AI has had many summers and winters. Proponents have overpromised, and there has been hype and disappointment. In recent years, however, we have watched with awe, surprise, and hope at the successes: Better than human capabilities of image-recognition; winning at Go; useful chatbots that seem to understand your needs; recommendation algorithms harvesting the wisdom of crowds. And with this success comes the spectre of danger. Machine behaviours that embed the worst of human prejudice and biases; techniques trying to exploit human weaknesses to skew elections or prompt self-harming behaviours. Are we seeing a perfect storm of social media, sensor technologies, new algorithms and edge computing? With this backdrop: is AI coming of age?

In Finance and Actuarial Science, the multivariate elliptical family of distributions is a famous and well-used model for continuous risks. However, it has an essential shortcoming: all its univariate marginal distributions are the same, up to location and scale transformations. For example, all marginals of the multivariate Student’s t-distribution, an important member of the elliptical family, have the same number of degrees of freedom. We introduce a new approach to generate a multivariate distribution whose marginals are elliptical random variables, while in general, each of the risks has different elliptical distribution, which is important when dealing with insurance and financial data. The proposal is an alternative to the elliptical copula distribution where, in many cases, it is very difficult to calculate its risk measures and risk capital allocation. We study the main characteristics of the proposed model: characteristic and density functions, expectations, covariance matrices and expectation of the linear regression vector. We calculate important risk measures for the introduced distributions, such as the value at risk and tail value at risk, and the risk capital allocation of the aggregated risks.

Throughout the past couple of decades, the surge in the sale of equity-linked products has led to many discussions on the evaluation and risk management of surrender options embedded in these products. However, most studies treat such options as American/Bermudian style options. In this article, a different approach is presented where only a portion of the policyholders react optimally due to the belief that not all policyholders are rational. Through this method, a probability of surrender is obtained based on the option moneyness and the product is partially hedged using local risk-control strategies. This partial hedging approach is versatile since few assumptions are required for the financial framework. To compare the different surrender assumptions, the initial capital requirement for an equity-linked product is obtained under a regime-switching equity model. Numerical examples illustrate the dynamics and efficiency of this hedging approach.

In this paper, we develop a methodology to automatically classify claims using the information contained in text reports (redacted at their opening). From this automatic analysis, the aim is to predict if a claim is expected to be particularly severe or not. The difficulty is the rarity of such extreme claims in the database, and hence the difficulty, for classical prediction techniques like logistic regression to accurately predict the outcome. Since data is unbalanced (too few observations are associated with a positive label), we propose different rebalance algorithm to deal with this issue. We discuss the use of different embedding methodologies used to process text data, and the role of the architectures of the networks.

In most industrialised countries, one of the major societal challenges is the demographic change coming along with the ageing of the population. The increasing life expectancy observed over the last decades underlines the importance to find ways to appropriately cover the financial needs of the elderly. A particular issue arises in the area of health, where sufficient care must be provided to a growing number of dependent elderly in need of long-term care (LTC) services. In many markets, the offering of life insurance products incorporating care options and LTC insurance products is generally scarce. In our research, we therefore examine a life annuity product with an embedded care option potentially providing additional financial support to dependent persons. To evaluate the care option, we determine the minimum price that the annuity provider requires and the policyholder’s willingness to pay for the care option. For the latter, we employ individual utility functions taking account of the policyholder’s condition. We base our numerical study on recently developed transition probability data from Switzerland. Our findings give new and realistic insights into the nature and the utility of life annuity products proposing an embedded care option for tackling the financing of LTC needs.

Variable annuities have become popular retirement and investment vehicles due to their attractive guarantee features. Nonetheless, managing the financial risks associated with the guarantees poses great challenges for insurers. One challenge is risk quantification, which involves frequent valuation of the guarantees. Insurers rely on the use of Monte Carlo simulation for valuation as the guarantees are too complicated to be valued by closed-form formulas. However, Monte Carlo simulation is computationally intensive. In this paper, we empirically explore the use of tree-based models for constructing metamodels for the valuation of the guarantees. In particular, we consider traditional regression trees, tree ensembles, and trees based on unbiased recursive partitioning. We compare the performance of tree-based models to that of existing models such as ordinary kriging and generalised beta of the second kind (GB2) regression. Our results show that tree-based models are efficient in producing accurate predictions and the gradient boosting method is considered the most superior in terms of prediction accuracy.

Raw mortality data often exhibit irregular patterns due to randomness. Graduation refers to the act of smoothing crude mortality rates. In this paper, we propose a flexible and robust methodology for graduating mortality rates using adaptive P-splines. Since the observed data at high ages are often sparse and unreliable, we use an exponentially increasing penalty. We use mortality data of England and Wales and model male and female mortality rates jointly by means of penalties, achieving borrowing of information between the two sexes.

The classical credibility theory is a cornerstone of experience rating, especially in the field of property and casualty insurance. An obstacle to putting the credibility theory into practice is the conversion of available prior information into a precise choice of crucial hyperparameters. In most real-world applications, the information necessary to justify a precise choice is lacking, so we propose an imprecise credibility estimator that honestly acknowledges the imprecision in the hyperparameter specification. This results in an interval estimator that is doubly robust in the sense that it retains the credibility estimator’s freedom from model specification and fast asymptotic concentration, while simultaneously being insensitive to prior hyperparameter specification.

It is well known that the presence of outliers can mis-estimate (underestimate or overestimate) the overall reserve in the chain-ladder method, when we consider a linear regression model, based on the assumption that the coefficients are fixed and identical from one observation to another. By relaxing the usual regression assumptions and applying a regression with randomly varying coefficients, we have a similar phenomenon, i.e., mis-estimation of the overall reserves. The lack of robustness of loss reserving regression with random coefficients on incremental payment estimators leads to the development of this paper, aiming to apply robust statistical procedures to the loss reserving estimation when regression coefficients are random. Numerical results of the proposed method are illustrated and compared with the results that were obtained by linear regression with fixed coefficients.

Conditional mean risk sharing appears to be effective to distribute total losses amongst participants within an insurance pool. This paper develops analytical results for this allocation rule in the individual risk model with dependence induced by the respective position within a graph. Precisely, losses are modelled by zero-augmented random variables whose joint occurrence distribution and individual claim amount distributions are based on network structures and can be characterised by graphical models. The Ising model is adopted for occurrences and loss amounts obey decomposable graphical models that are specific to each participant. Two graphical structures are thus used: the first one to describe the contagion amongst member units within the insurance pool and the second one to model the spread of losses inside each participating unit. The proposed individual risk model is typically useful for modelling operational risks, catastrophic risks or cybersecurity risks.