 Our next speaker is Ravenna Sost, she's a doctoral researcher at the University of Luxembourg and also a consultant of the IOM's Global Migration Data Analysis Center, so the GIMDAC, where she has recently co-authored two studies focusing on migration forecasts. In her first session, Ravenna will provide a comprehensive overview of the main predictive models that are currently in use, so forecasts, scenario studies and early warning and alert systems. Ravenna is the only one today who actually has the pleasure to be able to present for 20 minutes because she's really offering us a foundation so we can all speak from sort of the same level of knowledge for the rest of the day. So she will spend 20 minutes presenting and then we have 10 minutes for Q&A, but please feel free to write your questions in the chat whenever they occur to you and we'll be sure to organize them accordingly. So Ravenna, it is my pleasure to hand the virtual podium over to you. Hello, thank you very much for the introduction Julia. I hope you can hear and see me well enough. I would have loved to be in Vienna, but this is the time that we're living in and I'm happy to participate. So today I'm going to speak to you about the different approaches that exist to anticipate migration. My goal is really to lay around work for the follow up presentations that will take place throughout the day. So I will focus specifically on an overview of approaches and emphasize the relative strengths and weaknesses that come from them so that when the migration snapshots and the predictions that we will see later on in the day will become more intuitive relative to each other. I will speak for about 20 minutes, as Julia said, and there's time for some questions or clarifications should you have any in the end. So feel free to write anything in the chat or just afterwards. Okay, so interest in migration forecast and migration scenarios are clearly growing. What you see on the screen is a result from a systematic literature review that we conducted last year at Jim back. And that shows the number of forecast and scenario studies that were published each year since the 1940s approximately. So the number of and forth the forecast as shown in blue and the scenario studies as shown in yellow. And what you can clearly see is that the trend is going upward. So we're not the only ones here right now who are interested in and future migration. Over the last years and really the case, though, the amount of approach has been growing tremendously and there is now a great diversity. So it is helpful to think of these different approaches to belong to either of three groups, which would be first, early warning systems. Second forecast or statistical models. And lastly, foresight, you can understand them. For example, as ranging from a very short predicted horizon to one that goes to the medium term and the long term, ultimately. However, all these approaches are meant to help us anticipate migration, but they're fundamental differences that are really important to understand. For example, these approaches differ in how the conceptualized migration, some of the approaches that I'm going to speak about, for example, draw on causal theories of migration. They speak about drivers and push and pull factors, for example, whereas others are agnostic in that kind of sense. The approaches also differ in what time frames they're looking at. So I've shown that a little bit, but they can really, really range from a few days or weeks up to a few decades in the most long term. Approaches to anticipate migration also vary in what type of data are required, for example, and what and who is involved mostly in the production. So some might, for example, draw on their comprehensive administrative data sets, whereas others are more reliant on expert opinion in the very diverse stakeholders. And of course, all these approaches differ fundamentally in how they treat uncertainty. I'm not going to go very much into this because there is a talk by Jacoby Jack coming up who will provide you with much more detail in that. So now I'm going to go one by one through the fundamentals through one of the most common approaches, let's say, and give you a little bit more background and details about them. So let's start with the early warning system. As the European Commission's new pattern migration and asylum there was just really last week really showed is that early warning system has come so central to the European Commission thinking about migration management. And so I just brought that abstract here from this new pack. It says a structured migration management mechanism is necessary with real time monitoring and early warning. The EU should be moving from a reactive mode to one based on preparedness and anticipation. So what does that really mean and how does an early warning system function. As I said, early warning systems function on the short term, they really range from a few weeks to usually a few months up to one or two years at most. Early warning systems drawn a constant supply of information about migration trends and potential drivers. As such an early warning system could almost be more understood as migration monitoring or even now casting, as you will hear by Andrei Grueger later later today also. The type of information can be both qualitative or quantitative depending on the system. For example, it could be a network of field workers that are stationed across the world in interesting regions and that provide reports about changes that they observe. This is for example done in the IOM displacement tracking matrix, but it can also very well be quantitative. In that case, it could be, for example, in a timely information on board apprehensions, or it could be satellite data, or it could be more innovative data sources such as social media data or search engine results, as we will also hear. By combining all these different data sources, early warning systems allow to identify imminent migration surges and potential trigger events. Early warning systems have been mostly practitioner driven and so that focus on humanitarian migration. As we've seen, for example, with the EU migration tax, the goal is really to facilitate operational planning, allocate resources and therefore become more active in reacting to new events. And because of the relatively short time horizons, they're also presenting a relatively high accuracy, which is attractive to use with. The second, the second approach that I want to present are statistical model based forecast. They're more functioning on the midterm, so we speak about months, but also a few years, maybe 10 years. And really there are two approaches that are key in this field. One are econometric regressions and the other one are time series extrapolations. And they're quite technical and what is most usually referred to when we speak about forecast in a common sense. But there's one key difference between these two that is really important and that is the use of explanatory factors. So in econometric regressions that are used to forecast migration, so-called explanatory factors are drawn on that are usually taken from causal theories of migration and that explains why people move. This could for example include distance between the two countries, it can include income differentials between two countries, labor markets, situations, or also historical ties, shared common language and so on. So each of these factors in an economic forecast is quantified and then used to forecast migration into the future. In contrast, there's also time series extrapolation, which do not use any sort of explanatory variables, at least in a simple version, but instead the only type of data they draw on are historical time series data. So here in the bottom right of my presentation, you see one example that I've been recently working on with my colleagues from GEMDAC in Berlin. It's just to show that on the left side you see a thick black line, which are historic flows that have been observed, and on the right hand side you see a projection that is based solely on the historical time series data. So how it works is that it recognizes some sort of trend or pattern and projects it into the future. The shaded area that you see around them, there are prediction intervals and shows some sort of uncertainty spans that go along with that extrapolation. Let's go to the third approach that is commonly used. And these are expert and survey based forecasts. Again, there are two most important ones I would say. This is the Migration Attention Survey. It basically works by going into a country where you have potential immigrants and asking those people about their desires, plans or preparation to immigrate. For example, the Gala World Core Data is widely available and has a great comparability and asks these types of questions. It is important to keep in mind that any type of intention cannot be translated one to one in an actual action to immigrate and do that migration process. However, there are some of the papers that show that an intention and especially preparations are good predictors of actually migration behavior. Another type of expert and survey forecast relies more on a group of experts. And that's, for example, the Delphi Survey. Susanna Meldis is going to speak about that more in a few minutes. And the idea behind the Delphi Survey is to collect the estimates of a group of experts and by doing that in a systematic way reduce individual bias. And sort of distill a consensus of experts that creates a more reliable forecast in the end. These types of experts estimates can be used to complement statistical forecast like, for example, the time horizon, the time series extrapolations that I've shown previously. But it can also be used, for example, when data is scared about specific flows or when migration flow data are very volatile and therefore statistical models fail to make accurate predictions. Lastly, we're moving to the most long term of the approaches. This is fourth side. And there are different types of fourth side analysis like horizon scanning, for example, or trans analysis. But the most commonly used in migration, in the migration field of scenario building. Scenario building is really a long term strategic type of approach that's usually based on group work. Diversity of stakeholders come together, for example, experts from academia, policymakers, and against themselves. And they come together in some sort of discursive process