180 likes | 348 Views
Supply Chain Management: Market-Share. Matthew J. Olson CIS 690 Research Project Dr. Hsu. Introduction. Background: What is TAC-SCM The Base Agent: MinneTAC Why I Studied Market-Share Methodology: Winter’s Method Experiment Results Agent Variations The Second Experiment Results
E N D
Supply Chain Management: Market-Share Matthew J. Olson CIS 690 Research Project Dr. Hsu
Introduction • Background: What is TAC-SCM • The Base Agent: MinneTAC • Why I Studied Market-Share • Methodology: Winter’s Method • Experiment Results • Agent Variations • The Second Experiment Results • Conclusions
Background • The Trading Agent Competition – Supply Chain Management (TAC-SCM) • Each game has 6 autonomous agents buying computer parts, building computers, and selling them to customers. • A game is 220 days in length. • Each agent starts with a $0 balance and can take loans. • Interest on loans are charged. • Interest on positive balance is given. • A cost is associated with storage. • The winner is the one with the biggest bank balance at the end of the game.
MinneTAC • Minnesota’s TAC-SCM agent • Code was given under the open source license. • After coming in 25th in the 2004 competition, they made some improvements and released this version. • One documented short-coming was market-share.
Why I Studied Market-Share • Market-share was assumed at 1/6. (MinneTAC was one of six agents playing in a game.) • In the real world, market-share isn’t so quaint. • In the game, it isn’t a bad starting estimate, but as time goes on… • MinneKatTAC
Methodology: Winter’s Method • Level: Ei=U(Ei-1+Ti-1)+(1-U)Yi • Trend: Ti=VTi-1+(1-V)(Ei-Ei-1) • U & V derived from experimentation. • Y is the observed value (in our case, (our orders)/(total orders) )
double marketShare=1.0/6.0; double marketShareOld; double level=1.0/6.0; double levelOld; double u=0.995; double v=0.995; double minMarketShare=0.05; double maxMarketShare=0.5; CustomerOrderList orderList = repository.getCustomerOrders(); for(int i=0;i<catalog.size();i++){ totalOrders+= repository.getProductsOrdered(i); } ourOrders=(double)orderList.size(); if(totalOrders>0.0) { double orderRatio=ourOrders/totalOrders; marketShareOld=marketShare; levelOld=level; level=u*(levelOld+marketShareOld)+(1-u)*orderRatio; marketShare=v*marketShareOld+(1-v)*(level-levelOld); if(marketShare<minMarketShare) marketShare=minMarketShare; if(marketShare>maxMarketShare) marketShare=maxMarketShare; } Methodology: Winter’s Method(cont.)
Agent Variations • Ran 5 versions and the baseline • 0 – previously described • 1 – averaged market-share over 2 days • 2 – averaged market-share over 3 days • 3 – averaged market-share over 4 days • 4 – halved market-share for 1st 22 days
Conclusions • The modified agent is beating the baseline a majority of the time! • This shows that market-share is an important factor. • I would stand behind this agent in a competition!
References • Berenson, Mark L., et al. Basic Business Statistics: Concepts and Apllications. Tenth Ed. 679. • Collins, John, et al. The Supply Chain Management Game for the 2006 Trading Agent Competition. November 2005. <http://www.sics.se/tac/tac06scmspec_v16.pdf>. • Collins, John, et al. MinnieTAC. University of Minnesota. <http://www.cs.umn.edu/tac/source.html>. • Collins, John, et al. Component-based Design for a Trading Agent. July 2004. <http://www.cs.umn.edu/tac/release/minnetac_design.pdf>.