Non-parametric bounds for non-convex preferences

Choices from linear budget sets are often used to recover consumer's preferences. The classic method uses revealed preference theory to construct non-parametric bounds on the indifference curve that passes through a given bundle. We show that these bounds do not apply to non-convex preferences, and therefore may lead to erroneous predictions and welfare analysis. We suggest an alternative that is based solely on the assumption of monotonicity of preferences.