While there is a rich tradition of research in event studies, the automatic detection of events remains a huge challenge which is still mainly tackled using pattern-based approaches. Furthermore, as it has been generally acknowledged that decision-making on the financial market is often influenced by emotion and other psychological factors, these events often implicitly carry a positive or negative connotation, which cannot be detected by the currently predominant explicitly lexicon-driven sentiment analysis methodologies. In this talk, I will focus on our efforts to build a corpus of over 6,200 event schemata and 12,400 sentiment tuples to enable the data-driven extraction of economic events and implicit sentiment. I will also report on our experiments with different transformer architectures tackling both tasks separately and jointly from both a coarse-grained and fine-grained perspective. Finally, I will make an argument for the organization of a (series of) shared task(s) on financial NLP, for which our proposed corpus could be a stepping stone.