Migration Anticipation and Preparedness

8. How should forecasts be presented to policymakers and ensure effective interaction between policymaking and forecasting?

How to present forecasts and their inherent uncertainty to policymakers

The previous sections have presented the use case for forecasts and explored technical issues in their design and implementation. However, forecasts are only useful if policymakers understand their constraints and how to make appropriate use of the forecasts. This places an onus on modellers and those responsible for communicating results internally. This section provides indications on how to present forecasts and explain uncertainty to policymakers.

Communicating migration forecasts to policymakers effectively requires clear articulation of uncertainties and practical insights that can inform decision making. Given the complex and unpredictable nature of migration patterns, whatever the migration category under review, forecasts should be conveyed in a way that helps policymakers plan ahead while acknowledging their inherent limitations and the fact that uncertainty is a central element of the message.

One of the main challenges in presenting forecasts is indeed explaining uncertainty. Policymakers often request and expect definitive and precise numbers. Migration forecasts inherently involve uncertainty due to dynamic variables such as economic shifts, geopolitical events, and evolving policy landscapes. The most obvious solution is to present ranges of possible outcomes and scenarios rather than single‑point estimates, which better capture the uncertainty associated with migration drivers. Several options are available, including the traditionally-used variant scenarios, either based on some combinations of the underlying migration drivers (e.g. four migration scenarios combining economic convergence/divergence across countries and unilateral/multilateral international co‑operation in Acostamadiedo et al. (2020[1])), or full probabilistic outcomes, summarised through probability distributions (see Box 8.1) or a selection of predictive quantiles (Bijak and Wiśniowski, 2010[2]; Azose and Raftery, 2015[3]; Welch and Raftery, 2022[4]).

Presenting quantiles from predictive distributions to policymakers has one important advantage (see Bijak (2011[5]) and Raftery (2016[6])). Quantiles have a natural interpretation related to the frequency and magnitude of events: a median means that the occurrences below and above its value are equally likely, and each can be expected to occur on average half of the time. Quantile of order 0.95 means that higher values can be expected to occur only 1/20th of the time – so for annual forecasts, once every 20 years. Relative frequency of rare events (once in a decade, twice in a century, and so on, see Bijak (2024[7])), forms a base for analysing a range of migration scenarios.1 These are relatively simple concepts to communicate. For this interpretation to be correct, however, the predictive intervals must be well calibrated (see Chapter 7).

The graphic presentation can determine how policymakers understand uncertainty. One effective method for presenting uncertainty is through risk thresholds (e.g. the likelihood of exceeding a certain migration level) especially with visual tools such as fan charts (Figure 8.1), probability intervals (Figure 8.2), and density plots (Figure 8.3). An appropriate graphic representation can help policymakers grasp the range of potential migration outcomes.

Figure 8.1. Forecasts of total immigration to the United Kingdom, averaged univariate models (in thousands) 1975-2060

Source: Abel et al. (2013[8]), “Forecasting environmental migration to the United Kingdom: an exploration using Bayesian models”, https://doi.org/10.1007/s11111-013-0186-8.

Figure 8.2. Immigration of EU Citizens to the United Kingdom with different probability intervals, 1975-2013

Source: Bijak et al. (2019[9]), “Assessing time series models for forecasting international migration: Lessons from the United Kingdom”, https://doi.org/10.1002/for.2576.

Another option involves preparing a risk management matrix, juxtaposing the uncertainty and user-specific impact of migration flows (Bijak et al., 2019[9]). Such a matrix typically involves a variant of the highly-intuitive traffic-lights colour scheme, with ratings ranging for green for low-uncertainty or low-impact flows and forecasts, through amber as an intermediary category, to red for processes that are both highly uncertain and can have high impact, as is the case with asylum-related migration.

Ideally, the users of forecasts – policymakers – should receive some minimal training in understanding probabilistic forecasts, in order for them to effectively integrate migration projections into strategic planning and not to draw mistaken conclusions from misreading probabilities. One way to equip decision makers to make data-driven choices would be (if relevant to the institutional set up) to offer workshops and training sessions on forecast interpretation to help align insights with policy objectives. This is forecasters’ responsibility to teach policymakers how to read a model and that forecasting results are only possibilities of the future. A strong data culture helps policymakers better understand that forecasting models are the most efficient way to anticipate future migration flows, despite the inevitable and frustrating level of uncertainty. Some policymakers may remain reluctant to believe in forecasting outcomes; in this case, it’s important to note that a lack of prediction may be more costly. Collaboration and regular interaction between forecasters and policymakers are also key to improving the practical application of forecasts, ensuring they are both interpreted correctly and effectively used (Bijak, 2024[7]). The forecasters’ role should go beyond providing data and disseminating forecasting results to policymakers. They should regularly engage in discussions to ensure correct interpretation of the findings and to highlight emerging trends that require particular attention moving forward.

