I have had some experience in Pricing now, and I see a lot or new jobs talking about building Pricing Models. What does that entail?
I have done what I thought was traditional ratemaking work: producing indications, filing, exam 5 stuff, etc. But I haven’t built anything from scratch. How do you start building a pricing model from scratch and what skills do you normally need? And what is a Pricing Model?
So is a model like the rating plan? You think about rating variables and their factors like home age x roof condition x … any other variable that impacts your home rate?
How do you determine the rating factors for each variable?
I’m having a hard time conceptualizing how this all takes place in practice. Would you have a ton of data and do analysis to determine which factors are predictive and then include those in your pricing model?
And then to determine factors, regression analysis?
Are you taking exams? If so, what exam are you on? A lot of this is covered on the exams. Are you looking to break into the field and are just curious? Are you a student?
So I have taken 5 (and since forgotten the GLM section) and had some time as an EL analyst in Pricing. But I just can’t tie what I did in Pricing to “modeling”. When I think of models its usually predictive models which I didn’t touch. My pricing experience was working with indications, filings, etc. We have a rating sequence of factors to determine a policy premium, and now that is what I’m starting to think of as a pricing model - but it was always existing. I see new job postings about building Pricing Models and I’m so curious on how that happens since I never built one.
If I apply for one of these roles having pricing experience, I don’t want to be behind the ball if I can’t meet expectations for someone having been in pricing.
Indications and filing are NOT related to pricing models, unless you’re filing for a pricing model to get approved.
Pricing models involve…you guessed it, models. It doesn’t need to be complicated, but it does need to be a model.
If red cars have a loss ratio of 50%, and blue cars have a loss ratio of 70%, your model might indicate that blue cars need to have a rate relativity of 70/50 over red cars. This is when you are only looking at one variable - car color.
When you have more than 2 variables, it becomes near impossible to do this manually, so you leverage the advances of machine learning algorithms (which includes regression, but also much more).
I see. That is where my Pricing experience had been lacking. I hadn’t done any modeling in that role and it seems like all new roles do involve some form of modeling.
Thanks for the clarification! I will likely need to find some reading materials to better understand this and fill in my Pricing gaps.
Second, third, … whatever the count is, of people above. Syllabus material on ratemaking exams are definitely “basic education” level material, they only scratch the surface of what actuaries get asked to do.
“Pricing model” can mean all kinds of things. Can mean “this is how we’re going to price a policy to an insured” as happens in personal lines. Can mean “here’s how our policy would price out, relative to competitors.” Can mean “if we change factors, here’s the impact.” Can mean “if we add/delete items in a rating algorithm, here’s the impact.” Can be the basis for an indication. Can be the basis for making decisions on whether to market to certain risks, and what kinds of actions are needed to bring them to being profitable. Can mean a number of other things, depending on what other internal customers want to know and/or use.
“Model” is basically any “mathematical algorithm” that uses “known” input values to calculate some “unknown” outcome.
Examples:
Any rating plan
Credit scores–outcome is usually based on likelihood to default on credit of a certain amount
Insurance scores–like credit score, except the outcome is based on the likelihood of having a claim over a certain amount
Note that how the “insides” of the model is determined can be very wide ranging. I think you can boil most of them down into two categories:
Deterministic: use of all available (historical) data to “calibrate” the “insides” of the model
Predictive: splits the available data into a “training set” and a “test set”; if time is a key element of the model (like insurance pricing is), the test set should be “out-of-time” data.
“Traditional” ratemaking is basically deterministic modeling. CAS Exam 8 (Advanced Ratemaking) has material (Monograph on GLMs) to better understand predictive modeling in an insurance context.
In all likelihood, it is an Excel spreadsheet that takes input data and spits out totals, averages, ranges, estimates, and some other final number that is denoted in dollars and maybe cents.
Real answer: When actuaries and salespeople love each other very much, they share a special hug, and a Pricing Model is born.
Disclaimer: I know nothing about P&C “pricing models,” but I hear they are breathtaking.