slide1 l.
Skip this Video
Loading SlideShow in 5 Seconds..
Leticia Saldain Guadalupe Segura PowerPoint Presentation
Download Presentation
Leticia Saldain Guadalupe Segura

Loading in 2 Seconds...

play fullscreen
1 / 64

Leticia Saldain Guadalupe Segura - PowerPoint PPT Presentation

  • Uploaded on

The design of successful on-line auctions. Leticia Saldain Guadalupe Segura. Outline . E-bay basics Reputation Mechanisms Low-valued items vs. high-valued items Pennies: US Cent and Indian Head Paul Reed Smith Guitar Last-minute bidding. E-Bay basics. On-line auctions started in 1995

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about 'Leticia Saldain Guadalupe Segura' - PamelaLan

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

The design of successfulon-line auctions

Leticia Saldain

Guadalupe Segura

  • E-bay basics
  • Reputation Mechanisms
  • Low-valued items vs. high-valued items
    • Pennies: US Cent and Indian Head
    • Paul Reed Smith Guitar
  • Last-minute bidding
e bay basics
E-Bay basics
  • On-line auctions started in 1995
  • 1998: e-bay had > 3 billion transactions
  • Growth rate > 10% per month
  • Over 3 million individual auctions / week
  • 7 million unique individuals visit site / month
  • Over 2000 unique categories
e bay auction
E-bay auction
  • Second price auction
  • Ascending-bid (English) format

- fixed time and date

  • Reservation price
  • Auction lasts 3-10 days
  • “proxy bidding” system (Vickrey auction)
  • Seller chooses
    • Opening bid amount
    • Secret “reserve price”
    • Length of auction
  • Sellers pays two fees: non-refundable insertion fee and final value fee
  • E-bay takes no risk
reputation mechanisms
Reputation Mechanisms

Analyzing the Economic Efficiency of eBay-like Online Reputation Reporting Mechanisms

By Chrysanthos Dellarocas

effects of reputation
Effects of reputation
  • Studied by Economists
  • Little attention to the mechanisms for forming/communicating reputation
  • computer scientists have focused on design/implementation of reputations systems
on line reputation reporting systems
On-line Reputation Reporting Systems

Goal: to induce good behavior in markets with asymmetric information

  • Feedback profile = reputation
  • Quality signal and control

Allows a market to exist!

market for lemons akerlof 1970
Market for Lemons (Akerlof 1970)

Example: consider 9 used cars

Quality levels: 0, ¼, ½, ¾, 1, 1 ¼, 1 ½, 1 ¾, 2

Assume cardinality (e.g. car with value 1 has twice the quality of car with value ½)


  • quality of car known to seller
  • buyer only knows the distribution of quality
  • seller: reserve value = 1000*q
  • buyer: reserve value = 1500*q
market for lemons cont
Market for Lemons (cont’)

cars sold in an auction:

  • Initial price $2000/car (all owners are willing to sell their car)
  • Buyers : average quality = 1, bid <= 1500

Auctioneer must reduce price to 1500:

  • at this price seller 8 and 9 (best two cars) will withdraw from market (why?)
  • average quality of remaining cars fall (q = ¾)
  • buyers are only willing to pay $1125 (1500* ¾ )

Auctioneer must try a lower price…and so on


  • When potential buyers only know the average quality of used cars, market prices will be lower than the true value of top-quality cars
  • Owners of top-quality cars will withhold cars from sale


  • Cars will not be sold even though potential buyers value the cars

more than current owners

Result from asymmetric information!

e bay marketplace
E-Bay Marketplace
  • Asymmetric information
  • Incentive for seller to over-estimate quality and increase profits
  • Need to provide information to buyers

Reputation Mechanisms allow market to exist by reducing asymmetric information!

model to analyze efficiency of binary rm
Model to analyze efficiency of binary RM


  • real quality (qr) is unknown to buyer
  • buyer prefers high quality to low quality
  • advertised quality (qa) is controlled by seller

Seller : Max payoff function

π(x, qr, qa) = G (x, qr, qa) – c(x, qr)

x ≡ volume of sale

G() ≡ gross revenue

c() ≡ cost

ebay reputation system
eBay Reputation System

Feedback profile: R= (∑(+), ∑(-), ∑(no ratings))

