Analysis of Inventory Models with Limited Demand I f Information ti David Simchi-Levi Sources z The Bullwhip Effect )Drezner, Ryan, Simchi-Levi, “Quantifying the Bullwhip Effect: The Impact of Forecasting, Lead Times and Information Information.” )Chen, Ryan, Simchi-Levi, “The Impact of Exponential Smoothing Forecasts on the Bullwhip Effect.” z Minimax Inventory Models )Gallego, Ryan, Simchi-Levi, “Minimax Analysis for Finite Horizon Inventory Models Models.” Quantifying the B ll hi Effect Bullwhip Eff t z The Impact of Forecasting Methods )Moving Average Exponential Smoothing )Exponential z Multi-Stage Supply Chains )Centralized Information )Decentralized Information A Si Simple l Supply S l Ch Chain i z Single retailer, single manufacturer. )Retailer observes customer demand, Dt. )Retailer Retailer orders qt from manufacturer. )Lead time + 1 = L. Dt Retailer qt L Manufacturer A Simple Supply Chain: O d off Event Order E ts z At end of period t: )Retailer updates forecast based on Dt. Calculates order order-up-to up to point, yt+1. )Calculates )Places order qt+1. )O d arrives )Order i att start t t off period i d t+L. t+L )Demand is observed and filled. )Unfilled demand is backlogged. A Simple Supply Chain: Policy I Inventory t li z Retailer follows order-up-to policy based on inventory position. )Approximates Approximates the optimal policy under the assumption of normal demand. )A policy used frequently in practice practice. y t = L μ$ t + z LS t A Simple Supply Chain: Average Forecast M i A Moving t z Mean and standard deviation of demand are estimated using a moving g avera ge of p observations: p ∑ Dt−i $μ t = i=1 p p St = ∑ ( Dt−i − μ$ t ) i=1 p −1 2 Moving Average Forecasting: Quantity Od Q Order tit Excess demand is returned at no cost. z The order quantity for period t is: z qt = yt − yt−1 + Dt−1 L⎞ ⎛ ⎛ L ⎞ ⎞ = ⎜ 1+ ⎟ Dt−1 − ⎜ ⎟ Dt− p−1 + z L( S t − S t−1 ) ⎝ ⎝ p⎠ p⎠ Moving Average Forecasting: Variability Th V The i bilit off Orders Od z Determine the variance of q relative to the variance of demand, σ2: ⎛ 2L 2L ⎞ 2 2 Var( qt ) = ⎜ 1+ + 2 ⎟ σ + z LVar( St − St −1 ) ⎝ p p ⎠ 2 ⎛ 2L⎞ +2z L⎜ 1+ ⎟ Cov( Dt−1 ,St ). ⎝ p⎠ Moving Average Forecasting: Symmettriic Demand d Lemma: Let Di, i=1,...,p, be i.i.d. observations from a symmetric distrib bution with h variance σ2. Then Cov(Di,S)=0. Corollary: The increase in variability from the retailer to the manufacturer is: 2 Var (qt ) 2L 2L ≥1+ + 2 . 2 p p σ Moving Average Forecasting: N Normal lD Demand d Lemma: Let Di, i=1,...,p, be i.i.d. observations from a normall distrib d bution with h variance σ2. Var ( S t − S t −1 ) σ 2 ≥ 1 p 2 . Corollary: The increase in variability from the retailer to the manufacturer is: is: Var (qt ) σ2 2L 2L2 + z 2 L ≥1+ + . 2 p p 2 Var(q)/σ : L B d vs. Si l ti Lower Bound Simulation 3 25 2.5 2 1.5 1 0.5 0 0 10 20 Bound p 30 Simulation 40 2 Var(q)/σ : F V i L d Ti For Various Lead Times 14 L=5 12 10 L=3 8 6 L=1 L=1 4 2 0 0 5 10 15 20 25 30 The Bullwhip Effect: Manageriiall Insi I ights ht z z Exists, in part, due to the retailer’s need to estimate the mean and variance of demand demand. The increase in variability is an increasing function of the lead time and the smoothing parameter. )With With longer lead time need more demand data to reduce the bullwhip effect. M lti Stage Supplly Chai Multi-St Ch ins Consider an N stage supply chain: )Stage i places order qi to stage i+1. Li is lead time between stage i and ii+1. 1. )L qo=D Retailer Stage 1 q1 L1 Manufacturer Stage 2 q2 L2 Supplier Stage 3 Multi-Stage Supply Chain: C t lized Centrali d IInfformation ti At the end of period t-1, stage i: )Receives order qii-11 t. )Receives updated forecast from retailer. )Calculates the order-up-to point, yit. )Orders qit. Multi-Stage Supply Chains: C t lized Centrali d IInfformation ti Each stage in the supply chain uses: )The estimate of lead time demand received from the retailer. )An order-up-to inventory policy: y = Li μ$ t i t where μ$ t is received from the retailer. Multi-Stage Supply Chains: C t lized Centrali d IInfformation ti Lemma: When the retailer uses a moving average with p observations, observations the increase in variability at at stage k is: ⎡∑ L ⎤ 2 ⎡∑ L ⎤ 2 Var ( q ) ⎢⎣i=1 i ⎥⎦ ⎢⎣i=1 i ⎥⎦ =1+ + 2 σ p p k k k 2 Multi-Stage Supply Chains: D Decentralized t li d IInformation f ti z The retailer does not provide upstream stages with customer demand data data. )Stage i estimates the mean demand from the orders it receives from stage ii-1 -1, qti-1, i > 1 1. p −1 $ = μ i t ∑q j = 0 i −1 t − j p i = Li t y μ$ i t Multi-Stage Supply Chains: D Decentralized t li d IInformation f ti Lemma: When the retailer uses a moving average with i h p observations, b i the h increase i in i variability i bili at stage k is: Var ( q σ 2 k ) ≥ ∏ ⎡1 + 2 L k i=1 ⎢ ⎣ 2 Li ⎤ + 2 ⎥ p p ⎦ i 2 The Bullwhip Effect: Manageriiall Insi I ights ht z z z z Exists, in part, due to the retailer’s need to estimate the mean and variance of demand demand. The increase in variability is an increasing function of the lead time and the smoothing g parameter. )With longer lead time need more demand data to reduce the bullwhip effect. The more complicated the demand models and the forecasting techniques, the greater the increase. Centralized demand information can reduce the bullwhip effect, but will not eliminate it. MIT OpenCourseWare http://ocw.mit.edu ESD.273J / 1.270J Logistics and Supply Chain Management Fall 2009 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.