Forecasting Global Rice Consumption

This paper examines the time-series properties of global rice consumption data commonly used in studies of consumption trends and evaluates alternative time-series models for long-range forecasts. Using time-series data from 1961-2011, we find that per capita rice consumption and GDP per capita are non-stationary in levels (i.e., they are unit root processes) and are not cointegrated. Thus previous studies that have applied econometric models for stationary data suffer from the well-known spurious regression problem. Out-of-sample performance evaluation of appropriate, univariate time-series forecasting methods suggest that double exponential smoothing may be the preferred approach to forecast global rice consumption. Our forecast results suggest that global rice consumption is projected to increase from 450 million tons in 2011 to about 490 million tons in 2020, and to about 650 million tons by 2050. Moreover, forecast intervals for these point estimates tend be very wide (especially in 2050), reflecting the inherent uncertainty in making long-run forecasts using any approach. These wide forecast intervals suggest a great deal of caution is appropriate when interpreting such forecasts for formulating public policy.