Analysis of an Historic Market Dataset with Large and Sparse Data Missingness

The present work aims at applying the well-known LOOP theory to an historic agricultural market dataset. Collected data regard the local market of Friuli an historical north-eastern region of Italy. The prices of four commodities are observed with weekly frequency on a period of 115 years. The dataset present a large number of sparse missing bservations. In the multivariate time series context, solution for the missing data treatment have been mainly developed in the state space model context. Alternative solutions are also present in the classical likelihood framework. The present work applies the composite likelihood approach to the estimation of the particular Vector Autoregressive model connected with LOOP theory. In particular, the pairwise likelihood formulation is considered in order to recover the information included in the partially observed data vectors. The main advantage of this approach regards the improvement of estimators efficiency connected with observation recovery.