Migration Anticipation and Preparedness

1. Introduction

The multidimensional factors that drive migration trends are themselves difficult to predict. People migrate for a variety of interconnected reasons, which can include difficult-to-quantify considerations such as perceptions of safety or political freedom, changes in the economic situation at origin or destination, as well as personal reasons that cannot be easily captured. It is very difficult to look into the decision making of individuals before they move to pursue education, to start or unite with family, or to pursue individual aspirations. Migration also responds to policy changes in destination countries, and even in other countries as potential migration routes open or close. External factors may have an immediate impact on migration trends, while other times the impact may lag. Sometimes migration flows remain unchanged until a specific trigger point, while in other cases, emerging trends generate visible snowball effects. This extreme complexity makes predicting future flows particularly challenging.

Policymakers must still make an effort to anticipate migration flows, despite these difficulties. They are expected not only to exercise effective governance of migration but also to respond quickly and effectively to sudden changes. Designing effective policies for the future cannot be done without an attempt to understand what future flows might look like.

Forced migration (e.g. asylum, irregular migration, and movements combining elements of irregular and regular migration, such as mixed migration) in particular requires governments to conduct robust contingency planning and to be ready to rapidly adapt asylum and reception services. While regulated migration is by nature easier to predict – and less likely to catch policymakers off guard –, forecasting exercises are nonetheless important for programming and budgeting. Forecasts provide valuable insights for improved programming, financing, anticipating labour market needs, and making operational decisions, including staffing, integration support, facilities planning, establishing numerical limits or quotas coherent with policy objectives, and more. Improved anticipation of future migration trends also allows policymakers to inform public debate, explain choices and justify the costs of proactive investments and actions.

The track record for migration forecasting is improving, especially after several notable missed scenarios in recent memory. For example, the 2015/16 refugee surge from Syria in Europe and Russia’s war of aggression against Ukraine in February 2022 found most destination countries off guard. Looking back, however, visible early warning signals of mass displacement were missed. It is not just forced migration that reveals examples of missed signals: legal migration has been expanding recently to most OECD countries, sometimes putting unexpected pressure on housing or social services.

Enormous progress has been made in the past decade. Recognising the need for better forecasting, countries and international organisations or agencies have developed tools to predict short-term forced migration movements and assess migration uncertainties and risks. The EU, in particular, has established a robust early warning framework, which includes the Integrated Political Crisis Response (IPRC) Blueprint network, the well-established Frontex Common Integrated Risk Analysis Model (CIRAM), the European Union Asylum Agency (EUAA) Early Warning and Forecasting approach, and the most recent Situational Awareness, Early Warning and Forecasting Capacity Development Project, from the EU Joint Research Centre (JRC) in support of the implementation of the EU Pact on Migration and Asylum. Additionally, various national forecasting approaches in countries such as the Netherlands, Sweden, Switzerland and the United States have been developed over the last two decades.

Forecasting at first was at most notional or mechanical, relying on migration theories and causal (gravitational) models unable to fully capture the complexity of migration phenomenon. Today, it has increased in sophistication along with the development of the discipline, the massive expansion in computational capacity and the increasing availability and use of big data. However, these new forecasting efforts remain scattered, hard to generalise and narrowly focussed on short term effects, or even just on nowcasting – even with the right information systems in place, useful predictions are currently limited to a 12‑ to 18‑month time horizon. Moreover, for obvious reasons, most progress has been made on forced displacement (e.g. asylum applications). Regulated migration flows – such as family, work, and student migration – have generally not been the object of forecasting. Finally, much of the progress in modelling and data use has been occurring behind the scenes, for internal use by governments which understandably avoid publicising not only their data sources and the scenarios but often the models themselves.

The need for more, and better, forecasting, is clear. The request often comes down to civil servants and to modelers with shorter timelines and high expectations. There are no models that can be taken directly from the shelf and applied.

To provide more clarity, the OECD, with the financial support from the General Directorate for Foreign Nationals in France (DGEF), established the MAP (Migration Anticipation and Preparedness) task force. It aims to improve the anticipation of and enhance forecasting capabilities for both forced migration movements and regulated migration flows. The MAP Task Force complements other collaborative efforts cited above. The Task Force focusses on forecasting and does not engage in nowcasting (see Box 4.3), early warning, or foresight exercises. This Handbook is meant to support the growing demand for more timely and robust migration forecasting by providing practical guidance.

