Bilateral labor agreements and the migration of Filipinos: An instrumental variable approach

Though Philippine labor migration dates to the 16th century, the modern era of overseas employment was ushered in by the debt crisis under the Marcos regime (1965–1986). To ease surplus labor supply and calm civil unrest under martial law, Ferdinand Marcos devised his foreign policy strategy, referred to as “development diplomacy” (Punongbayan and Mandrilla, 2016). The administration predicated this plan on finding, establishing, and formalizing international labor markets for Filipinos to travel to (Santos, 2014). The Philippines began negotiating bilateral agreements with other nations for the expressed purpose of regulating temporary migration. By 1974, the Overseas Employment Program was adopted into the Philippine Labor Code, formalizing national labor export policy for the first time. Male migrants found work in construction in many Middle Eastern countries that were booming with oil money and investing in infrastructure. Filipinas began migrating in the 1970s to fill the need for English-speaking teachers (James, 1997). However, the demand for female migrants spiked in the late 1980s when health, sales, and domestic service sectors opened in Hong Kong, Singapore, South Korea, Japan, and Taiwan (Center for Migrant Advocacy, 2011). This system flourished when fresh markets were introduced, allowing OFWs to dominate certain labor markets.

The Philippines has been a consistent source of labor throughout all phases of migration and BLA ratification (Chilton and Posner, 2018). The state has a history of using diplomatic tools for development gains. Though there is variation in the type, depth, and clarity of their BLAs, the treaties appear to be the desired form of official policy. The government continues to prioritize labor agreements for deploying migrant workers, evidenced by the ongoing negotiations in Eastern Europe (CNN Staff, 2020). Due to this capital-producing system, the widespread desire to replicate the Philippines’ export model in other source states is growing (Mendoza, 2015). The following section situates BLAs and migration in theoretical contexts and explores the competing and commingling explanations as to why people move.

BLAs, much like other international treaties, are best understood through the international relations lens of liberalism. This is the belief that states achieve economic and social power through international cooperation rather than direct force. Though liberalists are not against military action, they prefer to utilize international diplomacy and bi- or multilateral agreements to achieve their national interests.

Countries use BLAs as diplomatic tools to regulate labor migration. These instruments take several forms with varying degrees of enforceability including Memorandums of Understanding (MOUs), Memorandums of Agreement (MOAs), Protocols, Annexes, Inter-Agency Agreements (IAAs), framework agreements, statements of mutual labor cooperation, and cross-border worker agreements (Wickramasekara, 2015). Though growing in popularity,

See Chilton and Posner (2018) and Wickramasekara (2015) for more information on global trends of BLA conclusion.

States enter BLAs for a variety of social and economic reasons. Reducing exploitation is a common social goal for labor-sending countries when entering BLAs. They wish to protect the rights of their nationals abroad, improve working conditions, and negotiate fair contracts with minimum standards (Blank, 2011; Chilton and Posner, 2018; Moraga, 2008). Labor-sending countries may wish to strengthen economic relations with certain states. Entering a BLA could establish other positive avenues for trade and investment down the road. Additional economic drivers for source countries include relieving a labor surplus, facilitating the return of remittances (Chilton and Posner, 2018), and preventing brain drain by negotiating contract terms and repatriation (Moraga, 2008).

Host or labor-receiving countries may wish to attract migrants to respond to a labor market need, as with the oil industry in the 1970s or the need for health-care workers to assist with aging populations in Japan, Canada, and the United States. Greater ability to control regular and irregular migration motivates hosts (Blank, 2011; Moraga, 2008). States can accomplish this by establishing specific types of visas and setting quotas relative to the labor shortage. Labor-receiving nations may also wish to secure a benefit unrelated to migration from a source country (Sykes, 2013), promote cultural ties, or gain symbolic political benefits (Chilton and Posner, 2018).

