Does Aid Drive Migration? Evidence from a Shift-Share Instrument ∗

Abstract

Foreign aid payments have been a key policy response by Global North countries to reduce increased migration flows from the Global South. In this article, we contribute to the literature on the relationship between aid and international migration flows and estimate the contemporaneous effect of bilateral aid payments on bilateral, international migration flows. The fundamental problem in analyzing this relationship is endogeneity, or reverse causality. To address this issue and achieve causal inference, we use a shift-share, or Bartik, instrument. Examining migration flows between 198 origin countries and 16 OECD destination countries over 36 years (1980−2015), we find a positive relationship between aid and migration. A ten-percent increase in aid payments will increase migration by roughly 2 percent. We further document non-linearity in the relationship between aid and migration and find an inverted U-shaped relationship between aid and migration flows. The findings presented here have implications for the design of bilateral and multilateral aid policies and for achieving various United Nations Sustainable Development Goals by stressing the importance of a better coordination between aid and immigration policies.

Keywords

Aid Bartik Migration

Introduction

Large flows of migrants from the Global South to the Global North have led to political conflict and polarization in the societies of the European Union (EU) and the United States (Liu et al., 2019; Stockemer et al., 2020). The European refugee crisis and the US immigration debate under the former Trump administration have resuscitated policy-makers’ interest in using aid and development programs to manage migration flows (Washington Post, 2017; Reuters, 2018). In 2017, for instance, the UK’s Department for International Development announced aid payments to Sudan, targeted at reducing migration to Europe and pledged payments of £121 million from 2017 to 2022.¹ Around the same time, Germany introduced a “Marshall Plan for Africa,” with the goal of “supporting migration management and governance including the protection of refugees and migrants’ rights.”² The plan included payments of up to 9 billion Euro from EU funds.³

Against this background, understanding the relationship between aid payments and migration flows is important to inform policy decisions and the design of aid projects themselves. In this article, we examine the relationship between bilateral foreign aid and bilateral, international migration flows. In doing so, we address, first, whether aid increases or decreases migration flows and, second, whether aid's effect on migration is linear or whether a turning point exists. More broadly, we contribute to the literature on the aid-migration relationship in two ways.⁴ First, we overcome the fundamental problem in this scholarship - endogeneity or, more precisely, reverse causality, which is often addressed by jointly estimating equations for aid and migration (e.g., Berthélemy et al., 2009; Lanati and Thiele, 2018). Because this standard approach suffers from uncertainty about the “correct” specification of the equation system, we use a shift-share, or Bartik, instrument which allows us to make a causal statement about the relationship between foreign aid and migration. Although Bartik instruments have been applied to the study of immigration's effects on labor-market outcomes (e.g., Card, 2001, Wozniak and Murray, 2012, Jaeger et al., 2018), to food aid's effects on conflicts (e.g., Nunn and Qian, 2014), and to the macroeconomic effects of government spending (e.g., Nakamura and Steinsson, 2014; Temple and Van de Sijpe, 2017), to the best of our knowledge, we are the first to use a Bartik instrument to study aid's effect on international migration. Second, we extend this literature on aid and migration by considering non-linearity in the aid-migration relationship and documenting a non-linear relationship between aid payments and migration flows. Usually, this literature ignores such non-linearity and assumes that the relationship between aid and migration is linear (e.g., Lanati and Thiele, 2018).⁵ As our analysis shows, however, this assumption is not supported by the data. Given our findings, the UK's Sudan spending program previously mentioned would have to be roughly four times larger to achieve its goal.

We begin this analysis by constructing a panel data set of international migration flows between 198 origin countries and 16 OECD destination countries over 36 years (1980-2015) and data on bilateral Official Development Assistance (henceforth, ODA). We, then, construct our Bartik instrument for aid, similar to the one used by Goldsmith-Pinkham et al. (2018). We interact the share of aid received by the origin country at time t with the growth rate of total aid spent by the destination country. Exogeneity of this instrument follows because the average growth rate across all donor countries will not exhibit correlation with the random term, while each country-pair specific growth rate of aid might be endogenous (Bartik, 1991). The standard Bartik approach relies on the identifying assumption that the initial aid shares are exogenous. Further, the role of the growth rates is mainly about the instrument's power, in contrast to achieving identification.

Using this instrument, we obtain two key results. First, we find a robust and positive relationship between aid and migration flows. These results support the finding by Azam and Berlinschi (2009), Berthélemy et al. (2009), and Belloc (2011) but contradict recent findings by Lanati and Thiele (2018), who report a negative relationship between aid and migration. Berthélemy et al. (2009) and Lanati and Thiele (2018), key contributors in this literature, use a system equation approach (three-stage least squares or 3SLS), simultaneously estimating two equations for aid and migration to deal with endogeneity. The key difference between our estimation strategy and theirs is that our results do not depend on the model specification.⁶ We employ a simple instrumental variable regression, using a strong, exogenous instrument (Bartik IV) to estimate aid's direct (or total) effect on migration. Second, we find evidence of non-linearities in the aid-migration relationship and an inverted U-shaped relationship between aid and migration. In our preferred specification, the turning point in the aid-migration relationship is calculated to be at $411 million in aid payments (per year). At this turning point, aid's effect on migration flows turns from positive to negative, and a further increase in aid payments reduces migration flows between countries. In our sample, only about 2 percent of aid payments exceed this threshold.

