savings, permitting more investment that may foster higher rates of economic growth that in turn increases absorption capacity of the recipient, including for imports. If aid is conditional on trade liberalisation there may be a direct effect of AfT on trade, while other reforms associated with aid may promote economic growth and indirectly enhance trade. Conversely, aid may crowd out domestic investment, result in real exchange rate appreciation, worsening export competitiveness and the terms of trade. One channel for this is if aid is tied to over-priced imports of capital goods from donors. 3 In sum, while aid may be expected to have a positive impact on aggregate trade, much depends on how it is designed and used. Given that the use of tied aid should in principle have declined over time following donor implementation of the Paris principles on aid effectiveness, there is a presumption that positive effects are more likely to be observed in more recent time periods. Of course, this is conditional on donors no longer tying aid, on both a de facto and a de jure basis.
Aid allocated to economic infrastructure (transport, ICT, and energy) is expected to have the most direct effect on economic growth and trade, especially on recipient exports (Cali & Te Velde, 2011;Vijil & Wagner, 2012). If donors target AfT by selecting infrastructure projects that primarily serve their own export interests, they may also enhance recipient imports . As the following section will show, economic infrastructure accounts for the bulk of AfT allocated to developing countries over 2002-2015 and all its three constituent sectors are classified as 'services' by the OECD Secretariat. This, together with the increasing servicification of economic activity, explains the a priori positive and most direct relationship between AfT allocated to services activities and trade of recipient countries (both merchandise and services trade).
Most empirical research on AfT is cross-sectional in nature and involves cross-country analysis, assessing the effects of AfT on (different dimensions of) merchandise trade. Cadot, Fernandes, Gourdon, Mattoo, and de Melo (2014) review much of the literature. 4 A particular focus within this literature has been on AfT in support for trade facilitation, reflecting the effort to negotiate an agreement on trade facilitation in the WTO and efforts by developing countries to reduce trade costs. Martínez-Zarzosa, Nowak-Lehmann, Klasen, and Larch (2009) find that aid for trade has a strongly trade-promoting effect, especially AfT for trade facilitation.
The present paper relates to several strands of this literature. Ferro et al. (2014) focus on AfT directed towards service sector-related projects and activities and investigate the relationship between such AfT and merchandise exports. They find that AfT allocated to services sectors increases manufactured exports of the recipient countries. Hoekman and Shingal (2020) use both aggregate and available bilateral data for a subset of countries on merchandise and services trade and AfT allocated to services and non-services sectors, respectively, to examine the effect of AfT on trade. Their dyadic analysis suggests that AfT, in particular, that allocated to economic infrastructure, has a positive effect on donors' merchandise imports from recipient countries. In this paper we use a different instrument for AfT, control for the potential endogeneity of non-AfT explanatory variables and undertake quantile analysis to examine the effects of AfT for small versus large-value trading economies. Martínez-Zarzoso et al. (2017) also examine the effects of AfT on goods and services exports using quantile regressions and find that AfT mainly promotes goods exports for the lower quantiles of the conditional export distribution, with no effects observed on services exports. We complement this paper by studying disaggregated effects of AfT by sector and examining the effects of AfT and its sub-types on imports. Moreover, we account for non-additive fixed effects in the quantile regressions Aid for trade and services trade 1725 using the single-step procedure of Powell (2015), which likely explains the difference in our findings relative to Martínez-Zarzoso et al. (2017) who use the two-step approach of Canay (2011).
Another strand of research has examined the impact of AfT on investment. Recent studies include Selaya and Sunesen (2012) and Donaubauer, Meyer, and Nunnenkamp (2016). These studies generally find positive associations between measures of AfT and investment. We contribute to this literature by analysing the relationship between AfT and Mode 3 services trade flows (sales through a commercial presence), utilising a new WTO dataset, TiSMoS, that breaks down trade in services by modes of supply. 5 Table 1 summarises estimation methodologies, the treatment of endogeneity, reporting of diagnostic statistics on IV/GMM specifications and statistical significance of findings in existing crosscountry AfT-trade studies that use aggregate trade data. Early work finding positive AfT effects on merchandise trade was based on OLS estimations. Subsequent research that accounts for endogeneity in the AfT-trade relationship reports weaker, but still positive results, although most studies do not report complete diagnostic statistics on the validity of the instruments used. None of these studies treats all non-AfT explanatory variables as endogenous in their estimation strategy, which we do.

