EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Metropolis‐Within‐Gibbs Bayesian Strategy for Forecasting Insurance Reserves With Claim Trend Covariates

Samuel Essamuah Assabil, Ali Abubakar, Richard Okyere

Journal of Probability and Statistics · 2026-01 · 2026 (1)

DOI: 10.1155/jpas/5839123 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Probability and Statistics, online erschienen am 15.11.2026. Verfasst von Samuel Essamuah Assabil und 2 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper develops a Bayesian claims reserving framework that combines macroeconomic covariates, catastrophe shocks and heavy‐tailed claim severity in a single predictive model. Inflation and interest rates enter through a low‐dimensional Fourier basis, claim severity follows a log‐Student‐ t law, and a latent Poisson process generates catastrophe‐induced heteroskedasticity; estimation is by Metropolis‐within‐Gibbs sampling. Because the log‐Student‐ t law has no finite claim‐scale moments, predictive means, standard deviations and TVaR are defined under an explicitly stated truncation, whereas quantile‐based measures such as VaR 0.95 need none. In‐sample, the model gives a total reserve of $3554.02 (SD $610.83) on a first benchmark triangle, against $3018.77 for chain ladder and $2925.00 for a standard Bayesian model; $20.5M on the classical 10 × 10 triangle, close to the de Jong estimate of $20.1M; and GHS 1,428,400 on Ghanaian nonlife data. Forecasting performance is assessed out‐of‐sample by truncating the first benchmark and Ghanaian triangles at a historical valuation date, re‐estimating every model and scoring forecasts of the withheld claims. Chain ladder attains the lowest continuous ranked probability score (a proper score for the whole predictive distribution) and absolute forecast error on both datasets. The proposed model attains a better log score than chain ladder on both, and 88.2% (15 of 17) coverage of nominal 95% intervals against 29.4% (5 of 17) for chain ladder on the first benchmark, with equal coverage on the Ghanaian data. Among nested Bayesian specifications, it has the lowest CRPS on both datasets but is not best on every metric. Macroeconomic covariates improve CRPS in Ghana but not on the shorter benchmark. The outputs support VaR 0.95 and TVaR 0.95 relevant to Solvency II and IFRS 17, although full regulatory implementation is outside the scope.

Zitieren

APA

Assabil, S. E., Abubakar, A., & Okyere, R. (2026). A Metropolis‐Within‐Gibbs Bayesian Strategy for Forecasting Insurance Reserves With Claim Trend Covariates. Journal of Probability and Statistics, 2026(1). https://doi.org/10.1155/jpas/5839123

Vancouver

Assabil SE, Abubakar A, Okyere R. A Metropolis‐Within‐Gibbs Bayesian Strategy for Forecasting Insurance Reserves With Claim Trend Covariates. Journal of Probability and Statistics. 2026;2026(1). doi:10.1155/jpas/5839123

BibTeX
@article{Assabil2026,
  author = {Assabil, Samuel Essamuah and Abubakar, Ali and Okyere, Richard},
  title = {A Metropolis‐Within‐Gibbs Bayesian Strategy for Forecasting Insurance Reserves With Claim Trend Covariates},
  journal = {Journal of Probability and Statistics},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/jpas/5839123}
}
Kostenlos starten →