The Port of Rotterdam is the biggest port in Europe and with its great location an important place for trade between Europe and other continents. Record numbers of intercepted illegal goods show that criminals use this port to traffic illegal goods. A lot is still unknown about the criminal supply chain in the Port of Rotterdam and about the chance to catch illegal goods that are smuggled. Therefore this research provides insights into this criminal supply chain by looking into the effects of the distribution of resources by law enforcement agencies between methods that can be used to catch illegal goods on the chance to catch illegal goods.

This research looks at the criminal supply chain from South America to the Port of Rotterdam where it focuses on the smuggling methods used inside the Europe Container Terminals in the Port of Rotterdam. Criminals make use of four smuggling methods which are the pincode fraud method, the switch and pincode fraud method, the extraction method and the empty depot method. Law enforcement agencies have a scan and a surveillance method to catch illegal goods. The methods of law enforcement agencies can catch different smuggling methods and both have a chance to catch illegal goods smuggled with the empty depot method. Whereas criminals choose per illegal container which smuggling method will be used, this is not possible for law enforcement agencies. Therefore, a linear relationship between the part of resources appointed to a method and the accuracy of that method is assumed for the methods of law enforcement agencies.

An agent-based model is built to capture the complexity of this criminal supply chain and show the behaviour of the cat-and-mouse-like situation between criminals and law enforcement agencies. This agent-based model is combined with the Nash equilibrium known from game theory, as game theory can give insights into this situation using a mathematical framework. As these research methods have not been combined for the distribution of resources among methods in the Port of Rotterdam in earlier research, this research provides insights into how these research methods can be combined and whether they will provide similar results.

When combining the Nash equilibrium from game theory with the agent-based model it is expected that the distribution of resources for the two players, criminals and law enforcement agencies, in the agent-based model will eventually end up in or around the Nash equilibrium. This is because, in the Nash equilibrium, no player can get a higher expected payoff by deviating from the equilibrium. The players in the agent-based model update their distribution of resources every period of four weeks to adapt to the behaviour of the other player. This adaption is modelled with an updating rule. As literature indicated that the modelling of this updating rule can cause different behaviour in agent-based models, six different updating rules are tested in this research.

Given a chance to catch illegal goods of the scan of 0.18 for smuggling methods that can be caught by the scan and a chance of 0.8 for surveillance for methods that can be caught by surveillance, the Nash equilibrium is reached when law enforcement agencies appoint 40/49 of their resources to the scan and 9/49 to surveillance and criminals appoint 40/49 of their resources to the pincode fraud method and 9/49 to the switch and pincode fraud method and the extraction method combined. In the Nash equilibrium criminals will not appoint any resources to the empty depot method.

The six updating rules used in the agent-based model show different behaviours. Some updating rules cause the players in the agent-based model to appoint all their resources to only one method and others cause the players to end up in a cyclic pattern around the Nash equilibrium. The agent-based model also shows that by changing parameters of certain updating rules, the behaviour can change completely. For some updating rules the distribution of resources can get appointed according to an equilibrium which is not a Nash equilibrium as the payoffs of the methods are not equal to each other. Various difficulties arise when designing an updating rule. None of the six updating rules can guarantee the agent-based models to end up in or around the Nash equilibrium.

These difficulties include that resources should be distributed according to the payoff of methods according to game theory and should not be done according to the success rate of methods. Other difficulties arise when methods get appointed little or no resources. As it is the best response for players to distribute all their resources to one method when the other player is not playing according to the Nash equilibrium, the updating rule needs to be able to appoint zero resources to a method. Updating rules should also ensure that this method can be reappointed more resources when the other player changes its distribution of resources. Other difficulties include equilibria that are not a Nash equilibrium as mentioned above and the fact that players do not adapt in the same way when they do not have the same number of methods.

From this research, it can be concluded that in order to use an agent-based model for the situation between law enforcement agencies and criminals the knowledge about Nash equilibria should be considered while creating the model. This will not happen automatically as seen by the multiple difficulties this research showed. This research also shows that updating the distribution of resources by law enforcement agencies is important as they risk having an unnecessarily low chance to catch illegal goods when criminals adapt, while law enforcement agencies would not.

Therefore, it is recommended for law enforcement agencies to update the distribution of resources when it is expected that criminals will adapt as well, to not risk having an unnecessarily low chance of catching illegal goods. Furthermore, it is recommended to consider the Nash equilibrium while updating the distribution of resources and to carefully observe the distribution of resources of criminals when appointing all resources to only one method.

Limitations of this research include that there are only a limited number of methods included and players can not learn new methods. Due to the limited data about the criminal supply chain and about the methods that can be used by law enforcement agencies, there is deep uncertainty in model parameters, especially in the chances of catching illegal goods of the methods of law enforcement agencies. Therefore the results are not entirely valid and should only be used to analyse expected behaviour. Further research is needed on these chances of catching illegal goods and to test more updating rules as this research does not provide an updating rule that will ensure that an agent-based model will always end up in or around the Nash equilibrium.

Illicit supply chain networks are not well mapped. Regulators do not know how goods flow from supplier to retailer. There is uncertainty about whom is involved, where goods originate from, and what quantities are being shipped. Simulation models are effective tools to find measures against the distribution of illicit goods such as personal protective equipment. However, simulation models often solely handle uncertainty through the variation of parameters. Structural uncertainty, which is uncertainty in the structure of the model, is often neglected. This study focuses on accounting efficaciously for structural
uncertainty in supply chain simulation models using model-driven exploratory modelling.

