KNOMAD PAPER 37 Migration Projections: The Economic Case Thomas Buettner and Rainer Muenz January 2020 i The KNOMAD Paper Series disseminates work in progress under the Global Knowledge Partnership on Migration and Development (KNOMAD). A global hub of knowledge and policy expertise on migration and development, KNOMAD aims to create and synthesize multidisciplinary knowledge and evidence; generate a menu of policy options for migration policy makers; and provide technical assistance and capacity building for pilot projects, evaluation of policies, and data collection. KNOMAD is supported by a multi-donor trust fund established by the World Bank.The European Commission, Germany’s Federal Ministry of Economic Cooperation and Development (BMZ), and the Swiss Agency for Development and Cooperation (SDC) are the contributors to the trust fund. The views expressed in this paper do not represent the views of the World Bank or the sponsoring organizations. Please cite the work as follows: Buettner, Thomas, and Rainer Muenz. 2020.“Migration Projections: The Economic Case.”KNOMAD Paper No. 37, World Bank, Washington, DC. All queries should be addressed to KNOMAD@worldbank.org. KNOMAD papers and a host of other resources on migration are available at www.KNOMAD.org. ii Migration Projections: The Economic Case* Thomas Buettner and Rainer Muenz† Abstract This paper adds an economic dimension to the projection of international migration flows. Using existing estimates of international migration flows and demographic projections from the United Nations, the paper analyzes the impact of economic development, expressed as gross domestic product (GDP) per capita, on international migration. The analysis was inspired by the migration transition hypothesis – also known as the “migration hump” theory – and confirmed ahypothesized nonlinear relationship between migration and GDP per capita. Despite the large variability of the data, nonparametric fits suggest that emigration rates are relatively low in low-income settings, rise with rising GDP per capita, and decline at high income levels. On the other hand, immigration rates seem to increase unabated with rising income levels. For the projection of international migration flows, the nonparametric curves were parametrized by logistic and bi-logistic functions. Migration projections for 183 countries with constant emigration rates and with migration rates augmented by (projected) GDP per capita were calculated and the results summarized. The results show that international migration flows might substantially increasing when countries pass from low- to high-income economies. The paper also considers possible interactions between labor force dynamics and international migration but finds insufficient evidence for the formal integration of employment dynamics into the formulation of assumptions of international migration. Labor force projections driven by demographic change and projections of labor force participation rates are calculated for 184 countries and summarized. This paper provides strong evidence that economic dynamics should be considered into the formulation of assumption for future trends of international migration. Key words: International migration, Population projection, Migration transition hypothesis, Migration scenarios, Labor force, GDP ___________________________ *This paper was produced for KNOMAD’s Thematic Working Group (TWG) on Data and Demographics. KNOMAD is headed by Dilip Ratha, the Data and Demographics TWG is chaired by Rainer Muenz, and the TWG’s focal point in the KNOMAD Secretariat is Supriyo De. †Thomas Buettneris a senior demographer who was working at the United Nations Population Division before his retirement.Rainer Muenzworked asan adviser onmigration and demography at the in-house think tank of the European Commission. The authors may be contacted at planetbuettner@gmail.com andrainer.munz@ec.europa.eu. The analysis and proposals expressed in this paper are the authors ’ personal views and do not represent the positions of their current or former employers. iii Contents 1. INTRODUCTION ..................................................................................................................................... 1 2. DATA ..................................................................................................................................................... 1 2.1. Migration....................................................................................................................................... 1 2.2. Gross domestic product ................................................................................................................ 2 2.3. Employment .................................................................................................................................. 3 2.4. Population ..................................................................................................................................... 3 3. GLOBAL MIGRATION DYNAMICS........................................................................................................... 3 4. ECONOMIC DEVELOPMENT AS A DRIVER OF INTERNATIONAL MIGRATION ........................................ 6 4.1. GDP and migrant stocks ................................................................................................................ 7 4.2. GDP and migrant flows ................................................................................................................. 8 4.3. Migration models ........................................................................................................................ 10 5. MIGRATION AND EMPLOYMENT ........................................................................................................ 13 5.1. Projected labor force .................................................................................................................. 15 5.2. Constant labor force ................................................................................................................... 16 5.3. Does the labor market drive international migration? ............................................................... 17 6. MIGRATION SCENARIOS...................................................................................................................... 18 6.1. Toward alternative KNOMAD migration projections .................................................................. 18 7. DISCUSSION......................................................................................................................................... 23 8. BIBLIOGRAPHY .................................................................................................................................... 25 Appendix A. Tables...................................................................................................................................... 28 Appendix B. Figures..................................................................................................................................... 30 Appendix C. Methodology .......................................................................................................................... 31 C.1 Bi-logistic migration model ............................................................................................................... 31 Tables Table 1: Parameters of the Bi-Logistic Emigration Model .......................................................................... 11 Table 2: Parameters of the Logistic Immigration Model ............................................................................ 12 Table 3: Projected Size of Labor Force, by World Bank Geographic Region, 2015–2100........................... 15 Table 4: Absolute Change in the Labor Force, by World Bank Geographic Region, 2015–2100 ................ 15 Table 5: Relative Change in the Labor Force (%), by World Bank Geographic Region, 2015–2100 ........... 15 Table 6: Labor Force Distribution, by World Bank Geographic Region, 2015–2100 .................................. 16 Table 7: Implied LFPR with Constant Labor Force (%), by World Bank Geographic Region, 2015–2100 ... 16 iv Figures Figure 1: Global Number of Migrants, by Five-Year Period, 1960–2015 ...................................................... 4 Figure 2: Global Migration Rates, by Five-Year Period, 1960–2015 ............................................................. 4 Figure 3: Crude Migration Rates (per 1,000), by Country Income Group, 1960–2015................................. 5 Figure 4: Migration Transition: Emigrant Stock (emigrants living abroad), 2005 ......................................... 7 Figure 5: Migration Transition: Immigrant Stock (share of foreign-born residents), 2005 .......................... 7 Figure 6: Migration Transition: Emigration Flows, 2005–10......................................................................... 8 Figure 7: Migration Transition: Immigration Flows, 2005–10 .................................................................... 10 Figure 8: Migration Transition: Emigration Flow Model ............................................................................. 11 Figure 9: Migration Transition: Immigration Flow Model .......................................................................... 12 Figure 10: Employment and GDP, 2010–14 ................................................................................................ 13 Figure 11: Employment and Emigration, 2010–14 ..................................................................................... 