I don’t get why the societies feel the need to add all this data science stuff to the exams. Unless we pass new laws mandating actuarial review of these techniques ain’t nobody gonna wanna sit for them when they can just leetcode their way into Google.
I mean, just keep the exams the same. If companies are running out of candidates they’ll just pay more to get people to learn the boring insurance stuff. Wages will adjust so it’s a self solving problem or not really a problem at all.
you and me both, man.
“Let’s redo our exams to try and make our candidates experts in everything, and in the process make them experts in nothing and less than qualified in key actuarial competencies.”
Nothing could possibly go wrong with that, right?
As a recent FCAS, I feel ashamed to say this but I stopped caring about the syllabus…
Lots of FCAS do that. I’m FCAS, I had to deal with all that bullshit, still think it’s all bullshit, but it’s not my problem now that I’m done - the rest of you all can suck it up and deal with it like I did.
I think a basic understanding of data science is becoming more important to the actuarial profession, and some of the items historically tested are becoming less relevant. The body of knowledge is constantly evolving. The CAS/SOA are just bad at adapting.
For example, I’d prefer to see a candidate who understands GLMs versus one who got a perfect score on exam 3F.
I do think they’re trying to cast too large a net though. It’s difficult to tell if they’re just trying to expand the membership base by creating actuaries who are “good” at non-actuarial job functions, or if they’re just bad at creating a curriculum because older actuaries with little/no expertise in data science are the ones who’ve been tasked with setting the data science standards.
To me when I was an exam taker, it was very annoying for them to make a new test. That meant that prior studying wasn’t useful and the potential risk that your prior exam credit would disappear.
From a perspective of learning useful knowledge. I think basic predictive analytics and R Coding would have been very helpful for my job. I would consider the CSPA for that exact reason.
I think the way the CAS have handled is thus far has been solid. I’m definitely of the view that it’s a fool’s errand to try to outcompete data scientists in data science because data science will inevitably be stocked with PhDs in statistics and math. Actuaries should focus on where actuaries provide the most value, by being that mathy business person who can synthesize the information together well and communicate it effectively to the business side to improve the business’ profitability.
I think of it much like CAT modeling. We don’t need to become experts in meteorology but we do need to be familiar with the inputs and general workings of a CAT model. Same deal with non-CAT GLMs. And actuaries should be able to take that super technical output and provide actionable insights to the business and so the exams should focus on equipping actuaries to be successful at that.
I agree with this. I think what the CAS is trying to do isn’t this. I think the CAS wants us to outcompete the PhDs and head them off, as if they’ll somehow turn and then invade our space. Having seen some data scientists at work, many are fantastic at building a model and their next step is “so, use it because it works” and skip getting buy-in from business units that will be affected. That’s how they end up with variables like heavy trucks / school ratio, say well the model says it’s highly predictive so I’m using it and then don’t understand why someone would object because but … it’s highly predictive!
Ask them about inconsistencies in their data and how that might impact their model results and how they accounted for it if at all, and you can practically cause the BSOD in some of them.
This is exactly the problem. The CAS is obsessed on making sure candidates know all of these advanced predictive analytics and data science/modeling work instead of actually testing what’s important. We will never be as good at model building as them and it baffles why the exams have this focus. There are other areas we add value and thats what the focus should be on.
Yet somehow the CAS thinks the answer is make the exams even more impossible with even lower passrates? Not to even mention the fact that passing the exams the way they are does not make you a data scientist…
I’m admittedly disappointed by this. You were someone who I felt always did a good job articulating the issues with the exams candidates were facing. I get why you don’t care anymore, but I do hope more people with an actual voice can advocate for us. These exams continue to get even more difficult and the bar continues to move even higher than ever was required before.
It’s not that I don’t want to care. It’s hard to keep up with all the CAS changes and I have not been proactive following it. I do hope to get more involved but I currently don’t have the capacity to volunteer. That may be something I do in the future.
I’ve gotten by fine without GLMs up until now and recently created a new working model with innovative variables. Although I have spent many hours dabbling with GLMs, traditional pricing is the main focus of my job (it is just me and my boss across the country, no semblance of a department) so it can only be pursued as a hobby when time permits.
I perused the CSPA but was concerned that there would be too much head knowledge and not enough examples, though I could certainly be wrong. What I need is a book with exercises that will enable me to complete the process from start to finish, and nothing I have found appears a very good job at this. Most resources read like a dull book for a doctorate student which isn’t great IMO. I did attend a GLM using R seminar many years ago and it was pretty good, but time passed and it was impossible for me to duplicate the exercise. I think a proper testing of GLM therory and experience would require a case study (i.e. module) rather than an in-person exam.
I don’t mind seeing the theory, but if it isn’t properly integrated with multiple real-world examples, it is lost on me. Is there anything that is a good resource? I doubt that with five or so working years left there would be a huge benefit to picking it up now, but I’m certainly open to it. And I have used the GLM function and gotten results, but not convincing or necessarily satisfactory ones.
So far I’ve liked R for Actuaries by Brian Fannin (recommended by our very own Meep!) but haven’t finished working through all the examples yet. The great thing about R is that it’s free to load.
GLMs is old school statistics. I wouldnt really classify it as a mainstream “data science” tool. For predictive analysis, the mainstream tools in data science tend to be non-parametric which is why people like using them so much (you dont have to think too much).
From an ideological point of view, domain knowledge should enable you to build parametric or semi-parametric models with as few variables as possible. If youre relying too much on non-parametric methods with lots of possible covariates then you can be replaced by a data scientist. Remember that machine learning is really just fancy “regression” with lots of variables.
I tend to agree with colonel: letting data science take over the actuarial curriculum would be a big mistake since it signals that traditional actuarial methodologies arent that good and actuarial expertise not that valuable. Actuarial societies should be careful here to make sure that actuarial candidates properly differentiate themselves from regular data scientists.
Strange comment. The bread and butter of an actuary is doing that kind of work. If you think that’s not important then the profession is doomed.
the curse of overfitting is real