Model Master

Q&A with Michael Ewald, predictive model builder at The Hartford

Photography: Jane Shauck

Q: Tell us about your background. How did you make the decision to become an actuary?

A: The probability of me becoming an actuary was very low. Having completed the bulk of my math requirements in high school, I graduated from The College of Charleston with a degree in economics and a minor in finance without taking any substantial math. After four years in the South, I returned home to be closer to my girlfriend, Katie, who is now my wife. I began working at a large, multinational Italian construction company and was responsible for analyzing the profitability of one of the company’s U.S. subdivisions. Like many jobs at most companies during the Great Recession, the work was nothing short of exciting. As construction spending dwindled in 2009, I was tasked with developing a restructuring strategy. Although the math seemed relatively straightforward, I quickly learned that you couldn’t blindly follow the numbers. I learned about the need to balance expenses, talent retention, employee morale, stress and a slew of other items you can’t learn from a textbook.

During this time I passed my first two CFA examinations and began to think about a new career path. Katie, following in her father’s footsteps, was taking actuarial exams. Because I already was taking exams in my spare time, she told me to consider a career as an actuary. My knee-jerk response was “that sounds boring,” but she pushed me to at least consider the profession. I learned that an actuary’s career options are seemingly limitless. For someone who thrives on change and learning about new areas, I decided this was the career for me. During the first half of 2010, I finished my CFA exams and passed the first two actuarial exams. In July 2010, I started my first actuarial job in The Hartford’s Executive Actuarial Training program, working in the financial reporting department of the company’s Japan variable annuity block.

Q: How did your work history segue into your interest in predictive analytics?

A: In March 2012, The Hartford announced it would focus on property and casualty, group benefits and mutual funds, and stop new annuity sales in April 2012. Working in fixed-indexed annuity pricing, my work abruptly halted; however, I transferred within The Hartford to a new predictive analytics position in the Group Benefits division. With predictive analytics well established within The Hartford, the position provided the necessary resources to support robust training.

The Hartford was one of the first group benefits carriers to develop a predictive analytics program. Because we were developing first-generation predictive models, most projects were blank canvases. What I liked about my new role was the level of creativity needed. It was similar to that of my product development role. I love solving puzzles and building things. (I was obsessed with Legos as a kid.) Predictive modeling is the perfect intersection of solving very complicated puzzles and building something meaningful from the ground up.

The first project in which I used predictive analytics was building a long-term disability termination model. I quickly found a niche facilitating communication between the modelers and the actuaries. Having substantial experience manipulating and aggregating large amounts of data, I helped with data preparation. I was able to leverage the skills developed in other areas, while honing my predictive analytics abilities.

Q: How did your professional experience lead you to a career in predictive analytics?

A: I don’t have any preconceived notions around what my career is supposed to look like. Four years ago I wouldn’t have expected to be working in predictive analytics, and seven years ago I barely knew what it meant to be an actuary. Having an open mind and a willingness to try new things naturally will lead actuaries to career paths in different areas.

Q: What are some challenges in working in predictive analytics?

A: One of my friends likes to say, “We can do a lot of cool stuff, but that doesn’t mean we should.” I think of this statement in two ways:

  1. Does our work pass The New York Times test?
  2. Does the marginal benefit of a more sophisticated model outweigh the marginal costs?

The first question asks how your work would be perceived if it were to be published in The New York Times. This will be asked more often as new data sources arise. Companies risk reputational damage if they do not properly protect their policyholders’ private information. We also need to understand whether the use of predictive analytics has legal ramifications, and how regulators will view these new methods. Finally, I think we have an ethical obligation to ask if we should do something, even if we know we can.

The second question tries to get at the optimal amount of time to spend on predictive models. I could spend months building a model using the most advanced statistical techniques available. However, I may have been able to develop an answer that is easier to explain and just as effective in a fraction of the time. The point is I have seen projects fail because the timeline couldn’t be met or the output of the models couldn’t be explained. To paraphrase Albert Einstein, if you can’t explain your model to a 6-year-old, you don’t understand it yourself.

Q: How did you learn the tools and techniques of modeling?

A: As a property and casualty company, The Hartford has a tremendous amount of predictive modeling expertise. I developed the foundation of my knowledge by working directly with the P&C actuaries and data scientists. I spent an enormous amount of time reading (and re-reading) articles, papers and textbooks. I asked a lot of questions.

Learning the mechanics of building a model can happen relatively quickly. Correctly interpreting the model, understanding the implications of the outputs and identifying anomalies takes years. Modeling is so much more of an art than an exact science. My manager always says it takes about a year before modelers get their feet under them. It takes even longer to develop an expertise.

Q: What are the main skills actuaries need for work in predictive analytics?

A: Actuaries have all of the skills necessary to perform predictive analytics—they just need to learn a different set of tools. I am a golfer, so I like to use a golf club analogy. For generations, most golfers utilized a 3-iron. In the last 10 years, technological advances produced a club called a hybrid. The hybrid replaced the low-numbered irons in most golfers’ bags because it is more versatile and easier to use. Golfers still played the same game; they just had a better club. Predictive analytics is a new club that the actuary needs to learn to swing. The irony in this analogy is that I still use a 3-iron, but I am thinking about making the switch!