14 Figure 12: Total Global Migrants, by Scenario and by Five-Year Period, 2005–2100 (in millions) ............. 19 Figure 13: Absolute and Relative Size of Emigrants, by World Bank Income Group, 2005–2100 .............. 20 Figure 14: Absolute and Relative Size of Emigrants, by World Bank Geographic Region, 2005–2100 ...... 20 Figure 15: Total Emigration from Nigeria and Niger, Constant Scenario (2005=100), 2005–2100 ............ 21 Figure 16: Growth of Total Emigrants, Nigeria, and Niger, by Scenario, 2005–2100 ................................. 22 Tables and Figures in Appendices Tables Table A.1: Total Migratory Movements, by Scenario and Five-Year Period, 2005–2100 ........................... 28 Table A.2: Total Emigrants, by World Bank Region, Selected Five-Year Period, 2005–2100 ..................... 28 Table A.3: Share of Migrants, by World Bank Region, Selected Five-Year Period, 2005–2100 .................. 28 Table A.4: Total Emigrants, by World Bank Income Group, Selected Five-Year Period, 2005–2100 ......... 29 Table A.5: Total Emigrants, by UN Region, Selected Five-Year Period, 2005–2100 ................................... 29 Table A.6: Total Emigrants, by UN Development Group, Selected Five-Year Period, 2005–2100 ............. 29 Table A.7: Average Labor Force Participation Rate, by World Bank Region, 2005–15 ............................... 29 Figures Figure B.1: Components of the Bi-Logistic Emigration Flow Model ........................................................... 30 Figure B.2: Geographic Labour Force Distribution, Absolute and Relative, 2015–2100 ............................ 30 v 1. INTRODUCTION Demographic change is a powerful driver of international migration. At the same time, international migration shapes the demographic development of many countries. This was shown in Buettner and Muenz (2018a) with simple assumptions and for six regions of the world. By using current (2010 –15) emigration and immigration rates, one can easily show that the still fast-growing population of Africa will generate a large volume of potential future migrants, but it is unclear how many of them will actually leave their countries and, ultimately, their continent. The population of Europe, on the other hand, is expected to change in the opposite direction, leading to demographic decline in the absence of significant immigration from other parts of the world. If Europe were to maintain its current rates of immigration over the next several decades, its population size would not change significantly. And even if all migrants arriving in Europe were to be of African origin, the migration potential of Africa could never be absorbed by Europe’s potential need for additional migrants. The paper on the demographic drivers of international migration quoted above (Buettner and Muenz 2018a) suggests a balancing solution to the apparent mismatch between projected or predicted emigration flows of sending countries vs. immigration flows of receiving countries that differential demographic trends would imply. However, most current global and national forecasts link either assumed net migration or both emigration and immigration to the dynamic of total population and age structure change. No systematic attempts have been made so far to link future migration forecasts to socioeconomic development and/or expected changes in the labor force. In this paper, we approach the problem differently by focusing on the interaction between economic development, measured as a change in gross domestic product (GDP) per capita over time, and international migration (emigration and immigration). We are also looking at the possible interaction between labor force developments and international migration. This offers a more nuanced and direct way to address the possible means and motivations of leaving a country of origin (emigration) or to accept or even recruit migrants in a country of destination (immigration). Calculations using an existing migration transition model (the so-called “migration hump” theory) and data on labor force participation rates from the International Labour Organization (ILO) were performed for all countries of the world. The results are summarized using the country groupings put forward by the World Bank Group and the United Nations. 2. DATA 2.1. Migration The empirical data on international migration remain very limited. In countries with a developed statistical system, immigrants are usually much better documented and accounted for than emigrants. Yet for those countries, international comparison is made more difficult by the fact that they apply different concepts and practices in the process of registering migration flows. Less developed countries often lack resources and stable institutions to account for international migration at all. Beyond the current migration statistics (measuring flows) produced by more developed countries, the only other data sources that allow the (indirect) estimation of migration flows are the 1 migrant stock tabulations produced by decennial censuses. Based on these, international population projections normally use net migration, estimated as residual, as a proxy for the migration component. In this context, Abel (2009, 2017) has pioneered the global estimation of migrant flows from migrant stock tables (using the stock-to-flow method). He has produced, in ever greater detail, migration flow estimates for the years 1960 to 2015 (Abel 2017). These data have been successfully used in global projection exercises (Buettner and Muenz 2018b; Lutz, Butz, and KC 2014a). But it also must be stressed that (indirect) estimates of international migration flows, while extremely useful, are also fraught with problems. The most significant challenge is the very heterogeneous statistical bases—that is, census tabulations that are not using comparable concepts and definitions of who is “foreign born” (see Özden et al. 2011 for attempts to clean and verify the raw census data). Another challenge is the difficulty of defining a meaningful spatial structure in the estimation model (Abel 2013, 2017), whichuses geographic distance as a “deterrence function.” In aglobal projection exercise, Lutz, Butz, and KC (2014b) adjust the international migration flow estimates generated by a stock-to-flow methodology to match the net migration estimates published by the United Nations Population Division (UNPD) (United Nations 2015). Such an approach avoids confusion and relies on reputable UN estimates. Yet it cannot be ignored that net-migration estimates are essentially residuals from a demographic accounting exercise and contain, in addition to the real (but unknown) balance of in- and outflows of migrants, measurement deficits from censuses and estimation errors from vital statistics (e.g., births and deaths). Preliminary analysis suggests that the stock-to-flow methodology tends to estimate lower net migration levels thandoes the UNPD, thus underestimating international migration. Sometimes the stock-to-flow methodology even yields estimates with a different sign (negative vs. positive net migration). A thorough investigation into the sources of these discrepancies is beyond the scope of this paper; but such an exercise could eventually help to improveglobal international migration estimates and solidify the basis for international migration projections. In addition to model-based estimates of migration flows, this paper also analyzes the migrant stock data that form the bases for the flow estimates. We use the 2017 revision of the International Migrant Stock dataset (United Nations, 2017a). Based on the regularly updated dataset of the UNPD, the 2017 revision contains stock data on bilateral migrants for the years 1990, 1995, 2000, 2005, 2010, 2015, and 2017 for all countries. The estimates are based on official statistics on the foreign-born or the foreign population as reported in national censuses. 2.2. Gross domestic product We use data from the ninth revision of the Penn World Tables1 (Feenstra, Inklaar, and Timmer 2015), namely the indicator “Expenditure-based real GDP” [ RGDPe ] (chained purchasing power parity in US$ 2011 million). The ninth revision of the Penn World Tables uses population data from the United Nation’s 2015 Revision of World Population Prospects as the denominator (United Nations 2015). For GDP projections, several sources are available. We chose to use the long-term projections of the Organisation for Economic Co-operation and Development (OECD) (Dellink et al. 2017), which are 1 The data are available in electronic form, together with extensive documentation, on www.ggdc.net/pwt. 2 compatible with the “middle-of-the-road scenario” (scenario 2) of the Shared Socioeconomic Pathways (SSP2).2The projections are available from 2000 to 2100, expressed in billionsof (2005) U.S. dollars per year. 2.3. Employment For our analysis of labor force participation, we employ ILO data (ILO, 2018) and the results of the 2017 Revision of the UNPD World Population Prospects (WPP 2017). We were aware of the rather wide definition of the labor force participation rate (LFPR)3 and explored, as an alternative indicator, the employment-to-population ratio (EPR)4. Ultimately, we decided to use the EPR modelling the relationship between employment and GDP butproceed with LFPR for the projection exercise later in the paper because it is less volatile and because more data are available (including projections). For greater flexibility in customizing aggregates, the ILO’s LFPRestimates5 were combined with the corresponding population data (for age groups 15–64) from the UNPD (United Nations 2017b). 