Migration scholars hypothesize a myriad of contributing factors and frameworks to explain why people move; however, there are three dominant theories. Researchers conceptualize the first as push and pull factors. The push-pull theory extrapolates key tenets from Ravenstein's 1885 work, The Laws of Migration. Determinants that push migrants to leave their home localities are war, violence, poor working conditions, meager wages, famine, and natural hazards. Services, food security, better job opportunities, and political asylum may also pull migrants to a new location. The applications of this push-pull theory of migration are vast and vary, but the central assumption is that migration is a function of sending and receiving countries’ economic performance (Cuamea Velázquez, 2000). Some scholars assume a symmetry hypothesis, stating that the push and pull factors between source and host countries impose approximately equal, “but opposite effects (elasticities)” (Schultz, 1982, p. 573). In her empirical study on the determinants of Filipino migration, Carlos finds evidence for asymmetry resulting from uneven information, natural barriers, and restrictive policies in host states leading her to conclude that “factors in the Philippines exert greater influence on the probability to migrate” (2002, p. 100). There is a significant overlap between push-pull migration determinants and the reasons for the BLA conclusion. BLAs are inherently flexible and versatile tools that account for asymmetry.

The second migration theory growing in popularity, the labor recruitment approach, builds on Piore's Dual Labor Market theory (1971). This theory depends on bilateral agreements to formalize, or private companies to orchestrate, direct recruitment to fill what Piore (1971) calls the “secondary sector.” This approach to explaining migration sets itself apart from push-pull theory by intentionally excluding income and wage gaps and focuses on employers’ recruitment strategies (Cuamea Velázquez, 2000). Piore develops the constructs of the Dual Labor Market to explain why Black Americans in urban areas occupy second sector jobs with poorer working conditions, lower wages, and less upward mobility. His core tenets challenge the push-pull supposition that migrants enter better living and working conditions inherent to their movement. Piore and others expand this theory to explain international migration patterns besides rural to urban in-country movement. When subscribing to the assumption that BLAs increase international migration flows, this is an apropos theoretical lens through which to view these agreements; unfortunately, it yields limited empirical results (Ashton and Maguire, 1984).

To address the shortcomings of the aforementioned theories, some migration scholars describe this phenomenon as a social process.

Conceptualizing migration as a social process creates space for analyzing the interaction between flows and structural features of host and source states. This perspective accounts for the macroeconomic factors that influence migration systems and considers a variety of incentives—beyond financial enticements—to motivate individuals. Accounting for the interconnectedness of decision-making is more comprehensive than other frameworks; however, it is only useful for ex post facto studies.

These three theories each provide an additional layer of explanation as to why people migrate. Unfortunately, they lack predictive power. This is a problem for states wanting to plan evidence-based immigration policy. It is possible to find case studies that prove or disprove each of these migration theories. Disney et al. (2015) find no reliable theory for predicting migration. Both deterministic forecasts (often called “projections”) and probabilistic (stochastic) forecasts have very high levels of error. Migration is difficult to forecast due to the myriad of political, social, economic, meteorological, and climatological drivers. Despite constraints in migration forecasting, policymaking must go on.

Varying theories and commingling reasons for seeking work abroad lead us to question BLAs’ ability to fulfill their objectives. If states are successful in achieving economic and social power through this diplomatic tool, it offers further evidence for continuing BLA use in international diplomacy and obliges researchers to expand their investigations. The following section reviews relevant empirical literature on global BLA conclusions and impacts in the Philippines.

Although academic attention to BLAs is increasing (Go, 2004, 2007; Gordon, 2010; Panizzon, 2010; Sáez, 2013), very few empirical studies test their strengths and weaknesses as a policy instrument. To my knowledge, there are three published articles on the topic that use econometric models. Chilton and Posner (2018) and Peters (2019) posit why countries sign BLAs and use unique global datasets to test several hypotheses. Chilton and Woda (2021) use administrative data from the Philippines to test the impact of BLAs on the flows of migrants to and remittances from cosignatory countries.

