1、1Chapter 5:Risk Pooling&Forecasting1Dr.YANG Ruina2Agenda Risk Pooling Case 1:ACME Forecasting 3Risk Pooling Demand variability is reduced if one aggregates demand across locations.More likely that high demand from one customer will be offset by low demand from another.Reduction in variability allows
2、 a decrease in safety stock and therefore reduces average inventory.6Acme Risk Pooling Case Electronic equipment manufacturer and distributor 2 warehouses for distribution in Massachusetts and New Jersey(partitioning the northeast market into two regions)Customers(retailers)receiving items from ware
3、houses(each retailer is assigned a warehouse)Warehouses receive material from Chicago Current rule:97%service level Each warehouse operate to satisfy 97%of demand(3%probability of stock-out)7 Replace the 2 warehouses with a single warehouse(located some suitable place)and try to implement the same s
4、ervice level 97%New Idea8Historical DataA slow-moving product9Summary of Historical Data10Inventory Levels11Savings in Inventory Average inventory for Product A:At NJ warehouse is about 88 units At MA warehouse is about 91 units In the centralized warehouse is about 132 units Average inventory reduc
5、ed by about 36 percent Average inventory for Product B:At NJ warehouse is about 15 units At MA warehouse is about 14 units In the centralized warehouse is about 20 units Average inventory reduced by about 43 percent Discussion Question Analyze the strengths and weaknesses of the current distribution
6、 system and the new distribution system.(e.g.delivery lead time,total inventory investment)13 Centralizing inventory reduces both safety stock and average inventory in the system.-Reallocation of items from one market to another easily accomplished in centralized systems.Not possible to do in decent
7、ralized systems where they serve different markets.Critical Points14 The higher the coefficient of variation,the greater the benefit from risk pooling.-The higher the variability,the higher the safety stocks kept by the warehouses.The variability of the demand aggregated by the single warehouse is l
8、ower.Critical Points15 The benefits from risk pooling depend on the behavior of the demand from one market relative to demand from another.-Risk pooling benefits are higher in situations where demands observed at warehouses are negatively correlated Critical Points17 Centralized vs.Decentralized Sys
9、tems Safety stock:lower with centralization Service level:higher service level for the same inventory investment with centralization Overhead costs:higher in decentralized system Customer lead time:response times lower in the decentralized system Transportation costs:not clear.Consider outbound and
10、inbound costs.18Contents of Forecasting Introduction Forecast Evaluation Subjective Methods Objective Methods -Causal Models -Time Series Models Summary 19Learning Objectives of Forecasting Understand commonly used forecasting techniques Learn to evaluate forecasts Learn to choose appropriate foreca
11、sting techniques20Introduction to Forecasting21Introduction to Forecasting22longtermDemandfulfillmentPurchasingProductioncontrolAggregate planningDemand forecastingInventorymanagementOperationsschedulingDistributionplanningTransportplanningFulfillmentimplementationDistribution network designSupply C
12、hain Management Product developmentmediumtermshorttermDistributionFacility location andlayoutManufacturingSupply net-work designPartnerselectionProduct portifolioDerivativeproductdevelopmentAdaptionsCurrent productsupportMaterialsorderingSupplycontractdesign Demand forecasting is the starting point
13、of all planning and control!23Characteristics of Forecast The forecast is always wrong It is difficult to match supply and demandThe longer the forecast horizon,the worse the forecast(Time horizon)It is even more difficult if one needs to predict customer demand for a long period of timeAggregate fo
14、recasts are more accurateChoosing appropriate aggregation levels,time horizons,and forecasting techniques is crucial24A Good Forecast is More than a Single Number25Long-term Forecasts are Always Wrong26What Makes a Good Forecast?27TWO FORECASTS1617181920Aug-02Sep-02Oct-02Nov-02Dec-02 Sales Which for
15、ecast is better?How can we evaluate the forecasting performance?Forecast quality28Forecast Errors29Evaluation of Forecast Accuracy30Measuring Forecast AccuracyForecast 131Measuring Forecast AccuracyForecast 232Evaluations of Two Forecasts33Bias in Forecast34Bias in Forecast35Reasons for Bias in Fore
16、cast Linear trend or non-linear trend Seasonality External factors,such as promotion and advertisementIf relevant elements are not considered in the forecast,the forecast can become biased.These elements can include:36Qualitative MethodsQualitativeMethodsSales ForceCompositePanel of ExpertsMarket Re