Model composability, which is a specific form of model driven exploratory modelling, is used in this study. The methodology is applied to a supply chain of illicit personal protective equipment. Using a model composer, many plausible models are generated of this supply chain. A model composer works by coupling model components in different configurations, while complying to preset constraints. Model components are submodels of a supply chain actors, for example, a retailer. Constraints help to restrict the way the model components can be coupled, making sure that every model generated by the model composer is plausible.

A ground truth is established to test the model composer on its efficacy to account for structural uncertainty. A ground truth is a simulation model of an illicit supply chain that functions as a benchmark. Five sets of 100 models are generated by the model composer to estimate the ground truth. Each set of models is generated with a different set of constraints. A constraint set consists of elements such as the maximum number of suppliers, the locations of supply chain actors, and the maximum number of customers of a supplier. These sets reflect different perspectives on an illicit supply chain.

Results show that structural uncertainty can result in significantly different simulation outcomes. The time in system, the production time, and the international transport time depend the most on changes in the constraints of the model composer. The time in system, the production time, and the international transport time of the models generated by the model composer are significantly different from the ground truth. The distributions of these outcomes have a different shape and have a wider range of possible values. Therefore, this study shows that model composability, a specific form of model-driven exploratory modelling, is efficacious in accounting for structural uncertainty in supply chain simulation models.

In the future, the methodology shown in this study can be used to model structural uncertainty in other fields such as water pipes networks, gas pipes networks, and telecom networks. Furthermore, the methodology can be used to identify robust measures to tackle the problem of illicit supply chains. Another recommendation is to use model composability for the individual components of the model. For example, a component such as a retailer can be build from several components: a cash register, a shelf, and a distribution area.

Trust has been identified in previous research as an important factor in criminals their decision-making when smuggling illegal goods. This research utilizes a case study of the criminal supply chain from South America to the port of Rotterdam, for simulating the relevance of trust in the criminal supply chain using an agent-based model. The agent-based model is able to simulate criminal decisions according to two decision-making models: (1) the risk vs. gain trade-off, and (2) transaction-cost theory. Interpersonal trust affects these trade-offs by minimizing risks and costs. The results show an insignificance of trust in the criminal supply chain for certain parameterizations of the model, when gains are relatively high compared to risks and/or costs. Different levels of trust have no effect on overal criminal supply chain performance. Furthermore, the results highlight the possibility of indirectly undermining trust within criminal networks when applying supply chain interventions, that hinder the smuggle of illegal goods. This implicates a possible reinforcing effect of supply chain interventions through the trust channel, on disrupting criminal supply chains when gains, risks, and costs are more in balance.

Being able to timely access and share accurate information to all stakeholders that can influence a supply chain is referred to as supply chain visibility. Supply chain visibility is crucial for increasing the resilience of a supply chain as this helps to proactively plan and design interventions to undesirable events. Various methods are available to increase supply chain visibility, often by increasing the accuracy of estimating future events. Many of these methods are relianton (large volumes of ) data, which is problematic because data on supply chains is often not accessible, expensive to acquire, and difficult to process. Therefore, methods to increase supply chain visibility in situations with limited data availability are needed.

 Modeling and simulation can be applied to situations with limited data availability, but has a limited predictive power for complex systems. Data assimilation in discrete event simulation (DES DA) is a method to increase the predictive power of modeling and simulation. This method aims to approximate system states based on real-time measurements to increase the accuracy of simulation models. DES DA has not yet been applied to supply chains and the dependency on data availability still needs to be explored. The goal of this research isto develop a DES DA algorithm that can accurately estimate future events in supply chains and test this algorithm for different levels of data availability. This will give insight in the relation between data availability and the accuracy of the proposed DES DA algorithm. During the development of this algorithm the typical supply chain challenges of high dimensionality and future state estimation are addressed.The proposed DES DA algorithm shows to have a higher predictive power (accuracy) than the results of a simulation exercise without assimilation. To allow for maximal reproducibility and facilitate future work, all adjustments to the simulation algorithm are extensively described in the body of this thesis.

Analysis of the performance of the proposed algorithm under different levels of data availability shows that limited data availability does not necessarily result in less accurate estimations. Less observations can can result in similar or even more accurate estimations. However, this can only be the case if the missing data is not part of the set of crucial data sources. If the missing data points are part of these crucial data sources the  accuracy of the data assimilation algorithm detoriates and the algorithm starts to under fit. Under fitting occurs when a model uses input variables that are not significant enough to determine a meaningful relationship between the input and output variables. 

If limited data availability is the result of omitting data sources that are not part of the crucial data sources, the accuracy of the data assimilation algorithm can increase. Using less datapoints helps combatting the curse of dimensionality which results in more accurate estimations. This research has shown that crucial data points for supply chains are the data sources that serve as arrival sensors of entities at the system boundaries. The arrival sensors reduce the uncertainty of future state estimations by accurately estimating the number of the entities in the system and the moment of arrival of these entities. Especially the accurate estimation of the moment of arrival reduces the variation in the estimations of the particles and thereby increases the accuracy of the estimation of future events.

The growing import of illegal products in the Netherlands via container ships causes increased violence, addiction, and tax evasion in the Dutch society. With the current criminal detection practices the Dutch National Police and Dutch Customs do not manage to intercept the majority of the illegal shipments. The resources used by the Dutch Police and Customs for criminal detection do not yet entail information from the criminal risk vs gain trade-off point of view. Therefore, this study set out to research the impact of the criminal risk vs gain trade-off on the international criminal supply chains.