The Effect of Electronic Commerce on Geographic Trade and Price Variance in a Business-to-Business Market

Size: px
Start display at page:

Download "The Effect of Electronic Commerce on Geographic Trade and Price Variance in a Business-to-Business Market"

Transcription

1 The Effect of Electronic Commerce on Geographic Trade and Price Variance in a Business-to-Business Market Eric Overby and Chris Forman, Georgia Institute of Technology 1

2 Motivation Supply and demand forces often cause the price for the same good to vary across geographic locations. I.e., geographic price variance (dispersion) is common. Economic theory suggests that the Law of One Price ( LOP ) should hold i.e., that the price for the same good should not vary across locations by more than the cost of transport. If the LOP does not hold, then supply and demand will shift until it does hold. $11,500 $9,000 $12,000 $9,500 $11,500 However, this assumes that buyers and sellers have the information necessary to shift demand and supply in an optimal way. Information is costly, and information search costs have long been posited as an explanation for violations of the Law of One Price (i.e., Stigler, 1961). Broad Research Question: How does information technology break down this type of geographic segmentation within markets? 2

3 Facilities Across Country Imbalances in supply and demand across facilities create geographic price dispersion and violations of the law of one price. 101% 95% 102% 3

4 How Should the Market React? Sellers should shift supply from low-price markets to high-price markets. However, there are two limitations to this: a) Seller Limitation 1: The ability to forecast prices at each facility. b) Seller Limitation 2: The cost of transporting vehicles. 97% 95% 101% 100% 102% 97% 95% 101% 100% 102% Buyers should shift demand from high-price markets to low-price markets. However, there are two limitations to this: a) Buyer Limitation 1: The ability to forecast prices at each facility. b) Buyer Limitation 2: The cost of transporting vehicles. c) Buyer Limitation 3: The cost of bidding at each facility, which traditionally has required physical attendance. 4

5 Information Technology Allows Buyers to Buy Remotely 5

6 How Should the Market React? Sellers should shift supply from low-price markets to high-price markets. However, there are two limitations to this: a) Seller Limitation 1: The ability to forecast prices at each facility. b) Seller Limitation 2: The cost of transporting vehicles. 97% 95% 101% 100% 102% 97% 95% 101% 100% 102% Buyers should shift demand from high-price markets to low-price markets. However, there are two limitations to this: a) Buyer Limitation 1: The ability to forecast prices at each facility. b) Buyer Limitation 2: The cost of transporting vehicles. c) Buyer Limitation 3: The cost of bidding at each facility, which traditionally has required physical attendance. 6

7 The Sample All transactions from January 2003 to June 2008 for vehicles with mileage between 15,000 and 21,000. Over 2.5 million transactions. # of vehicles > 100,000 50, ,000 10,000-50,000 5,000-10,000 < 5,000 7

8 Q103 Q104 Q105 Q106 Q107 Q108 Descriptive Evidence Proportion of Webcast Transactions Coefficient of Variation of Prices Geographic price variance is declining as webcast trading is increasing. I believe this is causal, and I will show you why I believe this. 8

9 Buyer Behavior H1: Buyers will use the webcast channel to shift demand from local facilities where prices are high to remote facilities where prices are low. $11,000 I.e., when using the webcast channel, buyers will be: - More sensitive to price - Less sensitive to distance U ijkt = β 0k + β 1 *NormPrice jkt + β 2 *Distance ik + $13,000 $11,500 β 3 *Supply jkt + β 4 *AvgCondition jkt + ε ijkt i = buyer; j =make/model; k = facility; t = day Assume the market average price is $12,000. Conditional on Physical Channel Coefficient Conditional on Webcast Channel Coefficient β 1 : NormPrice (0.039) *** (0.153) *** β 2 : Distance (0.000) *** (0.000) *** β 3 : Supply (0.001) *** (0.002) *** β 4 : AvgCondition (0.009) *** (0.029) *** β 0,k : Facility constants included included n (number of choices) 207,484 20,236 Log likelihood Robust standard errors in parentheses. *** p < 0.01 (We also estimated these simultaneously via a nested logit model; results are similar.) When using the electronic channel, a buyer is less sensitive to distance (because he doesn t have to travel) and more sensitive to price (because he has better price information across facilities.) 9

