Preference Learning and Demand Forecast

Abstract

Understanding consumer preferences is important for new product management, but isfamously challenging in the absence of actual sales data. Stated-preferences data are rel-atively cheap but less reliable, whereas revealed-preferences data based on actual choicesare reliable but expensive to obtain prior to product launch. We develop a cost-effectivesolution. We argue that people do not automatically know their preferences, but canmake an effort to acquire such knowledge when given sufficient incentives. The methodwe develop regulates people’s preference-learning incentives using a single parameter,re-alization probability, meaning the probability with which an individual has to actuallypurchase the product she says she is willing to buy. We derive a theoretical relation-ship between realization probability and elicited preferences. This allows us to forecastdemand in real purchase settings using inexpensive choice data with small to moderaterealization probabilities. Data from a large-scale field experiment support the theory, anddemonstrate the predictive validity and cost-effectiveness of the proposed method.