Moment-based density estimation of confidential micro-data: a computational statistics approach

Abstract Providing access to synthetic micro-data in place of confidential data protect the privacy participants is common practice. For be useful for analysis, it necessary that density function closely approximate data. Hence, accurately estimating based on sample important. Existing kernel-based, copula-based, and machine learning methods joint estimation may not viable. Applying multivariate moments’ problem sample-based has long been considered impractical due computational complexity intractability optimal parameter selection estimate when true unknown. This paper introduces a generalised form moment-based estimate, which can used functions only information empirical moments available. We demonstrate parametrisation solely by employing strategy selection. compare performance existing non-parametric parametric methods. The results show using provide reasonable, robust approximation comparable an example generation from resulting provides reasonable disclosure-protected alternative public release.