You are on page 1of 32

M07_CHOP0203_06_SE_C07.

indd Page 177 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780

C H A P T E R

7 7
Demand Forecasting
in a Supply Chain

LEARNING OBJECTIVES
After reading this chapter, you will be able to
1. Understand the role of forecasting for both 3. Forecast demand in a supply chain given
an enterprise and a supply chain. historical demand data using time-series
2. Identify the components of a demand methodologies.
forecast. 4. Analyze demand forecasts to estimate
forecast error.

A
ll supply chain decisions made before demand has materialized are made to a forecast.
In this chapter, we explain how historical demand information can be used to forecast
future demand and how these forecasts affect the supply chain. We describe several
methods to forecast demand and estimate a forecast’s accuracy. We then discuss how these meth-
ods can be implemented using Microsoft Excel.

7.1 THE ROLE OF FORECASTING IN A SUPPLY CHAIN
Demand forecasts form the basis of all supply chain planning. Consider the push/pull view of the
supply chain discussed in Chapter 1. All push processes in the supply chain are performed in
anticipation of customer demand, whereas all pull processes are performed in response to cus-
tomer demand. For push processes, a manager must plan the level of activity, be it production,
transportation, or any other planned activity. For pull processes, a manager must plan the level of
available capacity and inventory, but not the actual amount to be executed. In both instances, the
first step a manager must take is to forecast what customer demand will be.
A Home Depot store selling paint orders the base paint and dyes in anticipation of customer
orders, whereas it performs final mixing of the paint in response to customer orders. Home Depot
uses a forecast of future demand to determine the quantity of paint and dye to have on hand (a
push process). Farther up the supply chain, the paint factory that produces the base also needs
177

M07_CHOP0203_06_SE_C07.indd Page 178 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780

178 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO

forecasts to determine its own production and inventory levels. The paint factory’s suppliers also
need forecasts for the same reason. When each stage in the supply chain makes its own separate
forecast, these forecasts are often very different. The result is a mismatch between supply and
demand. When all stages of a supply chain work together to produce a collaborative forecast,
however, the forecast tends to be much more accurate. The resulting forecast accuracy enables
supply chains to be both more responsive and more efficient in serving their customers. Leaders in
many supply chains, from electronics manufacturers to packaged-goods retailers, have improved
their ability to match supply and demand by moving toward collaborative forecasting.
Consider the value of collaborative forecasting for Coca-Cola and its bottlers. Coca-Cola
decides on the timing of promotions based on the demand forecast over the coming quarter. Pro-
motion decisions are then incorporated into an updated demand forecast. The updated forecast is
essential for the bottlers to plan their capacity and production decisions. A bottler operating
without an updated forecast based on the promotion is unlikely to have sufficient supply avail-
able for Coca-Cola, thus hurting supply chain profits.
Mature products with stable demand, such as milk or paper towels, are usually easiest to
forecast. Forecasting and the accompanying managerial decisions are extremely difficult when
either the supply of raw materials or the demand for the finished product is highly unpredictable.
Fashion goods and many high-tech products are examples of items that are difficult to forecast.
In both instances, an estimate of forecast error is essential when designing the supply chain and
planning its response.
Before we begin an in-depth discussion of the components of forecasts and forecasting
methods in the supply chain, we briefly list characteristics of forecasts that a manager must
understand to design and manage his or her supply chain effectively.

7.2 CHARACTERISTICS OF FORECASTS
Companies and supply chain managers should be aware of the following characteristics of forecasts.
1. Forecasts are always inaccurate and should thus include both the expected value of the
forecast and a measure of forecast error. To understand the importance of forecast error, consider
two car dealers. One of them expects sales to range between 100 and 1,900 units, whereas the other
expects sales to range between 900 and 1,100 units. Even though both dealers anticipate average
sales of 1,000, the sourcing policies for each dealer should be very different, given the difference in
forecast accuracy. Thus, the forecast error (or demand uncertainty) is a key input into most supply
chain decisions. Unfortunately, most firms do not maintain any estimates of forecast error.
2. Long-term forecasts are usually less accurate than short-term forecasts; that is, long-
term forecasts have a larger standard deviation of error relative to the mean than short-term fore-
DBTUT 4FWFO&MFWFO +BQBO IBT FYQMPJUFE UIJT LFZ QSPQFSUZ UP JNQSPWF JUT QFSGPSNBODF5IF
company has instituted a replenishment process that enables it to respond to an order within
hours. For example, if a store manager places an order by 10 a.m., the order is delivered by
7 p.m. the same day. Therefore, the manager has to forecast what will sell that night only less
than 12 hours before the actual sale. The short lead time allows a manager to take into account
current information that could affect product sales, such as the weather. This forecast is likely to
be more accurate than if the store manager had to forecast demand a week in advance.
3. Aggregate forecasts are usually more accurate than disaggregate forecasts, as they tend to
have a smaller standard deviation of error relative to the mean. For example, it is easy to forecast
UIFHSPTTEPNFTUJDQSPEVDU (%1 
PGUIF6OJUFE4UBUFTGPSBHJWFOZFBSXJUIMFTTUIBOBQFSDFOU
error. However, it is much more difficult to forecast yearly revenue for a company with less than a
2 percent error, and it is even harder to forecast revenue for a given product with the same degree of
accuracy. The key difference among the three forecasts is the degree of aggregation. The GDP is an
aggregation across many companies, and the earnings of a company are an aggregation across sev-
eral product lines. The greater the aggregation, the more accurate the forecast.

M07_CHOP0203_06_SE_C07.indd Page 179 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780

 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 179

4. In general, the farther up the supply chain a company is (or the farther it is from the
consumer), the greater the distortion of information it receives. One classic example of this phe-
nomenon is the bullwhip effect (see Chapter 10), in which order variation is amplified as orders
move farther from the end customer. Collaborative forecasting based on sales to the end cus-
tomer helps upstream enterprises reduce forecast error.
In the next section, we discuss the basic components of a forecast, explain the four classi-
fications into which forecasting methods fall, and introduce the notion of forecast error.

7.3 COMPONENTS OF A FORECAST AND FORECASTING METHODS
Yogi Berra, the former New York Yankees catcher who is famous for his malapropisms, once
said, “It’s tough to make predictions, especially about the future.” One may be tempted to treat
demand forecasting as magic or art and leave everything to chance. What a firm knows about its
customers’ past behavior, however, sheds light on their future behavior. Demand does not arise in
a vacuum. Rather, customer demand is influenced by a variety of factors and can be predicted, at
least with some probability, if a company can determine the relationship between these factors
and future demand. To forecast demand, companies must first identify the factors that influence
future demand and then ascertain the relationship between these factors and future demand.
Companies must balance objective and subjective factors when forecasting demand.
Although we focus on quantitative forecasting methods in this chapter, companies must include
IVNBOJOQVUXIFOUIFZNBLFUIFJSGJOBMGPSFDBTU4FWFO&MFWFO+BQBOJMMVTUSBUFTUIJTQPJOU
4FWFO&MFWFO+BQBOQSPWJEFTJUTTUPSFNBOBHFSTXJUIBTUBUFPGUIFBSUEFDJTJPOTVQQPSU
system that makes a demand forecast and provides a recommended order. The store manager,
however, is responsible for making the final decision and placing the order, because he or she
may have access to information about market conditions that are not available in historical
demand data. This knowledge of market conditions is likely to improve the forecast. For exam-
ple, if the store manager knows that the weather is likely to be rainy and cold the next day, he or
she can reduce the size of an ice cream order to be placed with an upstream supplier, even if
demand was high during the previous few days when the weather was hot. In this instance, a
change in market conditions (the weather) would not have been predicted using historical
demand data. A supply chain can experience substantial payoffs from improving its demand
forecasting through qualitative human inputs.
A company must be knowledgeable about numerous factors that are related to the demand
forecast, including the following:
r 1BTUEFNBOE
r -FBEUJNFPGQSPEVDUSFQMFOJTINFOU
r 1MBOOFEBEWFSUJTJOHPSNBSLFUJOHFGGPSUT
r 1MBOOFEQSJDFEJTDPVOUT
r 4UBUFPGUIFFDPOPNZ
r "DUJPOTUIBUDPNQFUJUPSTIBWFUBLFO
A company must understand such factors before it can select an appropriate forecasting
methodology. For example, historically a firm may have experienced low demand for chicken
OPPEMFTPVQJO+VMZBOEIJHIEFNBOEJO%FDFNCFSBOE+BOVBSZ*GUIFGJSNEFDJEFTUPEJTDPVOU
UIFQSPEVDUJO+VMZ UIFTJUVBUJPOJTMJLFMZUPDIBOHF XJUITPNFPGUIFGVUVSFEFNBOETIJGUJOHUP
UIFNPOUIPG+VMZ5IFGJSNTIPVMENBLFJUTGPSFDBTUUBLJOHUIJTGBDUPSJOUPDPOTJEFSBUJPO
Forecasting methods are classified according to the following four types:
1. Qualitative: Qualitative forecasting methods are primarily subjective and rely on
human judgment. They are most appropriate when little historical data are available or when
FYQFSUTIBWFNBSLFUJOUFMMJHFODFUIBUNBZBGGFDUUIFGPSFDBTU4VDINFUIPETNBZBMTPCFOFDFT-
sary to forecast demand several years into the future in a new industry.

the rate of growth or decline in demand for the next period. The forecast error measures the difference between the fore- cast and actual demand.4 BASIC APPROACH TO DEMAND FORECASTING The following five points are important for an organization to forecast effectively: 1. A company cannot (and should not) forecast the direction of the random component. a good forecasting method has an error whose size is com- parable to the random component of demand. product pricing is strongly correlated with demand. a firm can combine time-series and causal methods to answer such questions as: What will be the impact of a price promotion? What will be the impact of a competitor opening a store nearby? Airlines simulate customer buying behavior to forecast demand for higher-fare seats when no seats are available at lower fares. The random component is the part of the forecast that deviates from the systematic part. 7. Time series: Time-series forecasting methods use historical demand to make a fore- cast. the predictable seasonal fluctuations in demand. Therefore. 3. any observed demand can be broken down into a systematic and a random component: Observed demand 1O2 = systematic component 1S2 + random component 1R2 The systematic component measures the expected value of demand and consists of what we will call level. A company may find it difficult to decide which method is most appropriate for forecast- ing. Forecast at the appropriate level of aggregation. there is always a random element that cannot be explained by his- torical demand patterns. 4. . the method has merged the historical random component with the systematic component. All a com- pany can predict is the random component’s size and variability. Causal: Causal forecasting methods assume that the demand forecast is highly cor- related with certain factors in the environment (the state of the economy. which provides a measure of forecast error. several studies have indicated that using multiple forecasting methods to create a combined forecast is more effective than using any one method alone. These are the simplest methods to implement and can serve as a good starting point for a demand forecast. For example. Understand the objective of forecasting. A manager should be skeptical of a forecasting method that claims to have no forecasting error on historical demand. On average. Integrate demand planning and forecasting throughout the supply chain. and any seasonal patterns.). Using simulation. etc. Companies can thus use causal methods to determine the impact of price promotions on demand. 3. These methods are most appropriate when the basic demand pattern does not vary sig- nificantly from one year to the next. 2. the current deseasonalized demand. As a result.indd Page 180 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 180 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 2. In this case. Establish performance and error measures for the forecast. The objective of forecasting is to filter out the random component (noise) and estimate the systematic component. They are based on the assumption that past demand history is a good indicator of future demand. trend. the forecasting method will likely perform poorly. growth patterns. Identify the major factors that influence the demand forecast. Simulation: 4JNVMBUJPOGPSFDBTUJOHNFUIPETJNJUBUFUIFDPOTVNFSDIPJDFTUIBUHJWF rise to demand to arrive at a forecast. we deal primarily with time-series methods. In this chapter. which are most appropriate when future demand is related to historical demand.M07_CHOP0203_06_SE_C07. In fact. interest rates. 5. 4. With any forecasting method. Causal forecasting methods find this correlation between demand and environmental factors and use estimates of what environmental factors will be to forecast future demand. and seasonality.

