There is a simple way to get overoptimism (and hence bubbles): suggest that the agent's payoff is correlated with his guess. Here's another:
The observer’s beliefs are different from agent A’s. They are drawn from the same distribution G but there is no reason that the observer’s beliefs are the same as agent A’s. In fact, the action agent A took will only be the best one from the observer’s perspective by accident. Actually, the observer’s beliefs will be the average of the distribution G which is lower than the belief of agent A since agent A deliberately took the action which he thought was the best. This implies that the agent A who took the action is “overoptimistic” relative to an arbitrary observer.