The doc for the default event loop policy states that """the default policy defines context as the current thread, and manages an event loop per thread that interacts with asyncio""".  What it doesn't mention, though, is that you get a RuntimeError if you call get_event_loop() in a non-main thread -- you first have to call set_event_loop(new_event_loop()) in that thread.  This wasn't expected by me.

As a side note, customizing the event loop policy isn't trivial either: it seems it's best to subclass DefaultEventLoopPolicy and make adjustments by overriding the API methods.  Writing your own wouldn't work, as e.g. the Unix default policy has custom initialization code for child watchers.