Louise Francis, past CAS VP-Research & Development and Alice Underwood, new CAS VP-Research & Development, co-authored this blog post.
Last August, David Menning wrote a blog post about how much statistics should be included within the basic education process. He asked if statistics were covered at the right level in the current basic education structure and whether there should be more coverage or less.
There was a lot of reaction to the post. Several respondents expressed that actuaries need to know more about statistics and be better prepared in the statistics related to predictive modeling. One commenter noted that actuaries are no longer the “go-to group” to build class plans and that this could occur in other areas of practice.
The issues surrounding the technical skill-set of actuaries are often viewed as an educational issue, whether basic or continuing education, but we believe this issue also affects research. CAS Call Paper Programs have featured papers about predictive modeling and data mining and the CAS Grants Task Force has funded some predictive modeling research projects — but is this enough? Are CAS actuaries positioned to take the lead in developing and applying advanced statistical approaches?
What can the CAS research division do to help actuaries to acquire and refine the skill set for advanced applications like predictive modeling? Please “Leave a Reply” below.
Identifying the characteristics of successful research actuaries (skill set and background) and the charactertistics of actuarial departments with successful research areas would be useful. It’s hard to get to where you want to go without identifying the destination. That type of information is something that the Research Committee may be able help provide.
I believe there is a good deal we can do both in our basic education and continuing education functions within the CAS to improve our members ability to compete in the market place for modeling positions in the insurance world. Getting feedback from reserarch actuaries on the issue of what can be done through our education programs or designing an on line program (or several programs) would be useful.
Some skills can be developed through requiring candidates to go through the CAS Syllabus. Webinars, research papers, one line courses and hands on seminars are useful in picking up new ideas. There are limits though on what we can do through our education programs, and getting feedback from the Research group on what skills are best picked up in an academic background (hiring Statisticians with advanced degrees) would be good. Then too, there is the issue of how much of this is a matter of getting on the job experience and first hand observations from the Research area would be valuable on this point.
There is a simple cheap solution: Develop a new exam that covers “developing and applying advanced statistical approaches”. This exam might be deployed in CBT format to allow the use of spreadsheets and/or other software on sample data.
The resources of the CAS research division could be used to compile syllabus material and an initial bank of prototype questions (eventually to be distributed to candidates as samples for study).
For older members not up for taking another exam, workshops covering the sample questions could be offered at CAS meetings.
Here are some thoughts on what the Research Group could do:
Create online learning:
Professor Andrew Ng is currently offering his Machine Learning class online for free. This course is very well done. The Research Group could develop similar online courses following his example. Or, even easier, they could develope a broader curriculum with courses like Professor Ng’s as subcomponents.
http://www.ml-class.org.
Hosting “classroom” modeling competitions:
Following the online learning approach, Kaggle-in-class is an online forum for an organization (like a university class, or a CAS training) to host a predictive modeling competition, to reinforce ideas taught in class. Research could develop such competitions. The winners might be invited to speak at a CAS event, or get some CAS event fee waived.
http://www.inclass.kaggle.com
Develop a credential the industry would recognize in the modeling space:
Following Menning’s post, of the several job postings listed, none asked for CAS credentials as a requirement; only one posting listed credentials as a possibility. One more exam won’t change the tide. It would take a concerted series of courses, application, and marketing for industry to start to recognize CAS in this space. Part of that marketing effort might be a separate CAS credential, like CAS Predictive Modeler. This credential might be earned through completion of the above curriculum, online learning courses, and successful Kaggle-in-class performance.
I think that if you put the question “Should all CAS members be required to develop the skill set to be a successful researcher?” to the general membership, the answer might be “yes” if you ask the question in the abstract, but if you specifically list some of requirements of the skill set, the answer “no” would appear pretty quickly.
That being said, I do think the CAS needs to: (1) Expose its members to some relevant statistical techniques that have been developed recently. The Bayesian MCMC modeling that I mentioned in round one of this discussion is one such example. (2) While not all CAS members need to be researchers, we need to encourage a critical mass of our members to be researchers. This might include a special designation similar to the CERA.
With respect to (1) we should update Exams 3 and 4 to include a survey of the “relevant statistical techniques” mentioned above. I am willing to participate in the effort to do this.
The CAS has always been doing the right kind of activities to encourage research. These include focused working parties, research committees, calls for papers, and competitions such as the COTOR Challenges. I might add that these activities serve mainly to introduce the problems. The real benefit comes afterwords when the researchers have had time to think. I like Mike Larsen’s suggestion of identifying the characteristics of successful researchers. That should lead to a better design of research activities for the CAS to promote.
The CAS needs to go the way of the SOA and offer specialized exams for different fellowship tracks. By the time you get to exam 9, you should know whether you’re a workers comp actuary or not. IMO, the time I spent learning about workers comp was not as valuable as an exam on advanced statistics would have been. Also, does everyone need such specialized investment knowledge? If there were 3 different exams for statistical, investment, and workers comp tracks, the CAS could produce more actuaries with specialized statistical knowledge.
I would suggest that we be cautious in reducing the amount of material on investments on the syllabus. Asset risk is a surprisingly large component of risk for P&C companies and the ability to speak and understand the language of finance in terms of risk and reward when evaluating product performance is valuable.
We should distinguish the base level skill set in terms of predictive modeling that all P & C actuaries should have vs. the skill set that is required to actually do the hands on modeling work. I believe it’s possible to change our basic education material, the CAS syllabus, to help insure that P&C actuaries have the vocabularly to talk with the statisticians. In many cases though, one cannot work examples by hand and , at least on a personal level, understanding the modeling concepts without working an example or two is difficult at best. We may need some form of an on line course (or set of courses) with paint by the number examples, which all candidates work through (maybe we should require this as continuing education too) and a pass/fail type test to measure that they understood what was going on in the examples.