Synthetic data for ratemaking: imputation-based methods vs adversarial networks and autoencoders
The study compared synthetic data generation methods for actuarial ratemaking, testing imputation-based MICE against complex deep generative models like CTGAN and VAEs. The findings show that Multivariate Imputation by Chained Equations using Random Forests is a competitive, user-friendly approach due to its simplicity and existing tools in R, making it preferable for actuaries over more complex methods.