On the product side. an accurate forecast will have great value. These estimates must be based on demand. a firm must know the number of variants of a product being sold and whether these variants substitute for or complement one another. among others. Although the decline in demand for the original product is not indicated by historical data.indd Page 181 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 181 Understand the Objective of Forecasting Every forecast supports decisions that are based on it. the supermarket must understand what the demand would have been in the absence of promotion activity and how demand is affected by promotions and competitor actions. For example. a retailer develops forecasts based on promo- tional activities. with members from each affected function responsible for forecasting demand—and an even better idea is to have members of different companies in the supply chain working together to create a forecast. 8BMNBSUTQMBOTUPEJTDPVOUEFUFSHFOUEVSJOHUIFNPOUIPG+VMZNVTUCFTIBSFEXJUIUIFNBOV- facturer. Identify Major Factors That Influence the Demand Forecast Next. and product-related phenomena that influence the demand forecast. how much to inventory. Failure to make these deci- sions jointly may result in either too much or too little product in various stages of the supply chain. supply. If demand for a product influences or is influenced by demand for another product. the transporter.M07_CHOP0203_06_SE_C07. A combination of these pieces of information will allow the supermarket to forecast demand for +VMZ HJWFOUIFQSPNPUJPOBDUJWJUZQMBOOFEGPSUIBUZFBS On the supply side. production planning. Integrate Demand Planning and Forecasting Throughout the Supply Chain A company should link its forecast to all planning activities throughout the supply chain. a company must ascertain whether demand is growing or declining or has a seasonal pattern. demand for the two products should be forecast jointly. comparable cereal brands was MPXJO+VMZ5IFTVQFSNBSLFUTIPVMEOPUVTFUIFTBMFTEBUBGSPNUPFTUJNBUFUIBUEFNBOEGPS UIJTCSBOEXJMMCFIJHIJO+VMZ CFDBVTFUIJTXJMMPDDVSPOMZJGUIFTBNFCSBOEJTQSPNPUFE BHBJOJO+VMZBOEPUIFSCSBOETSFTQPOEBTUIFZEJEUIFQSFWJPVTZFBS8IFONBLJOHUIF demand forecast. and how much to order. when a firm introduces an improved version of an existing product. This leads to a mismatch between supply and demand. All parties should come up with a common forecast for the promotion and a shared plan of action based on the forecast. it is a good idea for a firm to have a cross-functional team. All parties affected by a supply chain decision should be aware of the link between the decision and the forecast. For example. a firm must identify demand. and purchasing. Examples of such decisions include how much of a particular product to make. the historical demand is still useful in that it allows the firm to estimate the combined total demand for the two versions. whereas the demand for other. These include capacity planning. In one unfortunately common scenario. a company must consider the available supply sources to decide on the accuracy of the forecast desired. a highly accurate forecast may not be especially important. . resulting in poor customer service. If alternate supply sources with short lead times are available. if only a single supplier with a long lead time is available. whereas a manufacturer. promotion planning. not on sales data. unaware of these promotions. it is likely that the demand for the existing product will decline because customers will buy the improved version. as they all must make decisions that are affected by the forecast of demand. develops a different fore- cast for its production planning based on historical orders. On the demand side. the two forecasts are best made jointly. 'PSFYBNQMF BTVQFSNBSLFUQSPNPUFEBDFSUBJOCSBOEPGDFSFBMJO+VMZ"TBSFTVMU UIF demand for this cereal was high. However. To accomplish integration. Clearly. and others involved in filling demand. so an important first step is to identify these decisions clearly.

The equation for calculating the systematic com- ponent may take a variety of forms: r Multiplicative: 4ystematic component = level * trend * seasonal factor r Additive: 4ystematic component = level + trend + seasonal factor r Mixed: 4ystematic component = 1level + trend2 * seasonal factor The specific form of the systematic component applicable to a given forecast depends on the nature of demand. A better approach may be to forecast demand at the aggregate level when ordering with the supplier and ask each store manager to forecast only when the shirts are to be allocated across the stores. These measures should be linked to the objectives of the business decisions based on these forecasts. 4ystematic component = 1level + trend2 * seasonal factor A similar approach can be applied for other forms as well. the company must compare actual demand to forecasted demand to estimate the accuracy of the forecast. Establish Performance and Error Measures for the Forecast Companies should establish clear performance measures to evaluate the accuracy and timeliness of the forecast. At the end of the sales season.5 TIME-SERIES FORECASTING METHODS The goal of any forecasting method is to predict the systematic component of demand and esti- mate the random component. Companies may develop both static and adaptive forecasting methods for each form. Then plans for decreasing future forecast errors or responding to the observed forecast errors can be put into place. and seasonality within the systematic component do not vary as new demand is observed. a trend.M07_CHOP0203_06_SE_C07. 7. In this section. We assume that the systematic component of demand is mixed. the long lead time forecast (supplier order) is aggregate. we estimate each of these parameters based on historical data and then use the same values for all future forecasts. One approach is to ask each store manager the precise number of shirts needed and add up all the requests to get an order size with the supplier. given the supply chain decision that is driven by the forecast. when local market intelligence is likely to be most effective. we discuss a static forecasting method for use when demand has a trend as well as a seasonal component. In this case. we discuss techniques for static and adaptive time-series forecasting. Consider a mail-order company that uses a forecast to place orders with its suppli- ers. trend. The problem with this approach is that it makes store managers forecast well before demand arises at a time when their forecasts are unlikely to be accurate. The disaggregate store-level forecast is made close to the sales season. Consider a buyer at a retail chain who is forecasting to select an order size for shirts. thus lowering error. In its most general form. the systematic component of demand data contains a level. We begin with a few basic definitions: L = estimate of level at t = 0 (the deseasonalized demand estimate during Period t = 0) T = estimate of trend (increase or decrease in demand per period) .indd Page 182 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 182 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Forecast at the Appropriate Level of Aggregation Given that aggregate forecasts are more accurate than disaggregate forecasts. it is important to forecast at a level of aggregation that is appropriate. that is. The advantage of this approach is that it uses local market intelligence that each store manager has. In the next section. and a seasonal factor. The mail-order company must ensure that the forecast is created at least two months before the start of the sales season because of the two-month lead time for replenishment. In this case. Static Methods A static method assumes that the estimates of level. We now describe these static and adaptive forecasting methods. which take two months to send in the orders.

T.indd Page 183 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 183 TABLE 7-1 Quarterly Demand for Tahoe Salt Year Quarter Period. The level in Period t + l is the sum of the level in Period 0 (L) and (t + l) times the trend T. As an example. After meeting with JUTSFUBJMFST 5BIPFIBTEFDJEFEUPQSPEVDFBDPMMBCPSBUJWFGPSFDBTU5BIPF4BMUXBOUTUPXPSLXJUIUIF retailers to create a more accurate forecast based on the actual retail sales of their salt. 1 3.000 3 4 11 32.M07_CHOP0203_06_SE_C07. but the company has noticed that these retailers always overestimate their purchases. leaving Tahoe (and even some retailers) stuck with excess inventory.000 20.1) We now describe one method for estimating the three parameters L. the forecast in Period t for demand in Period t + l is a prod- uct of the level in Period t + l and the seasonal factor for Period t + l. 4 3. 3 1. 1 2.000 Demand 30.000 10.000 2 3 6 18. 4 4. The forecast in Period t for demand in Period t + l is thus given as Ft + l = 3L + 1t + l2T4St + l (7.000 40. 2 1.000 2 4 7 23. 3 3.000 3 3 10 13. and S. 4 2.000 2 1 4 34. t Demand. 3 2. Dt 1 2 1 8. 2 2.000 2 2 5 10.000 4 1 12 41. Quarterly retail demand data for the past three years are shown in Table 7-1 and charted in Figure 7-1.000 St = estimate of seasonal factor for Period t Dt = actual demand observed in Period t Ft = forecast of demand for Period t In a static forecasting method. consider the demand for rock salt used primarily to melt snow. 1 Period FIGURE 7-1 Quarterly Demand at Tahoe Salt .PVOUBJOT*OUIFQBTU 5BIPF4BMUIBTSFMJFEPOFTUJNBUFTPGEFNBOEGSPNB sample of its retailers. This salt is produced by a firm called 5BIPF4BMU XIJDITFMMTJUTTBMUUISPVHIBWBSJFUZPGJOEFQFOEFOUSFUBJMFSTBSPVOEUIF-BLF5BIPFBSFB PGUIF4JFSSB/FWBEB.000 3 2 9 12. 2 3.000 1 4 3 23.000 0 1. 50.000 1 3 2 13.000 3 1 8 38.

p =JTFWFO'PSt = 3.(p>2) + Dt + (p>2) + i = t + 1-(p>2) i=2 With this procedure. this method provides deseasonalized demand at a point between Period l + (p/2) and Period l + 1 + (p/2).1)>24 Di >p for p odd i = t . based on the change in demand over time: Dt = L + Tt (7. we can obtain deseasonalized demand between Periods 3 and 10 as shown in Figures 7-2 and 7-3 (all details are available in the accompanying spreadsheet Chapter 7-Tahoe-salt).2 as follows: a 2Di d n (2p) = D1 + D + a 2Di n 8 t-1 + (p>2)  D3 = c Dt . Estimate seasonal factors. The company estimates that growth will continue in the coming year at historical rates. We now describe the following two steps required to estimate each of the three parameters—level. Given that we are measuring demand on a quarterly basis.(p>2) Dt = e (7. for Period t.indd Page 184 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 184 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO In Figure 7-1. trend. By taking the average of deseasonalized demand provided by Periods l + 1 to l + p and l + 2 to l + p + 1.3. Dt represents deseasonalized demand and not the actual demand in Period t. and time t. If p is even. The following linear relationship exists between the deseasonalized demand. There is also a growth trend in the demand. Deseasonalized demand represents the demand that would have been observed in the absence of seasonal fluctuations.JDSPTPGU Excel (Data ! Data Analysis ! Regression). 1. with sales growing over the past three years. and T represents the rate of growth of deseasonalized demand or trend. We can estimate the values of L and T for the deseasonalized demand using linear regression with deseasonalized demand (see Figure 7-2) as the dependent WBSJBCMFBOEUJNFBTUIFJOEFQFOEFOUWBSJBCMF4VDIBSFHSFTTJPODBOCFSVOVTJOH. and seasonal factors. The periodic- ity (p JTUIFOVNCFSPGQFSJPETBGUFSXIJDIUIFTFBTPOBMDZDMFSFQFBUT'PS5BIPF4BMUTEFNBOE  the pattern repeats every year. we obtain the deseasonalized demand using Equa- tion 7.3(p . we obtain the deseasonalized demand for Period l + 1 + (p/2) if p is even. This sequence of commands opens the Regression .2) a t + 3(p . If p is odd. we take the average of p consecutive periods of demand. L represents the level or deseasonalized demand at Period 0.1)>24 In our example. ESTIMATING LEVEL AND TREND The objective of this step is to estimate the level at Period 0 and the trend. the periodicity for the demand in Table 7-1 is p = To ensure that each season is given equal weight when deseasonalizing demand. 2. Each cycle lasts four quarters. increasing from the second quarter of a given year to the first quarter of the following year. can be obtained as follows: a t . and the demand pattern repeats every year. The average of demand from Period l + 1 to Period l + p provides deseasonalized demand for Period l + (p + 1)/2. the deseasonalized demand. Dt. observe that demand for salt is seasonal.3) In Equation 7. Dt . The second quarter of each year has the lowest demand. We start by deseasonalizing the demand data.(p>2) + Dt + (p>2) + 2Di d n (2p) for p even i = t + 1 . this method provides deseasonalized demand for an existing period. Deseasonalize demand and run linear regression to estimate level and trend.1 + (p>2) c Dt . Thus.M07_CHOP0203_06_SE_C07.