17、searchDelphiMethodApplication Used to generate forecasts if historical data are not available(e.g.,introduction of new product)Used to modify forecasts generated by other approaches(e.g.,considering information not included in quantitative methods)37Sales Force Estimate RationaleSales force is close
18、 to customer and has good information on future demandsApproach Members of sales force periodically report their estimates.These estimates are then aggregated to generate the overall forecastMain advantages Sales force knows customer well Sales territories are typically divided by district/region.Sa
19、les forecasts can be broken down correspondingly38Sales Force Estimate Bias of sales force-Might have incentives to overestimate sales or underestimate sales-Might naturally be optimistic or pessimistic Sales force does not always have all information necessary to generate forecast-Features of produ
20、cts launched in future-Preferences of customers in new market segmentsTypical applicationMain drawbacksShort-term and medium-term demand forecasting39Executive OpinionRationaleUpper-level management has best information on latest product developments and future product launchesApproachSmall group of
21、 upper-level managers collectively develop forecasts Combine knowledge and expertise from various functional areas People who have best information on future developments generate the forecastsMain advantages40Executive Opinion Expensive No individual responsibility for forecast quality Risk that fe
22、w people dominate the group Typical applicationsMain drawbacksShort-term and medium-term demand forecasting41Market ResearchRationaleUltimately,consumers drive demandApproachDetermine consumer interests by creating and testing hypotheses through data-gathering surveys:Design questionnaireSelect cust
23、omer sample Conduct survey(e.g.,telephone,mail,or interview)Analyze information and generate forecast42Market Research Expensive Require considerable knowledge and skills Sometimes validity not guaranteed due to low response rates:For mailed questionnaires response rate often 30%Typical application
24、Systematic and fact-based approach Excellent accuracy for short-term forecasts Good accuracy for medium-term forecastsMain advantagesMain drawbacksShort-term and medium-term demand forecasting43Delphi MethodRationaleAnonymous written responses encourage honesty and avoid that a group of experts are
25、dominated by only a few membersApproachCoordinatorsends initialquestionnaireEach expertwrites response(anonymous)CoordinatorperformsanalysisCoordinatorsends updatedquestionnaireConsensusreached?CoordinatorsummarizesforecastNoYes44Delphi Method Slow process Experts are not accountable for their respo
26、nses Little evidence that reliable long-term forecasts can be generated with Delphi or other methods Long-term forecasting Technology forecasting Generate consensus Can forecast long-term trend without availability of historical dataMain advantagesMain drawbacksTypical application45Objective Forecas
27、ting Methods46Causal ModelsCausalModelsLinearRegressionNon-linearRegressionApplicationUsed to forecast the performance(demand,profit,etc.)of a business investment based on the observed data of existing and similar business activities 47A Simple Example48Linear Regression:ObjectiveObjectiveIdeaFind a
28、 linear function that represents the predicted variable y as a function of predictive variables x1,x2,xm and best fits the observed data4950515253EXAMPLE:m=1(1)a=1,796 x 2,710 132 x 35,29012 x 1,796 132 x 132=50.6 b=12 x 35,290 132 x 2,71012 x 1,796 132 x 132=15.9 Coefficientsxkykxk2xkyk7150491,0502
29、1004200613036780415016600142501963,500152702254,050162402563,840122001442,400142701963,780204404008,800153402255,1007170491,1901322,7101,79635,290Observed data and analysis54EXAMPLE:m=1(2)025050001020 Population Demandy(x)=a+bx=50.6+15.9xQuestion:What demand would we expect from investing in a busin
30、ess with a nearby population 10 thousand?Answer:y(10)=5556575859606162636465666768697172737475767778798081828384858687888990919293Other Factors of Forecasting94The Most Appropriate Technique Purpose of the forecast How will the forecast be used?Dynamics of system for which forecast will be made.How
31、accurate is the past history in predicting the future?95Determine the Most Appropriate Technique96Summary Demand planning/forecasting is the starting point of all planning The performance of forecasting approach can be evaluated based on various metrics -MAD -MSE -MAPE Various forecasting approaches exist.Which one is appropriate depends on the situation.-Qualitative methods,-Causal models,or-Time-series models
