Neighbouring countries and bilateral remittances: a global study

ABSTRACT

We measure to what extent neighbouring countries affect the amount of remittances between a source and a recipient country, controlling for the commonly used macro determinants of remittances. We provide novel evidence on the importance of neighbouring countries, with the parameter estimates capturing origin and destination spatial dependence being positive and significant. Disregarding the role of neighbouring countries leads to biased estimates and misprediction. Indeed, when we correctly account for the role of neighbouring countries, prediction errors decrease by 44% when we express bilateral remittances in nominal terms and by 31% when remittances are in logarithm. Next, we present evidence supporting the altruism, the investment and the financial friction motives to remit, with the altruism motive being the one that contributes the most to explain expected remittances. As an application of our model, we show that, following the Covid-19 shock, remittances are expected to decrease less in countries with smaller income per capita. This is good news for low- to middle-income countries.

KEYWORDS:

• bilateral remittances
• migrants
• network effects

JEL:

• F24
• F27
• R23

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1 Some examples of the political linkages between the two regions include: the strategic partnership formed since 1999 and the international cooperation in a number of issues, including macro-economic and financial matters; environment, climate change and energy; and science, research and technology. In economic terms, the European Union is the leading investor in the region and the second-largest trading partner after the United States.

2 According to the altruism hypothesis, migrants may send more remittances home when their home country’s income declines, so as to compensate for the lost income of family members owing to economic downturn back home, and when the relative income of the country where migrants work increases. In turn, the investment motive states that remittances are used to seize investment opportunities at the migrant’s home; therefore, when the home income of the migrant grows, remittances (to the migrant’s country of birth) should increase. The financial friction hypothesis states that financial frictions, such as restrictions on capital mobility, hinder remittances.

3 All country characteristics are lagged two years to mitigate reverse causality.

4 To give an order of magnitude, a 6% expansion in the electricity use per capita of a typical source country is expected to increase the remittance flows originating from that representative location by 6% to 8%, depending on the model specification.

5 To the best of our knowledge, we know of a single paper, Lee and Pace (2005), that considers different lists for origin and destination locations in a spatial gravity model. Their application is in the geo-marketing literature. However, their model includes a single spatial weight matrix. In contrast, in this paper, we demonstrate that a gravity model with spatial dependence can be defined with a more flexible specification.

6 Remittance-senders may wish to inherit to family members living in their countries of birth.

7 Hereafter we omit the subindex $t$ for notational convenience.

8 An alternative data source for the cost of remittances could have been the World Bank’s ‘Remittance Prices Worldwide’, which covers 48 remittance source countries, 105 recipient countries, and which results in 367 country corridors. Since we have 6619 country pairs in this paper, we prefer the Western Union dataset as it enables us to include a larger set of country pairs.

9 This choice is grounded on the extensive literature investigating the relationship between electricity consumption and GDP (Csereklyei et al., 2016; Csereklyei & Stern, 2015; Smulders & De Nooij, 2003; Stern, 2004, 2019).

10 The intuition for the 3000 km threshold is to recognize that, above a certain limit, we could end up having neighbours that make no sense to be so, for example, country pairs in different continents. Despite the latter, the to-be presented model estimates are robust to alternative distance thresholds.

11 In an unreported analysis, we compared our Bayesian MCMC estimates with the maximum likelihood estimates and the results were very similar.

12 Furthermore, results exhibited in Table 2 are robust to: (1) a different definition of proximity that combines nearest neighbours and contiguity; (2) alternative dependent variables, such as the logged ratio of remittances to the population of the source or the recipient country; and (3) a spatial Durbin model specification.

13 A total of 45 countries report migration flow data to the United Nations only.

14 In the case of electricity use per capita, since the variable is in logarithm, the 1% increase in the range of the variable we suppose represents a 6% in the electricity use per capita in level.

15 The to-be presented results are robust to considering other electricity-GDP elasticities (e.g., 0.7 as in Csereklyei et al., 2016).

16 It is important to add that in these exercises we are only computing the predicted effect on remittances, accounting for the impact on economic activity due to the pandemic. Therefore, we are not considering any increase in international aid that countries, in particular LMIC, might have received to mitigate the Covid-19 impact on the population.

Additional information

Funding

Thibault Laurent and Christine Thomas-Agnan gratefully acknowledge funding from the ANR [grant number ANR-17-EURE-0010 – Investissements d’Avenir programme].