Assessing the Gains from E-Commerce Paul Dolfen, Stanford Liran Einav, Stanford and NBER Pete Klenow, Stanford and NBER Ben Klopack, Stanford Jonathan Levin, Stanford and NBER Larry Levin, Visa Wayne Best, Visa June 2019 Technology Diffusion and Productivity Workshop Federal Reserve Bank of Richmond 1 / 45
What we do Document the rise of e-commerce using Visa data Estimate resulting consumer surplus > 1% of consumption Find gains are increasing in county population density Find gains are twice as big for incomes above $50k 2 / 45
Related literature Gains from e-commerce and the internet Brynjolffson and collaborators (2003, 2012, 2017) Goolsbee and Klenow (2006, 2018) Syverson (2016) Couture, Faber, Gu and Liu (2018) Allcott, Braghieri, Eichmeyer and Gentzkow (2019) Consumer surplus from new products more generally Feenstra (1994) Hausman (1997, 1999) Weinstein and collaborators (2006, 2010, 2018, 2019) 3 / 45
Outline Visa data and basic facts 1 Estimating the pure convenience gains from shopping online 2 Estimating the variety gains from e-commerce 3 4 / 45
Visa data Raw data is similar to line items in monthly statements: Transaction amount and day Unique card identifiers (credit and debit) Store name, NAICS, ZIP (longitude-latitude in recent years) January 2007 through December 2017 Merged with Experian data the last few years: Card income Card location 5 / 45
Visa data confidentiality All results have been reviewed to ensure that no confidential information about Visa merchants or cardholders is disclosed. Cards are anonymized, and we report no data on individual cards. Cardholder information is based solely on the card’s transactions. We report no data on specific merchants or from recent months – which is why the analysis sample ends in December 2017. 6 / 45
Visa data caveats No details on items bought or prices Cannot tie multiple cards to households Tremendous card turnover Will rely heavily on monetized distance to get at WTP 7 / 45
Visa summary statistics U.S. annual averages from 2007 through 2017 380 million cards 35.9 billion transactions $1.93 trillion in sales ◮ 55% credit, 45% debit 8 / 45
Flowing through Visa 22% 20% Visa share of consumption 18% 16% 14% 12% Visa as a share of GDP 10% 8% 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Sources: Visa and BEA 9 / 45
Experian data Consumer credit reporting agency Merged with Visa cards (only in recent years) Can match roughly 50% of Visa credit cards 2016–2017 Cardholder demographics (e.g. income and education) 10 / 45
E-commerce in the Visa data Visa transaction flags: CP ≡ Card Present (brick-and-mortar) CNP ≡ Card Not Present ◮ phone or mail order ◮ recurring bill payments ◮ ECI ≡ e-commerce indicator ◮ missing values For missing values we allocate within 3-digit NAICS years: ECI e-commerce = ECI + phone/mail/recurring × CNP 11 / 45
E-Commerce industries Retail Example Nonstore Retail Amazon Clothing Nordstrom Misc Retail Staples General Merchandise Walmart Electronics Best Buy Building Material, Garden Supplies Home Depot Furniture Bed Bath & Beyond Sporting Goods, Hobby Nike Health, Personal Care CVS Food Safeway Ground Transportation Uber Non-Retail Example Admin, Support Services Expedia Travel Air Transportation American Airlines Accommodation Marriott Car Parts AutoZone Rental Services Hertz Rent-A-Car 12 / 45
13 / 45
Estimating e-commerce in the U.S. overall U.S. Online Share = Total Card Spending · Visa Online Share Consumption Calculate e-commerce share in Visa as described above Assume Visa representative of all card transactions Assume non-card transactions are all offline 14 / 45
Share of U.S. consumption online 15 / 45
Estimating e-commerce by county-income group Fraction of households with cards: α cy ∝ # of Visa Cards cy � Tax Filers cy Fraction of all consumption on e-commerce for each county-income: s cy ∝ Visa online spending cy � · � α cy Total Visa spending cy 16 / 45
17 / 45
E-commerce share by population density and income Online share of all consumer spending: Below-median density counties 6.4% Above-median density counties 9.1% Cardholder income ≤ $50k 3.4% Cardholder income > $50k 9.7% 18 / 45
Outline Visa data and basic facts 1 Estimating the pure convenience gains from shopping online 2 Estimating the variety gains from e-commerce 3 19 / 45
Outline Visa data and basic facts 1 Estimating the pure convenience gains from shopping online 2 Estimating the variety gains from e-commerce 3 20 / 45