To develop these findings, the article is structured as follows. Section 2 discusses our empirical strategy and how we constructed our Bartik instrument. Section 3 presents the data set and descriptive statistics about the aid-migration relationship. Section 4 offers our main results, while Section 5 lays out aid's non-linear effects on migration. Section 6 concludes that more foreign aid will increase migration flows toward the country sending the aid but that this relationship is not linear and, instead, follows an inverted-U shape with a turning point around $400 million. These findings, we suggest, have implications for governments, non-governmental organizations (NGOs), and the wider research community and can inform the design of aid payments intended to manage migration flows by highlighting the role of non-linearity in the aid-migration relation. In our conclusion, we also identify various research questions for future work in this area.

Empirical Strategy

As a first step in our analysis, we derive an estimable equation from the Borjas (1987) model, which explains migration flows across countries.⁷ Using this model results in an equation that resembles an augmented gravity equation, where, for example, rich destination countries receive more migrants. The fixed-effects regression model used can be written as

\ln (M_{i j, t}) = α + β \ln (Ai d_{i j, t}) + γ X_{i j, t} + ν_{t} + ϕ_{i j} + ε_{i j, t},

(1)

where

M_{i j, t}

is migrant flow between origin country i and destination country j at time t,

α

is a constant,

ν_{t}

is the time fixed effect, and

ϕ_{i j}

is the country pair fixed effect. This specification of fixed effects follows the approach by Cai et al. (2016) and addresses the multilateral resistance problem. Multilateral resistance describes the idea that the decision to migrate to a country i depends not only on this country's attractiveness but also on other countries j's attractiveness and, thus, the attractiveness of country i relative to other potential destination countries.

There are various approaches in the literature to deal with the multilateral resistance problem (e.g., Baier and Bergstrand, 2009), which we address in the robustness section. Further, $β$ is the parameter of interest determining aid's effect on migration. Then, $X_{i j, t}$ is a matrix of control variables that includes economic, political, and disaster controls, $γ$ is an unknown parameter column vector, and $ε_{i j, t}$ is an independent and identically distributed (i.i.d.) error term. This model includes fixed effects that account for any origin time-invariant and country-pair specific determinants of migration.

The major problem in empirically modelling the aid-migration relationship is the potential endogeneity of migrant flows and aid flows, due to reverse causality (Wooldridge, 2010). The issue with reverse causality arises from the difficulty of determining the aid-migration relationship's directionality. For instance, we cannot determine whether increasing aid payments affects migration or whether the opposite holds. It is also possible that there is simultaneity in the aid-migration relationship, in that aid influences migration and concurrently migration flows impact aid flows. The conventional approach to addressing simultaneity uses instrumental variable estimation (e.g., Wooldridge, 2010). We argue that the instruments used in the literature studying the relation between aid and migration (e.g., lagged aid by Gamso and Yuldashev 2018, public expenditures on order and security by Azam and Berlinschi 2009, or population size by Rajan and Subramanian 2008) are not exogenous and, therefore, do not allow us to make statements about the direction of causality. In contrast to the existing literature on the aid-migration relationship, then, we employ a Bartik (or shift-share) approach.

Bartik (1991) first presented this method to deal with the issue of endogeneity in estimating the relationship between local wage growth and local employment growth. The Bartik instrument is typically constructed as the inner product of the endogenous regressor by interacting industry shares and the industry component of the growth rates (Bartik, 1991). Our instrument is similar to the one used by Goldsmith-Pinkham et al. (2018) and relates to Temple and Van de Sijpe (2017), who analyzed aid's effect on macroeconomic variables, such as consumption and investment. To do so, Temple and Van de Sijpe (2017) and Goldsmith-Pinkham et al. (2018) construct an instrument that interacts total donor budgets by the initial shares of recipients in those budgets, which they label as the supply-push instrument, and argue that this instrument is endogenous.

To estimate aid's effect on migration, we use country-level shares and recipient-specific aid growth rates. Our instrument is constructed according to

\begin{matrix} \ln (Ai d_{i j, t}) = Z_{i j, t} G_{i t} = z_{i j, t} \sum_{j = 1}^{J} g_{i j, t}, \end{matrix}

(2)

where

Z_{i j, t}

is a

(1 \times J)

column vector of recipient-time period ODA shares and

G_{i t}

is a

(J \times 1)

row vector of recipient-time period ODA growth rates.

The first important factor in our analysis is to understand the drivers of aid flows and the drivers of the initial aid share. Alesina and Dollar (2000) find that aid flows are driven by various factors, including previous colonial ties, the receiving country's UN voting behavior, and income. They argue that aid flows are strategically distributed and only show a weak relationship with poverty or democracy (see also Nunnenkamp and Thiele, 2006). Berthélemy (2006) shows that aid flows follow trade patterns, finding that some countries’ aid donations are driven by altruism (e.g., Switzerland, Austria, and most Nordic countries) while others are driven by egoism (e.g., Australia, France, and Italy). Alesina and Weder (2002) show that corruption has no effect on the distribution of aid.

