When I did my best attempt to fit the Hibbs model, which predicts incumbent-party vote share using change in per-capita income growth, I got a forecast of 52 percent of the two-party vote for the Democratic nominee.
Other forecasts give slightly different answers, as there are many ways of constructing a prediction model given past elections, but the general consensus is that the fundamentals predict the election is likely to be close. The economy is going okay but not great; the president is somewhat popular; there is a small minus for the Democrats in that the incumbent is not running for reelection, but a small minus for the Republicans in that they control both houses of Congress and, hence, represent a less appealing choice for centrist voters who prefer divided government.
So, to start with, no explanation is needed for why the election might be close. Yes, Donald Trump is an unusual candidate, but research by political scientist Steven Rosenstone and others suggests that when it comes to the general election, Americans vote party more than candidate.
To put it another way, the electoral landslides in postwar presidential elections were in 1964, 1972, and 1984 — all years in which the economy was booming. This year, the economy is doing okay but not booming, hence the more equivocal forecast.