Learning to Consume: Individual versus Social Learning

Obtaining the “exact” solution to the consumption-under-uncertainty problem can be an extraordinarily difficult mathematical problem. Allen and Carroll (2001), however, show that a simple linear approximation to the optimal policy rule can yield utility trivially close to that of the exact solution. Furthermore, Allen and Carroll show that consumers can consistently find the near-optimal approximate solution using simple trial-and-error statistical learning, which has an attractive and intuitive interpretation. Their result comes with a negative corollary: the amount of time necessary to consistently find a near-optimal rule is astronomical and unattainable under any plausible scenarios. This paper extends Allen and Carroll’s original agent-based model to incorporate social learning. A simple form of information sharing is introduced into the agents’ statistical learning process, and preliminary results indicate that this addition retains consumers’ ability to find an optimal rule while pushing the time required to find such a rule arbitrarily close to the lowest possible bound, within the “lifetime” of a single agent. Furthermore, a variation of the trial-and-error statistical learning process is identified which further decreases the amount of time required to find a near-optimal rule by a full order of magnitude. This variation also opens the door for the social learning process to incorporate heterogeneous initial endowment values across agents, a welcome extension to the original model.