How many people will likely move in the decades to come? And where will they come from and move to? Policymakers worldwide have a keen interest in these questions. While long-term developments in international migration patterns are relevant for the demography and economy of a country, sudden flows—for example, in the case of humanitarian emergencies—pose institutional challenges regarding reception capacities, health systems, housing, education, and training programs amongst others. This chapter reviews key concepts related to migration scenarios and forecasting. It outlines different qualitative and quantitative approaches, compares different studies, and discusses the potential use of various techniques for academic and policy audiences.Footnote 1

Although there is a general call to foresee and anticipate migration in order to develop adequate policy responses, migration patterns are notoriously difficult to predict. Not the least because of the range of different drivers of migration (at the individual and household level) as well as societal events (from war to climate disasters) that make people move across borders (see Chaps. 3 to 11 in this textbook). These often unforeseen global and local developments in socio-economic, environmental, and political terms can have a large impact on migration flows between and within countries and continents. Some events, such as conflicts and environmental disasters, may come as sudden ‘shocks’ that are hard to anticipate. Other geopolitical events like the extension of the European Union in the mid-2000s, were expected to affect migration, although the predictions of how it would do this were not always correct (see De Haas, 2011). Lastly, although some seemingly predictable, slow-onset developments such as climate change or digitalisation will increase, yet their likely effects on future migration are difficult to unpack. Despite the fact that, compared to other elements of population change (fertility and mortality), migration is hard to predict, several methods have been developed to help—at least to some extent—increase our understanding of different potential patterns of migration. These methods aim to describe potential volumes of people moving, but increasingly also include characteristics of those on the move. To understand the effectiveness of policy and the impact on societies, it is essential to also have estimates on the composition of flows, like, for example, by sex, education, or age.

In order to understand possible migration flowsrooted in events or change that one does not yet know will occur, the method of scenarios has been developed. Rather than forecasting, the aim of scenarios is to represent a ‘what if…’ plausible vision of the future and, as such, are thought experiments along different dimensions (Sohst et al., 2020). Migration scenarios are qualitative narratives about the future of migration that emphasise possible structural changes and their consequences for migration. In addition to scenarios, migration forecasts continue to serve as an important tool to estimate future migration flows. Even though one may not always be able to predict the unforeseeable causes of sudden migration flows, this does not mean that forecast models have no merit. These models are usually built on measurable causes of migration. The models are based on a reasonable assumption that migration flows generally follow certain rules: migrants have to come from somewhere in the world (origin countries) which means any predicted numbers should be constrained by population likely to migrate; migrants are often young and mostly driven by economic incentives to improve their lives and that of their family; instead of looking for new destinations, migrants tend to follow existing migration corridors, for instance by going to destinations where the same languages are spoken. Thus, among others, historical trends could be informative in forecasting models. What is less predictable, however, is how long migrants will stay, when and who returns, and to what extent are these decisions also influenced by new policies, economic conditions, or political changes among other factors. In fact, over the past decades forecast models have been developed that increasingly take into account the uncertainty in migration flow predictions (Bijak, 2011; Raymer & Wiśniowski, 2018).

In this chapter we describe, discuss, and illustrate different methods that have been developed to understand and predict future migration. We build on work conducted within the CrossMigration project—the project that established the Migration Research Hub (migrationresearch.com)—which had its primary focus on Europe. Although we acknowledge the work that has been done around the globe, we mainly use the European case for our examples here. We focus our review in this chapter on two different methods of providing information on future migration flows: scenarios and forecasts. Again, although we are aware that early warning systems have received ample (policy) attention in the past decade, it is beyond the scope of this work (for some essential reading on this, see OECD, 2016; Sardoschau, 2020). Below we will describe key concepts and underlying assumptions of scenarios and forecasts. We then proceed to explain the methodology in more detail and give examples of each method. Finally, we compare both methods, indicate how they complement each other, and then evaluate their use for predicting migration.

1 Scenarios

Scenario studies have a history in a range of other areas and disciplines before they were used and implemented for the study of migration. Its history in different practical work domains, shows that this method has been developed by practitioners and is now still dominantly used by practitioners and policymakers (Bradfield et al., 2005). When it comes to migration, policymakers are eager to know how flows develop as this may have implications on the future composition of the population in a country and the related impact on a range of policy domains from housing to education and health care. International organisations such as the International Organization for Migration, the Organisation for Economic Co-operation and Development (OECD) but also supranational unions, such as the European Union, are among those that develop migration scenarios for policy (see Sohst et al., 2020 for a list of studies). Also in the academic research on migration, scenario studies are more and more common (for an overview of types of scenario methods used in migration literature and reports, see Boissonneault et al. (2020)).