and met our contextual factors that will contribute to future migration. In the end, they come up with a narrative output, a qualitative output of what is scenarios. So they present not one future, but a field of possible futures. In that sense, scenarios and foresight in generally helps in thinking about assumptions, helps to challenge assumptions and highlight the complexities that are surrounding long term developments of future migration. Here, for example, is an extract from a scenario workshop from the International Migration Institute from 2011, where you see where participants group different factors really long to access. One access is how certain are we about these developments. And the other access is what has really the greatest impact on migration. And on the upper right hand side, there is a red circle and it shows one of the factors that I think is really a good example of how foresight, for example, differs from statistical modeling. It says EU fragmentation and sub-regional block formation. So this is a factor that has been considered very important for the future of migration to the European Union by the participants. But it would be typically something that's very difficult to model in, for example, econometric regression or time series extrapolation. So this is the type of arguments that can be discussed in foresight exercises. So this is just a sort of wrap up and last comparison of the three approaches that we did. Just to remind you, there were early warning system on the first as a fresh group of approaches. They were relatively short term, a few weeks to a few months. They can draw on a qualitative or quantitative information and can mix them easily. And they have a focus on mostly humanitarian migration. Second, we looked at forecast and statistical modeling with a midterm sort of outlook. And they usually mostly draw on quantitative data sources and then model regular migration flows such as labor, family or student migration. And lastly, we looked at foresight as an approach to anticipate migration with the most long term strategic approach and a qualitative narrative output. All types of migration flows can be considered in foresight exercises. So just to maybe launch the discussion and make a quick conclusion here is what can be then given all these different approaches realistically expect from migration workcast. One thing that is very clear is that there is no universally preferred approach. It really depends on the purpose of the exercise or of the forecast to tailor to what is needed. There are of course also trade-offs in terms of resources, time, people, data requirements, time horizons and so on that need to go into thinking about which forecast is the most adapted to us. And lastly, it is really important that forecast producers and forecast users work together and in a discussion produce something that is adapted and is also understood as what it can offer. Because every approach has something that can show and others which is simply cannot provide. So thank you. That is it from my side so far. And I'm looking forward to any questions that you might have any clarification. Thank you so much. That was a really excellent overview. Very comprehensive. Very to the point. I think we all learned a lot. So thanks. Thanks for this presentation. We have a few questions already that came in, which my fantastic team is helping organize here. The first question is from Arvid Zeng Norin from the Swedish Migration Agency. He's asking whether you've only investigated approaches to migration forecasts in research or whether you've also looked into the prognosis papers presented by migration agencies or ministries in different countries that are normally not conducted as research. So they don't have generally established methods. I assume the question is in regard to you mentioned in the beginning you did a systematic literature review. What kind of papers you looked at? Do you prefer to answer that or should I give you the next few ones as well? Maybe I answered directly. Great. Okay. Yeah. So we conducted the systematic literature review last year, which is also available and which can I share in the chat. And the focus in the systematic literature review was on the one hand, really on scenario studies. And in the scenario studies we looked at both academic and practitioner driven literature. Scenarios are mostly practitioner driven methodologies. So we actually had more studies coming from governments or from the EU, for example. But for the more quantitative forecast, so the statistical type of modeling, we focused on academic literature. It did not take into account very much of the practitioner approaches in that field. Excellent. Thank you very much. We also have another question from a PhD student and he's asking whether you see the possibility of using the machine learning techniques and predicting the immigration flows well in advance. Machine learning has many merits over the standard econometric approaches. So we will be speaking about machine learning. Actually in the next few presentations. But maybe you will have a comment. Yeah. Yeah. I mean, I think the following presentation will make that much clearer. But generally speaking, there is a great potential also for machine learning. And it has been increasingly used, let's say, they have been a few projects and they're developing greatly. But in its foundation, there is nothing against it speaking against that. So absolutely yet they can be used. Super. So if you hang on, then you will have the pleasure to listen to Andre Gruger who will touch upon this. This approach in more detail. Another question from Marie van Dresch from the Federal Planning Bureau of Belgium. And she says that econometric approaches or time series extrapolation are able to take what has already been observed into account. But how do we take not yet observed events into account. So for example, climate migrants. So how do we, how do we build scenarios around that? That is a very good question because it is really the fundamental difficulty I think of migration forecasting. Migration has very little regularity. So looking only at the past is unlikely to give us a sufficient picture for the future. And so I think if you if you're interested, for example, in in more recent changes, you would need to move away a little bit from the purely statistical modeling and take into account. For example, expert opinion as has been done in Bayesian frame rates. For example, the jack who's also going to speak in just a few minutes and or move even further into the foresight exercise. But as I said, it will be a trade-off in terms of certainty. So you can take into account these sort of events, but they will reduce the certainty of the forecast, especially when you look at longer time horizon. So there is another question from Alessandro Venturini, and she's asking about the problem of the quality of the data. So she says that South South migration is not well counted. What can we do? Yeah, it's true. My migration data are a difficult field, let's say. And so again, my my approach would be either to look for alternative approaches or combine approaches, combine data sources. For example, I spoke about the Delta video, which is usually one approach that can be used when data could be scarce, for example. Or another example would be the type of social media data or satellite data that can be drawn on, for example, when administrative data sources are really scarce. So I think the challenge has really become creative and from the diversity of approaches that we have picked the ones that we can use and use either one of them or combine them ideally. Very interesting. Revena, I actually also have a question. So at the European migration network in Austria, we consulted other European migration network member states or EU member states to get a sense whether migration forecasts are used across the European Union. And we received responses from 22 EU member states who responded to a survey and of those of those 11 reported that they do use forecasting methods at national level. So half of those that responded reported to do use it. And so one of the interesting findings for me was that almost three quarters of those use risk analysis or early warning systems. And only one third use sort of the longer term methods. So only one third go beyond two years. Now, you explained that with the different models, they're different, you use them depending on the policy approach. But do you see any methodological reason behind the fact that most member states are interested in the immediate to short term rather than the longer term? Well, yeah, this is exactly what I wanted to point out in my last slide on my last slide. So they're obviously trade-offs between the different advantages and time horizons. And one of them obviously is a certainty that's attached to forecast. So when you compare, for example, a foresight approach to an early warning approach, then the accuracy and the certainty that you can take from them is very different. Of course, there could be reasons to be found in the political economy and election cycles and so on that you can speculate about. But I don't think there is a purely methodological reason to prefer one approach over the other. There is no universally better approach in that sense. It really, really depends on the purpose of that forecast. But certainty is a key factor than a short-time horizon and therefore early warning systems seem preferable. Thank you very much. So there are more questions coming in from the participants which I'm very pleased about. There is one question by Dina Desa who is asking, first of all, thank you very much for a lucid and insightful presentation. The question is, where would you place probabilistic models of catastrophic risks that can provide the basis for displacement risk models? So the distinguishing between deterministic and probabilistic forecast methods is not something I mentioned in this presentation, but they can kind of cut through the different approaches. For example, a few of the ones that are presented could be presented both as deterministic or as probabilistic. I think in the foresight field, the uncertainty is something that's built into the method. There is no question about that. But for everything that's in the midterm, there's a question for a researcher or for someone wants to use a migration forecast, whether they want to have it in a deterministic or a probabilistic