2.4. Population Data on population dynamics were obtained from the 2017 Revision of World Population Prospects, the latest revision at the time this paper was written (United Nations 2017b). 3. GLOBAL MIGRATION DYNAMICS The flow of international migrants has increased over the past several decades. But did the volume of international migration grow because of pure demographic effects, namely population growth,or did it grow because the propensity/intensity of individuals’movement increased? To answer these questions, we employ data on migration flows (Abel 2017). Figure 1 shows the trends in the total number of emigrants and immigrants from 1960–65 through 2010–15.6 The absolute number of people on the move increased significantly, from less than 20 million in the five-year period of 1960– 65 to about 45 million by 2010–15 and 36 million during the following five year period. The data, albeit aggregates from more than 200 sovereign countries and other territories, show some mild temporal 2 The International Institute for Applied Systems Analysis (IIASA) maintains a reference database for the Shared Socioeconomic Pathways (SSPs) that contains long-term GDP projections produced by OECD and IIASA. For the database, see: https://tntcat.iiasa.ac.at/SspDb/dsd?Action=htmlpage&page=about; https://data.oecd.org/gdp/gross-domestic-product-gdp.htm. 3 The labor force participation rate: “… is a measure of the proportion of a country’s working -age population that engages actively in the labour market, either by working or looking for work; it provides an indication of the size of the supply of labour available to engage in the production of goods and services, relative to the population at working age.” See https://www.ilo.org/ilostat-files/Documents/description_LFPR_EN.pdf. 4 The employment-to-population ratio is defined as the proportion of a country’s working -age population that is employed.; see https://www.ilo.org/ilostat-files/Documents/description_EPR_EN.pdf. 5 For a methodological note about the LFPR, see ILO (2017). 6 The chart shows minor numerical discrepancies between total immigrants and total emigrants due to incomplete coverage (missing countries and territories). In reality (covered insufficiently by available data) the two figures should be the same, of course. 3 fluctuations that could be the result of better data estimation in years following a census.7 But despite this, the fitted linear curve shows a clear upward trend. Figure 1: Global Number of Migrants, by Five-Year Period, 1960–2015 Figure 2: Global Migration Rates, by Five-Year Period, 1960–2015 If we account for population growth, we still see a slight upward trend in the migration rates, but this is less pronounced than when looking at estimated absolute numbers (Figure 2). The overall development of the annual individual propensity to migrate (that is, the likelihood that a person will move), as represented by the linear trend line, is much less pronounced than the increase in the total flow of migrating persons (Figure 1). Between 1960–65 and 2010–15, the overall migration rate (emigration or immigration) fluctuated around the 50-year average of 1.2 migrants per 1,000 population. The drop in the migration rate for 2010–15 is possibly due to missing information, in the absence of a recent census, but it could also be the result of effects related to the global financial crisis. 7 The migrant stock data are available for quinquennial periods (years ending in 0 and 5) but are based on decennial censuses. The estimates for years ending in 5 are therefore interpolations and possibly less reliable than for years ending in 0. 4 Migration flows – expressed as crude emigration rates (CEMR) and crude immigration rates (CIMR) – show several characteristic trends when analyzed separately by income group, a feature well in line with recent migration theory (de Haas 2010a). Later in this paper, we will translate the theory into a mathematical model suitable for projecting international migration. Figure 3: Crude Migration Rates (per 1,000), by Country Income Group, 1960–2015 As Figure 3 shows, the average immigration rate in high-income countries showed a strong increase from the 1960s (with a clear drop, however, between 2005–10 and 2010–15), while the emigration rate remained largely unchanged and low. Therefore, the net migration rate also increased significantly over time. In contrast, middle-income countries showed a trend of increasingly negative net migration rates that reflect the consistently higher rates of emigration over immigration. The middle-income countries are prime sources of the immigration gains seen in high-income countries. Low-income countries exhibit no clear trend, with large ups and downs. In 1990–95, low-income countries, as a group, even registered a positive net migration rate, mainly due to large immigration and return-migration spikes in Afghanistan, Ethiopia, and the Democratic Republic of the Congo. Overall, low-income countries had a negative net migration rate that declined slightly over time. Low-income 5 countries appear to be a smaller player than middle-income countries in the global migration redistribution. In many places, people are just too poor and marginalized to migrate. 4. ECONOMIC DEVELOPMENT AS A DRIVER OF INTERNATIONAL MIGRATION In this section, we analyze the link between migration flows and socioeconomic development, measured as a change in GDP per capita over time. We are inspired by the hypothesis of a “migration hump” or, better, a migration transition. The notion of a link between migration and development is now well established (see IOM 2018).De Haas (2009, 2010a) presents a justification from first principles. Skeldon (2012) discusses the continued relevance of the migration transition hypothesis. In de Haas et al. (2018), the authors emphasize that the migration transition is linked to phases of the demographic transition.8 Clemens (2017) presents empirical evidence showing the inverted U-shape of the migration transition by combining the share of emigrants (defined as the fraction of people born in each country that live in all other countries) to the log of GDP per capita of the sending country. This proportion of migrant stock or diaspora by country of origin is a stock indicator and thus does not necessarily relate to recent migration flows. However, Clemens (2014) showed earlier that the relation between (decadal) emigration flows and GDP per capita also exhibits an inverted U-shape. In addition, an OECD study presented further empirical evidence of the migration transition and its typical inverted U-shape (OECD 2016). Another empirical analysis of the “migration hump” is presented by Natale, Migali, and Münz (2018). 8 “…the migration transition hypothesis ... links phases of the demographic transition (from high to low fertility and mortality) and concomitant development processes to distinctive phases in a ‘mobility transition,’ in which development initially leads to more internal (rural-to-urban) and international emigration. Only when countries achieve higher income levels, emigration levels tend to decrease alongside increases in nonmigratory mobility – such as commuting – and immigration, which leads to their transformation from net emigration to net immigration countries” (de Haas et al. 2018). 6 4.1. GDP and migrant stocks We reanalyzed the relation between migrant stock and national GDP per capita with updated migrant stock data (United Nations 2017a) and the most recent estimates of GDP per capita (Feenstra et al. 2015). Our analysis also found empirical evidence supporting the inverted U-shape of the migration transition (see Figure 4 for the year 2005). Figure 4: Migration Transition: Emigrant Stock (emigrants living abroad), 2005 Note: Function ksmooth, bandwidth = 2.0, n = 168. So far, the authors addressing the migration transition phenomenon have concentrated on the relationship between emigration and GDP per capita, usually leaving immigration aside. This emigration bias (Buettner and Muenz 2018b) is quite common in demographic research but misses the fact that a large proportion of emigrants, at the end of their journey, turn into immigrants if they do not return within a foreseeable period of time to their country of origin. As a result, when applying the same approach as above to immigrant stocks, Figure 5 shows a completely different picture. The relationship between immigrant stock and GDP per capita does not exhibit a hump or any diverging trends (Figure 5). Clearly, increased wealth (GDP per capita) is associated with an increasing share of immigrants in the receiving countries. Figure 5: Migration Transition: Immigrant Stock (share of foreign-born residents), 2005 Note: Function ksmooth, bandwidth = 2.0, n = 169. 7 Not surprisingly, the immigrant stock of receiving countries rises as GDP per capita increases. It must be remembered that these stock data show the accumulated results of migrants’ moves over a longer period, but not necessarily the effects of recent flows and events causing migration. 