We use a simple cost-benefit analysis to derive optimal similarity judgments – addressing the question: when should we expect a decision maker to distinguish between different time periods or different prizes? Our key premise is that cognitive resources are costly and are to be deployed only where they really matter. We show that this simple insight can explain a number of observed anomalies, such as: (i) time inconsistency, (ii) magnitude effects, (iii) interval length effects. For each of these phenomena, our approach allows to identify the direction of the bias relative to the benchmark case where cognitive resources are costless. Finally, we show that, when applied to choice under risk, the same insights predict anomalies such as the ratio and certainty effects, and rationalize Rabin's risk aversion paradox. This suggests that the theory may provide a parsimonious explanation of behavioral anomalies in different contexts.

We examine an analyst with career concerns making cheap talk recommendations to a sequence of traders, each of whom possesses private information concerning the analyst's ability. The recommendations of the analyst influence asset prices that are then used to evaluate the analyst. An endogeneity problem thus arises. In particular, if the reputation of the analyst is sufficiently high then an incompetent but strategic analyst is able to momentarily hide her type. An equilibrium in which the market eventually learns the analyst type always exists. However, under some conditions, an equilibrium also exists in which the incompetent analyst is able to hide her type forever.

We provide a universal condition for rationalizability by risk-averse expected utility preference in a demand-based framework with multiple commodities. Our test can be viewed as a natural counterpart of a classical test of expected utility, due to Fishburn (1975), in a demand setting.