The word ‘scenario’ as it is used in the analyses of migration refers to story line(s), which can be developed based on different input. The main goal of the scenario method is to obtain a probable future view (scenario) along different dimensions that provides indications of the potential migration that may take place if that scenario would occur. A scenario study does not predict the actual migration flows but compares different versions of a possible future. It is not a failure if the scenario(s) never occur(s) as it serves to think through different versions of potential futures. A scenario should thus not be judged on whether it becomes a reality or not, but rather on whether it sufficiently takes into account potentially different future lines, their drivers, and their relation with the outcome variable (in our case, migration). They should cover different versions of a future to also set the extreme boundaries in which migrations develop (from low to high migration flows, for instance). Also, scenarios that seem rather unlikely when they are developed are interesting to be included in a study, as we have seen with the COVID-19 pandemic; scenarios that seemed improbable became reality in a very short period of time across the globe.

1.1 Methodological Approach

While there are different ways in which scenarios are developed, generally they follow a similar structure. The scenario method is a qualitative method in which storylines are developed through discourse (Sohst et al., 2020). This means that through conversations between people, often practitioners and/or experts, the scenarios are developed and refined further. The process is often participatory, meaning that a group of experts and/or practitioners come together to discuss scenarios. One comprehensive methodological foundation for the development of a participatory methodology comes from the Global Migration Futures project (Paoletti et al., 2010; International Migration Institute & University of Waikato, 2013; Vezzoli et al., 2017). It involves multiple steps and several iterations that alternate between desk research of the research team and participatory elements (workshops, online surveys, interviews). During these meetings the different stakeholders in the process share their views and expectations on the likelihood of a particular scenario to occur and the associated migration flows they would expect. Knowledge is exchanged and the participants update their expectations along the way. Although different scenarios can be developed side by side without an evaluation of which scenario is the most likely, often a final aim is to reach some consensus on the likelihood of the different scenarios and rank them as such. In this case, the purpose of the scenario study is to sensitise policymakers to different futures and the potentially different impacts that policy may have on migration.

Most scenario studies construct a number of macro-perspective scenarios to approach the future of migration. This means that, for instance, anticipated changes on the societal or global level in terms of economic, environmental, or political stability are influencing changes in migration flows. The approach assumes that individual motivations for migration are shaped by these macro-level changes. Scenarios do not necessarily only include current developments, but can also describe potential developments in the future. For example, if countries in the European Union would not see any relevant economic growth after 2025, while Asian countries like China and India maintain high and sustainable growth rates, then these Asian destinations may become more attractive for migrant workers due to better economic opportunities and geographic proximity in the future after 2025 (Acostamadiedo et al., 2020).

Often scenario studies make use of a scenario matrix that represents different dimensions expected to influence migration flows. Often the dimensions are based on what are considered the most uncertain and most impactful drivers of future migration. This results in a selection of two intersecting macro-level axes which create four quadrants that each correspond to one scenario (Fig. 28.1).Footnote 2 To give an example, the Global Migration Future project for the Pacific region built two main scenario matrices to envision future migration in the Pacific countries (International Migration Institute and University of Waikato, 2013). One scenario matrix chose (a) level of human development and (b) regional integration and cooperation as the axes. The other one built on (c) economic growth and (d) political stability. These four dimensions are on the one hand uncertain yet at the same time very important for the future of migration. Thus, structuring the narratives according to different versions of these dimensions will result in the most useful scenarios.

Fig. 28.1
An illustration of a matrix has the high and low end of dimension A along the y-axis and the high and low end of dimension B along the x-axis.

Typical scenario matrix building

Scenario studies are usually applied to develop views on the more distant future. They often seek alignment with time horizons of broader political processes, such as the UN’s 2030 Sustainable Development Agenda (see Sohst et al., 2020). As a result, most migration scenario studies cover time spans of 10 to 30 years into the future. Typical outputs of scenario studies include narratives of future migration patterns and the development of associated factors. For example, the OECD (2016) uses participatory methods to explore future patterns of international migration between 2017 and 2030. One of their scenarios is “Slower shifting wealth” which foresees business continuing as usual, economic convergence between OECD and non-OECD countries continuing but at a slower pace than in the last 15 years, and global co-operation and coordinated action becoming more difficult.