fashion. In my opinion, there should always be a sense of uncertainty surrounding forecasts. No forecast is going to become reality like that, 100% certainty. And so the more we move into some sort of probabilistic sense, the better it is for forecast producers and users in the end. Thank you. So there is another question from Manfred Kohler from the Austrian Ministry of the Interior. And he's asking how much international conflict and the potential or the conflict potential to produce displacement is used in forecasting. So he's referring to the Brown University paper on displacement caused by the US, which shows that wars are a major driver of displacement since 9-11. So I assume this would be more the early warning system, but it would be interesting to get your take. Yeah, exactly. So in immigration forecasting, so-called shocks or events like natural disasters or violent conflict are very hard to forecast, especially with a longer time horizon. So exactly as you said, this would be a typical case for an early warning system to come into place, because there really you can see an event happening and can estimate the consequences of that event into a few weeks into the future. But that reminds into a short, it remains a short time horizon in the end. Excellent. Thank you so much. So, I think we have satisfied all questions of our attendees. I'm very pleased that with this presentation. Thank you so much, Rovenna, that we're all now kind of on the same page and level of knowledge. And I would like to maybe also go to the next panel or the next set of presentations from you. We've really learned that for the forecast are quantitative estimates derived more from statistical and econometric models. We also understand that scenario studies produce more the qualitative narratives and provide an area of different storylines and thought experiments. A future development and that early warning system present this third group, which is a combination of qualitative and quantitative, much more short term. And I would now like to introduce three fantastic speakers who will present insights into key findings of recent forecast and scenario studies and does share how these methods have actually been applied in practice. Our first speaker will be Susanne Melde. She's senior analyst at the IOM's Global Migration Data Analysis Centre in Berlin, and she currently also coordinates IOM's Global Migration Data Portal. She will present results of a recent study which looks at immigration scenarios for the European Union in 2013. Next, we will have André Grueger. He's an assistant professor of economics and Juan de la Sierra Felo at the Universidad Autónoma de Barcelona. And his research interests are in applied microeconomics, data science, development, labor and political economy. André will present a recent paper where he looks at Google trends data to shed light on how online search data can be used to predict migration. And finally, I have the pleasure to also introduce Jacopizak. Jacopizak is a professor of statistical demography at the University of Southampton, and he focuses his research on demographic uncertainty, population and migration models and forecasts, as well as the demography of armed conflict. He will present results from a recent paper drafted jointly with Matthias Taika, who we'll get to know later today, where they look at the critical element of uncertainty forecasting international migration. So each speaker will present for around 10 minutes. After each presentation, we will have a very brief Q&A session. So this is mainly for clarification questions. So if you do have questions during the talks, please don't hesitate and post them, but we will reserve 20 to 30 minutes at the end of the three presentations for discussion. So feel free to pose your questions whenever you want, and we will continue channeling them like we just did. So without further ado, Susanna, I would hand the word over to you. Thank you very much. And welcome also from my side. It's a great pleasure to be able to speak here at this very impressive conference with an impressive list of speakers. As I mentioned before, it is my pleasure to present to you and the study my colleagues conducted on immigration scenarios for the European Union in 2030, which will actually be launched today. So I'm really happy to present that those findings to you. My colleagues at Lauderovena, Jasper, Georgian, Harga, some of them are also joining. So if there's any questions after this, please do ask them as well. So, just a kind of background, this study was conducted as part of a EU funded 2020 project called Cross Migration, which we did together with the Netherlands Institute. It's a generic demographic institute. And we're just presenting the results today. So what did we do in terms of the methodology? We have been leading on and what we did there was to look at different scenario studies that were out there 21 European scenarios and looked at what were considered the most defining issues. And those were the economy and international corporations and multilateralism. And then the colleagues developed those four scenarios based on those storylines. So the first one on the naturalism and economic region, second on multilateralism and economic convergence. And so on as you can see there, these colors will come back when I present the results, but I'll, I'll explain that a bit more in detail. And so in terms of economic convergence, we were looking at, you know, how the EU and origin areas of migrants, such as in Africa, Latin America and Asia, you know, how they would be hearing. And so, as Ravenna was also mentioning, these scenarios are looking at what is what happens as a certain driver comes into play. And they look at alternative features. So they are kind of which we exclusive to the economy, one scenario selected, and they look at structural changes and consequences for migration. And we combined that with the Delta survey. These Delta surveys look a bit of, there's different approaches, but mostly look at how to write the consensus among experts on miracle essence on migration flows. And you do several rounds to see, to get to an agreement on a certain topic. And we combined those, one of these estimates of migration flows with different scenarios on migration drivers and looking at their likelihood. So we looked at, you know, the third and fourth group that Rana mentioned in her report site. And just a few words on the survey itself. So we looked at the probability of the four scenarios. And then also add some absolute flows on the total interest to be by 2030. We did two rounds of labor, my friends, I used for my friends, a regular flows and parent films. We did two rounds of the expert. Yeah, that's a survey and had an example of about a hundred and 10, 110 experts, because we only took those. These ideas in migration. So the interesting part of the results. So when we look at the four scenarios that I mentioned, what the experts rated us the least likely, 90%. Did you see here with four scenarios and to be equally probable, they have to arrive at about 25%. So, you know, we should arrive at 100% of each of those probabilities. And the one you see in the green, the least likely one is the one on economic convergence and multilateralism. So expert basis that international cooperation and less economic differences between countries is the least likely. And the other ones, in particular, when one's on unilateralism, so in blue, together with economic convergence or in yellow, together with economic divergence. They rated us more likely. So that is a bit depressing and that as hopeful, but yeah, that's what the experts rated. And in terms of annual migration inflows, what you can see here on the left is a graph with the black line showing what were the actual flows between 2008 and 2017. The dotted line represents the average over those 10 years. And then you can see on the right where experts rated under which scenarios, what would happen to those inflows. And you can see in red and green with multilateralism, there will be more influence. And with the natural approaches, which you can see, there's no yellow and blue, there would be less movement. But what is quite interesting is that on the very few cases or for a few scenarios, those flows will actually surpass the peak that we saw in 2015 and 2016. So flows are expected to increase under all scenarios between 2144% and what is not true, but especially for highly steadier and expected to be quite considerably. Now, when we look at asylum attitudes, we saw there were a couple of questions in the previous discussion on forced movement. And you can see, again, on the left in the graph, the actual figures from 2009 to 2018 with the peak that we saw in 2015 and 2016. And under all of these scenarios, experts expected to send applications to decrease by 2030. And you can see the red and yellow scenarios, look at economic divergence. You can see if there's more differences between you and other regions of the world, and they may be instability in those other regions. And so, you know, leads to more asylum applications. And under economic divergence, so the blue, blue scenarios, you can see that it's there, you know, expected to be in line with the 11 year average. And when we look at a regular border crossing from also over the past 10 years, based on context data, you can see the peak in 2015 and 2016. I'm not going to discuss some of the underlying issues with the data. But of course, you know, if there were more border controls, there's also more detections of the regular questions. But what you can see, interestingly, that under all scenarios, not under all scenarios, you can see a great difference according to economic differences. So in yellow and red, the scenarios, look at economic divergence. So you can see if there's more differences between origin and destination regions. Excerpt is an increase in regular border crossings, and the blue and green scenarios, and the expected which represent economic convergence and the naturalism, not a naturalism, respectively, they expected a decrease. And what you can also see in the graph on the left is that under all scenarios, experts would not expect to see similar peak as in 2015 and 2016. Now, when we look a bit at the experts responses, and we can see here, so-called violence and on in red, you see the first round of the survey and in blue, the second round. The first noteworthy aspect there is that experts