1
An Agent-Based Model and Simulation of the
operational decision effects on the journey of a refugee
Master Thesis
MSc Supply Chain Management
Faculty of Economics and BusinessUniversity of Groningen
Student Name: Juraj Fodor Student Number: S3514099 E-Mail: j.fodor@student.rug.nl
Supervisor: Dr. O.A. (Onur) Kilic
Second Assessor: Dr. K. (Kirstin) Scholten
1
Abstract
How do the refugees react to the different operational decisions introduced by countries? To which extent do this decisions affect their decisions to move from a safe country or to choose a destination? Through the better understanding of the operational decisions can the effectiveness of humanitarian logistics be improved? This paper uses the agent-based model and simulation tools supported by a temporal network to examine these questions. For the simulation, the migration crisis in Europe from January 2015 until March 2016 was chosen. By simulating the real situation with all the applied operational decisions during the crisis the results have been compared with the real asylum applicants` number and the numbers of arrivals in Greece to validate the model. Besides this, several scenarios are presented to figure out what could happen if some of the decisions were not made or were made coordinated among the Europe Union countries.
2
Content
1 Introduction ... 3 2 Theoretical Background ... 5 2.1 Humanitarian Logistics ... 5 2.2 Migration Movements ... 6 2.3 Capacity Management ... 8 3 Methods ... 93.1 Scope of the research ... 9
3.2 Migration Distribution Network ... 9
3.3 Agent-Based Model (ABM) ... 11
3.3.1 Factors used in the model ... 12
3.3.2 Refugee distribution model ... 15
3.3.3 Refugee decision model ... 16
3.3.4 Model Simulation ... 18
3.3.5 Simulated Scenarios ... 18
3.3.6 Data Sources ... 19
3.3.7 Assumptions in the model ... 20
4 Results ... 21
4.1 Impact of the different factors and parameters ... 21
4.2 Simulation of the situation during the migration crisis ... 23
4.3 Simulation of other scenarios ... 26
5 Discussion and Conclusion ... 29
References ... 32
APPENDIX A: Long-term factors per country ... 35
3
1 Introduction
In the last number of years, there have been large refugee movements, due to many conflicts, all over the world. The large influx of refugees in 2015 caused serious issues in Europe and tested the ability to solve difficult issues by the European Union (EU). Thousands of refugees arrived at the European borders every day.
The challenge for the EU was to get the refugee flows under control and to coordinate them. This could be achieved through a better understanding of their behavior and of the factors that can influence it. The ability to predict and coordinate the refugee movements is important for reaching a higher effectiveness of humanitarian logistics. Activities such as planning, implementing and controlling the efficient, cost-effectiveness flow of and storage of goods and materials as well as related information, from point of origin to point of consumption for the purpose of alleviating the suffering of vulnerable people are parts of the humanitarian logistics (Kopczak & Thomas, 2005).
The understanding of their behavior could improve the coordination of the refugees and the distribution of them among the countries could also have led to the smoother resolution of the crisis in Europe. Unfortunately, the EU did not make use of this potential and their responses have been ineffective to the influx of the refugees. The problem was that countries tried to affect the refugee movements through different decisions. Among them was the introduction of border controls in the Schengen area, which is the border-free zone between the EU countries, building of fences and barriers at the borders and making announcements towards refugees without knowing the effects of them. These operational decisions led to the slowing down of the migration flows because the refugees had to search for longer and more dangerous ways to their destination. Nevertheless, it almost resulted in a breakdown of the border and security system in the EU (Tasch & Nudelman, 2016).
4 (ABM) has been applied to the model and simulate the refugee movements during a crisis. The ABMs are very useful but also complex tools, because they are composed of autonomous agents, which have certain behaviors and can interact between each other as is common in the real world (MacAl & North, 2010).
In the area of humanitarian logistics, the research was focused on how to manage the humanitarian aid or how to bring in a most effective manner the shipments to the affected areas (Lysann, Kunz, & Gold, 2015; McCoy & Brandeau, 2011). Furthermore, many research studies have observed the different factors that attract a refugee to a country or to travel onwards from a country. The researchers found that there is a significant relationship between these factors and the refugee behavior. It has been found that economic factors, network factors, geographic factors and political factors are the most influencing (Kuschminder, 2017; Lin Junie, Carley, & Cheng, 2016; Schott, 2017). Furthermore, refugee journeys have been investigated little in the past. BenEzer and Zetter (BenEzer & Zetter, 2014) have shown that there are four conceptual challenges in understanding refugee journeys. These challenges are the temporal aspect, drivers and destinations, the process of the journey and the wayfarers` characteristics.
In this work a generalized simulation approach to understand the impact of different events that occurred in Europe during the migration crisis on the refugee movements is presented. The simulation is built upon a temporal network. By adding the different events, that occurred during the crisis, to the network, the structure of the network changes. The contribution of the research is to better understand the relationship between the different operational decisions and events and the migration movements. This should help to accelerate the humanitarian aid and optimize the humanitarian logistics, what is the aim of this research.
The result of this research provides an overview of the impacts of different events on the refugee movements and can be used to understand of them and the created model applied in this research can be replicable to future migration crisis. This can make them an important tool for solving the crisis in a more effective manner.
5
2 Theoretical Background
In this section, humanitarian logistics and migration movements are discussed. The focus lies on the description of them and on the previous researches, which has been done in both areas. Furthermore, the capacity management is shortly discussed.
