price dispersion in the gasoline market
play

Price Dispersion in the Gasoline Market Chandra and Tappata (2009) - PowerPoint PPT Presentation

Price Dispersion in the Gasoline Market Chandra and Tappata (2009) Beia Spiller Search Lunch Group October 2009 Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 1 / 26 Why do we see price dispersion in


  1. Price Dispersion in the Gasoline Market Chandra and Tappata (2009) Beia Spiller Search Lunch Group October 2009 Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 1 / 26

  2. Why do we see price dispersion in gasoline markets? Di¤erent search protocols: Nonsequential search (Varian, 1980) Sequential search (Stahl, 1989) Di¤erent consumer types: Commuters: Nonsequential search Occasional Drivers/ Tourists: Sequential search Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 2 / 26

  3. Model Homogenous good n …rms: compete on prices, constant unit production cost c unit mass of consumers: inelastic demands, valuation v λ 2 ( 0 , 1 ) : zero search cost consumers- Shoppers µ 2 ( 0 , 1 ) : search intensity Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 3 / 26

  4. Equilibrium Given µ , unique symmetric NE of mixed strategies. Firms draw simultaneously from price distribution: � ( 1 � µ ) ( v � p ) � 1 ( n � 1 ) F ( p ; c , v , n ) = 1 � µ n ( p � c ) where: � � p � = cn µ + ( 1 � µ ) v p 2 1 + ( n � 1 ) µ , v as µ ! 1: p * = c (competition) as µ ! 0: p * = v (monopoly) Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 4 / 26

  5. Equilibrium Amount of search is consistent with …rms’ pricing strategies Consumers calculate gains from search: GS = E [ p � p min j µ , c , v , n ] = Z v h 1 � n [ 1 � F ( p ; c , v , n )] n � 1 i p dF ( p ; c , v , n ) p � Gains from search is a concave function of µ (GS=0 if µ = (0,1)), and increases with number of …rms. Search costs are drawn from distribution G ( s i ) with s i 2 [ 0 , s ] Equilibrium: 1. Indi¤erent consumer: GS = e s 2. Search intensity: µ = λ + G ( e s ) Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 5 / 26

  6. Search Intensity, Price Dispersion and Number of Firms Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 6 / 26

  7. Search Intensity and Price Dispersion b µ : search intensity that maximizes GS µ < b µ : (+) search intensity = ) (+) price dispersion µ > b µ : (+) search intensity = ) (-) price dispersion Any relation between search costs and price dispersion is consistent with search. Need to compare price dispersion for products with di¤erent search costs Can identify e¤ect of number of …rms, production costs, and consumer search on price dispersion and price levels. Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 7 / 26

  8. Entry and Exit on GS If search intensity is exogenous: As n increases: fraction of uninformed consumers decreases probability of being lowest price decreases ) …rms more likely to set extreme prices: price dispersion increases with n gains from low prices < gains from high prices = ) price distribution shift towards higher prices Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 8 / 26

  9. Entry and Exit on GS If search intensity can change: As n increases, price dispersion increases If µ < b µ : (+) search = ) (+) price dispersion If µ > b µ : (+) search = ) (-) price dispersion, but can’t o¤set (+) e¤ect of n . However, increased search decreases price, so cannot sign total e¤ect on average prices. Final sign: relationship between average price and n is non-monotonic Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 9 / 26

  10. Production Costs and Price Dispersion Holding µ constant: As costs increase, p m -p * decreases = ) …rms set higher, less dispersed prices Allowing µ to change: As costs increase, searching decreases. New equilibrium: higher and less dispersed prices Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 10 / 26

  11. Search Costs and Price Dispersion Higher search costs: less shoppers; change in distribution of search costs. Di¤erent fuels have di¤erent search costs: Premium fuels SC > Regular fuels SC Greater SC = ) (-) µ = ) (+) Prices � � µ Q b Price dispersion depends on initial equilibrium µ If initial equilibrium µ < b µ : higher search costs = ) (-) dispersion If initial equilibrium µ > b µ : higher search costs = ) (+) dispersion Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 11 / 26

  12. Search Costs and Price Dispersion Need to identify local equilibrium of price dispersion and search intensity Use the di¤erence in search costs between Regular and Premium to identify equilibrium Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 12 / 26

  13. Data Gasoline Data: Oil Price Information Service Daily service station data for 120,000 stations, in 4 states Station-date-fuel-type triple Costs: Spot prices from Energy Information Administration Proxy for changes in wholesale costs faced by stations Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 13 / 26

  14. Data Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 14 / 26

  15. Data Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 15 / 26

  16. Data Market de…nitions: Market center: gas station reporting prices on a given date Entire market: center station plus all surrounding stations falling within a speci…ed region. Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 16 / 26

  17. Results: Price Dispersion and Search Intensity Observed data: price dispersion is 43% higher for Premium Fuel than for Regular Fuel This implies µ > b µ Thus, gasoline market is closer to a competitive outcome than monopoly Search intensity decreases price dispersion Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 17 / 26

  18. Temporal Price Dispersion Consumer search models predict sales because of mixed strategies. Look at price ranking over time: sales and price dispersion would ‡ip ranking of stores. Rank reversals between stations i and j: T ij r ij = 1 ∑ I f p jt > p it g T t = 1 > 90% of all station pairs have positive rank reversals Cheaper station charges higher price � 15% of the time High average price spread, increases with rating: intensity of price competition is di¤erent across fuel types. Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 18 / 26

  19. Temporal Price Dispersion Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 19 / 26

  20. Imperfect Information and Price Dispersion Test: is imperfect information driving observed price dispersion? Natural Experiment: compare separate stations with kitty-corner stations 1. Kolmogorov-Smirnov Test: F 1 ( r ) > F 2 ( r ) Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 20 / 26

  21. Imperfect Information and Price Dispersion 2. OLS/Quantile Regression: r ij = β 0 + β 1 I ( corner ij ) + β 2 X ij where X ij are controls Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 21 / 26

  22. Results: Price Dispersion and Key Parameters How do costs a¤ect price dispersion? PRICEDISP jt = β 0 + β 1 MC t + β 2 AVPRICE jt Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 22 / 26

  23. Results: Price Dispersion and Key Parameters PRICEDISP jt = β 0 + β 1 STATEAVERAGE t + β 2 AVPRICE jt Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 23 / 26

  24. Results: Prices and Number of Firms More …rms = ) more price dispersion Test the concavity of this relationship How do number of …rms a¤ect markup? MARKUP it = β 0 + β 1 MC t + β 2 STATIONS i Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 24 / 26

  25. Robustness Checks Work with cleaned prices rather than actual/raw prices Regress prices on station FE, then run the original regression with residuals Assign single price observation to many overlapping markets = ) correlation in regressors and error term Run model on non-overlapping markets Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 25 / 26

  26. Conclusion Price dispersion and …rms’ markups are non-monotonic functions of: Search costs Number of …rms in market But, only if allow consumers to adjust search intensity Production costs: Decrease price dispersion Decrease markups Consumers could save 5% by shopping around for 1 mile Sources that alleviate imperfect information will reduce prices and price dispersion Beia Spiller (Search Lunch Group) Price Dispersion in the Gasoline Market 10/09 26 / 26

Recommend


More recommend