10 Q103 Q104 Q105 Q106 Q107 Q108 The Effect on Geographic Price Variance To examine how this change in buyer behavior affects geographic price variance, we focus on the specific mechanism through which this should occur: the cross-facility purchase. These occur when a buyer local to facility A purchases from facility B. 101% 95% 102% 100% 75% 50% 25% 0% 60% 55% 50% 45% 40% % Cross Facility Purchases - via Electronic Channel (Left Axis) % Cross Facility Purchases - via Physical Channel (Left Axis) Percentage of Electronic Purchases (Left Axis) % Cross Facility Purchases - Total (Right Axis) 10

11 The Effect on Geographic Price Variance: Specification CVPrice jklt = α + β 1 CrossFacilityPurchases jklt + β 2 CrossFacilityPurchases jklt * Distance kl + β 3 SameFacilityPurchases jklt + β 4 SameFacilityPurchases jkl * Distance kl + β t time t + c jkl + ε jklt We measured price variance between facilities as follows. 1) Calculated the average price of each vehicle model at each facility. 2) Calculated the coefficient of variation of the two average prices for each auction pair where the same vehicle model was traded. Ford Taurus: $11,000 Coefficient of Variation: Ford Taurus: $13,000 11

12 The Effect on Geographic Price Variance: Results Coefficient (Std. Error) - All Coefficient (Std. Error) With Trades β 1 : CrossFacilityTrans jklt (0.0001) *** (0.0001) *** β 2 : CrossFacilityTrans jklt * Distance kl (0.0002) *** (0.0002) *** β 3 : SameFacilityTrans jklt (0.0000) *** (0.0000) *** β 4 : SameFacilityTrans jklt * Distance kl (0.0000) *** (0.0000) Intercept (0.0005) *** (0.0007) *** Time dummies included included Vehicle model / Facility pair fixed effects included included R 2, including fixed effects N 3,986,978 1,100,421 Cross-facility trades, which are enabled by the webcast channel, are associated with lower price variance (β 1 ), although the effect attenuates with distance (β 2 ). At distance = 250 miles (the average distance for a webcast purchase), the marginal effect of one cross-facility trade is to reduce price variation by 1.3%. 12

13 Testing the Entire Market CVPrice jt = α + β 1 CrossFacilityPurchases jt + β 2 SameFacilityPurchases jtt + β t time t + c j + ε jt. Variable Description Mean (St. Dev.) CVPrice jt Coefficient of variation of price for vehicles of model j in 0.23 (0.23) time period t. CrossFacility Number of cross-facility purchases for vehicles of model j (307.12) Purchases jt in time period t. SameFacility Purchases jt Number of same-facility purchases for vehicles of model j in time period t (338.05) Coefficient (Std. Error) - All β 1 : CrossFacilityTrans jklt (0.0000) *** β 2 : SameFacilityTrans jklt (0.0000) Intercept (0.0056) *** Time dummies included Vehicle model / Facility pair fixed effects included R 2, including fixed effects 0.66 N 8,017 A one st. dev. increase in cross-facility trades, is associated with a 13% decrease in price variation throughout the country. 13

14 Reconciling the Two Prior Results The reduction in price variance attenuates with distance, yet shows up for the market as whole. Why? 1) Unusually high (low) prices are eliminated. 2) Transitive property 14

15 Conclusions The webcast channel allows buyers to shift their demand from local facilities where prices are high to remote facilities where prices are low. This reduces the geographic price variance in the market. Due to the cost of transportation, this phenomenon attenuates with distance, but the effects are evident across the market. Sellers are reacting to the increased market efficiency by becoming less strategic about how they distribute their vehicles. Contributions include: - Examining and documenting the micro-level mechanism. - Considering the role of location. - Using transaction prices determined by auction. 15