The second important consideration is the Bartik instrument's exogeneity. The standard Bartik approach relies on the identifying assumption that the industry shares are exogenous (Bartik, 1991). From this perspective, the role of the growth rates is predominantly about the instrument's power and less about identification. Our instrument is computed as the interaction of the share of bilateral aid received by the origin country at time t with the growth rate of total aid spent by the destination country. Exogeneity follows from the argument that while each country-pairing specific growth rate of aid will be endogenous, as it is correlated with $ε_{i j, t}$ in equation (1), the average growth rate across all donor countries will not exhibit correlation with $ε_{i j, t}$ .⁸

Goldsmith-Pinkham et al. (2016) argue that updating the industry shares each period, as is the case in the standard Bartik approach presented in equation (2), can result in biased estimates, should the changes in industry shares be partially driven by endogenous shocks. Therefore, they propose that researchers use the earliest possible version of their available shares to avoid these potentially biased estimates. Goldsmith-Pinkham et al. (2016) conclude that $z_{i j, t}$ should be selected for the earliest available period $t,$ essentially codifying and reiterating existing estimation practices in the use of Bartik instruments (e.g., Card, 2001; Nunn and Qian, 2014; Temple and Van de Sijpe, 2017).

In our dataset, 150 countries received aid in at least one period, although not all these countries received aid disbursements in every time period. Of these 150 countries, 124 countries received aid in 1980. Therefore, selection of the initial period t that determines the aid share component $z_{i j, t}$ of the Bartik instrument is not straightforward. We constructed our Bartik instrument, using the share of aid these 124 aid recipients received in 1980. Consequently, this process reduces our sample by 36 countries that received no aid in 1980. Notice that when we run a Poisson Pseudo-Maximum Likelihood (PPML) regression (due to observations of zeros in our data), trying to explain aid as a function of migration and its first two lags, we do not find significant effects of migration on aid.

Data

In this article, we combine data on migration flows with bilateral aid payments (flows), constructing a panel data set of migrant flows between 198 origin countries and 16 destination countries from 1980 to 2015 (36 year observations, 3,168 cross-sectional observations for each year, a total of 114,048 observations). For aid, we observe the opposite direction: from the 16 destination countries toward the 198 origin countries. The selection of migrant destination and origin countries, along with the selected timeframe, is due to data availability.

Data on migration are taken from the data set developed in Aburn and Wesselbaum (2019), which merges annual net migration flows from the 2015 Revision of the United Nations’ Population Division with data provided by the OECD.⁹ This data set, as is usual in the literature studying international migration flows (e.g., Beine and Parsons, 2015; Cai et al., 2016), does not capture unauthorized immigration, leading to a likely underestimation of true migrant flows. The 16 selected migrant destination countries are all OECD, highly developed countries and include seven of the 20 destinations with the largest number of international migrants in 2015 (Australia, Canada, Germany, Italy, Spain, the United Kingdom, and the United States). All destination countries are members of the OECD Development Assistance Committee (OECD-DAC), meaning that they share similar economic development goals and act as aid donors.

Our measure of aid is gross ODA disbursements (measured in millions of 2016 US Dollars), as used by Berthélemy et al. (2009), Temple and Van de Sijpe (2017), and Lanati and Thiele (2018). Aid disbursements are sourced from the OECD-DAC database.¹⁰ We use ODA disbursements, rather than ODA commitments, because we argue that injections of ODA funds into a local economy should have a larger impact on the migration decision than the commitment of funds. ODA commitments will still be likely to have an effect by fostering stronger international ties between a donor and recipient; however, these effects will be largely captured by the ODA disbursements as well. Further, ODA disbursements may provide a more accurate representation of the resources that reach the recipient countries’ population.

Table 1 presents the top five countries sending and receiving aid in our sample over the entire time period. Egypt, Mexico, and Iraq are the top three countries receiving ODA. Turkey and India follow with some distance. Countries like Afghanistan (47,620) or Bangladesh (34,303) do not make the top five list. The largest donor countries are the United States, Germany, and the United Kingdom. Interestingly, the United States spent almost twice as much as second-placed Germany, which spent more than twice as much as the United Kingdom and Italy. The Netherlands ranked fifth, with a sizably lower spending of only about one-eighth of what the United States spent.

Table 1.

Top five aid receiving countries and top five aid sending countries (full sample), aid measured in 2016 USD (millions).

Country	Aid Received	Country	Aid Send
Egypt	130,688.4	United States	824.990.3
Mexico	116,832.2	Germany	463.129.3
Iraq	108,000	United Kingdom	215,101
Turkey	92,429.3	Italy	208,669.5
India	92,134.1	Netherlands	109,310.2

While this article focuses on the link between aid and migration, we also want to understand the driving forces of aid flows between countries. To do so, we ran a simple regression with log aid flows as the dependent variable (results shown in the Online Appendix, Table A4). We find that distance had a significant, negative effect on aid payments and that sharing a border had a positive effect. These findings imply that countries send more aid to neighboring countries (see also Aburn and Wesselbaum, 2019). This finding could be motivated by a desire for regional stability, as well as easier implementation of aid projects. Further, the proxy for cultural closeness (language) has a positive effect, while colonial relationships (after WWII) increased aid flows as well. Destination countries at war with an origin country sent more aid. Further, more aid flowed to countries that suffered from weather- and non-weather-related disasters.