greatly disagreed on, you know, the flows of migrants coming to Europe by 2030. You can see the future variety in the boxes. And they represent the black line in the middle of those boxes represents the majority of 50%. So what 50% of the experts responded? And you can see according to the different scenarios they expect between two different people will enter the EU by 2030, but there are huge differences. And the larger on those boxes, the more diverse the responses were by experts. So there's large disagreement and uncertainty about what will happen in 10 years. And what's quite interesting is with those two rounds we were also looking at, did experts actually change their opinion once they had seen what other experts answered. And we can say they did not only 9% actually revise their assessment. They did not only stick to their own views and even when seeing what others were responding and they weren't very confident in their own responses on a scale from 0 to 100, they only said they would rate their responses at about 40. Quite some uncertainty. So you probably asked yourself, well, then why should we look at those scenarios? We've seen experts are not very sure what will happen and how certain migration drivers will pan out. But the impact is likely to be ambiguous on the volume and the position and direction of migration flows. And interestingly, experts don't answer in a different way than the general population. And so any questions. We can reach out to experts and I'll come back to that. And another finding was that the best case scenario for international cooperation. Convergence integration was rated as least likely by 2030. And in terms of drivers, international cooperation, multilateralism was considered to lead to higher migration, which could be through the movement or for your movement. But also working together. And while economic divergence and differences were considered to lead to a regular migration and driving up asylum applications. So, not as much as in the 2016 to that we've seen also in the data. So that's quite an interesting finding that this crisis or so-called crisis in 2016 was not expected to. So what are some of the policy implications? Coming back to the list and that was also leading to this is more for a strategic long term thinking, like a thought process. We've seen there's quite some disagreement, experts don't really agree and also didn't really revise their kinds of view that once they saw the other experts responses. But what is useful about this approach is to highlight this uncertainty. Probably not what a policymaker would want to hear and rather refer to cut responses. So I think that these drivers, how they will work out and depend on a number of factors. So, any also on the old bias and approach of experts and there's different approaches. So there's a lot of itself to know that there is disagreement that there is uncertainty. So I think it's useful to think about, you know, what are different scenarios, what are different storylines, what could happen if this and that occurs and right input to rather trends, generate a discussion among policy makers, but maybe also some less likely scenarios that we could have mentioned in 2020 as well. This service conducted before and there is a scenario that was very unlikely before it has now happened. And so, yeah, looking at the interdependencies that different migration drivers have. And with that, I would like to thank you. Thank you very much. This is really insightful. So I don't think there is a question from the chat. Or maybe there is, and I haven't appeared for me so maybe you can get it, but I wanted to ask you. Maybe you can. So you explained very nicely the methodology and the model, but maybe you can sort of give us an example of what this would mean in the real world. So what does the world look like that is economically divergent with more multilateralism and what does this mean is, for example, you want to look at migration from Africa to Europe or something. I don't know, take that as an example. What does this mean? What does this mean in this sort of exemplary forms? Yeah, I mean, I think I guess the multilateralism, you know, the countries work together, there's regional integration, and not everyone takes their own decisions. But we will try to have a common approach in terms of economic convergence. I mean, those scenarios, of course, in terms of how we visualize them are simplified. There's a lot of other factors like technology, democracy, market inequality, social policy, the environment, but also conflict that went into the storylines of those scenarios beyond the time here to explain what that would look like. But yeah, in terms of economic convergence, shifting wealth, so that there's not that many or that huge differences that some regions are very well off, others getting poorer and poorer, but there's, yeah, shifting wealth between different regions. So what we've seen quite interestingly that expertise to look a lot at, you know, pushing full factors. So if there's, you know, if there's higher wages or incomes in certain regions like the European Union, that that will attract people to come here. And the differences are not that stark. They wouldn't go there, which reflects the common migration theory. So that's interesting to see that in the underlying assumptions and experts are informed by the moments we have today. And common models, but maybe it is a bit more complicated to think outside the box and what this would mean. Thank you, Susanne. So there is a question by Jose Ignacio Carrasco. And he's thanking you for the great talk. And his question is that he wants to know whether observed the observed large disagreement is due to acknowledged uncertainty of the future or to due to a greater amount of methods and forecast approaches. And it's probably due to a number of factors. One is uncertainty. So, what do we know now about drivers, you would think, you know, that this evidence space that there's not that much agreement people or experts come from different backgrounds. Interestingly, one factor that did not play a role, in fact, for now, is that it didn't really vary across the different backgrounds. So it didn't really matter whether an expert was an academic or more from traditional background. So it really seemed to be more the high level of uncertainty. And that surrounds those scenarios. And also, you know, we looked at certain data, but you also need to consider that there's shortcomings in all data sets that once we take into consideration, either to some regular migration, there's issues there. So, if you've looked at migration data, it's nothing that 100% reliable. So, the only as good as that data that he uses as a basis. Excellent. Thank you very much. So in the interest of time, I would like to hand the word to Andre who will talk about using online search data to predict migration. Yes. Thank you so much for having me here. It's my pleasure to share some recent work of ours today, which is on online search data to predict migration. And let me just say, so this is going to be a two part. So, first of all, recent work that we've already published and and I towards the end, I'm going to give a little outlook on what we're currently working on in a follow up project financed by the horizon 2020 program. Okay, so, so let me just remind you as an introduction that whenever we do quantitative migration predictions, what we need in order to train and use these models is some migration flow data. And predictors, as we call them, and you can think of those as the typical push or pull factors, right? These may be, for example, GDP in the origin and or destination country or demographics, right? So, problem being, and I think this has been pointed out a number of times already that often there's a lack of migration data. So, here, specifically, specifically talking about the flows, which is the outcome variable in our prediction model and reliable predictors. So these GDP data for demographics to train our prediction model. So, just to give you an idea, when it comes to migration flows, typically what we use is OECD data. And here, just to point out the lack or the weaknesses of this data is that often this is published only with a very decisive lack of up to two years. Then also, it is only available at least officially at the yearly level. So there's also quite a restriction on the frequency. And when it comes to the predictors, then of course, taking about GDP reliable GDP data and developing countries or employment statistics is very, very hard to get, if anything. So in recent work that we have published is already is that we use geolocated digital trace data. So specifically here, Google Trends. And I'm going to show you a couple of examples just to make clear what we're talking about if you may not know the Google Trends engine yet. And we use this data to predict a lateral migration flows. Just to give you an idea why Google Trends, you can see that quite impressively here, I would say, first of all, if you take a look at the left hand side graph, you see that this has very, very high time variation. So the frequency of this data and you can you can see that we extracted. These are the trends. This is the data of people searching for the term visa in Mexico and starting in 2004, which is the time from which these this data is available. And as you can see, this is you can you can go to a frequency down to the daily level, which is unprecedented in other type of predictors. On the other hand, we also have very nice, not complete but very high coverage. So looking around the world. We have data for many countries, including developing countries. And of course, why is Google Trends maybe the right engine to do that. It's because it's the most used widely used search engine worldwide right so whenever we talk about Google. This represents a market share of 80% 97% even on mobile devices. So this is quite frequency and and the cover test provides some very nice advantages to research. And what we're claiming or what we're saying is that this index reflects daily search intensities through Google for a given keyword and geographical area. So if we choose these keywords correctly or in the in the migration context, then we we can use it as a proxy for migration intentions. And I want just to give you a brief example. So here you see the the trends index for the terms immigration and United States in Mexico between 2016 and 17 can see this very