| The first essay examines the connection between finite memory and biases in information processing. A single, infinitely-lived agent must eventually make a decision; the correct decision depends on the state of the world, which is unknown. An information process generates signals which are informative about the state, and terminates after each signal with probability eta. The decision-maker cannot remember all of the signals that he observes: he is constrained to use a finite memory, consisting of a finite set of memory states, and a rule which specifies the transition between states as new information is received. It is shown that the optimal memory rule may perform very poorly in the short run, and can explain several biases that psychologists have observed. In particular, it is shown that if eta is close to zero, then the decision-maker optimally ignores information with very high probability in some memory states. As a result, he appears to display a confirmatory bias (tendency to ignore or misinterpret information which does not support his first impressions), and an overconfidence/underconfidence bias (tendency to infer too much from ambiguous information, too little from precise information).; The second essay examines a dynamic voting game, in which three agents decide whether to amend an existing policy. The policy space is one-dimensional; the voters have single-peaked preferences, and differ in their ideal points. In addition, each agent may be either rational or behavioral: a rational agent cares only about the policy implemented, while a behavioral agent incurs an additional cost from supporting policies which he dislikes. The objective is to determine whether rational agents have an incentive to mimic the behavioral types, and whether this can result in inefficient agreements. It is shown that if votes are observable, then there is no inefficiency: rational voters will immediately agree to implement the median ideal point. If votes are unobservable, then reputation effects can arise: if the status quo is sufficiently far to the left extreme, then the voter with the right-most preferences will pretend to be a behavioral type; this will result in a new agreement too far to the right. |
