The accuracy of Nate and other quantitative modelers' forecasts depends on the quality and accuracy of the polls they use. Drew Linzer of votamatic.org has written some really good pieces the past few days examining poll accuracy and the potential for "systemic polling failure":
Nate's model accounts for presumed inaccuracy in the polls but it is partly reliant in those polls' historical leanings. I think it would be more correct to say that his model depends on the quantity and consistency of polls.
http://votamatic.org/can-we-trust-the-polls/
http://votamatic.org/another-look-at-survey-bias/
http://votamatic.org/pollsters-may-be-herding/