Your SlideShare is downloading. ×
Euro2012presentation
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Euro2012presentation

46
views

Published on

Yusuke Goto (iwate Pref. Univ.) and Shingo Takahashi (Waseda Univ.) …

Yusuke Goto (iwate Pref. Univ.) and Shingo Takahashi (Waseda Univ.)
Agent-Based Simulation Analysis of Performance Measurement Systems Considering Uncertainties of a Learning Model
The 25th European Conference on Operational Research
July 9, 2012 (Vilnius, Lithuania)

Published in: Technology, Business

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
46
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
1
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Agent-based Simulation Analysis of Performance Measurement Systems Considering Uncertainties of a Learning Model EURO XXV The 25th European Conference on Operational Research July 9, 2012 Vilnius, Lithuania Yusuke Goto* and Shingo Takahashi** *Iwate Prefectural University, Japan **Waseda University, Japan 1
  • 2. Introduction •Performance Measurement System (PMS) is a major tool of management control •PMS indirectly controls member s behavior by evaluation •We have little knowledge about which PMS design is effective in a given situation •Effect of PMS is highly contingent on the situation: - organizational culture - environmental factors - individual characteristics ... •Traditional positivistic approaches play only a limited role Simulation analysis is a promising approach 2
  • 3. Agent-based Simulation To simulate the effect of PMS... •We must capture individual behavior (micro level) and organizational performance (macro level) •We must develop a human behavioral model •We must consider the variety of individual value and attitude in the behavioral model Agent-based simulation (ABS): •Micro-macro link of the organization system is considered •Members in an organization are autonomous agents •Agents determine their own behavior by referring to their internal decision-making model 3
  • 4. Modeling agent s behavior Behavioral framework: •Agents have their own attitude for their actions •Their actions are based on their attitude •Their attitude has variety •They learn their attitude to improve their evaluation by the PMS agent s attitude PMS evaluation agent s behavior learn / change Uncertainties in agent s learning model •When do agents learn? Always? •Who learns? Everybody? •How do agents learn? Are there some noises? 4
  • 5. Purpose and Method Research questions: •Does agent s learning model have an affect on the effect of PMSs? •Is this affect contingent to PMS design? Research method: •Modeling the target organization: a sales organization •Simulation analysis: 8 different learning models 3 different PMS designs •Discussion •Summary 5
  • 6. Sales organization Sales organization •The sales division has 10 groups •Every group has 10 agents •Every group has 100 customers initially Agent (sales person) •capability: cp ( 0 cp 1) agent s sales capability is equal to the probability of sales success •attitude: •aggressiveness: ag {0,1,..., 7} •cooperativeness: co {0,1,..., 7} •innovativeness: in {0,1,..., 7} 6
  • 7. Flowchart Sales related activities by agents 1.Visit 2.Market cultivation 3.Education 4.Training Evaluation by the PMS Learning of agents attitude 7
  • 8. Sales related activities Agent s activities 1.Visit: •selling of a good to a customer •The probability of sales success is equal to agent s sales capability cp 2.Market cultivation: •seeking new customers for a good mi + + 3.Education of teammates: •increase the cp of all teammates by EO ( 0 EO 1) 4.Training: •increase sales person s cp Attitude •aggressiveness: ag •cooperativeness: co •innovativeness: in by ES ( 0 ES 1) Behavior •market cultivation •education of teammates •training 8
  • 9. PMS (Performance Measurement System) •PMS evaluates every agent based on his/her behavior and result •PMS intends to control agent s behavior and organizational performance by the evaluation for agents 1.Individual sales: 2.Group sales: (agent i belongs to group j) 3.Behavioral control: 9
  • 10. Learning model Basic algorithm •Learning as an imitation process based on Genetic Algorithm (GA) •Agents imitate another agent s attitude whose evaluation is higher •Sometimes agents changes the attitude randomly Parameters in a learning model 1.Who? Number of agents who learn their attitude 2.When? Agents learn their attitude until their evaluation th value meets their learning threshold 3.How? Probability of random change 10
