Graph representations for aggregate insurance risk modeling and prediction

As a reinsurer, the partner organization backs risky and/or high face-value policies that are otherwise detrimental to insurance companies to underwrite. However, currently the available information on the policyholder is based on prior information stored in the company databases and on previous policies held or claims administered and/or what is presented on the forms provided to the partner organization underwriters. As such, the risk visibility to underwriters is minimal and does not allow underwriters to accurately gauge on how incoming policies can be priced. Underwriters, in general, view risks in different settings: family, occupation, physical activities, demographical and geographical. For instance, if members of a family are applying for high-value life insurance policies, underwriters would like to know that as a family, what is the accumulated risk of life insurance payout should anything happen? How are the different policyholders linked to each other (relationships, living at the same address, working at the same company, company risk profile, etc.)? Is a particular applicant engaging in risky physical activities (e.g., skydiving, parkour, etc.) that is otherwise not stated in the applications? Is the policyholder living in an area of high crime statistics?

Faculty Supervisor:

Carson Leung;Lorenzo Livi

Student:

Partner:

Munich Re

Discipline:

Computer science

Sector:

Finance and Insurance

University:

University of Manitoba

Program: