Actuarial Applications of Graph Theory and Networks
Explore how graph theory and network analysis apply to actuarial problems on Exam MAS-I.
Graph Theory Fundamentals
A graph G = (V, E) consists of a set of vertices (nodes) V and a set of edges (connections) E. Each edge connects two vertices. In an undirected graph, edges have no direction: if vertex A is connected to vertex B, then B is also connected to A. In a directed graph (digraph), edges have orientation: an edge from A to B does not imply an edge from B to A. Weighted graphs assign a numerical value (weight) to each edge, representing cost, distance, strength of connection, or some other quantity.
The degree of a vertex is the number of edges attached to it. In a directed graph, the in-degree counts incoming edges and the out-degree counts outgoing edges. A path is a sequence of vertices where each consecutive pair is connected by an edge. A graph is connected if there is a path between every pair of vertices. A connected component is a maximal connected subgraph. A tree is a connected graph with no cycles, and it has exactly n - 1 edges for n vertices.
The Adjacency Matrix
The adjacency matrix A is an n x n matrix where A_{ij} = 1 if there is an edge between vertices i and j, and A_{ij} = 0 otherwise. For weighted graphs, A_{ij} equals the weight of the edge. The adjacency matrix is symmetric for undirected graphs. Powers of the adjacency matrix have useful interpretations: the (i, j) entry of A^k gives the number of paths of length k from vertex i to vertex j.
The degree of vertex i equals the sum of row i of the adjacency matrix. The Laplacian matrix L = D - A, where D is the diagonal matrix of degrees, has properties important for spectral graph theory. The eigenvalues of the Laplacian reveal information about the graph's connectivity: the number of zero eigenvalues equals the number of connected components.
Graph Metrics and Centrality
Several metrics characterize graph structure. The diameter is the longest shortest path between any two vertices, measuring the "width" of the network. The clustering coefficient of a vertex measures the proportion of its neighbors that are also connected to each other, capturing local connectivity. A high average clustering coefficient indicates a network where groups of nodes tend to form tightly connected clusters.
Centrality measures identify the most important nodes in a network. Degree centrality simply counts connections. Betweenness centrality measures how often a vertex lies on shortest paths between other pairs of vertices; high betweenness nodes are "bridges" whose removal would disrupt communication. Eigenvector centrality assigns importance based on connections to other important vertices. PageRank, the algorithm behind Google search, is a variant of eigenvector centrality for directed graphs.
Fraud Detection Applications
Insurance fraud detection is one of the most prominent actuarial applications of network analysis. Fraud rings often involve multiple connected parties: a claimant may be linked to a specific attorney, medical provider, body shop, and witness who appear together across multiple claims. Building a network of connections between claim participants reveals these patterns. Clusters of highly interconnected entities that appear across multiple claims are flagged for investigation.
Network-based fraud detection goes beyond analyzing individual claims in isolation. A claim that appears legitimate on its own may become suspicious when its connections to other claims are revealed. The network structure provides information that claim-level models miss. Centrality measures identify the key organizers of fraud rings: the vertices with the highest betweenness centrality are often the coordinators who connect otherwise separate parts of the network.
Reinsurance and Systemic Risk
The global reinsurance market can be modeled as a network where insurers and reinsurers are vertices and reinsurance contracts are edges. The structure of this network affects systemic risk: if a major reinsurer fails, the shock propagates through the network to all its cedants. A highly concentrated network (where a few reinsurers absorb most of the risk) is more vulnerable to systemic failure than a diversified network.
Contagion models study how defaults or losses propagate through networks of interconnected financial institutions. These models use the adjacency matrix and edge weights (representing exposure amounts) to simulate cascading failures. Stress testing involves removing a vertex (simulating a default) and analyzing the impact on connected vertices. Network topology determines whether a shock is absorbed locally or amplifies through the system.
Other Actuarial Applications
Social network effects influence insurance markets in several ways. Purchasing behavior is affected by peer influence: people are more likely to buy insurance if their social connections have it. Moral hazard may be affected by network norms: communities where insurance fraud is socially accepted may have higher fraud rates. Agent and broker networks affect distribution strategy: understanding the structure of relationships between agents, brokers, and clients helps insurers optimize their distribution channels.
Network analysis also applies to understanding claim causation in complex systems. In product liability or environmental liability, the chain of causation from manufacturer to end user can be modeled as a directed graph. In healthcare, the referral network between providers affects treatment patterns and costs.
Common Pitfalls
A common mistake is over-interpreting network visualizations. A graph that looks "interesting" (lots of connections, apparent clusters) may not be statistically significant. Always compare observed network metrics to what would be expected under a random graph model. Another error is confusing the adjacency matrix with the weight matrix in weighted graphs, or applying unweighted algorithms to weighted networks without modification.
Study Tips
Exam MAS-I introduces graph theory concepts and their potential applications in actuarial data analysis and risk assessment. Focus on understanding the basic definitions (vertices, edges, degree, paths, components), the adjacency matrix, and centrality measures. Know how these concepts apply to fraud detection and reinsurance network analysis. Practice computing basic graph metrics from small example graphs. Be prepared to explain how network analysis provides information that traditional claim-level analysis misses.