An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace - - PowerPoint PPT Presentation

▶
an empirical analysis of algorithmic pricing on amazon
SMART_READER_LITE
LIVE PREVIEW

An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace - - PowerPoint PPT Presentation

Dec 09, 2023 151 likes •1.27k views

An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace Le Chen, Alan Mislove, Christo Wilson Northeastern University The rise of e-commerce E-commerce Sales over time (US Commerce Department) 400B 300B Sales in Dollar 200B 100B

slide-1
SLIDE 1

An Empirical Analysis of Algorithmic Pricing

  • n Amazon Marketplace

Le Chen, Alan Mislove, Christo Wilson Northeastern University

slide-2
SLIDE 2

The rise of e-commerce

2

E-commerce Sales over time (US Commerce Department)

Sales in Dollar 0B 100B 200B 300B 400B Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

slide-3
SLIDE 3

The rise of e-commerce

2

E-commerce Sales over time (US Commerce Department)

Sales in Dollar 0B 100B 200B 300B 400B Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

slide-4
SLIDE 4

Algorithmic pricing

3

slide-5
SLIDE 5

Algorithmic pricing

  • Dynamic pricing algorithms or Revenue/Yield Management

3

slide-6
SLIDE 6

Algorithmic pricing

  • Dynamic pricing algorithms or Revenue/Yield Management
  • Use softwares to update prices in real time

3

slide-7
SLIDE 7

Algorithmic pricing

  • Dynamic pricing algorithms or Revenue/Yield Management
  • Use softwares to update prices in real time
  • Challenging in traditional retailers
  • Traditional retailers lack competitor prices in real time
  • Constraint of manually relabel prices of products on scale

3

slide-8
SLIDE 8

Algorithmic pricing

  • Dynamic pricing algorithms or Revenue/Yield Management
  • Use softwares to update prices in real time
  • Challenging in traditional retailers
  • Traditional retailers lack competitor prices in real time
  • Constraint of manually relabel prices of products on scale
  • Available in e-commerce

3

slide-9
SLIDE 9

Algorithmic pricing

  • Dynamic pricing algorithms or Revenue/Yield Management
  • Use softwares to update prices in real time
  • Challenging in traditional retailers
  • Traditional retailers lack competitor prices in real time
  • Constraint of manually relabel prices of products on scale
  • Available in e-commerce

3

slide-10
SLIDE 10

Problems in algorithmic pricing

4

slide-11
SLIDE 11

Problems in algorithmic pricing

  • Market failure

4

slide-12
SLIDE 12

Problems in algorithmic pricing

  • Market failure

4

slide-13
SLIDE 13

Problems in algorithmic pricing

  • Market failure

4

slide-14
SLIDE 14

Problems in algorithmic pricing

  • Market failure

4

slide-15
SLIDE 15

Problems in algorithmic pricing

  • Market failure

4

  • Price of seller 1

= 0.998 * (price of seller 2)

slide-16
SLIDE 16

Problems in algorithmic pricing

  • Market failure

4

  • Price of seller 1

= 0.998 * (price of seller 2)

  • Price of seller 2

= 1.27 * (price of seller 1)

slide-17
SLIDE 17

Problems in algorithmic pricing

  • Market failure

4

  • Price of seller 1

= 0.998 * (price of seller 2)

  • Price of seller 2

= 1.27 * (price of seller 1)