impressive peak here in November 2016. Well, just what happened there exactly is that Trump got elected and our hypothesis is that migrants search for information so relevant information terms that are that they may reflect information retrieval for going to the United States. Online and prior to departure in their country of origin and of course you could say now well this could be just reflecting a general interest in in in that in in these terms which is driven by Trump's election. But also impressively if you take a look at the Google at the OECD migration data you see a peak in arrivals to the United States just shortly after this peak in in Google trends in Google searches for the term immigration in the United States. So recent work that has just been published in January this year. We developed this idea. So the article is called searching for better life predicting international migration with online search keywords and that's this joint work with to be a stir at the IFW and who is also presenting in this afternoon session as in the following session. So, the, just to give you the summary findings from this article well we find that the Google trends has strong predictive performance compared to the typical predictors say GDP and or population statistics in gravity benchmark models which are often used to do migration predictions. And this so just to kind of give you an interpretation of this of this article. This is a proof of concept. It does not rigorously implement practical forecasting yet, I would say this is think this is subject to work that we're currently developing. The promise, however, is that using Google trends to measure migration intentions and allows us to do now casting and short term forecast right so whenever sorry whenever I say prediction should have said that before I mean now casting and short term forecast and the terminology may be a little different to what we saw in the previous talk but so now forecasting is basically up to the current period and short term forecast are a few months or if any a year ahead of time. In the following slide I just want to show you a couple of graphs that kind of reflect the performance or the predictive power of this data that we're using. And what you see here is the evolution of immigration from Venezuela to Spain and predictions based on a classical migration flow model and an augmented one. So basically what you see here in the bold blue dark blue line is the total number of immigrants from Venezuela to Spain as reflected by the OCD data and two predictions that are based on a benchmark model which includes GDP and demographic indicators at the origin and destination and using that same model, but adding our Google Trend variables for migration relevant keywords and what you can see is that the dashed light blue line here is pretty much relatively flat and does not respond very accurately or closely to the total number of immigrants that the migration flows that you see here, whereas the dotted line here the magenta colored one is relatively closely following the evolution of the migration flows here. And now predictive the predictive power basically for us or said just to say it simply is the degree that this model simulates the flow of migrants from Venezuela to Spain here. And in other words, the smaller deviations between the solid line and the dashed or dotted line, the better the fit of our model and the better the predictive power. And by the way, just to point out here, you can see clearly the turnaround in the Spanish economy in 2011 2012, whereas also coinciding with the start of the crisis in Venezuela, which led to increases in migration flows towards Spain. And this turnaround has also been it's also been predicted relatively nicely by our Google Trends model, I would say. And of course, this is a bilateral just one bilateral corridor, you can expand that to other countries and show similar patterns that are overall taking, giving the oppression that as I said the classical predictors are relatively slow moving, as we know, and they don't have much of predictive power to in these models, whereas adding Google Trends for migration related keywords does do a much better job as reflected here by the difference in the dashed and the dotted and the solid line. And you can also bring this to the next level and look at corridors at say an aggregation of corridors here for six origin countries to 35 states of the U.C. We see similar patterns again, let me just point out interestingly that we have all that this also works if you take a look at the center lower panel for Spain. This also works within the U. So this is, these are flows, I say within the OECD these are flows representing from Spain originating from Spain going to the OECD of course, and that's particularly here for Pakistan if you take a look at the left lower panel, you see a lot of turning points so a lot of spikes, and here it works particularly well, relatively well of course there are arrows in these models still, but the turning points are nicely captured. So let me just give you a brief outlook because I'm running out of time I think. So next steps, as I said, this was a proof of concept study so far that have that has provided the groundwork for a new project which is called it flows it tools and methods for managing migration flows by horizon 2020 to collaborate this project with 40 nodes running just started this month basically so very fresh. ongoing work, led by our university. And the idea is to do finally implement these migration forecasting techniques using, including digital trace data using Google trends Twitter, etc. other of course here I focused on those and there are other big data sources or other digital trace data we could think of policy objective is to improve management capacity of migration flows during different stages of the migration process such as reception relocation and settlement. And let me just point out at the last slide here, a couple of challenges that we're facing of course again related to data limitations. One again is the availability and the frequency of training data. And here, particularly the migration flows so we're really hoping for as much as collaboration as possible. And on behalf of the member states of the United European Union to obtain higher than you the frequency migration data which is kind of one bottleneck that we're facing. And another thing is of course the keyword selection. There's a semantic approach which we followed in previous work that I just pointed out. Whereas new alternative strategies are crowdsourcing so say, looking at terms in Twitter or other social networks that migrants are effectively using to to include in the as a keyword. Time horizon, as always, just very neatly following up on the previous talks now casting so I so predicting up to the current period which is, if you have a lack of two years of migration, publishing, migration data publishing then this may already be very important and helpful versus short to medium or long term forecast right so I think what we're trying to do here is to do now casting and short term forecast so up to one year ahead of time. Last but not least, my taking migration routes into council in our work so far we have mostly taken the bilateral view and just looked at the origin versus the destination. Of course, as not migration, or especially refugee migration is not always that simple. And there may be several transit countries in, in along the routes that also, yeah, that also matter and that we will try to take into account here in that in this inflow project as well. So, thank you very much. I hope this was interesting and you can have further information here on these sites. Thank you. Thank you, Andre. What a fascinating project. I'm really intrigued to see how this develops, particularly how it moves from proof of concept to something else. We have one question by Susanna Falka Janssen from DG home, and she's asking what the margin of error is with the Google search method. So, as I said, when looking at, I think there, I need to, I need to differentiate a little bit so the, what we did so far in the proof of concept. I can, I can say that the, the error margins are much using Google trends are much lower than than just relying on the typical predictors, as I said, something like GDP per capita or and or population development in an origin and destination. When it comes to, to the migration implementation of these forecasts. I cannot, I cannot yet say much about the margin of error because this project has just started and we have not been able to systematically explore that but along the same lines of the proof of concept, I expect this to be much superior. So, including little trace data much superior to, to the typical prediction models that we're facing. I cannot yet quantify this. I'm sorry for that, but stay tuned. So we're going to, this is something on that we're currently working on. And we are very excited also to explore this more systematically. Super. Then I have one question by Freddie Norma, who's asking which keywords are used in Google trends to generate the graphs represented. And another question, which is also one that I wanted to ask from the league. What is the time lag between the intention to migrate for the Google search and the actual migration event. Very well, very good question. So, so I didn't mention this due to time limitations, but so the graph that I showed you, they're based on Google trends for. For 67 different keywords, as far as I remember, so we follow the semantic approach here and the idea was to use. Turns that are related to migration, semantically in the Wikipedia and Wikipedia. And, of course, we had to kind of limit up the number of terms that we were using. So, 67 was kind of a past dependent number. This, so this is basically based on such a large list. And to the second question. This is very interesting question because we don't know exactly, of course, so this is also relating to potential future research, I would say. So what is the exact lack time lag between people in the origin country searching, and finally, say, leaving that country and arriving somewhere at the destination is is say we don't we don't know yet due to time due to data of six is that we are facing. With the migration flows and the data, we were using 1 lack. So, the corresponding to 1 year between search and arrival. But of course, we would go, we would like to go much finer. I can imagine that