2.1 Humanitarian Logistics
The core of the humanitarian logistics is to provide aid to vulnerable people. Humanitarian logistics aim is to help affected people by using resources in an effective way during and also after a disaster (Van Wassenhove, 2006). It is a challenging research field because of the uncertainty and complexity which characterizes the different disasters (Seifert et al., 2015). According to Van Wassenhove ( 2006) is the disaster management composed of four different phases. The phases are mitigation, preparedness, response and rehabilitation. The first two phases occur before a disaster. In these first phases, it is important to avoid or to minimize the negative impacts of a disaster. The other two phases are after a disaster occurs. They deal with the short-term response to a disaster and the long-term reconstruction to bring the affected people back to his condition before the disaster occurred.
There were different researches done in the field of humanitarian logistics with the focus on refugees and migration. An overview of all the different researches has been provided by Seifert et al. (2015). They were observing and analyzing the researches between 1989 and 2016 with the main focus in humanitarian logistics and refugees and found 53 valid articles that deal with both topics. The articles were then grouped into four different groups. In the first group were articles assigned to performance measurements, in the second group articles assigned to logistics and operations, in the next group articles assigned to public health and in the last one articles assigned to human right and refugee protection. This overview provides a good source of information about what has been already researched in this area.
6 Another research investigated, how the efficiency of UNHCR emergency response to beneficiaries through appropriate central coordinated stock levels and shipping policies can be improved. The researchers focused on the needs of beneficiaries in refugee camps beside the efficiency factors. They constructed a simulation tool for the operational decision-making process, to optimal use the limited relief resources and needs of refugees (McCoy & Brandeau, 2011).
In the past, a lot of various research in the humanitarian logistics with a focus on refugees have been done. Most of the researches observed how to distribute humanitarian aid most effective to the refugees but no one focused on the fact how to take the migration flows under control and to coordinate them. This could improve the humanitarian logistics in different ways because for example the camps will not be built where the refugees will go but the refugees will be coordinated through different operational decisions to the refugee camps for example. Though this is possible just during long-term migration movement and not by uncertain short-term disasters like earthquakes or tsunamis.
2.2 Migration Movements
Migration movements are extensive events and it’s important for our understanding of them to figure out the causes of the migration, the size of the movement or the country of origin of the refugees. BenEzer and Zetter (2014) pinpointed in their research four conceptual challenges of a refugee movement. These challenges are drivers and destination, the process of the journey, the characteristics of the wayfarers and the temporal characteristics. Furthermore, as well as in different service settings, the demand for humanitarian aid or the flow direction during a migration crisis are unpredictable, uncertain and difficult to control. The reason for this is the behavioral aspect of the actors (Pansiri, 2014). All the characteristics of the migration movements make them complex events to research. Hence, little research has been done to investigate them.
7 assumption they held in their model is that refugees do not choose which country they are settled in. They move until they will be accepted by a country or until the capacity limit of a country has not been reached. In our model, the refugees can free choose the way they go to each country and based on different parameters they will choose their destination. Furthermore, capacity management approaches are set in the network that influence their decision or the way they choose to come to their destination.
In the research from Lin Junjie et al. (2016), an ABMS has been used to examine shifts in population network relations among countries, which influence overall population change. In their research, they examined several types of countries` networks and successfully explained migration and population shifts as a confluence of country networks.
Groen (2016) created an agent-based model for the exploration of the refugee patterns of refugee movements during the Mali conflict. The refugees in his model move with a defined probability to different places and want to arrive in one of the refugee camps in the safe neighbouring countries. To verify his model, he compared the results with the data from the UNHCR and found that the number of refugees in the refugee camps according to the simulation is slightly less than in the available data. What`s different to the model used in this research is that the model used by Groen was stable without any temporal changes in the network, that could affect the decision for the destination. The only parameter which affected the refugee movements was the capacity of the chosen refugee camps.
There is other research like the one from Suleimanova et al. (2017a), who used ABM to predict the refugee destination, or the automated framework agent-based simulation established by Suleimanova et al. (2017b) to enable research to construct simulations of refugee movements more quickly and systematically, who used ABM to understand the refugee movements. However, nobody in the previous research added into the ABM model dynamic factors that influence the movements and supported their research by a temporal network.
8 2.3 Capacity Management
The focus of capacity management is to ensure sufficient capacity to meet the demand (Pullman & Rodgers, 2010). Capacity management decisions are important by balancing the demand and the capacity of different environments (Grummit & Van Bon, 2009). The capacity management decisions for migration crisis can be, as mentioned in the introduction, in many ways handled as capacity management decisions by managing of services (Klassen & Rohleder, 2002). These decisions can be handled for long-term or short-term perspectives (Pullman & Rodgers, 2010). The effects of short-term capacity management decisions on the refugee journey are observed in this research. Short-term capacity management focuses on relatively short time periods and specific processes. During a refugee crisis the introduction of border controls, building of fences or political announcements can be seen as short-term capacity management approaches that affect the refugee flows.
9
3 Methods
In this section, a detailed overview of the scope of and data used for the research, research network and research models and their parameters can be found.
3.1 Scope of the research
This research focuses on the refugee journey to Europe. The journey the refugees make until they crossed the European borders is not taken into consideration. The focus lies just on the analyzing of the EU countries. According to this, no refugees stay in the countries that are not members of the EU. Nevertheless, the refugees must cross these countries, which influences the duration of their journey. For the analysis the Mediterranean route was chosen. This route was in 2015 the main route for the refugees to Europe (Frontex, 2018) and also the route which was most affected by different operational decisions during the observed period. The start of the journey is set in Turkey and the end is set as the destination in which the refugees decide to stay.
