Pre-print, Working paper: While the recent empirical literature on macroeconomic crises focused on a limited subset of events (e.g. banking, currency and sovereign), macroeconomic crises are usually characterized by large scale domino effects that involve a much wider and heterogeneous array of sectors and transform them into highly complex events. This data limitation, in turn, hampers the understanding of these chaotic and painful episodes for researchers and policymakers alike. After building a raw corpus of roughly 23,000 International Monetary Fund country reports, we harness the power of text mining to produce a new database on crises discussion: the database covers 20 different types of economic, financial and non economic events for a sample of 181 countries over the period 1950-2019. We document a substantial rise in complexity of macroeconomic crises throughout the X X and X X I t h century and a higher centrality of the non-fundamental channel in the system.