Scholarlink

<p>The adoption of telematics data into auto-insurance pricing enables dynamic risk assessment by capturing individual driving behavior. This study introduces a telematics-driven dynamic pricing system that integrates real-time driving behavior data with a learning effect framework to address limitations in traditional auto-insurance models. Utilizing a dataset of 9,934,265 trips from China, we conduct a logit-reduced mixture-of-experts (LRMoE) model to investigate the combined effects of driving behavior and the learning effect on accident frequency. Our findings indicate that the LRMoE model outperforms Poisson and Zero-Inflated Poisson (ZIP) benchmarks in terms of goodness-of-fit, prediction accuracy, and risk segmentation. Compared to the conventional no-claims discount (NCD) system, our framework substantially reduces premium misalignment through daily updates based on near-miss events and driving duration. These results highlight the potential of the proposed LRMoE-driven pricing system to enhance actuarial fairness and precision, enabling insurers to align premiums with real-time risk exposure while fostering transparency and policyholder retention in automobile insurance markets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Beyond no-claims discount: a learning-embedded telematics-driven pricing system for dynamic premium adjustments

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月9日

查看全文
  • Heng Xiong,
  • Yuxuan Guo,
  • Ričardas Zitikis

摘要

The adoption of telematics data into auto-insurance pricing enables dynamic risk assessment by capturing individual driving behavior. This study introduces a telematics-driven dynamic pricing system that integrates real-time driving behavior data with a learning effect framework to address limitations in traditional auto-insurance models. Utilizing a dataset of 9,934,265 trips from China, we conduct a logit-reduced mixture-of-experts (LRMoE) model to investigate the combined effects of driving behavior and the learning effect on accident frequency. Our findings indicate that the LRMoE model outperforms Poisson and Zero-Inflated Poisson (ZIP) benchmarks in terms of goodness-of-fit, prediction accuracy, and risk segmentation. Compared to the conventional no-claims discount (NCD) system, our framework substantially reduces premium misalignment through daily updates based on near-miss events and driving duration. These results highlight the potential of the proposed LRMoE-driven pricing system to enhance actuarial fairness and precision, enabling insurers to align premiums with real-time risk exposure while fostering transparency and policyholder retention in automobile insurance markets.