Refugees from different countries can have different drivers to travel onwards. They can travel because of conflicts, violence or a better economic situation. These drivers can affect the behavior of the refugees in different ways and it is necessary to differentiate between them (BenEzer & Zetter, 2014). According to this, the decision was made that in this research, the focus is not on all the refugees that came to Europe through the chosen route but only on the Syrian refugees. This will increase the reliability of our results because we focus on one particular group of refugees with the same reason to migrate and with the same nationality. The same nationality is import because in the model a factor related to it was used.
Not all the EU countries are included in the simulation. The reason for this is that there are other routes to Europe that the refugees preferred to take to these countries. For example, there is a small probability that a refugee who wants to go to Spain will travel through Greece because he could take the southern route through Africa. For this analysis, Spain, Portugal, Italy, France, United Kingdom, Ireland, Finland, Malta, Cyprus and Luxembourg were excluded but all other EU countries are included.
3.2 Migration Distribution Network
10 dynamics at all (Holme & Saramäki, 2012). Often the arcs between the points in a network are continuously active, these types of networks are called static networks. By temporal networks, it is important that the arcs between the points are transitive. This means that for example if country A is connected to country B and country B to country C then country A is directly connected to C as well in a static network, but in a temporal network if the edge between country A and B is inactive for a period of time then in the same period of time, country A and C aren’t connected as well. The time when the arcs are active or inactive can matter a lot and their correlations do have effects that go beyond what can be captured by static networks (Holme & Saramäki, 2012). Temporal networks are an extension of static networks and are not so easy to construct. The use of temporal networks is a relatively new field, but they have been used in many cases to better understand for example the spreading of infectious disease, opinions, rumors, in social networks (Holme, 2015).
11 attractiveness of a country for the refugees and is influenced by several factors. These factors are mentioned in the next section. The refugees can take each route just one time and the probability to go from one country that has a good reputation to one that has a very bad reputation will be zero because it is not expected that this decision could be made by a refugee. For example, if a refugee arrives in Austria he will not travel to Slovakia if all borders are open, in the case that the borders from Austria to Germany and Czech Republic are closed then he could decide to travel through Slovakia. A draft of the network can be seen below.
3.3 Agent-Based Model (ABM)
To model the movements the ABM tools were chosen for this research. The ABM is a powerful simulation modeling technique. This technique is used for real-world problems and systems with autonomous decision-making entities called agents. In an ABM the agents make decisions
12 based on a basic set of rules, it can capture emergent phenomena, it is flexible and provides natural descriptions of a system. Hence, it is an appropriate tool to observe flow management. The ABM is a useful technique when the interactions between the agents are complex, when space is crucial and the agents’ positions are not fixed when the interactions between agents are heterogeneous and complex, and when the agents exhibit complex behavior (Bonabeau, 2002). Following on from the mentioned characteristics, the ABM is a suitable technique to simulate and model migration movements.
3.3.1 Factors used in the model
The ABM model used in this research includes two agents. These agents are refugees and countries. Each of the countries has a reputation. The reputation determines how attractive the country for the refugees is and is calculated by several parameters. These parameters represent the pull-facts – the factors that attract a refugee to a specific country. The factors can be grouped into four groups:
• Economic Factor
Economic factors influence the refugee journey and the decisions to travel onwards to the largest extent (BenEzer & Zetter, 2014). The reason is that the refugees cannot find a job in the arrival country or the salaries are low and they travel onwards (Crawley et al., 2016). Keogh (2013) investigated the relationship of the country’s GDP in the EU-15 to the number of asylum applicants and found that a higher GDP has a positive effect on the number of applications. Economic factors going in hand with the living conditions, education and health-care system in the destination country (Crawley et al., 2016; Kuschminder, 2017). Kuschminder (2017) found in her research that the lack of employment in Turkey and Greece and bad living conditions are the main reasons for Afghan refugees to travel onwards from these countries. As a result, three economic parameters were chosen for determining the economic factor per country. These parameters are the GDP per Capita, Average Net Salary and Unemployment. GDP per Capita and Net Salary are calculated by the percentage of the average of the EU-28 countries in 2014.
• Geographic Factor
13 as well. How the duration between the countries was set you can find in the methodology under the network description.
• Network Factor
Previous research has shown that cultural ties and similarities, linguistic similarities, recognition rates or social networks have a positive impact on the attractiveness of a country. The reason for this is that refugees don`t feel foreign and the similarities to their home country help them to integrate faster in the country (Crawley et al., 2016; James & Mayblin, 2016; Kuschminder, 2017; Lin Junie et al., 2016). According to James and Mayblin (2016) moving from one country to another represents uncertainty and risk a for refugee. This can be decreased if they know somebody in the country. They can have a family member or a friend in the country. Middleton (2005) says that network factors stop acting as pull factors, and start deflecting future migrant numbers. The parameters - Refugee population, Syrian Asylum Applicants and Muslim Population were chosen to determine the influence of the network factors in the research.
• Political Factors
14 Below you can find a table with the list of operational decisions made during the migration:
Date Country Event Event type
20-22 August Macedonia Sealed its border to Greece and declared a state of emergency
Barrier
5 September Germany Angela Merkel announced that there are no limits on the number of asylum seekers Germany will take.
Announcement
13 September Germany Introduced temporary border controls with Austria
Border control
14 September Austria Introduced border controls on the borders with Hungary
Border control
15 September Hungary Closed their border with Serbia. Started to build a barrier along its border with Croatia.
Fence effect as a Barrier 17 September Slovenia Introduced temporary border controls on its
border with Hungary.