While scenario studies often follow the participatory method explained earlier, there are a few other approaches used in migration scenario research as well. We distinguish five which we list in Table 28.1 summarising the main methods, their features, strengths, and weaknesses (see also Sohst et al., 2020). First, in addition to starting from self-defined axes for the development of migration scenarios these can also be based on existing well-developed scenarios in related domains. Two such existing scenarios are for example the Intergovernmental Panel on Climate Change’s Special Report on Emission Scenarios and the Global Environmental Outlook of the UN Environment Programme. Both scenarios provide comprehensive narratives about the future state of the world considering a range of environmental and structural factors such as the degree of international cooperation, cultural shifts, population growth and technological advances (Intergovernmental Panel on Climate Change, 2000; United Nations Environment Programme, 2007). These factors can also potentially easily be translated in and used for migration scenarios.

Table 28.1 Methods used in migration scenarios (building on Sohst et al., 2020)

Second, scenarios can also be developed using and mixing both qualitative and quantitative approaches (e.g. Sohst et al. (forthcoming); Acostamadiedo et al., 2020). For instance, the European Spatial Planning Observation Network (ESPON) and Netherlands Interdisciplinary Demographic Institute (NIDI) report (2010) Demographic and Migratory Flows affecting European Regions first provides conventional demographic projections of the European population up to the year 2050. Then, in a second step, it applies a scenario framework to four individual components of their projections: mortality, fertility, migration, and labour force participation. This method provides a more comprehensive picture of possible future migration but requires a greater amount of resource and capacity to perform both scenario and other research.

Third, instead of generating scenarios from participatory activities, scenarios that were developed by Frontex (2016) combine information from experts with the use of computer software to generate all possible combination of scenarios. After this they selected the most consistent ones for their further exploration. Although this process stresses to be more objective considering different elements (and therefore may result in less bias), it still relies on a review and validation from experts. Fourth, another method is actor analysis, where the scenarios were developed by combining an analysis of the influencing factors with an analysis of shaping actors (European Asylum Support Office, 2019). This approach is based on the Causal Layered Analysis method used in futures research (Inayatullah, 1998).

Finally, scenario studies often use the Delphi method to either develop scenarios but also to evaluate scenarios. This method is similar to the participatory method (see above) but aims to reduce (cognitive) biases as much as possible (see Acostamadiedo et al., 2020). It focuses on iteratively improving one single estimate or understanding of future migration. Instead of having experts freely discuss, the participating experts are asked to fill in a survey on future migration patterns anonymously. After the experts have filled in the survey, the participants then receive feedback on each other’s answers after which the involved experts have the option to revise their initial responses to the survey taking the feedback they received into account (or not). This whole process can be repeated several times. The Delphi method became popular to understand potential migration from Eastern European countries to Western Europe (Lachmanova & Drbohlav, 2004; OECD, 2001, p. 26; Drbohlav, 1996, 1997). The Delphi method has been used in both migration scenario and in forecast studies, which will be discussed in the next section. At the same time, the reliability of expert views, also, or especially, in the field of migration, has been debated in the literature (for a further overview and discussion on this, see Acostamadiedo et al., 2020).

2 Forecasts

Migration forecasts, as opposed to scenarios, are quantitative assessments of future migration trends (Sohst et al., 2020). When speaking of forecasts many may think about weather forecasts, but population researchers have also used forecast methods for quite some time (Zipf, 1946). However, up until the 1990s, these were not so much geared towards migration specifically. This started to change with the policy discussions surrounding the EU enlargement (Booth, 2006) and other population dynamics including declining fertility and improved life expectancy in many OECD countries (OECD, 2019).

Forecast methods are used to produce numerical estimates of future migration. With the need to produce more accurate numbers, forecast models become increasingly sophisticated to cope with an unstable and non-linear future. As a result, academics are among those who lead the development and application of migration forecasting, with experts from international bodies (e.g. Eurostat, UN Population Division, the World Bank, and the OECD), national statistical offices, think tanks, and research institutes actively engaged (Sohst et al., 2020).

