Baruch thinks that sell-side analysts deliberately game the earnings-expectations system:
Analysts with buy ratings on a stock will have lower earnings numbers going into quarters than their peers, so they don't inadvertently raise the average of published sell side expectations, "the consensus number" as calculated by Bloomberg or Reuters.
It's definitely the case that an analyst with some kind of buy rating -- and most ratings are positive, remember -- has a vested in having the company in question "beat expectations" and thereby ratify his buy rating. And if he keeps his earnings estimate low, that will help to keep expectations low, and thereby maximize the chances that those expectations will be beaten. Cunning.
I'm not convinced this is a big problem. Partly because, as Baruch notes, the really important number to beat isn't the consensus earnings number as published by Reuters or Bloomberg, but rather the "whisper number." And partly because most analysts are sheep anyway, and their earnings estimates are always within a pretty narrow range around whatever the company has indicated to them that its earnings are going to be. In other words, earnings estimates aren't a good guide to what analysts think, but they are a good guide to what the company in question wants analysts to think.
Even so, if you wanted another reason to ignore sell-side earnings estimates, here you go. It's a perfectly good one.