For many U.S insurers, particularly smaller and midsize carriers, the standard approach to rate changes is to compare various rate scenarios and choose the one that best aligns with business objectives and regulatory considerations. These rate scenarios are generally rebalanced to an overall target.
However, an important facet that is missing in those cases is the impact of customer behavior. Machine learning can help insurers understand customer behavior. But before insurers begin using the technology, we believe they need to consider two main issues: off-balancing books of business and long-term goals.
01
The first comes down to how carriers go about off balancing their books of business. For example, if you want a 5% overall rate increase, but you want to vary rates more for chosen segments based on experience of their relative riskiness, you really want to know what proportion of customers each of those segments represents and what the effect of those targeted increases will be on business acquisition, renewals and cancellations.
Using a personal auto business as an example, if you apply to raise rates by 15% for drivers under 21 years of age, how many existing customers are going to start shopping around and potentially cause premium leakage? The acquisition and renewal probabilities differ for various customer segments, so it pays to be able to model those specific impacts.
02
Keep your long-term goals in mind. Let’s continue with the example of a 5% overall rate increase. While such an increase may help achieve immediate profit targets, how might it affect demand and the business longer-term? Suppose you’re considering four different rate scenarios for an upcoming rate filing. By considering the impact of customer demand, you can track the true profit impact of each scenario over multiple periods as the book begins to reshape. Without this insight, you may be left in the dark when forecasting future performance.
Of course, all of this is possible using generalized linear modeling (GLM) techniques. But it does need a skilled modeler to specify the overlapping demographic traits that tend to cause variations in buying behavior. This is where machine learning techniques such as gradient boosting machines (GBMs) are extremely useful.
GBMs work by dividing data sets into segments to produce homogenous groupings known as decision “trees.” As the GBM continues to scan and learn from the trees it creates, it identifies further discrepancies in the data within them and forms new trees that eventually make up a multi-layered “forest” of instances where risks interact within the data. The potential cost and workload savings are immense.
And while some companies may have balked at what is seen as the chore or difficulty of collecting data on new business quote acceptance, renewals and policy cancellations, a clearer picture of how customers – both existing and new – are likely to respond to rate changes should definitely make the effort worthwhile.
To learn more, we are presenting on this topic at the Casualty Actuarial Society Ratemaking, Product and Modeling Seminar on March 16 at 2:15 p.m. ET.
