The Efficient Market Hypothesis (EMH) was famously defined by Fama (1991) as "the simple statement that security prices fully reflect all available information."

That is, you can't open the Wall Street Journal, read a news article from this morning about Google's great earnings numbers that were just released, and make money by buying Google stock. The positive information contained in the earnings numbers would already have been incorporated into Google's share price.

To put it another way, the EMH simply says that there is no such thing as a free lunch for investors.

Does this imply that stock prices (or other asset prices) are unpredictable? No! The EMH unequivocally does not mean that prices or returns are unpredictable.

This fallacy arises all the time. Some author claims to have found a way to predict returns and so declares, "The EMH is dead." Return predictability does not invalidate the EMH. This is important – the empirical evidence shows that returns are indeed eminently predictable.

The key lies with risk premia.

I. What are risk premia?
The price of a stock (or any other asset) can be decomposed into two parts:

1. The (discounted) expected value of the stock
2. A "risk premium"

The first part is the standard discounted present-value that you might read about in an accounting textbook. The second is the compensation required by the stock investor in order to bear the risk that the stock might drop in value, known as a risk premium.

To understand risk premia, suppose that I offer you the following deal. You can pay me $x, and then get to flip a coin: heads I give you $100, tails you get nothing. How much would you be willing to pay to have this opportunity?

Although the expected value of this bet is $50, you're probably only going to be willing to pay something like $45 for the chance to flip the coin, if that. The five dollars difference is the compensation you demand in order to bear the risk that you could lose all your money – the risk premium.

II. Return predictability is compensation for risk
The above decomposition suggests that return predictability can either be the result of

1. The ability to truly predict movements in the underlying value of the stock
2. The ability to predict risk premia

If the first type of predictability were possible, this would in fact invalidate the EMH. However, the second sort of predictability – predictability of risk premia – allows for stock returns to be predictable, even under the EMH.

This is because, if only risk premia are predictable, then there is still no free lunch.

Sure, you can predict that a stock portfolio will outperform the market over the next year. However, this excess return is simply compensation for the fact that this set of stocks is extra risky – i.e., the portfolio has a high risk premium.

As an extreme example, consider the well-known fact that buying and holding a diverse basket of stocks predictably has higher expected returns than buying and holding short-term Treasury bills.

Is this a free lunch? Does the existence of the stock market invalidate the EMH? No. This return predictability exists only because equities are fundamentally riskier than T-bills.

III. Summing up
This is all to say that while returns may be predictable, it is likely that any profits earned from such predictable strategies are merely compensation for extra risk.

The EMH says that there is no free lunch from investing. Just because returns are predictable does not mean you can eat for free.

Postscript. There is another (outdated) theory, the "random walk hypothesis", defined as the claim that returns are not predictable. This is different from the EMH, which says that asset prices reflect all available information. The random walk hypothesis has been shown to be clearly empirically false, per links above.