Buyer Utility: U = θ *q – p

p = price

q = level of quality after consumption

θ = buyer’s quality sensitivity

Buyer estimated quality qe = f( qa, R)

qa – advertised quality

R - Reputation

Buyer: max E(utility) = Ue = θ *qe – p

After purchase: buyer observes q = qr + ε

ebay reputation system15
eBay reputation system

Error term (ε) represents:

  • buyers misinterpreting qa
  • sellers may vary in actual q from one transition to another
  • buyers may have small difference in q, based on outside factors like weather
  • some aspects of q depends on factors outside seller’s control (i.e. post office delays)
ebay reputation system16
eBay reputation system

Buyer satisfaction: S = U – Ue = θ (qr – qe + ε), ε ~ N(0, σ)

Rating function r(S) : ‘+’ if S>0

‘-’ if S <= - λ

no rating if –λ <S<= 0

  • Ratings as a function of a buyer’s satisfaction relative to expectations
  • λ accounts for e-bay buyers giving few ‘-’ ratings to sellers

Possible Explanations:

    • Fear of reciprocal ratings
    • outside network communication between sellers and buyers
    • “culture of praise” : buyers feel a moral obligation to give “+” ratings
conditions for a well functioning reputation mechanism rm
Conditions for a “well-functioning” reputation mechanism (RM) :
  • If there exists an equilibrium of prices and qualities under perfect information (qe=qa=qr), then in markets where qr is private to sellers, the existence of a RM makes it optimal for sellers to settle down to steady-state pair of real and advertised qualities (qr, qa)
  • Assuming (1) holds, under all steady-state seller strategies (qr, qa) the quality of sellers as estimated by buyers before transactions take place must be equal to the true quality

(i.e. qe =qr)

-in competitive markets:

if qr > qe, then sellers would leave the market

if qr < qe, then buyers would leave the market

can they be well functioning
Can they be well-functioning?
  • If given a rating function whether condition 2 is satisfied depends on the relationship between this rating function and the quality estimation function
  • Seller’s find it optimal to settle down to steady-state advertised quality levels if buyers are lenient when rating seller’s profiles
estimated vs real qualities in steady state
Estimated vs. Real Qualities in Steady State
  • The focus is on binary reputation mechanisms satisfying condition 2
  • Denote an estimated quality function qe and the deception factor ξ
  • If ξ > 0, buyers will overestimate seller’s true quality
  • If ξ < 0, buyers will underestimate seller’s true quality
  • Let N be total no. of sale transactions [N= ∑(+) + ∑(-) + ∑(no ratings)]
  • Note: eBay does not specify quality assessment function f, it just publishes ∑(+) and ∑(-) allowing buyers (users) to use function they see fit. It also does not publish ∑(no ratings) thus N is not known.
estimated vs real qualities in steady state continued
Estimated vs. Real Qualities in Steady State Continued
  • We will explore whether binary reputation systems can be well functioning
  • Let ξ(R) be an estimate of seller’s deception factor based on information contained in the seller’s profile
  • A binary reputation system where buyers
    • Rate according to r(S)
    • Assess item quality according to (5)
    • Have reliable rule for calculating ξ(R) for a given seller satisfies condition 2
  • Aside: We would not expect any profit maximizing seller to under-advertise
estimated vs real qualities in steady state continued21
Estimated vs. Real Qualitiesin Steady State Continued
  • There are three ways buyers may use ∑(+),

∑(-), and N to estimate ξ(R)

