Model misspecification, Bayesian versus credibility estimation, and Gibbs posteriors. (English) Zbl 1448.91261

Summary: In the context of predicting future claims, a fully Bayesian analysis – one that specifies a statistical model, prior distribution, and updates using Bayes’ formula – is often viewed as the gold-standard, while Bühlmann’s credibility estimator serves as a simple approximation. But those desirable properties that give the Bayesian solution its elevated status depend critically on the posited model being correctly specified. Here we investigate the asymptotic behavior of Bayesian posterior distributions under a misspecified model, and our conclusion is that misspecification bias generally has damaging effects that can lead to inaccurate inference and prediction. The credibility estimator, on the other hand, is not sensitive at all to model misspecification, giving it an advantage over the Bayesian solution in those practically relevant cases where the model is uncertain. This begs the question: does robustness to model misspecification require that we abandon uncertainty quantification based on a posterior distribution? Our answer to this question is No, and we offer an alternative Gibbs posterior construction. Furthermore, we argue that this Gibbs perspective provides a new characterization of Bühlmann’s credibility estimator.


91G05 Actuarial mathematics
62P05 Applications of statistics to actuarial sciences and financial mathematics
62F15 Bayesian inference
Full Text: DOI


[1] Alquier, P., PAC-Bayesian bounds for randomized empirical risk minimizers, Mathematical Methods of Statistics, 17, 4, 279-304 (2008) · Zbl 1260.62038
[2] Bailey, A. (1950). Credibility procedures. Proceedings of the Casualty Actuarial Society, Stockbridge, MA, USA. XXXVII 7-23 and 94-115.
[3] Berk, R. H., Limiting behavior of posterior distributions when the model is incorrect, The Annals of Mathematical Statistics, 37, 51-58 (1966) · Zbl 0151.23802
[4] Bissiri, P.; Holmes, C.; Walker, S., A general framework for updating belief distribution, Journal of the Royal Statistical Society: Series B (Statistical Methodology), 78, 5, 1103-1130 (2016) · Zbl 1414.62039
[5] Brown, L. D., Fundamentals of statistical exponential families with applications in statistical decision theory, 9 (1986), Hayward, CA: Institute of Mathematical Statistics, Hayward, CA · Zbl 0685.62002
[6] Bühlmann, H., Experience rating and credibility, ASTIN Bulletin, 4, 199-207 (1967)
[7] Bühlmann, H.; Gisler, A., A course in credibility theory and its applications (2005), New York: Springer, New York · Zbl 1108.91001
[8] Bunke, O.; Milhaud, X., Asymptotic behavior of Bayes estimates under possibly incorrect models, The Annals of Statistics, 26, 2, 617-644 (1998) · Zbl 0929.62022
[9] Dahalhaus, R.; Wefelmeyer, W., Asymptotically optimal estimation in misspecified time series models, The Annals of Statistics, 24, 3, 952-974 (1996) · Zbl 0865.62063
[10] De Blasi, P.; Walker, S., Bayesian asymptotics with misspecified models, Statistica Sinica, 23, 169-187 (2013) · Zbl 1257.62026
[11] Diaconis, P.; Ylvisaker, D., Conjugate priors for exponential families, The Annals of Statistics, 7, 2, 269-281 (1979) · Zbl 0405.62011
[12] Fraser, D. A. S., Is Bayes just quick and dirty confidence?, Statistical Science, 26, 3, 299-316 (2011) · Zbl 1246.62040
[13] Ghosh, J. K.; Ramamoorthi, R. V., Bayesian nonparametrics (2003), New York: Springer, New York
[14] Grünwald, P. D. (2012). The safe Bayesian: Learning the learning rate via the mixability gap. Proceedings of the 23rd International Conference on Algorithmic Learning Theory, Lyon, France. P. 169-183. · Zbl 1255.68086
[15] Grünwald, P. D., Safe probability, Journal of Statistical Planning and Inference, 195, 47-63 (2018) · Zbl 1383.62014
[16] Grünwald, P. D. & Mehta, N. A. (2019). Fast rates for general unbounded loss functions: from ERM to generalized Bayes. https://arxiv.org/abs/1605.00252.
[17] Grünwald, P. D.; van Ommen, T., Inconsistency of Bayesian inference for misspecified linear models, and a proposal for repairing it, Bayesian Analysis, 12, 4, 1069-1103 (2017) · Zbl 1384.62088
