Paper

E-Loyalty Networks in Online
Auctions
INBAL YAHAV
 WOLFGANG JANK
R.H. SMITH SCHOOL OF BUSINESS,
UNIVERSITY OF MARYLAND
Motivation
Bidders
Sellers
Actors
Objective
 High profit
 High conversion rate
 Get
Get the
the product
product
product (quality)
? Low price?
 Auction design (e.g., open
price, duration, etc.)
Means
 Feedback score
Lit
IS THAT ENOUGH??
 Trust
Research Questions
1. How to define and measure e-loyalty?
2. How does loyalty impact auction outcome
(price, conversion)?
3. What factors drive loyalty in online auctions?
Data
~350 Sellers
~700 Repeating Buyers
Loyalty in the Literature
 Definition: repeating purchases
 Brand-switch literature:
 Probability of switching to another brand
 Distribution of purchases across different brands (commonly 2
brands)
Research Questions
1. How to define and measure e-loyalty?
2. How does loyalty impact auction outcome
(price, conversion)?
3. What factors drive loyalty in online auctions?
Define and Measure eLoyalty
 Three steps measurements



Construct eLoyalty network
Transform network into loyalty distribution
Transform the distribution into quantifiers using PC analysis
Define and Measure eLoyalty
 eLoyalty networks
 Bipartite graph with:


First nodes set: sellers (red)
Second node set: buyers (white)
 Arcs: purchases, with the width
corresponding to the number of
interactions
Define and Measure eLoyalty
 eLoyalty disribution
100%
100%
70%
30%
80%
100%
Sellers
Buyers
1.
Measure proportion of interactions per buyer (~normalized distribution of out-degree)
2.
Measure the perceived loyalty per seller (~distribution of the weighted in-degree)
Define and Measure eLoyalty
Transform the distribution into two quantifiers (PC1, PC2) that measure
the difference between the sellers’ perceived loyalty.
PCA
Input
First & Second PCA
Scores (~80% of the
variation)
m
sellers
(discrete grid)
Sellers’ Perceived eLoyalty: PCAs
Very little weight on low scores , very large
weight on high scores (between 0.8 and 1
PC1 contrasts distributions of sellers with
extremely loyal bidders with those that are
little loyal
Most weight on medium-scores  PC2
contrasts the moderate loyalty distribution
from the extremes – distinguishes sellers
that have neither extremely loyal nor
extremely disloyal bidders
Research Questions
1. How to define and measure e-loyalty?
2. How does loyalty impact auction outcome
(price, conversion)?
3. What factors drive loyalty in online auctions?
Modeling eLoyalty : Effect of eLoyalty on Price
 OLS/ WLS regression
 High volume sellers have multiple, inter-dependent auctions
 Low-volume sellers have only few auctions
 Violates regression assumption
Modeling eLoyalty : Effect of eLoyalty on Price
 Random-effects regression model
 Account for seller-specific variation
 Heteroscedasticity
Modeling eLoyalty : Effect of eLoyalty on Price
 Segment sellers into three groups
Modeling eLoyalty : Effect of eLoyalty on Price
 Segment sellers into three groups: model fit
Low volume
R2=0.81
Medium volume
High volume
R2=0.77
R2=0.83
Effect of eLoyalty on Price
Coefficient
(Intercept)
StartPrice
AuctionDuration
log(ItemQuantity)
Bidcount
log(Pieces)
Size
log(SellerFeedback)
log(Volume)
PC1
PC2
Medium volume
Low volume
High volume
-0.21
0
4
0.08
0.04
0.05
0
0
0
0.12
0.14
0.08
0.11
0.13
0.07
0.19
0.08
0.36
0.03
0.03
0.07
0.04
0.04
0.12
-0.04
0.01
-0.81
0.21
-0.24
2.74
0.02
-1.72
-15.41
 The effect of loyalty depends strongly on size of the
seller:


High volume sellers can extract huge price-premiums from loyal
bidders
The impact of loyalty is much smaller for sellers of smaller scale
Summary
 Define and measure eLoyalty
 eLoyalty network
 Buyers loyalty ~ normalized distribution of out-degree
 Seller perceived loyalty ~ distribution of the weighted indegree
 Transform the distribution into quantifiers using PC analysis
 Modeling eLoyalty: data segmentation
 Conclusions
 Loyalty has higher impact on high volume sellers
 Saturated market
Discussion
 The analysis can be replicated to other products; the
results might change
 Temporal networks
 Examine the evaluation of eLoyalty
 We did not observe temporal effect in our data
More Information?
Inbal Yahav
[email protected]
http://www.rhsmith.umd.edu/faculty/phd/inbal/