Title: Add the beef and combine: Dynamic density combinations from point forecasts
Authors: Oliver Grothe - Karlsruhe Institute of Technology (Germany)
Laura Hersing - Karlsruhe Institute of Technology (Germany) [presenting]
Abstract: A real-time forecasting problem is faced, where one-period-ahead point forecasts for a certain variable of interest are available at each point in time. Such forecasting problems are relevant in many areas of the economy, e.g. in energy and financial markets. Typically, different commercial providers offer point forecasts for the corresponding variables of interest. We aim at constructing density forecasts out of these point forecasts and at combining these forecasts dynamically to a joint forecasting density. Transforming the point forecasts to density forecasts allows for considering more advanced risk metrics than in the point forecasting case. Forecasting the density allows the user of the forecast to dynamically calibrate its risk over time, particularly in a high-frequency setup. Our approach consists of two main steps. First, we transform the available point forecasts to univariate, conditionally optimal density forecasts (adding the beef to the bone). Second, we infer the time dynamic copula between the forecast densities and use it to combine the density forecasts (combination part). We show that our method outperforms other techniques that are using or combining the different point forecasts with respect to a variety of loss measures.