Principled imputation for missing innovation components prior to multivariate KDE simulation

Identify statistically principled methods for imputing missing components of multivariate regression-residual series used as innovations, in a manner that preserves marginal distributions and cross-sectional dependence for subsequent multivariate kernel density estimation–based simulation, providing an alternative to the paper’s ad hoc linear‑regression‑plus‑resampling approach.

Background

To simulate 7‑dimensional IID innovations via multivariate kernel density estimation, the authors need complete residual series but face missing values in several components. They currently fill missing entries by regressing the incomplete component on the complete ones and resampling residuals—a pragmatic but non-standard approach.

The authors note the absence of clear, established methods in the literature for this specific task and invite alternatives that maintain dependence structure and are compatible with KDE-based simulation.