Search Presentations
STAY TUNED! If you are anticipating additional search filters by attribute and level to align with the CAS Capability Model, it is coming later this Summer. As the CAS begins to code recorded sessions by specific attributes and levels (starting with the 2023 Annual Meeting), these will be tagged in the CAS database of presentations going forward and should be searchable.
But you may use the Capability Model now to help you identify topics. For example, if you want to move up one level under the content area “Functional Expertise,” you may search topics in the particular functional area to expand your knowledge.
Recorded content is searchable by Capability Model attribute and level in the CAS Online Library.
Refine Your Search
The presentation materials are offered in connection with CAS professional education offerings. © 2022 Casualty Actuarial Society. All Rights Reserved. The presentation materials may contain copyrighted content the use of which has not been specifically authorized by the copyright owner. You are permitted to view and print the materials for personal/professional noncommercial research purposes. Except for the foregoing, you agree not to reproduce, distribute, modify, create derivative works, or commercially exploit the presentation materials without prior written permission from CAS. Please direct any copyright permission inquiries regarding use of the presentation materials to acs@casact.org.
STAY TUNED! If you are anticipating additional search filters by attribute and level to align with the CAS Capability Model, it is coming later this Summer. As the CAS begins to code recorded sessions by specific attributes and levels (starting with the 2023 Annual Meeting), these will be tagged in the CAS database of presentations going forward and should be searchable.
But you may use the Capability Model now to help you identify topics. For example, if you want to move up one level under the content area “Functional Expertise,” you may search topics in the particular functional area to expand your knowledge.
Recorded content is searchable by Capability Model attribute and level in the CAS Online Library.
How "Smart" is Your Competitive Intelligence?
Most business professionals will tell you that one key to running a successful business is understanding your competition. Who are they? What is their product offering? What are they charging?
Competitive analysis is one of the key elements in measuring the performance of a rate algorithm. Over the years, much work has been performed in capturing the data necessary to do competitive analysis; however, there has often been a significant lack of sophistication in analyzing it. For those companies that have invested significant time and energy in the analysis, questions arise regarding return on investment.
This session will begin by discussing the sources and challenges of acquiring good competitive information. We will then explore different ways companies leverage competitive information and offer recommendation to extract the most value from this business intelligence. There will be a focus on how to generate actionable insights that improve companies’ pricing decisions. This session will also utilize live polling to better quantify themes such as:
• What business unit is responsible for competitive intelligence?
• Do you foresee investing more or less in competitive intelligence in the next two years?
• How confident are you in the premium information produced by your competitive intelligence unit?
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Kevin Donnelly
Panelists:
Klayton Southwood, Howard Kunst, Drew Lawyer, Adi Bar-lev
Product Architecture
Many P&C insurance companies are developing modular product architectures to facilitate better speed to market and consistency in product development. A product architecture is a comprehensive mapping of the components, dimensions, and rules of an insurance product with a focus on isolating the reusable assets. Especially when combined with leading product management practices and emerging technologies, the use of product architecture can greatly enhance a company’s flexibility in launching new products and reducing the workload associated with product enhancements.
This session will focus on practical applications and lessons learned, where a panel of product managers and actuaries will share experiences in developing and implementing product architectures across commercial, specialty, and personal lines.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Kevin Donnelly
Panelists:
Kelly Cusick, Kathy Olcese, Bharat Govindaluri, Brendan Smyth
Information Graphics for Actuaries (Live Streaming)
This session will cover visualization principles and examples, e.g. what charts to use, how to scale, how to use color, how to use text, etc. This will extend beyond mapping and use real examples to demonstrate how to create data visualization exhibits.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Kevin Donnelly
Panelists:
Margaret Brinkmann, Cody Webb, Garrett Bradford
Machine Learning: A Primer for Actuaries
In a 2001 paper, famous statistician Leo Brieman classified statistical modeling in two classes: one, stochastic data model (- requires a complete understanding of the underlying data generating process) and two, algorithmic models (- treats DGP to be unknown). He argued that if "our goal as a field is to use data to solve problems, then we need to move away from exclusive dependence on data models and adopt a more diverse set of tools". In this session speaker will give a practitioners level overview of these "diverse set of tools" along with philosophical motivations and an illustrative case study.
