With regard to predicting earnings, analysts have actually been consistently, but slightly, conservative. That is, they make earnings predictions that end up being slightly lower than the eventual actual earnings. Sort of surprising, isn't it? It is especially surprising given their known penchant for over- optimism . These "conservative" earnings predictions are a well-known phenomena, and it is not difficult to explain. There are two factors involved. First, companies like to meet or beat the earnings expectations. Members of management will be viewed as being good at their jobs, and the company will be viewed as being as good as, or better than, expected. Second, for analysts to do a good job predicting earnings, they need information. If analysts can get full access to the firms that they follow, such as having personal meetings with the CEO or with other top executives, it really helps their task. However, will a CEO be 100 percent cooperative with an analyst who will not make an estimate that is either makeable or beatable? Probably not. In fact, Bill Gates and sales chief Steve Ballmer of Microsoft once purposely criticized their own firm to analysts in order to depress their expectations. Later, upon being told by one analyst that they had succeeded in painting a grim picture, Gates and Ballmer gave each other a high-five!  What is the general outcome of these two factors? The answer is that analysts make slightly conservative estimates because this is what the management of the firm they are analyzing wants.  This makes the CEO happy and willing to grant further and future access. Analysts end up being "off" on their estimates by only a very tiny margin, so they are still considered good analysts. The company will either make or beat the estimate, so it will be considered a good company. Everyone wins. "Under-promise, over-deliver" is the name of this game.
So, analysts may be able to predict rather accurately, but they have far more superior information than the rest of us. With the private information that they have, they can probably make more precise earnings estimates than anyone else. However, despite their ability to be very accurate, they will publicly and knowingly come out with slightly conservative estimates to keep the firms that they follow happy. While analysts are going to end up being systematically wrong by a small amount, their behavior is rational.
However, the ability of analysts to predict earnings accurately may suffer in the future. Since October 2000, the SEC has prevented firms from divulging privileged information to any analyst. Any information that the firm wishes to convey to an analyst must simultaneously be conveyed to the public. The SEC thought it was unfair that some investors, through analysts, were getting private information that other investors were not getting. The SEC policy creates a level playing field for all investors. Without privileged access to information, it is likely that analysts' forecasting accuracy will go downhill. However, John Coffee, securities law professor at Columbia University, suggests that forecasts will now become more honest assessments of future earnings.  It's too early to tell what kind of effect this SEC regulation is going to have on analysts' forecasts, but one academic study finds that analysts' forecasts since the passing of the SEC regulation have become less accurate. 
What about analysts' ability to recommend stocks? It is unclear whether analysts are any good at picking stocks. Older academic studies from the 1970s contended that analysts did not have good stock-picking abilities .  However, the more recent studies suggest that analysts may have some marginal ability as stock pickers.  If you were to have bought the stocks recommended as a "strong buy" during 1985 to 1996 and held them until the rating was downgraded, you would have outperformed the market by 4.3 percent per year if transactions costs were not considered. Analysts did indeed pick good stocks! However, if transaction costs were considered, you would have underperformed the market by 3.6 percent. While the picks were good, they were not good enough to implement a successful trading strategy.
Probably the most well-known evidence of analyst stock-picking abilities is from the Wall Street Journal 's "Dartboard Competition," which is no longer printed. Throughout the 1990s, the Wall Street Journal pitted the stock picks of buy-side analysts against stocks picked by a random throw at a stock dartboard.  Out of 146 contests, the pros beat the dartboard 62 percent of the time. When pitted against the Dow Jones Industrial Average, the pros beat it 55 percent of the time. This evidence indicates that analysts can pick stocks, but, at the same time, it also shows that they're not that great at it.
But what about Blodget and Grubman, whom we mentioned earlier, and the famous Mary Meeker? Meeker is the star Internet analyst at Morgan Stanley who was once dubbed Queen of the Net by Barron's . Before and during the 85 to 97 percent price decline of Priceline, Amazon, Yahoo!, and FreeMarkets, Meeker never downgraded them.  Why were all three analysts so terribly off on many of their recommendations? Why were they being so well compensated at the same time? To understand this, you should know how analysts' compensation has been structured.