In the following, we are mainly interested in aid's total (or direct) effect on migration and include several control variables to investigate the channels through which aid might work. We use a wide selection of standard economic control variables in the aid-migration literature (e.g., Beine and Parsons, 2015; Cai et al., 2016; Cattaneo and Peri, 2016), taken from the World Bank's database (the Online Appendix provides details), including income (GDP per capita), trade measures (export and import shares), and data on total population.¹¹ Further, we control for policy variables, such as a measure for conflicts and wars between country pairs and a measure for political framework, drawn from different sources. The political framework measure, Polity2 variable, comes from the Centre for Systematic Peace and ranges between 10 for a highly democratic society and -10 for a highly autocratic country.¹² The criterion for conflicts requires that the conflict kills at least 1,000 people annually.

Finally, we control for the number of weather (storms, floods, extreme temperature events, droughts) and non-weather (earthquakes, epidemics, landslides, wildfires, volcanic events) disasters. Disaster data come from the Centre for Research and Epidemiology of Disasters EM-DAT database.¹³ Studies also sometimes include the stock of migrants in a year to proxy for network (or diaspora) effects (e.g., Grogger and Hanson, 2011; Beine and Parsons, 2015). However, we do not have data on the stock of migrants in a country for our set of countries and years. In the following, we provide a first look at the relationship between migration and aid in our data set. Figure 1 plots total migration flows into our 16 destination countries alongside total ODA disbursements from the same 16 destination countries.¹⁴

Figure 1.

Relationship between total migration flow (left axis) and total aid flow (right axis).

As Figure 1 shows, over the 36-year sample period, migration tended to be increasing, while aid payments appeared more volatile and did not follow a distinct upward trend. The spike in aid in 2005 is due to exceptionally high debt relief operations in Iraq and Nigeria from the Paris Club, of which all donor countries studied here, other than New Zealand, are members.¹⁵ Total aid payments decreased toward the end of the sample period, which we can attribute to the Global Financial Crisis (beginning late 2007) and the European Sovereign Debt Crisis (beginning in 2009), both of which had significant impacts on donor countries’ economies (Chen et al., 2019). A further notable point in Figure 1 is the spike in both aid and migration in the late 1980s and early 1990s, which could be attributed to the Soviet Union's collapse, along with unrest in places like Sub-Saharan Africa and the Middle East.

Table 2 presents descriptive statistics for all variables in our data set on our Bartik IV sample. We lose observations due to limitations in the availability of aid data. Destination-country variables are denoted (d), and origin-country variables are denoted (o). Migration and aid are written in logarithmic scale.¹⁶ The mean aid disbursement was $3.19 million (US); however, there is large variation in the size of the aid disbursements, with the largest being $13,599,190,000 US from the United States to Iraq in 2005.

Table 2.

Descriptive Statistics.

Variables	Obs.	Mean	Std. Dev.	Min.	Max.
Migration (flow)	28,317	4.75	2.34	0	13.76
Aid (flow)	28,317	1.89	2.44	-4.61	9.52
GDP (d)	28,317	10.67	0.28	10.02	11.43
GDP (o)	26,654	7.50	1.16	4.75	11.65
Population (d)	28,317	16.84	1.32	14.95	19.59
Population (0)	28,313	16.28	1.78	8.92	20.99
Policy (d)	27,565	9.98	0.21	8	10
Policy (o)	26,459	1.27	6.26	-10	10
Export Share (o)	26,007	31.51	21.94	0.005	216.42
Import Share (o)	26,007	38.55	23.34	0.02	424.82
Export Share (d)	28,317	35.65	16.94	6.98	83.43
Import Share (d)	28,317	32.95	14.20	9.03	81.99
War (o)	28,317	0.14	0.35	0	1
War (do)	28,317	0.008	0.09	0	1
Weather Disaster (o)	28,315	1.60	2.55	0	27
Non-Weather Disaster (o)	28,315	0.79	1.43	0	15
Weather Disaster (d)	28,317	3.35	6.05	0	33
Non-Weather Disaster (d)	28,317	0.59	1.28	0	10

Main Results

Table 3 reports the direct effect of aid disbursements on migration flows for the different estimation strategies. Models (1) to (4) present aid's direct (or total) effect on migration, while models (5) and (6) include control variables and present aid's indirect effect on migration. For the instrumental variable models, the first and second stage test results are presented in the Online Appendix, Table A2. All test results point toward a reasonable instrument, passing various statistical tests. The estimated parameters for the controls are shown in the Online Appendix, Table A3. The values of these control variables are in line with the existing literature (see Beine and Parsons, 2015; Aburn and Wesselbaum, 2019).

Table 3.

Main Regression Results.