000 20. . and the trend. is obtained as the X variable coefficient (or the slope) from the sheet containing the regression results.4) It is not appropriate to run a linear regression between the original demand data and time to estimate level and trend because the original demand data are not linear and the resulting linear regres- sion will not be accurate.indd Page 185 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 185 Cell Cell Formula Equation Copied to C4 =(B2+B6+2*SUM(B3:B5))/8 7. This new sheet contains estimates for both the initial level L and the trend T. The demand must be deseasonalized before we run the linear regression.000 40.2 C5:C11 FIGURE 7-2 Excel Workbook with Deseasonalized Demand for Tahoe Salt 50.39 + 2t (7. T. A new sheet containing the results of the regression opens up (see work- sheet Regression-1). L. For the Tahoe 4BMUFYBNQMF XFPCUBJOL = BOET = BMMEFUBJMTBSFBWBJMBCMFJOUIFXPSLTIFFU Regression-1 and numbers are rounded to integer values). The initial level.M07_CHOP0203_06_SE_C07. deseasonalized demand Dt for any Period t is thus given by Dt = 18.000 Actual Demand Deseasonalized Demand Demand 30. For this example.000 10. is obtained as the intercept coefficient.000 0 1 2 3 4 5 6 7 8 9 10 11 12 Period FIGURE 7-3 Deseasonalized Demand for Tahoe Salt EJBMPHCPYJO&YDFM'PSUIF5BIPF4BMUXPSLCPPLJO'JHVSF JOUIFSFTVMUJOHEJBMPHCPY XF enter Input Y Range:C:C11 Input X Range:A:A11 and click the OK button.

We can now obtain the forecast for the next four quarters using Equation 7.7 S2 = 1S2 + S6 + S10 2 >3 = 10.868 F1 = 1L + 1T2S1 = 118. Given the periodicity p. we obtain the seasonal factor for a given period by averaging sea- sonal factors that correspond to similar periods. We obtain seasonal factors using Equation 7.68 = 17.22 >3 = 0. for all periods of the form pt + i. we have estimated the level.5) Dt 'PSUIF5BIPF4BMUFYBNQMF UIFEFTFBTPOBMJ[FEEFNBOEFTUJNBUFEVTJOH&RVBUJPOBOEUIF TFBTPOBMGBDUPSTFTUJNBUFEVTJOH&RVBUJPOBSFTIPXOJO'JHVSF TFFXPSLTIFFUFigure 7-4).68 + 1.7 + 0. In the example.4 C3:C13 D2 =B2/C2 7.68 S3 = 1S3 + S7 + S11 2 >3 = 11.M07_CHOP0203_06_SE_C07. For example.67 + 0.27 .1 + 1. if we have a periodicity of p =  1FSJPET  BOEIBWFTJNJMBSTFBTPOBMGBDUPST5IFTFBTPOBMGBDUPSGPSUIFTFQFSJPETJTPCUBJOFE as the average of the three seasonal factors.6 as S1 = 1S1 + S + S9 2 >3 = 10.indd Page 186 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 186 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Cell Cell Formula Equation Copied to C2 =18439+A2*524 7.5 D3:D13 FIGURE 7-4 Deseasonalized Demand and Seasonal Factors for Tahoe Salt ESTIMATING SEASONAL FACTORS We can now obtain deseasonalized demand for each QFSJPEVTJOH&RVBUJPO TFF'JHVSF 5IFTFBTPOBMGBDUPSSt for Period t is the ratio of actual demand Dt to deseasonalized demand Dt and is given as Dt St = (7.83 + 0.0 + 1. trend.322 >3 = 1.662 >3 = 1.1. Given r seasonal cycles in the data.39 + 13 * 220. the forecast for the next four periods using the static forecasting method is given by F13 = 1L + 13T2S13 = 118. we obtain the seasonal factor as a Sjp + i r-1 j=0 (7.2 + 0. and all seasonal factors.7 = 11.17 S = 1S + S8 + S12 2 >3 = 11. 1 … i … p.2 >3 = 0.39 + 1 * 220.67 At this stage.6) Si = r 'PSUIF5BIPF4BMUFYBNQMF BUPUBMPGQFSJPETBOEBQFSJPEJDJUZPGp =JNQMZUIBUUIFSF are r = 3 seasonal cycles in the data.66 + 1.

67 = . 2.M07_CHOP0203_06_SE_C07. It is desirable that the modification be such that if the demand is lower than forecast.39 + 16 * 221. . and seasonal factor (St+p+1). .8) 4. and seasonal factors (S1.770 F16 = 1L + 16T2S16 = 118. . . with periodicity p. and 4UFQT  BOEBSFSFQFBUFEVOUJMBMMIJTUPSJDBMEBUBVQUP1FSJPEn have been covered. when the systematic component of demand data has the mixed form and contains a level. and seasonality are updated after each demand observation. Given quarterly data. however.7) The four steps in the adaptive forecasting framework are as follows: 1. and a variety of production and inventory inefficiencies would result. The framework can also be specialized for the case in which the systematic component contains no seasonality or trend.39 + 1 * 221. Sp) from the given data. Adaptive Forecasting In adaptive forecasting. trend.17 = 30. wherein the pattern repeats itself every year. given the error Et+1 in the forecast. forecast demand for Period t + 1 using Equa- tion 7. 3.Dt + 1 (7. The main advantage of adaptive forecasting is that estimates incorporate all new data that are observed. The revised estimates in Period t + 1 are then used to make a forecast for Period t + 2. Modify estimates: Modify the estimates of level (Lt+1).indd Page 187 19/09/14 8:35 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 187 F1 = 1L + 1T2S1 = 118. The error for Period t + 1 is stated as Et + 1 = Ft + 1 . trend (Tt+1).7. and seasonal factor at Period 0. . This is done exactly as in the static forecasting method discussed earlier in the chapter with L0 = L and T0 = T. and a seasonal factor. Estimate error: Record the actual demand Dt+1 for Period t + 1 and compute the error Et+1 in the forecast for Period t + 1 as the difference between the forecast and the actual demand. the estimates are revised upward. this supply chain would have a less accurate forecast. Forecast: Given the estimates in Period t. Initialize: Compute initial estimates of the level (L0). We assume that we have a set of historical data for n periods and that demand is seasonal. trend. The framework is provided in the most general setting. We now discuss a basic framework and several methods that can be used for this type of forecast. the forecast for Period t + l in Period t uses the estimate of level and trend in Period t (Lt and Tt respectively) and is given as Ft + l = 1L t + lTt 2St + l (7. the estimates of level. the estimates are revised downward. we have a periodicity of p = We begin by defining a few terms: Lt = estimate of level at the end of Period t Tt = estimate of trend at the end of Period t St = estimate of seasonal factor for Period t Ft = forecast of demand for Period t (made in Period t − 1 or earlier) Dt = actual demand observed in Period t Et = Ft − Dt = forecast error in Period t In adaptive methods. Our first forecast is for Period 1 and is made with the estimates of level. The esti- mates at Period n are then used to forecast future demand.79 5BIPF4BMUBOEJUTSFUBJMFSTOPXIBWFBNPSFBDDVSBUFGPSFDBTUPGEFNBOE8JUIPVUUIFTIBS- ing of sell-through information between the retailers and the manufacturer. trend (T0). It can easily be modified for the other two cases. whereas if the demand is higher than forecast. a trend.

The forecast is stated as Ft + 1 = L t and Ft + n = L t (7. the moving average becomes less responsive to the most recently observed demand. In each case. EXAMPLE 7-1 Moving Average A supermarket has experienced weekly demand of milk of D1 = 120.2 "GUFSPCTFSWJOHEFNBOEJO1FSJPE UIFSFWJTFEFTUJNBUFPGMFWFMGPS1FSJPEJTHJWFOCZ L  = 1D + D + D3 + D2 2 > = 112 + 122 + 11 + 1272 > = 122 SIMPLE EXPONENTIAL SMOOTHING The simple exponential smoothing method is appropri- ate when demand has no observable trend or seasonality. we simply add the latest observation and drop the old- est one. The moving average corresponds to giving the last N periods of data equal weight when forecasting and ignoring all data older than this new moving average.7 5IFGPSFDBTUPGEFNBOEGPS1FSJPE VTJOH&RVBUJPO JTFYQSFTTFEBT F = L  = 120. we revise the estimates as follows: L t + 1 = 1Dt + 1 + Dt + ### + Dt . As we increase N. This represents an N-period moving average and is evaluated as follows: L t = 1Dt + Dt .10) After observing the demand for Period t + 1. the level in Period t is estimated as the average demand over the most recent N periods. D3 =  BOE D = HBMMPOTPWFSUIFQBTUGPVSXFFLT'PSFDBTUEFNBOEGPS1FSJPEVTJOHBGPVSQFSJPE NPWJOHBWFSBHF8IBUJTUIFGPSFDBTUFSSPSJGEFNBOEJO1FSJPEUVSOTPVUUPCFHBMMPOT Analysis: 8FNBLFUIFGPSFDBTUGPS1FSJPEBUUIFFOEPG1FSJPE5IVT BTTVNFUIFDVSSFOUQFSJPEUPCF t =0VSGJSTUPCKFDUJWFJTUPFTUJNBUFUIFMFWFMJO1FSJPE6TJOH&RVBUJPO XJUIN = XF obtain L  = 1D + D3 + D2 + D1 2 > = 1122 + 11 + 127 + 1202 > = 120. In this case. 4ystematic component of demand = level . The revised moving average serves as the next forecast.D = 12 .1 + g + Dt . Ft + 2 = L t + 1 To compute the new moving average.indd Page 188 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 188 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO We now discuss various adaptive forecasting methods. 4ystematic component of demand = level In this method.N + 2 2 >N. We illustrate the use of the moving average in Example 7-1.9) The current forecast for all future periods is the same and is based on the current estimate of level.M07_CHOP0203_06_SE_C07. MOVING AVERAGE The moving average method is used when demand has no observable trend or seasonality.N + 1 2 >N (7.7 = .7 gallons "TEFNBOEJO1FSJPE D JTHBMMPOT XFIBWFBGPSFDBTUFSSPSGPS1FSJPEPG E = F . The method that is most appropriate depends on the characteristic of demand and the composition of the systematic component of demand. D2 = 127. we assume the period under consideration to be t. In this case.120.

