Fake Agents Play the Beer Game Eliminate the Bullwhip Effect and Whip the MBAs

Presentation Transcript

Counterfeit Agents Play the Beer Game Eliminate the Bullwhip Effect and Whip the MBAs Steven O. Kimbrough D.- J. Wu Fang Zhong FMEC, Philadelphia, June 2000; record: beergameslides.ppt

The MIT Beer Game Players Retailer, Wholesaler, Distributor and Manufacturer. Objective Minimize framework wide (chain) long-run normal cost. Data sharing: Mail. Request: Deterministic. Costs Holding cost: $1.00/case/week. Punishment cost: $2.00/case/week. Leadtime: 2 weeks physical postponement

Timing 1. New shipments conveyed. 2. Orders arrive. 3. Take care of requests in addition to overabundance. 4. Choose the amount to arrange. 5. Figure stock expenses.

Game Board …

The Bullwhip Effect Order fluctuation is opened up upstream in the inventory network. Industry illustrations (P&G, HP).

Observed Bullwhip impact from students amusement playing

Bullwhip Effect Example (P & G) Lee et al., 1997, Sloan Management Review

Analytic Results: Deterministic Demand Assumptions : Fixed lead time. Players act as a group. Maker has boundless limit. "1-1" arrangement is ideal - arrange whatever sum is requested from your client.

Analytic Results: Stochastic Demand (Chen, 1999, Management Science ) Additional presumptions: Only the Retailer acquires punishment cost. Request circulation is regular information. Settled data lead time. Diminishing holding costs upstream in the chain. Arrange up-to (base stock establishment) approach is ideal .

Agent-Based Approach Agents fill in as a group. No specialist has learning on request circulation. No data sharing among specialists. Operators learn through hereditary calculations. Settled or stochastic leadtime.

Research Questions Can the operators track the request? Could the operators dispense with the Bullwhip impact? Could the specialists find the ideal arrangements on the off chance that they exist? Could the specialists find sensibly great approaches under complex situations where investigative arrangements are not accessible?

Flowchart

Agents Coding Strategy Bit-string representation with altered length n . Furthest left piece speaks to the indication of " + " or " - ". The rest bits speak to the amount to arrange. Govern " x+1 " signifies "if request is x then request x+1 ". Lead look space is 2 n-1 – 1.

Experiment 1a: First Cup Environment: Deterministic request with settled leadtime. Alter the strategy of Wholesaler, Distributor and Manufacturer to be "1-1". Just the Retailer operator learns. Result: Retailer Agent discovers "1-1".

Experiment 1b All four Agents learn under the earth of analysis 1a. Über lead for the group. Each of the four specialists discover "1-1".

Result of Experiment 1b All four specialists can locate the ideal "1-1" strategy

Artificial Agents Whip the MBAs and Undergraduates in Playing the MIT Beer Game

Stability (Experiment 1b) Fix any three operators to be "1-1", and permit the fourth specialist to learn. The fourth operator minimizes its own long-run normal cost as opposed to the group cost. No specialist has any motivating force to stray once the others are playing "1-1". Along these lines "1-1" is evidently Nash.

Experiment 2: Second Cup Environment: Demand consistently conveyed between [0,15]. Settled lead time. Each of the four Agents settle on their own choices as in trial 1b. Specialists kill the Bullwhip impact. Specialists discover preferable strategies over "1-1".

Artificial specialists dispose of the Bullwhip impact.

Artificial specialists find a superior strategy than "1-1" when confronting stochastic request with punishment costs for all players.

Experiment 3: Third Cup Environment: Lead time consistently dispersed between [0,4]. The rest as in analysis 2. Operators discover preferable strategies over "1-1". No Bullwhip impact. The polices found by operators are Nash.

Artificial operators find preferred and stable strategies over "1-1" when confronting stochastic request and stochastic lead-time.

Artificial Agents can take out the Bullwhip impact when confronting stochastic request with stochastic leadtime .

Agents learning

The Columbia Beer Game Environment: Information lead time: (2, 2, 2, 0). Physical lead time: (2, 2, 2, 3). Beginning conditions set as Chen (1999). Operators locate the ideal approach: arrange whatever is requested with time move, i.e., Q 1 = D (t-1), Q i = Q i-1 (t – l i-1 ).

Ongoing Research: More Beer Value of data sharing. Coordination and participation. Haggling and arrangement. Elective learning components: Classifier frameworks.

Summary Agents are equipped for playing the Beer Game Track request. Take out the Bullwhip impact. Find the ideal arrangements if exist. Find great approaches under complex situations where scientific arrangements not accessible. Clever and light-footed production network. Multi-specialist undertaking displaying.

A structure for multi-specialist savvy endeavor demonstrating