Preference Discovery

Is the assumption that people automatically know their own preferences innocuous? We present a theory and an experiment that study the limits of preference discovery. Our theory shows that if tastes must be learned through experience, preferences for some goods will be learned over time, but preferences for other goods will never be learned. This is because sampling a new item has an opportunity cost. Learning is less likely for people who are impatient, risk averse, low income, or short-lived, and for consumption items that are rare, expensive, must be bought in large quantities, or are initially judged negatively relative to other items. Preferences will eventually stabilize, but they need not stabilize at true preferences. A pessimistic bias about untried goods should increase with time. Agents will make choice reversals during the learning process. Welfare loss from sub-optimal choices will decline over time but need not approach zero. Results from an online experiment show that learning occurs as predicted and with the expected biases, but with even more error than our theory suggests. Overall, our results imply that undiscovered preferences could confound interpretation of choice data of all kinds and could have significant welfare and policy implications.