A Study on Catastrophe Bonds and Index Insurance

In the presence of natural disasters, such as earthquakes, hurricanes, and floods, insurance companies may face tremendous losses beyond their capacities. To address this situation, catastrophe bonds (CAT bonds) have been developed to transfer catastrophic losses from the insurance community to the capital market. Typically, insurance companies (or their agents) issue CAT bonds to investors seeking high-risk, high-return investment opportunities and are willing to reimburse the bond issuers in the event of specific catastrophe events. Index insurance is an instrument similar to CAT bonds but transfers risks within the insurance community, from the insureds to insurers. While these two instruments have been widely used in practice, they have several limitations that call for improvements in their design to enhance applicability.

In this project, we aim to (i) collect literature and real-life examples of CAT bonds and index insurance products, (ii) summarize common product design and pricing models, and analyze their limitations, and (iii) develop new models to address those limitations and implement them with case study demonstrations.

Supervisor: Wei Wei
Graduate Supervisor: Andres Medina Landeros

Actuarial Models for Cyber Risks

Cyber risks have been posing increasing concerns to both public and private sectors. While cyber insurance has naturally emerged as a market solution to mitigate cyber risk in the recent decade, its development is still in an early stage. The underdevelopment of cyber insurance market is attributable to the complex yet unknown nature of cyber risks. One major challenge is the potential dependence among cyber risks. Due to the cyber nature, the dependence could widely exist among a large scale of cyber risks regardless of locations. This poses substantial insolvency risk to insurance providers and thus discourage their participation. To make it worse, there is yet no sufficient historical data to uncover the dependence structure.

In this project, we aim to tackle this challenge from a physical simulation approach. That is, to employ cyber engineering techniques to generate reliable data and identify the root cause of dependence, and thus better capture the dependence characteristics. In a later stage, we anticipate utilizing the data and dependence characteristics to develop interactive actuarial models for pricing and risk management, with the aim to promote healthy development of the cyber insurance market and enhance the overall social welfare for all stake holders.

This project is a continuation of the IRisk project of “Modeling dependence of cyber risks and its applications”, conducted in Fall 2023. In the previous project, literature has been searched, and some preliminary models have been established and investigated. In this phrase, we shall continue to develop the models and enhance their applicability. The involvement of the previous project is not required but would be a plus.

Supervisor: Wei Wei
Graduate Supervisor: Jiajie Yang

Climate Risk Measures

The significant impact that extreme weather events have on the financial, insurance, agriculture, and insurance sectors is no longer news, and companies now prioritize physical risk and transition risk in addition to macroeconomic risk when managing their risk exposure. In light of climate change, it is important that companies develop their risk management tools, incorporating climate risk and existing scenario models for their analysis. However, there is no specific model that is widely adopted across sectors to perform scenario analysis related to climate change. With that in mind, the goal of this project is to investigate and create a model that reflects expert knowledge, research, and data in a systematic manner. The first step is to identify the sector of interest, for example, the agriculture sector and the crop of interest, or claim modeling for insurance companies, followed by the weather variables and the geographical region. Then we carried out a literature review by learning from existing literature on papers that referenced the impact of climate change on the subject of interest. Further, the different approaches and methodologies that are widely adopted across industries to perform scenario analysis related to climate change will be explored.

Supervisor: Tolulope Fadina

Decentralized Insurance Market Analysis (Continued)

Cryptocurrency risk is considered an emerging and dynamic area of concern, and most traditional insurers tend to steer clear of this market due to the scarcity of data and the difficulty in risk assessment. On the other hand, by leveraging the blockchain technology, decentralized insurance has become the pioneers to offer covers for hacks and exploits related losses, and gained significant attention because of its advantages in energy efficiency, transparency, and scalability. In contrast to traditional insurance where the insurance company serves as the central entity, decentralized insurance utilizes the Proof-of-Stake (PoS) consensus for risk and claim assessment purposes.

During the previous semester, our focus centered on examining the pricing and claim assessment frameworks employed by the majority of decentralized insurers in the market. We collected real data of transactions, coverage, claims, and various parameters from the leading decentralized finance (DeFi) insurer, Nexus Mutual. In this semester, we will continue our investigation of the decentralized insurance, with an emphasis on the inherent risks associated with decentralized products and protocols. Our approach will involve conducting thorough analyses for potential factors influencing cryptocurrency prices, pricing of the covers, and claim assessments. Eventually, we will integrate these components to gain a comprehensive understanding of the entire landscape of decentralized insurance.

Supervisor: Xiaochen Jing

Neural Networks from Scratch

This project creates a neural network from scratch using basic coding techniques with Python or R, specifically leveraging the basic library or packages, such as NumPy and basic R functionalities. The primary objectives include providing an educational experience to understand the fundamentals of neural networks, offering practical exposure to coding in Python and R, and enabling customization of the network architecture. The proposed methodology involves research, environment setup, data preparation, implementation of neural network components, activation functions, loss functions, training algorithms, testing, evaluation, and thorough documentation. Additionally, time permitting, the exploration of attention mechanisms and a few transformer-based architectures will be included.

Supervisor: Eric Icaza

Data Discovery and Consolidation

As data-driven decision-making becomes increasingly important in all fields, it is crucial to have a comprehensive understanding of the datasets at our disposal to facilitate research, analysis, and innovation across disciplines. The primary objective of this project is to conduct an extensive data discovery process within UIUC to identify and catalog various datasets that exist across different units, research centers, and government entities. By consolidating these datasets into a centralized repository, we can provide researchers, students, and faculty members with a unified platform to access a wide range of data for their projects and initiatives. We anticipate that this project will require dedicated personnel, access to relevant systems, perform data visualization, creating a relational database.

Supervisors: Zhiyu (Frank) Quan, Eli O’Donohue
Graduate Supervisor: Litong Liu