Reference-Dependent Pricing

I provide the first field causal evidence that consumers’ reference prices, shaped by past price exposures, affect current demand. In a large-scale two-stage field experiment across over 800 vending machines, I first randomly assign machines to different base price levels for 3 to 6 months, then implement common discounted prices across all machines. Arrivals dip immediately after the price increase, then recover over time, consistent with consumers forming new reference prices. In the second stage, consumers with a 10% higher reference price generate 1 to 3% more net revenue when facing identical current prices. I estimate a structural demand model that separates price sensitivity from reference dependence, enabling counterfactual policy analysis. In my setting, a firm that correctly accounts for reference dependence earns about 28% higher profits than one relying on a standard A/B test, which conflates loss aversion with price sensitivity. Most of this gain (about 21 percentage points) comes from correcting the demand model; dynamic pricing adds another 7 points on top.