Bài giảng môn Quản trị kinh doanh - Chapter 10: Demand forecasting: building the foundation for resource planning

Chapter 10Demand Forecasting: Building the Foundation for Resource Planning1Learning ObjectivesDescribe the benefits of effective resource planning. Explain how the planning horizon affects planning tasks. Describe how lead times determine the planning horizon. Explain how product and service life cycles can aid in the planning process.Describe the benefits of collaborative planning, forecasting & replenishment Describe the different types of forecasting methods. Compute a causal forecast using simple linear regression. Recognize the components of a time series and appropriate forecasting techniques for each component. Compute forecasts using averages, exponential smoothing, seasonal indexes, and a multiplicative model. Compute measures of forecast accuracy. Describe how enterprise resource planning (ERP) systems benefit businesses. shot2Resource Planning - Determining what is needed, and making arrangements to get it, in order to achieve objectives.Contingency Plans – Alternative or back-up plans to be used if an unexpected event makes the normal plans infeasible.Operations Management Framework3 Increasing Alternatives Management has more options if it plans ahead. Profitability EnhancementPlanning can both reduce costs and increase sales.The further ahead we plan, however, the less we know about future conditions. There is a tradeoff between increasing alternatives and increasing uncertainty.Financial Benefits of Effective Planning4 Planning HorizonThe distance into the future one plans.Looking into the Future:The Planning Horizon5A business may have many different planning horizons depending on the resources in question Inventory- Usually very shortEmployees - Generally pretty shortTemps, new hires, etc.Equipment - A little longerPurchasing and installation lead timesFacilities - LongestPurchase property, build the buildingLooking into the Future:The Planning Horizon6Life Cycle: A pattern of demand growth and decline that occurs from the introduction of a product to its obsolescence.The five stages of a life cycle:IntroductionGrowth – Demand begins to increase. Maturation – Demand begins to level off. Saturation – Demand shifts to the beginning of its decline. Decline – Final stage as demand disappears.Exhibit 10.2 Product Life CycleProduct and Service Life Cycles7Market leaders sometimes try to create entry barriers by replacing products and maintaining intentionally short life cycles.Exhibit 10.3 Product Life Cycles Interrupted by New Product IntroductionProduct and Service Life Cycles8Demand ForecastingQualitative ForecastsDo not use past data. Usually used when such data is not available (such as planning for a new product).Customer surveys, expert opinions, etc. Quantitative ForecastsDivided into causal forecasting and time series forecasting techniques.9Collaborative Planning, Forecasting, and Replenishment (CPFR)A shared process of creation between two or more parties with diverse skills and knowledge delivering a unified approach that provides the optimal framework for customer satisfaction.CPFR requires that data be shared among supply chain partners and that partners collaborate on developing demand forecasts..10Causal TechniquesUses external data to predict future demandLooking for the factors that “cause” demandLinear regression is often used.Time Series TechniquesUse past demand to predict future demandDemand Forecasting: Quantitative Analysis11Some external variable is a leading indicator (independent variable) for the demand you want to predict (dependent variable)The example (10.1) uses temperature as the independent variable, but you could use others as well.e.g., exam schedule, promotions, sporting events, day of the week, etc.Causal Models12Demand Forecasting: Simple Linear Regression ExamplePredicted High Temp.Beer SalesPredicted High Temp.Beer Sales624,000636,1508513,0008814,800809,0009018,500582,5009217,100687,0008613,000727,4008913,8008211,6009419,1008612,9009118,4509318,0008716,7009118,2008215,100799,100718,3508410,200778,9008511,00013Demand Forecasting: Simple Linear Regression ExampleIf we believe that fluctuations in demand for beer, Y, are partly due to changes in X, the predicted temperature: Given a particular temperature prediction, what will demand be?Predicted14Demand Forecasting:Simple Linear Regression ExampleX1, Y1X2, Y2Regression analysis provides the formula for the line that best fits through the data points.Underlying model: Y = a + bx15Regression Line: Demand = -23,535 + 438.44 (Predicted Temperature)Demand Forecasting:Simple Linear