4.2. GDP and migrant flows To analyze recent migration trends, we turn to migration flow estimates and their relationship to the level of economic output. This is for two reasons: • We try to analyze recent migration flows that are not distorted by past events/mass flows related to violent conflicts and large-scale natural disasters/extreme weather conditions. • We use the results of the analysis to propose a migration projection model better suited to making population projections than are most existing approaches. Population projections ideally require information or substantiated assumptions about total migration flows. If this is not possible, one must rely, at least, on net flows. We apply the same methodology used to model the migration transition now to migration flows. Clemens shows that emigrant flows exhibit an inverted U-shape vis-à-vis GDP per capita (Clemens 2014). For updated data covering the period 2005–10, the estimated curve (Figure 6) unsurprisingly shows the inverted U-shape, too. Figure 6: Migration Transition: Emigration Flows,2005–10 Note: Function ksmooth, bandwidth = 1.0, n = 160. We are also interested in examining the opposite relationship: that is, between immigrant flows and GDP per capita. In that constellation, a country’s actual admission of migrants is set in juxtaposition to the country’s economic position at that particular point in time (Figure 7). Here, the relationship between migration and economic performance appears to be of a different shape. Even if one considers the large fluctuations, it is apparent that the greater a country’s economic performance, the greater its draw for immigrants. This is not, of course, a completely new finding. In its functional form, though, it gives a handle on formulating projection assumptions (here for the immigration dynamics driven by GDP per capita). 8 9 Figure 7: Migration Transition: Immigration Flows, 2005–10 Note: Function ksmooth, bandwidth = 2.0, n = 156. Although the original data (the dots in the figures) do not immediately reveal any clear pattern, the estimation method (kernel regression estimator) is able to summarize the empirical data into a smooth and coherent form. It must be noted, however, that the result is dependent on the choice of the bandwidth and thus is loaded with some discretionary elements. It is also important to note that the resulting curves are a “cleaned” average. As such, the curves are not immediately suitable for predicting migration flows for individual countries. Indeed, the curves obtained in this section need further adaptation to become a more suitable tool for modeling assumptions of future projections. 4.3. Migration models In this section, we present parameterized models (orfunctions) representing the curves obtained through Kernel regressions.9 We employ a re-parameterized logistic function introduced by Fisher and Pry (1971) and extensively used for modeling global change (Grübler 1998; Marchetti 1997; Marchetti, Meyer, and Ausubel 1996; Meyer, Yung, and Ausubel 1999; Riahi, Grübler,and Nakicenovic 2007). The Fisher/Pry functional form has parameters that are easier to interpret and fit to the empirical data as they have realistic meaning, such as the growth rate of the S-curve and the length of time the curve takes to reach the midpoint of the growth trajectory (AppendixC). A logistic function exhibits an S-shape and describes a diffusion or growth process rising from an initial level to an upper or lower asymptote. However, most migration transition curves show a clear trend reversal. Such processes may be effectively modeled by a combination of two logistic functions whereby one diffusion process approaches an upper asymptote, and a second and delayed process approaches a lower asymptote. Combined, these two processes then represent trends with a reversal. 9 Nadaraya–Watson kernel regression estimates were calculated using the ksmooth function in the stats package available in the core implementation of the R language for statistical computing (R Core Team 2017). In addition, the package npregfast was used for validation (Sestelo et al. 2017). 10 Emigration model We present first the results of fitting a bi-logistic model to the curve of the emigration-flow-based migration transition. Figure 8: Migration Transition: Emigration Flow Model Figure 8 shows the estimated curve still exhibiting some (arbitrary) fluctuation, which the bi-logistic model (red curve) smoothed. We argue that such smoothing is advantageous for modeling and especially for forecasting, as it removes artefacts from becoming part of the input-affecting results. The bi-logistic model depicted in Figure 8 is itself a combination of two logistic curves, one approaching an upper limit of 3.8 emigrants per 1,000 population (see Table 1). The second curve starts at midrange to counteract the process represented in the first curve. The combination of the two curves (or processes) results in the emigration flow model shown in Figure B.1 (AppendixB). For modeling purposes, it may be easier to express the relationship between emigration flows and GDP per capita directly, without first transforming GDP per capita into log scale. The parameters of that model are also shown in the column “GDP” in Table 1. Table 1: Parameters of the Bi-Logistic Emigration Model Parameters Log (GDP) GDP k1 3.8 129.9 Delta_t1 2.1 4915.0 tm1 6.8 -3443.5 k2 -1.5 -127.6 Delta_t2 1.2 43873.1 tm2 9.4 -39004.3 11 Immigration model Each emigration event turns, at its destination, into an immigration event. In other words, the countries of origin and the countries of destination interact, through migrants, with each other (for a discussion, see Buettner and Muenz 2018b, 2018a). Analogous to the parametrized emigration flow models, we now present a logistic model for immigration flows, or admissions, into the destination country. A visual inspection of the smooth regression curve Figure 9) seems to suggest the use of a single logistic, as a clear trend of reversal seems to be absent. Figure 9: Migration Transition: Immigration Flow Model The virtual unchecked propensity of higher immigration rates as the GDP per capita rises may be caused, in part, by a few countries that have unusual immigration and economic histories, such as many countries in the Gulf and several Asian city-states. We therefore excluded these countries from the fit andfound a more plausible curve as a result. In addition, we implemented a ceiling value for the immigration (admission) rate into the model with GDP per capita at unit scale (Figure 9). Such a ceiling or upper limit introduces some degree of arbitrariness into the model. It is, however, preventing the model from producing unreasonable results. In any case, such ceilings should be verified by careful analysis. It may be useful to also allow for separate immigration models for certain groups of countries (not done here). The logistic immigration model depicted in Figure 9 is represented by a single logistic function that appears sufficient to model the immigration flow depending on GDP per capita (see Table 2). The parameters of the logistic model are fitted to GDP per capita at unit scale but shown in log scale in Figure 9. Table 2: Parameters of the Logistic Immigration Model Parameters Value k1 20 Delta_t1 60,000 tm1 36,000 12 5. MIGRATION AND EMPLOYMENT In Section 4 we showed that GDP per capita is associated with certain trends in corresponding migration flow rates. The labor market, as an integral part of the economy, should also exert an impact on migration, both emigration and immigration. In this section, we first discuss some trends in labor force participation vis-à-vis future demographic trends, and then hypothesize how these may inform the development of alternative migration scenarios. A country’s level of economic development is associated with the proportion of people engaged in work. This is obvious, though the relationship between the two factors is not clear-cut. Plotting employment - here expressed as the employment to (working age) population ratio - against GDP per capita (Figure 10) for 2010–14 (single years), we see that at lower levels of economic development, the proportion of people employed initially declines amid large variations in employment, but then rises to almost 80 percent at very high levels of GDP per capita. The latter may be an artefact, caused by relatively small countries/territories and city-states with a high proportion of foreign workers (Bahrain, Brunei Darussalam, Hong Kong, Luxembourg, Kuwait, Macao,Norway,Qatar, Singapore, Switzerland, and the United Arab Emirates). Figure 10: Employment and GDP, 2010–14 Note: Function npregfast, bandwidth = 1.0 (determined by cross validation), confidence interval 95%, n = 835. Given the nonlinear relationship between employment and economic performance, is it possible to link employment to international migration directly? Is there an empirically verifiable association between employment and migration (emigration)? When plotting the share of employed persons against the crude emigration rate, no clear tendency appears (Figure 11). For almost the whole range of employment/population ratio, the fitted regression line is virtually a parallel to the x-axis, except for the tails with fewer observations. This preliminary evidence suggests that the relationship of employment to international migration is complex and appears not to allow for an easy analytical expression needed for the formulation of migration projection scenarios. We will therefore concentrate the following discussion on a comparison between select labor market scenarios and our international migration projections developed in Section6. We still argue that the challenges of maintaining and expanding the number of 13 jobs in fragile economies in the context of rapid population growth, especially in Sub-Saharan Africa, are significant. These challenges could largely determine future migratory movements. Figure 11: Employment and Emigration, 2010–14 Note: Function npregfast, bandwidth = 0.62 (determined by cross validation), confidence interval 95%, n = 158. This paper develops scenarios for the growth or decline ofthe absolute size of the economically active population and the corresponding labor force participation rates. The calculations are performed for all countries.10 The results are then subsequently aggregated into common geographic or economic groupings (put forward by the World Bank and United Nations) for the reader’s convenience. For this projection exercise, the ILO's projections of future Labour Force Participation Rates are employed. Scenarios: 1. Labor force trends are estimated using ILO’s projected country-specific LFPRsfor the years until 2030; for the remaining time until the end of the century the rates are held constant. The LFPRs are multiplied by the population aged 15–64 from the medium variant of the 2017 Revision of the UNDP’s projections, both sexes combined, up to 2100. The assumed constant LFPRs after 2030 and the moderate changes modeled by ILO before 2030 give the underlying demographic dynamics priority. 