Evaluating the role of mixed frequency real-time weather data in economic forecasts

This thesis examines the role of weather forecasts in economic decision making and proposes a state-space model for evaluating multi-horizon forecasts. Our first application sets the scene by using mixed frequency real-time weather data to improve the predictive performance of retail sales forecasts. Our findings illustrate that significant improvements in predictive accuracy can be achieved when weather variables are incorporated into a mixed frequency forecasting model. However, once we move to a real-time environment, substituting meteorological forecasts for observed weather variables, meteorological forecast uncertainty leads to a somewhat muted improvement in predictive accuracy. The implication is that real-time data constraints are a limitation of forecasting models that use weather variables as an input. This points to a role for forecast evaluation and the need to better understand the structure of forecasting error. Our state-space model of multi-horizon forecasts is based on the forecast revision structure implied by forecast rationality. The parameters of this model allow us to identify rational forecast revision components, that we call information content, as well as residual forecasting error components, that we call implicit forecasting error. A key contribution of this thesis is the demonstration of how our model based approach to forecast evaluation nests existing test based approaches as special cases. We illustrate our proposed multi-horizon forecast evaluation approach using two empirical applications. First, we employ our model based approach to examine the structure of forecasting error in multi-horizon meteorological forecasts of daily temperature. Multi-horizon temperature forecasts are found to contain multiple sources of forecasting error. The structure of meteorological forecasting error depends on the length of the forecast horizon. Second, we employ our model to examine the information content of multi-horizon electricity demand forecasts for the New South Wales region of the Australian National Electricity Market. We discover that the information content of electricity demand forecasts depends not only on the length of the horizon at which the forecasts are made, but also the time of day the forecasts are made.