Credibility Theory Extensions for Sparse Data Environments: Applications to Nigerian Insurance Pricing

This study develops extensions of classical credibility theory tailored for sparse data environments characteristic of developing insurance markets and applies them to improve risk classification and premium rating in the Nigerian insurance context. Credibility theory, which provides a statistically optimal framework for combining individual risk experience with portfolio-wide mean estimates, is a cornerstone of actuarial premium rating. Classical Bühlmann and Bühlmann-Straub credibility models perform well in markets with rich historical data but produce unstable and unreliable estimates when applied to the small sample sizes and high noise levels typical of Nigerian insurance portfolios, particularly for specialist or regional risk classes. This study extends the classical credibility framework in three directions: a Bayesian nonparametric credibility estimator that places fewer restrictions on the prior distribution of risk parameters; a longitudinal credibility model that borrows strength across geopolitical zones using hierarchical spatial structure; and a transfer credibility approach that formally incorporates international experience data as informative priors calibrated to Nigerian risk environment differences. All three extensions are derived theoretically and validated on Nigerian motor and property insurance datasets from 2015 to 2023, with cross-validation performance comparisons against classical credibility and GLM benchmarks. Findings demonstrate that the hierarchical spatial credibility model achieves the best out-of-sample prediction accuracy on the Nigerian data, with performance gains largest for northern states with sparse local data. The study contributes original credibility theory extensions and a software implementation for sparse market applications, recommending adoption by NAICOM as part of minimum pricing adequacy standards.

Keywords: credibility theory, sparse data, Bayesian nonparametric, hierarchical model, actuarial pricing.