Chilton and Posner (2018) assemble an original dataset of 582 BLAs between 1945 and 2015 and test several assumptions using state-state dyads. The primary focus of their study is to test hypotheses around why states conclude labor agreements with one another. One such argument is that agreements draw countries with sizeable differences in the level of democracy and strength of economy into concordance on migration policy. The authors use logit regressions to examine patterns of ratification and find evidence against the belief that countries with dissimilar political regimes and relative levels of wealth are more likely to sign BLAs unless the host state is Middle Eastern. The researchers also conduct some exploratory analysis on their country dyads to determine the effect of BLAs on migration. They find that across their global dataset, countries with BLAs experience higher levels of migration; however, these trends appear to begin slightly before the BLAs ratification. Thus, Chilton and Posner (2018) do not claim a causal relationship between BLAs and migration.

In another global study, Peters (2019) uses a rare events logit model to test the hypothesis that hosts, or labor recipient states, enter BLAs only when they cannot fulfill their labor needs unilaterally. She argues that host states have all the negotiating power; hence, it will not benefit them to give up some control over immigration if the country meets its own labor needs. Using BLAs to serve as recruitment tools, instead of third-party middlemen, solves the “matching problem” Peters presents. By signing treaties, hosts can reduce screening costs and vacancy costs associated with filling positions. Here she assumes that “costs of screening tend to rise when the economy needs more skilled labor” (Peters, 2019, p. 290). Peters uses labor productivity to support the hypothesis for reducing screening costs, and remoteness and an increase in labor force participation to measure vacancy costs.

Peters’ most notable finding in relation to this study is that a BLA is more likely when the dyad signs another treaty (trade or aid) in the same year. She does not attribute this to linkage, because none of the BLAs she examines include links to other treaties but to the existing relationship between the two countries. This finding supports the IV identification strategy I discuss in the next section. Additionally, Peters (2019) finds that hosts are less likely to sign labor agreements with source states from which they already have a large stock of migrants, supporting her hypothesis that receiving states do not enter BLAs when they can fulfill their labor needs independently. This result explains the holdout by Saudi Arabia from signing a BLA with the Philippines (2005) despite the source country's decades-long effort.

Finally, her analysis reveals a negative and statistically significant relationship between BLA conclusion and distance, showing that a remote host country is nearly three times more likely to sign an agreement than a non-remote host. The logic behind this finding is that distance is a barrier to migration; clauses requiring recruitment agencies or employers to cover the associated travel costs can mitigate that barrier. This may be relevant to OFWs because many of their destinations are island nations.

Chilton and Woda (2021) attempt to mitigate past criticism of empirical evaluations of international treaties by employing an event study research design. This method allows them to model control and treatment groups to test the impact of Philippine BLAs on emigration and remittance flows. Event studies must satisfy three fundamental assumptions: (1) parallel trends that are independently and identically distributed through time, (2) events are unexpected, (3) no other events can occur during the event window. Unlike the two studies above, these authors use OFW labor migration data specifically rather than general migration variables. Their dataset includes the 68 land-based BLAs originally from Mangulabnan and Daquio's report (2019).

Their findings indicate that BLAs between the Philippines and cosignatories do not increase the deployment of migrant workers, nor the return of capital in the form of remittances. Chilton and Woda (2021, p. 20) argue that their “results suggest that signing more BLAs may not be an easy solution for countries wishing to increase their labor migration, and the financial benefits of that labor migration.” This finding is critical for policymakers who are actively negotiating labor agreements as part of the nation's COVID recovery plan (CNN Staff, 2020).

The authors recognize four limitations of their study: non-random BLA conclusion, imperfect data, inability to test other reasons the Philippines signs such treaties (namely, worker protections), and lack of generalizability to other countries’ BLAs. Like Chilton and Woda (2021), this study uses specific labor migration data from the World Bank and the Philippine Overseas Employment Administration (POEA) and has the same limitations. However, an IV's research design has different identifying assumptions than an event study design and allows other events (e.g., bilateral treaties) to be instrumented in place of the endogenous BLA. It is promising that by using different estimation techniques with similar data, we reach the same conclusions.

These studies are all critically important pieces to the BLA-migration puzzle. The following analysis uses the first IV approach to test whether BLAs truly impact migration. The paper proceeds with a thorough explanation of the methodological contribution and details my identification strategy, including two suitable instruments.

The theoretical model is macroeconomic in nature and allows migrant stock of OFWs to be a function of a BLA with a host country, in addition to a set of control variables, (1) yit=β0+β1BLAit+†2′xit+αi+δt+uit {y_{it}} = {\beta _0} + {\beta _1}{BLA_{it}} + {{\it \dagger} _2}^\prime {\boldsymbol {x}_{it}} + {\alpha _i} + {\delta _t} + {u_{it}} where subscript it denotes the country i in time t. BLA_it is the independent variable of interest, a dummy variable for the years a BLA is in effect between the Philippines and country i. For example, BLA for Saudi Arabia is 0 from 1960 to 2004 and 1 from 2005 to 2018. β1 is the coefficient of interest, x_it is a vector of country characteristics, α_i is an unobserved time-invariant country-fixed effect, δ_t is a time trend, and u_it is the error term. The dependent variable, denoted by y_it, is the number of temporary Filipino migrant workers in country i at time t.

The ordinary least squares (OLS) method produces biased results since the independent variable, BLA, suffers from two types of endogeneity. First, it is plausible that reverse causality affects the model. One situation where this may occur is between the Philippines and Saudi Arabia. Filipino migration to Saudi Arabia increased by 560% from 1960 to 1970 and rose another five-fold between 1970 and 1980 due to the oil boom. To expedite deployment the Philippines proposed a BLA in 1988 but the Kingdom declined to take part in negotiations. Migration continued to rise, and the Philippines remained consistent in its efforts to formalize the labor relations between the two countries. The parties eventually signed a BLA in 2005 when migration peaked at around one million. Though Saudi Arabia remains a top destination for OFWs, there has been a decline in the number of migrants since the BLA. In this dyad, it seems that migration was the driver for the BLA conclusion and is consistent with reverse causality.

Similarly, reverse causality may also be present with Japan, a top destination country for OFWs, especially seafarers (Go, 2004). The Philippines proposed an MOA on Mobilization of Manpower in 1988. Japan did not outrightly decline the BLA; instead, they chose the course of inaction. Deployment increased year after year from the late-1980s to mid-2000s until Japan tightened its entertainer visa restrictions to address human trafficking. Deployment of newly hired OFWs dropped by 81.7% from 2005 to 2006. Negotiations of the Japan-Philippines Economic Partnership Agreement (JPEPA), a unique bilateral agreement concerning investment, trade, and labor, were ongoing. According to Amante, “the employment of Filipino nurses and caregivers in Japan was the main item requested by the Philippine side” (2007, p. 26). This indicates an effort to increase deployment, dominate a new sector, and relieve a nursing surplus. JPEPA includes language for seafarers and nurses and ultimately led to the conclusion of two specific MOUs with Japan in 2009 for these industries.

Besides reverse causality, OLS likely suffers from omitted variable bias caused by other unobserved factors driving the migration patterns of Filipinos that are also correlated with the adoption of BLAs. Arguments exist in qualitative literature on the strength of social networks and imitation of kin migration patterns, also known as stepwise migration (Francisco-Menchavez, 2020), which is unaccounted for in the data. The Philippines tends to conclude agreements where its citizens are already working hoping to expand protections, encourage remittance transfer, and maintain the migration stream. If OFWs are migrating to countries where their family members work and there are systematically more BLAs with those destinations, that omitted variable would cause positive bias in the BLA coefficient. The unmeasured migration would over-inflate the effect of the agreement on migration.

Exploitation of migrant workers is an issue the Philippine government contends with regularly. Scholars cite increasing protections for OFWs as a reason for BLA conclusion (Blank, 2011; Chilton and Woda, 2021; Go, 2004). It is also an issue that I expect to bias the OLS model. Exploitation is likely to have a negative effect on the dependent variable and positively impact the independent variable. To my knowledge, there are no empirical studies that analyze this issue, nor does the Philippines distribute quantitative data on violation of migrant workers’ rights.

