John Quiggin, author of Zombie Economics, graciously engages this post I made in response to his book in the comments. He also points to an interesting and easy to read paper he wrote on the equity premium puzzle. My real critique is that the equity premium puzzle isn't an EMH puzzle it's a macroeconomics puzzle. Let me try to explain the difference.
In Quiggin's book and definition of EMH he is confounding two ideas:
Efficient Markets Hypothesis - prices fully reflect all available information.
Market Efficiency - a well functioning market will allocate all resources in the best possible way
The second idea is often strictly defined as the First Fundamental Welfare theorem, which basically says that if you make a bunch of assumptions (that aren't true) like markets are perfectly competitive, markets are complete, no externalities and perfect information, then the market outcome is efficient. In this case, efficient means pareto efficient, a weak notion that means we can't make one person better off without hurting anyone else.
This confusion explains a lot of odd statements he makes, like, "The Efficient Markets Hypothesis implies that governments can never outperform well-informed financial markets."
The first fundamental theorem does say that (under very restrictive assumptions) and if our notion of improvement is limited to pareto efficiency, but the EMH just says markets adjust in the background. Let's say there is an negative externality called pollution. A government may be able to enact a policy to limit pollution and make everyone better off, say a tax on gasoline. The EMH just says that stocks will adjust to that new policy. The tax on gasoline will limit the amount of gasoline consumed, so companies like Exxon and BP will see their profits decrease. People will probably drive less, so Ford's stock will go down. The EMH just says that security prices will reflect all available information.
The EMH is still not a weak hypothesis. If we imagine that the government is deciding on whether or not to enact the policy, the market prices must be constantly adjusting to the updating chances that the knew policy will be enacted. The prices must be the best guess for the properly discounted value of future dividends for every stock in the market, even stocks that will only be barely affected by the law, like a theme park that will lose profits, because it is far out of town and a few less people will want to make the drive.