What model would be best suited for cases where policyholders are likely not reporting claims due to fear of higher premiums?

The zero-inflated model is particularly well-suited for situations where there is an excess of zeros in the dataset, meaning many observations have a zero count, which aligns with the scenario of policyholders not reporting claims. In this context, the presence of unreported claims is essentially creating a situation with more zeros than would normally be expected based on the underlying distribution of claims.

This model effectively combines two different processes: one for the actual data-generating process of the claims, which may produce positive claims, and another model that determines whether a claim is reported at all. The zero-inflated model captures the idea that there are inherently two groups: those who will report claims and those who will not, potentially due to fear of higher premiums. Therefore, it can account for the excess zeros in the claims data as a result of deliberate non-reporting.

In contrast, the hurdle model treats the process of reporting and the count of claims differently but does not necessarily account for the additional complexity of having a latent (unseen) group that never reports claims. The logistic regression model, while useful for binary outcomes, may not fully capture the count nature of the claims data. The negative binomial model is typically employed for overdispersed count data