An actuary relies on data and assumptions to price products, calculate reserves and capital, to form strategies and more.
An actuary with lots of data, is indeed a happy actuary. Now what do we do with the data?
We analyse it, of course. Similar to the adjacent field of data science, actuaries analyse data to uncover explanatory variables, trends and patterns. This area of practice is called experience analysis/studies.
How do we go about it?
Let's assume we start one from scratch and are not just updating one that is already in place.
The Big Picture - envision the end result
It's good to start with the end in mind, or at least a hazy picture of it. What do we intend to do with the analysis - is it to:
Build a basis for pricing and valuation
Do a deep-dive to get insights on the drivers of losses
Find mitigating strategies to stem losses
Uncover explanatory variables for a new type of risk
Find the optimal way to structure reinsurance
Etc etc
Depending on the goal we have, the approach we take will differ. There's no one size fits all, and tailoring our approach will result in a more efficient and insightful study.
Understand the data, the core system and the insurance policies underlying the experience
We need to spend time understanding the data that we have at hand.
Some questions we need to ask are:
How reliable is the data, and which data fields are useable?
For what time period can we rely on the data?
What are the various policy statuses coded, and what do they mean?
What adjustments need to be made to the data before using it?
Are there multiple sources of data? If yes how do we combine them?
We also need to understand the core system which processes transactions and generates the data:
How long has this system been in use, and what was the system used prior to this?
Are all policy functionalities built into the system or are there manual workarounds?
Were there errors in the system configurations and fixes made after that?
And we also need to understand the various insurance policies underlying the experience.
What types of products are they?
What type of underwriting do they require?
What distribution channels are used?
What client segments do they target
We may need to segment the analysis based on some of these criteria above.
Determine an appropriate study period
Depending on data availability and the answers to some of the questions above, we will get to know the feasible time period for which the study can be done.
The goal of the experience study also guides us in choosing this period, which is usually between 1 to 5 years for longer-term risks and can be as short as 4 weeks for short-term risks with lots of data.
Determine an appropriate credibility threshold
By credibility, here I'm referring to statistical credibility. We typically use results from classical/limited fluctuation credibility to decide the acceptable number of claims/exposure for the study.
We use this to determine useable grouping sizes/cohorts to study too.
Perform the experience study
Finally we can get to actually doing the study.
Actually, once we have done the groundwork above, this part can be surprisingly quick.
For small quantities of data we can use Excel and pivot tables. For larger quantities of data we may be better off using SQL, R or Python, depending on our choice of poison.
Piece together the story
This is the fun part and also arguably the most crucial part of our work. What does this all mean? We need to understand the analysis in depth, and provide recommendations to our stakeholders.
A wide range of insights can come out of this, including:
Problematic client groups
Drivers of poor experience
Impacts of product designs
Changes in pricing strategy
Impact of underwriting approaches
Claims management insights
We need to be story-tellers to effectively communicate to various technical and non-technical audiences.
We can do the perfect analysis, but it won't be of value to the business if we can't communicate it effectively.
Many times we may not need to do much beyond simple tables and explaining them with stories.
More complex insights may be better explained using charts. We can also build interactive charts using for example, the Shiny package.
Experience analysis is..
a complex but really fun aspect of actuarial work. This is where we can bring in many aspects of our actuarial techniques and derive value for our stakeholders.
We can (and should) also incorporate techniques from data science as well in this.
Comments