Forecast demand for Period VTJOHTJNQMFFYQPOFOUJBMTNPPUIJOHXJUIa = 0. EXAMPLE 7-2 Simple Exponential Smoothing Consider the supermarket in Example 7-1.a2 nDt + 1 .1 × 127) + (0. The revised value of the level is a weighted average of the observed value of the level (Dt+1) in Period t + 1 and the old estimate of the level (Lt) in Period t.11) =1 The current forecast for all future periods is equal to the current estimate of level and is given as Ft + 1 = L t and Ft + n = L t (7. the revised estimate of level for Period 1 using Equation 7.9 * 120.7 The observed demand for Period 1 is D1 = 120.7 With a = 0.12) After observing the demand. L0.68) = 121. Given that D2 = 127. Dt+1.12) is thus given by F1 = L 0 = 120. This gives F3 = L2 = 121.a2L 0 = (0.7  i=1 The forecast for Period 1 (using Equation 7.31. D3 =  BOED = 122 gallons over the past four weeks. We thus obtain F2 = L1 = 120. We illustrate the use of exponential smoothing in Example 7-2.a2L t (7.68 Observe that the estimate of level for Period 1 is lower than for Period 0 because the demand in Period 1 is lower than the forecast for Period 1. with recent observations weighted higher than older observations.1.13. is taken to be the average of all historical data because demand has been assumed to have no observable trend or seasonality.9 × 120.indd Page 189 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 189 The initial estimate of level.120 = 0.7 .31. in which weekly demand for milk has been D1 = 120.68. We can thus rewrite Equation 7. Given demand data for Periods 1 through n. we can express the level in a given period as a function of the current demand and the level in the previous period. Analysis In this case. A higher value of a corresponds to a forecast that is more responsive to recent observations.D1 = 120. Using Equation 7. for Period t + 1. whereas a lower value of a represents a more stable forecast that is less responsive to recent observations.1.13 is given by L 1 = aD1 + 11 .1 * 120) + (0. we have the following: n ia 1 n L0 = Di (7. we obtain L2 = (0. The forecast error for Period 1 is given by E1 = F1 .n + 11 .a2 tD1 t-1 n=0 The current estimate of the level is a weighted average of all the past observations of demand. we have demand data for n =QFSJPET6TJOH&RVBUJPO UIFJOJUJBMFTUJNBUFPG level (rounded to 2 decimals) is expressed by L 0 = a Di > = 120.7) = 120. we revise the estimate of the level as follows: L t + 1 = aDt + 1 + 11 .13 as L t + 1 = a a11 . .M07_CHOP0203_06_SE_C07.13) where a (0 6 a 6 1) is a smoothing constant for the level. D2 = 127.

16) where a(0 6 a 6 1) is a smoothing constant for the level and b (0 6 b 6 1) is a smoothing constant for the trend. EXAMPLE 7-3 Holt’s Model An electronics manufacturer has seen demand for its latest MP3 player increase over the past six months. we revise the estimates for level and trend as follows: L t + 1 = aDt + 1 + 11 .961.732. Period t. b = 0.72. running a linear regression between demand and time periods is appropriate because we have assumed that demand has a trend but no seasonality. we obtain L = (0.L t 2 + 11 . The estimate of initial level L0 is obtained as the intercept coefficient.1. we obtain L 0 = 7. of the form Dt = at + b In this case. This gives F = L = 120.367 and T0 = 673 5IFGPSFDBTUGPS1FSJPE VTJOH&RVBUJPO JTUIVTHJWFOCZ F1 = L 0 + T0 = 7.00 . In this case. Observed demand (in thousands) has been D1 =  D2 = 8. For the MP3 player data. We first run a linear regression (using the Excel tool Data ! Data Analysis ! Regression) between demand and time periods. The underlying relation- ship between demand and time is thus linear. The slope a measures the rate of change in demand per period and is our initial estimate of the trend T0.1 × 122) + ¤ = 120. but no seasonality. and time. given estimates of level Lt and trend Tt.72. the revised estimate (of level or trend) is a weighted average of the observed value and the old estimate. we have 4ystematic component of demand = level + trend We obtain an initial estimate of level and trend by running a linear regression between demand. Observe that in each of the two updates. Forecast demand for Period 7 using trend-corrected expo- nential smoothing with a = 0.9 × 121.808.14) After observing demand for Period t. We illustrate the use of Holt’s model in Example 7-3 (see associated spreadsheet Examples 1–4 Chapter 7). The constant b measures the estimate of demand at Period t = 0 and is our estimate of the initial level L0.2. D =  BOED6 = 11.M07_CHOP0203_06_SE_C07. TREND-CORRECTED EXPONENTIAL SMOOTHING (HOLT’S MODEL) The trend-corrected exponential smoothing (Holt’s model) method is appropriate when demand is assumed to have a level and a trend in the systematic component. Dt.15) Tt + 1 = b1L t + 1 .b2Tt (7. D3 =  D = 9.367 + 673 = 8. and the trend T0 is obtained as the X variable coefficient (or the slope) in the spreadsheet Examples 1-4 Chapter 7 (there is some variation between the spreadsheet and the results shown here because of rounding). Analysis The first step is to obtain initial estimates of level and trend using linear regression.a21L t + Tt 2 (7.31) =5IJTHJWFTF = L3 = (JWFOUIBUD = 122. In Period t. the forecast for future periods is expressed as Ft + 1 = L t + Tt and Ft + n = L t + nTt (7.indd Page 190 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 190 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Given that D3 = XFPCUBJOL3 = ¤ + (0.

L t 2 + 11 .399.1. . In Period t. trend.078 T1 = b1L 1 . .19) St + p + 1 = g1Dt + 1 >L t + 1 2 + 11 .M07_CHOP0203_06_SE_C07. . b (0 6 b 6 1) is a smoothing con- stant for the trend. and seasonal factors. and g (0 6 g 6 1) is a smoothing constant for the seasonal factor. the forecast for future periods is given by Ft + 1 = 1L t + Tt 2St + 1 and Ft + l = 1L t + lTt 2St + l (7. T = 661. .9 * 8.1.36724 + 10.7. L  = 10. .79 Continuing in this manner.b2Tt (7. .1 = -37 With a = 0. L 6 = 11. This gives us a fore- cast for Period 7 of F7 = L 6 + T6 = 11. Observe that in each of the updates (level.18) Tt + 1 = b1L t + 1 .676.D1 = 8.AND SEASONALITY-CORRECTED EXPONENTIAL SMOOTHING (WINTER’S MODEL) This method is appropriate when the systematic component of demand has a level. Lt.7.399 + 673 = 12. or seasonal factor). given estimates of level. and sea- sonal factors as follows: L t + 1 = a1Dt + 1 >St + 1 2 + 11 .2 * 18. we revise the estimates for level. our updates have increased the estimate of level L1GPS1FSJPEGSPN UP BOEUIFFTUJ- NBUFPGUSFOEGSPNUP6TJOH&RVBUJPO XFUIVTPCUBJOUIFGPMMPXJOHGPSFDBTUGPS Period 2: F2 = L 1 + T1 = 8.L 0 2 + 11 . . and seasonal factors (S1.072 TREND.393.a21L t + Tt 2 (7. g = 0. the revised estimate of level and trend for Period 1 using Equa- UJPOTBOEJTHJWFOCZ L 1 = aD1 + 11 . a trend.039.8 * 6732 = 681 Observe that the initial estimate for demand in Period 1 is too low. Tt .b2T0 = 30. St. To begin.indd Page 191 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 191 The observed demand for Period 1 is D1 = 5IFFSSPSGPS1FSJPEJTUIVTHJWFOCZ E1 = F1 . b = 0. and a sea- sonal factor.a2 1L 0 + T0 2 = 10. .078 + 681 = 8. the revised estimate is a weighted average of the observed value and the old estimate. trend.20) where a (0 6 a 6 1) is a smoothing constant for the level. trend (T0).1. L  = 10.8. b = 0.00 . trend. Sp). . L 3 = 9. T = 666. In this case we have 4ystematic component of demand = 1level + trend2 * seasonal factor Assume periodicity of demand to be p. T6 = 673.1 * 8. St+p-1. we need initial estimates of level (L0).g2St + 1 (7.078 .17) On observing demand for Period t + 1. we obtain L 2 = 8. T2 = 680. T3 = 672. EXAMPLE 7-4 Winter’s Model $POTJEFSUIF5BIPF4BMUEFNBOEEBUBJO5BCMF'PSFDBTUEFNBOEGPS1FSJPEVTJOHUSFOEBOE seasonality-corrected exponential smoothing with a = 0.2. We obtain these estimates using the procedure for static forecasting described earlier in the chapter.2.002 = 8. As a result. We illustrate the use of Winter’s model in &YBNQMF TFFXPSLTIFFUExample 7-4).12 + 10.

7692 4 + 10. The forecast error for Period 1 is thus given by E1 = F1 .7 = 8. the forecasting method is overestimating the systematic component and should be corrected.1.8. because its demand experiences both a trend and seasonality.6 MEASURES OF FORECAST ERROR As mentioned earlier. 2.39 + 22 4 = 18.39 T0 = 2 S1 = 0.1. we describe how a manager can estimate and use forecast error. and seasonal factors exactly as in the static case. Managers use error analysis to determine whether the current forecasting method is pre- dicting the systematic component of demand accurately.8 * 22 = 8 S = g1D1 >L 1 2 + 11 .000 = 913 With a = 0. All contingency plans must account for forecast error.68 = 13.L 0 2 + 11 . the revised estimate of level and trend for Period 1 and TFBTPOBMGBDUPSGPS1FSJPE VTJOH&RVBUJPOT  BOE JTHJWFOCZ L 1 = a1D1 >S1 2 + 11 .a21L 0 + T0 2 = 30.9 * 0.093 The forecasting methods we have discussed and the situations in which they are generally applicable are as follows: Forecasting Method Applicability Moving average No trend or seasonality Simple exponential smoothing No trend or seasonality Holt’s model Trend but no seasonality Winter’s model Trend and seasonality *G5BIPF4BMUVTFTBOBEBQUJWFGPSFDBTUJOHNFUIPEGPSUIFTFMMUISPVHIEBUBPCUBJOFEGSPN its retailers. They are expressed as follows: L 0 = 18.72 = 0. Forecast errors contain valu- able information and must be analyzed carefully for two reasons: 1.18. trend.39 + 220. A good forecasting method should capture the systematic component of demand but not the random component. The first is in the Far East and has a lead time of two months.b2T0 = 30.D1 = 8. The second is .769 + 82 * 0.g2S1 = 30. *GXFEPOPULOPXUIBU5BIPF4BMUFYQFSJFODFTCPUIUSFOEBOETFBTPOBMJUZ IPXDBOXFGJOE out? Forecast error helps identify instances in which the forecasting method being used is inap- propriate.000>18.3924 + 10.17) is thus given by F2 = 1L 1 + T1 2S2 = 118.17) is thus given by F1 = 1L 0 + T0 2S1 = 118.17 S = 1.2. g = 0. every instance of demand has a random component.769 T1 = b1L 1 .68 S3 = 1.7 The forecast of demand for Period 2 (using Equation 7.913 The observed demand for Period 1 is D1 = 8. Consider a mail-order company with two suppliers.769 . b = 0. For example.72 4 + 30.2 * 118.1 * 18. The random component manifests itself in the form of a forecast error. 7.7 S2 = 0.67 The forecast for Period 1 (using Equation 7.1 * 18. Winter’s model is the best choice. if a forecasting method consistently produces a positive error.913 .M07_CHOP0203_06_SE_C07. In the next section.000>0.9 * 118.indd Page 192 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 192 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Analysis We obtain the initial estimates of level.000.