To overcome endogeneity concerns, I use an IV estimator. The impact of BLAs on the migration of Filipinos is identified by two instruments, bilateral investment treaties (BITs) and formal alliances. IV regressions are a well-tested approach in econometrics that allows researchers to identify causal effects in the presence of endogeneity. The scarcity of empirical literature on the impact of BLAs on migration is due, in part, to the difficulty of finding a suitable instrument that must satisfy two assumptions, instrument relevance and the exclusion restriction.

Equation (2) estimates BIT and Alliance as instruments for BLA. The country characteristics included in x_it are GDP ratio, unemployment rates, government spending, population, imports, exports, and Polity2

“Revised Combined Polity Score: This variable is a modified version of the POLITY variable added in order to facilitate the use of the POLITY regime measure in time-series analyses. It modifies the combined annual POLITY score by applying a simple treatment, or ‘fix,’ to convert instances of ‘standardized authority scores’ (i.e., −66, −77, and −88) to conventional polity scores (i.e., within the range, −10 to +10)” (Center for Systemic Peace, 2020).

The independent variable of interest is an indicator for the presence of a labor agreement with the Philippines in country i at time t. The variable includes all forms of diplomatic instruments used to reach an agreement for the Philippines to supply labor to a host country, such as MOAs, their annexes and protocols, and less-binding forms such as MOUs. Seventy-one BLAs have been identified between the Philippines and other states or provinces, 69 concern land-based labor and the remaining two are sea-based. Of these, there are 29 unique country pairs, and the remaining 42 are updated protocols, renewals, or brand-new agreements with previous partners. As a robustness check, I test two forms of this variable, BLA-onset and BLA-real, both dichotomous. BLA-onset assumes that once an agreement is in place for a country dyad it remains in effect until the end of the sample period. The benefit of using the onset form is to test for the lasting effects of a BLA on migration; this captures the persisting impact on migration, even if the agreement has expired.

BLA-real refers to the actual years that the agreements are in effect. This variable measures the impact of the agreement on migration during the window the agreement is valid and does not consider persisting effects. Unfortunately, some information is missing for Liberia (33 obs.), Norway (16 obs.), and Qatar (15 obs.). The text of the Liberian BLA concerning seafarers is publicly available but does not specify the length of the terms. Texts for the BLAs with Norway and Qatar (1981) are not publicly available, and I could not identify if the agreements remain in force or expired. These data are considered missing at random, there is no indication that agreement texts or status details are not available systematically, thus the missing values are not likely to bias results. Using the “real” form identifies 461 observations as having a BLA in effect while using the “onset” form increases this figure to 556 observations. This difference results from missing observations (64) and expired agreements (31 obs.) (see Table A1 in Appendix A).

The following control variables, denoted by X, measure wealth, trade relationships, and governance schemes. I divide the GDP per capita of the host countries by the GDP per capita of the Philippines to calculate a ratio for each country dyad. The natural log of the host-to-Philippines GDP ratio is in the model to capture the relative wealth between the host and source countries. There is evidence that workers migrate to countries with growing economies (Ghosh, 2013; Rass, 2012; Wickramasekara, 2015). A simple glance at the World Bank's GDP per capita indicator graph demonstrates the upward trend of global growth across the sample period. Overall, this does not provide much information. The ratio, however, gives some insight into how the economies of host countries are growing relative to the Philippines.

The model contains national estimates of unemployment because this variable has a negative relationship with migration, according to Geis et al. (2008). Host countries with high levels of unemployment may be less likely to sign BLAs and may be less desirable destinations for OFWs. The model controls for the population because it is correlated with both x and y. Population growth of developed countries is slowing because of aging and decreasing fertility while populations of developing nations are booming. This is the precise reason that countries like Canada, Germany, and Japan are recruiting Filipino nurses and care workers via bilateral agreements. According to Brush and Sochalski, “[o]ver the past 60 years, the Philippines has led the world in preparing nurses explicitly for export to meet the demand for nurses in the United States and other developed nations” (2007, pp. 37–38).