AfT definitions and allocations
In the empirical analysis that follows we use data from the OECD Creditor Reporting System (CRS). The CRS spans data on official development assistance (ODA) that is committed and disbursed by donor countries in recipient countries. The dataset spans a large sample of countries and sectors for the 2002-2015 period. AfT is a subset of total ODA and comprises the following three categories: (i) technical assistance for trade policy (for example, helping countries to develop trade strategies and negotiate trade agreements); (ii) trade-related infrastructure (for example, roads, ports, and telecommunications networks); and (iii) productive capacity building for trade (for example, supporting the private sector to expand exports). 6 The CRS does not provide data that exactly match these AfT categories. Only parts of ODA data are reported as aid going to building economic infrastructure and to the creation of 'productive capacity'. Infrastructure includes several services sectors -for example, transport, storage, and information and telecommunications networks -for which data are reported separately. Aid for productive capacity spans all sectors of the economy, and thus includes services. Three services activities are split out in the CRS for this productive capacity AfT category: banking and financial services, business and other services, and tourism. Note that CRS data are proxies at best for aid targeting trade-related infrastructure and productive capacity building, as not all of ODA reported under these headings is trade-related. This said, ODA statistics reported under these headings are the closest approximation of AfT going to services. 7 Total AfT disbursements increased from US$9.1bn in 2002 to an average of US$21bn in 2006to 39.8 USDbn in 2015(OECD and WTO, 2017. Asian and African countries have been the major recipients of AfT disbursements, each region accounting for around 40 per cent of total AfT global aid since 2002. The global distribution is qualitatively similar when we look at AfT allocated to services sectors. We define AfT for services to span the following six categories of AfT: (1) assistance to economic infrastructure in three sectors, transport/storage; ICT and energy; and (2) assistance for productive capacity building in financial services, business services, and tourism activities. We do so largely because these six categories are identified in the OECD data on AfT as services. 8 Globally, AfT mapped to these six categories increased from 59 per cent of total AfT to 72.4 per cent in 2015. Thus, most AfT over the period was allocated to services sectors, a feature of AfT that is generally not emphasised in AfT reporting or analysis. 9 Within services, the transport and energy sectors have been the largest recipients of global ODA disbursements, accounting for 45.9 and 30.2 per cent, respectively, of total AfT in services disbursed over 2012-2015 on average.
In the empirical analysis that follows, we abstract from the quality and suitability of the data on hand. As is stressed in the literature on trade in services, data quality is a concern and a constraint. Aid for trade and services trade 1727 The availability and quality of even aggregate services trade data in low-income countries are poor and the coverage of bilateral disaggregated services trade data amongst non-OECD countries is extremely limited. 10 We also take the OECD sector definitions of AfT as given but recognise there may be concerns whether annual AfT disbursements adequately capture the allocation and implementation of AfT within recipient countries and across sectors. There is considerable variation across countries between types of AfT. Similarly, there is variation in the time required for disbursing commitments, implementing projects and the duration of AfT projects. There will also be variable lags in the impact of completed projects on exports and imports of different types, all of which could have a bearing on estimates from empirical analysis. These data quality problems could affect the results of the analysis. In the case of trade data, use of bilateral data will greatly improve the quality of inference. But as these do not exist for many of the countries that are the focus of AfT, 11 there is no choice but to use aggregate data. This has the advantage of being consistent across countries -similar classifications are used by all countries. Lags in disbursement and average length of implementation of projects are both project-and countryspecific. Infrastructure takes more time than AfT going to trade policy regulation, so estimates of the trade effects of AfT allocated to economic infrastructure may be biased downwards. Similarly, countries with weak institutions and governance may experience longer lags between commitments and completion of projects. In the absence of detailed data on projects/countries/donors we rely on our explanatory variables and their variation to control for these factors.
Concerns have also been expressed by practitioners about the quality of AfT data, notably the definition of AfT (that is, differentiating between trade and non-trade-related ODA), which inherently embodies an element of subjectivity. While we recognise these concerns, data compilation by the OECD has been informed by regular consultations and review by donors and recipients, including through the bi-annual Aid for Trade review meetings that are held at the WTO and that bring together the international development community working on trade. Moreover, whatever biases may be embedded in the data as a result of categorisation decisions will apply to all flows reported by the OECD, and thus should not be a concern in terms of influencing only our results, as they will have a bearing on the findings observed in the entire empirical AfT-trade literature.

