Tokyo:

Abstract

This presentation will highlight the applications of AI in insurance claims management by focusing on two types of applications: (1) Natural Language Processing (NLP) – The unstructured text associated with claims file notes and documentation are a gold mine that can provide important information for may possible decision-support predictive models. How does NLP enable this gold mine to be tapped? (2) Computer vision – This technology will be evolving to support claims activities ranging from producing an accurate vehicle repair process and cost based on photos only; enhancing underwriting based on evaluating the condition of the property; documenting building contents; and estimating property damage from perils such as water, hail, and wind. Artificial Intelligence Applications in Insurance Claims, Jinjie Duan (30 minutes) A Tractable Method for Unravelling and Modelling Unobservable or Complex Dependence Drivers with Granular Data, Bernard Wong (30 minutes) Social Media Data and Catastrophe Warnings, Professor Zheng Sujin (30 minutes) Bio: Bernard Wong, PhD, FIAA, is Head of School, Risk and Actuarial Studies, at the University of New South Wales (Australia). Bernard is a Fellow of the Institute of Actuaries of Australia, a member of the Actuaries Institute Data Analytics Practice Committee, and serves on the Board of ASTIN. His current research interests span two main areas: modelling of insurance processes, optimal capitalisation policies for risk businesses, as well as the interaction between the aforementioned problems. His research has been recognised via the award of numerous prizes. Abstract: The estimation of claim and premium reserves is a key component of an actuary's role and plays a vital part of any insurance company's operations. In practice, such calculations are complicated by the stochastic nature of the claims process as well as the impracticality of capturing all relevant and material drivers of the observed claims data. In the past, computation limitations have promoted the prevalence of simplified, but possibly sub-optimal, aggregate methodologies. However, in light of modern advances in processing power, it is viable to increase the granularity at which we analyse insurance datasets so that potentially useful information is not discarded. By utilising more granular data that is usually readily available to insurers, model results and predictions may become more accurate and precise. The estimation of claim and premium reserves is a key component of an actuary's role and plays a vital part of any insurance company's operations. In practice, such calculations are complicated by the stochastic nature of the claims process as well as the impracticality of capturing all relevant and material drivers of the observed claims data. In the past, computation limitations have promoted the prevalence of simplified, but possibly sub-optimal, aggregate methodologies. However, in light of modern advances in processing power, it is viable to increase the granularity at which we analyse insurance datasets so that potentially useful information is not discarded. By utilising more granular data that is usually readily available to insurers, model results and predictions may become more accurate and precise. Unfortunately, detailed analysis of large insurance data sets in this manner poses some unique challenges. Firstly, there is no standard framework to which practitioners can refer and it may be challenging to tractably integrate all modelled components into one comprehensive model. Secondly, computation requirements are a material concern when processing such large volumes of data. Finally, analysis at this greater level of detail requires more intense scrutiny as trends and drivers that were previously masked by aggregation may emerge. This is particularly an issue with claim drivers that are either unobservable to the modeller or very difficult to model. We propose a Markov-modulated non-homogeneous Poisson model to overcome the above problems in the practical implementation of a detailed "micro-level" claim count model. We incorporate a flexible exposure measure to explicitly allow for known claim drivers while the hidden component of the Hidden Markov model captures the impact of unobservable or practicably non-modellable information. Theoretical findings are illustrated and validated in an empirical case study using Australian non-life insurance data in order to highlight the benefits of the proposed approach. This is joint work with Benjamin Avanzi, Greg Taylor, and Alan Xian. Bio: Dr. Zheng is Doctor of Economics, Director of the Center for Risk Management Research, China Academy of Finance and Economics; Central University of Finance and Economics, Associate Professor; and Member of the China Association of Actuaries. 2010-2011 Visiting Scholar, Department of Mathematics and Accounting, Michigan State University. She studies at Beijing Normal University, Renmin University of China and Nankai University. Abstract: Social networks are playing an increasingly important role as an early warning system, which will aid the rapid assessment of disaster and post-disaster reconstruction. First responders can use the streams of data generated by social media in addition to other information needed to estimate damage caused by an upcoming crisis event and implement future solutions. In this paper, we seek to answer the following questions: What kind of information combined with data mined from social network will lead to a more accurate result in terms of damage estimation? More specifically, is it geolocation or the data of capital loss? And what functions do social media serve in the early, middle and late stages of natural disasters respectively? Social networks are playing an increasingly important role as an early warning system, which will aid the rapid assessment of disaster and post-disaster reconstruction. First responders can use the streams of data generated by social media in addition to other information needed to estimate damage caused by an upcoming crisis event and implement future solutions. In this paper, we seek to answer the following questions: What kind of information combined with data mined from social network will lead to a more accurate result in terms of damage estimation? More specifically, is it geolocation or the data of capital loss? And what functions do social media serve in the early, middle and late stages of natural disasters respectively? Session 2: Cyber Risk ERM Successes and Failures, Raymond Cheung (45 minutes) Bio: Raymond Cheung brings more than eighteen years of experience in FinTech business strategy, actuarial and capital modeling, product development, merger and acquisition, credit ratings, fund management, as well as risk and compliance advisory roles. Raymond currently is an independent director for an SGX-listed entity, a director of a regulated remittance company, and the risk and strategy advisor to several InsurTech companies in Asia. Previously, he was the Regional Insurance Lead of Grab, and Chief Risk Officer for AIG Asia Pacific and Asia Capital Reinsurance Group. Raymond is an Associate Member of the Institute & Faculty of Actuaries (UK), and a Associate of the Singapore Actuarial Society Abstract: This session will track the development and maturity of ERM implementation – from merely a regulator-driven compliance model to a board-driven business excellence model. The presentation aims to cover the objectives and benefits of ERM, its development and trend over time from countries that have implemented ERM for several years, including success stories from the insurance industry as well as failures of ERM – and how to fix them. The presentation ends with a discussion of the future of ERM. Session 3: Insurance ERM and Solvency Update on ASTIN Working Party on Cyber Risk, Eric Dal Moro (60 minutes) This session will track the development and maturity of ERM implementation – from merely a regulator-driven compliance model to a board-driven business excellence model. The presentation aims to cover the objectives and benefits of ERM, its development and trend over time from countries that have implemented ERM for several years, including success stories from the insurance industry as well as failures of ERM – and how to fix them. The presentation ends with a discussion of the future of ERM. Session 3: Insurance ERM and Solvency Update on ASTIN Working Party on Cyber Risk, Eric Dal Moro (60 minutes) Bio: Eric Dal Moro is Head of Group P&C Reserving at SCOR. Eric has over 20 years of experience in reserving and risk management. Prior to SCOR, he worked in the consulting industry both in Paris and Zurich, and also for AXA in France, Japan and Italy in different functions. He also gained some good experience in credit and surety pricing when working at Swiss Re. Eric has been serving as the Chairman of ASTIN Committee of the International Actuarial Association between 2014 and 2017 and is an active researcher within the global actuarial community. Abstract: This presentation will provide an update on the ASTIN Working Party related to Cyber Risk. As cyber events have virtually no geographical limitations and can result in economic losses on a global scale, the assessment of return periods for such economic losses is currently debated among experts. The potential accumulation of consequential insurance losses is one of the major reasons why the (re-)insurance industry has limited risk appetite for cyber related risks. One way to increase the available cyber capacity of the risk transfer market and to achieve an additional atomization of this accumulation exposure are alternative risk transfer (ART) solutions involving the capital market via insurance linked securities (ILS), i.e. cyber cat bonds and parametric insurance and industry loss warranties (ILW). While indemnity triggered products already exist, index triggered ART solutions are currently “uncovered t