Border Control
16 October Hungary Announced the completion of the fence along its border with Croatia.
Fence
16 October Bulgaria An Afghan migrant was shot dead by a Bulgarian border guard.
Announcement
5 November Austria Began to build a barrier along a part of its border with Slovenia.
Barrier
11 November Slovenia Began to build a barrier along its border to Croatia.
Barrier
24 November Sweden Introduced temporary border controls. Border Control 4 January Sweden Introduced border controls on its border
with Denmark.
Border Control
4 January Denmark Tightened border controls with Germany. Border Control 15 February Bulgaria Announced that they will close all their
external EU borders and deport anyone who does not meet the criteria for asylum.
Announcement
17 February Austria Announced introducing border controls on its borders with Hungary, Slovenia and Italy.
15 9 March Balkan Balkan Route was closed. Slovenia, Croatia,
Serbia and Macedonia agreed that only migrants who plan to seek asylum in their countries or those with clear humanitarian needs will be allowed entry.
Fence
Table 1: Operational decisions made by countries towards refugee crisis from year January 2015 until March 2016
All the above-mentioned political restrictions, announcements and agreements have an impact on the country reputation as well. The political factors can be also used as handles for the demand management, because of the fast reactiveness on the migration flows. Security is also related to the politic factors of the country and has an impact on their attractiveness (Kuschminder, 2017).
The impact of each political factor is determined through analysis of his impact on the refugees’ arrivals in the affected countries before and after the introduction of the event. The factor is then multiplied by the reputation of the country for the fixed parameters.
In Appendix A table with an overview of all fixed factors per country can be found.
3.3.2 Refugee distribution model
Based on the four types of factors a refugee distribution model is developed to determine the flows. Each fixed parameter is multiplied by a weight. The parameters unemployment and distance have a negative impact on the reputation. From the fixed parameter the reputation per country fix is calculated. The fixed reputation per country is calculated by summation of the positive parameters per country multiplied with their weights divided by the sum of the negative factors multiplied by their weights.
Political factors are multiplied by determined weights but besides them, they are also with the fixed country reputation in countries where they have been introduced and where they have an impact. For example, an announcement made by a single country influences the reputation only of the one specific country but a barrier on a country border can extend the duration, not only to the one particular country but to other countries as well.
16 SAAi – Syrian Asylum Applicants per country (i)
RPOPi – Refugee Population per country (i) MPOPi – Muslim Population per country (i) DURi – Duration per country (i)
ANNi – Announcement per country (i) FENi – Fence per country (i)
BARi – Barrier per country (i)
TPBi – Temporal Border per country (i) Wj – Weight for the particular factor (j)
Refugee distribution model: 𝑅𝐸𝑃𝐹𝑖 =
𝐺𝐷𝑃𝑖∗ 𝑊𝑗+ 𝑅𝑃𝑂𝑃𝑖 ∗ 𝑊𝑗+ 𝑀𝑃𝑂𝑃𝑖 ∗ 𝑊𝑗+ 𝐴𝑆𝑆𝑖∗ 𝑊𝑗
𝑈𝑁𝐸𝑖 ∗ 𝑊𝑗 + 𝐷𝑈𝑅𝑖 ∗ 𝑊𝑗
𝑅𝐸𝑃𝑖 = 𝑅𝐸𝑃𝐹𝑖 ∗ (±𝐴𝑁𝑁𝑖 ∗ 𝑊𝑗± 𝐹𝐸𝑁𝑖 ∗ 𝑊𝑗± 𝐵𝐴𝑅𝑖 ∗ 𝑊𝑗 ± 𝑇𝑃𝐵𝑖 ∗ 𝑊𝑗)
The weights have been determined by using the solver in excel with the objective function to minimize the deviation of the asylum applicant per country in the year 2015. The reputations per country you can find in Appendix B.
3.3.3 Refugee decision model
17 The decision to travel onwards is not determined just by the number of refugees that arrive to their destination but also on the short-term events that have to be taken into consideration during the observed time period. If one of these events occurs, it can influence positively or negatively the number of refugees. The introduction of a barrier or temporal borders extends the time needed to arrive at the destination by one. The time in the simulation is set as an absolute number and represents days. For example, an introduction of border controls extends the duration of the journey by one day because the refugees have to wait at the border checks. How strong a decision affects the choice to travel onwards also depends on the reputations of the affected countries. When for example a border control at the border between Romania and Bulgaria is introduced, the effect of it is smaller than a border control between a Balkan state and Greece because this border check affects the traveling time to almost all destinations. In the case of the introduction of a fence, the decision to travel onwards is decreased following to the reputation of the countries to that the refugees cannot arrive anymore and following the time their journey is extended by due to the use of another way.
The following parameters are used in the model:
ARRt – Number of refugees that arrived at the final destination in period t REPi – Reputation per country (i)
ANNi,t – Announcement per country (i) in period t FENi,t – Fence per country (i) in period t
BARi,t – Barrier per country (i) in period t
TPBi,t – Temporal Border per country (i) in period t DEPR – Departure Rate
Refugee decision model:
𝑅𝐸𝐹(𝑡) = 𝐴𝑅𝑅𝑡−1∗ (𝐷𝐸𝑃𝑅 ± 𝑅𝐸𝑃𝑖 ∗ [𝐴𝑁𝑁𝑖 ± 𝐹𝐸𝑁𝑖 ± 𝐵𝐴𝑅𝑖± 𝑇𝑃𝐵𝑖])
18 and refugees, refugees interact with each other. When a refugee arrives, he motivates another one to travel onward. Though refugees cannot influence the decisions made by a country. This means that the model does not simulate for example, that if a number of refugees arrive at country A the country introduces a barrier or builds a fence at their borders. Furthermore, refugees cannot predict the decisions made by other agents and all the decisions made are fixed during the simulation.