  • 11. Simulation experiment 8 different learning models: = {100, 90} 1.Who?: Number of learners 2.When?: learning threshold 3.How? th = {1000, 1.1} = {0.005, 0.001} 3 different PMSs 1.Individual sales: 2.Group sales: 3.Behavioral control: •1,000 trials for each pattern •Performance to control: agent s sales 11
  • 12. Result (1/3): Individual Sales 30 25 20 15 Avg. # of sales 0 5 10 25 20 15 10 0 5 Avg. # of sales 30 35 PMS (individual sales): P1_1000_100_0.001 35 P1_1000_100_0.005 0 10 20 30 40 50 60 0 10 20 30 40 50 60 40 50 60 25 20 60 0 10 20 30 30 25 20 15 Avg. # of sales 20 15 0 5 10 10 25 30 35 P1_1.1_100_0.001 35 Cycle 5 Avg. # of sales 50 30 30 10 20 30 40 50 60 0 10 20 30 P1_1.1_90_0.005 P1_1.1_90_0.001 30 25 20 15 Avg. # of sales 20 15 0 5 10 0 5 10 25 30 35 Cycle 35 Cycle Avg. # of sales 40 15 Avg. # of sales 20 0 0 •Width of possible outcomes 60 5 10 Cycle •Bifurcation 50 0 0 P1_1.1_100_0.005 Model dependent trend: 40 10 30 25 20 15 Avg. # of sales 10 0 •Sales decline gradually •Sales vary considerably 60 P1_1000_90_0.001 5 General trend: 50 35 P1_1000_90_0.005 40 Cycle 35 Cycle 0 10 20 30 Cycle 40 50 60 0 10 20 30 Cycle 12
  • 13. Result (2/3): Group Sales 30 25 20 15 Avg. # of sales 0 5 10 25 20 15 10 0 5 Avg. # of sales 30 35 PMS (group sales): P11_1000_100_0.001 35 P11_1000_100_0.005 0 10 20 30 40 50 60 0 10 20 30 40 50 60 40 50 60 30 25 20 15 Avg. # of sales 20 30 40 50 60 0 10 20 30 30 25 20 15 Avg. # of sales 20 15 0 0 5 10 10 25 30 35 P11_1.1_100_0.001 35 Cycle P11_1.1_100_0.005 Avg. # of sales 60 5 10 Cycle 0 10 20 •Some declinations 30 40 50 60 0 10 20 30 P11_1.1_90_0.005 P11_1.1_90_0.001 30 25 20 15 Avg. # of sales 20 15 0 5 10 0 5 10 25 30 35 Cycle 35 Cycle Avg. # of sales 50 0 0 5 Model dependent trend: 40 10 30 25 20 15 Avg. # of sales 10 0 •Sales improve gradually •Sales variation 60 P11_1000_90_0.001 5 General trend: 50 35 P11_1000_90_0.005 40 Cycle 35 Cycle 0 10 20 30 Cycle 40 50 60 0 10 20 30 Cycle 13
  • 14. Result (3/3): Behavioral control 30 25 20 15 Avg. # of sales 0 5 10 25 20 15 10 0 5 Avg. # of sales 30 35 PMS (behavioral control): P9_1000_100_0.001 35 P9_1000_100_0.005 0 10 20 30 40 50 60 0 10 20 30 40 50 60 40 50 60 30 25 20 15 Avg. # of sales 20 30 40 50 60 0 10 20 30 30 25 20 15 Avg. # of sales 20 15 0 5 10 0 5 10 25 30 35 P9_1.1_100_0.001 35 Cycle P9_1.1_100_0.005 Avg. # of sales 60 5 10 Cycle 0 10 20 30 40 50 60 0 10 20 30 P9_1.1_90_0.005 P9_1.1_90_0.001 30 25 20 15 Avg. # of sales 20 15 0 5 10 0 5 10 25 30 35 Cycle 35 Cycle Avg. # of sales 50 0 0 and saturate •Sales converge 40 10 30 25 20 15 Avg. # of sales 10 0 •Sales improve gradually 60 P9_1000_90_0.001 5 General trend: 50 35 P9_1000_90_0.005 40 Cycle 35 Cycle 0 10 20 30 Cycle 40 50 60 0 10 20 30 Cycle 14
  • 15. Discussion Learning model effect 1.width of possible outcomes 2.bifurcation 3.declination • Learning models have an effect on organizational behavior • The learning models effect is complicated PMS effect 1.Individual sales: Different learning models generate different outcomes 2.Group sales: Different learning models generally generate same outcomes 3.Behavioral control: All learning models always generate same outcomes Effects of learning models are contingent on the PMSs 15
  • 16. Summary • We developed learning models of agents behavior under a PMS • 8 different learning models and 3 typical PMSs are simulated • Learning models have an effect on organizational behavior • Effect of learning models are complicated and contingent on the PMSs The uncertainties of agent s learning model have an impact on PMS s effects 16
  • 17. References Deguchi, H. (2009) Dawn of Agent-Based Social Systems Sciences. In Deguchi, H. and Kijima, K. (Eds.) Manifesto: Agent-based Social Systems Sciences. Keiso-Shobo (in Japanese). Goto, Y., S. Takahashi, and Y. Senoue (2009) Analysis of Performance Measurement System for Knowledge Sharing under Intraorganizational Competition. Journal of the Japan Society for Management Information 18(1): 15-49 (in Japanese). Hales, D., J. Rouchier, and B. Edmonds (2003) Model-to-Model Analysis. Journal of Artificial Societies and Social Simulations 6(4). North, M. and C. M. Macal (2007) Managing Business Complexity: Discovering Strategic Solutions with Agent-Based Modeling and Simulation. Oxford University Press. Otomasa, S. (2003) On Use of Performance Measurement Indices in Japanese Companies. Rokkodai-Ronshu Management Series 49(4): 19-54 (in Japanese). Richiardi, M., R. Leombruni, N. Saam, and M. Sonnessa (2006) A Common Protocol for Agent-Based Social Simulation. Journal of Artificial Societies and Social Simulation 9(1). 17

×