Speaker will use a hospital readmission dataset with an objective of predicting patient's likelihood of coming back to hospital during the first 30 days of last release. He will first talk about the good-old techniques e.g. General Sessionized Linear Models along with regularization concept (lasso, ridge and elastic net) - something philosophically fascinating and practically tremendously useful - as it is widely used for variables selection and dealing with multicollinearity.
The goal of the next section is to show how machine learning captures non-linear relationships in the data and also, incorporates interactions between two or more predictors. Speaker will first introduce classification and regression tree (CART) - an important building block of many modern machine learning techniques and then, finally will talk about two most important ML techniques: Gradient Boosted Machines (GBM) and Random Forest (RF).
All concepts will be illustrated using a hospital readmission dataset.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Gary Wang, Satadru Sengupta
Combining Models and Ensembles
One size so rarely fits all. And anyone who has based a rating plan on “a GLM” knows that there are really multiple GLMs (and other techniques) working in concert, first to normalize the data, then to get those new territory groupings, then to solve for the other rating factors, and so on. Sometimes one model is built on the residuals of another. At other times a model’s output is used as a predictor in another model. And then there are Ensembles which in a different way involves combining models. This session will discuss various ways of combining models and situations in which they might be useful, and will also introduce the basic ideas and approaches of Ensembles.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Christopher Cooksey, Eliade Micu, Shan Lin
Overview and Practical Application of Non-GLM and GLM Methods in Insurance
GLMs are a powerful tool with very practical benefits in pricing and other insurance applications. Other statistical methods can aid in further improving GLM results or more broadly to bring valuable insights to complex problems. This session will review three commonly used methods (most likely decision trees, random forests, gradient boosting) and illustrate different ways they are being applied in insurance. This session will focus on the high-level mechanics of each method and the benefits/challenges of their application – as opposed to the underlying technical details.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Claudine Modlin, Duncan Anderson
Comparing Machine Learning and Conventional Statistical Techniques in Claims Models
Machine learning methods are becoming popular in insurance modeling applications. In the session, the panelists will discuss a few machine learning methods (gradient boosting, random forests, neural network, LASSO, Elastic Net, etc.) and compare them with conventional regression methods (GLM, General Sessionized mixed models, etc.) in the context of claims triage models. A case study will illustrate the pros and cons of those techniques.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Luyang Fu, Peter Wu, Josh Brady, Richard Shafer
Emerging Issues in Homeowners Insurance – A Panel Discussion
There are numerous emerging issues that are currently challenging companies writing homeowners insurance. Issues that will be discussed by the panel will include:
-Risk from home-sharing services like Airbnb and Homeaway.
-More Refined Rating Plans.
-The Smart Home and the internet of things.
-Flood coverage in voluntary market.
-An additional issue selected via a live poll by the attendees.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Robert Curry, Paul Anderson, Terri Dalenta, Robert Lee
Vehicle History: Insurance Scoring for Your Car
To evaluate the risk associated with personal auto coverage, insurers have traditionally primarily used information about the driver, such as age, gender, marital status and location; the intended use of the vehicle, such as business or pleasure; and information about the vehicle when new, such as original cost new. Insurers then added detail to driver information using sophisticated credit scoring and demographic models. They added detail to vehicle use information with telematics and usage-based insurance pricing. Now, insurers have begun to add detail about the vehicle itself using vehicle history.
