Actuarial aspects of the application of artificial intelligence in non-life insurance pricing in the Republic of Srpska

This paper focuses on the assumptions that AI models must satisfy for the purpose of determining non-life insurance pricing in the Republic of Srpska. When applying Generalized Linear Models, which are most common in modern non-life insurance pricing practice, actuaries are aware of the risk that inadequate choice of variables and their interrelationships will result in poorer model performance. This limitation was one of the reasons why experts in this field began developing models that combine neural networks and Generalized Linear Models for insurance pricing. In this paper, we analyze two hybrid models: the Combined Actuarial Neural Network and LocalGLMnet. Summarizing the results presented in the selected literature, we conclude that these models can deliver better results compared to Generalized Linear Models; however, given the limited transparency of insurance pricing procedures, further research is necessary to develop models aligned with regulations and actuarial practice standards.