[18] Hong, L.; Martin, R., Discussion on ‘Credibility estimation of distribution functions with applications to experience rating in general insurance, North American Actuarial Journal, 20 (2016) · Zbl 1414.91200
[19] Hong, L.; Martin, R., A flexible Bayesian nonparametrics model for predicting future insurance claims, North American Actuarial Journal, 21, 228-241 (2017) · Zbl 1414.91201
[20] Hong, L.; Martin, R., A review of Bayesian asymptotics in general insurance applications, European Actuarial Journal, 7, 231-255 (2017) · Zbl 1394.62142
[21] Hong, L.; Martin, R., Dirichlet process mixture models for insurance loss data, Scandinavian Actuarial Journal, 6, 545-554 (2018) · Zbl 1416.91188
[22] Jewell, W. S., Regularity conditions for exact credibility, ASTIN Bulletin, 8, 336-341 (1975)
[23] Kalman, R. E., A new approach to linear filtering and prediction problems, Transactions of the ASME-Journal of Basic Engineering, 82, D, 35-45 (1960)
[24] Kleijn, B. J. K.; van der Vaart, A. W., Misspecification in infinite-dimensional Bayesian statistics, Annals of Statistics, 34, 2, 837-877 (2006) · Zbl 1095.62031
[25] Kleijn, B. J. K.; van der Vaart, A. W., The Bernstein-Von Mises theorem under misspecification, Electronic Journal of Statistics, 6, 354-381 (2012) · Zbl 1274.62203
[26] Klugman, S. A.; Panjer, H. H.; Willmot, G. E., Loss models: from data to decisions (2008), Hoboken: Wiley, Hoboken · Zbl 1159.62070
[27] Lyddon, S.; Holmes, C.; Walker, S., General Bayesian updating and the loss-likelihood bootstrap, Biometrika, 106, 2, 465-478 (2019) · Zbl 1454.62098
[28] Martin, R., False confidence, non-additive beliefs, and valid statistical inference, International Journal of Approximate Reasoning, 113, 39-73 (2019) · Zbl 1471.62236
[29] Müller, U. K., Risk of Bayesian inference in misspecified models, and the sandwich covariance matrix, Econometrica, 81, 5, 1805-1849 (2013) · Zbl 1291.62069
[30] Patilea, V., Convex models, MLE and misspecification, Annals of Statistics, 29, 1, 94-123 (2001) · Zbl 1029.62020
[31] Ramamoorthi, R. V.; Sriram, K.; Martin, R., On posterior concentration in misspecified models, Bayesian Analysis, 10, 4, 759-789 (2015) · Zbl 1335.62022
[32] Richardson, R.; Hartman, B., Bayesian nonparametric regression models for modeling and predicting healthcare claims, Insurance: Mathematics and Economics, 83, 1-8 (2018) · Zbl 1406.62143
[33] Schervish, M. J., Theory of statistics (1995), New York: Springer, New York · Zbl 0834.62002
[34] Syring, N.; Martin, R., Gibbs posterior inference on the minimum clinically important difference, Journal of Statistical Planning and Inference, 187, 67-77 (2017) · Zbl 1391.62251
[35] Syring, N.; Martin, R., Calibrating general posterior credible regions, Biometrika, 106, 2, 479-486 (2019) · Zbl 1454.62105
[36] Syring, N. & Martin, R. (2019b). Robust and rate-optimal Gibbs posterior inference on the boundary of a noisy image. Annals of Statistics, https://arxiv.org/abs/1606.08400. · Zbl 1454.62209
[37] Syring, N.; Hong, L.; Martin, R., Gibbs posterior inference on value-at-risk, Scandinavian Actuarial Journal, 2019, 7, 548-557 (2019) · Zbl 1422.91376
[38] van der Vaart, A. W., Asymptotic statistics (1998), New York: Cambridge University Press, New York · Zbl 0910.62001
[39] Wang, Z. & Martin, R. (2019). Model-free posterior inference on the area under the receiver operating characteristic curve, https://arxiv.org/abs/1906.08296. · Zbl 1441.62176
[40] Whitney, A., The theory of experience rating, Proceedings of Casualty Actuarial Society, 4, 274-292 (1918)
[41] Zhang, T., From ϵ-entropy to KL-entropy: analysis of minimum information complexity density estimation, Annals of Statistics, 34, 2180-2210 (2006) · Zbl 1106.62005
[42] Zhang, T., Information theoretical upper and lower bounds for statistical estimation, IEEE Transaction on Information Theory, 52, 1307-1321 (2006) · Zbl 1320.94033
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. It attempts to reflect the references listed in the original paper as accurately as possible without claiming the completeness or perfect precision of the matching.