This session will discuss how carriers use the specific vehicle history to more accurately price and underwrite risks. We will discuss obvious data elements, such as salvage title brands, verified average annual mileage and severe damage. We will also discuss data that may not be as obvious, but has proven predictive of loss in the past. Finally, we will discuss using vehicle history as an entry point for UBI programs.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Donald Hendriks, Patrick Foy
Next Generation Enhancements to Credit-Based Insurance Scores
Early adoption of credit based insurance scores created large improvements in segmentation ability, but recent gains have been incremental. New credit data, including trended, time series credit measures, and alternative credit data, present the opportunity for substantial gains in credit based insurance score segmentation.
Traditional credit report information is static making it difficult to know where a consumer has come from, or where they are going. Trended measures such as balance and payment history help predict future behavior and support improved decisions. The use of traditional credit report information has resulted in over or under estimation of the risk of consumers that have changed or will change scores.
As many as 50 million consumers have very limited, or no credit history. These thin file and no-hit consumers represent a large and growing consumer base. The lack of traditional credit information for these consumers has led to limitations in ability to objectively evaluate the consumer, and General Sessionized treatment. Alternative credit information such as banking account, utility, and alternative financing account information can supplement traditional credit information, and help to improve the segmentation of consumers not found in traditional credit sources.
Discover how these new sources of consumer credit information can shape the future of credit based insurance scores.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Donald Hendriks, Stothard Deal
Is Mileage the Most Powerful Variable in Personal Auto?
Mileage is a powerful variable in rating personal auto, but is seldom used because customer-reported mileage tends to be inaccurate. Recent changes in the industry have given access to accurate, verifiable mileage data. This session will discuss how verified mileage data can be used in a rating plan and whether it is as predictive as other traditional rating variables.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Douglas Wing
Panelists:
Ryan Morrison, Donald Hendriks, Dan Hill, Fred Blumer
Citizens Insurance of Florida – Rating and Underwriting for Hurricane Exposure
Citizens Property Insurance Corporation is Florida’s state-owned residual property insurer. After 8 hurricanes struck Florida during the 2004 and 2005 storm seasons, Citizens’ entered a period of rapid growth during which the number of policies it insured increased from under 600,000 at the end of 2002 to over 1.4 million at the end of 2011, when it insured 24% of personal lines policies in Florida. Citizens has since seen many of its policies find coverage from private insurers, and it now insures less than 600,000 policies. Citizens’ status as Florida’s residual market creates many special considerations for its policies. Examples include a statutory requirement to consider both actuarial soundness and affordability; and Citizens’ requirement to pay claims and how this could result in assessments. In this session, we discuss these considerations, with special attention paid to Citizen’s +50% drop in policy count.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Xiangfei Zeng
Panelists:
Brian Donovan, Paul Kutter
Territorial Risk Classification Using Spatially Dependent Frequency-Severity Models
Ratemaking in property-casualty insurance often requires territory-based risk classification. This paper proposes a spatially dependent frequency-severity modeling framework to produce territorial risk score. The framework applies to aggregated insurance claims where the frequency and severity component model the occurrence rate and average size of insurance claims in each geographic unit, respectively. We employ the bivariate conditional autoregressive (CAR) models to accommodate the spatial dependency in the frequency and severity components, as well as the cross-sectional association between the two components. Using a town-level claims data of automobile insurance in Massachusetts, we demonstrate the quantification of territorial risk score and the classification of rating territories with various clustering techniques based on the score distribution.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Joe Griffin
Panelists:
Peng Shi, Dion Oryzak
Modeling High-Dimensional Interactions in Personal Auto   
In recent years, significant improvements have been made in personal auto pricing model through the use of GLM modeling technique. However, it is still difficult to achieve credible results in high- dimensional interactions in a typical GLM modeling setting. Meanwhile, modern automated programs can capture local patterns, but it is always difficult to explain the modeling results. In this session, we will discuss the difficulties and approaches to modeling high-dimensional interactions.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Joe Griffin
Panelists:
Xiangfei Zeng, Yue Fan, Takeshi Yamaguchi
What’s it Really Worth? Two Unconventional Approaches to Measuring Model Value
Predictive modeling teams usually understand the models’ theoretical measures – predictive power, lift, credibility – much better than the practical measures which determine the model’s projected ROI in application. Through two real-world case studies, participants will learn some new ways to estimate the value of a model in application, and the importance of using the correct approach in communicating model value to the sponsor.
