Exact Discrete Representations of Linear Continuous Time Models with Mixed Frequency Data

The time aggregation of vector linear processes containing (i) mixed stock-flow data and (ii) aggregated at mixed frequencies, is explored, focusing on a method to translate the parameters of the underlying continuous time model into those of an equivalent model of the observed data. Based on manipulations of a general state-space form, the results may be used to model multiple frequencies or aggregation schemes. Estimation of the continuous time parameters via the ARMA representation of the observable data vector is discussed and demonstrated in an application to model stock price and dividend data. Simulation evidence suggests that these estimators have superior properties to the traditional approach of concentrating the data to a single low frequency.