World Bank trade loans and export performance of recipient countries

Abstract

Since 1980 several developing countries have received World Bank structural loans, aimed at opening their economy to international trade. By estimating a gravity equation on a panel of 180 countries, observed from 1962 to 2010, we investigate whether the Bank’s programs have affected the export performance of beneficiaries in the subsequent years. According to our results, trade loans have been ineffective in the shorter run while, in the longer, they appear to have hindered the export performance of recipient countries. The Bank’s new trade policy approach, however, seems to have some potential for inverting the negative influence that we have detected.

Keywords:

• World Bank
• trade adjustment loans
• export performance of developing countries
• gravity model

JEL Classifications:

• F13
• F35
• C2

Notes

1. We refer to Parida and Sahoo (2007) for a deeper discussion of the export-led growth hypothesis and a review of the related literature.

2. Since most developing countries’ exports go to industrial nations and a relevant part of developing countries’ imports come from advanced nations, reciprocity of trade reform is crucial for developing economies. For instance, less developed countries call for easier access to the richer nations’ agricultural markets, which have been historically protected by tariff and non-tariff barriers (such as high quality and sanitary standards).

3. IEG (2006) highlights that, from 2000 to 2004, almost half of all trade lending was represented by institutions-related trade facilitation.

4. Conversely, according to the “infant industry” argument, trade protection is desirable for developing countries in order to shield new industries, allowing the latter to develop and become competitive. Besides, if a large part of exports is made up of primary commodities, an increase in exports may be associated with a fall in export revenues, as the demand tends to be price-inelastic. In addition, cutting taxes on trade affects the fiscal balance of developing countries, as trade taxes are an important source of government revenues, due to limited administrative capacity to collect other taxes on consumption and income. Furthermore, higher levels of foreign competition imply the decline of inefficient domestic firms, with a consequent increase in the unemployment rate and the social costs of liberalization. Finally, trade liberalization could damage the environment of developing countries if associated with greater exploitation of natural resources and trading of toxic waste to countries with weak environmental laws.

5. For the sake of conciseness, we avoid reviewing empirical studies which, although related to the influence of aid on exportable industries, do not analyze the influence of aid on trade flows. Among the others, it is worth mentioning Rajan and Subramanian (2011), who provide robust evidence that aid negatively affects the competitiveness of the tradable goods sector by appreciating the real exchange rate. We refer to them also for a review of the empirical literature on aid, exchange rates, and manufacturing.

6. Nkusu (2004) draws the same conclusion, indicating what may have prevented Dutch disease effects in the Uganda experience: “Besides the availability of unused production factors (…), structural reforms and prudent macroeconomic management, leading to both an easing of supply bottlenecks and a stronger foreign exchange reserve position, have contributed to curbing the prospects of a resource transfer against the tradables sector” (p. 22).

7. We refer to Easterly (2005) for an historical review of this lending instrument.

8. The basic gravity equation, initially specified without an explicit theoretical foundation, has been provided with different theoretical bases. In this paper, we also estimate an equation consistent with the “gravity with gravitas” model of Anderson and Van Wincoop (2003, 2004).

9. By using a dummy variable for each year after the event of interest, Head, Mayer, and Ries (2010) estimate the impact of independence on trade between former colonizers and colonies; Glick and Rose (2002) analyze the effect on bilateral trade of currency union dissolution; Martin, Mayer, and Thoenig (2008) investigate whether militarized disputes between country pairs have delayed effects on bilateral trade barriers. Jinjarak, Salinas, and Tsikata (2005), by using dummy variables individuating 3-year intervals since trade reforms, assess the influence of trade reforms on macroeconomic indicators.

10. According to IEG (2006) more than half of the considered liberalizations need from four to eleven years to be accomplished. The same evaluation deems as rapidly implemented those reforms completed within three years.

11. Besides the transfer of knowledge, these variables control for the fact that, as Easterly (2005) points out, adjustment may require several loans to be completed. The author suggests that, as a patient may need multiple doses to fully recover, a country may need successive adjustment loans to gradually improve its performance, or it may experience an improvement after a threshold is overcome.

12. See De Benedictis and Taglioni (2011) for a discussion of this model and the empirical strategy to estimate it.

13. When we employ the tetrad method, each of the dummies TAL_1d, TAL_2d, and TAL_3d is defined as a “dyadic” variable, coded 1 if at least one of the trading partners has received World Bank trade assistance (one, two, or three decades ago). When the couple is made up of two developing countries, it could be that only the importer has received this kind of loan; hence, part of the observed increase (decline) in exports over time may be due to greater (lower) openness of the developing countries partners induced by TALs. If the import demand of recipient countries has generally increased after TALs, the positive (negative) export response to TALs may be overestimated (underestimated). The latter effect, though, is expected to be negligible, since most developing countries’ exports are absorbed by developed countries (in our sample, the average export flow towards high-income countries is about five times that corresponding to the other countries). As a matter of fact, as illustrated in the next section, our results are analogous to those obtained when we specify the TAL dummies as monadic effects and estimate the gravity equation by the PPML estimator.