- In a Homeowners’ condition hazard application, what impact will the model have on discovery rate, action rate, loss ratio, and expense ratio?
- In a WC pricing model, how can we evaluate the potential impact on profitability without relying too heavily on many arbitrary assumptions?
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Panelists:
Bret Shroyer
Point Estimation Models vs. Ranking Models
Point estimation and ranking modeling are two types of predictive modeling approaches used in many industries. In the P&C insurance, class plan rating, territory rating, and vehicle symbol rating are typical areas for point estimation, while underwriting automation, claim fraud detection, credit scoring and Telematics scoring are typical areas for use of ranking models.
The two types of modeling approaches are fundamentally different in many ways. In this presentation, we will discuss the major differences of point estimation and ranking modeling in the following areas:
- Model design – Business considerations and technical considerations
- Predictive variable creation and predictive variable structure
- Model development and model validation
- Business implementation and IT implementation
We will use class plan development and underwriting modeling as an example to demonstrate the differences in details.
In the presentation, we will also list and describe many interesting applications of the two types of modeling approaches in life insurance, government affairs, and other industries.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Panelists:
Gary Wang, Jun Yan
Application of Text Mining In Claims Analytics: A Case Study
This presentation will discuss applications of text mining in predictive modeling. Text mining can be a valuable tool to add insights to data based on free text entry and unstructured information. When text data is considered, analyses like marketing and consumer sentiment, claims analytics, and even pricing can be enhanced.
We will focus on a case study in claims triaging for workers compensation and personal auto lines. Traditional claims triaging requires experienced adjusters to predict which claims require the most attention to minimize losses. To facilitate this process, senior adjusters can manually review claim notes and descriptions. By using modern text mining techniques on claim notes, however, insurers can better triage and see the big picture of their claims.
Text mining is a set of statistical and machine learning tools in the field of natural language processing. These methods examine a collection of documents and systematically identify important words and patterns. Topic modeling, a newer method in text mining, can also be used to predict which claim types are represented in a given claim. By understanding claim types, this advanced analysis can predict, e.g., whether an apparent “fender-bender” is more like a typical single-payment whiplash claim or more like a multi-payment claim with possible nerve damage.
The presentation will focus on applications with intuitive and visual results, with technical info reviewed at a high-level.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Joe Griffin
Panelists:
Will Frierson, Peter Lowth
The Biggest Problem with Your Pricing Model is Your Reserving Model
Using historical data to develop a rate plan requires assumptions about loss development. Often these assumptions are broad and do not distinguish between very different loss development tendencies across rating variables. Biased pricing indications are the natural result. If the company is conservatively reserved, with case savings common, slower developing classes of business are likely to be overpriced. If case-incurred losses tend to develop upward, the slower developing classes are likely to be underpriced. By changing the reserving model paradigm from triangle squaring to one that reflects differences across all of the rating variables, these issues are revealed. The differences in indicated rating factors are often dramatic.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Len Llaguno
Panelists:
Christopher Gross, Dion Oryzak
How Predictive Analytics Can Change Your Market Footprint (Live Streaming)
There are many statistical techniques that are available to identify patterns in existing data. Your existing data reflects the book of business that you have historically written – it represents a cross section of customers that identify with your brand and price. But what if you wanted to expand your brand and grow your book in market niches that don't currently resemble your existing distribution? The purpose of this session is to discuss analytical approaches to extrapolating beyond your existing distribution thereby altering your market footprint. We will discuss statistical techniques on internal data and an analytical approach in using external data to assist in improving extrapolation. Finally we will conclude with a General Session discussion of how changing your market footprint is more than just an analytical problem.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Fan Yang
Panelists:
Serhat Guven, Evan Petzoldt
Hypothesis Testing and Statistical Models: Avoiding the Journal of Irreproducible Results
When your favorite statistical package says that a parameter estimate has a p-value of 0.045, is that evidence that you should include that effect in your model? This session will focus on some of the key challenges of the hypothesis testing framework, especially
that, in many situations, “standard” procedures reject true null hypotheses more often than advertised and that p-values are very unintuitive to interpret. Multiple testing will also be addressed.
