Paul Cornell, God bless him, was introducing me to passers-by at Worldcon as the guy who predicts the Hugo winners. I retaliated by accusing him of writing fiction, of course. I don't actually claim to do any more than analyse the statistics already available to us, and those statistics are few and misleading.
The two tools I have been using are the Goodreads/LibraryThing stats of nominated books and the prominence of the contenders on blog posts. The former measure had a bad year this year. Ancillary Justice was third of five on the Goodreads list for the Kitschies' Golden Tentacle (second on LibraryThing, but a long way behind Mr Penumbra's 24-Hour BookstoreNebula shortlist by both systems; and was far behind all of the Wheel of Time volumes for the Hugos (third on Goodreads, second on LibraryThing). Having said that, Ancillary Justice was top on Goodreads, and second on LibraryThing, for both the BSFA and Clarke shortlists and won both (the former jointly with the novel which placed third on Goodreads and fourth on LibraryThing).
My conclusion is that the Goodreads/LibraryThing stats are a decent guide to what has already sold well, but only the vaguest of indications as to what will actually win. (NB however that the Kitschies' Red Tentacle winner, A Tale for the Time Being, was also the clear leader on Goodreads/LibraryThing stats.)
The bloggers were a much better guide this year. Ancillary Justice was far ahead in my blogging tally for Best Novel, and "The Water That Falls on You from Nowhere" similarly for Best Short Story; "Equoid" was joint top for Best Novella, and "The Lady Astronaut of Mars" in a strong second place for Best Novelette. So that's two and half hits, and a near miss, out of four; better than my blogging surveys for last year (two hits, but two distant misses) or 2011 (one clear hit, three clear misses).
The sample size was anyway for too small to try any such tally for the other categories, or the Retro Hugos, and I also wonder if, because so many people are writing up their commentary these days in places that I can't see (or rather that Google can't see), my sample may be skewed by search algorithms. When it comes down to it, a survey of blogs about the Hugo shortlists will reveal only the preferences of those surveyed, which may (or may not) be a good guide to the preferences of the voters overall.