MAD is an appropriate choice when selecting forecasting methods if the cost of a fore- cast error is proportional to the size of the error.4&UPDPNQBSFGPSFDBTUJOHNFUIPETJGUIFDPTUPGBMBSHFFSSPSJTNVDIMBSHFSUIBOUIF HBJOTGSPNWFSZBDDVSBUFGPSFDBTUT6TJOHUIF. It is important that a manager estimate the error of a forecast made at least as far in advance as the lead time required for the manager to take whatever action the forecast is to be used for. Even when the error distribution is sym- metric. this may be another signal that the firm should change its forecasting method. In this case the standard deviation of the random component is s = 1. At. Finding an error that is well beyond historical estimates may indicate that the forecasting method in use is no longer appropriate or demand has fundamentally changed.21) =1 5IF. that is.4&BTBNFBTVSFPGFSSPSJTBQQSPQSJBUFXIFO forecast error has a distribution that is symmetric about zero. In a situation with a six-month lead time. . the error in Period t is the difference between the forecast for Period t and the actual demand in Period t. where the following holds: Et = Ft . The local supplier is more expensive than the Far East supplier.21 can also have n − 1 instead of n): n ta 1 n 2 MSEn = Et (7.2 MAD (7.or underestimate demand. As long as observed errors are within historical error estimates. At = ! Et ! Define the mean absolute deviation (MAD) to be the average of the absolute deviation over all periods. a manager should estimate the error for a forecast made six months before demand arises. As defined earlier. The decision regarding the quantity of local capacity to contract is closely linked to the size of the forecast error with a two-month lead time. and 20.4&QFOBMJ[FTMBSHF errors much more significantly than small errors because all errors are squared. Thus. to be the absolute value of the error in Period t. The mail-order company wants to contract a certain amount of contin- gency capacity with the local supplier to be used if the demand exceeds the quantity the Far East supplier provides.22) =1 The MAD can be used to estimate the standard deviation of the random component assum- ing that the random component is normally distributed.4&JGUIF forecast error does not have a symmetric distribution.23) We then estimate that the mean of the random component is 0. there is no point in estimating errors for a forecast made one month in advance.indd Page 193 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 193 local and can fill orders with one week’s notice.Dt That is. If all of a firm’s forecasts tend to consistently over. For example."%JTBCFUUFSNFBTVSFPGFSSPSUIBO. Thus. and the standard deviation of the random component of demand is s. if a forecast will be used to determine an order size and the supplier’s lead time is six months.4& BNFUIPEXJUIBGPSFDBTUFSSPSTFRVFODFPG   BOE 9 will be preferred to a method with an error sequence of 1.4&DBOCFSFMBUFEUPUIFWBSJBODFPGUIFGPSFDBTUFSSPS*OFGGFDU XFFTUJNBUFUIBUUIF SBOEPNDPNQPOFOUPGEFNBOEIBTBNFBOPGBOEBWBSJBODFPG. forecast error for Period t is given by Et.4& XIFSFUIFGPMMPXJOHIPMET (the denominator in Equation 7. One measure of forecast error is the mean squared error . as expressed by n ta 1 n MADn = At (7. if we select GPSFDBTUNFUIPETCZNJOJNJ[JOH. 3. 2.M07_CHOP0203_06_SE_C07.4&5IF. Define the absolute deviation in Period t. firms can continue to use their current forecasting method. it is a good idea UPVTFUIF.

If we are aiming for a long-term smoothing constant of a = 1 .rt In the long term. the slope of the best straight line passing through should be 0. one needs a method to track and control the forecasting method. if we plot all the errors.FUIPEIBT BMPXFS. it makes sense to increase the weight on current data relative to older data when making forecasts. The tracking signal 54 JTUIFSBUJPPGUIFCJBTBOEUIF.24) MAPEn = n The MAPE is a good measure of forecast error when the underlying forecast has signifi- cant seasonality and demand varies considerably from one period to the next.FUIPETIPVMECFQSFGFSSFEUP Method 1. where the following holds: biasn = a Et n (7. When a forecast method stops reflecting the underlying demand pattern (for instance.26) MADt *GUIF54BUBOZQFSJPEJTPVUTJEFUIFSBOHF {6.indd Page 194 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 194 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO The mean absolute percentage error (MAPE) is the average absolute error as a percentage of demand and is given by a ` D ` 100 Etn t=1 t (7."1& QFSDFOU*OUIJTJOTUBODF JUDBOCFBSHVFEUIBU. . Ideally.FUIPEBOEXPVMECFQSFGFSSFEJGFJUIFSDSJUFSJPOXBTVTFE *GEFNBOEJTIJHIMZTFBTPOBM IPXFWFS BOEBWFSBHFT      BOE JOUIFGPVS periods.1 1 .r. if demand drops considerably as it did for the automotive industry in 2008–2009). Because trend is not included."%BOEJTHJWFOBT biast TSt = (7."%SFMBUJWFUP. Method 2 results in a MAPE of 9.1 1 .M07_CHOP0203_06_SE_C07.4&BOE. the forecast errors are unlikely to be randomly distributed around 0.FUIPESFUVSOTGPSFDBTUFSSPSTPG   BOEPWFSGPVSRVBSUFST. the smoothing constant will converge to a = 1 .25) t=1 The bias will fluctuate around 0 if the error is truly random and not biased one way or the other.r at = = r + at . One JOTUBODFJOXIJDIBMBSHFOFHBUJWF54XJMMSFTVMUPDDVSTXIFOEFNBOEIBTBHSPXUIUSFOEBOEUIF manager is using a forecasting method such as moving average. a declining alpha approach would be to start with a0 = 1 and reset the smoothing constant as follows: at . making historical data less rele- vant. McClain (1981) recommends the “declining alpha” method when using exponential smoothing when the smoothing constant starts large (to give greater weight to recent data) but then decreases over time. this is a signal that the forecast is biased and is either underforecasting (TS 6 -6) or overforecasting (TS 7 +6). Consider a sce- nario in which two methods are used to make quarterly forecasts for a product with seasonal EFNBOEUIBUQFBLTJOUIFUIJSERVBSUFS. This may happen because the forecasting method is flawed or the underlying demand pattern has shifted. One approach is to use the sum of forecast errors to evaluate the bias.FUIPESFUVSOTGPSFDBTUFSSPSTPG   BOE . UIFBWFSBHFPGIJTUPSJDBMEFNBOEJTBMXBZTMPXFSUIBOGVUVSFEFNBOE5IFOFHBUJWF54EFUFDUT that the forecasting method consistently underestimates demand and alerts the manager. The tracking signal may also get large when demand has suddenly dropped (as it did for many industries in 2009) or increased by a significant amount.r with the forecasts becoming more stable over time. whereas Method 1 results in a much higher . In general. If demand has suddenly dropped.9 percent.

7 SELECTING THE BEST SMOOTHING CONSTANT When using exponential smoothing.M07_CHOP0203_06_SE_C07. The initial level is estimated using Equation 7. In general. 6). it is best to use a smoothing constant that is no larger than 0.4& ."% BOE. The smoothing constant aJTPCUBJOFEVTJOH4PMWFS FIGURE 7-5 Selecting Smoothing Constant by Minimizing MSE .2. If a manager has a good sense of the underlying demand pattern."1&*OUIFBCTFODFPGBQSFGFSFODFBNPOH FSSPSUFSNT JUJTCFTUUPQJDLTNPPUIJOHDPOTUBOUTUIBUNJOJNJ[FUIF.indd Page 195 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 195 7.4& We illustrate the impact of picking smoothing constants that minimize different error mea- TVSFTVTJOHUIFQFSJPEEFNBOEEBUBTIPXOJODFMMT##PG'JHVSF BDDPNQBOZJOH spreadsheet Chapter 7-Tahoe-salt and worksheet Figures 7-5. it is best to pick smoothing constants that minimize the error term that a manager is most DPNGPSUBCMFXJUIGSPNBNPOH. the value of the smoothing constant chosen has a direct impact on the sensitivity of the forecast to recent data.11 and is shown in cell C2.

"%EFDSFBTFTUP  DPNQBSFEUPJO'JHVSF BOEUIF. The forecasts and errors with the resulting a = 0. we show the results from minimizing MAD (cell G13)."1&EFDSFBTFTUPQFSDFOU DPNQBSFEUP QFSDFOUJO'JHVSF 5IFNBKPSEJGGFSFODFCFUXFFOUIFUXPGPSFDBTUTJTJO1FSJPE UIFQFSJPE XJUIUIFMBSHFTUFSSPS TIPXOJODFMM% XIFONJOJNJ[JOH. 5IFTNPPUIJOHDPOTUBOUDBOBMTPCFTFMFDUFEVTJOH4PMWFSCZNJOJNJ[JOHUIF."%= BOE MAPE = 2.4&QJDLTBTNPPUIJOHDPOTUBOU that reduces large errors.M07_CHOP0203_06_SE_C07.4&=  . In this case. UIF.indd Page 196 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 196 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO FIGURE 7-6 Selecting Smoothing Constant by Minimizing MAD CZNJOJNJ[JOHUIF.4&JODSFBTFTUP  DPNQBSFEUP JO'JHVSF XIFSFBTUIF. whereas minimizing MAD picks a smoothing constant that gives equal weight to reducing all errors even if large errors get somewhat larger. .4& DFMM' BUUIFFOEPGUIFQFSJPETBTTIPXOJO'JHVSF5IFGPSF- DBTUTIPXOJO'JHVSFVTFTUIFSFTVMUJOHa =BOEHJWFT.32 are shown in Figure 7-6.1 percent. In Figure 7-6."%PSUIF MAPE at the end of 10 periods.

