A High-Frequency Algorithm for Battery Storage Trading on Continuous Intraday Electricity Markets - Replication, Optimization, and Empirical Comparison of the Rolling-Intrinsic Policy Across Three European Intraday Markets

Abstract

Battery storage earns revenue on continuous intraday electricity markets only with a trading policy that reoptimizes as the order book moves. This thesis replicates the rolling-intrinsic dynamic-programming (DP) policy of Schaurecker et al. and adapts it to Flower’s operational setting. Schaurecker et al. study hourly products on the German EPEX market. We move the rolling intrinsic to quarter-hour contracts and backtest it on roughly a year of Nord Pool order-book data in each of three bidding zones: Sweden SE3, Sweden SE4, and Germany Amprion. The quarterhour switch makes the DP backward pass roughly 64 times heavier than the hourly reference, because the number of stages, the action set, and the storage grid each grow fourfold. A C++/OpenMP kernel keeps event-resolution backtests tractable, reaching about 2.6 ms per re-solve on the SE3 benchmark. On a single 10 MWh, one-hour battery, re-solving on every order-book event earns roughly EUR 0.75M on Sweden SE3 and EUR 0.93M on Sweden SE4 over twelve-month windows, and EUR 1.08M on Germany Amprion over a longer fourteen-month window, and beats every fixed re-solve timeline tested in all three zones. Over the window common to all three zones, SE4 earns the most, ahead of Germany Amprion and then SE3; Germany’s larger raw total reflects only its longer window. Set against the dominant ancillary-market alternative, the rolling intrinsic earns more on only 8% of days in Sweden SE3, 4% in Sweden SE4, and 10% in Germany Amprion. It is therefore a conditional trading policy for the battery, chosen day by day against ancillary opportunities, rather than a default replacement for ancillary-market participation.