ACTUARIES AND PREDICTIVE MODELING: PAST, PRESENT AND FUTURE
General (non-life, P&C) insurers operate in a data driven business where predictive models are essential. Insurers are constantly confronted with the challenges created by rapidly increasing technical and computer facilities for data collection, storage and analysis. Debates in society (e.g. the need for an affordable insurance product for young, safe drivers), internal risk management, marketing needs, changing supervisory guidelines (i.e. the European Solvency II framework), and the increasing demand for analytical modeling (e.g. in fraud detection) motivate the relevance of research in actuarial predictive modeling (or: insurance analytics). Formulating an adequate response to these challenges requires interaction between the insurance business and the fields of statistics, data analytics and insurance regulation. Moreover, the pricing and design of insurance contracts is currently undergoing major transitions driven by the growing impact of new technical developments (e.g. black box driving devices and smartphone apps enabling telematics insurance) and regulations at national and European level (e.g. the recent gender directive). This keynote lecture will give an overview of evolutions in predictive models used for pricing in the presence of small and large data, and different types of risk factors (e.g. factor, continuous, spatial and multi-level). We also present recent work on analyzing a telematics data set from a European insurer. This data allows motor insurers to use real driving exposure when pricing the contract. We build flexible claims frequency models combining traditional and telematics information and discover the relevance and impact of adding the new telematics insights. (The latter is joint work with Gerda Claeskens and Roel Verbelen from KU Leuven.)
ACTUARIAL SCIENCE NEEDS BEHAVIORAL ECONOMICS
Each of us knows a good friend who shows up very risk adverse and such that in the same time we always astonished to see her/him take thoughtless risks. This inconsistency calls us about the positive value on risk aversion measures. The common foundations of actuarial science and economic theory were constructed assuming that details about the functioning of the brain – the ultimate ‘black box’ –would never be known. As a consequence, fear and anxiety do not need to be modeled, as if an evaluation of risk aversion would suffice. The asifers’ defense (Milton Friedman and Leon Savage 1954) states pool player acts “as if” s/he knows geometry and mechanics. “As if” optimization comes from somewhere (evolution, learning, education, imitation…). But “as if” assumption may predict badly when conditions change (e.g., warped table, lighter balls, fatigue). Many important economic decisions will not necessarily be computed correctly : housing; marriage/divorce; children; education and career choices; violence (crime, war, terrorism); health and longevity; ….
It is time to stop talking biases and anomalies as the ultimate tailings of risk analysis. Unfortunately the financial and economic behavior are not only based on logical or rational foundations. Moral sentiments and emotions play a role that we have to incorporate into a behavioral approach of actuarial science.
UPDATE ON ACTUARIAL STANDARDS AND IMPACT ON MODELLING PRACTICES
The actuarial profession is giving a new push to Actuarial Standards as professional quality tools defining and helping harmonisation of best practices at European or global levels. As an example, the Solvency 2 directive article 48 is expecting the Actuarial Function to evidence relevant experience in line with professional standards.
The aim of the presentation is to review why we need standards and how they help the profession. It will be an opportunity to review what are the subjects currently under review at local, European and global level: Best Estimates, ERM, Modelling, ICAs etc… and will focus in particular on how new standards might impact current actuarial modelling practices.
DO WE NOW HAVE A FOURTH KIND OF LIE ?
Capital models have been used by European insurers for internal purposes for 15 or more years. Only with Solvency II did it become possible to use the model to determine the regulatory capital requirement. In this session we will consider the benefits of internal capital models, but also their limitations and dangers. The way in which the Solvency II internal model regime anticipates and tries to deal with these risks will be explained, focusing in on some particular examples whilst trying to maintain a high-level view of the challenges. Lastly, the speaker will put on his supervisory hat and share the challenges of supervising internal models, including some of the quantitative initiatives underway.
HETEROGENEITY IN A LIFE ANNUITY PORTFOLIO: MODELING ISSUES AND RISK ASSESSMENT
This talk focuses on some critical biometric aspects underlying risk identification and risk assessment for life annuity portfolios and pension funds. On the one hand, statistical evidence shows, in many populations, a deceleration in mortality increase at very old ages, in particular a non-exponential increase in the age-pattern of mortality. On the other, causes of this feature of the age-profile of mortality is a rather controversial issue. Nevertheless, a deceleration in the mortality increase can be explained by the (reasonable) assumption of heterogeneity with respect to mortality inside a population, and, in particular, in terms of non-observable risk factors, which can be represented, for each individual in the population, by his/her “frailty” level.
The presence of heterogeneity heavily impacts on the riskiness of a life annuity portfolio (or a pension fund), and hence should carefully be taken into account in the risk management process. In particular, appropriate parametric models can help in assessing the impact of heterogeneity among annuitants, while differentiated annuity pricing can mitigate the portfolio riskiness.