Advancements in Actuarial AI Affecting Insurance Pricing Models
The rapid advancement of actuarial artificial intelligence is challenging many insurance carriers as their pricing systems struggle to keep pace, potentially resulting in mispriced risk. Peggy Brinkman, a principal actuary at Milliman, outlines the progression of modeling techniques from univariate methods through generalized linear models (GLMs), to more advanced models like gradient-boosted machines, and onto explainable boosting machines (EBMs). These EBMs offer the accuracy of gradient boosting with the necessary interpretability for regulatory rate filing approval, as seen across the industry at varied rates.
"New modeling techniques can extract more value from the same data in terms of risk understanding," Brinkman stated, highlighting the importance of evolving methodologies alongside new data sources. The evolution of these methodologies is fueled by increased computing power, which has expanded dramatically since 2010, according to a UK government report. Computing capabilities for training machine learning models have grown exponentially, far surpassing Moore’s Law, with projections showing continued expansive growth.
"Computing power and tools' ability to take advantage of it remain primary constraints on data processing," Brinkman explained, adding that cloud and distributed computing environments have significantly improved processing capabilities, allowing insurers to update pricing models more efficiently. While generalized linear models have long been essential in actuarial pricing due to their transparency and ease of regulatory compliance, their limited ability to capture intricate risk interactions has driven a shift to gradient-boosted machines.
These advanced models offer enhanced risk segmentation, particularly helpful in complex scenarios like wildfire risk, though their lack of explainability posed challenges for regulatory approval. Explainable boosting machines now offer a solution, enabling carriers to achieve high modeling accuracy while maintaining transparency required by regulators, marking a significant development in actuarial modeling within a regulated framework.
Recent advancements in data sources also play a crucial role in pricing models. Over the past five years, satellite and aerial imagery have been incorporated to assess property characteristics on a large scale, while telematics and credit data continue to enhance auto insurance pricing models. The Internet of Things (IoT) represents the next wave of property risk data, though its current limited scale does not yet support statistically credible pricing, Brinkman notes—pointing to a trajectory similar to the historical adoption of satellite imagery.
The investment in advanced pricing models yields returns when new models are accepted by regulators, a process that can be resource-intensive. Judson Boomhower from UC San Diego identifies regulatory approval as a significant barrier to adopting sophisticated pricing systems. However, recent regulatory reforms in California, which facilitate the use of advanced catastrophe models and reinsurance cost inclusion in rate filings, may signal a changing environment. Carriers operating in California should stay informed about current guidance on these changes to fully assess their impact on market participation.