Price dispersion with multi-product firms and vertical differentiation

ABSTRACT

Empirical evidence from the gasoline market suggests that prices and markups for regular gasoline are higher in stations offering both premium and regular compared to stations without premium. Also, the difference is larger in affluent areas. This implies that consumers pay higher prices for the same low-quality product when they visit stores offering both high- and low-quality products in areas that are populated with more high-income consumers.

KEYWORDS:

• Gasoline
• price dispersion
• vertical differentiation
• nearest-neighbor matching
• average treatment effect

JEL CLASSFICATION:

• D31
• D43
• L11

Acknowledgments

I would like to thank Kathryn Graddy, George Hall, Taehoon Kim, Raphael Schoenle, Benjamin Shiller, anonymous referees, and seminar participants at Brandeis University and the 2018 Asian Meeting of the Econometric Society for helpful comments and conversations.

Disclosure statement

No potential conflict of interest was reported by the author.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

¹ The information about a car wash, a repair shop, and a convenience store is reported voluntarily by stations. This information is available for around 70% of stations.

² KNOC provides the monthly consumption of regular and premium gasoline for each district.

³ For the stations using non-major brands, I use the average wholesale prices of the four major brands, because these stations purchase gasoline from the major oil companies. There could be other factors that influence the wholesale price, including transportation costs, wages, and contract types. However, the portion of wages in marginal costs is very small, because most stations pay only the minimum wage rate. Also, I control for geographical differences in cost factors by comparing nearby stations. The effect of different contract types can be controlled by having brand fixed effects. Moreover, the data includes more than one million observations and the stations with non-major brands take only 5% of total observations. The results do not change qualitatively when these stations are dropped.

⁴ In search models, non-searchers are randomly assigned to sellers (Varian 1980; Burdett and Judd 1983).

⁵ See, for example, Hottman, Redding, and Weinstein (2016) and Midrigan (2011).

⁶ $\varphi$ can be considered as quality. Even though gasoline is a homogeneous good, stations with locational advantages will have a high $\varphi$.

⁷ The expenditure on each product is normalized to one.

⁸ The optimal prices of single-product firms are $P_{f_h}=\frac{{ϵ}_{hh}}{{ϵ}_{hh}-1}M{C}_{f_h}$ and $P_{f_l}=\frac{{ϵ}_{ll}}{{ϵ}_{ll}-1}M{C}_{f_l}$.

⁹ The daily prices are aggregated to the monthly level.

¹⁰ There could be other unobserved characteristics that are not captured by the baseline specification, such as cleanliness and types of consumer goods sold. Because I compare two stations selling both premium and regular gasoline (the treatment group) and selling only regular gasoline (the control group), these factors will affect the results only when the two groups of stations have systemically different characteristics. However, there is no evidence of this. Therefore, I assume that these factors are randomly distributed and captured by the random error term. In addition, it is easy to find a convenience store in Korea and most of the revenues of gas stations come from selling gasoline. Therefore, types of consumer goods sold are not an important factor in the Korean gasoline market.

¹¹ Most of the results are significant at the 1% level.

¹² Results are available upon request.