AfT and aggregate trade
Following previous studies, we assess the relationship between AfT and aggregate goods and services trade by estimating the following augmented export supply and import demand functions using fixed effects specifications 12 : where x jt is the log of services (goods) exports of AfT recipient j in year t; m jt is log of (goods) services imports of recipient j in year t; aft jt-1 is the log of AfT in recipient j in year t-1; z kjt-3 is a vector of recipient-time varying non-AfT controls lagged by three time periods to mitigate endogeneity-related concerns in estimation; δ j are recipient fixed effects; δ t are year fixed effects and ε jt is the error term. Consistent with other papers analysing the AfT-trade relationship, we allow trade flows to respond to AfT with a lag and also experiment with alternative lag structures. To accommodate zero AfT flows in the analysis (which are more prevalent in the disaggregated decompositions of AfT data), following the methodology suggested by Wagner (2003), we define aft jt-1 as ln(max{1,AfT jt-1 }) and include a NAfT jt-1 dummy in the estimating equations, which takes the value of 1 when AfT = 0 and is zero otherwise. Thus, the coefficient of aft jt-1 measures the elasticity of exports (or imports) where AfT is positive while the coefficient of NAfT jt-1 serves as an adjustment to the constant in cases where AfT is zero. The log of trade when AfT is positive exceeds the log of trade when AfT is zero by α 1 lnAfT À α 2 i:e: x jt AfT>0 À x jt To enhance comparability of results, we also follow the existing literature in the choice of explanatory variables (Cali & Te Velde, 2011;Martínez-Zarzoso et al., 2017). The control vector comprises a measure of country size -the log of population(POP jt-3 ); a measure of geographic distance to global markets -the log of market penetration (MP jt-3 ), computed as a distance (d ij ) weighted measure of other countries' GDP (GDP it-3 ), 13 that is MP jt-3 = Σ i (GDP it-3 /d ij ); a measure of domestic prices -(log of) the consumer price index (CPI jt-3 ), 14 and a measure of government effectiveness (GE jt-3 ) that reflects the institutional strength of the AfT recipient country. We expect each of these variables to be positively correlated with aggregate goods and services exports and imports, which justifies their choice as controls in the estimating equations.
Following Hoekman and Shingal (2020), we also include inward foreign direct investment (FDI jt-3 ) in the recipient country as an additional control. This is motivated by the observation that some twothirds of international provision of services occur through sales of services by foreign affiliates that have established a commercial presence in export markets (WTO, 2019). Moreover, FDI is a key element of many global value chains (GVCs) and a driver of the associated cross-border flows of services that occur within GVCs, both directly through the provision of 'headquarters services' and indirectly through embodiment of services in the value of the products that are produced.
Finally, in the spirit of Hühne et al. (2014), we include trade costs as additional controls in Equations (1) and (2). The TC jt-3 variable is constructed from a structural gravity model of bilateral (services and merchandise) trade over 1999-2012 with time-varying importer (j) and exporter (i) fixed effects and standard gravity controls (bilateral distance, contiguity, common language, common colonial antecedents, and common legal systems) as well as membership of preferential trade agreements (PTAs). The estimated coefficients (k1-k7) are used as weights in constructing TC jt-3 as follows: where n is the total number of exporting countries per importing country. The time-varying exporter and importer fixed effects in the gravity model also control for GDP and population of the partner countries. Since GDP (through market potential) and population are already included as control variables in Equations (1) and (2), we do not include these variables in constructing TC jt-3 as that is likely to lead to multicollinearity. To study trade effects by type of aid, we follow the OECD classification and decompose aggregate AfT into two parts, 'services' and 'non-services' as well as into three broad categories: AfT for economic infrastructure, AfT for productive capacity building, and AfT for trade policies and regulation. In addition, AfT in productive capacity building is further decomposed into services and non-services. Finally, we also examine the sectoral relationship between trade and AfT for seven disaggregated services sectors: business, communications, computer-and-related services, energy, financial, tourism, and transport services. 15 To control for endogeneity in the aggregate AfT-trade relationship, we deploy both IV and GMM (difference and system) estimations. Following Uberti and Jackson (2020), we instrument for AfT in each recipient country j at time t by the average AfT received by all its neighbouring countries in the geographical neighbourhood at the same time t. 16 This differentiates the IV analysis from that in Hoekman and Shingal (2020) who follow the synthetic instrument approach of Temple and Van de Sijpe (2017). In a departure from existing literature, we also treat non-AfT explanatory variables as Aid for trade and services trade 1729 endogenous in our estimation strategy in both GMM and IV specifications. In the latter, these variables are also instrumented using their 'neighbourhood-averages' as done in the construction of the AfT instruments.

