Spatial data are a rich source of information for actuarial applications: knowledge of a risk’s location could improve an insurance company’s ratemaking, reserving or risk management processes. Relying on historical geolocated loss data is problematic for areas where it is limited or unavailable.

This talk presents a method to construct spatial embeddings within a complex convolutional neural network representation model using external census data and use them as inputs to a simple predictive model. We will also discuss how one may adapt the representation learning approach with image data and use multiple sources of unstructured information within the same homeowner’s insurance ratemaking model.