The common approach to calibrating stochastic age-period-cohort (APC) models is to decompose the data into various APC components, choose which components to model and finally to fit ARIMA processes to them. This unfortunately discards information on the uncertainty of the observations and obscures both identifiability issues and the common aspects of smoothing and fitting ARIMA processes. We present a new approach that calibrates models in a single step, fully incorporates information on observation uncertainty, treats smoothing and ARIMA processes consistently (calibrating both), calculates information criteria based on all information, and enables a coherent treatment of identifiability for the whole model. Speaker: Tim Gordon, Aon