Quantile analysis
We use quantile regressions to examine whether the trade effects of AfT depend on the magnitude of the trade of AfT recipient countries, to reflect the possibility that AfT effects are more likely to be observed for countries with low levels of exports or imports (that is, the marginal effect of AfT may be higher for such countries) and/or countries that already have an established 'trade footprint' and can use AfT more effectively than other nations to scale up trade flows further. In both cases, such effects could work through the intensive or the extensive margin.
In quantile regression models, the quantiles of the conditional distribution of the dependent variable are expressed as functions of the observed covariates. Their main advantage lies in interpreting potentially different solutions at distinct quantiles as differences in the response of the dependent variable to changes in the regressors at various points on the conditional distribution of the dependent variable. In the context of this paper, quantile regressions allow us to trace the entire export and import distribution, for goods and services, respectively, conditional on the regressors included in Equations (1) and (2).
The estimation of these equations based on the q th quantile regression (0 < q < 1) and the set of covariates Z jt minimises the absolute value of the residual. The objective function is as follows (Cameron & Trivedi, 2009): where y jt is the dependent variable and β is the vector of estimated parameters.
To account for non-additive fixed effects in the quantile regressions, we use the single-step procedure of Powell (2015). This is an improvement over the two-step approach of Canay (2011), as it addresses a fundamental problem posed by fixed-effect quantile estimators, that is inclusion of fixed effects alters the interpretation of the estimated coefficient on the treatment variable. In conditional quantile models, the parameters of interest are assumed to vary based on a nonseparable error term. Canay (2011) assumes the fixed effects to be location shifters and uses estimates of these fixed effects from a within-FE model in stage one to demean the dependent variable, the transformation of which is then used as the dependent variable in quantile analysis in stage two. This treatment of fixed effects alters the structure of the quantile function, causing a bias even if the treatment variable is randomly assigned. The single-step estimation of Powell (2015) circumvents this problem by maintaining the nonseparable error term commonly associated with quantile estimation.
The analysis is carried out on 159 ODA recipients over 2002-2015; the sample of recipients is reported in Annexe Table A. Fourteen countries in our sample witnessed a change in their AfTrecipient status over the period of analysis, a fact that we exploit in identification. 17 Summary statistics are reported in Annexe Table B. The dataset has over 1800 observations on goods and services trade and the aid variables.

Aggregate analysis (IV)
Equations (1) and (2) are estimated separately for goods and services exports and imports, using IV and GMM specifications with three decompositions of AfT: (i) distinguishing AfT allocated to services (AfT_Ser) from AfT going to non-services sectors (AfT_Non_Ser); (ii) splitting AfT into four types: AfT allocated to economic infrastructure (AfT_EI), productive capacity building in services (AfT_PCB_Ser), productive capacity building in non-services (AfT_PCB_Non_Ser), and support for trade policies and regulations (AfT_TPR); and (iii) differentiating between the sectoral allocation of services AfT. The data permit us to distinguish between AfT allocated to transport services, communications services, financial services, energy services, computer-and-related services, other business services, and travel services. The AfT variables were lagged by one, two and three time periods, in distinct specifications, while the non-AfT regressors were lagged by three time periods. All regressions controlled for country (recipient) and year fixed effects, with standard errors clustered by country-year.
The 2SLS IV estimates for AfT and its sub-types for both goods and services exports and imports are found not to be statistically different from zero across all lag structures (see Table 2). This is a significant departure from the positive, albeit weak, effects of AfT on merchandise trade that have been found in the IV results in existing literature. This is likely attributable to our treatment of the non-AfT regressors as endogenous.
GMM results (not reported) show some evidence of positive effects for disaggregated and sectoral AfT, especially on the export side, but the validity of the instruments was consistently rejected in the diagnostic statistics. 18 As services trade effects of AfT might vary across modes of supply used, we also used a new database released by the WTO Secretariat, TiSMoS, breaking down aggregate trade in services across modes of supply for 200 economies over 2005-2017 to replicate the analysis for different modes for the overlapping time period (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012). The IV estimates were found not to be statistically significant irrespective of the mode of service delivery and lag structure.