For whom is this handbook?

Migration forecasting involves many actors – the politicians and policymakers who ask for forecasts, the civil servants who must figure out how to produce forecasts, the modellers who must identify and secure data and choose and apply methodologies, the officials responsible for explaining the results of the forecasts, and the policymakers, politicians and communication teams who make use of them.

This handbook attempts to provide a guide for all these actors. In particular, it has been drafted primarily with five categories of users in mind:

• The civil servant, policymaker or official who has been tasked with producing forecasts, who needs to understand the state of the art of forecasting and how to meet the request with the resources at hand

• The modeller who has experience with forecasting in different domains but is relatively new to migration forecasting, who must identify data sets and understand specificities of migration

• The migration data analyst with little to no experience in time series modelling, who needs to better understand the different approaches and tools necessary for forecasting migration

• The migration forecaster who has been asked to improve the accuracy of their forecasts using new techniques or drawing on developing practices

• The civil servant, policymaker or official who receives the output of the modeller and must translate it into a form that can be used correctly, with all the uncertainties inherent to forecasting.

In some cases, the user of this Handbook may find themselves playing more than one role – both modeller and communicator, for example, or both commissioner and communicator. Policymakers who are considering whether and how to establish a forecasting approach may also draw on this Handbook in order to understand better the need for – and limits of – forecasting in migration.

The Handbook is structured into questions, which can be grouped into four general categories. First, what is migration forecasting, why is it useful and how is it currently being used? Second, how can a forecast be conducted, in terms of models, resources, data, development and evaluation? Third, how can forecasts be presented to policymakers so that they understand the complex results and the uncertainties and make proper use of the forecast? Finally, the Handbook points to some emerging questions in forecasting that may help policymakers and modellers alike to understand how the discipline may evolve in the coming years.

Forecasting for migration flows by categories versus (net) migration as a whole?

Deterministic scenarios of international migration are used by the vast majority of the statistical institutes in the developed countries notably to estimate population forecast. These efforts most often focus on net migration (immigration of both national and foreigners minus emigration of both groups) offering only aggregate trends without disaggregation by category. These are usually based on past demographic trends and does not attempt to predict the impact of policies or economic changes or of other factors potentially affecting migration.

Recently, as a result of rapidly increasing asylum applications, some efforts have been made to better anticipate specifically forced migration. The very nature of this migration component requires taking into account specific factors that go beyond demographic variables and are linked notably to the situation in countries of origin and transit. This has led to new forms of modelling that can not be directly applied to forecast other categories of migration. Several key factors explain why forecasting needs to distinguish between different migration categories.

First, different migration flows respond to different driving forces and show very different structural characteristics. For example, forced displacement due to conflict or natural disasters tends to be highly volatile and non-stationary. In contrast, labour migration between two high-income countries often follows more regular, predictable patterns (Bijak et al., 2019[1]). Forecasting models must reflect these differences. Time‑series models that assume stationarity are unsuitable for forecasting highly volatile flows such as asylum applications; while applying a non-stationary model to more stable, rule‑based labour migration may lead to overestimated variability and uncertainty.

Second, data availability and structure vary significantly across migration categories. Very high frequency data (daily or weekly) are typically available on asylum applications or irregular border crossings, while other administrative data based on permits are at best available monthly or just annually. The level of information attached to the migration headcounts also varies, with more or less details on socio economic profiles of migrants. These differences affect the choice of models and the robustness of forecasts. Selecting an appropriate forecasting approach requires an understanding of the scope, granularity, and quality of available data.

Third, while total net migration projections – which balance immigration and emigration figures – can be informative for long-term demographic planning, they offer limited utility for migration policymaking. Effective migration management depends on knowing who is migrating and why, not just how many. For example, preparing integration services, setting visa quotas, or adjusting labour market programmes all require category-specific insights. Forecasting future changes of status (for example transitions from study permits to work permits) might also require specific models (see Chapter 9 for a discussion on category interactions). Aggregated figures, while useful for demographic purposes are often too coarse to guide operational or strategic decisions. Tailoring models to the specific features of each migration category is therefore essential for producing meaningful and actionable forecasts.

References

[1] Bijak, J. et al. (2019), “Assessing time series models for forecasting international migration: Lessons from the United Kingdom”, Journal of Forecasting, Vol. 38/5, pp. 470-487, https://doi.org/10.1002/for.2576.