Do you have any citation for that? I use BitcoinWisdom exclusively because it's the only forecast I've seen that doesn't swing wildly (like baddw said), and as it gets closer to the change, it's rarely off by more than 0.1%. For this last change, near the last few blocks it forecast that difficulty would go down 0.88%, but it only dropped 0.73%, so they were only off by 0.15%, not too terribly bad if you ask me. All forecasts are "wrong" in the sense that it's the future, so it's unknowable, but BitcoinWisdom seems to do a really decent job at providing a reliable number that more closely matches how difficulty actually changes, unlike many other mining forecast sites.