Box 8.1. How the Netherlands presents its forecasts to policymakers

The Netherlands has been conducting asylum application forecasts in a new, systematic way since 2018. The Dutch mixed approach combines Bayesian probabilistic models with expert intuition inputs, eventually providing four or more different forecast scenarios. Each scenario is then presented in a Probability Density Function (PDF) figure, which shows each possible forecasting value with its probability of occurring, according to this specific scenario. A combined PDF, summing all scenarios, weighed according to their estimated total probability of occurring, is also presented (Figure 8.3) and integrated into a cumulative PDF (Figure 8.4). In the end, this approach provides a continuous function of predicted values, rather than one value within a confidence interval. Nonetheless, what interests most policymakers is a unique value. Such a unique value can be derived from the combination-PDF by also providing a so-called “loss-function”, i.e. a function that quantifies the costs (both in money and in e.g. political unrest) of aiming “too low” versus aiming “too high”. The combination of these two functions yields an optimum number to consider in the preparations. The forecast is used in budget planning: the Dutch budget for asylum reception has to match with a bandwidth between the 30th and the 70th percentiles of the cumulative probability density function.

Figure 8.3. Probability Density Function for Prognosis – Forecasted first asylum applications 2024 in the Netherlands

Source: Dutch Ministry of Justice (2024).

Figure 8.4. Cumulative probability for the Dutch number of first asylum applications in 2024

Source: Dutch Ministry of Justice (2024).

Against that background, forecasters in the Netherlands regularly present their results by trying to combat typical probabilistic misunderstandings. Using the PDF representation, they emphasise the wide bandwidth of probability that might occur: a 10% probability does not mean that this outcome will not happen, but rather indicates than once every ten years, something unexpected might happen. The COVID‑19 crisis was a recent example of such a situation. Indeed, the number of actual 2020 asylum applications was under 15 000, which had only a 10% probability to occur according to forecasts, but still fell well within the PDF-indicated range predicted in 2019.

One important part of the Dutch approach is the central role of expert intuitions. One of the main arguments for the use of expert elicitation is that migration is such a complex phenomenon that predicting the consequences of various developments will never be fully objectifiable. As expert elicitation gathers many different insights and arguments, it provides a formal and informal exchange of ideas that are continuously challenged by one another. Even hard-to-conceive scenarios (extremely high or low migration flow levels) must be discussed by experts to take into account all possibilities.

The Dutch experience has also served to highlight the kind of misunderstandings policymakers may have, which can lead to wrong anticipations. It has been necessary for forecasters to e.g. insist that monthly forecasts must not be derived from annual forecasts by simply dividing the predicted value into 12 similar pieces. Explaining the seasonality of migration flows can help avoid this risk. Forecasters have also had to reign in the expectation that asylum influxes can be predicted with any precision or reliability beyond the horizon of two years in the future.

More importantly, forecasters have had to be pedagogical in repeating the aim of forecasting. An important task of the forecasters is to keep communicating clearly and transparently about the purpose and use of forecasts. Regular discussions are necessary to make sure policymakers are aware that forecasters are attempting foremost to provide a basis for sensible policy actions, rather than produce a headline figure. Predictions should be shared with all involved services, to make them aware of all hypothetical changes and possibilities. In that sense the forecasts can also be seen as a way to quantify ignorance. In the Dutch forecasts, the “tail” values indeed help policymakers to be better prepared and flexible and must therefore be taken into consideration to adapt the capacity of the asylum system in the extreme‑case scenarios.

Even if a wide range of possibilities are forecast, it’s still quite possible for reality to fall outside any forecast. In this case, forecasters have the difficult task of explaining how their model failed to policymakers.