	(1)	(1a)	(2)	(3)	(4)	(5)	(6)
Variables	OLS	OLS	PPML	IV-2SLS	Bartik	OLS	Bartik
log(Aid)	0.06***	0.02***	0.08***	0.113***	0.188*	0.048***	0.113
	(0.006)	(0.005)	(0.03)	(0.012)	(0.108)	(0.006)	(0.095)
Fixed Effects
Year	Yes	No	Yes	Yes	Yes	Yes	Yes
Couple	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Origin*Year	No	Yes	No	No	No	No	No
Destination*Year	No	Yes	No	No	No	No	No
Controls	No	No	No	No	No	Yes	Yes
Observations	40,370	40,370	43,934	37,229	28,284	33,350	23,308
R-squared	0.898	0.93	0.64	0.254	0.270	0.90	0.30

Standard errors clustered at the country-pair level and are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is log migration flows (except for the PPML estimation, where the level is used). Controls include: war, political freedom, weather and non-weather related disasters, GDP per capita, population, exports and imports and are presented in Table A3 in the appendix.

We begin by discussing our different estimation approaches: Column (1) presents a standard OLS estimation, ignoring endogeneity concerns. Similarly, column (2) uses a different estimator (Poisson Pseudo Maximum Likelihood) that explicitly accounts for zero observations, as suggested by Santos Silva and Tenreyro (2006), but also ignores potential endogeneity. Column (3) uses lagged aid as an instrument, as in, for example, Gamso and Yuldashev (2018), and is the first approach accounting for endogeneity by using an instrumental variable (IV). This IV is most likely not exogenous and will result in biased estimates. Finally, column (4) presents our preferred specification, using the Bartik instrument constructed previously.

We find that across all four approaches, aid had a positive, significant, and direct effect on migration. The effect's size varied across the approaches, from 0.06 (OLS) to 0.19 (Bartik). Our Bartik approaches gives the largest point estimate of aid's effect on migration, showing that the other empirical approaches likely underestimate aid's effect on migration. The difference in the effect size is quite substantial between the (biased) OLS estimate of 0.06 and the Bartik-IV estimate of 0.19. There are various explanations for this difference in the effect size. An omitted variable (e.g., bilateral migration policies or variables describing information flows) might be negatively correlated with migration flows, which would downwardly bias the OLS estimate. A less likely reason for the difference in the effect size could be measurement error in foreign aid, which would not be present in the Bartik IV. Finally, a more technical reason is that the IV computes the local average treatment effect, while OLS computes the average treatment effect across the entire population.

We find that a ten-percent increase in aid payments would increase migration flows by 1.88 percent. This result supports the findings by Berthélemy et al. (2009), who use a larger panel data set, and Belloc (2011), who uses data on flows from Sub-Saharan countries. In contrast, Gamso and Yuldashev (2018) and Lanati and Thiele (2018) find a negative effect of aid on migration. When we consider the effect's magnitude, we find that our effect is sizably larger than the one obtained by Lanati and Thiele (2018), who report an effect around 1 percent, but smaller compared to Berthélemy et al. (2009), who report an effect of roughly 3 percent. Both studies use a 3SLS approach, rather than a Bartik instrument. The main differences between Berthélemy et al. (2009) and Lanati and Thiele (2018) are the use of stocks vs. flows and the use of fixed effects. Importantly, the Bartik approach avoids the issue of specifying equations for aid and migration simultaneously, as needed in the 3SLS approach. For example, it is not clear why income would affect aid, but not migration, in the specification chosen by Lanati and Thiele (2018), or why disasters are not included, as in Berthélemy et al. (2009). This specification problem is not present in the Bartik approach.

Models (5) and (6) include control variables, which reduce the coefficient's size whether we account for endogeneity (Bartik, column 6) or not (OLS, column 5). We lose statistical significance in column (6), due to a sizable reduction in the number of observations. Given this large reduction in the number of observations, we argue that aid's effect on migration is, in fact, positive and significant but imprecisely estimated (and, therefore, insignificant). The reduction in the coefficient's size accounts for the fact that we now estimate aid's partial (or indirect) effect on migration.

Finally, we conducted several robustness checks. The positive relationship we show extends to total aid disbursements received, as shown in Table A5 in the Online Appendix. Further, a different approach to the multilateral resistance problem is to use couple, destination-year, and origin-year fixed effects. Model (1a) uses this specification and finds that aid's effect on migration is still highly significant. Finally, we also apply different Bartik instruments (i.e., Bartik instruments with different initial shares) to address potential endogeneity in the initial aid share. We used the average of aid payments from 1980 to 1990 and from 1985 to 1990 to compute the initial share. In both specifications, we find a positive parameter (0.07 and 0.04, respectively), albeit insignificant, most likely due to fewer observations used.

Non-Linearity

The regressions so far do not account for the possibility of non-linearity in the aid-migration relationship. However, it is plausible that aid's effect on migration varies by the size of aid payments. As argued in Cattaneo and Peri (2016), binding liquidity constraints are a key factor in the migration decision. Along this line, Faini and Venturini (1993) argue that for low-income levels of potential migrants, a small increase in aid payments will lift budgetary constraints and increase migration. We hypothesize that for higher levels of aid payments, aid's effect on migration becomes smaller, as (i) binding liquidity constraints are less of an issue and (ii) the positive effects of the aid payments in the origin country materialize. Hence, we expect an inverted U-shaped relationship with one turning point.

Table 4 tests for aid's potential non-linear effect on migration by including a squared log aid term into the baseline specification presented before (see Table 3).

Table 4.

Non-Linear Effects.

