مقاله سال ۲۰۱۶

Food prices have shown strong correlations in the past, even before their upward co-movement over recent decades. For instance, the commodities included in the World Bank’s food price index for the period 1990–۲۰۱۳ (on a monthly basis) show price correlations of well over 0.60, and even up to 0.85 for some subsets. We therefore explore whether or not the forecast accuracies of a subset of these commodity prices could be improved by taking their cross-dependence into account. Some might think that this is an old question that has already been answered, as simultaneous modelling did not survive after the 1973 oil crisis, due to the poor forecasting performances of macro models relative to naïve forecasts. However, we now have a better understanding of the effects of breaks on forecasting. Various different devices and methods, such as robust transformations and updating, may be useful for forecasting in the presence of breaks, and can also be applied for joint and other models that consider crossdependence.

We focus on three food prices which are strongly correlated: corn, soybeans and wheat. These are agricultural commodities that, whether directly or indirectly, feed a large part of the world’s population. There has been a special interest in understanding the common behaviours of their prices since the 2000s, when their downward trend reversed, as the demand for them started to increase significantly, driven by the unprecedented growth of emerging economies such as China and India. The demand for oilseeds has also increased greatly, due to their competing use as biofuels. Because of these effects, and also due to various macro and financial developments, their prices have experienced a long-term boom, along with many other food, mineral and energy commodities.

We are interested mainly in developing conditional forecasts of food prices in which the out-of-sample values of the weak exogenous variables will come from outside the model; that is, will be provided by the forecaster (e.g., from organizations such as the World Bank, FAO, IMF or USDA). These values should respond to conjectural scenarios about the future behaviours of the regressors, in order to quantify what would happen to corn, soybean and wheat prices if, for example, the economy of China decelerated at a given rate, or the US monetary policy changed. Thus, using the conditional forecasting models should also make it possible to project what might happen to foodprices given a range of assumptions regarding the paths of the explanatory variables employed in the model. To allow for the effects of these variables, a necessary condition is to evaluate the forecasting accuracy of the econometric models over a given pseudo out-of-sample period and forecast horizon.