this lag is very rise a lot in the function of the bilateral corridor. So, say, if we're talking about intra European migration, then this lack will probably be very small. Right. So, someone can can just freely decide to migrate somewhere else within the U. However, if we're talking about extra European into international migration, then of course this may be very long process and eventually a lack of more than 1 year would be the more accurate Right. So, we're currently working on exploring this empirically and understanding. Well, what type of lag is kind of predominant in which type of corridors. Super. Thank you so much. So there are 2 more questions, but I would like to park them for now. I'm in the interest of time and hand over to Jack who will present on dealing with uncertainty in migration forecasting across a range of time horizons. Jack, over to you. Thank you very much. Good morning to everyone. I'm delighted to be able to speak here also as a former IOM staffer who started working on migration forecasting 15 years ago. I work in IOM also. So, so it's also a homecoming for me. And this talk will be about uncertainty in migration forecast across a range of time horizons. And this is, this is part of a project that another horizon 2020 project quantity about quantifying migration scenarios that we've also recently started similarly to the one that Andre has been talking about. So, a big shout out to the European Commission. Thanks for funding all these migration related initiatives. And this is a piece of work based on a, on a paper that we produced recently with Matthias Stryker and all the references will follow in later in the presentation. So, to start with, I'd like to propose to look at migration uncertainty according to the main type. And the, we'd like to think about two types of uncertainty that draws from the, from the uncertainty which I say, which are one is epistemic. So one is related to the knowledge of the processes or lack of knowledge of the processes. And this is something that where we can, we can think that if, if there is an area of uncertainty that's related to imperfect knowledge, then it can be helped. No, some of these unknowns are at least in principle knowable if we do more research and try to shed light on that. But there is also a second type, the aleatory or intrinsic uncertainty, which is irreducible. So these are the unknowable, not only unknown but unknowable unknowns. And this distinction will be crucial in terms of the, what we can say about future migration and also in terms of the policy implications on. So what can go where we had a first go at trying to approximate the division of different types of uncertainty and different factors that come under the epistemic and aleatory had us. So the epistemic is everything that has to do with how we define, how we conceptualize migration, how we measure it through whatever imperfect instruments we have at our disposal. And also, some things to do with the drivers of migration with the imperfect, our imperfect knowledge of the drivers and the environment and configurations. And finally, the models. So how is migration represented? How are migration decisions drivers put together in a forecasting model? Then on the other hand, we have the aleatory part. And this is much more, much more tricky. So here we have all the unpredictable shocks, shocks to migration, shocks to migration drivers in the past instance. We also have step changes in data in modeling and the arrival of digital traces is an example of such a step change. This is something that has appeared quite recently on the horizon. Then we have the unpredictable human behavior, especially at the individual level. And then finally, since we're talking about the future, the future being open, it's fundamentally concerned. So the models and methods that Ravenna talked about this morning, the experts or expert-based methods or survey-based methods or extrapolative approaches or scenario-based ones, they all have uncertainty hidden in them, sometimes in quite different places. So for the experts and for the expert studies and survey-based approaches, what do we know? And this follows from the discussion on the chat that we already started when talking about Susanna's presentation. Expert judgment is also uncertain. So if you have a collection of experts, they will rarely agree on anything. So what comes into the model or comes into the picture is the uncertainty of the expert opinion, yet another source. Then we have the intentions. Ravenna mentioned the intentions to migrate. Andres Tok also alluded to that in the form of Google searches. The thing is that with intentions, they are necessary but not sufficient conditions for migration. So they do not easily translate to reality. They will translate into reality only for a small fraction of people who form an intention. The thing with extrapolations and early warnings and similar methods based on either statistical or econometric modeling is that they assume some stability of trends at some level. And if they use drivers, the drivers themselves are uncertain and the drivers themselves are very difficult to predict. So if we are not careful, we can end up trying to predict migration by using drivers which are even similarly unpredictable if not worse. And finally, the scenario-based approaches. So we've seen examples in Susanna's talk. The questions to ask in terms of trying to tease out the uncertainty of scenarios and narratives. Are they imaginative and coherent? Do they stretch the world of possible futures broadly enough? We can use simulation models. We can use micro-simulations to try to formalize the scenarios. But then they will be data-hungry and quite heavily assumption-driven. So here we introduce yet another source of uncertainty into the mix. And with drivers, as Susanna mentioned, it will come later also in the talk by Matthias. So the drivers are many varied and interrelated. And it's sometimes very difficult to tease out what impacts on migration, on any particular migration flow because it's so interrelated. This brings us to the question of prediction horizons. So how far ahead can we make meaningful statements about the future of migration? And in formal terms, in statistical terms, it comes back to the question of the concept of stationarity or non-stationarity. And many migration, if not most migration processes, exhibits what is called non-stationarity. So basically the process is changing on the time. Whenever a shock happens, this brings about a new migration equilibrium. And this happens constantly, which is one of the reasons why it is so difficult to predict migration. And it also means that the uncertainty increases, typically increases with the time horizon. And that has implications for the decision makers. So what we can really look at across the range of time horizons here is in the very short term, we can, as Ravenna alluded to in her talk, we can make statements that are probably more precise than the longer term ones, and they can be used for operational purposes. So this is the whole idea behind an early warning system. Try to detect changes in patterns, changes in processes as quickly as we can. In the short to mid-term, so a few years ahead, still, forecast can be used for planning, as long as the uncertainty in them is somehow acknowledged. But the long-term scenarios, the long-term visions, the narratives are crucial for the more high-level strategic and policy type of considerations. So the key challenge now becomes how to choose an appropriate method for the different horizon and for the different assets. And, of course, the two types of uncertainty, the aleatory and epistemic also play different roles across these different time horizons. So what I would like to conjecture here is that the returns from knowledge diminish. So in a sense, the role of the epistemic uncertainty decreases over time, but then the aleatory one, which is related to the complexity of the migration processes, which compounds the uncertainty, is increasing very rapidly. So you can sort of almost imagine that in the very short horizons, it's the epistemic part that is most crucial. What do we not know now versus, in the long run, it is all about the unpredictable things that may happen once we, on our way from now to the future. And this also maps to the choice of techniques and methods for different tasks. So here on this style of example of an immigration focus to Germany, you can see how narrow a horizon for short-term and now-casts and early warnings is. Beyond that, what we are faced with is an ever-increasing range of uncertainty. And beyond a few years' horizon, we can rely pretty much only on some imaginative scenarios and narratives of the future. So what about the data? What about the data? This is an interesting point because whatever we feed into our model and our assumptions and our scenarios is also biased and also birth errors. So this is yet another source of uncertainty. We can at least try to approximate it. We can try to measure the errors in the data. So there have been attempts to produce probabilistic estimates of migration flows. You know, the IMEM project that we run at Southampton a few years back is one example. Another is a full probabilistic matrix of flows estimated by John Azos and Adrian Rafteri last year. And of course, there is now a huge appetite for making the most of the digital traces and other big data. They're quite volatile, which means that they're useful in the short horizons as Andre's talk has quite convincingly illustrated. But ideally, what we'd like to do is also to couple them with traditional sources, which are better understood because they have been around for longer. But also this can help us make the most of the relative advantages of different types. So the timeliness of the digital trace data coupled with the appropriate benchmarking that comes from, I don't know, an established survey or a register. So there is quite a lot of work to be done in that area. In terms of levels of predictability, what we proposed in a piece of work that was published last year in general forecasting was to actually look at migration forecasting and migration forecasting uncertainty through the lens of a risk management approach. So to classify different types of migration flows according to the levels of uncertainty they exhibit. And note that in the world of migration, we are nowhere in the area of low uncertainty. So the leftmost