	(7)	(8)	(9)
Variables	OLS	IV-2SLS	Bartik
log(Aid)	0.07***	0.15***	0.361
	(0.006)	(0.013)	(1.794)
log(Aid) - Squared	−0.005***	−0.016***	−0.030
	(0.001)	(0.003)	(0.326)
Turning Point (log Aid)	7.00	4.69	6.02
Observations	40,370	37,229	28,284
R-squared	0.90	0.25

Standard errors clustered at the country-pair level and are reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1. All models include both country pair and year fixed effects. The R-squared terms are pseudo R-squared values. The dependent variable is log migration flows. No controls included.

As before, we find a positive, significant linear effect of aid on migration when using simple OLS (model 7) and lagged aid as an instrument (model 8). We lose significance, however, when using our Bartik instrument (model 9), due to a drop in the number of observations. Again, we argue that the coefficient is non-zero but imprecisely estimated. Further, we find a significant squared term of aid. The consistency across the four estimation methods is reassuring and supports the existence of non-linearities in aid's effects on migration. With the negative squared term, we find the expected inverted U-shaped relationship between aid and migration.

Table 3 also reports the turning point of log aid computed from each regression. The turning point varies considerably across the specifications from 4.7 ($105 million) to 7 ($1,096 million). In our preferred specification, the Bartik instrument (model 9), the turning point is 6.02 ($411 million). Only about 2 percent of aid payments exceed this threshold in our sample, which is a substantive finding, as several aid flows in the dataset exceed this threshold. The most extreme is the $13,599,190,000 US aid flow between the United States and Iraq in 2005, which places downwards pressure on the positive aid coefficient estimates.

Some cautious notes are in order here about the interpretation of the threshold. First, this calculated threshold should not be interpreted as a definitive or precise threshold above which an aid donor will receive reduced migration flows. From our data, we see that countries that receive aid beyond this value are generally in a state of civil unrest or experiencing a major shock to their economy, which will have an impact on the number of migrants from those countries. Therefore, other factors must be considered that can affect aid and migration flows jointly. These factors could be related to macroeconomic trends (e.g. recessions in the origin country), weather- and non-weather-related disasters hitting the origin country (e.g. storms or earthquakes), other types of disasters (e.g. diseases such as Ebola or Zika), and conflict (e.g. war). However, in this type of scenario, it becomes much more difficult to find an exogenous instrument, as these types of events are, most likely, related to one another and related over time. Second, the turning point's estimate also implies that most aid payments fall below this threshold. Even if we used the $105 million turning point, more than 92 percent of aid payments would fall below the turning point. Policy projects like the UK’s Department for International Development (2017), according to this finding, must include substantially larger aid payments to reach their goals.

Figure 2 shows aid's marginal effect on migration for different levels of GDP in the migrant origin country.¹⁷

Figure 2.

Conditional marginal effects taken from OLS regression for aid on migration with 90 percent confidence intervals.

Aid's marginal effect on migration is positive until GDP in the origin country reaches an income of about 7 (roughly $1,100) log GDP per capita, at which point it reverses and becomes negative. Countries below the turning point include, for example, the Central African Republic, Burundi, D.R. Congo, Liberia, and Malawi. Roughly, 75 percent of our origin countries, over time, are above this threshold. The fact that we have few observations below the income threshold explains why the effect at the tails is imprecisely estimated, as the number of observations for very low and very high-income levels is low.

The change in aid's effects on migration from positive to negative as GDP in the origin country increases aligns with the surrounding literature (e.g., Faini and Venturini, 1993), in that it indicates the existence of a peak in migration. This migration peak is also clearly observed in Figure 2, as aid's marginal effect on migration follows a hump shape as GDP in the origin country changes. The turning point for the migration hump is calculated by both Berthélemy et al. (2009) and Lucas (2005) to be close to $7,500 US. Our threshold is much lower, which could be because of a different sample of countries (migration flows toward 16 OECD countries) or because of the econometric methodology (we consider the direct effect without control variables).

We also want to acknowledge that due to data limitations, our analysis is based upon the underlying assumption that aid's marginal effectiveness on income is constant. We do not make this assumption because we think that the marginal effectiveness is in fact constant, but because we do not have disaggregated data to properly address this issue. Further, we do not have details about the aid projects themselves. For example, if Italy and the United States jointly financed a project in Iraq, we would have to treat these aid payments differently. We leave investigating the aid-migration relationship at a finer level of data aggregation to future research.

Conclusion

This article extends the literature on the aid-migration relationship along two dimensions. First, we employ a Bartik (or shift-share) instrumental variable approach to achieve causal inference. While Bartik instruments have been applied to various econometric problems (c.f., Card, 2001, Nakamura and Steinsson, 2014, Nunn and Qian, 2014), we offer a novel application of this approach to the aid-migration relationship. The advantage of using a Bartik approach over existing approaches (e.g., Berthélemy et al., 2009, Lanati and Thiele, 2018) is that it does not rely on the correct specification of the model. In addition, the instrument's exogeneity is ensured because the average growth rate across all donor countries will not exhibit correlation with the shock, despite the fact that each country-pair specific growth rate of aid might be endogenous. Using our Bartik instrument, we find a robust, positive relationship between aid and migration flows, supporting the findings by Berthélemy et al. (2009) and Belloc (2011) and contradicting the recent finding by Lanati and Thiele (2018).

