Stochastic Mortality Modelling for Small and Data-Scarce Populations: A Systematic Literature Review

Abstract

Stochastic mortality models developed for large national populations face challenges when applied to small populations with volatile death counts, short data series, and potentially biased parameter estimates. This systematic review synthesizes modelling strategies for small and data-scarce populations from an actuarial perspective, examining uncertainty quantification, model evaluation, and remaining research gaps. Following PRISMA 2020, records were identified through Semantic Scholar and citation snowballing. Of 40 unique records screened, 18 studies published between 2011 and 2026 met the eligibility criteria; included journal articles appeared in Scopus-indexed journals. Data were extracted using a structured form and synthesized thematically. Five categories emerged: reference-population and two-population models, credibility and Bayesian methods, analyses of small-population bias, subnational and small-area methods, and hybrids incorporating machine learning. Most approaches stabilized projections by borrowing information from larger, related populations. However, the reviewed machine-learning studies seldom reported forecast intervals, and none of the included studies used Southeast Asian data. The evidence supports selecting an appropriate reference population and reporting distinct sources of uncertainty separately when modelling mortality in small populations. By consolidating these approaches and their limitations, the review clarifies the methodological foundations of small-population mortality modelling and provides practical guidance for actuaries in data-scarce markets, including Indonesia. It also identifies uncertainty reporting and Southeast Asian applications as priorities for further research.