2. Constant absolute labor force is then compared with the growth of that portion of the population that is at a productive age. Such a scenario assumes a constant economy vis-à-vis a changing population. This scenario is expressed using a fictive LFPR that would occur given the labor force of 2015 vis-à-vis demographic change through 2100. Though the EPR used earlier in the modelling exercise issupposedly a more concrete expression of an economy’s capacity to provide jobs, it was not available in ILO's projections. However, the differences between the usual LFPR and the EPR appear to be small (see Table A.7in AppendixBfor a comparison). 10 Some smaller countries/territories are omitted, because they were not included in ILO’s modeling. 14 5.1. Projected labor force Under the assumption of ILO’sLFPR projected until 2030 and held constant thereafter,11 combined with the demographic dynamics predicted by the United Nations, the geographic regions of the world exhibit strikingly different trends in terms of the implied size of their future labor force (Table 3). Table 3:Projected Size of Labor Force, by World Bank Geographic Region, 2015–2100 Region 2015 2030 2050 2100 Labor Force (15-64) East Asia & Pacific 1,186,922,785 1,154,498,499 1,047,916,294 805,820,018 Europe & Central Asia 425,654,091 410,967,093 382,974,704 337,360,724 Latin America & Caribbean 290,388,788 338,582,724 348,385,860 272,138,147 Middle East & North Africa 141,533,336 176,726,832 208,745,554 231,573,031 North America 170,554,489 174,902,295 189,514,725 201,206,219 South Asia 643,064,494 782,387,709 875,119,472 724,874,579 Sub-Saharan Africa 373,934,935 590,670,498 970,148,716 1,856,966,549 While some geographic regions exhibit a strong increase in their labor force (Sub-Saharan Africa), other show only little change (South Asia) or even a decrease over the period (Europe, East Asia and Pacific; Table 4, Table 5). The most dramatic increase in the labor force is expected to happen in Sub-Saharan Africa, where the size of the potentially economically active population would more than double by 2050, and even grow fivefold by the end of the 21st century. At the same time, Europe and Central Asia would see a significant decline. Table 4:Absolute Change in the Labor Force, by World Bank Geographic Region, 2015–2100 Region 2015 2030 2050 2100 Labor Force Change (absolute) East Asia & Pacific 0 -32,424,285 -139,006,491 -381,102,767 Europe & Central Asia 0 -14,686,998 -42,679,387 -88,293,368 Latin America & Caribbean 0 48,193,937 57,997,073 -18,250,641 Middle East & North Africa 0 35,193,496 67,212,218 90,039,695 North America 0 4,347,805 18,960,236 30,651,730 South Asia 0 139,323,215 232,054,978 81,810,085 Sub-Saharan Africa 0 216,735,563 596,213,781 1,483,031,614 Table 5: Relative Change in the Labor Force (%), by World Bank Geographic Region, 2015–2100 Region 2015 2030 2050 2100 Employment Change (2015=100, %) East Asia & Pacific 100 97 88 68 Europe & Central Asia 100 97 90 80 Latin America & Caribbean 100 117 120 94 Middle East & North Africa 100 125 147 163 North America 100 103 111 118 South Asia 100 122 136 113 Sub-Saharan Africa 100 158 261 502 11 Had we kept LFPRs (averaged over 2005–15) constant across the entire projection horizon, the global labor force would have been about 4 percent smaller when compared to our hybrid approach that included the ILO projection until 2030. 15 Thus, demographic changes are forecast to drive an unprecedented change in the geographic distribution of the global labor force (Table 6). Currently, the largest portion (38 percent) of the global labor force resides in the East Asia and Pacific region. By 2050, the share of the East Asia and Pacific region is projected to decline to 26 percent, almost on par with a rapidly increasing share in Sub-Saharan Africa (24 percent). By 2100, Sub-Saharan Africa would be home to 42 percent of the global labor force, an even larger proportion than the currently leading East Asia and Pacific region. Almost all other regions in this geographic classification would see a decline in their share of the global labor force, except for the Middle East and North Africa (MENA) region. While South Asia is projected to experience a growing labor force, its share would decrease by 2100. Table 6:Labor Force Distribution, by World Bank Geographic Region, 2015–2100 Region 2015 2030 2050 2100 Labor Force Share (15-64) (in %) East Asia & Pacific 36.7 31.8 26.0 18.2 Europe & Central Asia 13.2 11.3 9.5 7.6 Latin America & Caribbean 9.0 9.3 8.7 6.1 Middle East & North Africa 4.4 4.9 5.2 5.2 North America 5.3 4.8 4.7 4.5 South Asia 19.9 21.6 21.8 16.4 Sub-Saharan Africa 11.6 16.3 24.1 41.9 World 100 100 100 100 The consequences of such a dramatic reorganization of the global supply of labor are not easy to anticipate and build into a model. How will the global economy deal with such a dramatic change? It may be comparatively easier to adjust to a shrinking volume and share of the labor force, as projected for East Asia and Pacific, Europe and Central Asia, and Latin America and the Caribbean. Technological progress could be the answer. Another option would be to fill any gaps that open with immigrants. For Sub-Saharan Africa, the challenges seem rather formidable, even unsurmountable (see subsection5.2). It would require the region, possibly, to become the next “work bank for the world,” in combination with increased emigration into other regions of the world. It seems impossible to lessen the need for additional jobs by migration alone, however. It is hard to say how much of the pressure to employ ever-larger cohorts in the African labor force would result in increased emigration to other continents. 5.2. Constant labor force Another way to illustrate the impact of demographic change on prospective labor markets is to assume a stagnant economy, here expressed as a constant volume of the labor force. Table 7 shows how the LFPR would evolve under such a scenario. Table 7: Implied LFPR with Constant Labor Force (%), by World Bank Geographic Region,2015–2100 Region 2015 2030 2050 2100 LFPR with Constant Labor Force (%) East Asia & Pacific 75 74 81 106 Europe & Central Asia 71 73 78 90 Latin America & Caribbean 69 60 59 75 Middle East & North Africa 51 40 34 31 North America 72 70 65 61 South Asia 57 47 42 50 16 Sub-Saharan Africa 69 44 27 14 The assumption of a constant labor force can lead to seemingly implausible results: In the case of East Asia and Pacific, the current size of the labor force is larger than the forecasted working-age population in 2100, resulting in LFPRs greater than 100 percent. Declining populations, such as in this region, could imply a need to actively seek immigration, to expand the participation of the native population, or to reduce the labor force. The assumption of a constant labor force does, on the other hand, show how dramatic the impact of fast-growing populations can be. In Sub-Saharan Africa, the LFPR would fall from its current average of 69 percent to less than half in 2050 (27 percent) and even to 14 percent by the end of the century. 5.3. Does the labor market drive international migration? We find insufficient evidence for the formal integration of employment dynamics into the formulation of assumptions of international migration. Clearly, more empirical and theoretical work is urgently needed. This is important precisely because the labor market plays such a dominant role in the past and current narratives of international migration. We have shown that the demographic dynamics in certain world regions, combined with expected changes in levels of economic performance, may result in large quantitative increases in the number of migrants originating in, for example, Sub-Saharan Africa. The increase in the number of migrants originating in Sub-Saharan Africa might even be larger if the economies in that region fail to provide enough employment for the fast-growing working-age populations. We have already shown that demographically driven changes in the absolute labor force in some regions may indicate a need to replenish the labor force by several means, immigration being one of them. We offer some preliminary thoughts on how to get a clearer picture of possible future emigration- immigration interactions in Section 6, below. 