Examining the trade potential of the UAE using a gravity model and a Poisson pseudo maximum likelihood estimator

Abstract

When the United Arab Emirates (UAE) started exporting oil in 1962, it leveraged off the country's economy to emerge as one of the world's fastest-growing nations. Recently, however, worries about the vulnerability of the economy from shocks in international oil prices have effectuated a rapid transformation based on production and trade diversification. Trade agreements and liberalization policies turned the country into a regional trade hub. Given the recent changes in the economy's structure and the importance of trade for the UAE economy, we examine opportunities for the country to further expand trade. We use a gravity equation on 2002–2016 panel data and a Poisson pseudo maximum likelihood estimator to examine the determinants of trade and the trade potential. We compare the results obtained from a theory-consistent econometric approach for the estimation of gravity with the results obtained from past ‘intuitive’ approaches to gravity estimation and infer on the bias present. Results reveal that the UAE has exhausted trade potential with some of its major trading partners including many member countries of the GCC and PAFTA. However, potential for trade expansion exists with many other countries, including Japan and India that could dictate future policy efforts for trade expansion.

KEYWORDS:

• Gravity model
• trade potential
• United Arab Emirates
• pseudo Poisson maximum likelihood
• PPML

JEL CLASSIFICATIONS:

• F14
• F10
• C23

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 We further discuss SFA in the footnotes of Section 3 and defer this topic to the extensions and future research proposed in the conclusions.

2 PAFTA members include Bahrain, Egypt, Iraq, Jordan, Kuwait, Lebanon, Libyan Arab Jamahiriya, Morocco, Oman, Qatar, Saudi Arabia, Sudan, Syrian Arab Republic, Tunisia, UAE and Yemen.

3 The conventional two-step method for estimating the trade potential, which we employ in this research, uses the centered values of a dataset for (gravity) predicted trade which differs from the maximum possible trade that a country can achieve when there are no frictions (Kumar and Prabhakar 2017) as required by the definition of trade potential. Kalirajan (2007) notes that to estimate trade potential, an approach is required that can estimate the upper limit of the data rather than mean values of the dataset. He proposes the use of SFA that defines trade potential as the maximum level of trade. SFA employs a two-part error term: a conventional error and a one-sided error that represents losses due to trade barriers (Kumar and Prabhakar 2017). These represent the trade inefficiency. Adding to the contribution of this article, the use of PPML estimator has not been employed using the conventional methodology. A comparison of our results with past research allows us to infer on the validity of conclusions reached with the help of ‘intuitive’ gravity models that do not employ the full structure of the gravity model presented by Anderson and Wincoop (2003). Further extensions of this research are discussed in the conclusions of this article.

4 However, they are also more likely to reflect cases where a country is not suitable for the production of a good, in which case an observation can be dropped (see WTO Practical Guide 2012a).

5 This is due to Jensen’s inequality which shows that the expected value of the logarithm of a random variable is not equal to the logarithm of the expected value of the same random variable.

6 For a more detailed discussion, see Head and Mayer (2014) and Feenstra (2016). For a summary of the advantages and disadvantages of each method, see Gomez-Herrera (2013).

7 Thirty-two exporting countries were selected mainly from the MENASA region or main destinations from the UAE. These are listed in Table A1 in the Appendix. During the estimation of a structural gravity model, we could have used the UAE as our only exporter combined with all of the countries in the world where export data were available as importers. In other words, we could adjust the sample to a traditional (and more intuitive) panel, where we have only one exporter (the UAE) and numerous importers over time. However, this would not allow us to (fully) take into account the multilateral resistances, hence the model would not be theoretically sound. By restricting the sample to the bilateral exports and imports of the UAE, we would miss out on possible trade diversion effects between third countries that would not be taken into account. And this is exactly the point that Anderson and Wincoop (2003) made when they introduced multilateral resistances. The two-way trade relationship between trading partners is not what determines trade flows. Rather, bilateral trade depends on multilateral resistances which means that there is a dependence on all other trading partners of both the exporting and the importing countries. Multilateral resistances define the choices that countries make based on opportunity costs. The main idea is that every sale has multiple possible destinations similar to every purchase which has multiple possible origins. Therefore, any bilateral sale involves all bilateral frictions. Changes in trade cost in one of the bilateral routes affect trade flows on every other route as there are relative price effects. In trade, general equilibrium forces are important. We therefore need to estimate the gravity model with multiple exporters even if we are interested in discussing the results for only one of them. If we were interested in a different country, we would hence adjust the list of exporters to account for more relevant trade routes for the country in question.

8 Data for GDP and exports for the econometric estimations were employed in nominal form. However, in Table 1, and in order to allow intertemporal comparisons, data were converted to constant US dollars using the country $i$ GDP deflator, also downloaded from the World Bank database.

9 While we use direct export trade values, we must note that the literature has employed in the past mirrored flows of exports reported by the trading partners when large differences between exports reported by a country to the UN and imports reported by partner countries are present (see Proenca, Martinez-Galan, and Fontoura 2017; Yotov et al. 2016).

10 Past studies have found that currency unions, one of the deepest forms of integration, result in doubling/tripling of bilateral trade between countries (Rose 2000). Rose and Stanley (2005) find smaller increases in the range of 30–90%. On the other hand, Baldwin and Taglioni (2006) employing both multilateral resistances and unobserved bilateral heterogeneity, find no effect from a currency union to the level of bilateral trade flows.

11 Augmenting variables that have been employed in the past, such as the openness of the importer and exporter countries, population or exchange rates were not added to the specification. De Benedictis and Taglioni (2011) suggest that appropriate measures of attractors should be selected on the basis of theoretical considerations. Similarly, Shepherd (2008, 2013) discuss the inclusion of variables in model specification and further state that ‘expedients sometimes encountered in the older gravity literature, such as using population and per capita GDP as separate regressors, should therefore be avoided’. If other variables, such as population or country openness, are to be included, we would first have to consider the theoretical implications.

12 Similarly, if population, exchange rates, openness of the country, etc. were included, these would have been absorbed by the time-varying FEs.

13 Following Naryan and Nguyen (2016), we examined the model for multicollinearity by excluding one income variable at a time and also including a variable that consists of the product of the GDP from importers and exporters. Results were sensitive to the inclusion of the interaction term with the coefficients on distance and FTA agreements consistent throughout the model.

14 Similar differences were obtained when ET/IT FEs were included in the OLS model.

15 PPML (4) and GPML (8) estimated both with regular FE as well as PPML (6) and GPML (9) estimated with ET/IT FEs carry strikingly similar coefficients and also pass the RESET test.

16 GCC-EFTA countries are not included in the FTA dataset obtained from the World Bank.

17 Converge could not be achieved when ET/IT FEs were included, so results from the Heckman model include regular FEs.