Scenario-based forecasting presents migration projections as a range of possible futures rather than a fixed outcome, allowing for greater policy flexibility. It is important to incorporate scenario analysis into migration forecasting, helping policymakers anticipate various potential outcomes and prepare for uncertainty, rather than relying on a single definitive prediction.

However, when actual migration figures deviate from forecasts, transparent communication is crucial. In most cases, using a great range of possible estimates, such as the approach recommended in the previous section, helps to make sure that forecasts encompass the full range of possible outcomes. It’s easier for forecasters to explain deviation when the latter still falls within the boundaries (even extreme) of the interval (see Box 8.1 for the interesting example of a forecast that falls within the COVID‑19 lockdown). When forecasts fall outside the predicted values, forecasters have to explain to policymakers the underlying reasons for forecast errors, whether they stem from data limitations, unexpected policy changes, or external shocks such as conflicts or natural disasters.

Regularly updating forecasts and engaging in continuous dialogue about their limitations enable more adaptive decision making. Forecasters should clearly explain how external drivers influence migration projections, ensuring that policymakers interpret forecasts as dynamic and evolving rather than fixed predictions.

Checklist:

• Have I presented to policymakers the uncertainty of my forecasts in a clear manner?

  • Use charts and density plots showing quantiles from predictive intervals to clearly interpret the frequency of events, or alternatively traffic-light colour schemes in results tables.

• Do policymakers fully understand that forecasts are dynamic and not fixed predictions?

  • Regular interactions between forecasters and policymakers are necessary at each forecast updates. Training session explaining probabilistic forecasts may also help.

• What should I do if forecasts fell outside the range of estimates provided?

  • Precisely explain forecast errors, whether due to data limitations, unexpected policy changes or external shocks.

How to use the forecast properly and effectively

Migration forecasts are generally commissioned in order to inform policy action, so their conclusions should be policy actionable. In some cases, action needs to be taken quickly: when early warning models identify large additional influxes, the policy response needs to be quick and efficient. An effective way to increase the impact of forecasting is to include forecasters – or at least forecasting results – in emergency budget allocation meetings, which define how to allocate financial resources (e.g. staff, infrastructure, see Chapter 3). A dissemination policy developed in advance can identify which national practitioners to prioritise to receive forecasting results. When warned in advance, these actors and their budget experts can use their knowledge to better anticipate the financial consequences of the migration forecast and better allocate the funds for future needs. Although on a less urgent basis, forecasts made with a longer time horizon (a few years) can support an efficient budget allocation as well with the right dissemination policy followed by a budget allocation meeting in due time.

Forecasts may identify future needs that go beyond budget allocation issues. Forecasts help anticipate the implementation of new policies or rules that enable the country to adapt to possible or expected future flows. For instance, if a mass influx is predicted that could exceed budgetary constraints, forecasting results can provide the impetus for policies that either ease legal frameworks and administrative procedures to better manage larger flows, or alternatively, prompt governments to adopt stricter measures to deter arrivals. Similarly, if governments aim to forecast future labour shortages, policymakers will also need to anticipate corresponding labour migration flows and consider facilitating migration policies accordingly. The examples of Japan and Korea (Box 8.2) are labour migration projections rather than forecasts, but they illustrate how such scenario-based exercises can still inform migration policymaking – for example, by helping to set visa quotas or adapt labour migration schemes. In practice, both forecasts and projections become policy-relevant when their results are clearly communicated and effectively linked to decision making processes.

To make sure forecasts are taken into account for policy action, the forecast results have to be disseminated to a clearly identified network of national practitioners, following a whole‑of-society approach. Forecasting results can be shared with all stakeholders responsible for policy action on the ground (from national to local level), supporting their efforts to efficiently anticipate the necessary future action. However, not all forecasts should be made public. As the dissemination of forecasting results is a critical component of an effective policy response, it must adhere to established confidentiality and security protocols when shared with relevant stakeholders.

Box 8.2. Labour migration projections in Asian OECD countries

In Japan, the JICA Ogata Institute projected foreign labour needs for 2030 and 2040. The study began with GDP growth targets set by the Ministry of Health, Labour and Welfare and calculated the achievable GDP based on demographic projections and capital stock scenarios, including automation. The gap between achievable and target GDP was attributed to foreign labour requirements, which were further disaggregated by industry and region. On the supply side, the model estimated potential migrants from 13 origin countries, using UN population projections, IMF growth forecasts (assuming emigration declines as incomes rise), Japan’s historical migrant share, and retention rates. The study concluded that the stock of foreign workers would need to increase by 236% by 2040, yet only around 420 000 workers could realistically be supplied, leaving shortages of 0.77 million in 2030 and 0.97 million in 2040.

