Download - Heterogeneity and the Winner’s Curse
Heterogeneity and the Winner’s Curse
Mike Huwyler
What is the Winner’s Curse?
• Win the auction, but overpay relative to true value
• Three assumptions:– Imperfect information scenario– Common values setting– Impact of increasing bidder count
Contemporary Examples of the Winner’s Curse
• Professional sports (free agency)
• Initial public offerings– Google example
Online Setting
• Is it still an imperfect information scenario?– Feedback, product reviews, other listings
• Diverse range of participants – Income and experience
• Increased number of participants
Literature Review
• Uncertainty– Product misrepresentation (Jin and Kato, 2002)
• Pictures (Hou et. al, 2009)• Product quality (Adams et. al, 2011)• Timing strategies– “Sniping” (Easley and Wood, 2005)
• Secret reserve prices (Bajari and Hortascu, 2003)
Dataset
• 6,000 eBay auctions
• Bidder, auction, seller, and product characteristics
• Corvettes (all different models)– Most popular car sold on eBay
Tests
• Divide dataset into experience and income groupings
• Primary test– Relationship b/w bid amount and bidder count
• Secondary tests– Relationship b/w bid amount and individual and
product characteristics
Hypotheses
• Goal: Determine how different individuals respond to the winner’s curse– Do bidders optimally respond to an increase in the
number of bidders?• Hypotheses: – High income and high experience bidders should
respond optimally– Secondary test results will be mixed (horizontal vs.
vertical characteristics)
Regression Model
• For three experience and three income groupings (low, medium, and high):– Y1 = β0 + β1x1 + β2x2 + β3x3 + β4x4 + β5x5 + β6x6 + ϵ1
• Dependent variable = bid amount• Independent variables = number of bidders,
bidder income/experience, seller feedback, vehicle mileage, vehicle condition (dummy), vehicle transmission (dummy)
Experience Model Results
• Hypothesis partially supported• Negative, statistically significant relationship
b/w bid amount and number of bidders for ALL experience groups
• Secondary tests mixed– Universal response to mileage, condition– Seller feedback more important to high
experienced bidders
Income Model Results
• Hypothesis fully supported• Negative, statistically significant relationship
b/w bid amount and number of bidders for high income; Positive, insignificant for low income
• Secondary tests remain mixed
Adjustment #1
• Low R-squared values– Addition of three new variables: year, color
(dummy), and model (dummy)– Adjusted R-squared increased– Same Results
Adjustment #2
• Switch number of bidders to auction length– Proxy for the expected number of bidders
• Results support experience hypothesis, conflict with previous income findings– Negative relationship b/w bid amount and auction
length for medium and high experienced bidders, positive relationship for low experienced
– Income models scrapped
Real World Applications
• Can bidders improve their situation?– Education– Personality– Third Parties
Future Adjustments
• As results indicate, model is far from perfect• Future adjustments would include:– Interaction model– More accurate way to represent expected number
of bidders – Examine different products– New bidder variables (education)– Split dataset into quantiles, not by standard
deviation