To mitigate omitted variable bias and improve the precision of the estimates, I include direction of trade statistics. Peters’ study (2019) finds that country pairs are more likely to sign a BLA when they sign another economic or aid treaty in the same year. Furthermore, trade is correlated with migration (Campaniello, 2014; Uprety, 2019). Bilateral trading between all possible country dyads has increased from 13% in 1950 to 58% in 2014 (Ortiz-Ospina and Beltekian, 2018). Though there is still one-quarter of country pairs in a non-trade relationship and nearly 17% unilaterally trading, the effects of globalization on mutual export partnerships have been positive. Eaton and Kortum (2002) document a negative relationship between trade and distance, speaking to the strength of the Philippines’ ability to trade with every country in the sample.

The regression includes a variable for the absolute value of the difference between the Polity2 score of the Philippines and the Polity2 score of the host country. Center for Systemic Peace (2020) rates countries on their level of democracy and then assigns an annual Polity score. The scores range from −10 (total autocracy) to 10 (consolidated democracy). The absolute value of the difference captures the governance scheme similarity of the country dyads across the sample period. The coefficient on the variable estimates the effect of the relative level of democracy on migration flows of Filipinos and provides a test of whether OFWs migrate to host countries with political systems like their own in a given year.

The difference in Polity2 scores without taking the absolute value is substituted for this variable as a robustness check discussed in Section 5.1.

To proxy for a variety of country characteristics likely to attract migrants, such as strong health-care systems, wages, education systems, and other goods and services which have a positive effect on migration, the model includes government spending (Böheim and Mayr, 2005; Dinbabo and Nyasulu, 2015; Geis et al., 2008). Migrants’ preferences may differ based on their skill level and length of the employment contract. Filipinos respond to labor market needs in both high- and low-skilled categories. The 2018 Survey of Overseas Filipinos reports nearly 75% of all OFWs holding low-skilled occupations (Philippine Statistics Authority, 2019).

The model includes a set of region dummies to further control for variation and regional shocks across time. For consistency and replicability, I group countries using the World Bank's region classifications: East Asia and Pacific, Europe and Central Asia, Latin America, Middle East and Northern Africa, North America, South Asia, and Sub-Saharan Africa. Model 3 interacts these region dummies with a time trend (Table 2).

Results from a series of robustness checks are available in Appendix B. Using the onset version of the BLA indicator variable instead of the real version produces qualitatively similar results (See Table B1). The results are also robust to redefining the instruments for the IV estimation. The first model in Table B2 in Appendix B uses BIT as the only instrument for BLA. This specification is nearly identical to the main model (Table 2, Model 3); GDP per capita ratio and exports are still significant at 10% and 5%, respectively. The second model in Table B2 in Appendix B presents estimates when a formal alliance is the only instrument for BLA. There are no significant variables in this model and all the standard errors, except for unemployment, are inflated compared to the main model. For Model 3 of Table B2 in Appendix B, the instrument is a single dummy variable for either a BIT or a formal alliance. The coefficients and standard errors are relatively similar to the main model. GDP ratio and exports keep their significance and population becomes significant at 10%.

Allowing the difference in Polity2 scores to take on positive or negative values, rather than using the absolute value of this difference, changes the sign of the coefficient from negative to positive. The coefficient indicates that as the Polity score increases by one (becomes more democratic) there is an expected reduction of approximately 253 migrants. Labor demand, which is high in Middle Eastern countries that have low polity scores, likely drive this negative relationship rather than migrants seeking less democratic conditions. Redefining the polity variable leads to an unremarkable change in other coefficients. Finally, since government spending is part of GDP and unemployment is a percentage of the population, I conduct checks that remove the controls. Dropping them from the model independently or together leads to no change to signs or significance of the coefficients.