Regression Example16Y = -23,535 + 438.44xFor an 80-degree day, the demand forecast would be:Y = -23,535 + 438.44(80) = 11,540.2Demand Forecasting:Simple Linear Regression Example17There are four potential components of a time series:Cycles A pattern that repeats over a long period of time (such as 20 years). Cycles are less important for demand forecasting, since we rarely have 20 years’ worth of data.TrendSeasonalityRandomnessDemand Forecasting:Components of a Time Series18Demand Forecasting:Components of a Time SeriesTrend – Component of a time series that causes demand to increase or decrease. Exhibit 10.6 Example of a Time Series with Trend19Seasonality – A pattern in a time series that repeats itself at least once a year.Exhibit 10.7 Time Series with SeasonalityDemand Forecasting:Components of a Time Series20Random Fluctuation – Unpredictable variation in demand that is not due to trend, seasonality, or cycle.Exhibit 10.8 Time Series with Random FluctuationDemand Forecasting:Components of a Time Series21Time Series Techniques: AveragesAveraging is used to remove random fluctuations in historical data.Various kinds of averages can be usedDifferences between them are exploited to create varying degrees of responsiveness. Responsiveness: The degree to which the forecast responds to the most immediate change in demand.22Time Series Techniques: AveragesAverages constructed from bigger data sets (i.e., more history) are less responsive to sudden changes.An average that uses the eight most recent data points is less responsive than one that uses the past three:Exhibit 10.11 Three-Period and Eight-Period Moving Average Forecasts23Time Series Techniques: AveragesMoving averages can be weighted to change responsivenessWeights must sum up to 1.0For more responsiveness, assign heavier weights to more recent data pointsPeriod 1 2 3 4Demand133130134146Example 10 .2Use weighted averages and the past four weeks’ demand to predict the next week’s demand. Demands for the past four weeks are:24Time Series Techniques: Averages25A variant of moving average (weighted average):Premise: More recent observations are better indicators of future demand than past observations.Reduces the need to hold lots of data.Uses a smoothing constant, ‘alpha’ () to weight the previous demand and establish the responsiveness of the forecast.Ft+1 = At + (1- )FtWhere:Ft+1 = The forecast for the next time period = A smoothing constant, between 0 and 1 At = The actual demand for the most recent periodFt = The forecast for the most recent periodTime Series Techniques:Exponential Smoothing26A higher alpha makes the forecast more responsive to changes: Exhibit 10 .13 Comparison of .1 and .4 Alpha Values for Exponential Smoothing Time Series Techniques:Exponential Smoothing27Example 10 .3: To forecast February’s demand using exponential smoothing with an alpha of .3. (assume an initial January forecast of 90)Ft+1 = At + (1- )Ft = .3(100) + (1-.3)90 = 30 + 63 = 93 Time Series Techniques:Exponential Smoothing Example28Continue the process until the forecast for July is determined:Time Series Techniques:Exponential Smoothing Example29Trend-adjusted Exponential Smoothing adds a smoothing constant to account for trendAlso called “forecast including trend” (FIT)FIT t+1 = Ft + TtWhere Ft is the smoothed forecast, Tt is the trend estimate andFt = FIT t + (At – FITt)Tt = Tt-1 + (FITt - FITt-1 - Tt-1)Time Series Techniques:Trend-Adjusted Exponential Smoothing30Example 10 .4: Using the following data = 0.2 = 0.9Initial trend (F1) = 3Initial forecast (T1) = 25Calculate the demandFor period twoFor period threeFor subsequent periods48464241383934302925Demand (A)10987654321Week (t)Time Series Techniques:Trend-Adjusted Exponential Smoothing31For period 2FITt+1 = Ft + TtInitial forecast (F1) = 25Initial Trend (T1) = 3FIT2 = F1 + T1 = 25 + 3 = 28 48464241383934302925Demand (A)10987654321Week (t)FIT t+1 = Ft + TtFt = FIT t + (At – FITt)Tt = Tt-1 + (FITt - FITt-1 - Tt-1)Time Series Techniques:Trend-Adjusted Exponential Smoothing32For period 3FITt+1 = Ft + TtFIT3 = F2 + T2 F2 = FIT2 + (A2 – FIT2)F2 = 28 + .2(29-28)F2 = 28 + .2 = 28.2T2 = T1 + (FIT2 - FIT1 - T1)T2 = 3 + .9(28 - 25 - 3)T2 = 3 + .9(0) = 3FIT3 = F2 + T2FIT3 = 28.2 + 3FIT3 = 31.2048464241383934302925Demand (A)10987654321Week (t)FIT t+1 = Ft + TtFt = FIT t + (At – FITt)Tt = Tt-1 + (FITt - FITt-1 - Tt-1)Time Series Techniques:Trend-Adjusted Exponential Smoothing33For period 4FITt+1 = Ft + TtFIT4 = F3 + T3 F3 = FIT3 + (A3 – FIT3)F3 = 31.2 + .2(30-31.2)F3 = 31.2 + .2(-1.2) = 30.96T3 = T2 + (FIT3 - FIT2 - T2)T3 = 3 + .9(31.2 - 28 - 3)T3 = 3 + .9(.2) = 3.18FIT4 = F3 + T3FIT4 = 30.96 + 3.18FIT4 = 34.1448464241383934302925Demand (A)10987654321Week (t)FIT t+1 = Ft + TtFt = FIT t + (At – FITt)Tt = Tt-1 + (FITt - FITt-1 - Tt-1)Time Series Techniques:Trend-Adjusted Exponential Smoothing34Time Series Techniques:Trend-Adjusted Exponential Smoothing35Time Series Techniques:Using the Linear Trend EquationIdentical to using linear regression as a causal techniqueTime period is the independent variableDemand is the dependent variableExample 10.5:Consider the following 10-month time series with an apparent trend:Month: 12345678910Demand:30831536039141242344545647148236Time Series Techniques:Using the Linear Trend EquationExhibit 10.16 Partial Excel Regression Analysis Output for Backpack Sales for Example 10.537Seasonality: a common component in time seriesSell more ski equipment in Fall and WinterSeasonality is described by using a ratio of the average demand for a period to the average demand across all periodsIf July has a seasonal index of 1.8, it means that average July demand is 1.8 times greater than overall average monthly demandTime Series Techniques:Including Seasonality38Calculate average demand for each “season” (period average)e.g. all Mondays, all January, etc.Compute average of all observations (global average)Divide period averages by global averageTime Series Techniques:Including Seasonality39So if in general I forecast 20 visitors per day, I adjust by the seasonal index to estimate what I expect on a particular dayTime Series Techniques:Including Seasonality40A regression-based approach (Multiplicative model)Compute seasonal indexes for each periodRemove seasonal component from the time series“Deseasonalize” the dataModel the trend using linear regression on the deseasonalized dataDetermine the forecast by using the trend equation and seasonal indexesTime Series Techniques:Dealing with Seasonality and Trend41Calculate seasonal indexes to deseasonalize dataTime Series Techniques:Dealing with Seasonality and Trend42The regression analysis determines the best-fitting line through the deseasonalized demand. The general equation for that line is:Y = a + btWhere: Y = a point on the trend line a = Y intercept b = slope t = time period Dealing with Seasonality and TrendUsing Regression43Regression analysis result: Y = 280.48 + 2.30 (period)Forecast (deseasonalized): Y = 280.48 + 2.30 (17) = 280.48 + 39.10 = 319.58Forecast (seasonal):Multiply back by the appropriate seasonal index Y = 319.58(Q1 index) = 319.58(1.67) = 533.70Example: Given the trend in demand over the past 4 years and the effects of seasonality, what do we expect demand to be in period 17?Time Series Techniques:Dealing with Seasonality and Trend44Percentage (79%) of the variation in demand for tax services is explained by the time periodRate of demand growth per periodBase-level service demand (period 0)Time Series Techniques:Dealing with Seasonality and Trend45Forecast AccuracyForecast error is the actual demand minus the forecast demand.Absolute Error: how far “off” are we, in absolute terms?Measured by mean absolute deviation (MAD) or mean squared error (MSE)Forecast Bias: Are we consistently high or low?A forecast should be unbiased (low forecasts are as frequent as high forecasts)Bias is measured by mean forecast error (MFE) or running sum of forecast error (RSFE)46The ideal value for both is zero, which would mean there is no forecasting errorThe larger the MAD or MSE, the less the accurate the model Forecast AccuracyTwo similar approaches are used to measure absolute forecast errorMAD is the mean of the absolute values of the forecast errorsMSE is the mean of the squared values of the forecast errors47Example 10.8: Calculating MAD48Example 10.9: Calculating MSE49Forecast BiasForecast Bias: Tendency of a forecast to be too high or too low.Mean forecast error (MFE)The mean of the forecast errors Running sum of forecast errors (RSFE)The sum of forecast error, updated as each new error is calculated.Ideal measure is zero which indicates no bias.Positive means forecast tends to lowNegative means forecast tends to high50Forecast Bias: Calculating MFE and RSFEMean Forecast Error = 1.00RSFE (period 8) = 851Tracking SignalsThe size of the cumulative forecast error expressed as MADsTS = RSFE/MAD52Integrated Resource Planning Systems Enterprise Resource Planning (ERP) SystemsPlanning for resources done from common databaseAllows decisions to be made from enterprise perspectiveEveryone uses the same numbers.ERP solutions are typically “off the shelf” – not customized to business Major providers: SAP, BAAN, PeopleSoft, Oracle53Integrated Resource Planning SystemsExhibit 10.28 Conceptual View of a Generic ERP System54