Polls are unreliable a year before an election.

[Drunken Irishman]

6. Head-to-head polls against an incumbent this far out are actually pretty reliable...

Or have been in the last few presidential elections. There are some extreme exceptions to this, like with Ronald Reagan in 1983, as the US was coming out of a pretty significant recession, and even 1991, when Bush was riding high based on the success of the Gulf War - but since? It's been pretty stable and this is because the dynamics of a race generally do not change between now and election time. Beyond a recession (which will devastate Trump) or a war (which could boost his popularity), the numbers start to set, even a year out.

Case in point:

In Sept., 1995, Clinton led Dole by two nationally. It was not a significant lead, but by the end of 1995, in December, his lead would balloon to double-digits. Even the last polls were basically a year out from the election and they proved pretty reliable.

In a Zogby International Poll conducted between Aug. 16-19, 2003, Bush beat Kerry 50-41.

Between July 4 and August 30, 2011, Obama led Romney in ten national polls, while Romney led Obama in just two - and they were tied in three.

These last three elections prove one thing: the incumbent was leading at this point in the campaign. Some by a wide margin (Obama and Bush) and another by a narrower margin (Clinton), though his numbers would improve greatly by the end of the year.

FWIW, Reagan led Mondale by ten at the end of 1983.

So, really, the only campaign where there was significant shift a year out was 1992. Bush absolutely started that campaign with a comfortable lead (he led by double-digits through May of 1992) and then faded. But there's not a lot of examples of incumbents being down like Trump right now who've come back to win. Every incumbent who won reelection was leading at this point sans H.W. Bush.