Microsoft Research New York City successfully predicted 21 out of the 24 Oscar winners which is a great improvement over last years predictions as David Rothschild an economist working for Microsoft Research then predicted 19 out of the 24 successfully.
The categories they predicted wrong were Best Animated Short Film which was won by Mr. Hublot Best Documentary Feature which was won by 20 Feet from Stardom and Best Live Action Short Film which was won by Helium.
All though they had 3 wrong, David wasn’t surprised by this error as he lets know in a comment to Predictwise
The top nominee in each category averaged 86.5% likelihood of victory. Thus, multiplied out over 24 categories we expected to pick the winner 20.76 categories. Winning 21 out of 24 means that our predictions were perfectly calibrated.
They had already predicted that of the 24 predicted 20.76 (21) was going to be accurate which was true. In order to predict 24 out of 24 accurately the likelihood should have been 100%. Here is the image of the predictions it made several months back updated to show which ones were accurate and which ones werent. (click image to enlarge)