Spatial independent component analysis for heteroskedastic random fields – Rodrigo Morales Martínez (University of Helsinki)

Independent Component Analysis (ICA) is a powerful tool for uncovering latent signals, but classical methods often fail under spatial heteroskedasticity. This talk introduces spFOBI and spJADE, novel estimators designed to recover independent components in heteroskedastic random fields by exploiting the spatial structure of fourth-order cumulants. To build intuition, we will briefly delve into the novel spARCH and spGARCH models to explore the underlying dynamics of spatial volatility clustering. We then show through simulations how our spatial extensions successfully navigate these complex environments and outperform their classical, non-spatial counterparts. Finally, we conclude with a practical proof of concept, we will conclude with a brief proof of concept showing how these methods can be used to assess climate risk in real estate.