  • 0.998 * 1.27 > 1
slide-18
SLIDE 18

Problems in algorithmic pricing

5

slide-19
SLIDE 19

Problems in algorithmic pricing

5

  • Collusive strategies
slide-20
SLIDE 20

Problems in algorithmic pricing

5

  • Collusive strategies
slide-21
SLIDE 21

Goal of this study

6

slide-22
SLIDE 22

Goal of this study

  • Detect sellers who use algorithmic pricing (algo

sellers) and uncover the strategies they use

6

slide-23
SLIDE 23

Goal of this study

  • Detect sellers who use algorithmic pricing (algo

sellers) and uncover the strategies they use

  • Challenging: No ground truth

6

slide-24
SLIDE 24

Goal of this study

  • Detect sellers who use algorithmic pricing (algo

sellers) and uncover the strategies they use

  • Challenging: No ground truth
  • Quantify the impact of algo sellers in the

marketplace

6

slide-25
SLIDE 25

Goal of this study

  • Detect sellers who use algorithmic pricing (algo

sellers) and uncover the strategies they use

  • Challenging: No ground truth
  • Quantify the impact of algo sellers in the

marketplace

  • Scale, competitiveness, sales volume, etc.

6

slide-26
SLIDE 26

Goal of this study

  • Detect sellers who use algorithmic pricing (algo

sellers) and uncover the strategies they use

  • Challenging: No ground truth
  • Quantify the impact of algo sellers in the

marketplace

  • Scale, competitiveness, sales volume, etc.
  • Feedback loops of algorithms

6

slide-27
SLIDE 27

Outline of Amazon study

  • Introduction
  • Data collection
  • The Buy Box
  • Algorithmic pricing detection and analysis
  • Summary
slide-28
SLIDE 28

Data collection

8

slide-29
SLIDE 29

Data collection

  • Focus on the marketplace on

8

slide-30
SLIDE 30

Data collection

  • Focus on the marketplace on
  • Largest e-commerce site in the US and Europe

8

slide-31
SLIDE 31

Data collection

  • Focus on the marketplace on
  • Largest e-commerce site in the US and Europe
  • A marketplace populated by 3rd-party sellers and

Amazon itself

8

slide-32
SLIDE 32

Data collection

  • Focus on the marketplace on
  • Largest e-commerce site in the US and Europe
  • A marketplace populated by 3rd-party sellers and

Amazon itself

  • API provides capability to algorithmic pricing

8

slide-33
SLIDE 33

Data collection

  • Focus on the marketplace on
  • Largest e-commerce site in the US and Europe
  • A marketplace populated by 3rd-party sellers and

Amazon itself

  • API provides capability to algorithmic pricing
  • Near real-time price/product update

8

slide-34
SLIDE 34

Data collection

  • Focus on the marketplace on
  • Largest e-commerce site in the US and Europe
  • A marketplace populated by 3rd-party sellers and

Amazon itself

  • API provides capability to algorithmic pricing
  • Near real-time price/product update
  • Aggregated price information for product

8

slide-35
SLIDE 35

Data collection

9

slide-36
SLIDE 36

Data collection

  • Buy Box

9

slide-37
SLIDE 37

Data collection

  • Buy Box

9

slide-38
SLIDE 38

Data collection

  • Buy Box

9

slide-39
SLIDE 39

Data collection

  • Buy Box

9

slide-40
SLIDE 40

Data collection

  • Buy Box

9

  • Seller pages
slide-41
SLIDE 41

Data collection

  • Buy Box

9

  • Seller pages
slide-42
SLIDE 42

Data collection

  • Buy Box

9

  • Seller pages
slide-43
SLIDE 43

Data collection

  • Buy Box

9

  • Seller pages
slide-44
SLIDE 44

Data collection

  • Buy Box

9

  • Seller pages

82% of sales go to the Buy Box…

slide-45
SLIDE 45

Data collection

  • Buy Box

9

  • Seller pages

82% of sales go to the Buy Box… Win the Buy Box!

slide-46
SLIDE 46

Data collection

10

slide-47
SLIDE 47

Data collection

  • Cover Buy Box and seller pages

10

slide-48
SLIDE 48

Data collection

  • Cover Buy Box and seller pages
  • First crawl
  • 09/2014 — 12/2014, crawl every 25 minutes
  • 837 best selling products, all seller pages

10

slide-49
SLIDE 49

Data collection

  • Cover Buy Box and seller pages
  • First crawl
  • 09/2014 — 12/2014, crawl every 25 minutes
  • 837 best selling products, all seller pages
  • Second crawl
  • 08/2015 — 09/2015, crawl every 25 minutes
  • 1000 best selling products, first two seller pages

10

slide-50
SLIDE 50

Outline of Amazon study

  • Introduction
  • Data collection
  • The Buy Box
  • Algorithmic pricing detection and analysis
  • Summary
slide-51
SLIDE 51

Key questions

12

slide-52
SLIDE 52

Key questions

  • What features does Amazon use for choosing the

winner of Buy Box?