3.3.4 Model Simulation
For the simulation Microsoft Excel is used. This software allows us to create a spreadsheet with all the necessary models and tools for the simulation. The simulation starts with a set of 300 refugees traveling onwards Europe from their starting point in Turkey. Every day when a refugee arrives to a country another travels onwards to Europe. After a short time, it leads to a continued flow towards Europe from Turkey. The model provides by changing of the weights of parameter a simple overview about the effect on the distribution of the refugees among the chosen countries and the arrival rates to Greece, also the arrivals per day and month can be observed from the spreadsheet.
3.3.5 Simulated Scenarios
To observe and better understand the relationship between the capacity management and the refugee movement different scenarios were simulated:
• Scenario 1: drawing the real flows from 2015 until 2016
• Scenario 2: excluding the introduction of the fence at the Hungarian border to Serbia that one of the countries determined as Balkan countries in the simulation and Croatia
• Scenario 3: without the German announcement about the “welcome policy” • Scenario 4: to reallocate the refugees to Eastern Europe and Baltic countries
19 The second scenario was chosen because Hungary was often criticized because of its attitude towards refugees and regarding the building of the fence at their borders to Serbia and Croatia. This scenario should show how it would affect the refugees’ movement in Europe if Hungary followed the Dublin convention and did not build a fence at their borders.
The next one was chosen to simulate the case, that Germany would not make the announcement that refugees are welcome. This political decision was the most controversial during the migration crisis because it increased the refugee influx greatly. Still this is a decision which is often a topic of discussion and there is no research that could provide an answer about the effect of it.
With the last one there was an attempt to redirect as many refugees as possible to the countries that are not preferred by them. Each of the countries has made several positive announcements in a row towards refugees that should increase their reputation. Nevertheless, in the attractive countries no decisions are made to avoid the refugee flow. All the decisions from the real case are used exclusive of the building of the fence at the Hungarian borders that made it impossible to come to almost all Eastern Europe countries and Baltic countries and the announcement made by Germany which made this a preferred country. The results from this model show if there can be a diplomatic way to relocate the refugees without force or resettle them to the less attractive countries.
3.3.6 Data Sources
20 The following secondary data about the long-term and short-term factors for the development of the model were chosen:
Data Source of data Period Country
GDP per capita Eurostat yearly per country
AW earnings Eurostat yearly per country
Unemployment Eurostat yearly per country
Muslim Population PEW 2010 & 2016 per country
Syrian FT applicants Eurostat yearly
2008-2014
per country
Refugee Population World Bank yearly per country
Duration Google maps/Reports and
Articles
Actual From country to country Data about operative
events/decisions
Different sources during the migration crisis
per country
Table 2: Overview of the used data
3.3.7 Assumptions in the model
To decrease the complexity of the model the following assumptions were made:
1. When a barrier or a temporal border is introduced, it extends the duration of the journey by a specific time. It can be expected that when a number of refugees come to a border control they will have to wait there until they will be allowed to cross the borders. The waiting time depends on the number of people and on the extent of inspection. For this research, the time was determined to one day.
2. If an announcement is made, or the temporal border controls are introduced by a country it affects the network for two months. It can be expected that an announcement will not have the same effect in the first few months as half a year later. Regarding the temporal border controls, there was no information available when the border controls have been stopped, due to this the two-month period has been chosen.
3. In Hungary, every fourth refugee who crossed their border stays because Hungary followed the Dublin convention which says, that responsibility for the refugees in the country in which they register first. This means that if the refugees crossed Hungarian borders on the legal routes, they have been registered there and if they moved onwards to other countries they could be sent back to Hungary (European Commission, 2018). The rate of 100% was not chosen because many of the refugees crossed Hungary illegally without registration or after agreement transported to the neighboring countries.
21 5. When there was no data or reports available about the duration of the journey from country to country, for each 500km, measured by Google_Maps by observing the distance from one country to another, the duration of one has been set.
4 Results
The results of the simulation are described in this section. Among them, the different weights for the chosen parameter are presented and the different cases that have been observed are discussed and compared to the real situation from 2015.
4.1 Impact of the different factors and parameters
Through the use of the excel solver for calculating the importance for each parameter presented in the previous section, the following results for the different factors have been captured:
Factor Parameter Impact
Long-term Factors Network
Factors
Syrian Asylum Applicants (2008-2014) 39,68 %
Refugee Population (2014) 29,62 % Muslim Population (2010) 0,01 % Economic Factors GDP per Capita (2014) 0,00 % Unemployment (2014) 30,14 % Net Salary (2014) 0,17 % Distance Factor 0,39 % Short-term Factors Political Factor Announcements 81,63 % Fence 60,60 % Barrier 13,31 % Temporal Border 10,10 %
Table 3: Parameters and their weights
22 The economic factors do not influence the decision on a large scale. Based on the provided calculation their impact is just a little bit over 30%. However, the previous research showed, as in the section discussed above, that economic factors have tremendous effects on the decisions made by refugees. In the table, it can be seen that GDP per capita and country does not have an effect on the country`s attractiveness and the net salary is just a weak one. The reason why this is the case in the research is that the network factors are correlated with the economic factors. These parameters are the number of Syrian Asylum Applicants between the years 2008 and 2014 and the number of refugees per country in 2014. Both parameters include the number of people that made their decision based on several parameters and one of these parameters could be the GDP per Capita or the Net Average Salary for example. To find the exact weights it is necessary to search for the economic data before the Syrian refugees traveled to the different countries or to exclude both correlated factors. For this research, the decision was made to use the latest data and to include both network parameters in the analyses.
