The research explores strategies for short-term forecasting of photovoltaic (PV) power in energy communities, focusing on the aggregator's perspective. It investigates the economic value of various data types in forecasting accuracy, using both physical and machine learning methods. The study employs models like PVWatts for physical forecasting and advanced machine learning models for data-driven approaches. A case study of a Dutch energy community with rooftop PV systems is analyzed, highlighting the economic benefits of detailed data for forecasting. The research concludes with implications for aggregators' data collection and modeling strategies, considering regional privacy laws and the effectiveness of data-driven approaches. Future work aims to develop machine learning algorithms that utilize live meter data for enhanced forecasting accuracy.