In Korea, labour demand – supply projections are used to guide the Ministry of Justice’s employment visa quota pre‑announcement system. In 2024, the quota for skilled worker visas (E‑7‑4) was raised to 30 000, up from 2 000 in 2023. A study commissioned to Seoul National University projected labour supply by sex, age, education, sector, and occupation, while labour demand was derived from models incorporating productivity and technological change. By comparing projected supply and demand for 2026‑2031 under four scenarios with different assumptions about substitutability across sectors and skill levels, the government estimated varying levels of labour shortages. These results serve as the basis for setting annual visa quotas by sector, with higher caps considered where shortages persist and lower caps where substitution is feasible.

Taken together, the Japanese and Korean cases illustrate how projections can inform migration planning. They do not predict actual inflows, but provide structured scenarios linking labour demand, domestic supply, and long-term migration policy needs.

One use case for forecasts expressed as full probability distributions is to input them into formal statistical decision analysis to make informed decisions based on costs incurred to address the different scenarios. The information provided by the whole distributions of results can be combined with national practitioners’ knowledge of the possible consequences of the decisions that are based on the forecasts. Such knowledge is best expressed in terms of loss functions, which describe the costs (losses) incurred when the reality departs from the forecasts – in other words, whenever the forecast either underpredicts or overpredicts the actual migration flows. The tacit assumption is that these costs are minimal if the forecast comes true, which however formally speaking has near-zero (or even zero) probability. The shape of the loss function and the relative magnitude of the costs of overprediction and underprediction, together with the probability distribution describing the forecast, jointly determine which predicted value should form the basis for decisions.

To illustrate this approach, for the numbers of asylum seekers, forecasts may serve as a useful tool for planning the resources needed to cover processing costs, as well as provide adequate reception services, registration support and other integration services if any. In this case, overpredicting asylum applications will likely lead to incurring additional costs through providing redundant reception capacity which ends up unused. On the other hand, having insufficient capacity will result in last-minute additional costs to scale up the reception and registration services, likely done not in the most efficient manner, as all emergency costs. In this case, and in most other practical applications, these costs are typically not symmetric: depending on the exact context, overprediction may end up being either more or less costly (per migrant) than underprediction by the same magnitude (see Box 8.3 for a practical example).

Box 8.3. A Practical Rule from Decision Theory

The methodology and technical aspects of the statistical decision theory is discussed more fully elsewhere (e.g. Robert (2007[10])), but there are some simple and approximate rules, which can be easily applied in practice. One such rule refers to the situation when the costs (losses) scale up approximately linearly with the magnitude of overprediction and underprediction, and that the costs (losses) of underprediction are k times higher than for overprediction of the same size. In such cases, it can be mathematically shown that the optimal predictions, minimising the expected loss under a given predictive distribution, are quantiles of order k/(k+1) from the predictive distributions. For example, if k = 4, the optimal prediction is the fourth quintile (quantile of order 4/(4+1) = 0.8), whereas for k = 0.25, indicating that in fact overprediction is four times costlier, the first quintile (0.25/(1+0.25) = 0.2) is optimal. A special case occurs when the costs of overprediction and underprediction are linear and symmetric (k = 1): the optimal decision is then the median (quantile of order 1/(1+1) = 0.5) of the predictive distribution. The value of k needs to be also elicited, this time from the users of the forecasts (the national practitioners), who will be best positioned to judge the relative costs of over- and underprediction.

One important point to make is that if forecasts serve as input for policy decisions, then these decisions can, by their very nature, influence the migration flows and invalidate the forecasts, at least to some extent (see the Lucas critique, Lucas (1976[11])). Still, the policy effects on migration are uncertain and difficult to quantify – and so are the effects of migration on policy (Czaika et al., 2024[12]). One possible way of dealing with such challenges is to create several conditional forecasts, which can be in fact seen as scenarios, created under the assumptions of different decisions, or different political or policy developments. One example of implementing such an approach is a set of conditional forecasts of total migration to and from Scotland depending on the results of the 2014 referendum on Scottish (Wiśniowski, Bijak and Shang, 2014[13]). In this example, the calculated expert-based forecasts were conditional on the outcome of the referendum and subsequently averaged over both possible outcomes (Yes and No), based on a prediction of the referendum outcome at the time of preparing the forecast.