    • Based on the number of positives
    • Based on the number of negatives
    • Based on the ratio between negatives and positives
estimate based on positives
Estimate based on positives
  • Require a fraction of positive ratings exceed a threshold, η^
  • We will use statistical hypothesis,
    • test null hypothesis Ho: η ≥ 0.5 given η^
    • We get new quality assessment function, qe
  • Method is appealing due to its relatively simplicity
  • Although, it is difficult without knowledge of N (Recall N is not specified by eBay)
  • Conclusion, method is rarely used by eBay users
estimate based on negatives
Estimate based on Negatives
  • This method is similar to previous, except we are know looking at fraction of negative ratings, ζ
  • Again using statistical testing
    • Null hypothesis, Ho’: ζ ≤ k* , where k* is a monotonically decreasing function of the leniency factor λ
    • New quality assessment function, qe
  • Conclusion, satisfaction of condition 2 is always possible. In order to find k* we need parameters λ, θ, and σ. However, this are not available to buyer’s in practice and the right k* is very important to the well-functioning of the mechanism. But parameters can be derived from ∑(+), ∑(-), and N.
  • Overall, this function is a rather fragile rule for assessing the seller’s quality efficiently
estimate based on negatives continued
Estimate based on Negatives Continued
  • Other things to consider:
    • What methods they use to compute threshold
    • Whether their trustworthiness thresholds do indeed come close to satisfying condition 2
  • These open questions invite to further explore empirical and experimental results to complement this paper
estimate based on ratio between negative and positives
Estimate based on ratio between negative and positives
  • We can just try to find a quality assessment function between positive and negative ratings, such a function will not exists
  • Let ρ(ξ) = ∑(-) / ∑(+), so this function is non-negative and monotonically increasing in ξ
  • Again using statistical testing
    • Null hypothesis, Ho”: ρ^ ≤ 2*Φ[-λ / (θσ)] where Φ() is the standard normal CDF
    • New quality assessment function qe
  • Once again we need knowledge of parametersλ, θ, and σ which is not known but can be substituted by N if buyers have knowledge of it.
  • Conclusion, this problem is just as difficult as estimation on negative ratings.
existence of steady state
Existence of Steady-State
  • We want to show function based on number of negative ratings to find quality assessment function are preferred to the one’s using just the number of positive ratings
  • To show let’s observe what would occur if sellers oscillate between good and bad quality
    • Let’s say in period 0 seller had N transactions with a good reputation, q*
    • If in period 1, she milks reputation earned during period 0 (in order to make a little more profit since item being sold this period is not as in good conditions)
    • Seller’s subsequent estimate quality will fall to zero. But in order to re-gain their good reputation, seller will have to reduce the ratio ∑(-) / N and the threshold k* (this action will occur when seller’s places a good item at a lower price)
existence of steady state continued
Existence of Steady-State Continued
  • Conclusions
    • A profit maximizing seller will oscillate if profit of ‘deceiving’ transaction exceeds the loss from ‘redeeming’ transaction both relative to steady-state profit
    • However, seller’s will require many more ‘redeeming’ transactions after a ‘deceiving’ one. Thus, if λ is sufficiently large sellers will find it optimal to settle down to steady-state real and advertised quality levels
reality checks
Reality Checks
  • Assumptions:
    • All buyers have the same quality sensitivity, θ and leniency factor, λ
    • Buyers always submit ratings when satisfaction is above 0 or falls below –λ

Not likely to hold in real market!

reality checks29
Reality Checks:
  • Some buyers never rate
  • Buyers differ in sensitivity and leniency
  • Relax both assumptions


Modify r(S) to account for some buyers never rating

  • Binary Reputation Systems can be well-functioning provided buyers find the right balance between leniency and quality assessment
  • Finding this balance when judging seller’s profiles is necessary for the well-function of the system, otherwise the resulting market outcome will be unfair
low valued vs high valued items
Low-Valued vs. High-Valued Items

Pennies from eBay: the Determinants of Price in Online Auctions

By: David Lucking-Reiley, Daniel Reeves and Doug Bryan/Naghi Prasad (2000 draft)

Valuing Information: Evidence from Guitar Auctions on eBay

By: David H. Eaton (2002)

u s cents
U.S. Cents
  • Collected data over 30-day period, July-Aug 1999
  • 20,292 observations (referred to as the large set)
  • Subset of these used, on auctions of “U.S. Indian Head Pennies”
    • 461 such auctions (referred to as the small set)
experiments conducted
Experiments Conducted
  • Experiments and regression analysis on three types of parameters:
    • Effect of positive and negative feedback
    • Effect of auction length
    • Effect of minimum bid and reservation prices
  • Result 1-
    • A 1% increase in positive feedback → 0.03% increase in auction price
    • Effect of 1% negative feedback → 0.11% decrease in auction price (this statistically significant at 5%)
  • Result 2-
    • Also found length of auction positively influenced price, longer auctions higher prices
    • 3 & 5-day auctions almost had same prices, but 7-day auctions increased by 24% while 10-day auctions increased by 42% (both statistically significant)
  • Result 3-
    • The presence of reserve prices increased price by 15%
    • as minimum price bid increases by 1% final price increases by less than 0.01%
low valued vs high valued items35
Low-Valued vs. High-Valued Items