While there are many different types of forecast modelling, all of them share some common characteristics. First, any forecast model requires data input. Forecasts often use large data sources, such as administrative data, population census data, large scale surveys, and recently also cellphone and online data (e.g. Alexander et al., 2020; Tjaden et al., 2021). The majority of forecast models are building on past migration data and previous patterns, including annual migration flow or migrant stocks, to predict future migration levels. Second, each forecast requires some type of statistical modelling, although the kind of modelling is diverse across the different types of forecast methods (see below). Third, forecast models usually make a set of quantifiable assumptions on how past data can be applied and extrapolated to the future. For example, economic models often assume that individuals primarily base their migration decision on costs and benefits of migration (Bauer & Zimmermann, 1999). Gravity models assume other factors, such a geographical distances and network size of migrant group in destination country, shape migrants’ moves (Zipf, 1946; Arranz, 2019). Migration intention surveys assume that intention to migrate is a strong predictor for actual migration and thus can shed light on future migration (e.g. Tjaden et al., 2018).

In Table 28.2, we list seven types of methods used in migration forecast research. First, argument-based models are believed to be the most widely spread method for migration forecasts in official statistics and are usually treated as a component of population projections (Bijak, 2010). There are no strict rules about how to arrive at these assumptions but to illustrate a range of possible future outcomes. In most cases, these assumptions are produced in three or more variants labelled as “low”, “middle” and “high” scenarios. Even though these variants are often also described as ‘scenarios’ by these studies (Boissonneault et al., 2020), they do not provide a storyline and are a different method than what we described as the scenario method in the previous section. Argument-based models are easy to implement, but the uncertainty about future migration is obviously large. As a result, most often the “middle” variant is taken as the most likely projection, even though this is not necessarily the most likely given that the variants are not equipped with probabilities.

Table 28.2 Methods used in migration forecasts, (building on Sohst et al., 2020)

A second forecast method that is used are migration intention surveys, in which emigration intentions of the population are asked. It can be a valuable source for migration estimation in the absence of migration flow data (International Organization for Migration, 1998; Krieger, 2004; Tjaden et al., 2018). However, intentions do not necessarily materialise to future emigration flows and those who have already left a country are obviously not represented. Furthermore, surveys like these cannot do justice to the demand side of migration that may change over time and may make people move even when they originally did not express migration intentions.

Third, explanatory econometric models were originally used to verify economic theories about migration but have increasingly gained popularity for forecasting too (Sohst et al., 2020). It is characterised by the inclusion of drivers that researchers believe are related to migration. Frequently used variables include GDP or GDP per capita, income differentials, and labour market performance.

The fourth type of forecasting are those that are built around the spatial interaction model. The difference between spatial and econometric models is that spatial models see migration happening in a bilateral or multiregional system so that origins are connected with destinations (Zipf, 1946; Arranz, 2019). Depending on the type of spatial model, origin-destination interactions may refer to distances between origins and destinations (gravity model), shares of emigrants by destinations (generation-distribution model), and difference between expected and observed bilateral flows (multiplicative component model). Spatial interaction models always produce bilateral migration flow (or stock) estimates.

Fifth, time-series extrapolations are used and are solely reliant on past migration data to produce future estimates of migration. The Autoregressive Integrated Moving Average (ARIMA) model and its many variants are most frequently used (e.g. de Beer, 1993, 1997; Calian, 2013). This data-driven model is often used by statisticians and distinguishes from previous deterministic models by providing forecast uncertainty (i.e. predictive intervals) and has the possibility to be combined with other quantitative or qualitative data under a Bayesian framework (Bijak, 2011; Raymer & Wiśniowski, 2018). It, however, needs extensive and complete migration data which are often not at hand.

The sixth approach is a Bayesian framework in which expert opinions are incorporated with time-series models. Over the past decade this has been developed further building on time-series extrapolations but dealing with imperfect data by adding experts’ knowledge and expectations. Bayesian statistics understand probability as the degree of belief in an event. Bayesian studies do not use an experimental setting to establish the probability of a given event, instead, this method allows researchers to insert prior beliefs into their calculations. For example, Wiśniowski et al. (2013) documented how experts’ knowledge gathered in Delphi surveys can be translated into Bayesian priors to improve estimations of international migration in Europe. However, while it is a strong point that these models include uncertainty, questions remain on how to select experts and how meaningful the input of experts is when opinions between experts vary to a large degree. Bayesian approaches have gained popularity in migration forecasts because they offer innovate ways to deal with the large amount of uncertainty connected to forecasting, while also making use of different sources of information (i.e. expert opinion and migration data analysis). Bijak and Winiowski (2009) combined a time-series model with expert opinion solicited from a two-round Delphi survey in a Bayesian framework to predict 2010–2015 annual immigration flow to seven European countries. Their forecast model is one of the methodologically most advanced approaches currently available. It addresses and quantifies uncertainty, applies a statistical model with relatively few restrictions, and enhances the limited data with expert knowledge.