According to the table, the distance factor is with the lowest impact. This is because of the correlation with the networking factors but also because of the fact that the refugees were not attracted to the countries that could be reached in a short time but more to the countries far away from Turkey, like Sweden or Germany, which were preferred due to the other observed reasons. This could be changed by the addition of the weather parameter to the model. As a result of this, it could be possible that distance would play a more significant part in the refugee`s decisions in the winter months. The summation of the long-term factors is 100%.
23 The last calculated factor in this research is the departure rate. The optimal departure rate calculated by the solver in Excel is 143,5%. This means that the number of refugees that decided to travel onwards from Turkey increases compared to the arrivals in the previous period by 143,50%.
4.2 Simulation of the situation during the migration crisis
In this section, the results from the real scenario with all the operational decision made from January 2015 until March 2016 are discussed. Firstly, to prove the reliability of the designed model the simulation results from the first scenario are discussed. The graph 1 provides an overview of the real development of the number of refugees that arrived in Greece in the observed period and the results from the simulation of the real scenario.
Graph 1: Arrivals in Greece - Real situation versus Simulation
The total number of refugees that arrived in Greece in the simulation in comparison with the real arrival number is slightly higher. There can be observed some monthly deviations from the real situation. What`s important to mention are the months August 2015 and February 2016, where, according to the simulation, less refugees arrived than in the real situation with a difference of about 10.000 people. Contrary to this the simulation shows that in the month of November the highest positive deviation was about 10.000 refugees. This can be the impact of different factors that were not been taken into consideration in this research. Such factors can be the weather or especially in February 2016, the fact that the refugees predicted the closing of the borders between the Balkans (Macedonia) and Greece and moved in bigger numbers to
20 000,00 40 000,00 60 000,00 80 000,00 100 000,00 120 000,00 140 000,00 N u m b er o f r ef u ge es Months
Arrivals in Greece
24 Europe. Besides these three bigger deviations from the real situation, the model has proven that it can simulate the arrivals in Greece from the real situation with small discrepancies.
Table 4 shows the number of the redistributed refugees among the countries. This table includes in the second column the real number of asylum applicants per country from the year 2015. In the third column the results gathered by the simulation and in the last one there is the difference between the results from the simulation and the real situation.
Countries Real situation Real flow simulation Errors to the real situation Germany 162 495 206 099 127% Sweden 51 310 75 460 147% Netherlands 18 690 26 880 144% Austria 25 015 32 233 129% Belgium 10 415 11 245 108% Denmark 8 585 17 475 204% Bulgaria 5 975 9 053 152% Poland 300 1 372 457% Hungary 64 585 60 249 93% Czech Republic 135 1 358 1006% Romania 550 2 819 513% Lithuania 10 81 810% Slovakia 10 87 870% Croatia 25 179 716% Slovenia 15 352 2347% Latvia 5 50 1000% Estonia 15 95 633% Greece 3 500 3 284 94%
Table 4: Number of Asylum Applications per country - Real situation versus Simulation
25 be that many refugees that arrived in their destination applied for asylum first in 2016 or they traveled to the countries that are out of the scope of this research. To compare the data in a better way the fraction of the real asylum applications was calculated by the overall asylum applicants number and this percentage was multiplied with the total arrivals in the year 2015. The results of the calculation you can find in the table 5 below and they were used for the comparison of the different scenarios as well.
Country Number of asylum applicants
based on arrivals from 2015
Errors to the real situation Germany 230 461 89% Sweden 72 771 104% Netherlands 26 507 101% Austria 35 478 91% Belgium 14 771 76% Denmark 12 176 144% Bulgaria 8 474 107% Poland 425 322% Hungary 91 599 66% Czech Republic 191 709% Romania 780 361% Lithuania 14 571% Slovakia 14 613% Croatia 35 505% Slovenia 21 1655% Latvia 7 705% Estonia 21 447% Greece 4 964 66%
Table 5: Number of Asylum Applicants equal to the number of Arrivals to Greece - Real situation
26 Hungary. There the deviation from the real situation reaches more than 30,000 refugees. This can mean that our assumption is that every fourth refugee who crossed the Hungarian border was registered and had to apply for asylum in Hungary was underestimated and could be set higher. Despite this, the results show that this model can be applied in the best way for countries with the same attitudes towards refugees or by the addition of this parameter to the overall model.