The source of these challenges will be explored, and ways of dealing with them will be suggested. The attendee should expect to come away with an understanding of what statistics is about process and not just math, and how to avoid some of the mistakes that are made by accepting computer output uncritically.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Fan Yang
Panelists:
Christopher Monsour
Predictive Analytics in Strategy and Decisions – A Tour Around the Globe
Predictive analytics can shape key aspects of insurers’ business strategies and thus is gaining popularity world-wide. Regional diversity in practice, market conditions and regulation is influencing the way analytics is being implemented around the globe. The availability of data, analytical sophistication and the maturity level of customer-centric views can also influence many business problems. These problems can be addressed by predictive analytics, and the efficiency and precision of actuarial decisions. We can learn from the global differences as well as from the communalities.
This session will discuss various analytical challenges around pricing, underwriting, retention management, customer life time value, etc and provide real life examples acquired through numerous projects in over 20 countries around the globe.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Fan Yang
Panelists:
Reuven Shnaps, Adi Bar-lev, Nicholas Chee Lek Yeo
Price Optimization Outside of the U.S.: Techniques and Implementation Strategies in Different Markets (Live Streaming)
This session will focus on how price optimization implementation works in the U.K., with contents covering what price optimization is and its key aspects such as inputs, algorithm and implementation. The panel will discuss the business benefits and wider implications of price optimization.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Fan Yang
Panelists:
Jeremy Benson, Francesco Daboni
Discussion of China's P&C Insurance Industry and Its Recent Market Development
In this session, the panelists will conduct an open forum, round table format of discussion on the China P&C insurance industry and its recent interesting market development. The topics for discussion will include but not limited to: the current Auto regulation updates, recent development of market distribution channels, adoption of UBI, direct market distribution channel, demand of international actuarial professionals, and culture adjustments in doing business in China. The panelists will share their experience on these interesting topics. On the same time, the audience and attendees are encouraged to participate the discussion for their experience and questions.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Concurrent Session
Moderators:
Fan Yang
Panelists:
Peter Wu, Jun Yan, Yuxiang Lei, Xin Li
Disney Analytics: The Mouse, The Magic and The Math
The Walt Disney Company is known for many things some of which include: theme parks, memorable characters, storytelling, and world-class guest experiences. What likely won’t come to mind is analytics. Over the last decade, analytics has become ingrained in the Disney culture as a way to increase operational efficiency, drive value across all of our lines of businesses, and bring more magic to resort guests. Within Walt Disney Parks and Resorts, various teams of decision scientists, analytics consultants, and industrial engineers have successfully tackled problems ranging from dynamic pricing for the Disney Theatrical Group, revenue management of Disney Park restaurants, ad optimization for ESPN, and maximizing marketing spend of Disney Studios. Join me as we explore how Disney is leveraging analytical techniques to solve its most challenging business problems.
Source:
2016 Ratemaking and Product Management Seminar
Type:
Featured Speaker
Moderators:
Fan Yang
Panelists:
Michael Akeroyd
Price Optimization (Live Streaming)
What is Price Optimization, and why is it so controversial? This session will try to answer those questions, and more. The session will begin with an introduction of thought-provoking questions about price optimization from an actuarial/professionalism perspective, followed by short presentations, both in support and against the use of price optimization. The remainder of time will be allocated for questions to our panel of industry experts.
Source:
2016 Ratemaking and Product Management Seminar
Type:
General Session
Moderators:
Fan Yang
Panelists:
Howard Eagelfeld, James Lynch, Brent Petzoldt
