Advancing Actuarial Science: Leveraging Synthetic Data for Privacy-Preserving Modeling

The actuarial profession increasingly relies on detailed data for risk assessment and decision-making, yet balancing data utility with privacy remains a key challenge, particularly under the GDPR. Traditional anonymization methods often degrade dataset quality, hindering actuarial modeling. We propose a novel synthetic data approach using kernel density estimation to generate datasets that preserve the multivariate statistical properties of original actuarial data. We discuss use cases for synthetic data in actuarial settings and demonstrate the fidelity, as well as the privacy, of the data generated. Our approach can enable actuaries to leverage realistic datasets without compromising privacy, and conduct collaborative research while maintaining regulatory compliance.