Any global predictor that claims to have accurately determined Trump's victory should keep in mind that he lost the popular vote. Only a state-based predictor can be plausibly said to have been able to make such a prediction.
I won't mention the signal because I want to keep my job. However I'd be surprised if there weren't at least a dozen companies that have access to some honest signal with such a good sample of the population at large as to be better at predicting the election than any pollster.
Big ISPs have access to these "honest" signals on a large scale.
I have a great many hunches and some facility with visualization of directed graphs and other analytics, but am not clear on how to scrape and normalize the datasets I need.
But there might be "hidden" location information in datasets that dont appear to have it, such as Twitter.
Word choice carries regional and social information.
So it's conceivable Twitter data could correctly predict the election with ML techniques that essentially are estimating the electoral college without being trained on an explicit model of it.