Lastly, we mention machine-learning models which have recently been applied to migration flows. To test the predictive power of this novel approach, Robinson and Dilkina (2018) use machine learning models to estimate 2004–2014 inter-county migration flows in the United States and 1960–2000 international migration flows. This method is still very much in development and has not yet been widely applied in migration forecasts.

To summarise, typical outputs of migration forecast are concrete numbers. These can be the expected future stock of immigrants in a given country, the expected migration flow from and to a country, or the expected flow of migrants from one country to another in a given year. Depending on who produces the numbers, the focus of forecast is either on migration as a component of population projections (in general more demographers’ interest) or on the size and impact of future migration flows on labour markets and welfare (often more economists’ interest). Different from scenarios, the time span of forecasts depends more on temporal coverage of available data and the aim of the work. The time frame for specific flows can vary between 1 year to more than 100 years. For instance, migration forecasts produced as a component of population projection usually stretch a few decades to 100 years into the future, whereas econometric models are usually (but not always) designed to predict migration futures for shorter time periods, such as 1 to 5 years in the future.

3 Summary and Evaluation of the Scenario and Forecast Approaches

Migration scenarios and forecasts are two approaches that equip researchers and policymakers with scientific tools to predict and understand the future of migration. In Table 28.3, we briefly summarise and compare the two approaches. First and foremost, scenarios approach the future of migration qualitatively while forecasts do so more quantitatively. Therefore, what we can get out of the two approaches is quite different: scenarios produce narrative storylines; forecasts produce numbers. The future is full of uncertainty and both approaches try to address that in different ways. The scenario approach produces “what if…” scenarios to better prepare for migration resulted from unexpected events: environmental changes, conflicts, and other types of crises. The goal is not to provide accurate predictions but sensitise participants as well as readers about future possibilities. Forecast studies, on the other hand, aim at producing numbers about future migration levels. The only certainty about the future, however, is uncertainty (Acostamadiedo & Tjaden, 2020). Many forecast studies do not address uncertainty (in detail), because the models used were not designed to quantify uncertainty. Recent developments see more sophisticated models (e.g. time-series model, Bayesian approach) being used in migration research on forecasts, to incorporate uncertainty measurements (see for example Bijak, 2011; Azose & Raftery, 2015; Disney et al., 2015).

Table 28.3 Comparison of scenario and forecast approaches in migration studies along different dimensions

Both methodological approaches have their advantaged and disadvantages. The scenario approach sensitises participants for various directions of potential long-term futures. To communicate lengthy and often highly abstract narrative outputs from scenario studies to non-participants, however, is rather difficult. The strength of forecast studies, lies in its aim to provide a probable prediction on future migration levels. Though the accuracy of forecasting relies on continuity of past migration trends, which may be unrealistic, the exercise provides a quantifiable base for users to understand future migration.

Evaluating the strengths and weaknesses of both approaches reveals their potential usefulness in different situations. The choice of methods depends on the purpose of research and the (data) resources. If one wants to get a most accurate number for future migration and has access to comprehensive, high-quality data on past migration trends, forecasting models are the potentially better option. If the goal is to engage an institution in an open discussion and challenge taken-for-granted assumptions, or to examine hard-to-measure factors like political stability, migration scenarios might be the right tool.

Reflecting on the purpose of the exercise, time coverage of the different methods usually varies. Due to its preparedness goal, scenario studies often align time horizons with key dates in broader political process, for instance UN 2030 Sustainable Development Goals. Forecasts on the number of asylum applicants can be shorter than a year as the flow of asylum seekers is particularly volatile and thus unpredictable. Migration forecasts as an integral part of population projections can go up to 100 years. However, errors of forecast models increase with forecast duration. To obtain a reasonable prediction, one should be conscious about limitations of the chosen forecast model we discussed in Sect. 28.2. For instance, to make the best use of an econometric model, the forecasts should be applied to short time horizons (around 5 years) and be interpreted with careful attention to changing contexts that might diminish validity of the chosen migration drivers. Overall scenarios are more useful when a long-term perspective is taken whereas forecast serve more short-term future visions and planning (see also Acostamadiedo et al., 2020).

