By modeling reserves with microlevel models, individual claims information is better preserved and can be more easily handled in the fitting process. Some of the claim information is available immediately at the report date and remains known until the closure of the claim. However, other helpful information changes as claims develop; for example, the previously observed number of payments. In this article, we seek to model payment counts discretely based on past information, in terms of both claim characteristics and previous payment counts. We use a dynamic score that weighs the risk of the claim based on previous claim behavior and that gets updated at the end of each discrete interval. In the model used in this article, we will also distinguish between the different types of payments. We evaluate our model by fitting it into a dataset from a major Canadian insurance company. We will also discuss estimation procedures, make predictions, and compare the results with other models.

通过使用微观模型对储量进行建模，可以更好地保存个人索赔信息，并且可以在拟合过程中更轻松地处理这些信息。一些索赔信息在报告日期立即可用，并且在索赔结束之前一直为人所知。然而，其他有用的信息会随着索赔的发展而变化；例如，之前观察到的付款次数。在本文中，我们寻求根据过去的信息，根据索赔特征和之前的付款计数，对付款计数进行离散建模。我们使用动态评分，根据之前的索赔行为来衡量索赔的风险，并在每个离散间隔结束时更新。在本文使用的模型中，我们还将区分不同类型的支付。我们通过将模型拟合到加拿大一家主要保险公司的数据集中来评估我们的模型。我们还将讨论估计程序、做出预测并将结果与​​其他模型进行比较。