Pennies from eBay: the Determinants of Price in Online Auctions

By: David Lucking-Reiley, Daniel Reeves and Doug Bryan/Naghi Prasad (2000 draft)

Valuing Information: Evidence from Guitar Auctions on eBay

By: David H. Eaton (2002)

paul reed smith guitar auctions
Paul Reed Smith Guitar Auctions
  • High valued item (price > $1000)
  • Some knowledge of item based on reputation of original product

( i.e. manufacturer reputation)

  • Information signals:
    • feedback profile
    • availability of escrow services/ fraud protection
    • pictures
prs guitar auctions
PRS Guitar Auctions


  • auctions between January – April 2001
  • four model classes
  • 325 successful auctions
prs guitar auctions38
PRS Guitar Auctions

Empirical Results:

  • pictures attract more bidding action and increase final bid price

(added value $60-232 / bid)

  • Use of escrow services send a negative signal to buyers
  • Negative feedback:

- decreases the likelihood of sale

- increases the final bid price for item

(added value ~ $504)

last minute bidding
Last-minute bidding

Last-Minute Bidding and the Rules for Ending Second-Price Auctions: Evidence from eBay and Amazon Auctions on the Internet

By Alvin E. Roth and Axel Ockenfels

late bidding
Late bidding

Rules for ending auction:

  • E-bay: fixed end-time
  • Amazon: automatic 10 minute extension on end time whenever bidding continues
late bidding41
Late bidding

Observations: the effect of experience in late bidding

  • More late bidding on e-bay than on Amazon
  • e-bay: experienced bidders submit late bids more often than less experienced bidders (opposite for Amazon)
  • e-bay: more late bidding for antiques than for computers
last minute bidding sniping
Last-minute bidding (“sniping”)
  • E-bay advices buyers to bid early (i.e. proxy bidding)
  • Late bids: risk of not being successfully transmitted
  • lower expected revenues for sellers

“sniping” is a best response to e-Bay fixed deadline!

last minute bidding43
Last-minute bidding
  • Theorem 1:

A bidder in continuous-time second-price private value auction doesn’t have any dominant strategies



m = min initial bid

s – smallest increment possible

Vj – willingness to pay for bidder j ~ F

Consider two bidders : i, j

Show: bidder j with value Vj > m+s has no dominant strategy to every strategy of bidder i

proof cont
Proof (cont’)

Case 1:

player i strategy:

bid m at t=0, not bid if she remains the highest bidder

bid B (with B> Vj+s) whenever she is not the highest bidder

player j best response:

Not bid at any time t<1

Bid Vj at t=1 (end of auction)

Payoff to player j = p*(Vj – m- s)>0, p=probability bid is transmitted

Case 2:

Player i strategy: not bid at any time

If player j uses her previous strategy: E[payoff]= p*(Vj – m) < Vj –m

Player j has no dominant strategies!

last minute bidding46
Last-minute bidding
  • Recall Theorem 1:

A bidder in continuous-time second-price private value auction doesn’t have any dominant strategies

last minute bidding47
Last-minute bidding
  • Theorem 2:

There can exist equilibria in which bidders do not bid their true values until last moment (t=1), at which time there is only a probability p (p<1) that a bid will be transmitted

