The results of Tuesday’s election stunned most people – including internal analysts within the Trump camp –  because the polling seemed to give Clinton an insurmountable lead.  She was ahead of Trump in many states, and though there was great room for uncertainty in each poll, everyone assumed that even if there were some polling errors, there were not enough to make a difference in outcome.  I.e., she could lose Ohio and Florida and still win so long as she held Pennsylvania and so forth, so it seemed as though even accounting for polling error, there was little chance she could lose.

That assumption, however, ignored the possibility that all of the errors were correlated – so that an error in one state’s polls meant that the same error would be replicated across multiple states.  That’s something that Nate Silver accounted for in his model, however, and others rejected, and it contributed to Silver’s more bearish Clinton predictions.

And of course, that’s what happened with respect to mortgage backed securities as well.  Everyone knew that some mortgages would fail – and that some RMBS tranches would fail – but the assumption was that there were enough of them that any errors would not damage an entire pool.  What many (though, as with Clinton, not all) people failed to foresee was that the errors were correlated, so that they stood or fell in unison.  What seems so obvious in retrospect was difficult to understand at the time.