10 D7:D13 E6 =D6-B6 7.24 J7:J13 K6 =Sum($E$6:E6)/H6 7.8 E7:E13 F6 =Abs(E6) F7:F13 G6 =Sumsq($E$6:E6)/(A6-4) 7.22 H7:H13 I6 =100*(F6/B6) I7:I13 J6 =Average($I$6:I6) 7.M07_CHOP0203_06_SE_C07.8 FORECASTING DEMAND AT TAHOE SALT 3FDBMMUIF5BIPF4BMUFYBNQMFFBSMJFSJOUIFDIBQUFSXJUIUIFIJTUPSJDBMTFMMUISPVHIEFNBOEGSPN JUTSFUBJMFST TIPXOJO5BCMF5IFEFNBOEEBUBBSFBMTPTIPXOJODPMVNO#PG'JHVSF TFF BTTPDJBUFETQSFBETIFFUChapter 7-Tahoe-salt 5BIPF4BMUJTDVSSFOUMZOFHPUJBUJOHDPOUSBDUTXJUI TVQQMJFSTGPSUIFGPVSRVBSUFSTCFUXFFOUIFTFDPOERVBSUFSPG:FBSBOEUIFGJSTURVBSUFSPG:FBS "OJNQPSUBOUJOQVUJOUPUIJTOFHPUJBUJPOJTUIFGPSFDBTUPGEFNBOEUIBU5BIPF4BMUBOEJUTSFUBJM FSTBSFCVJMEJOHDPMMBCPSBUJWFMZ5IFZIBWFBTTJHOFEBUFBN‡DPOTJTUJOHPGUXPTBMFTNBOBHFST GSPNUIFSFUBJMFSTBOEUIFWJDFQSFTJEFOUPGPQFSBUJPOTGPS5BIPF4BMU‡UPDPNFVQXJUIUIJTGPSF DBTU5IFGPSFDBTUJOHUFBNEFDJEFTUPBQQMZFBDIPGUIFBEBQUJWFGPSFDBTUJOHNFUIPETEJTDVTTFEJO UIJTDIBQUFSUPUIFIJTUPSJDBMEBUB5IFHPBMJTUPTFMFDUUIFNPTUBQQSPQSJBUFGPSFDBTUJOHNFUIPE BOEUIFOVTFJUUPGPSFDBTUEFNBOEGPSUIFOFYUGPVSRVBSUFST5IFUFBNEFDJEFTUPTFMFDUUIFGPSF DBTUJOHNFUIPECBTFEPOUIFFSSPSTUIBUSFTVMUXIFOFBDINFUIPEJTVTFEPOUIFRVBSUFSTPG IJTUPSJDBMEFNBOEEBUB Cell Cell Formula Equation Copied to C5 =Average(B2:B5) 7.indd Page 197 09/10/14 6:05 PM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 197 *O HFOFSBM  JU JT OPU B HPPE JEFB UP VTF TNPPUIJOH DPOTUBOUT NVDI MBSHFS UIBO  GPS FYUFOEFEQFSJPETPGUJNF"MBSHFSTNPPUIJOHDPOTUBOUNBZCFKVTUJGJFEGPSBTIPSUQFSJPEPGUJNF XIFOEFNBOEJTJOUSBOTJUJPO*UTIPVME IPXFWFS HFOFSBMMZCFBWPJEFEGPSFYUFOEFEQFSJPETPG UJNF 7.26 K7:K13 FIGURE 7-7 Tahoe Salt Forecasts Using Four-Period Moving Average .21 G7:G13 H6 =Sum($F$6:F6)/(A6-4) 7.9 C6:C13 D6 =C5 7.

is 1.13. have a fairly large MAD12 of 9. the forecast for Periods 13 through 16 (using Equation 7.1 does not indicate any significant bias.00 Given that MAD12 is 9.12 to forecast demand for the succeeding period. using a four-period moving average.2 * 9.9 to estimate level and Equation 7.90 Thus. to fore- cast demand. the forecast for the next four quarters (using Equation 7. Using Equation 7. MAD12 is 10.10) is given by F13 = F1 = F1 = F16 = L 12 = 2. The esti- mate of level is updated each period using Equation 7. Trend-Corrected Exponential Smoothing (Holt’s Model) The team next investigates the use of Holt’s model. with a MAPE12PGQFSDFOU'SPN'JHVSF PCTFSWFUIBU L 12 = 23. the team estimates the initial level for Period 0 to be the average demand for Periods 1 through 12 (see worksheet Figure 7-8).719 = 12.90 In this case. "TJOEJDBUFECZUIF54 XIJDISBOHFTGSPN-UP UIFGPSFDBTUVTJOHTJNQMFFYQP- nential smoothing with a = 0. however. observe that L 12 = 2. Thus. the standard deviation of forecast error is fairly large relative to the size of the forecast.760. However. the standard deviation of forecast error is fairly large relative to the size of the forecast. Simple Exponential Smoothing The forecasting team next uses a simple exponential smoothing approach. All calculations are shown in Figure 7-7 (see worksheet Figure 7-7 in spreadsheet Chap- ter 7-Tahoe-salt) and are as discussed in the section on the moving-average method earlier in this chapter.11. it has a fairly large MAD12 of 10.208 = 12. with a MAPE12PGQFSDFOU'SPN Figure 7-7.719.indd Page 198 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 198 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Demand in this case clearly has both a trend and seasonality in the systematic component. the team initially expects Winter’s model to produce the best forecast. "TJOEJDBUFECZDPMVNO. In this case.208.1.12) is given by F13 = F1 = F1 = F16 = L 12 = 23. This method is also tested on the 12 quarters of historical data.083 The team then uses Equation 7. Moving Average The forecasting team initially decides to test a four-period moving average for the forecast- ing. In this case. the systematic component of demand is given by 4ystematic component of demand = level + trend . In this case. the estimate of standard deviation of forecast error.10 to forecast demand. It does.M07_CHOP0203_06_SE_C07.00 Thus. which indi- cates that the forecast using the four-period moving average does not contain any significant bias. The initial level is the average of the demand entries in cells B3 to #JO'JHVSFBOESFTVMUTJO L 0 = 22.19.JO'JHVSF UIF54JTXFMMXJUIJOUIF {6 range. using a four-period moving average. The team uses Equation 7. The results are shown in Figure 7-8.719.2 * 10. with a = 0.208 and MAPE12JTQFSDFOU5IVT UIFFTUJNBUFPGTUBOEBSE deviation of forecast error using simple exponential smoothing is 1.

21 G4:G14 H3 =Sum($F$3:F3)/A3 7.1)*C2 7.1*B3+(1-0.M07_CHOP0203_06_SE_C07.12 D4:D14 E3 =D3-B3 7."1&12PGQFSDFOU'SPN'JHVSF  PCTFSWFUIBU L 12 =  44 aOE T12 = 1 541 .22 H4:H14 I3 =100*(F3/B3) I4:I14 J3 =Average($I$3:I3) 7.'SPNUIFSFHSFTTJPOPGUIFBWBJMBCMF EBUB TFFXPSLTIFFUholts-regression UIFUFBNPCUBJOTUIFGPMMPXJOH L  = 12 15 aOE T = 1 549 5IFUFBNOPXBQQMJFT)PMUTNPEFMXJUIa = 1BOEb = 2UPPCUBJOUIFGPSFDBTUTGPS FBDIPGUIFRVBSUFSTGPSXIJDIEFNBOEEBUBBSFBWBJMBCMF TFFXPSLTIFFUFigure 7-9 5IFZ NBLFUIFGPSFDBTUVTJOH&RVBUJPO VQEBUFUIFMFWFMVTJOH&RVBUJPO BOEVQEBUFUIF USFOEVTJOH&RVBUJPO5IFSFTVMUTBSFTIPXOJO'JHVSF "TJOEJDBUFECZB54UIBUSBOHFTGSPN-UP USFOEDPSSFDUFEFYQPOFOUJBMTNPPUI JOHXJUIa = 1BOEb = 2EPFTOPUTFFNUPTJHOJGJDBOUMZPWFSPSVOEFSGPSFDBTU)PXFWFS  UIFGPSFDBTUIBTBGBJSMZMBSHF.24 J4:J14 K3 =Sum($E$3:E3)/H3 7.8 E4:E14 F3 =Abs(E3) F4:F14 G3 =Sumsq($E$3:E3)/A3 7.26 K4:K14 FIGURE 7-8 Tahoe Salt Forecasts Using Simple Exponential Smoothing 5IFUFBNBQQMJFTUIFNFUIPEPMPHZEJTDVTTFEFBSMJFS"TBGJSTUTUFQ JUFTUJNBUFTUIFMFWFMBU 1FSJPEBOEUIFJOJUJBMUSFOE"TEFTDSJCFEJO&YBNQMF UIJTFTUJNBUFJTPCUBJOFECZSVOOJOH BMJOFBSSFHSFTTJPOCFUXFFOEFNBOE Dt BOEUJNF 1FSJPEt.indd Page 199 09/10/14 6:06 PM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 199 Cell Cell Formula Equation Copied to C3 =0."%12PG  XJUIB.13 C4:C14 D3 =C2 7.

1*B3+(1-0.2*(C3-C2)+(1-0.24 K4:K14 L3 =Sum($F$3:F3)/I3 7.8 F4:F14 G3 =Abs(F3) G4:G14 H3 =Sumsq($F$3:F3)/A3 7.21 H4:H14 I3 =Sum($G$3:G3)/A3 7.indd Page 200 09/10/14 6:06 PM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 200 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Cell Cell Formula Equation Copied to C3 =0.15 C4:C14 D3 =0.16 D4:D14 E3 =C2+D2 7.1)*(C2+D2) 7.2)*D2 7.22 I4:I14 J3 =100*(G3/B3) J4:J14 K3 =Average($J$3:J3) 7.14 E4:E14 F3 =E3-B3 7."% = 5IVT UIFFTUJNBUFPGTUBOEBSEEFWJBUJPOPGGPSFDBTUFSSPS VTJOH)PMUTNPEFMXJUIa = 0BOEb = 02JT25 * 8 836 =  05*OUIJTDBTF UIFTUBO- EBSEEFWJBUJPOPGGPSFDBTUFSSPSSFMBUJWFUPUIFTJ[FPGUIFGPSFDBTUJTTPNFXIBUTNBMMFSUIBOJUXBT XJUIUIFQSFWJPVTUXPNFUIPET)PXFWFS JUJTTUJMMGBJSMZMBSHF  "TBSFTVMUPGSPVOEJOH DBMDVMBUJPOTEPOFXJUIPOMZTJHOJGJDBOUEJHJUTTIPXOJOUIFUFYUNBZZJFMEBEJGGFSFOUSFTVMU5IJT JTUIFDBTFUISPVHIPVUUIFCPPL .M07_CHOP0203_06_SE_C07.26 L4:L14 FIGURE 7-9 Trend-Corrected Exponential Smoothing 5IVT VTJOH)PMUTNPEFM &RVBUJPO UIFGPSFDBTUGPSUIFOFYUGPVSQFSJPETJTHJWFOCZ UIFGPMMPXJOH: F3 = L 2 + T2 = 30 3 +  5 = 3 98 F = L 2 + 2T2 = 30 3 + 2 *  5 = 33 525 F5 = L 2 + 3T2 = 30 3 + 3 *  5 = 35 066 F6 = L 2 + T2 = 30 3 +  *  5 = 36 607 *OUIJTDBTF .