By analyzing the temporal network, it shows that the refugees react to the decisions by changing their routes and deciding on the fastest ones. After the introduction of a fence at the Hungarian border with the Balkan countries (Serbia) the refugees changed their route and crossed into Hungary from the Croatia borders. This led to the decision to extend the fence on the Hungarian borders. Besides this, the network shows that after the closing of the route through Hungary many countries stayed disconnected from the network. Countries like Slovakia, Poland and the Baltic countries only got linked to the network through Hungary. This shows that in a temporal network a disconnection of one link can have bigger effects on the whole network structure. 4.3 Simulation of other scenarios
27
Graph 2: Overview of the Arrivals to Greece - different scenarios
By observing the third and fourth scenarios where no announcement by Germany was made it is clear that the arrivals in Greece are lower in comparison with the real situation. Furthermore, it shows that the policy applied for in scenario 4, in that all the lower preferred countries make positive announcements during the whole year has not had a big effect on the arrivals. Besides the arrival numbers, the asylum applicant number for both scenarios resulted in a much lower total application in the year 2015 than in the real situation. Furthermore, in scenario 4 it can be seen that due to the announcements made by the countries the number of applications increased by 1000%. For example, in the case of Slovakia in the year 2015 just 15 refugees applied for asylum but if they changed their attitude towards refugees and made positive announcements towards refugees, the number could increase to almost 310 refugees what presents an increase of more than 2000%. 20 000,00 40 000,00 60 000,00 80 000,00 100 000,00 120 000,00 140 000,00 N u m b er o f r ef u ge es Months
Arrivals in Greece under different scenarios
28 Real situation Scenario 1 Scenario 2 Scenario 3 Scenario 4 Germany 249 825 206 099 207 651 120 973 121 537 Sweden 78 886 75 460 76 851 54 518 54 575 Netherlands 28 735 26 880 27 195 19 171 19 255 Austria 38 459 32 233 32 772 22 815 23 758 Belgium 16 012 11 245 11 381 7 991 8 026 Denmark 13 199 17 475 17 681 12 439 12 506 Bulgaria 9 186 9 053 9 175 6 660 9 196 Poland 461 1 372 2 620 1 208 1 936 Hungary 99 295 60 249 104 223 51 645 78 033 Czech Republic 208 1 358 1 373 929 2 217 Romania 846 2 819 2 871 1 949 2 394 Lithuania 15 81 190 66 336 Slovakia 15 87 182 65 309 Croatia 38 179 186 84 107 Slovenia 23 352 353 208 266 Latvia 8 50 115 43 213 Estonia 23 95 203 80 368 Greece 5 381 3 284 3 335 2 197 3 071 Total 540 615 448 371 498 357 303 041 338 103
29
5 Discussion and Conclusion
The model developed for this research proved according to the results of the simulations that it can predict the number of refugees based on different parameters. Under these parameters, we can count operational decision or events made by countries during the migration crisis in 2015 and the first quarter of the year 2016. This model is easily replicable for similar situations with comparable environments which is the biggest strength of it and parts of the model can be adopted for other purposes. In particular, the model to calculate the country`s reputation or the decision model could be reused in different contexts, such as economic migration or migration due to environmental issues. Besides this, the model also provides the answers about which way the refugees take their journey and how long the journey takes. This can help a lot with the allocating of humanitarian aid, refugee camps and volunteers on the way along their journey. The important thing is that this model was not developed for the simulation of the direct migration from the affected country but from a country where refugees are already safe but due to different reasons like job opportunities or the economic situation are attracted to another country.
The comparison of asylum applicants of the real situation with the real simulated scenario showed that there are deviations in the number of refugees that arrive in Europe and refugees that applied for asylum in one of the countries. The difference is larger than 190,000 which makes up almost one-third of the arrived refugees in Europe. As discussed in the result sections, the causes for such a high difference can be that the refugees applied for asylum first in the year 2016 or they did not apply for asylum and are in Europe illegally. This provides us with an overview of the border policy in the European Union. Due to the poor control of the external borders and refugee flows, many refugees are illegally in Europe and present a potential safety risk for the European countries.
30 the number of arrivals could have been lowered by almost one third and the resolution of the crisis could have been smoother and better coordinated. The results from the third scenario show that the introduction of the fence at the Hungarian borders does not have a big impact on the overall refugee arrivals number but makes clear that it influenced the number of asylum applicants in Hungary and in other Eastern Europe and Baltic countries. This has proven that a fence can be a useful tool to redirect the refugee movements but not a decision that could stop the refugee movements when this decision is not in collaboration with the neighbouring countries. If besides Hungary Slovenia had built a fence on their borders it would have led to the same effect as the introduction of the fence at the Balkan borders with Greece that caused the stopping of the refugee influx to Europe. Scenario four tested if the reallocation of the refugees could be made smoother and not forced by different fixed quotes. The simulation showed that with the changing of the attitudes in the countries through making positive announcements to refugees, the relocation of them could be fairer. The numbers for most of the countries increased rapidly but the fraction of the total number of refugees arrived in Europe is still small. This means that besides operational decisions it`s necessary to increase the long-term country reputation.
To recap, this research has proven that as in the service management, also in the relief management, especially during the migration crisis, the capacity management approaches can be an effective strategy to increase the coordination and control the flows. The results show that the better understanding of the effects of the operational decision could positively affect the issues during the migration crisis. Especially through better predictability of the causes of the decision, the uncertainty could be minimalized and the preparedness of the humanitarian aid or volunteers increased. Nevertheless, the simulation has proven that through the effective application of the operation capacity management decisions a better reallocation of the refugees among the countries could be reached.
31 lead to more exact results with lower deviations to the real situation. Though the complexity of the simulation would increase.
32
References
BBC. 2015. From Turkey to Sweden: Syrian migrant’s perilous journey - BBC News. BBC. https://www.bbc.com/news/world-europe-34118978.
