From https://www.pgcon.org/events/pgcon_20...
One of the new features in PostgreSQL 10 is the ability to create multi-column statistics, helping the optimizer understand dependencies between columns. PostgreSQL 12 further improved the feature by supporting more complicated statistics types.
This talk will explain a bunch of important questions - why we need this capability at all for some queries, how it works, which cases it can address currently, and which improvements are in the queue for PostgreSQL 13.