While applying these tools, researchers need to be aware of the constraints and limitations of the methods. In current practice, the scenario method does not necessarily require (big) data input (even though it can use statistical input) and thus is suitable to produce and include different types of migration futures. The forecast model, however, is bounded by available data. As a result, seasonal or irregular migrants are usually not counted in forecasting models because they are not captured in most data sources.

On the other side, the scenario method relies heavily on participants and therefore potentially reflects bias from them (see Acostamadiedo & Tjaden, 2020, p. 22). Experts are subject to their own cognitive biases and are often not better at predicting the future than the average population as some research has suggested (Tetlock, 2017). The bias in forecast models more likely comes from imperfect data (e.g. undercount and different definitions of migrants). Scenario studies heavily depend on researchers’ assumptions on uncertainties (e.g. how to define the scenario matrix), while forecasts results are sensitive to the choice of data, assumptions and models. Thus, a transparent and detailed documentation is essential for both approaches.

While it is useful to learn from past examples on the usage of scenario and forecast methods, it is important for the researcher to realise in what context these methods were used. The scenario methods on migration we review and included here were limited to those developed mainly by practitioners in the European Union. Scenario studies are often used to understand the factors that drive migrants to move to Europe. However, realities on the impact of immigration and roles of migrants vary depending on the context of migration. Migration realities are very different across the globe and thus, it is important for researchers to be aware of the surrounding political, societal, environmental, health, and economic factors that shape migration. This should inform the decision-making process for choosing a certain method, but also on how these methods are applied and matched with local realities.

4 Conclusion and Future Developments

To understand and shed light on potential future migration, there needs to be an understanding of the different methods that can be used but also of what the limitations of the diverse methods are. We have discussed different approaches to understand future migration by describing scenario and forecast methods in more detail and evaluating the two. Depending on the goal of the study and the questions on the table, combined with the available resources and data, the appropriate method should be selected.

When using scenario and/or forecast methods, one should always be aware of the surrounding political, societal, environmental, health, and economic factors, the role and bias of the researcher, as well as how each of these factors may affect the results. Researchers may not be able to address all these influences, but it is good practice to raise awareness on these issues and make them explicit in the study design and findings.

As we have shown, both scenario and forecast methods have their unique strengths and weaknesses. For this reason, there are several likely avenues for the field to make advances in the coming years (see also Chap. 29). First, a possible development in studying future migration is to integrate these two methods further than has been done so far. Recently some steps were taken to indeed achieve this. A clear example is the Bayesian framework, where forecast models are improved by inserting expert knowledge solicited from Delphi surveys to assess the reliability of migration data and incorporate expectations on future migration trends. Instead of aiming for a more uniform prediction on migration, one could embrace the diversity of assessments on future migration. Then, potentially, a next step can be achieved in creating multiple forecast predictions which are based on different future migration scenarios that have been proposed in scenario discussions. This could serve as a way to incorporate uncertainty. Machine-learning techniques may then be used to further improve the accuracy of the predictions. Triangulating and cross-checking predictions from various studies is crucial to identify a convergence in the general tendency of results which can then be interpreted with more confidence.

Second, another development that is expected to develop further in the future is the continuous improvement of longitudinal migration data. Limited data availability and quality has often been lamented in the field of migration (Willekens et al., 2016; Raymer et al., 2019), however, recent advances in making harmonised migration data available at the regional level (e.g. Eurostat and ASEAN) provides the opportunity for time series forecasting on migration to become more accurate and reliable over time.

Third, the increasing accessibility of large-scale digital trace data (Cesare et al., 2018), such as Facebook and Google searches, provide great potential for the forecasting community, especially when innovative data sources can be combined with traditional ones (Carammia et al., 2022, Hsiao et al., 2020; Rampazzo et al., 2021; Tjaden et al., 2021).

In conclusion, combinations of scenario and forecast methods appear to be a promising avenue for research on future migration predictions. While these methods may never be able to accurately predict migration patterns, the insights that can be derived from the migration projections can still be valuable for practitioners and policymakers, helping them understand population change and design more accurate policies. Even though there are shortcomings to each method, forecasts and scenarios (or the combination of both) can provide valuable, though not complete insights into future migration. Even when we cannot fully cope with the uncertainties in future events, each method prepares us to at least partially understand future migration.