BenEzer, G., & Zetter, R. 2014. Searching for directions: Conceptual and methodological challenges in researching refugee journeys. Journal of Refugee Studies, 28(3): 297–318. Bonabeau, E. 2002. Agent-based modeling: methods and techniques for simulating human
systems. Proceedings of the National Academy of Sciences, 99(suppl. 3): 7280–7287. Clay Whybark, D. 2007. Issues in managing disaster relief inventories. International Journal
of Production Economics.
Crawley, H., Düvell, F., Jones, K., McMahon, S., & Sigona, N. 2016. Destination Europe? Understanding the dynamics and drivers of Mediterranean migration in 2015. MEDMIG Final Report, (November): 1–84.
European Commission. 2018. Country responsible for asylum application (Dublin). European Commision - Migration and Home Affairs. https://ec.europa.eu/home-affairs/what-we-do/policies/asylum/examination-of-applicants_en.
Frontex. 2018. Migratory routes map. European Border and Coast Guard Agency. http://frontex.europa.eu/trends-and-routes/migratory-routes-map/.
Groen, D. 2016. Simulating refugee movements: Where would you go? Procedia Computer Science, 80: 2251–2255.
Grummit, A., & Van Bon, J. 2009. Capacity Management: A Practitioner Guide. Van Haren Publishing.
Hattle, A., Yang, K. S., & Zeng, S. 2016. Modeling the Syrian refugee crisis with agents and systems. The UMAP Journal, 37(2): 195–213.
Holme, P. 2015. Modern temporal network theory : a colloquium. https://doi.org/10.1140/epjb/e2015-60657-4.
Holme, P., & Saramäki, J. 2012. Temporal networks. Physics Reports, 519(3): 97–125. Independent. 2015. Germany`s Angela Merkel says no numbers limits to right to asylum.
33 James, P., & Mayblin, L. 2016. Factors influencing asylum destination choice: A review of
the evidence, 1–8.
Keogh, G. 2013. Modelling asylum migration pull-force factors in the EU-15. Economic and Social Review, 44(3): 371–399.
Klassen, J. K., & Rohleder, R. T. 2002. Demand and capacity management decisions in services: How they impact on one another. International Journal of Operations & Production Management, 22(5): 527–548.
Kopczak, L. R., & Thomas, A. S. 2005. From Logistics to Supply Chain Management: The Path Forward in the Humanitarian Sector. San Francisco, CA: Fritz Institute.
Kovács, G., & Spens, K. 2009. Identifying challenges in humanitarian logistics. International Journal of Physical Distribution and Logistics Management, 39(6): 506–528.
Kuschminder, K. 2017. Afghan Refugee Journeys: Onwards Migration Decision-Making in Greece and Turkey. Journal of Refugee Studies, (February).
https://doi.org/10.1093/jrs/fex043.
Lin Junie, L., Carley, M. K., & Cheng, S.-F. 2016. An Agent-Based Approach To Human Migration Movement. Winter Simulation Conference.
MacAl, C. M., & North, M. J. 2010. Tutorial on agent-based modelling and simulation. Journal of Simulation, 4(3): 151–162.
McCoy, J. H., & Brandeau, M. L. 2011. Efficient stockpiling and shipping policies for humanitarian relief: UNHCR’s inventory challenge. OR Spectrum, 33(3): 673–698. Middleton, D. 2005. Why asylum seekers seek refuge in particular destination countries: an
exploration of key determinnants. Global Commision on International Migration (GCIM). http://www.refworld.org/docid/42ce54774.html.
Nicholls, S., Amelung, B., & Student, J. 2017. Agent-Based Modeling: A Powerful Tool for Tourism Researchers. Journal of Travel Research, 56(1): 3–15.
Pansiri, J. 2014. Tourist Motives and Destination Competitiveness: A Gap Analysis
Perspective. International Journal of Hospitality and Tourism Administration, 15(3): 217–247.
34 review of current approaches. International Journal of Hospitality Management, 29(1): 177–187.
Schott, S. 2017. Migration: The Push & Pull Factors | Sciencing. https://sciencing.com/push-pull-factors-migration-8069131.html.
Seifert, L., Kunz, N., & Gold, S. 2015. Humanitarian supply chain management responding to refugees: a literature review.
Suleimenova, D., Bell, D., & Groen, D. 2017a. A generalized simulation development approach for predicting refugee destinations. Scientific Reports, 7(1): 1–13.
Suleimenova, D., Bell, D., & Groen, D. 2017b. Towards and Automated Framework for Agent-based Simulation of Refugee Movements. Winter Simulation Conference, 8247870.
Tasch, B., & Nudelman, M. 2016. Map of border fences and controls across Europe -
Business Insider. Business Insider. http://uk.businessinsider.com/map-refugees-europe-migrants-2016-2?international=true&r=UK&IR=T.
Van Wassenhove, L. N. 2006. Humanitarian Aid Logistics: Supply Chain Management in High Gear. Journal of the Operational Research Society, 57(5): 475–489.
Warren, R. 2015. Here Is The Long Route Many Refugees Take To Travel From Syria To Germany. BuzzFeed-News. https://www.buzzfeed.com/rossalynwarren/here-is-the-long-
route-many-refugees-take-to-travel-from-syr?utm_term=.xj0LWxNKpW#.fqVKZLWeqZ.
Yoo, E., & Koo, J. W. 2014. Love thy neighbor: Explaining asylum seeking and hosting, 1982-2008. International Journal of Comparative Sociology, 55(1): 45–72.
35
APPENDIX A: Long-term factors per country
dd Network Factor Economic Factor Distance
37
APPENDIX B: Reputation per country
This reputation is estimated from the economic, network and distance factors.