12

slide-53
SLIDE 53

Key questions

  • What features does Amazon use for choosing the

winner of Buy Box?

  • Can we predict the winner of Buy Box?

12

slide-54
SLIDE 54

Buy Box features

13

slide-55
SLIDE 55

Buy Box features

  • Features Amazon considers for each seller
  • Pricing: competing prices
  • Availability: stock of product, plan for vacations
  • Fulfillment: shipping fees, Fulfilled By Amazon (FBA)
  • Customer service: rating, positive feedbacks

13

slide-56
SLIDE 56

Buy Box features

  • Features Amazon considers for each seller
  • Pricing: competing prices
  • Availability: stock of product, plan for vacations
  • Fulfillment: shipping fees, Fulfilled By Amazon (FBA)
  • Customer service: rating, positive feedbacks

13

slide-57
SLIDE 57

Buy Box features

  • Features Amazon considers for each seller
  • Pricing: competing prices
  • Availability: stock of product, plan for vacations
  • Fulfillment: shipping fees, Fulfilled By Amazon (FBA)
  • Customer service: rating, positive feedbacks
  • Random forest

13

slide-58
SLIDE 58

Buy Box winner prediction

Accuracy 20 40 60 80 100 Number of sellers 1 5 10 15 20

Baseline: Lowest Price Prediction

14

slide-59
SLIDE 59

Buy Box winner prediction

Accuracy 20 40 60 80 100 Number of sellers 1 5 10 15 20

Baseline: Lowest Price Prediction

14

slide-60
SLIDE 60

Buy Box winner prediction

Accuracy 20 40 60 80 100 Number of sellers 1 5 10 15 20

Baseline: Lowest Price Prediction

14

slide-61
SLIDE 61

Buy Box winner prediction

  • Feature weights

Weight 0.1 0.2 0.3 0.4 Feature P r i c e D i f f L

  • w

e s t P r i c e R a t i

  • L
  • w

e s t P

  • s

F e e d b a c k F e e d b a c k c n t A v g r a t i n g I s F B A ?

15

slide-62
SLIDE 62

Buy Box winner prediction

  • Feature weights

Weight 0.1 0.2 0.3 0.4 Feature P r i c e D i f f L

  • w

e s t P r i c e R a t i

  • L
  • w

e s t P

  • s

F e e d b a c k F e e d b a c k c n t A v g r a t i n g I s F B A ?

15

slide-63
SLIDE 63

Buy Box winner prediction

  • Feature weights

Weight 0.1 0.2 0.3 0.4 Feature P r i c e D i f f L

  • w

e s t P r i c e R a t i

  • L
  • w

e s t P

  • s

F e e d b a c k F e e d b a c k c n t A v g r a t i n g I s F B A ?

15

slide-64
SLIDE 64

Buy Box winner prediction

  • Feature weights

Weight 0.1 0.2 0.3 0.4 Feature P r i c e D i f f L

  • w

e s t P r i c e R a t i

  • L
  • w

e s t P

  • s

F e e d b a c k F e e d b a c k c n t A v g r a t i n g I s F B A ?

15

slide-65
SLIDE 65

Outline of Amazon study

  • Introduction
  • Data collection
  • The Buy Box
  • Algorithmic pricing detection and analysis
  • Summary
slide-66
SLIDE 66

Key questions

17

slide-67
SLIDE 67

Key questions

  • Can we develop a methodology to detect sellers

who are adopting algorithmic pricing?