These caveats notwithstanding, successfully presenting migration forecasts to policymakers requires a careful balance between clarity and complexity. Ensuring transparent communication about forecast uncertainty, offering explanations for deviations, and integrating forecasts into policy discussions are all crucial for enhancing the role of migration forecasting in decision making. By utilising probabilistic models, scenario-building techniques, and continuous updates, forecasts can become a valuable tool in anticipating and effectively managing migration flows.

Checklist:

• How to make sure that forecasts are translated in budget planning and policy action?

  • Disseminate forecasting results to the right network of practitioners and integrate forecasters in budget allocation meetings and policy discussions is key to take the most informed decision.

• Has the magnitude of costs’ underprediction versus overprediction been clearly defined?

  • To minimise the loss, the optimal forecast value has to be taken within the predictive distribution bandwidth, after an expert discussion defining the cost of underprediction versus overprediction, For asylum reception for instance, the cost of urgent reception capacity versus unused reception capacity.

• Have I built conditional forecasts depending on different scenarios?

  • Where some policy decisions or development are ongoing, create scenarios based on their different outcomes may help grasp the full extent of migration predictions.

References

[8] Abel, G. et al. (2013), “Forecasting environmental migration to the United Kingdom: an exploration using Bayesian models”, Population and Environment, Vol. 35/2, pp. 183-203, https://doi.org/10.1007/s11111-013-0186-8.

[1] Acostamadiedo, E. et al. (2020), Assessing Immigration Scenarios for the European Union in 2030 – Relevant, Realistic and Reliable?, International Organization for Migration and the Netherlands Interdisciplinary Demographic Institute.

[3] Azose, J. and A. Raftery (2015), “Bayesian Probabilistic Projection of International Migration”, Demography, Vol. 52/5, pp. 1627-1650, https://doi.org/10.1007/s13524-015-0415-0.

[7] Bijak, J. (ed.) (2024), From Uncertainty to Policy: A Guide to Migration Scenarios, Edward Elgar Publishing, https://doi.org/10.4337/9781035319800.

[5] Bijak, J. (2011), Forecasting International Migration in Europe: A Bayesian View, Springer Netherlands, Dordrecht, https://doi.org/10.1007/978-90-481-8897-0.

[9] 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.

[2] Bijak, J. and A. Wiśniowski (2010), “Bayesian Forecasting of Immigration to Selected European Countries by using Expert Knowledge”, Journal of the Royal Statistical Society Series A: Statistics in Society, Vol. 173/4, pp. 775-796, https://doi.org/10.1111/j.1467-985x.2009.00635.x.

[12] Czaika, M. et al. (2024), “European migration governance in the context of uncertainty”, International Migration, Vol. 62/6, pp. 93-110, https://doi.org/10.1111/imig.13308.

[11] Lucas, R. (1976), “Econometric policy evaluation: A critique”, Carnegie-Rochester Conference Series on Public Policy, Vol. 1, pp. 19-46, https://doi.org/10.1016/s0167-2231(76)80003-6.

[6] Raftery, A. (2016), “Use and communication of probabilistic forecasts”, Statistical Analysis and Data Mining: The ASA Data Science Journal, Vol. 9/6, pp. 397-410, https://doi.org/10.1002/sam.11302.

[10] Robert, C. (2007), The Bayesian Choice: From Decision-Theoretic Foundations to Computational Implementation, Springer New York, New York, NY, https://doi.org/10.1007/0-387-71599-1.

[4] Welch, N. and A. Raftery (2022), “Probabilistic forecasts of international bilateral migration flows”, Proceedings of the National Academy of Sciences, Vol. 119/35, https://doi.org/10.1073/pnas.2203822119.

[13] Wiśniowski, A., J. Bijak and H. Shang (2014), “Forecasting Scottish Migration in the Context of the 2014 Constitutional Change Debate”, Population, Space and Place, Vol. 20/5, pp. 455-464, https://doi.org/10.1002/psp.1856.

Note

← 1.  For an example, see the QuantMig Migration Scenarios Explorer: https://www.quantmig.eu/data_and_estimates/scenarios_explorer/ (as of 15 November 2024).