Second, we document non-linear effects in the aid-migration relationship. This finding is important because it implies that there is a turning point in the aid-migration relationship (we estimate this tipping point to be at about $411 million). When we are below the turning point (i.e., aid payments below $411 million), increasing aid payments will lead to an increase in migration flows. In contrast, when we are above the turning point, further increases in the aid payments will reduce migration flows. This non-linearity is important for policy makers to consider when they decide to use aid payments as a tool to manage migration flows, as the marginal effect of aid on migration depends on the level of aid payments. For example, the aforementioned Sudan spending program by the UK government would need to be about four times larger to achieve its goal, according to our results.

In general, our findings are of interest to policy makers and NGOs and can inform the design of aid payments intended to manage migration flows - a topic generating substantial attention in the EU and the United States in recent years (e.g., Washington Post, 2017; Reuters, 2018). Aid policies must be seen in combination with other policies and cannot be conducted independently. Managing foreign relations related to development, the economy, the flow of goods, services, capital, and people is interrelated and requires a complex policy mix. For example, according to our estimations, cutting foreign aid payments which are above the turning point will increase migration flows, but cutting foreign aid payments which are below the turning point will lower migration flows. Therefore, foreign aid policies and immigration policies must be coordinated to deal with higher (e.g., stress on infrastructure) or lower (e.g., lower labor supply) migration flows.

Along this line, our findings also contribute to achieving the UN Sustainable Development Goals, especially policy actions related to achieving goals 1 (No Poverty), 8 (Promote Growth), and 10 (Reduce Inequality Within and Among Countries), by highlighting both the imperatives and the challenges of better coordinating aid and migration policies. Further, our results document potentially unintended side-effects of cutting foreign aid, as aid policy affects migration and, hence, immigration policies. An example here is cutting foreign aid with the intention to reduce migration, but because the level of foreign aid is above the turning point, migration, in turn, increases. This situation would lead to the opposite, compared to the desired, outcome.

Based on the findings presented here, future research should work to further disentangle the effect of aid on migration. In particular, an important extension of this literature would be to study the effects of different types of aid payments (or projects) on migration. While studies such as Lanati and Thiele (2018) go in this direction, exploring recently available geo-coded data sets at the aid-project level could offer new insights when combined, for example, with nationally representative surveys. Further, since aid projects are often coordinated across countries (i.e., within the EU), it would be interesting to study whether these joint projects have different effects on migration and, if they should increase migration, where these migrants choose to locate. These considerations could also help us better understand the reasons for the non-linearity we find in the aid-migration relationship. Finally, studying whether aid projects have persistent, rather than temporary, effects on migration flows would be an interesting research question.

Supplemental Material

sj-pdf-1-mrx-10.1177_01979183211069316 - Supplemental material for Does Aid Drive Migration? Evidence from a Shift-Share Instrument

Supplemental material, sj-pdf-1-mrx-10.1177_01979183211069316 for Does Aid Drive Migration? Evidence from a Shift-Share Instrument by Hamish Fitchett, Dennis Wesselbaum in International Migration Review

Footnotes

ORCID iD

Dennis Wesselbaum

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Aburn

Wesselbaum

. 2019. “Gone with the Wind: International Migration.” Global and Planetary Change 178: 96–109.

Alesina

Dollar

. 2000. “Who Gives Foreign Aid to Whom and Why?” Journal of Economic Growth 5(1): 33–63.

Alesina

Weder

. 2002. “Do Corrupt Governments Receive Less Foreign Aid?” American Economic Review 92(4): 1126–1137.

Azam

J. P.

Berlinschi

. 2009. “The Aid-Migration of Trade-Off.” Working Papers 538, Institut d’Économie Industrielle Toulouse.

Baier

Bergstrand

J. H.

. 2009. “Bonus vetus OLS: a simple method for approximating international trade-cost effects using the gravity equation.” Journal of International Economics 77(1): 77–85.

Bartik

1991. “Who Benefits from State and Local Economic Development Policies?” W.E. Upjohn Institute for Employment Research .

Beine

Docquier

Özden

. 2011. “Diasporas.” Journal of Development Economics 95(1): 30–41.

Beine

Parsons

. 2015. “Climatic Factors as Determinants of International Migration.” Scandinavian Journal of Economics 117(2): 723–767.

Belloc

2011. “International Economic Assistance and Migration: The Case of Sub-Saharan Countries.” International Migration 53(1): 187–201.

10.

Berthélemy

J.-C.

2006. “Bilateral Donors’ Interest vs. Recipients’ Development Motives in Aid Allocation: Do All Donors Behave the Same?” Review of Development Economics 10(2): 179–194.

11.

Berthélemy

J.-C.

Beuran

Maurel

. 2009. “Aid and Migrations: Substitutes or Compliments.” World Development 37(10): 1589–1599.

12.

Borjas

1987. “Self-Selection and the Earnings of Immigrants.” The American Economic Review 77(4): 531–553.

13.

Bound

Holzer

H. J.

. 2000. “Demand Shifts.” Population Adjustments, and Labor Market Outcomes during the, 1980s. Journal of Labor Economics 18(1): 20–54.