Quantile analysis (IV)
The lack of a statistically significant relationship between aggregate trade in goods or services and AfT received by a country is consistent with the theoretical AfT-trade literature which concludes that the sign of the association is ambiguous for both recipient country exports and imports. In this Aid for trade and services trade 1731 subsection, we examine an alternate proposition, that trade effects of AfT and its sub-types may more likely be observed along the conditional distribution of trade, than at the conditional mean. The economic intuition underlying this proposition was discussed in Section 1.
We examine this hypothesis using fixed effects IV quantile regressions. 19 The full set of results from these regressions, for AfT variables lagged by one time period, are reported in the online Table 2: Impact of Non-AfT, AfT and its sub-types on goods and services trade (IV estimates) appendix Table 1-6 for exports and imports, respectively. The estimated coefficients and associated confidence intervals depicting the main findings from these results are shown in Figure 1. Estimates of all the non-AfT regressors were found to be statistically significant with the expected signs, suggesting that the empirical model is well-specified for both exports and imports. In all cases, the R-squares range from 75 to 85 per cent, indicating that the explanatory variables account for substantial variation in the dependent variables.

Aid for trade and services trade 1733
The results reported in online appendix Tables 1 and 3 reveal that the effects of AfT allocated to services, including economic infrastructure and productive capacity building, as well as AfT allocated to trade policies and regulation, are both larger and more precisely estimated for lower quantiles of the services export distribution. As the countries that account for this part of the distribution tend to be lower-income and smaller economies these results suggest such AfT is more likely to achieve its purported objective -trade promotion -for certain types of recipients. Illustratively, a doubling of AfT allocated to services activities is associated with a 3.5 per cent increase in services exports at the first quantile, an 18.6 per cent increase at the third quantile, 1.3 per cent increase at the fifth quantile, and a 4.8 per cent increase at the seventh quantile (see columns 1, 5, 9 and 13 of online appendix Table 1). The effects of AfT and its broad sub-types on services imports, reported in Tables 3 and 5, follow a similar pattern as that for services exports but are smaller in magnitude.
In contrast, evidence on the relationship between AfT in services and merchandise trade in the results reported in appendix Table 1-4 is more limited and the effects are also smaller in magnitude relative to those for services trade. We observe a positive effect of AfT allocated to services activities only on merchandise imports for the second and eighth quantiles of the merchandise imports distribution -a doubling of such AfT is associated with a 3.4 and 0.96 per cent increase, respectively (see columns 4 and 16 of appendix Table 2). In contrast, AfT allocated to productive capacity building in services reports larger effects on both merchandise exports and imports at the lower quantiles, with the maximum impact observed for the first quantile in the case of exports (estimated coefficient of 0.05 in column 2 of appendix Table 3) and for the second quantile in the case of imports (estimated coefficient of 0.11 in column 4 of appendix Table 4). The effects of AfT allocated to trade policies and regulation also decline along the distribution of merchandise exports and imports and are smaller in magnitude to those for services exports (see appendix Table 3).
AfT allocated to non-services activities does not have a positive effect on either exports or imports of goods and services -the quantile analysis results for non-services AfT are not statistically significant in the results reported in appendix Table 1-4. The complete absence of a statistically significant positive effect of AfT allocated to non-services activities on trade in both aggregate and quantile analysis is likely attributable to a 'volume' effect given that AfT allocated to the six services categories accounted for more than 70 per cent of total AfT in 2015.
Our findings contrast with Martínez-Zarzoso et al. (2017), where the trade effects of AfT decline along the conditional distribution and are observed primarily for goods exports. This is likely attributable to accounting for non-additive fixed effects in the quantile regressions using the singlestep procedure of Powell (2015), vis-a-vis the two-step approach of Canay (2011) employed by Martínez-Zarzoso et al. (2017).
