Working Paper: NBER ID: w30778
Authors: Amil Petrin; Mark Ponder; Boyoung Seo
Abstract: The standard Berry, Levinsohn, and Pakes (1995) (BLP) approach to estimation of demand and supply parameters assumes that the product characteristic observed by consumers and producers but not the researcher is conditionally mean independent of observed characteristics. We extend BLP to allow all product characteristics to be endogenous, so the unobserved characteristic can be correlated with the observed characteristics. We derive moment conditions based on the assumption that firms choose product characteristics to maximize expected profits given their beliefs at that time about market conditions and that the “mistake” in the amount of the characteristic that is revealed once all products are on the market is conditionally mean independent of the firm’s information set. Using the original BLP dataset we find that observed and unobserved product characteristics are highly positively correlated, biasing demand elasticities upward, as average estimated price elasticities double in absolute value and average markups fall by 50%.
Keywords: Discrete choice models; Demand estimation; Product characteristics; Endogeneity
JEL Codes: C25; L0
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| traditional BLP approach (C51) | biased demand elasticities (D11) |
| higher prices (D49) | higher unobserved quality (L15) |
| higher unobserved quality (L15) | consumers appear less price-sensitive (D12) |
| correlation between observed prices and unobserved characteristics (C10) | upward bias in demand elasticity estimates (D12) |
| extended model (C59) | estimated price elasticities double in absolute value (D11) |
| extended model (C59) | average markups fall by 50% (D43) |
| correlation between observed and unobserved characteristics (C10) | understanding consumer behavior and market dynamics (D12) |