14.

Cai

Feng

Pytliková

Oppenheimer

. 2016. “Climate Variability and International Migration: The Importance of the Agricultural Linkage.” Journal of Environmental Economics and Management 79: 135–151.

15.

Card

2001. “Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of Higher Immigration.” Journal of Labor Economics 19(1): 22–64.

16.

Cattaneo

Peri

. 2016. “The Migration Response to Increasing Temperatures.” Journal of Development Economics 122: 127–146.

17.

Chen

Mrkaic

Nabar

. 2019. “The Global Economic Recovery 10 Years After the 2008 Financial Crisis.” IMF Working Paper, 19/83.

18.

Clemens

2014. “Does development reduce migration?” In International Handbook on Migration and Economic Development, edited by R. Lucas, 152-185. Cheltenham, UK and Northhampton, MA: Edward Elgar.

19.

de Haas

2010. “Migration transitions: A theoretical and empirical inquiry into the developmental drivers of international migration.” Working Paper 24, International Migration Institute, University of Oxford.

20.

Department for International Development 2017. “UK aid tackling global migration crisis.” Press Release.

21.

Espisito, A. 2018. “U.S. pledges billions in aid to develop Central America, curb migration.” Reuters , December18.

22.

Faini

Venturini

. 1993. “Trade, aid and migrations: some basic policy issues.” European Economic Review 37(2): 435–442.

23.

Gamso

Yuldashev

. 2018. “Targeted Foreign Aid and International Migration: Is Development Promotion an Effective Immigration Policy?” International Studies Quarterly 62(4): 809–820.

24.

Goldsmith-Pinkham

Sorkin

Swift

. 2016. “Notes on Bartik Instruments.” Unpublished manuscript .

25.

Goldsmith-Pinkham

Sorkin

Swift

. 2018. “Bartik Instruments: What, When, Why, and How.” American Economic Review 110(8): 2586-2624.

26.

Grogger

Hanson

. 2011. “Income maximization and the selection and sorting of international migrants.” Journal of Development Economics 95(1): 42–57.

27.

Jaeger

Ruist

Stuhler

. 2018. “Shift-Share Instruments and the Impact of Immigration.” NBER Working Paper. No. 24285 .

28.

Lanati

Thiele

. 2018. “The impact of foreign aid on migration revisited.” World Development 111: 59–71.

29.

Liu

Andris

Desmarais

B. A.

. 2019. “Migration and political polarization in the U.S.: An analysis of the county-level migration network.” Plos One 14(11): e0225405.

30.

Lucas

R. E.

2005. International Migration and Economic Development: Lessons from Low-Income Countries. Cheltenham, UK and Northhampton, MA: Edward Elgar Publishing.

31.

Martin-Shields, C., S. Angenendt, and B. Schraven. 2017. “When and how can foreign aid slow migration?” The Washington Post, November 2.

32.

McFadden

. 1984. Econometric Analysis of Qualitative Response Models. In Handbook of Econometrics, edited by Z. Griliches and M. Intriligator, 1395-1457. Amsterdam: ElSevier.

33.

Nakamura

Steinsson

. 2014. “Fiscal Stimulus in a Monetary Union: Evidence from US Regions.” The American Economic Review 104(3): 753–792.

34.

Nickell

S. J.

1981. “Biases in dynamic models with fixed effects.” Econometrica 49(6): 1417–1426.

35.

Nunn

Qian

. 2014. “US Food Aid and Civil Conflict.” The American Economic Review 104(6): 1630–1666.

36.

Nunnenkamp

Thiele

. 2006. “Targeting Aid to the Needy and Deserving: Nothing but Promises?” The World Economy 29(9): 1177–1201.

37.

OECD (Organisation for Economic Co-operation and Development). 2010. “Development aid reaches an historic high.”

38.

Parsons

Winters

. 2014. “International Migration, Trade and Aid: A Survey.” In International Handbook on Migration and Economic Development edited by R.E.B. Lucas, 65-112. Cheltenham, UK and Northhampton, MA: Edward Elgar Publishing.

39.

Rajan

R. G.

Subramanian

. 2008. “Aid and growth: what does the cross-country evidence really show?” The Review of Economics and Statistics 90(4): 643–665.

40.

Roy

1951. “Some Thoughts on the Distribution of Earnings.” Oxford Economic Papers 3(2): 135–146.

41.

Santos Silva

. and Tenreyro

. 2006. “The Log of Gravity.” The Review of Economics and Statistics 88(4): 641–658.

42.

Stockemer

Niemann

Unger

Speyer

. 2020. “The 'Refugee Crisis,' Immigration Attitudes, and Euroscepticism.” International Migration Review 54(3): 883–912.

43.

Temple

Van de Sijpe

. 2017. “Foreign aid and domestic absorption.” Journal of International Economics 108: 431–443.

44.

Wooldridge

J. M.

2010. Econometric Analysis of Cross Section and Panel Data. Cambridge: MIT Press.

45.

Wozniak

Murray

. 2012. “Timing is everything: Short-run population impacts of immigration in US cities.” Journal of Urban Economics 72(1): 60–78.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.45 MB