14. For a more detailed discussion of the selection bias problem, see Conway (2003), Easterly (2005) and Agostino (2007, 2008). See also Baier and Bergstrand (2007) for a discussion of the endogeneity bias potentially implicit in some gravity variables. To address the problem, the latter authors adopt panel estimators (fixed effects, random effects, and first differencing). In a gravity framework as well, Head, Mayer, and Ries (2010) acknowledge that unobserved factors (specific to the colonizer and to the colony countries and to their bilateral relationship) may affect not only bilateral trade between colonies and metropoles, but also the probability of independence; hence the problem of endogeneity bias arises. Analogously to the present study, they eliminate these specific effects by removing time-varying country effects and non-time-varying bilateral effects.

15. Besides TAL_1d, TAL_2d, and TAL_3d (see note 13), the other dyadic variables in Equation (1) are RTA, GATT, and CURR.

16. From 1 January 2008 a temporary unilateral scheme (denominated Market Access Regulation, signed by 35 countries) has replaced the Cotonou regime, which will be definitely superseded by Economic Partnership Agreements.

17. In our Poisson estimations, the error term is decomposed to distinguish an idiosyncratic part from exporter and importer fixed effects. The latter factors, initially suggested by Matyas (1997) to control for unobserved country heterogeneity, make cross-sectional gravity equations consistent to Anderson and Wincoop’s (2003) theoretical model, as country fixed effects serve as proxies for multilateral resistance terms. Besides, the decomposition that we adopt represents a special case of a general specification, featuring non-symmetric country-pair fixed effects, advocated by Cheng and Wall (2005). Adopting a general specification in our case is unfeasible, as it would imply considering more than 21,000 dummies (analogously, as already highlighted, it is unfeasible to consider exporter-year and importer-year fixed effects). However, to account for country-pair specific shocks, implying a potential correlation in the error terms of the same couple of trading partners, we cluster observations at the country-pair level.

18. In other words, the ratio between the (average) exports observed between 10 and 20 years after the program start and the (average) exports observed before the first decade is about 87%. Indeed, as the coefficient of the dummy TAL_2d is −0.1403, its anti-log is 0.8693.

19. We apply a Fisher-type test (Maddala and Wu 1999) to the per capita GDP and population series, as the relative panels are not balanced, while we apply the Harris-Tzavalis (1999) test to the exports variable, for which panels are strongly balanced. As advocated by Fidrmuc (2009), both tests are corrected to mitigate cross-sectional dependence.

20. According to a World Bank historical classification (data.worldbank.org/about/countryclassifications/a-short-history), in our data-set these countries are: Aruba, Bahrain, Barbados, Croatia, Cyprus, Czech Republic, Equatorial Guinea, Estonia, Gibraltar, Greece, Hungary, Republic of Korea, Macao, Malta, New Caledonia, Northern Mariana Islands, Oman, Poland, Portugal, Saudi Arabia, Slovak Republic, Slovenia, and Trinidad and Tobago.

21. The first two intervals parameters are statistically significant 20 over 72 times, while the other intervals parameters are statistically significant 64 out of 144 times. The last lustrum coefficient tends to be negative and significant less often than the others; further, it is positive and significant at 10% level in four cases (GBR_PAK, GBR_MYS, ITA_PAK, and ITA_BRA) giving support to the PPML estimates. Furthermore, when considering five-year intervals, we carry out the same sensitivity checks that we have performed for the decades case, namely we add preferential dummies and we extend the estimation sample to those exporters that were not high-income countries before 2010.

22. As far as the tetrad output is concerned, we obtain estimates relative to the first two reference countries we consider (USA and China) and, further, we retrieve the average estimated coefficients of 36 estimations (as already mentioned, six different reference importers to six reference exporters are matched). In the case of the USA and China, there appears a significant and negative difference between prior and post-programs trade for the 8–28 year range (excluding only year 16). In the average case, there is no significant difference between prior and post trade, suggesting that if trade doesn’t tend to fall relative to “normal” trade, neither does it tend to increase in the aftermath of TALs. Averaging estimated parameters, however, hides a pattern which emerges when looking more closely at the estimation results. First of all, the negative sign of the dummies of interest is preponderant across all 36 estimations. Further, in 13 cases the TALs dummies are negative and significant for more than half of the considered period (reaching two-thirds of the span in 11 cases). In six other cases, the TALs are negative and significant for at least 4 years. The very few positive and significant coefficients are concentrated when Pakistan or Brazil is the reference importer. The output of these further robustness checks is available upon request.

23. For this investigation, we cannot employ the tetrad method, as it automatically eliminates the dummy individuating the time period 1995–2010.