17 6. MIGRATION SCENARIOS This section deals with the applicability of the migration transition models outlined in this paper. We address the challenge of transforming a model of a general tendency (e.g., the bi-logistic models) into a method suited to the diversity and variability of the global migration experience and propose a very simple method for adapting the model to individual cases (countries). We also discuss possible variations of that approach. We select the migration transition model for emigrant flows, expressed in terms of the crude emigration rate, and then transform this into projected total migrants using the latest UNDP population projections. We choose to explore only the emigration model by combining economic (GDP) and demographic projections and projected crude emigration rates and total emigrants for all countries. This example of the migration transition theory is contrasted with a reference scenario of constant emigration rates. The inclusion of the flip side of emigration movements – immigration – by using the model proposed in Section 4 is beyond the scope of this paper. A practical implementation would have to follow similar steps as outlined below for the emigration case. We suggest, however, that in a future exercise both emigration and immigration projections using the models presented in this paper should be simultaneously projected and then combined into consolidated projections of international migration flows. We have proposed, in a previous paper, a mechanism to implement a consolidation between emigration potential and immigration admission by means of suitable transfer functions (Buettner and Muenz 2018b, 2018a). As could be expected, we found, in Section 5, evidence of a (nonlinear) relationship between GDP and employment, but no or very little association between employment and emigration. Integrating the employment dimension into our migration projection model (as part of a demographic projection) seems to be a challenge. We take some assurance from our simple labor force projections (Section 5) that a combination with demographic projection is feasible (see subsection 6.2, below). 6.1. Toward alternative KNOMAD migration projections The migration transition scenario The migration transition hypothesis (the “migration hump”) implies a nonlinear relationship between migration intensity (here, crude flow rates) and changes in the economic performance of countries. In our model, emigration flows are simultaneously affected by the underlying population dynamics and the level of economic performance. As will be shown, demographic and economic trends may reinforce one another, but also may act in opposite directions. The projection of international migrants (emigrants) is organized in three distinct steps. 1. The future crude emigration rates (CEMR) for each quinquennial projection period are calculated by inserting the projected GDP per capita data (see subsection 2.2) into the bi-logistic model for the emigration transition model (see subsection 4.1). The results are CEMR for each country and period that express the average trend, or central tendency, of the emigration transition model relative to the country’s GDP per capita. This initial CEMR estimate is not suitable for projecting total emigrants because it disregards the countries’ actual base level of CEMR. Instead of using the initial CEMR directly, we use the force of change implied in the average migration transition model, implemented as a simple scaling factor (see below). 18 2. The CEMR obtained in step 1 is scaled upwards or downwards by using a scaling factor. This scaling factor – calculated by dividing the CEMR of the base period by the modeled CEMR – expresses the degree to which the model is adjusted (upward or downward). In a first scenario, the scaling factor is kept constant throughout the projection period. This procedure keeps the overall shape of the migration transition model but moves the curve parallel to the abscissa or x-axis. 3. In a next step, total emigrants for each projection period are calculated by multiplying the CEMR obtained in step 2 by the person-years for each projection period. When setting the crude emigration rates for each country as constant, we see the full effect of demographic change in isolation. In the migration transition model, both forces act together. This is shown in Figure 12. Since at the global level emigration and immigration must be equal, the figure depicts the total amount of flows for each quinquennial projection period. Migration flows in both scenarios show a clear upward trend. Assuming constant migration rates, this trend is slowing somewhat at the end of this century (due to a slowing of population growth) and reaches 77 million migrants over the period 2095–2100, an increase of 171 percent (Table A.1, AppendixA). For a projection of future migration flows that combine population dynamics and economic change, the volume of migration flows is clearly larger than in the other case but is peaking at about 2070. This is attributable to the downward swing of the migration model once a certain level of economic performance is reached and passed. Figure 12: Total Global Migrants, by Scenario and by Five-Year Period, 2005–2100 (in millions) In sum, the projections suggest that between 2015 and 2100, a total of 1.3 billion (in the transition model) or 1.1 billion (constant migration rate) moves could occur. This is not a small amount compared with the demographic components of births and deaths: the 2017 Revision UNPD projects a total accumulated sum of 11.3 billion births and 8 billion deaths for the period 2005–2100. 19 Among these total moves, a clear majority of migration is expected to originate in middle-income countries. Meanwhile, migrants originating in low-income countries are projected to increase significantly in size and share (Figure 13). Figure 13:Absolute and Relative Size of Emigrants12, by World Bank Income Group, 2005–2100 The projection results also indicate significant changes in the geographic distribution of migrants. While the share of migrants originating in Sub-Saharan Africa is currently about 11 percent, it is projected to rise to 48 percent. All other regions will reduce their share of the total amount of migration during the projection period (see Table A.2 and Table A.3 in AppendixA). Figure 14:Absolute and Relative Size of Emigrants13, by World Bank Geographic Region, 2005–2100 12 By country of origin 13 By country of origin 20 To illustrate the dynamics in Sub-Saharan Africa, we look at the migration projection for two fast- growing countries: Nigeria and Niger. Nigeria’s total population is projected to grow from currently 181 million (2015) to a staggering 794 million by the end of the century, a more than fourfold increase. Niger, with about 20 million people in 2015, much smaller than Nigeria, is projected to reach 192 million in 2100, an almost tenfold increase. Our projection of migration originating in those two countries reflects the effect of demographic force in the constant rate projections (Figure 15). Figure 15: Total Emigration from Nigeria and Niger, Constant Scenario (2005=100), 2005–2100 Combining the demographic change with economic growth from the GDP projections, a different perspective emerges. For Nigeria and Niger, the inclusion of economic growth has a different impact. Until the middle of the 21st century, the number of migrants originating from Nigeria is higher than in a constant (demographic growth only) scenario. After 2055, the increased GDP per capita is decelerating and thus the number of migrants from Nigeria is decreasing. At the end of the 21st century, there is a sizable gap – about 500,000 persons over a period of five years – between the potential number of Nigerian migrants according to the demographic model and the lower number according to the transition model. 21 Figure 16: Growth of Total Emigrants, Nigeria, and Niger, by Scenario, 2005–2100 The case for neighboring Niger is quite different. Here, both forces – demographic change and economic growth – push in the same direction. The transition model for Niger produces a total number of migrants that is about 2.5 times larger than the number projected in the demographic model, and even larger than those projected for Nigeria in the transition model. Scenario with labor force extension We suggest that it be possible to perform projections as outlined in the preceding subsection with an extension that includes the labor force. Our proposed extended projection model tries to avoid the challenges of extreme complexity by performing the demographic projections in two steps instead of fully integrating the labor force dimension into the state space of the multistate projection model. We recall the demographic projection model used in a previous paper (Buettner and Muenz 2018b). It consisted of populations by country, age, and sex at the base year, plus age patterns of female fertility, age patterns of male and female mortality, and age patterns of male and female migration between countries of origin and destination for all projection periods. This model also included a transfer function used for consolidating emigration and immigration. This model could be amended to include additional nondemographic factors that act as relative weights of attraction and repulsion (Schoen 2006, p. 190). It is theoretically possible to add more dimensions to this demographic projection model, such as employment, or even employment by educational level. Even if the LFPR were only considered as an additional dimension, it would be necessary to have fertility, mortality, and migration indicators by labor force status. This may be possible for a limited number of countries with strong statistical institutions, but doubtful for a global exercise. We therefore suggest a hybrid or tandem approach that limits complexity by splitting the projection into two steps. In a first step, a demographic projection is performed that includes international migration as 22 a true interaction, driven by demographic change and the level of economic performance (see Section 4). In a second step and using LFPR projections (see Section 5), the population by age and sex calculated in each projection step would be distributed into the labor force (by age and sex) and those outside the labor force (by age and sex). The result would be populations by age, sex, and labor force status. The possibilities of specifying scenarios that address other dimensions of interest are not exhausted by our suggestions. It seems interesting, for example, to add migration status instead of employment status to the model. Such a model would project not only migration flows but also migrant stocks. 