  • A mutual delay until the last-minute of the auction can raise the E [profit] of all bidders because of the positive probability that another bidder’s last-minute bid will not be successfully transmitted
  • At this equilibrium, E [bidder profits]> than at equilibrium at which each player bids his true values early
strategic reasons for late bidding
Strategic Reasons for Late Bidding
  • To avoid “bidding wars” with incremental bidders
  • To avoid “bidding wars” with other like-minded bidders
  • To protect information
non strategic reasons for late bidding
Non-strategic Reasons for Late Bidding
  • Procrastination
  • To retain flexibility to bid in other auctions (same item)
comparisons of ebay amazon
Comparisons of eBay & Amazon
  • Most differences come from the different auction rules
  • Data Description
    • Both make data publicly available
    • Downloaded data in two categories
      • Computers: retail price of most items are usually available
      • Antiques: retail prices are not usually known and value is in most cases ambiguous
comparisons of ebay amazon continued
Comparisons of eBay & Amazon Continued
  • Data set
    • Auctions were randomly selected during a certain time period
    • Criteria
      • Two or more bidders
      • Auctions with reserve price were selected if it was met
  • Selected 480 auctions with 2279 bidders
    • 120 eBay and 120 Amazon Computer Auctions
    • 120 eBay and 120 Amazon Antiques Auctions
comparisons of ebay amazon continued53
Comparisons of eBay & Amazon Continued
  • Timing
    • Data on seconds last bid submitted by each bidder before auction closed (if bid was within last 12 hours)
    • For Amazon computed ‘hypothetical’ deadline
  • Feedback Number
    • For eBay as explained before
    • For Amazon, users (both buyers and sellers) place a 1-5 rating. The sum of these ratings is the ‘feedback number’ in Amazon
  • Included a survey consisting of 8 questions
    • Targeted at buyers who have been successful last moment bidders
  • Questions:
    • Do you plan early on to be a late bidder? Why?
    • Bid by hand or bidder software?
    • What % of your late bids were not submitted due to auction closing? Due to something else coming up?
    • On average number of bids per auction?
    • Idea of max you are willing to pay for an item, early on
    • What % of time do you wish you had bid higher (when not highest bidder)?
survey results
Survey Results
  • Conclusion, most often late bidding is part of a planned strategy even knowing of late bidding risks.
    • 91% confirm late bidding is planned early (Q1, N=65)
    • 65% say its to avoid ‘bidding wars’ or to keep prices down (some experienced Antique-bidders use it to avoid sharing valuable info) (Q1, N=49)
    • 88% know early on what they are willing to pay (Q6, N=65)
survey results continued
Survey Results Continued
  • Amateurs late bidding due to confusing eBay with English auction (< 10%). Also some do feel regret for not bidding higher. (Q7)
  • Most bidders, 93%, do not use sniping software but have many windows open to improve late bidding performance. (Q2, N=67)
  • 86% say at least once they were not able to make bid (Q3, N=65) and 90% say sometimes something else comes up (Q4, N=63)
conclusions from prior experiment
Conclusions from Prior Experiment
  • Many causes for multiple and late bidding
    • Differences in the number of late bids on eBay and Amazon is evidence that rational strategic considerations play a significant role
    • Additional differences between categories suggests bidders respond to strategic incentives for late bidding in markets with unknown values
    • The large number of late bidding on Amazon also shows non-strategic causes for late bidding
further research
Further Research
  • Differences in auction outcomes due to negative feedback related to seller characteristics as compared to negative feedback related to product characteristics
  • Analyze impact of internet-based companies that provide price information to auction participants
u q reputation systems
U-Q: Reputation Systems
  • Are Reputation Mechanisms truly reliable?
  • How should users use information provided for their decision-making?
  • Do they promote efficient market outcomes?
  • To what extent are they manipulated by strategic users?
  • What is the best way to design reputation systems?
u q rules for ending auctions
U-Q: Rules for Ending Auctions
  • Does a fixed-deadline auction of a private value good raise less revenue than one with the same number of bidders in automatic extended deadline auction?
  • How about for a public value good?
  • Could the increased entertainment value of a fixed

deadline attract sufficiently many bidders to overcome


  • Dellarocas, Chrysanthos, Analying the Economic Efficiency of eBay-like Online Reputaion Reporting Mechanisms, MIT, 2001
  • Eaton, David H., Valuing Information: Evidence from Guitar auctions on eBay, Murray State, 2002
  • Lucking-Reiley, David, Bryan Doug, and Reeves, Daniel, Pennies from eBay: The Determinants of Price in Online Auctions, Vanderbilt, 2000
  • Melnik, Mikhail I. and Alm, James, Does a Seller’s eCommerce Reputation Matter? Evidence from eBay Auctions, Georgia State
  • Roth, Alvin and Ockenfels, Axels, Last-Minute bidding and the Rules for Ending Second-Price Auctions: Evidence from eBay and Amazon Auctions, Harvard, 1999
  • Wilcox, Ronald T., Experts and Amateurs: The Role of Experience in Internet Auctions, Carnegie Mellon, 2000