column is empty. But that uncertainty being juxtaposed with the level of potential level of societal impact changes in a specific flow might have. So here is a British example with low, medium and high impact of different flows on the policy area. This was based on the work done for the Migration Advisory Committee through the perspective of the Home Office. But this is something that will be very much user specific. So depending on the policy question, depending on the user, this matrix may look different. So from the point of view of someone who works in operations and humanitarian relief, the matrix will look completely different than the one from the perspective of someone who looks at labor markets. This brings me to the question about how to measure predictability. So the standard ways of doing that is by looking at, by trying to get a handle on the errors, that sort of ex ante errors. So how large we expect the errors to be given the model that we run. So it's a statistical property of the model. But also we can look at exposed errors. So once we observe the reality, we can compare it with our prediction and see how the error actually manifested itself in the real life. So having these two pieces of information also can help us tell how well calibrated our models are. So how well do the ex ante and exposed errors align? Is something that we predicted was a 50-50 event over a longer horizon? Did it really happen about 50% of the time? This is the kind of question that we're looking at. And there has been quite a lot of work in statistics on designing so-called scoring rules that combine errors and calibration. So for example, one rule could be to minimize errors that are well calibrated. And this is yet to be propagated into the world of migration forecasting or population forecasting in more detail. The additional thing to bear in mind here is that the errors will have different meaning to different users. So for some users, it may be more costly to overestimate. For some users, it may be more costly to underestimate future migration. And this is something that we can also bring into the mix by using statistical decision theory and so-called loss functions. So there has been a little work being done in that area, but not much in practical application. So what are the options? In the short horizons and for the flaws that are somewhat better predictable, the epistemic uncertainty, so the one that's related to our imperfect knowledge tends to dominate. So what we can do here is to look at the sort of cost-benefit or risk-benefit assessment of different possibilities and bring in the decision analysis to help us and help the policymakers and users actually decide between different policy options. But what we can look at in longer horizons is much more limited because here is the aleatory. We are in the realm of the aleatory uncertainty. And here is the place for scenarios which can serve for what-if-type stress testing for designing contingency plans and going back to today's building capacity and resilience of the migration system and migration policy. There's of course a question of can we reduce uncertainty in migration forecasting and with the epistemic part, the answer is yes, but only for the epistemic part. So the aleatory part is something that will stay there no matter what. And we can think of better data. We can think about better domain knowledge, new research, better judgment. There's a nice book about super forecasting that shows that you can actually improve the judgment about the future events slightly by following some codified principles. But the main point is that the aleatory uncertainty will always be there. And it's something to acknowledge and to manage. So there are two things here depending on what we are talking about. We can either try to help reduce it or we just have to live with it somehow. The challenge here, of course, is to know which is which. And I'd like to finish on one of my favorite quotes from a British philosopher Calvert Reid who said, it's better to be vaguely right than exactly wrong. And with that, I thank you. Thank you very much, Jakub, for this somewhat sobering presentation and the view into reality, which I think is really important. And certainties are such an important driver of predictions. I have quite a few questions, but before I pause, and there's also one direct question to you Jakub from Manfred Kohler of the Austrian Ministry of the Interior. I would like to know whether you ever looked into social media platforms, which are by now also able to predict behavior have been used to predict behavior. And they might be able to reduce uncertainty. The aleatory uncertainty. Well, I think the the it's a very it's a very good point and it sort of goes back to the wider discussion about digital digital prices. And actually, we are looking at social media data with one of my PhD students and Francesco Rampatta, who is working with Facebook data. And the thing is that there is an interesting signal coming from these data sources. But it's it requires some careful treatment. So, so what you what you need to be mindful of that these these data are only valid for a subgroup of the of the population that that's the one that uses a particular social media themselves. And also that we don't know what the data really signify. So, so that there is a there is a variable on on Facebook that tries to ascertain whether someone is a migrant or not. But this is all this is all proprietary knowledge of Facebook and Facebook algorithms how this, how this indicator is defined and created. And we do not really know what is being measured. You can you can follow some sentiments analysis and and again, you know, this I can imagine that this will be something that very useful in the very short, very short horizons just to pick up some possible changes in the trend. And we are back in the early warning systems type of type of approaches, whether they will be given, especially given the high frequency of these data and volatility, whether they will be much of use in longer time horizons. I think I need to be convinced. I think at the moment I'm quite skeptical on that. But, you know, that that's definitely something to explore further. That's why that's a very interesting point. In fact, so we have another question by each. I'm sorry if I'm not pronouncing this name correct, as you would like to know what the techniques are to manage aleatory uncertainty. Is this even possible. She asks. And there is there is a there is a whole there is a whole literature on actually how to behave in the context of high and high unpredictability and though the sort of rampant aleatory uncertainty. That the usual that the usual precautions. Making sure that we have reserves that's costly, right, but that if we have spare capacity, we can easily accommodate shock whenever it comes, whatever the shot may be right we don't know when we don't know. We don't know what we don't know how, how large, but, but we can anticipate that something might happen. You know, COVID-19 is a case in point. Right. This is this is this is something that even though they were warnings about epidemics getting out of control for for quite a while now. No one could have reasonably predicted that that actually in 2020 the world will be in facing something like, like we are today. So, so reserve sharing the burden is another thing. So, so it's pulling the pulling the resources in order to be able to to respond to the challenges of the other answer. And this is, you know, this is a huge role to play for international organizations and and supranational bodies such as the European Union, such as the, you know, there's a there's a power in strength and in numbers. We can think about, you know, about about other things if you think about reinsurance business right reinsurance also the business of ensuring the insurance against catastrophic risk. This is something that that can with with with some careful thinking something similar could be translated into the policy world in the area of migration management. So that there are a few ideas about how to how to work with other uncertainty. The key point is that we cannot pretend that it doesn't exist. And I think that the pandemic, the pandemic actually helped hammer the point home. So, so now pretty much everyone recognizes uncertainty forward. Absolutely. I think that's a crucial point. We have another question from from yesterday, then I am Jim back and he would like to know whether you've ever looked into prediction markets where experts compete to make the most accurate predictions. And what your opinion is on those. Yeah, so we looked at them. I think the prediction markets have quite mixed performance we looked at them in the context of trying to predict referendum results in Britain. And so we have, we had, you know, one on Scottish independence that actually the prediction markets were quite successful and then on the, on the, you know, you membership the Brexit, in which the prediction markets were horribly unsuccessful. And I think in the, in the context of migration predictions where because the, the, the prediction markets work based on an assumption of basically frictionless information exchange. And, and, you know, hyper efficiency in how information spreads. I don't think we have this assumption is met in the, in the context of migration and also, even if it was, I think that the best we can hope for with this prediction markets is a short term horizon. So again, early warnings might be worth, might be worth having an eye on, you know, some betting markets, just to see whether there is any, any, you know, some, some uptick that may, may herald a change in trend. Another thing is that to make it work, people would have to bet with real money. So, so there have to be some real financial incentives and this incentives for taking part, because otherwise that's a purely academic in exercise it wouldn't work. So maybe there can be a research grant with this idea. Yeah. So, I would like to go into the panel discussion right now. And I have one question to start off, but please all participants are more than welcome to also ask some general questions to, to all the great speakers. We have online with us. So, in a recent conversation I had with Andre Kröger, he said, I haven't met anyone who can predict the future, not even economists and had really made me laugh. And now Jacob, you also said it's better to be vaguely right than exactly wrong. And