290 likes | 383 Views
Determinants of Aggressive Bidding in the “Buying a Company” Task. Andy Lockett - School of Business, University of Nottingham Elke Renner - School of Economics, University of Nottingham Martin Sefton - School of Economics, University of Nottingham
E N D
Determinants of Aggressive Bidding in the “Buying a Company” Task Andy Lockett - School of Business, University of Nottingham Elke Renner - School of Economics, University of Nottingham Martin Sefton - School of Economics, University of Nottingham Deniz Ucbasaran - School of Business, University of Nottingham
Winner’s Curse In general, when bidding for an object of unknown value the person who most overestimates true value most likely to be winner … Winning bidder often makes expected losses • Auctions (e.g. Oil field drilling rights) • Acquisitions • Start-ups
Buying a Company(Samuelson and Bazerman, 1985) Company worth v to current owner and 3v/2 to potential buyer v uniformly distributed on [0, 100], observed by owner Potential buyer makes bid, b Current owner accepts • Owner receives b - v • Buyer receives 3v/2 - b or rejects • Owner receives 0 • Buyer receives 0
Optimal bid Owner: accepts if bv, rejects otherwise Buyer: expected profit = E[3v/2 – b|b v] Pr{b v} Optimal bid: b* = 0
Experimental Evidence Samuelson and Bazerman (Research in Experimental Economics, 1985). Subjects generally bid between E(v) = 50 and E(3v/2) = 75. Ball, Bazerman and Carroll (Organizational Behavior and Human Decision Processes, 1991). Learning from own outcome over multiple trials does not reduce winner’s curse. Holt and Sherman (American Economic Review, 1994). Vary distribution of v and find data consistent with a model of naiive bidding: maximize E[3v/2 – b] Pr{b v}. Charness and Levin (2005). Frame task as an individual decision problem, vary complexity of distribution of v, measure mathematical sophistication. Winner’s Curse survives, is lower with less complex distributions, lower for more mathematically sophisticated.
Experimental Evidence Selten, Abbink and Cox (Experimental Economics, 2005): Frame task as individual decision problem, vary lower bound of uniform distribution (u), focus on learning.
Goal of this experiment 1. Can winner’s curse be reduced by learning from others? • We compare bids made by individuals with bids made by pairs who can discuss problem • We study effect of giving bidders information about others’ outcomes 2. How are bids related to individual characteristics? • We elicit risk attitudes, measure mathematical sophistication, demographic data.
The Experiment 4 Sessions: • INDIVIDUAL: 27 subjects • TEAM: 25 x 2 subjects • INDIVIDUAL + INFO: 39 subjects • TEAM + INFO: 31 x 2 subjects 178 Subjects from University of Nottingham School of Business Each Session: • Buying a Company • Eliciting Risk Attitudes • Assessing Probabilities • Questionnaire Average earnings £ 17.55 for sessions lasting approx. one hour
Notes: This is one of the parameterizations used in Selten, Abbink, Cox (2005) We also had subjects make decisions for their other parameterizations: Task B (same as A but chips 11-99) Task C (same, but chips 21-99) Task A: optimal bid = 1 Task B: optimal bid = 22 Task C: optimal bid = 42 All three tasks played out at end of experiment
Notes: This is similar to risk elicitation procedure used in Dohmen, Falk, Huffman, Sunde, Schupp and Wagner (2005) Switching point gives measure of risk attitude Switching point = amount up to which subject chooses lottery (e.g. 2.5 means subject chose lottery when safe option is 2.5 or less, then chooses safe option when safe option 3 or more)
INDIVIDUAL TEAM 30 20 Frequency 10 0 0 .5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 .5 1 1.5 2 2.5 3 3.5 4 4.5 5 Switching Point No significant difference between INDIVIDUAL and TEAM (p=0.559) 68 risk averse decisions, 47 consistent with risk neutrality (2.5 or 3), 4 risk loving decisions, 3 inconsistent decisions (not shown)
Based on INDIVIDUAL data: Male (n = 49) average switchpoint = 2.21 Female (n = 17) average switchpoint = 1.74 Women more risk averse than men (Fisher exact test: p = 0.026)
Notes: This is same as used in Charness and Levin (2005) INDIVIDUAL TEAM 30 20 Frequency 10 0 0 1 2 3 4 0 1 2 3 4 # correct (part III) Significant difference between INDIVIDUAL and TEAM (p=0.007) Based on INDIVIDUAL data, males score higher than females (1 vs. 0.59) but difference marginally significant (p = 0.13)
Histograms of Bids: INDIVIDUAL (n=27) av = 40 s.d. = 26 Task A av = 45 s.d. = 21 Task B av = 57 s.d. = 20 Task C
Statistical Comparisons 1. Bids increase significantly across tasks as predicted: • bidA < bidB (signrank p =0.001) • bidB < bidC (p=0.000) 2. Significant over-bidding. Median bid significantly above predicted bid: • Median Bid A = 40 (signtest p=0.000) • Bid B = 45 (p=0.000) • Bid C = 58 (p=0.001)
Histograms of Bids: TEAM (n=25) Histograms of Bids: TEAM av = 37 s.d. = 20 Task A av = 44 s.d. = 20 Task B av = 53 s.d. = 21 Task C
Statistical Comparisons 1. Bids increase significantly across tasks as predicted: • bidA < bidB (signrank p =0.000) • bidB < bidC (p=0.000) 2. Significant over-bidding. Median bid significantly above predicted bid: • Median Bid A = 37 (signtest p=0.000) • Bid B = 48 (p=0.001) • Bid C = 53 (p=0.011) 3. No significant differences between INDIVIDUAL and TEAM: • Bid A: INDIVIDUAL = TEAM (ranksum p=0.8185) • Bid B: INDIVIDUAL = TEAM (p=0.9051) • Bid C: INDIVIDUAL = TEAM (p=0.3540)
+ INFO Sessions Subjects given selective information from earlier sessions: • INDIVIDUAL + INFO: 19 subjects given negative information, 20 given positive information • TEAM + INFO: 15 teams given negative information, 16 given positive information
Statistical Comparisons INDIVIDUAL: No statistical effect: • bidA (Kruskal-Wallis p =0.590), bidB (p = 0.491), bidC (p=0.274) TEAM: only bid C significantly differs across information conditions: • bidA (p = 0.264), bidB (p = 0.346), bidC (p = 0.038) • bid C: negative info leads to higher bid than no info
Preliminary Conclusions • Robust Winner’s Curse – individuals systematically overbid • Two Heads not Better than One – teams make statistically indistinguishable bids • No Information effect – individuals not affected by pos/neg info • Risk attitudes have small and insignificant effect on bids • Mathematical sophistication has small and insignificant effect on bids • Females bid more aggressively