Consumer problem � M � σ � σ − 1 ( q m · x m ) 1 − 1 max U = σ m =1 subject to M � M φ b F b + M φ o F o + p m · x m ≤ w m =1 q m = “quality” of merchant m x m = quantity purchased from merchant m p m = price per unit at merchant m M = M b + M o = total merchants bought from M b ( M o ) = # of merchants shopped at in-store (online) F b ( F o ) = scale of fixed costs for shopping in-store (online) 21 / 45
Comments on the consumer problem Merchants are either online or offline (not both) ◮ Broadly consistent with low merchant overlap within cards σ > 1 is the elasticity of substitution across merchants ◮ σ < ∞ ⇒ “love of variety” φ governs how fast fixed shopping costs rise with the # of online and brick-and-mortar merchants shopped at ◮ φ > 1 so we get an interior solution despite love of variety 22 / 45
Producer problem max π m = p m y m − wL m − wK j p m subject to M j y m = Lx m and y m = Z m L m M j,market j = o or b M j ≤ M j,market Brick-and-mortar (online) sellers split their market evenly K j = overhead labor needed to operate 23 / 45
Free entry and market clearing For each market j : E j [ π m ] = 0 Labor market clearing: � L = L m + L b + L o + M b,market K b + M o,market K o m 24 / 45
Shopping technology L · M φ = Y b = A b L b b L · M φ o = Y o = A o L o Perfectly competitive so marginal cost pricing: w F b = A b w F o = A o 25 / 45
Symmetric technologies Process efficiency: Z m = Z Quality offline: q m = q b for m ∈ M b,market Quality online: q m = q o for m ∈ M o,market 26 / 45
Symmetric outcomes Pricing: σ − 1 · w σ p m = p = Z Spending per merchant online ( o ) and offline ( b ): � q o � σ − 1 o b = q b Profits: M o L · o π o = σ − wK o M o,market M b L · b π b = σ − wK b M b,market 27 / 45
Merchants in GE � q o � φ − 1 ( σ − 1) � A o � φ 1 φ − 1 Define k ≡ q b A b 1 + k · 1 1 1 + ( σ − 1) φ · L ( σ − 1) φ M b,market = σ · K b 1 + k · 1 k 1 + ( σ − 1) φ · L ( σ − 1) φ M o,market = σ · K o � � 1 1 1 φ M b = 1 + ( σ − 1) φ · 1 + k · A b � � 1 1 k φ M o = 1 + ( σ − 1) φ · 1 + k · A o 28 / 45
GE comparative statics M o,market M o o M b,market M b b A o + + 0 A b q o + + + q b 29 / 45
Online spending share Let s o denote the share of card spending online: oM o k s o ≡ = oM o + bM b k + 1 � � φ − 1 ( σ − 1) � � φ 1 q o A o φ − 1 where k ≡ q b A b s o rises with q o /q b and A o /A b Consumers gain from rising s o if it is due to a combination of better (rising q o ) and easier to access (rising A o ) online options 30 / 45
Welfare Consumption-equivalent welfare is proportional to Z · M 1 / ( σ − 1) · ¯ q where average quality is � q bσ − 1 · M b + q oσ − 1 · M o � 1 / ( σ − 1) q ≡ ¯ M 31 / 45
Welfare gain from e-commerce In terms of exogenous variables, welfare is proportional to � � φ − 1 1 1 1 φ φ φ σ − 1 φ − 1 ( σ − 1) A φ − 1 ( σ − 1) A φ − 1 φ − 1 Z · q b + q o o b For given Z , q b , and A b , welfare is increasing in s o : � � φ − 1 1 1 φ ( σ − 1) φ ( σ − 1) Z · q b · A b 1 − s o 32 / 45
Quantitative strategy Calibrate: φ = convexity of fixed shopping costs σ = elasticity of substitution across merchants Then infer the welfare gain from the path of s o 33 / 45
Estimating φ (convexity of fixed shopping costs) According to the model, we can estimate φ using one of two regressions that yield the same answer by construction: ln M = α + 1 φ · ln ( oM o + bM b ) � oM o + bM b � = η + φ − 1 ln · ln ( oM o + bM b ) M φ Extensive and intensive margin Engel Curve slopes should reflect φ Caveat : This assumes idiosyncratic fixed costs are uncorrelated with a card’s total expenditures 34 / 45
Estimates of φ (convexity of fixed shopping costs) 2007 2017 � φ 1.73 1.75 # of cards 283M 462M R 2 0.67 0.67 Standard errors are tiny ... 35 / 45
Estimating σ Assuming distance is uncorrelated with preferences (controlling for merchant fixed effects), we can use how visits change with distance to estimate σ Aggregating merchant pairs { j , k } with the same { dist ij , dist ik } : � Trips j � � q j � � p jk + τ ij � ln = ln − σ · ln Trips k q k p jk + τ ik ◮ p jk = average ticket size at merchants j , k ◮ τ = transportation costs for i to j or k ◮ τ = 0 for online transactions ◮ Capture relative quality with cross fixed effects ◮ Regress on both online-offline and offline-offline samples 36 / 45
Transactions online vs. distance to a physical store 37 / 45
Recommend
More recommend
