An auction with many anonymous bidders can bring big surprises. All the better, of course, if one could predict the outcome of an auction or at least estimate it more accurately. Our colleague Stefan Rameseder, working together with Jun Prof. Dominik Liebl from the University of Bonn, made a big step in this direction. In their article "Partially Observed Functional Data: The Case of Systematically Missing Parts", published in the journal "Computational Statistics and Data Analysis", they abstract human strategies into a statistical model, which the cooperation partner Entelios AG uses in everyday energy trading.
The method proposed by the two authors is motivated by the practical problem of Entelios AG of unpublished prices and the consequent difficulty in forecasting demand curves. This problem arises, for example, on the so-called control reserve market. This electricity market is to acquire power and energy in order to compensate for fluctuations in the frequency of the electricity grid, since the amount of electricity fed in there must always be the same as the amount withdrawn. The auction design of the standard power market trades around 500 million euros a year, but the price curves can only be partially observed, and the mechanisms used so far are often wrong, as they are based on ordinary system trading strategies. They counteract this problem with a novel estimation method.
The project is a cooperation with Entelios AG, a supplier of demand response solutions in the energy market. With its help, electricity customers, but also producers can market their flexible electricity loads.
Our colleagues Stefan and Dominik Liebl have succeeded in estimating the demand curves better than before, thus optimizing the pricing mechanism.
This project shows again that science and practice are not incompatible. Entelios' data has enabled the development of a state-of-the-art energy trading model that is both methodologically up-to-date and of high economic relevance.