This section first discusses the difference-in-differences (DID) framework used in the analysis, including the control group and its validity (sub-section 3.1). It then presents the model specification upon which we rely to explore the effect of the Waiver on LDCs' services exports (sub-section 3.2).
3.1. Difference-in-Difference framework
The analysis uses the difference-in-differences (DID) framework to uncover the causal effect of the Waiver on LDCs' services exports by comparing the services export performance by the beneficiaries of the Waiver (i.e., the LDCs) - referred to as the treatment group - with services export performance of a control group, over two time periods (pre-versus post-intervention, i.e., pre-versus post-operationalization of the Waiver). The control group contains countries that did not benefit from the Waiver. However, the DiD framework rests on the strong parallel trends assumption, which requires that the difference between the outcome (here, services export performance) of the treatment group (i.e., LDCs) and the control group would be constant over time in the absence of the intervention (i.e., in the period preceding the operationalization of the Waiver). In other words, as per this assumption, the average outcomes of treated and control units would follow parallel trends over time in the absence of the intervention (Abadie, 2005). While observable and unobservable factors may cause the level of the outcome variable (here services exports) to differ between the treatment group and the control group, such a difference (in the absence of the reform in the treatment group, i.e., in the absence of the operationalization of the Services Waiver) must be constant over time. In reality, given that the treatment group is only observed as treated, the assumption is fundamentally untestable (e.g., Fredriksson and Oliveira, 2019).
Given the unique features of the LDCs compared to other countries in the world, it was difficult to find a control group that would fit the purpose of the present analysis, including fulfil the parallel trends assumption. In fact, as all countries in the group of LDCs (designated as such by the United Nations) are categorized as low-income countries (LICs) by the World Bank, it was not possible to constitute a control group using the World Bank's group of LICs. We finally opt for relying on the group of countries defined by the International Monetary Fund (IMF) as LICs. These are countries eligible for the Poverty Reduction and Growth Trust (PRGT[13]) facilities. They include overall 69 countries located in Africa, Asia, and Latin America (see IMF, 2021: p34). When compared the countries included in the category of LDCs to those in the category of LICs (as defined by the IMF), we notice that all LDCs except for 'Angola' are in the category of the PRGT-eligible LICs. Therefore, our control group includes LICs (as defined by the IMF) that are not LDCs, and for which data is available to perform the analysis. Overall, 22 countries have been included in the control group (see the list of these countries in Appendix 3).
To assess the validity of the parallel trends assumption, we start by visualising the plot of the services exports indicators in the treatment and the control groups, over the period from 2005 to 2019. However, Wing et al. (2018, p. 459) have noted that while the visual plot can serve as a precursor to a statistical test of the parallel trends assumption, it has the drawback of being less compelling when the data are noisy or when the time series is short. This is because in these instances, it is difficulty to make a distinction between statistical noise and genuine deviations from the common trends. As a result, we also perform a formal statistical test of the parallel trends assumption. As explained by Wing et al. (2018, p. 459), this test entails first adding specific time trend-terms for the treatment and control groups in the baseline regressions (i.e., the time trend interacted with each group). Second, as the common trends model is nested in the group-specific trend model, the test of the common trends model is an F-test of whether all the coefficients of the group-specific linear trends are jointly zero (null hypothesis). The common trend hypothesis is valid if we do not reject the null hypothesis. The results of the statistical test (F-test) of the parallel trends assumption are reported at the bottom of the relevant results' Tables.
To have a first idea on the validity of the parallel trends assumption, we perform analyses of a number of graphs, using as indicators of modern services exports, the share (not in percentage) of modern services exports in GDP (denoted "MSE"), and alternatively the performance in terms of modern services exports relatively to world performance in terms of modern services exports (denoted "MSEPERF") (see for example, Okafor and Teo, 2019: p3457). For traditional services exports, we use the share (not in percentage) of traditional services exports in GDP (denoted "TSE") and alternatively, the performance in terms of traditional services exports relatively to world performance in terms of traditional services exports (denoted "TSEPERF").
