@jeppe_and Hi, there are 4 options for backtesting:
processing all data in one pass. Very fast but you risk to overfit (by taking a global average for example); processing all data in many passes using the built-in backtester; same, but preserving the state at each iteration; for example, you can use the weights of yesterday to fix the weights today. In this case submission will take longer as parallelization is not possible; if yo use machine learning methods and you need to retrain your model periodically, you can use the ml extension for the call to the backtester.J