I would like to maybe ask sort of the unaskable. Can we actually predict migration? Should we really be talking about predictions? For me, it would be really interesting to learn more about the limits of predictability. So we heard obviously about the big issue of uncertainty. But also about adequate policy responses, given these limitations. And maybe we can start in reverse order. So maybe Jacob Andre and then Susanne. Thank you. So I think it's helpful to realize that prediction is only the means, not the end. So it's crucial to start from the policy question. What do we need the prediction for? And this will then guide the choice of methods, the choice of possible data, and also illuminate the limitations. So the different types of methods that Ravenna mentioned in her talk, and I alluded to later on, that maps quite neatly onto the levels of policymaking. So from the low-level operational to the mid-level practical to high-level strategic policy aims. And of course, in a strict sense, prediction is impossible. But what we can get out of such an exercise is at least some form of an approximation of what the future might like. And then following through to the policy question, what different policy options might imply under different circumstances. So what, where it also can help is that the predictions can help us illuminate the trade-offs between different policy options, different social values that underpin them. You know, the thing about the freedom versus security debate that's currently with COVID pandemic is quite heavily featuring. And what we can do with predictions is to throw everything in the open and be more transparent about it as well. So we can stop here and then we can come back to the world and follow up. Yes. So thank you. I think Jacob already showed extensively and very convincingly that there's a trade-off between the time horizon and uncertainty. And I think this is exactly what you need to talk about. And then, of course, also the policy objective as you just said. So when it comes to, yeah, when it comes to Google trends and digital trade data, I think as we could see, this is something very helpful in the current horizon. So with now casting and or with short-term predictions, and that's what typically has a relatively low uncertainty band. So this is what I mostly believe in and or say this is what I would use this data for. And of course, once we go to a longer time horizons, then I think what I send to Julia in terms of my statement, I would stick to it. And especially, I mean, the best example, but I don't want to anticipate too much with Tobias talk later about COVID. But of course, as who would have guessed that we are in the situation that we are, and we're doing conferences online these days and instead of meeting in Vienna. And so this is precisely limiting completely migration, right? And there was also a question in the chat about whether and how we control for this migration policy or ease of migration. And I just answered in the chat as well that we didn't, at least in this recent work, but in ongoing work, we will certainly do this. And I think the current situation is the best situation to show us that this is actually very important, especially now this. Thank you. Thank you, Julia for the three interesting questions. And I think my other panelists have already alluded to a lot of very pertinent points. And the question is maybe not what exactly can we predict, but what can we do with the information we have when we look at scenarios to think about what are some possible effects. And so also to think about not just, you know, the policy field that maybe a person is usually working on, but also interrelated mess and connectedness with other policies. We've seen many countries where migration is looked at as one field economic approaches, labor or social issues. And I kind of looked at in isolation, the environment climate change, but it's important to look at what would happen if one of those factors changes if we have political population or not. How does that possibly affect others? And yeah, thinking about also what are the explanations of policy, where would we like to go and then what factors do we have to consider. And I know policy makers like to think or consider, you know, what's happening now and different regions of the world, where are they doing more movements be generated. But I think those scenario building exercises can really help us to think about, okay, but an ideal world where would we want to go? Would we want to be in 10, 15, 20 years? And what are some factors that pose more risk than others? We know about, you know, demographic changes that are more likely to predict than some other factors. We don't know what is going to happen completely. So those, yeah, it can be quite useful policy tool to start that discussion and really, yeah, consider some options of futures that could happen, even if they're as we've seen in the talks, considered very unlikely now they may become a reality next year. Thank you so much. So I will ask three questions now. You can choose which one you prefer to answer. We have six minutes left. So please limit your response to two minutes. So the first question is one from me, which is just what is the policy relevant of now casting. If it's already happening now, how quickly can policymakers respond? So I would like to see, understand this a bit better. Then there's another question from Tina Holt from the EML in Norway. She's asking, could you make some reference to the refugee crisis in 2015? Could Europe have been better prepared with better predictions? The third question coming from Manfred Kohler, which I think was already discussed in the chat that maybe you want to comment as well, was may disagreements among experts not also depend on ideological dispositions of researchers themselves. Maybe we can start with André, then Susanne and then Jakob. So I would just go for your first question, I think, and I think now casting, although it happens now and we may be able to observe it currently, I think it has a merit, especially when it comes to when thinking about the lack between the departure and the arrival. So discovering a signal in migration tensions as reflected by Google Trends in a certain origin countries, say, in North Africa or in the Middle East, will certainly not lead to direct increases in refugee immigration to the European Union. Since there is typically these migrants do not arrive by plane, but they choose other routes and they take time, right? So exactly by bridging, now casting can help bridging this gap, this time gap between departures and arrivals and therefore it has practical policy relevance for improving preparedness and management of refugee flows, particularly I would say. Thank you, André, that's very clear. Susanne? Yes, sure. On the question on the minister's interior on the, yeah, ideological background of the experts, I mean, we looked at, you know, whether the regional expertise, the academic background of the use of working on migration hadn't, you know, just to be able to kind of affect on how these experts waited certain scenarios, and we didn't really find anything specific that was a bit of a difference between maybe developers who are a bit more in agreement than sociologists and political scientists who were looking for qualitative approaches. But what we really saw is it didn't really matter whether some of these, a lot of experience and not with young or old, and it was quite useful to see that there is this variety of it. And of course, and others have moved to this, you know, a lot of underlying bias and assumptions that people bring, but we didn't use quite a large group of experts for the survey. So we did, you know, get a range of opinions on that. But that is, yeah, always part of it. And just to put you to the one question in the chat on whether the Delta survey affected the establishment of change of scenarios. We developed the scenarios first and then the Delta survey, but I think for future approaches, then we would take some other aspects more into conservation in the environment or certain prices or shops that we see and whether it would affect the scenario. Thank you. So, yeah, cool. I know I gave you the option to choose the question that's in your core panelists dodged the difficult question of the so-called refugee crisis. I would really kindly ask you to touch upon that. Absolutely. And actually, I think that it goes well together with the one about the relevance of now casting because they are they are they are interlinked. So could we be better prepared. One way is through through developing an early warning system that that could have helped us at least pick up the signal a bit earlier. So looking at something that changes on a more rapid scale than migration and Google intentions is a prime example or social media or anything else. But also, you know, so what what we could better do better to respond that that's going back to the previous discussion. We enter into the area of, you know, politics, political choice and and values and will right. So, so that was that was what was not not really visible in the early days of the crisis. And that's that's something that we could definitely do better next time. And in the in the sort of broader picture what what can we do is we can we can still try to reduce the uncertainty in by by looking at the regularities by looking at different data sources combining them them them them together. But in order to manage the next crisis, or the crisis after next, better, we really need to start thinking more in terms of, you know, preparedness, contingency planning, risk risk mitigation, all these, all these kind of all these kind of concepts are well known to the in the area of migration, but the discourse has to move into that into that direction. So I think I'll stop here just noting a small point on the experts and agreeing or disagreeing, even how uncertain migration is if the experts in the in the Delta survey agreed. Thank you. What a nice way to close this panel. Thank you so much to all of you. What a wonderful way to start this morning. I thought these were really inspiring presentations, very interesting discussions and questions. We will now take a coffee break. Please do get your coffee and tea, although we cannot invite you. We do hope to see you back at 1130 we will start start sharply at 1130 and very much look forward to seeing you back here in 15 minutes. Thank you.