Figure 1 shows the plot of the share of modern and traditional services exports in GDP, in the treatment and control groups. Figure 2 presents the plot of the performance of the treatment and control groups in terms of modern and traditional commercial services exports (relatively to the world). Figures 3 to 5 compare the development of the 'average' values of the indicators of commercial services exports in the treatment and control groups over the full period. Figure 3 presents the development of the share (not in percentage) of total commercial services in GDP (which is the sum of "MSE" and "TSE") in the treatment and control groups. This helps get an insight into how the total commercial services exports (expressed as a share of GDP) had evolved in these two groups of countries. Figures 4 and 5 show respectively the developments of "MSE" and "TSE" in the treatment and control groups, and the developments of "MSEPERF" and "TSEPERF" in the treatment and control groups.
[Insert Fig. 1, here]
[Insert Fig. 2, here]
[Insert Fig. 3, here]
[Insert Fig. 4, here]
[Insert Fig. 5, here]
The comparison of the trends of each indicator in the treatment and control groups in Figs. 1 and 2 lends support for the parallel trends assumption. We observe from Fig. 3 that the share of total commercial services in GDP steadily increased in both LDCs and the control group, and additionally tended to move in tandem in the two groups of countries – and this confirms the parallel trends assumption. However, countries in the control group consistently experienced on average, a higher share of total commercial services in GDP than LDCs. This share moved from 5.6% in LDCs (against 13.9% for countries in the control group) in 2005 to 6.8% in LDCs (against 20.4% for countries in the control group) in 2019.
We note from Fig. 4 that the parallel trends assumption (in particular between 2005 and 2013 (i.e., before the operationalization of the Services Waiver) tends to be more valid for "MSE" between the LDCs and the control group than for "TSE" between these two groups. In addition, in both groups, the share of traditional services exports in GDP is well higher than the share of modern services exports in GDP. The share of traditional services exports in GDP increased from 4.5% in LDCs (against 11.8% for countries in the control group) to 5.7% in LDCs (against 18.7% for countries in the control group) in 2019. Concurrently, countries in the control group experienced a higher share (slightly more than the double) of modern services exports in GDP than LDCs. The share of modern services exports in GDP increased from 1.1% in LDCs (against 2.53% for countries in the control group) in 2005 to 0.96% in LDCs (against 2.46% for countries in the control group) in 2019. The patterns of the developments of the indicators of performance in terms of modern and traditional services exports (relatively to the world) in LDCs and the control group displayed in Fig. 4 are quite similar to the ones observed in Fig. 3. Overall, building on the graphical analyses of Figs. 1 to 5, we can accept the hypothesis of parallel trends of indicators of modern and traditional services exports between LDCs and the control groups, even though a word of caution may be in order here. This is because as noted by Wing et al. (2018), the presence of noise in the data or the relatively short time-period of our dataset may not provide a compelling evidence of the validity of the parallel trends assumption.
3.2. Model specification
In light of the discussion in the previous section, and building on previous works on the macroeconomic determinants of services exports (e.g., Eichengreen and Gupta, 2013a; Sahoo and Dash, 2014; 2017), we postulate a simple model specification that includes as control variables, three main key determinants of services exports, and where the dependent variable is an indicator of modern or traditional commercial services exports. The three control variables are the share (not in percentage) of foreign direct investment inflows in GDP (denoted "FDIGDP"), the real per capita income (denoted "GDPC") and the population size (denoted "POP"). Appendices 2a, 2b and 2c present the descriptive statistics on these variables respectively over the full sample, the treatment group (i.e., LDCs) and the control group. Appendix 3 reports the list of countries contained in the treatment and control groups.
We consider the following parsimonious model specification:
$$Log({SEXP)}_{it}={\alpha }_{1}{Log\left(SEXP\right)}_{it-1}+{\alpha }_{2}\left[{(LDC}_{i})*{\left(WAIVER\right)}_{t}\right]+{\alpha }_{3}{LDC}_{i}+{\alpha }_{4}{WAIVER}_{t}+{\alpha }_{5}{FDI}_{it}+{\alpha }_{6}{Log\left(GDPC\right)}_{it}+{\alpha }_{7}{Log\left(POP\right)}_{it}+{\alpha }_{8}{DUMOUT}_{it}+{\mu }_{i}+{\delta }_{t}+{ϵ}_{it}$$
1
The subscripts i and t refer respectively to a country, and a year. The panel dataset is unbalanced and contains overall 66 countries, of which 44 LDCs (i.e., the treatment group) and 22 countries (i.e., the control group). \({\alpha }_{1}\) to \({\alpha }_{8}\) are coefficients to be estimated. The parameter \({\alpha }_{2}\) represents the difference-in-difference effect between control countries pre-and-post-operationalization of the Waiver, and treated countries pre-and post-operationalization of the Waiver. It uncovers the causal effect of the Waiver on services exports. In the empirical analysis, we denote \(DiD=\left[{(LDC}_{i})*{\left(WAIVER\right)}_{t}\right]\).
