Super Exogeneity and Forecasting Energy Demand with High and Low Frequency Data

An econometric model to be useful for policy analysis should pass the super exogeneity test. In the paper we study the linkage between super exogeneity failure and bias in forecasts and suggest that super exogeneity is an important property also for a forecasting model. Building up on a previous work of Engle, Granger and Hallman (1989), we show that the forecasting performance of a seasonal model that fails the super exogeneity test may be improved by allowing error correction relatively to the long-run estimated from an annual super exogenous model. Then, merging low and high frequency information is found to be important for forecasting not only when the seasonal process is integrated, but also when other forms of instability characterise the data