17

slide-68
SLIDE 68

Key questions

  • Can we develop a methodology to detect sellers

who are adopting algorithmic pricing?

  • Limited information, such as, price, rank, etc.

17

slide-69
SLIDE 69

Key questions

  • Can we develop a methodology to detect sellers

who are adopting algorithmic pricing?

  • Limited information, such as, price, rank, etc.
  • Non-trivial because there is no ground truth.

Therefore, we locate sellers behaving like “bots”

17

slide-70
SLIDE 70

Key questions

  • Can we develop a methodology to detect sellers

who are adopting algorithmic pricing?

  • Limited information, such as, price, rank, etc.
  • Non-trivial because there is no ground truth.

Therefore, we locate sellers behaving like “bots”

  • What are the different behaviors between the algo

sellers and non-algo sellers?

17

slide-71
SLIDE 71

Algorithmic pricing detection

18

slide-72
SLIDE 72

Algorithmic pricing detection

  • How can algo sellers stay competitive?

18

slide-73
SLIDE 73

Algorithmic pricing detection

  • How can algo sellers stay competitive?
  • Matching to a “competitive” price

18

slide-74
SLIDE 74

Algorithmic pricing detection

  • How can algo sellers stay competitive?
  • Matching to a “competitive” price
  • Target price time series

18

slide-75
SLIDE 75

Algorithmic pricing detection

  • How can algo sellers stay competitive?
  • Matching to a “competitive” price
  • Target price time series
  • Lowest price

18

slide-76
SLIDE 76

Algorithmic pricing detection

  • How can algo sellers stay competitive?
  • Matching to a “competitive” price
  • Target price time series
  • Lowest price
  • Second lowest price

18

slide-77
SLIDE 77

Algorithmic pricing detection

  • How can algo sellers stay competitive?
  • Matching to a “competitive” price
  • Target price time series
  • Lowest price
  • Second lowest price
  • Amazon price

18

slide-78
SLIDE 78

Algorithmic pricing detection

Product Price $1 $1.1 $1.2 $1.3 $1.4 Timeline t1 t2 t3 t4 t5

Lowest Price Algo Seller Price Non-algo Seller Price

19

slide-79
SLIDE 79

Algorithmic pricing detection

Product Price $1 $1.1 $1.2 $1.3 $1.4 Timeline t1 t2 t3 t4 t5

Lowest Price Algo Seller Price Non-algo Seller Price

19

slide-80
SLIDE 80

Algorithmic pricing detection

Product Price $1 $1.1 $1.2 $1.3 $1.4 Timeline t1 t2 t3 t4 t5

Lowest Price Algo Seller Price Non-algo Seller Price

19

slide-81
SLIDE 81

Algorithmic pricing detection

Product Price $1 $1.1 $1.2 $1.3 $1.4 Timeline t1 t2 t3 t4 t5

Lowest Price Algo Seller Price Non-algo Seller Price

19

slide-82
SLIDE 82

Algorithmic pricing detection

20

slide-83
SLIDE 83

Algorithmic pricing detection

  • Detection criteria
  • Spearman’s rank correlation between seller and target price
  • Product must have more than 1 seller
  • Correlation coefficient >= 0.7
  • p-value <= 0.05
  • Number of price changes >= 20

20

slide-84
SLIDE 84

Algorithmic pricing detection

  • Detection criteria
  • Spearman’s rank correlation between seller and target price
  • Product must have more than 1 seller
  • Correlation coefficient >= 0.7
  • p-value <= 0.05
  • Number of price changes >= 20
  • Discovered 543 sellers covering 513 products
  • 2.4% of total sellers
  • 31% of the bestselling products

20

slide-85
SLIDE 85

Algorithmic pricing detection

21

slide-86
SLIDE 86

Algorithmic pricing detection

$10 $12 $14 $16 $18 $20 11/22 11/24 11/26 11/28 11/30 12/02 12/04 12/06 12/08 12/10 Price Timeline (Year 2014) Algo Seller Seller1 Seller 2