As expected, there is considerable heterogeneity in the quantile analysis results for AfT allocated to different services sectors in the results reported in appendix Tables 5 and 6, for exports and imports, respectively. Effects of AfT allocated to communications and travel services (as well as computer-related services and services imports) follow a U-shape for both goods and services along the conditional distribution of exports and imports. In the case of financial services (as well as computer-related services and merchandise exports), the effects exhibit a declining pattern. Thus, AfT allocated to these sectors seem to matter for smaller trading economies, which presumably are the primary focus of AfT efforts. In contrast, AfT allocated to energy and other business services is more effective for larger services trading economies, with apparent limited salience for merchandise traders. Estimates for AfT allocated to transport services are not statistically significant for either exports or imports of goods and services.
In sum, AfT allocated to services sectors and activities is found to enhance services exports of smaller exporting countries, suggesting that such AfT meets its claimed objective. This is particularly true of AfT allocated to economic infrastructure and productive capacity building in services and AfT allocated to communications, financial and travel services at the sector-level. The finding that larger exporting countries benefit from AfT allocated to trade policies and regulation makes intuitive sense given that such countries are likely to be relatively less capacity constrained and hence, more capable of utilising aid to their advantage.
An implication of these results for AfT design and implementation is that the heterogeneity of trade matters for responses to AfT. From a donor perspective, the same volume of aid allocated to services activities may be more effective for small services exporters and importers. If the objective includes maximising returns on aid allocation from both the individual donor and the international donor community perspectives, then these results indicate that the marginal gains from AfT to smallvalue trading economies may be larger, which has important implications for aid-targeting. Moreover, in so far as some of this aid also enhances recipient imports from donors (and even recipient exports to donors in a world of GVCs where cheaper imported inputs matter), it also strengthens the political economy argument in favour of providing aid.

Conclusion
Many dimensions of the potential relationship between AfT and the trade performance of recipient economies have been studied in the literature on this subject. A common characteristic of this research is that it mostly focuses on the effects of AfT on merchandise trade, and to a lesser extent, on investment flows. The analysis in this paper complements Ferro et al. (2014) and Hoekman and Shingal (2020) by focusing on AfT and trade in services, and Martínez-Zarzoso et al. (2017) by examining the AfT-trade relationship along the conditional distribution of exports and imports using quantile regression analysis.
Our results suggest that the effects of AfT allocated to services, including economic infrastructure and productive capacity building, as well as AfT allocated to trade policies and regulation are both larger and more precisely estimated for smaller services exporting and importing countries. Thus, AfT disbursement seems to meet its purported objective -to expand recipient participation in global trade. The results also indicate that the heterogeneity of trade, especially trade in services, matters for its response to AfT. Thus, smaller services trading economies may be more responsive to the allocation of AfT and its major sub-types, which can be a useful take-away for AfT design and implementation. At the same time, AfT allocated to non-services activities does not have a positive effect on either exports or imports of goods and services, which again is relevant for AfT design and implementation. The effects of a given type of AfT may vary across recipient countries, suggesting that policy-makers need to target AfT carefully and avoid a 'one-size-fits-all' approach in determining where to allocate AfT.
The finding of limited effects of services AfT on merchandise trade in the quantile analysis suggests limited complementarities between goods and services. This could be for several reasons. Most AfT recipients in our analysis already have an established 'trade footprint' in merchandise but are much more recent services traders. Their services trade is thus also at a much lower base relative to their merchandise trade, which makes any marginal effect of AfT more observable. Moreover, barriers to their services trade are much larger than those to their merchandise trade (Jafari & Tarr, 2017;WTO, 2019), which again makes the marginal impact of AfT on their services trade larger than that for their merchandise trade. This is suggestive of AfT allocated to services activities being welltargeted and the salience of the sector definitions used by the OECD Secretariat. Finally, the effects of services AfT on merchandise trade may be less pronounced than expected insofar as servicification of manufacturing is more likely to be observed in services value-added data than in the gross services trade data analysed in this study.
From a methodological perspective, our analysis reveals the need to consider the possibility that endogeneity can affect the AfT-trade relationship not just through the AfT variables but also through the non-AfT controls. A more complete treatment of this endogeneity is important for identification of treatment effects. Finally, given the potential time lags involved in the impacts of AfT disbursements, we conclude with a caveat. It may well be the case that the time period for empirical analysis in this paper is not long enough to observe effects of AfT and its sub-types on aggregate trade flows at the mean. Thus, fourteen years and three lags may not be sufficient to examine longer-term effects Aid for trade and services trade 1735 of AfT on trade at the aggregate level. Assessing such longer-term effects of AfT on trade remains an important agenda for future research. 20