\({\mu }_{i}\) are time invariant countries' fixed effects, and \({\delta }_{t}\) are year dummies that represent global shocks affecting all countries' services exports. The introduction of these dummies in the baseline model allows eliminating time-related shocks from the error term, and hence avoiding the presence of contemporaneous correlation in the error term. \({ϵ}_{it}\)is a well-behaving error-term.
The variable "SERV" is the indicator of services exports, notably modern services exports, and traditional services exports. As noted above, the indicators of modern services exports are alternatively "MSE" and "MSEPERF". In reference to Okafor and Teo (2019: p3457), the indicator "MSEPERF" (the performance index in terms of modern services exports relatively to the world) is computed for a given country in a given year, as the ratio of that country's share of modern services exports in the world's modern services exports, to the country's share of GDP in the world's GDP. In other words, it is computed as follows: \({MSEPERF}_{it}= \frac{{MSE}_{it}}{{MSEW}_{it}}\), where the variable "\({MSE}_{it}\)" is a country's share of modern commercial services in GDP, and "\({MSEW}_{it}\)" is the share of the world modern commercial services exports in the world GDP.
Likewise, the indicator of traditional services exports is either the share (not in percentage) of traditional services exports in GDP (denoted "TSE") or the performance in terms of traditional services exports. The latter (denoted "TSEPERF") measures how a country is competitive in terms of traditional services exports relatively to the world. It is computed for a given country, in a given year, as the ratio of that country's share of traditional commercial services exports in the world's traditional services exports to the country's share of GDP in the world's GDP. In other words, it is defined as follows: \({TSEPERF}_{it}= \frac{{TSE}_{it}}{{TSEW}_{it}}\), where the variable "\({TSE}_{it}\)" is a country's share of traditional commercial services in GDP, and "\({TSEW}_{it}\)" is the share of the world traditional commercial services exports in the world GDP.
The main problem when computing these indicators of services exports is to define services export items would be included in the categories of "modern services exports" and "traditional services exports". As a matter of fact, the literature has not provided a clear delineation between these two concepts (e.g., Eichengreen and Gupta, 2013a). For example, according to Eichengreen and Gupta (2013b), the category of 'modern services' includes communications, computer, information and other related services, and the category of 'traditional services' covers trade and transport, tourism, financial services and insurance. The authors have pointed out that insurance and finance services can be classified in either category (see Eichengreen and Gupta, 2013a: page 2 - footnote 5). A slightly different classification of services sectors has been adopted by Sahoo and Dash (2014, 2017). For example, Sahoo and Dash (2017) have built on the work by Baumol (1985), Ghani and Kharas (2010) and Eichengreen and Gupta (2013a), and considered that the category of "modern services" includes transportability and tradability, financial services, insurance, business processing and software services. On the other side, "traditional services" cover transport and travel services.
Building on these previous works (e.g., Eichengreen and Gupta, 2013a; Sahoo and Dash, 2014; 2017), and using the dataset on services exports developed by WTO/UNCTAD in cooperation with the International Trade Centre and the United Nations Statistics Division, the present analysis considers that the category of "modern services" covers the major sub-sectors of 'Insurance and pension services'; 'Financial services'; 'Telecommunications, computer, and information services'; 'Charges for the use of intellectual property n.i.e'; and 'Other business services'. The category of "Traditional services" covers the following major sub-sectors: 'goods-related services (i.e., manufacturing services on physical inputs owned by others and Maintenance and repair services)'; 'Transport'; 'Travel'; 'Construction'; and 'Personal, cultural, and recreational services'.
All four indicators of services exports described above have been computed using these categorizations of modern and traditional services exports. The computed indicators have been taken in natural logarithm because of their high skewed distributions. The one-year lag of the dependent variable has been introduced as a right-hand side regressor in the model (1) in order to take into account the strong persistence of commercial services indicators over time.