1)*D2 7.and Seasonality-Corrected Exponential Smoothing Trend.26 M4:M14 FIGURE 7-10 Trend.21 I4:I14 J3 =Sum($H$3:H3)/A3 7.22 J4:J14 K3 =100*(H3/B3) K4:K14 L3 =Average($K$3:K3) 7.05*(B3/E3)+(1-0.19 D4:D14 E7 =0.1*(C3-C2)+(1-0.indd Page 201 09/10/14 6:06 PM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 201 Cell Cell Formula Equation Copied to C3 =0.1*(B3/C3)+(1-0.24 L4:L14 M3 =Sum($G$3:G3)/J3 7.18 C4:C14 D3 =0.M07_CHOP0203_06_SE_C07.05)*(C2+D2) 7.17 F4:F18 G3 =F3-B3 7.1)*E3 7.8 G4:G14 H3 =Abs(G3) H4:H14 I3 =Sumsq($G$3:G3)/A3 7.and Seasonality-Corrected Exponential Smoothing (Winter’s Model) 5IFUFBNOFYUJOWFTUJHBUFTUIFVTFPG8JOUFSTNPEFMUPNBLFUIFGPSFDBTU"TBGJSTUTUFQ JUFTUJ- NBUFTUIFMFWFMBOEUSFOEGPS1FSJPE BOETFBTPOBMGBDUPSTGPS1FSJPETUISPVHIp =5PTUBSU  UIFEFNBOEJTEFTFBTPOBMJ[FE TFFXPSLTIFFUdeseasonalized 5IFO UIFUFBNFTUJNBUFTJOJUJBM MFWFMBOEUSFOECZSVOOJOHBSFHSFTTJPOCFUXFFOEFTFBTPOBMJ[FEEFNBOEBOEUJNF TFFXPSLTIFFU winters-regression 5IJTJOGPSNBUJPOJTVTFEUPFTUJNBUFUIFTFBTPOBMGBDUPST TFFXPSLTIFFU deseasonalized 'PSUIFEFNBOEEBUBJO'JHVSF BTEJTDVTTFEJO&YBNQMF UIFUFBN PCUBJOTUIFGPMMPXJOH L  = 8 439 T = 54 S = 47 S = 68 S3 = 7 S4 = 67 *UUIFOBQQMJFT8JOUFSTNPEFMXJUIa = 5 b =  g = UPPCUBJOUIFGPSFDBTUT "MMDBMDVMBUJPOTBSFTIPXOJO'JHVSF TFFXPSLTIFFUFigure 7-10 5IFUFBNNBLFTGPSF- DBTUTVTJOH&RVBUJPO VQEBUFTUIFMFWFMVTJOH&RVBUJPO VQEBUFTUIFUSFOEVTJOH&RVB- UJPO BOEVQEBUFTTFBTPOBMGBDUPSTVTJOH&RVBUJPO .20 E8:E18 F3 =(C2+D2)*E3 7.

This helps firms respond quickly to changes in the marketplace and avoid the costs of a delayed reaction.74 to 4. the team forecasts the following demand for the coming four quarters: 4FDPOE2VBSUFS :FBS  5IJSE2VBSUFS :FBS  'PVSUI2VBSUFS :FBS  'JSTU2VBSUFS :FBS  The standard deviation of forecast error is 1.0.902 F1 = 1L 12 + 2T12 2S1 = 12. The team compiles the error estimates for the four forecasting methods as shown in Table 7-2.indd Page 202 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 202 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO TABLE 7-2 Error Estimates for Tahoe Salt Forecasting Forecasting Method MAD MAPE (%) TS Range Four-period moving average 9. It is not surprising that Winter’s model results in the most accurate forecast. A good forecasting package provides forecasts across a wide range of products that are updated in real time by incorporating any new demand information. In this case. the fre- quency with which forecasting is performed.469 8 –2.1.719 49 –1.9 THE ROLE OF IT IN FORECASTING There is a natural role for IT in forecasting.2 * 1.17 S16 = 1.17 = 30.836.68 S1 = 1.67 = . the standard deviation of forecast error relative to the demand forecast is much smaller than with the other methods. thus incorporating the most current data into the demand forecast.21 Simple exponential smoothing 10.791 + 2 * 322 * 0. b = 0.00 *OUIJTDBTF UIF. Using Winter’s model.1 is 1.15 Holt’s model 8.69 = 1. Based on the error information in Table 7-2.68 = 17.791 T12 = 32 S13 = 0. Although this tech- nical improvement can help produce better forecasts.67 Using Winter’s model (Equation 7. because the demand data have both a growth trend as well as seasonality.836 52 –2. the forecasting team decides to use Winter’s model. 7.9 In this case.791 + 322 * 0."%PG BOE. A positive outcome of the investment in ERP systems has been a significant improvement in supply chain transparency and data integration.791 +  * 322 * 1.873 F16 = 1L 12 + T12 2S16 = 12. From Figure 7-10.M07_CHOP0203_06_SE_C07. observe that L 12 = 2. and the importance of getting the highest quality results possible. .52 to 2. firms must develop the organizational capabilities required to take advantage of this improvement.7 = 11.81 F1 = 1L 12 + 3T12 2S1 = 12."1&PGQFSDFOUBSFTJHOJGJDBOUMZMPXFSUIBOUIPTF obtained with any of the other methods.208 59 –1. MAD12 = 5IVT UIFFTUJNBUFPGTUBOEBSEEFWJBUJPOPGGPSFDBTUFSSPS using Winter’s model with a = 0.38 to 2.7 S1 = 0. the forecast for the next four periods is F13 = 1L 12 + T12 2S13 = 12.791 + 3 * 322 * 1. and g = 0.15 to 2. given the large amount of data involved.836. Good demand planning modules link not only to customer orders but often directly to customer sales information as well. thus allowing potentially better forecasts.00 Winter’s model 1.17).

M07_CHOP0203_06_SE_C07. However. Good demand planning modules contain tools to perform what-if analysis regarding the impact of potential changes in prices on demand. Even with all these sophisticated tools. other purchases.indd Page 203 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 203 Besides providing a rich library of forecasting methodologies. in some cases. As a result. and replenishment are discussed in greater detail in Chapter 10). the supply chain benefits of collaboration are often an order of magnitude greater than the cost (collaborative planning. This link is discussed in greater detail in Chapter 9 under sales and operations planning. The value of data depends on where one sits in the supply chain. A purchase of “cocoa-butter lotion. A good IT system should help track historical forecast errors so they can be incorporated into future decisions. is that most forecasts do not even account for all the information available across the different functions of a firm. However.com/ PSNTTVSWFZT'44GTTGSIUNM 7. The manufacturer finds aggregate demand data to be quite valuable. sometimes it is better to rely on human intuition in forecasting. spur demand. a good demand planning module should provide support in helping select the right forecasting model for the given demand pattern. however. Keeping the data shared to what is truly required decreases investment in IT and improves the chances of successful collaboration. These tools help analyze the impact of promotions on demand and can be used to determine the extent and timing of promotions. these modules facilitate the shaping of demand. can significantly improve decision making. a purse large enough to double as a diaper bag. A retailer finds point-of-sale data to be quite valuable in measuring the performance of its stores. Target predicted that women were pregnant based on other products they were purchasing.g. Share only the data that truly provide value. February 16.45PEBZGPSFDBTUJOH software survey. An important development is the use of demand correlated data (e. A well-structured forecast.10 FORECASTING IN PRACTICE Collaborate in building forecasts. This has become particularly important as the available library of forecasting method- ologies has grown. price. but remember that they cannot assess some of the more qualitative aspects about future demand that you may be able to do on your own. The reality today. Keep in mind that none of these tools is foolproof. weather. Forecasts are virtually always inaccu- rate. in turn. One of the pitfalls of these IT tools is relying on them too much. 2012. with marginally more value coming from detailed point-of-sale data. . along with a measure of error. which eliminates the human element in forecasting. and a discussion of each vendor is available at http://www. Collaboration with one’s supply chain partners can often create a much more accurate forecast. "EFUBJMFEMJTUPGGPSFDBTUJOHTPGUXBSFWFOEPSTJTSFQPSUFEJOUIF03. In a well- publicized case. a manufacturer selling to a distributor that.2 Target then used this information to send suitable coupons to entice these women or UIFJSIVTCBOETUPWJTJU5BSHFUBOEQVSDIBTFCBCZSFMBUFEQSPEVDUT4PQIJTUJDBUFETZTUFNTTVDIBT this can be used to not only improve forecast accuracy but also identify suitable marketing oppor- tunities to spur future demand. As the name demand planning suggests. firms should aim to put a sales and operations planning process in place (discussed in Chapter 9) that brings together the sales and operations functions when planning. zinc and magnesium supplements and a bright blue rug” was a strong predictor of the woman’s pregnacy. Use the forecasts and the value they deliver.lionhrtpub. It takes an investment of time and effort to build the relationships with one’s partners to begin sharing information and creating collaborative forecasts. 2 $IBSMFT%VIJHH i)PX$PNQBOJFT-FBSO:PVS4FDSFUT uNew York Times. forecasting. sells to retailers does not need all the point-of-sale detail. social data) to improve forecast accuracy or.

Forecast error measures the random component of demand. 7. Holt’s model. Identify the components of a demand forecast. Adaptive methods include moving averages. Level measures the current deseasonalized EFNBOE5SFOENFBTVSFTUIFDVSSFOUSBUFPGHSPXUIPSEFDMJOFJOEFNBOE4FBTPOBMJUZJOEJDBUFT predictable seasonal fluctuations in demand. Enterprises have always forecast demand and used it to make decisions. Moving averages and simple exponential smoothing are best used when demand displays neither trend nor seasonality. What role does forecasting play in the supply chain of a mail. How could Apple use collaborative forecasting with its sup. competitor actions. supply chain stages farther from demand will likely have poor forecasts that will lead to supply chain inefficiencies and a lack of responsiveness. companies make the mistake of looking at historical sales and assuming that this is what the historical demand was.4& . The systematic component consists of level. Often. Fore- casting is a key driver of virtually every design and planning decision made in both an enterprise and a supply chain. Discussion Questions 1. is to create collaborative forecasts for an entire supply chain and use these as the basis for decisions. What role does forecasting play in the supply chain of a 6. Analyze demand forecasts to estimate forecast error. A relatively recent phenomenon. Demand consists of a systematic and a random component. trend. The random component measures fluctuations in demand from the expected value. Give examples of products that display seasonality of build-to-order server manufacturer such as Dell? demand."1&BSFVTFEUPFTUJNBUFUIFTJ[FPGUIFGPSFDBTUFSSPS5IFCJBTBOE54BSFVTFEUPFTUJNBUFJG the forecast consistently over. 8IBUJOGPSNBUJPOEPUIFCJBTBOE54QSPWJEFUPBNBOBHFS  forecast historical demand without any forecast error? How can the manager use this information? . 7. What systematic and random components would you expect 9.L. The systematic component measures the expected value of demand. Bean? 8. Forecast demand in a supply chain given historical demand data using time-series methodologies. pricing. Without collabo- ration."% BOE. Time-series methods for forecasting are categorized as static or adaptive. Understand the role of forecasting for both an enterprise and a supply chain.M07_CHOP0203_06_SE_C07. This measure is important because it reveals how inaccurate a GPSFDBTUJTMJLFMZUPCFBOEXIBUDPOUJOHFODJFTBGJSNNBZIBWFUPQMBOGPS5IF."1&QSPWJEFUP in demand for chocolates? a manager? How can the manager use this information? 5. Holt’s model is best when demand displays a trend but no seasonality.11 SUMMARY OF LEARNING OBJECTIVES 1."% BOE .4& . Collaborative forecasting greatly increases the accuracy of forecasts and allows the supply chain to maximize its performance. To get true demand. the estimates of parameters and demand patterns are not updated as new demand JTPCTFSWFE4UBUJDNFUIPETJODMVEFSFHSFTTJPO*OBEBQUJWFNFUIPET UIFFTUJNBUFTBSFVQEBUFE each time a new demand is observed. and promotions. ing year? order firm such as L. What is the problem if a manager uses last year’s sales data pliers to improve its supply chain? instead of last year’s demand to forecast demand for the com- 3.or underforecasts or if demand has deviated significantly from historical norms.indd Page 204 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 204 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO Be sure to distinguish between demand and sales. 8IBUJOGPSNBUJPOEPUIF. 3. adjustments need to be made for unmet demand due to stockouts. however. Winter’s model is appropriate when demand displays both trend and seasonality. 4. and seasonality. however. Failure to do so results in forecasts that do not rep- resent the current reality. How do static and adaptive forecasting methods differ? 4. Why should a manager be suspicious if a forecaster claims to 10. simple expo- nential smoothing. 2. 2. In static methods. and Winter’s model.

