A few misconceptions here: the efficient market hypothesis does *not* mean that the market will fairly or correctly value a stock, nor does it imply that sharp, discontinuous movements will not occur as investor sentiment changes. It is simply an equilibrium argument about the current traded value, ie:

  • If information existed that investors agreed should give the company a higher or lower valuation, it would naturally be exploited by timely buying or selling, which would drive the price toward a new equilibrium value.
  • Such trading will continue until all new information is figured into the stock price.
  • Hence, the market price reflects all (public) information about the stock

That's it. This is basically tautological given its assumptions: that investors will preferentially buy or sell based on positive or negative news and prospects, and that there is "perfect information" (that is, information is disclosed or evident to all investors more-or-less simultaneously) in the marketplace.

EMH is a pretty weak theory as theories go because it is almost completely equivalent to its assumptions. So you shouldn't try to use it for much. It is wrong, in particular, to use the EMH to conclude that a stock will always be fairly valued, or that its value will not change markedly over time even when no new information about the stock itself comes to light. This is because lots of other factors influence equity prices: the current levels of interest and exchange rates, liquidity preference, foreign trade and balance-of-payments, investor confidence/preference/exuberance, money flowing in and out of mutual funds, etc.

You could of course consider all of these as market factors that are part of the broader panorama of "information" the theory is talking about, but then you would be back where you started: at a theory that basically begs the question of information. If investors buy or sell the stock, for whatever reason, you can conclude that it is a reflection of new information in whatever form, and therefore the theory works. But you haven't learned anything useful.

As weak as it is, EMH does succeed rather well in demonstrating that technical analysis of asset prices is in principle not a very good idea; in fact this is part of why EMH was formulated, as Blackthorn correctly points out. That is, if there really were information about the future performance of an asset in the history of its price then investors would immediately take advantage of it, causing the price change and negating the effect of the information. Technical analysts rebut this argument by claiming essentially that only a Highly Trained Technical Analyst can perform such analysis, and therefore the information is not public to the investing community and EMH does not apply. You can tune in to CNBC during the trading day and watch them give their analysis and a defense of it along with about a million-or-so other investors; I leave it as an exercise to the reader to spot the irony in this.

Finally, a comment about alex.tan's point that people make (or lose) money trading in the market: of course they do. This is why people invest in the first place. If there were no tradeoff of risk vs. reward, then there would be no point in investing. In short-term trading (holding the asset minutes, hours or days) the volatility in the asset price that investors trade against is a measure of the lag, market timing and asynchronous dispersal of information in the marketplace. In the long term (months or years) the change reflects more fundamental trends in the business prospects of the company, which have a lot of uncertainty surrounding them. The fact that money can be made this way in no way disproves EMH: in the short-term case, it simply means that investors can take advantage of short-term swings in the price as a stock settles to a new information equilibrium, and in the long term case it say that investors are rewarded for investing their money. Nothing is surprising about this.