21

slide-87
SLIDE 87

Algorithmic pricing detection

22

slide-88
SLIDE 88

Algorithmic pricing detection

$7 $8 $9 $10 10/30 10/31 11/01 11/02 11/03 11/04 11/05 11/06 11/07 11/08 11/09 Price Timeline (Year 2014) Algo Amazon Seller 1 Seller 2 Seller 3

22

slide-89
SLIDE 89

Behaviors of algo-sellers

23

slide-90
SLIDE 90

Behaviors of algo-sellers

  • Number of products sold by algo/non-algo sellers

23

slide-91
SLIDE 91

Behaviors of algo-sellers

  • Number of products sold by algo/non-algo sellers

CDF 20 40 60 80 100 Number of products 100 1000 10000

Algo sellers Non-algo sellers

23

slide-92
SLIDE 92

Behaviors of algo-sellers

  • Number of products sold by algo/non-algo sellers

CDF 20 40 60 80 100 Number of products 100 1000 10000

Algo sellers Non-algo sellers

23

slide-93
SLIDE 93

Behaviors of algo-sellers

24

slide-94
SLIDE 94

Behaviors of algo-sellers

  • Feedbacks received for algo/non-algo sellers

24

slide-95
SLIDE 95

Behaviors of algo-sellers

  • Feedbacks received for algo/non-algo sellers

CDF 20 40 60 80 100 Number of feedbacks 1 100 10000 1000000

Algo seller Non-algo seller

24

slide-96
SLIDE 96

Behaviors of algo-sellers

  • Feedbacks received for algo/non-algo sellers

CDF 20 40 60 80 100 Number of feedbacks 1 100 10000 1000000

Algo seller Non-algo seller

24

slide-97
SLIDE 97

Feedback loops

25

slide-98
SLIDE 98

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

25

slide-99
SLIDE 99

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

25

slide-100
SLIDE 100

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

25

slide-101
SLIDE 101

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

  • Feedback loops:

25

slide-102
SLIDE 102

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

  • Feedback loops:

25

Feedback

slide-103
SLIDE 103

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

  • Feedback loops:

25

Feedback Buy Box

Wi more wins

slide-104
SLIDE 104

Feedback loops

  • Algo sellers sell less types of products but receive

much more feedbacks than non-algo sellers

  • Recall the feature weights

Weight 0.2 0.4 Feature P r i c e D i f f P r i c e R a t i

  • F

e e d b a c k # F e e d b a c k A v g r a t i n g I s F B A ?

  • Feedback loops:

25

Feedback Buy Box

Wi more wins Wi more sales

slide-105
SLIDE 105

Behaviors of algo-sellers

  • Probability of winning the Buy Box for algo/non-algo sellers

26

Probability of Winning the Buy Box 14 28 42 56 70 Rank 1 6 11 15 20

Algo seller Non-algo seller

20 15 10 5 1

slide-106
SLIDE 106

Behaviors of algo-sellers

  • Probability of winning the Buy Box for algo/non-algo sellers

26

Probability of Winning the Buy Box 14 28 42 56 70 Rank 1 6 11 15 20

Algo seller Non-algo seller

20 15 10 5 1

slide-107
SLIDE 107

Summary

27

slide-108
SLIDE 108

Summary

  • We tracked the price and sellers of best selling

products on Amazon for 5 months

27

slide-109
SLIDE 109

Summary

  • We tracked the price and sellers of best selling

products on Amazon for 5 months

  • We developed a methodology to detect algo sellers

27

slide-110
SLIDE 110

Summary

  • We tracked the price and sellers of best selling

products on Amazon for 5 months

  • We developed a methodology to detect algo sellers
  • We uncovered their strategies for winning the Buy

Box, and potentially creating the feedback loops

27

slide-111
SLIDE 111

Thanks!

  • Questions?