Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
p. 1/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Prediction Markets for Business Forecasting – Results from the Lab and a
Case Study
Christian Horn, Björn Sven Ivens Otto-Friedrich-University of Bamberg, Department of Marketing and
Chair of Innovation Management, Germany
Alexander Brem University of Erlangen-Nuremberg, Germany
International Symposium on Forecasting,
Seoul, 6/25/2013
p. 2/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Designed for „the primary purpose of aggregating information, so that market prices forecast future events“ (Berg, Nelson and Rietz 2003)
They bring groups of informants together and let them trade contracts whose payoff depends on the outcome of uncertain future events (Luckner 2012)
Prediction Markets work (possibly) with experts (scientists, managers, customers) and non-experts (e.g. customers)
Theoretical background
Prediction Markets (PM):
p. 3/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Adidas & sporting goods characteristics
• Low to medium priced products • Short product life cycles (64 % of portfolio <4 seasons on the
market) • Extreme seasonal influence on sales • Aims for adidas:
• Production planning • Logistics planning • Price management • Sales margin forecasting
p. 4/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Market Share
• Market share splits of products of one design line (t-shirt, pants, poloshirt, hodded pullover)
• Regional split of Graphic Tee sales
• Market share portfolio of product Graphic Tee
Price
• Sales optimum price of product Track Top
• Impact of own price increase on sales
• Optimal price products: replica tee, replica jersey, authentic jersey
• Price change of Jeresey: sales change
Innovation/Timing
• Ideal product launch date for product Club Track Top
• Product Jersey‘s most important features 1
• Most important fesatures 2
• Specific product version for soccer league/CL
Motivation/Benchmark
• Euro/Dollar Exchange Rate
• Sports Results of soccer matches of Euro-Championship 2012
Virtual Stocks and Contracts in PMs
Prediction Questions
p. 5/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
What would be the optimum sales price for the products shown below? Please invest your virtual money.
Prediction Markets: Example
p. 6/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
p. 7/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Research design
The PM experiments consisted of three steps:
Methodology
Steps for Experiment
1. Participants received an explanation of the software (Video)
2. Participants could trade for the specified trading time
3. Participants fill out questionnaire
p. 8/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Find out differences of short-running markets (<2h) and long runnig scenarios (~3w) 1
Find differences of laboratory study and practical study
Identify questions/contracts applicable in a PM 3
2
Study aims
p. 9/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Research design of Prediction Markets (PM)
During the PM experiments: participants had the possibility to trade 12 to18 shares and to bet on
predictions explained above Participants were could win 60 amazon.com-giftcards from 10 € up to 100 €
and 50 adidas-shirts Trading groups were set up with identical stocks for longer trading and shorter
trading time. Each group didn’t trade two stocks of the possible ones. The experiment took place in Germany at the University of Erlangen-Nuremberg and the University of Bamberg in May/June 2012
13 experimental groups with 15 to 461 persons
Methodology
Research design of Prediction Markets (PM)
Benchmark for MAPE: sales of Q3+Q4 2012 and adidas management decisions
Continous Double Auction and Simple Bet trading menachnisms
p. 10/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Trading numbers (net) and sample size (net)
Groupname <1 hour transactions
~ 3 weeks transactions
participants Net
Transactions per Participant
1 Lab 1 400,00 14,00 28,6 2 Lab 2 488,00 20,00 24,4 3 Lab 3 391,00 19,00 20,6 4 Lab 4 677,00 17,00 39,8 5 Lab 5 447,00 20,00 22,4 6 Adidas Managers 1 167,00 10,00 16,7
7 Adidas Managers 2 87,00 7,00 12,4
8 B1:L 547,00 13,00 42,1
9 B2:L 868,00 20,00 43,4
10 B3:L 261,00 19,00 13,7
11 B4:L 754,00 17,00 44,4
12 B5:L 314,00 20,00 15,7 13 Public:L 13.159,00 451,00 29,2
Sum 2.657,00 15.903,00 647,00 24,6
p. 11/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Example for Case Study Group „Adidas Managers 1“ (15 Adidas-Managers“) <1hour Trading Time
Stock-Question Group Subject Given Range Final Stock Price Benchmark APE Price change of product: influence sales volume
Adidas Managers
Sales Volume -2,19 -
Track Top: best retail price for sales optimum
Adidas Managers
Sales Price 75,57 65 0,16
Product feature or not: impact on sales volume
Adidas Managers
Product Managem
ent 5 5 0,00
Optimum Product Price Adidas
Managers Innovation -
Optimum Product Price Adidas
Managers Innovation
Authentic Jersey (80€ - 150€) 123,91 110 0,13
Optimum Product Price Adidas
Managers Innovation Cotton Tee (30€ - 50€) 36,28 33 0,10
Optimum Product Price Adidas
Managers Innovation Replica (70€ - 100€) 80,38 80 0,00
Optimum Product Price Adidas
Managers Innovation Replica Tee (35€ - 60€) 43,91 48 0,09
Euro-Dollar-Rate Adidas
Managers Economic 1,26 -
vorläufige Ergebnisse
p. 12/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
MAPE results of PMs at end of trading
Groupname MAPE < 1 hour
transactions MAPE ~ 3 weeks
transactions % improvement participants Net
1 Lab 1 0,15 14,00
2 Lab 2 0,29 20,00
3 Lab 3 0,12 19,00
4 Lab 4 0,15 17,00
5 Lab 5 0,10 20,00
6 Adidas Managers 1 0,25 10,00
7 Adidas Managers 2 0,33 7,00 % improvement MAPE
Lab1-5/ adidas1-2 -0,45
8 B1:L 0,11 0,42 14,00
9 B2:L 0,12 1,43 20,00
10 B3:L 0,15 -0,21 19,00
11 B4:L 0,13 0,11 17,00
12 B5:L 0,12 -0,17 20,00
13 Public:L 0,10 451,00
% improvement MAPE Public/MAPE Longs - - 0,28 648,00
p. 13/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Conclusion
Improvement in MAPE for 3-week markets compared to <2h markets
Improvement for larger group
Experts (Adidas-employees) produced weakest results
Stocks perform differently well, some topics could not be solved
p. 14/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Implications
Non-experts perform slightly better than experts, but
differences are slight.
Large numbers of participants are not necessary
Short running prediction markets can be used for
accurate forecasting in several fields.
p. 15/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Limitations
motivation of Adidas-managers was lower (less trades)
sports products are low- to medium priced consumer
products
Mostly German participants
Only two expert groups
p. 16/16 Prediction Markets | Christian Horn | Otto-Friedrich University of Bamberg, Department of Marketing
Questions?
Christian Horn Department of Marketing Otto-Friedrich-University of Bamberg Tel. +49 951 / 863 2564 Mobile +49 176 / 20993285 Fax + 49 951 / 863 2566 Email: [email protected] Web: www.uni-bamberg.de/bwl-marketing