Also, in light of the skewed distributions of the variables "FDIGDP", "GDPC" and "POP", these variables have been transformed. "GDPC" and "POP" have been transformed using the natural logarithm as they contain strictly positive values. As the variable "FDIGDP" contains both positive and negative values, it has been transformed using the procedure proposed by Yeyati et al. (2007), which is as follows: FDI \(=sign\left(FDIGDP\right)*\text{l}\text{o}\text{g}(1+\left|FDIGDP\right|)\) (2), where \(\left|FDIGDP\right|\)refers to the absolute value of "FDIGDP" (which to recall, is not expressed in percentage).
"DUMOUT" is a dummy variable that takes the value of 1 for outliers identified in the dataset for each indicator of services exports share (see for example Figs. 1 and 2 to visualize these outliers).
Effect of the real per capita income
An increase in the real per capita income reflects an improvement in the development level, and the existence of economies of scale (e.g., Li et al., 2005; Nyahoho, 2010; Schulze, 1999). Linder (1961) has argued that the per-capita income reflects the demand structure for goods and services. Krugman (1981) has proved theoretically that economies of scale is a critical determinant of trade in goods and services. Therefore, we postulate that the presence of economies of scale can be associated with an increase in the demand for services, including for new services and consequently enhance services production. However, the extent to which the rise in services production could translate into higher or lower services exports may depend on several factors, including the relative price of the services items in the domestic and exports markets. Overall, the effect of the real per capita income on services exports is to be determined empirically.
Nonetheless, we can expect that the LDC Services Waiver would lead to higher services exports, including modern services in LDCs (such as Bangladesh) that enjoy a higher per capita income than in LDCs with a relatively lower per capita income (hypothesis 2). This can be because relatively high income LDCs (like Bangladesh) are likely to export more modern services exports than traditional ones. The exports of modern services exports could be driven by the export by these countries of manufacturing goods (although light ones such as textile) (see for example, Eichengreen and Gupta, 2013a; Gnangnon and Priyadarshi, 2016; Gnangnon, 2021b). In contrast, very low-income countries among LDCs are likely to export more traditional services than modern services, notably because their reliance on primary export products is likely high.
Effect of the population size
The population size, which reflects the country's size, can affect services exports. The international trade literature has established that countries with larger populations tend to trade less than smaller countries, as the small population size provides firms with less opportunities to expand their sales in the domestic market (e.g., Alesina and Wacziarg, 1998; Bleaney and Neaves, 2013). As far as services exports are concerned, Goswami et al. (2012) have argued that an increase in the population size can translate into a higher demand for services by final consumers, and lead to a larger services sector. Kong et al. (2021) have uncovered that a greater size of the domestic market can significantly boost a country's service exports. However, countries with larger populations may not necessarily export more services than countries with smaller populations, in particular if domestic trading firms opt for primarily satisfying the demand for services in the domestic market (and thus export less services items). Thus, the effect of the population size on modern and traditional services exports is to be determined empirically. Similarly, the extent to which the effect of the Waiver on LDCs' services exports would depend on these countries' population size is to be addressed empirically. For example, if an increase in the population size is associated with a higher demand for new services items, including modern services, the Waiver may induce a higher export of modern services items (hypothesis 3).
Effect of FDI inflows
The variable "FDI" represents the foreign direct investment flows to a country, and has been introduced in model (1) in light of the potential of FDI inflows to promote services exports (e.g., Eichengreen and Gupta, 2013a,b; Grünfeld and Moxnes, 2003; Lancheros and Demirel, 2012; Srivastava, 2006; Wong et al., 2006), and notably modern services exports (e.g., Sahoo and Dash, 2014; 2017; Tharakan et al., 2005). Maurer and Magdeleine (2008) have pointed out that two-thirds of international trade in services takes place through Mode 3 of services (i.e., commercial presence), including FDI inflows. The services sector tends to be the predominant destination of FDI inflows, and represents now two-third of global FDI stock (e.g., UNCTAD, 2016). FDI inflows can enhance services exports through its positive effect on the supply-side determinants of modern services exports, including the quality of physical capital and workers' skills (e.g., Sahoo and Dash, 2014; De Gregorio, 1992). The presence of multinational firms in the host countries can also allow local firms to have access to better inputs (e.g., Eck and Huber, 2016; Javorcik et al., 2018) and benefit from improved technical and organizational competencies, including through knowledge transferred by MNEs (e.g., Annique and Rodríguez, 2018; Chen et al., 2022; Havranek and Irsova, 2011; Saliola and Zanfei, 2009). All these positive spillovers of the presence of MNEs in the host countries' local economy can significantly promote goods exports (e.g., Antonietti and Franco, 2021; Harding and Javorcik 2012; Javorcik et al. 2018; Zhu and Fu, 2013), and services exports, although for the latter, the effect can be both direct and indirect, including through the export of goods. In fact, several works have emphasized that trade in goods (including goods exports) is complementary with trade in services (e.g., Ceglowski, 2006; Eichengreen and Gupta, 2013b; Gnangnon and Priyadarshi, 2016; Karmali and Sudarsan, 2008; Kimura and Lee, 2006; Sahoo and Dash, 2014).