7. DISCUSSION In this paper, we have approached the projection of international migration in a novel way. Inspired by recent theoretical and empirical advancements in the analysis and conceptualization of international migration (Clemens 2014, 2017; de Haas 2009; de Haas, Vargas-Silva, and Vezzoli 2010; IOM2018), we propose functional migration models driven by the level of economic performance of each country, in addition to their demographic trends. In line with the theoretical postulates of the migration transition (or “migration hump”) hypothesis, our migration models result in a migration trend with two phases. In an initial phase, increasing GDP per capita adds, in de Haas’s terms, migration capabilities to those with migration aspirations (de Haas 2010b). Once a certain level of economic development is reached, aspiration declines and capabilities to leave are offset by opportunities at home. Migration propensities become nonlinear, and the results become more realistic than in linear models. We have implemented in this paper the emigration model alone and discussed results for the world’s countries and selected country groupings up to the end of the 21st century. For the time being, in this model, the migratory moves are linked to the originating country and ignore changes in the potential destinations of these migratory movements. We have also tried to explore possibilities of informing international migration models with data on labor market performance, expressed as the proportion and growth of the labor force. Here, although we did not find a clear empirical formulation for that link, we believe it is an important factor shaping migration flows. We did show the extremely varied trends of employment across countries and groups of countries that result from demographic change. We have identified Africa, and particularly the Sub- Saharan region, as the most challenging region of the world. Although a formal link to expected migratory moves remains elusive for now, the sheer amount of jobs that would be necessary in order to match rapidly growing populations and potential labor forces suggests a growing migration potential for that part of the world. One possibility to factor in labor market dynamics could be to relate the growth rates of the labor force in origin and destination countries (plus economic development). Demographically speaking, this would favor a flow from young populations (with a relatively large proportion of people in the labor force) to aging populations (with a declining proportion in the labor force). One way to formalize this would be to employ weighted harmonic averages for the consolidated emigration and admission(immigration) flows, affording higher weights to the countries with surplus labor, and less weight to countries in need of additional labor (i.e., countries with a stagnant or declining labor force). This paper has opened new avenues for improving international migration projections. But additional options remain to be explored. The most important of these is to combine and reconcile the emigration model with an immigration model into a comprehensive population projections model (Buettner and 23 Muenz 2018b, 2018a). This would complete the implementation of the migration models presented in this paper with the immigration portion and establish the international migration system as a model of interactions between countries, their economies, and population dynamics. Additional scenarios may also be explored in future work: • Migration intensities could be assumed to converge to the average level embodied by the migration model by adjusting the scaling factors. Such a convergence option may be warranted for long-term projections, associated with growing uncertainties over time. • The migration transition models have been determined based on empirical evidence for a certain period. Analysis of historical trends could reveal a drift in the relationship between migration intensity and economic performance. In the simplest implementation of such a scenario, the existing migration transition model would be shifted along the x-axis over time. • A major challenge for contemporary migration projection models is the virtual constancy of the spatial dimension of migration flows, e.g. established sending-receiving country relations remain unchanged. By including the immigration transition model, it may be possible to find a way to change the global distribution of migrants in a transparent and data-driven way.14 According to theoretical considerations and empirical findings, growing economic performance at first increases the share of population with means to emigrate, but in the second place also increases the potential of countries to absorb international migrants coming from other countries. As economic performance develops differently across countries, such differences could be used to explain changing migration distribution patterns. We are aware that the proposed migration models, even with their extensions, can only contributefirst insights but not paint a full picture of the future. In this paper, and in the models, international migration is treated as experienced by homogenous groups of people. This is, of course, a stark simplification. The various types of international migrants would warrant more differentiated approaches to establish valid projection models. The more general approach in this paper is therefore acontribution to a more complex model, allowing for better forecasts of both migration patterns and changes in population size. 14 Clemens discussed the need to foster new emigration destinations, as growing concentrations of international migrants may find traditional receiving countries insufficiently prepared to manage their rising numbers (Clemens 2017, p. 13). 24 8. BIBLIOGRAPHY Abel, G. J. 2009. International Migration Flow Table Estimation (PhD). University of Southampton, Southampton. Abel, G. J. 2013. “Estimating Global Migration Flow Tables using Place of Birth Data.”Demographic Research28 (March): 505–546. https://doi.org/10.4054/DemRes.2013.28.18 Abel, G. J. 2017. “Estimates of Global Bilateral Migration Flows by Gender between 1960 and 2015.”International Migration Review, 1–44. https://doi.org/10.1111/imre.12327 Buettner, T., and R. Muenz. 2018a. “International Migration Projections: Methodology Brief.” (KNOMAD Working Paper No. 30 Annex. World Bank, Washington DC. Buettner, T., and R. Muenz. 2018b. “Modeling Alternative Projections of International Migration.” (KNOMAD Working Paper No. 30). World Bank, Washington DC. Clemens, M. A. 2014. “Does Development Reduce Migration?” In R. E. B. Lucas (ed.), International Handbook on Migration and Economic Development (pp. 152–185). Edward Elgar Publishing. https://doi.org/10.4337/9781782548072.00010 Clemens, M. A. 2017. “Migration Is a Form of Development: The Need for Innovation to Regulate Migration for Mutual Benefit.” Population Division Technical Paper No. 2017/8. UNDP, New York. de Haas, H. 2009. Mobility and Human Development. Human Development. (Human Development Research Paper 2009/01). New York: UNDP de Haas, H. 2010a. “Migration and Development: A Theoretical Perspective.”International Migration Review, 44 (1), 227–264. https://doi.org/10.1111/j.1747-7379.2009.00804.x de Haas, H. 2010b. “Migration Transitions: A Theoretical and Empirical Inquiry into the Developmental Drivers of International Migration.” (International Migration Institute Working Papers No. 24). Oxford de Haas, H., Czaika, M., Flahaux, M.-L., Mahendra, E., Natter, K., Vezzoli, S., and Villares-Varela, M. 2018. “International Migration. Trends, determinants and policy.” (International Migration Institute Working Papers No. 142). Oxford de Haas, H., Vargas-Silva, C., and Vezzoli, S. 2010. “Global Migration Futures. A conceptual and methodological framework for research and analysis.” (International Migration Institute, University of Oxford). Oxford Dellink, R., Chateau, J., Lanzi, E., and Magné, B. 2017. “Long-term Economic Growth Projections in the Shared Socioeconomic Pathways.”Global Environmental Change, 42, 200–214. https://doi.org/10.1016/j.gloenvcha.2015.06.004 Feenstra, R. C., Inklaar, R., and Timmer, M. P. 2015. “The Next Generation of the Penn World Table.”American Economic Review, 105(10), 3150–3182. https://doi.org/10.1257/aer.20130954 Fisher, J. C., and Pry, R. H. 1971. “A Simple Substitution Model of Technological Change.”Technological Forecasting and Social Change, 3, 75–88. https://doi.org/10.1016/S0040-1625(71)80005-7 Grübler, A. 1998. Technology and Global Change. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9781316036471 25 ILO. 2017. ILO Labour Force Estimates and Projections: 1990-2030 (2017 Edition). Methodological description. Geneva. ILO. 2018. Labour market projections - ILO modelled estimates, May 2018. ILO.http://www.ilo.org/ilostat/faces/oracle/webcenter/portalapp/pagehierarchy/Page3.jspx?MBI_ ID=27 IOM. 2018. World Migration Report 2018. (M. McAuliffe and M. Ruhs, eds.), World Migration Report. Geneva: International Organization for Migration. Natale, F., Migali, S., and Münz, R. 2018. Many more to come? Migration from and within Africa.” Science for Policy. Joint Research Centre. Luxembourg: Publications Office of the European Union. http://publications.jrc.ec.europa.eu/repository/handle/JRC110703 Lutz, W., Butz, W. P., and KC, S. 2014a. World Population and Human Capital in the Twenty-First Century. (W. Lutz, W. P. Butz, and S. KC, eds.). Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198703167.001.0001 Lutz, W., Butz, W. P., and KC, S. (eds.). 2014b. World Population and Human Capital in the Twenty-First Century. Oxford University Press. Marchetti, C. 1997. Longevity and Life Expectancy. Technological Forecasting and Social Change, 55 (3), 281–299. https://doi.org/10.1016/S0040-1625(97)00024-3. Marchetti, C., Meyer, P. S., and Ausubel, J. H. 1996. “Human Population Dynamics Revisited with the Logistic Model: How Much Can Be Modeled and Predicted?”Technological Forecasting and Social Change, 30, 1–30. Meyer, P. S., Yung, J. W., and Ausubel, J. H. 1999. “A Primer on Logistic Growth and Substitution: The Mathematics of the Loglet Lab Software.”Technological Forecasting and Social Change, 61 (3), 247–271. https://doi.org/10.1016/S0040-1625(99)00021-9 OECD. 2016. Perspectives on Global Development 2017: International Migration in a Shifting World. Paris: OECD Publishing. https://doi.org/10.1787/persp_glob_dev-2017-en Özden, Ç., Parsons, C., Schiff, M., and Walmsley, T. L. 2011. “Where on Earth is Everybody? The Evolution of Global Bilateral Migration 1960 – 2000.”The World Bank Economic Review, 25 (1), 12– 56. R Core Team. 2017. R: A Language and Environment for Statistical Computing. Vienna, Austria. Riahi, K., Grübler, A., and Nakicenovic, N. 2007. “Scenarios of Long-Term Socio-economic and Environmental Development under Climate Stabilization.”Technological Forecasting and Social Change, 74, 887–935. https://doi.org/10.1016/j.techfore.2006.05.026 Schoen, R. 2006. Dynamic Population Models. Dordrecht: Springer Netherlands. https://doi.org/10.1007/1-4020-5230-8 Sestelo, M., Villanueva, N. M., Meira-Machado, L., and Roca-Pardiñas, J. 2017. “npregfast : An R Package for Nonparametric Estimation and Inference in Life Sciences.”Journal of Statistical Software, 82 (12), 1–27. https://doi.org/10.18637/jss.v082.i12 Skeldon, R. 2012. “Migration Transitions Revisited: Their Continued Relevance for The Development of Migration Theory.”Population, Space and Place, 18 (2), 154–166. https://doi.org/10.1002/psp.667 26 United Nations. 2015. World Population Prospects: The 2015 Revision, DVD Edition. New York: Department of Economic and Social Affairs, Population Division. United Nations. 2017a. Trends in International Migrant Stock: The 2017 Revision. Department of Economic and Social Affairs, Population Division. United Nations. 2017b. World Population Prospects: The 2017 Revision, DVD Edition. New York: Department of Economic and Social Affairs, Population Division. 27 Appendix A. Tables Table A.1: Total Migratory Movements, by Scenario and Five-Year Period, 2005–210015 Period Transition model Const. migration rate Total migrant moves 2005-2010 44,842,398 44,842,398 2010-2015 49,882,245 47,786,280 2015-2020 54,638,442 50,825,137 2020-2025 59,309,809 53,790,703 2025-2030 63,802,436 56,627,276 2030-2035 67,985,664 59,302,231 2035-2040 71,821,925 61,805,551 2040-2045 75,305,832 64,132,850 2045-2050 78,318,699 66,268,494 2050-2055 80,736,904 68,193,311 2055-2060 82,478,539 69,899,858 2060-2065 83,551,681 71,397,132 2065-2070 84,040,702 72,702,165 2070-2075 84,063,477 73,825,383 2075-2080 83,754,857 74,771,386 2080-2085 83,242,487 75,548,756 2085-2090 82,629,323 76,173,391 2090-2095 81,982,597 76,659,445 2095-2100 81,340,612 77,015,188 Table A.2: Total Emigrants, by World Bank Region, Selected Five-Year Period, 2005–2100 Region 2005-2010 2015-2020 2025-2030 2045-2050 2095-2100 Total emigrants East Asia & Pacific 11,264,861 12,925,141 13,547,930 13,411,915 10,721,830 Europe & Central Asia 7,388,612 7,337,928 7,164,885 6,787,819 5,982,332 Latin America & Caribbean 3,959,734 4,229,163 4,433,539 4,448,130 3,530,104 Middle East & North Africa 3,684,794 4,451,575 4,927,928 5,552,890 6,030,314 North America 1,889,896 2,039,400 2,174,079 2,396,694 2,763,460 South Asia 11,763,552 16,049,753 19,067,467 19,966,623 13,106,738 Sub-Saharan Africa 4,890,949 7,605,481 12,486,607 25,754,629 39,205,835 Table A.3: Share of Migrants, by World Bank Region, Selected Five-Year Period, 2005–2100 Region 2005-2010 2015-2020 2025-2030 2045-2050 2095-2100 Regional Distribution of Migrants (%) East Asia & Pacific 25 24 21 17 13 Europe & Central Asia 16 13 11 9 7 Latin America & Caribbean 9 8 7 6 4 Middle East & North Africa 8 8 8 7 7 North America 4 4 3 3 3 South Asia 26 29 30 25 16 Sub-Saharan Africa 11 14 20 33 48 World 100 100 100 100 100 15 On the world level, emigrants must equal immigrants and thus are referred here to as migratory moves. 28 Table A.4: Total Emigrants, by World Bank Income Group, Selected Five-Year Period, 2005–2100 Group 2005-2010 2015-2020 2025-2030 2045-2050 2095-2100 Total emigrants High income 8,205,217 8,714,824 9,011,285 9,390,160 9,650,626 Middle income 32,844,686 39,673,593 44,109,234 46,066,564 39,086,084 Low income 3,792,494 6,250,025 10,681,916 22,861,975 32,603,903 Table A.5: Total Emigrants, by UN Region, Selected Five-Year Period, 2005–2100 Region 2005-2010 2015-2020 2025-2030 2045-2050 2095-2100 Total emigrants Africa 6,323,429 9,239,945 14,203,453 27,480,425 40,765,060 Asia 26,324,723 32,845,966 36,822,177 38,052,369 28,785,913 Europe 5,777,461 5,609,151 5,411,722 5,072,683 4,526,924 Latin America 3,959,734 4,229,163 4,433,539 4,448,130 3,530,104 Northern America 1,889,896 2,039,400 2,174,079 2,396,694 2,763,460 Oceania 567,154 674,817 757,465 868,399 969,153 Table A.6: Total Emigrants, by UN Development Group, Selected Five-Year Period, 2005–2100 Group 2005-2010 2015-2020 2025-2030 2045-2050 2095-2100 Total emigrants Least developed countries 10,175,036 16,133,595 23,228,377 35,310,689 39,471,378 More developed regions 8,112,470 8,144,701 8,122,176 8,072,641 8,007,490 Other developed regions 26,554,892 30,360,146 32,451,883 34,935,368 33,861,744 Table A.7: Average Labor Force Participation Rate, by World Bank Region, 2005–15 Region Labor force participation rates, 2005–15 (%) Total labor force Employment based East Asia & Pacific 75 72 Europe & Central Asia 71 65 Latin America & Caribbean 69 64 Middle East & North Africa 51 46 North America 72 67 South Asia 57 55 Sub-Saharan Africa 69 64 29 AppendixB. Figures Figure B.1: Components of the Bi-Logistic Emigration Flow Model Figure B.2: Geographic Labour Force Distribution, Absolute and Relative, 2015–2100 30 Appendix C. Methodology C.1 Bi-logistic migration model For the parameterization of estimated migration transition curves, we employed the class of logistic functions as flexible and coherent mathematical models. The curves obtained from the kernel regression needed to be transformed into a parameterized form, for which logistic functions are well suited. A logistic function exhibits an S-shape and describes a diffusion process growing from an initial level to an upper or lower asymptote. The general form of a logistic can be expressed as: k P(t ) = (1) 1 + exp[− (t −  )] k Saturation level or asymptote of the diffusion process  Growth rate of the s-curve  Length of time the curve takes to reach the midpoint of the growth trajectory. We use a re-parameterized logistic function introduced by Fisher and Pry (1971) and extensively used for global change modeling (Grübler 1998; Marchetti 1997; Marchetti et al. 1996; Meyer et al. 1999; Riahi et al. 2007). Th Fisher/Pry logistic function has parameters that are easier to interpret and are better suited to complex fitting exercises:16 k P(t ) = (2) Ln(81) 1 + exp[− (t − tm )] t tm Midpoint of the growth/diffusion process t Duration for the growth process to proceed from 10 percent to 90 percent of the asymptote (k). However, most migration transition curves show trends with a clear trend reversal. Such processes may be effectively modeled by a combination of two logistic functions where one diffusion process approaches an upper asymptote, and a second and delayed process approaches a lower asymptote. Combined, these two processes represent trends with a reversal. k1 P (t ) = Ln(81) 1 + exp[− (t − tm1 )] t1 (3) k2 + Ln(81) 1 + exp[− (t − tm 2 )] t 2 The migration transition models are indexed by the level of GDP per capita, not time. Hence, the parameters tm and  t represent, in this setting, the midpoint of the GDP per capita growth/diffusion process, and the time that it takes for GDP per capita to grow from 10 percent to 90 percent of the asymptote (k), respectively. The time parameter t is replaced with GDP per capita (either in log-scale or unchanged). ln(81) 16 This function relates to the general form by substituting  = tm ; t = .  31 32 Page 33 of 38