Towards a new correlation

"recommendations affected by the changes on the 2000-2002 tapes are too optimistic, while those on the 2003-2005 tapes are too pessimistic"

Just reading a paper about sell-side analyst recommendations and how they are being affected ex-post, after the fact. So on a tip from Thompson three researchers from NYU, Harvard and UVA searched through Thompson's database. The three found that if you downloaded the data set in 2000 it would be different from 2001 and both different from 2002. Based on their research Thompson is making a move to create an un-blanched data set that will not have revisions.

Now not all the revisions were changes from hold to buy, or sell to buy which the authors deem alterations, they also found deletions, additions and anonymizations. Now in hindsight it would seem to only be a historical triviality only interesting to researchers, but any investor following a strategy of buying and selling based on changes in opinion from analysts will look at the historical record to extrapolate how much she might gain from an analyst upgrading a stock. Unfortunately, the data, as the authors put it "the historical bedrock of empirical research" would seem to have no foundation.

More darkly, for star analysts it marks a method of enhancing their record or prescience. Just imagine a golfer changing his score after the round. The justification, well I would have made par except for that nasty sand trap. In the analysts case par is the correct recommendation and the sand trap is the bear market of 2000-2002 or even worse the bull market of 2003-2007. In this new history there outstanding pay is more correlated to their "enhanced" prognostications.