Modeling analysis and prediction method based on marginal zero truncation Poisson model

Abstract

A modeling analysis and prediction method based on a marginal zero truncation Poisson model comprises the steps of modeling a total mean value of zero truncation Poisson distribution instead of a standard Poisson parameter, constructing a random representation-based EM-FS algorithm for parameter estimation, and developing the modeling analysis and prediction method based on the marginal zero truncation Poisson model. And establishing a screening criterion for determining the optimal weight combination of the candidate model and providing a candidate model optimization strategy. The marginal zero truncation Poisson model is mainly used for carrying out modeling analysis and prediction on observation indexes with zero truncation counting characteristics in the financial insurance field, and the marginal zero truncation Poisson model can provide direct utility evaluation, instead of indirect evaluation, of potential influence factors on a target overall mean value, so that the explanation ability of the model is improved; the model average prediction method of the marginal zero truncation Poisson model can effectively reduce prediction loss and improve prediction performance, thereby providing theoretical basis and data support for making related decisions and schemes.

Description

technical field [0001] The invention relates to the field of count data analysis in financial insurance, in particular to a modeling and predictive analysis method for count data with zero truncation feature. Background technique [0002] Zero-truncated count data generally refers to positive integer data with observations such as 1, 2, 3, 4, etc. greater than 0. Such data are widely used in the fields of financial insurance, traffic safety, and medical health. For example, whether it is property insurance or personal insurance, the number of insurance policies purchased by the policyholder in a coverage year is at least 1, which has a typical zero-truncated count feature. For insurance companies, whether it is property insurance or personal insurance, how to dig out the potential factors that affect the number of policies purchased by policyholders and make accurate predictions is one of the key issues to effectively increase the number of policyholders' insurance targets a...

AI Technical Summary

Problems solved by technology

However, the current research on the zero-truncated Poisson model is mainly based on the standard Poisson parameters to establish a logarithmic regression model rather than the overall mean in the zero-truncated case, so that the impact of the interested factors on the overall mean cannot be determined. Measured directly, which can lead to large errors and even misleading results

In addition, there is a lack of systematic research on how to improve the prediction performance of the model under the zero-truncated Poisson model. The usual practice is to simply show the prediction effect based on the selected model

However, model selection itself is unstable, and there is no evidence that the model with the best fit for the sample data must also have the smallest prediction error

Therefore, there is a certain risk in the prediction of the model selected by the model selection method.

Images