Trends and cycles in primary commodities: state space modelling and the Kalman filter
Ashesi University College, published by Mot Juste
Decomposing economic time series into their temporary and permanent components have followed two broad paths: trend versus difference stationary models and detrending versus filtering. Whereas the former breaks down due to their inability to capture the underlying data generation process (dgp), the latter are either one sided filters or are based on ad hoc procedures in achieving parsimony. In this paper, we propose structural time series models in which trends, cycles, seasonal components are treated as stochastic, and which contains the traditional approach as a special case. Cast in state space form, and estimated using maximum likelihood via the Kalman filter, these models accurately predict the behaviour of commodity prices through time. Using data on agricultural raw materials and metal price indices for the 1957(1) to 2008(4) period we document the frequency and duration of commodity prices, key elements for designing policies aimed at smoothing terms of trade shocks and the resulting macroeconomic effects associated with price disruptions. We found that the individual dgp have varied over time and are best captured as stochastic rather than deterministic trends. Moreover, we uncover multiple structural breaks and outliers, far beyond what extant results would like us to believe. Finally, the models remain robust in an out of sample forecast.
commodity prices, term of trade shocks, trends, cycles
Alagidede, P . (2012) “Trends and cycles in primary commodities: state space modelling and the Kalman filter” Ashesi Economics Lectures Series Journal. 1, (1):1–6