It is important to recognize that predictive analytics is just a single club in an actuary’s golf bag. We have numerous other skills that are complemented by predictive analytics.

Ultimately, some actuaries will need to be as comfortable with advanced analytics as they are with Microsoft Excel. Other actuaries only will need to understand it at a high level. It all depends on your role and the business challenges you are trying to solve. Generally speaking, I believe actuaries will be at a competitive disadvantage if they cannot do the following:

  • Manipulate large amounts of data
  • Perform advanced statistical techniques
  • Simplistically communicate the not-so-simple

Q: So, we have the basic skills. How do we learn the tools?

A: For a tactical answer, learn a software. There are so many free resources online that you don’t need to spend money to learn predictive analytics. While not endorsing any particular resource, Google has made its introductory Python course available online for free. Kaggle has a great introductory course that explains how to use R, Python or Excel to predict the likelihood of dying on the Titanic. If you prefer more structure, Coursera has a number of advanced analytic classes you can take; however, not all are free.

I find that a simple Google search is the best place to start. Last weekend, I decided I wanted to analyze car prices at dealerships across New England. A few Google searches later, I built a program that scrapes the internet, aggregates car prices at selected dealerships and models where I would get the most favorable price. There is so much information at our fingertips that with a little drive and some time, an actuary can learn a lot about predictive analytics.

Q: Where do you see opportunities for actuaries in the predictive analytics arena?

A: In the short term, companies are building out predictive modeling teams, and they need actuaries to bridge the gap between the traditional business and cutting-edge analytics. In the long term, I don’t see there being a huge distinction between actuaries and those who perform predictive analytics. Again, I think predictive analytics is a club we all should have in our golf bags.

For the creative actuary, opportunities exist wherever there is a substantial amount of data. The usual suspects like lapses, mortality, cross-selling, up-selling, client retention, marketing, etc., are well-documented, so I am not going to belabor the point. Actuaries are the natural facilitators of data analysis because they understand the business. One of my employees recently built a model in which we were seeing lapses drop dramatically at a given duration. The phenomena passed validation, so it must be real, right? As hard as we tried, we couldn’t rationalize why this was occurring. We sliced and diced the data, and eventually we found a flaw that resulted in us throwing away a large chunk of the data. Without business context, it would have been easy to take the indications at face value and run with the model.

Actuaries and statisticians can learn a lot from one another. There is talk of competition for these jobs. Diversity of thoughts and ideas is imperative for a company’s success. Why wouldn’t we want to have statisticians and actuaries working together to solve a company’s most complex problems? Data is only going to get more complicated. Insurance companies are going to need very technical people to manipulate and model the increasing amount of data, and the field is only going to grow. Our predictive analytics department at The Hartford has doubled in size in the last three years. These are new jobs that are being filled by both actuaries and statisticians. Companies are going to need strong communicators to bridge the gap between the technical analytics and business leaders. Actuaries have the background to choose where in the process they want to be involved.

Q: What are some of your best professional experiences as an actuary?

A: Aside from sitting in my dining room for weeks at a time studying with my wife? As twisted as it may sound, I actually do miss those days.

We are very fortunate as actuaries to work with some of the most talented, dedicated, intelligent and genuinely good people in this world. I have fond memories and lifelong friendships that are a result of my day job. That being said, we all have to go to work, so I will focus on other rewarding aspects of this career.

I am very proud of the research that my colleague and I published with the Society of Actuaries (SOA). The paper, “Predictive Modeling—A Modeler’s Introspection,” details how to build generalized linear models by describing a long-term disability pricing project.

I also spoke at various industry conferences and joined the SOA’s Predictive Analytics Advisory Group. The advisory group is identifying opportunities to train actuaries on this critical skill. As mentioned, I believe this skill set is very important, and I am happy to be involved in educating the greater actuarial community.

Working with the SOA’s Candidate Connect, I have spoken with aspiring actuaries about my career path. Given that many of the Candidate Connect attendees are career changers, I hope my path can help provide guidance on how to leverage unique backgrounds to land a career in an actuarial student program.

I also am very involved with the Actuaries’ Club of Hartford/Springfield. The committee organizes semi-annual meetings for almost 400 local actuaries, hosts networking opportunities and provides support for our local universities. We also provide support to the local Actuarial Bootcamp, an organization that educates high school students on the actuarial field. This position has allowed me to meet industry leaders, such as the Connecticut Insurance Commissioner and the chief actuary of the U.S. Social Security Administration.

The best advice I can give is don’t just get involved, but get involved in areas in which you are passionate. If you are passionate about a topic, it becomes less like work and more like a hobby. To have a hobby that opens career doors is an easy way to expand your network and have some fun.

Michael Ewald, FSA, CERA, CFA, is a director at The Hartford, where he has been responsible for building predictive models for the Group Benefits division.