As far as LDCs are concerned, a recent report by the UNCTAD (UNCTAD, 2022) has noted that LDCs continue to attract significant FDI flows in extractive industries, as this sector remained the top recipient for international project finance deals in 2019. This is not surprising as LDCs' exports are highly commodity-dependent[14] (WTO, 2021a). In spite of the important volume of FDI flows located in extractive industries and natural resource sectors in LDCs, several LDCs had experienced some sectoral diversification of FDI over the last decade, both in the non-resource-based manufacturing industries[15] and in the services sector, including targeted utilities, ICT and the financial sector (UNCTAD, 2022: page 3). For example, in Bangladesh, the lion's share of inward FDI has been in the services sector over the past decade, notably in communications and financial services. FDI in services represented more than 60 percent of total FDI over the past decade (e.g., Chanda and Raihan, 2016).
UNCTAD (2022) has also noted that during the period from 2011 to 2021 (i.e., after the adoption of the Waiver Decision by WTO Trade Ministers), LDCs adopted investment promotion and facilitation measures that represented about one third of all measures. For example, the introduction of investment incentives measures represented 48 per cent of all facilitation measures adopted by African LDCs, and 44 per cent of measures implemented by Asian LDCs were devoted to the establishment of one-stop shops and other measures to facilitate business permits. At the sectoral level, measures adopted by African LDCs in favour of the services sector represented 30 per cent of all measures, just behind policies to promote investment in the extractive sectors (39 per cent of the total measures). Measures adopted by Asian LDCs focused essentially on the promotion of FDI flows to the manufacturing sector, and accounted for 44 per cent of the sectoral measures in the region. Thus, even though FDI flows in the goods sector (extractive industries for African LDCs and manufacturing sector for Asian LDCs) tend to dominate FDI flows in the services sector, the number of measures implemented by LDC governments (in particular after 2011) to attract foreign investors in the services sector have been increasing, and FDI has been increasing flowing to the services sector of LDC economies.
In light of the above-mentioned positive effect goods exports on services exports, one could expect that even measures adopted by LDC governments to attract FDI in extractive and manufacturing sectors can exert positive spillovers in the services sector, and ultimately contribute to promoting services exports. However, we could not determine a priori whether these spillovers effects would generate higher modern services exports relatively to traditional services exports, although compared to exports of primary commodities, the expansion of manufactured exports may generate higher exports of modern services than of traditional services. Gnangnon and Priyadarshi (2016) have found that the diversification of export products promotes services exports. Multinational enterprises (MNEs) that wish to take advantage of the Waiver may locate either in the LDCs' services sectors of comparative advantage (e.g., traditional services sectors for many LDCs), and/or in other services sectors, including modern services sectors, where the LDCs have a future competitive advantage.
Overall, while we can expect positive effect of FDI inflows on services exports, the extent and direction of these effects on modern services exports and traditional services exports are to be determined empirically. Additionally, we can expect any eventual positive effect of the Waiver on modern and traditional services exports) could potentially be dependent on FDI inflows (hypothesis 4).
[13] The PRGT is the main vehicle used by the IMF to provide concessional loans (currently at zero interest rates) to LICs (as defined by the IMF). See further information online at: https://www.imf.org/en/About/Factsheets/IMF-Support-for-Low-Income-Countries
[14] UNCTAD (2022: page 3) has particularly noted that 39 of the 46 LDCs are considered commodity-dependent.
[15] The number of FDI projects in non-resource-based manufacturing industries increased from 26 per cent at the beginning of the decade to 31 per cent of in 2015-2019.