000 8.000 February 3.000 August 6.000 98.000 March 3.000 16.000 20. Weekly demand figures at Hot Pizza are as follows:   'PSFDBTURVBSUFSMZEFNBOEGPSZFBSVTJOHTJNQMFFYQPOFO- tial smoothing with a = 0.000 5.000 14.000 15.000 July 7.000 5.M07_CHOP0203_06_SE_C07.indd Page 205 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 205 Exercises 1.000 3.1 as well as Holt’s model with a = 0.1&WBMVBUFUIF. Quarterly demand for flowers at a wholesaler are as shown.000 10."1& .000 2.000 June 6.000 October 12."% .000 April 3. Which of the two methods do you Week Demand ($) prefer? Why? 1 108 Year Quarter Demand ($000) 2 116 1 I 98 3 118 II 106 4 124 III 109 5 96 IV 133 6 119 2 I 130 7 96 II 116 8 102 III 133 9 112 IV 116 10 102 3 I 138 11 92 II 130 12 91 III 147 IV 141   &TUJNBUFEFNBOEGPSUIFOFYUXFFLTVTJOHBXFFLNPW.000 6.000 8. Which of the two methods do you prefer? III 165 Why? IV 173 3.000 20.4&&WBMVBUFUIFRVBMJUZPGUIFGPSFDBTU TABLE 7-3 Monthly Demand for ABC Corporation Sales Year 1 Year 2 Year 3 Year 4 Year 5 January 2. .000 10.000 3.000 7.000 2.000 10.000 20.1 and b = 0.000 6. Consider monthly demand for the ABC Corporation.000 8.000 5.000 5.000 113."% .000 8.000 10.000 7.4& CJBT BOE54 II 142 in each case.000 September 10.000 12.000 15.000 115.000 May 4.000 2.000 8.000 22.000 December 8.000 4.000 2.000 3.000 5. as VTJOHUIFTUBUJDNFUIPEGPSGPSFDBTUJOH&WBMVBUFUIFCJBT 54  shown in Table 7-3.000 7.000 89.000 3. Forecast the monthly demand for Year 6 .000 5.000 4.000 5.000 2.000 4.000 16."1& BOE.000 November 14.000 12.000 Total 78. 4 I 144 ing average as well as simple exponential smoothing with a = 0.000 18.1.000 12.000 16.000 3.000 4.

D(SBX)JMM  +VOF m $IBNCFST +PIO$ 4BUJOEFS. two methods do you prefer? Why? mance of simple exponential smoothing with a = 0. David M. Forecast the monthly demand for Year 6 ing model."1& BOE.. and Robert G. Statistical Forecasting for Inventory Control. Oakland.0 and b = 0. and Winter’s model.+FOLJOTTime Series Analysis: -FXBOEPXTLJ +/FXUPO &1BS[FO BOE38JOLMFSi5IF Forecasting and Control.167 10 445 12 2.4& CJBT BOE54 nential smoothing with a = 0. 3d ed. "DDVSBDZPG&YUSBQPMBUJPO 5JNF4FSJFT .BLSJEBLJT 4 ""OEFSTFO 3$BSCPOF 3'JMEFT .191 Estimate demand for the next four weeks using a five-week moving average. Bruce L.. Monthly demand at A&D Electronics for flat-screen TVs are nation of smoothing constants do you prefer? Why? as follows: 8. Forecasting and %FNBOE4VEEFOMZ$IBOHFTuJournal of Operations Manage- Time Series: An Applied Approach. Belmont.VMMJDL BOE%POBME%4NJUI . Business Review +VMZm"VHVTU m 4BGGP 1BVMi4JY3VMFTPG&GGFDUJWF'PSFDBTUJOHuHarvard Busi- Forecasting with Regression Analysis.000 Week Demand (units) 2 1. as well as simple exponential smoothing Estimate demand for the next two weeks using simple expo. “Is Forecasting a Waste of Time?” Supply Forecasters Worth Listening To?” Harvard Business Review Chain Management Review +VMZm"VHVTU m 4FQUFNCFSm0DUPCFS m . CA: ment 0DUPCFS m Duxbury.3 and Holt’s model with in each case. For the Hot Pizza data in Exercise 2."1& . compare the perfor. Michael.” Harvard Methods for Management. and b = 0. and Richard T.445 3 557 5 1.558 4 498 6 1. Using the A&D Electronics data in Exercise 6.271 2 510 4 1. For Holt’s model. Robert G. Weekly demand for dry pasta at a supermarket chain is as follows: Month Demand (units) 1 1.1."% .9. level at Period 0 to be L0 =BOEUIFUSFOEJO1FSJPEUPCF 54 . New York: Wiley. “Manager’s Guide Today +VOF m4PGUXBSFTVSWFZBWBJMBCMFBUIUUQ to Forecasting. Murdick.4& CJBT BOE54JOFBDIDBTF8IJDIPGUIF 5. simple exponential smoothing. 1993.)JCPO 3 #PY (FPSHF&1 BOE(XJMZN..724 6 444 8 1.M07_CHOP0203_06_SE_C07. use model. .2&WBMVBUFUIF. Which of the two methods do you prefer? Why? Bibliography #FSOTUFJO  1FUFS -  BOE 5IFPEPSF ) 4JMCFSU i"SF &DPOPNJD Gilliland. evaluate the bias. use the level at Period 0 to be L0 =  UIF using moving average.1.indd Page 206 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 206 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 4.113 1 517 3 1..864 8 441 10 2. For the simple exponential smooth- shown in Table 7-3. Cambridge. MA: Harvard ness Review +VMZ"VHVTU m #VTJOFTT4DIPPM/PUFmm  :VSLJFXJD[  +BDL i'PSFDBTUJOH "O 6QXBSE 5SFOE u ORMS Georgoff."1& . Consider monthly demand for the ABC Corporation as a = 0. a Forecasting Competition. O’Connell.850 7 526 9 1.076 9 541 11 2.D$MBJO  +PIO 0 i3FTUBSUJOH B 'PSFDBTUJOH 4ZTUFN 8IFO Bowerman. CA: Holden-Day."% .BLSJEBLJT  4QZSPT  BOE 4UFWFO $8IFFMXSJHIU Forecasting “How to Choose the Right Forecasting Technique.4&8IJDIGPSFDBTUJOHNFUIPEEP T0 = 109 (both are obtained through regression). repeat Holt’s a = 0."% . Which combi- 6. Evaluate the you prefer? Why? . In each case.648 5 498 7 1. 1989. Compare the performance Which of the two smoothing constants do you prefer? of Holt’s model with a = 0. 1976.FUIPET3FTVMUTPG Brown. What difference in forecasts do you observe? model with a = 0.0 and b = 0. with a = 0. Holt’s average demand over the 12 months).” Harvard Business Review +BOVBSZm'FCSVBSZ XXXMJPOISUQVCDPNPSNTTVSWFZT'44GTTGSIUNM 1986): 2–9.1 and 7. .” Journal of Forecasting (April– /FX:PSL.

As a result. as managers could demand for each of the two types of containers (clear use this more accurate forecast for their production BOE CMBDL  5IF UFBN NPEJGJFE 41$T TBMFT EBUB CZ QMBOOJOH *NQSPWFE GPSFDBTUT XPVME BMMPX 41$ UP accounting for lost sales to obtain true demand data. Making the food containers containers for years 6 to 8. 8JUIPVUUIFDVTUPNFSTJOWPMWFEJOUIJTUFBN 41$XPVME never have known this information. which converts it into a polystyrene sheet that BGUFSXIJDIJUJTFYQFDUFEUPQMBUFBV+VMJFNVTUTFMFDUUIF is wound into rolls. Demand for containers key customers.M07_CHOP0203_06_SE_C07. it wants to resin is unloaded from bulk rail containers or overland forecast quarterly demand for each of the two types of trailers into storage silos. Capacity on the extruders is not suffi- WBMVFUPUIJTQSPDFTT4IFXBTUPMECZIFSEJWJTJPONBO. TIFXPVMECFBTTJHOFEUPBUFBNDPOTJTUJOHPG41$T Over the past five years. forecast demand for years 6 rolls are loaded onto thermoforming presses. Caterers and grocery stores use meet demand effectively over the previous several the black plastic trays as packaging and serving trays. The rolls are either used immediately forecast error. and restaurants. Which method should she choose? Why? UP NBLF DPOUBJOFST PS QVU JOUP TUPSBHF 4FDPOE  UIF Using the method selected. SPC 41$UVSOTQPMZTUZSFOFSFTJOJOUPSFDZDMBCMFEJTQPTBCMF Forecasting containers for the food industry. the plant is forced to build inventory of PSBUJWF GPSFDBTU VTJOH EBUB GSPN CPUI 41$ BOE JUT each type of sheet in anticipation of future demand.indd Page 207 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780  $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO 207 CASE STUDY Specialty Packaging Corporation +VMJF8JMMJBNTIBEBMPUPOIFSNJOEXIFOTIFMFGUUIF form the sheet into containers and trim the containers DPOGFSFODF SPPN BU 4QFDJBMUZ 1BDLBHJOH $PSQPSBUJPO from the sheet. as the company did not keep track of lost orders. resin is conveyed to an demand is expected to continue to grow until year 8. the plastic packaging marketing vice president and staff members from their business has grown steadily. The plastic comes in two forms— appropriate forecasting method and estimate the likely clear and black. customers. sons. The two manufacturing steps are shown 41$ )FSEJWJTJPOBMNBOBHFSIBEJOGPSNFEIFSUIBU in Figure 7-11. is a two-step process. The As a first step in the team’s decision making. which to 8. Based on historical trends. improve delivery performance. Polystyrene is pur- chased as a commodity in the form of resin pellets. bak- TVQQMZDIBJOQFSGPSNBODF BT41$IBECFFOVOBCMFUP eries. whereas demand for black plastic containers 41$T DVTUPNFST BOE XPOEFSFE IPX TIF XPVME BEE peaks in the fall. extruder. The goal of this team was to improve made from clear plastic comes from grocery stores. First. ZFBST5IJT PGUFO MFGU 41$T DVTUPNFST TDSBNCMJOH UP Demand for clear plastic containers peaks in the summer NFFUOFXDMJFOUEFNBOET+VMJFIBEMJUUMFDPOUBDUXJUI months. cient to cover demand for sheets during the peak sea- ager that the team’s first task was to establish a collab. Step 1 Step 2 Thermo- Resin Extruder Roll forming Storage Storage Press FIGURE 7-11 Manufacturing Process at SPC . This forecast would serve as the basis for 5BCMFBOE'JHVSFEJTQMBZIJTUPSJDBMRVBSUFSMZ improving the firm’s performance.

250 3.742 II 2.695 IV 7.269 2.696 16.indd Page 208 19/09/14 8:36 AM f-445 /203/AW00176/9780133800203_CHOPRA/CHOPRA_SUPPLY_CHAN_MANAGEMENT06_SE_9780 208 $IBQUFS r %FNBOE'PSFDBTUJOHJOB4VQQMZ$IBJO TABLE 7-4 Quarterly Historical Demand for Clear and Black Plastic Containers Black Plastic Clear Plastic Year Quarter Demand (000 lb) Demand (000 lb) 1 I 2.186 III 4.680 III 3.485 5 I 5.315 5.459 5.M07_CHOP0203_06_SE_C07.843 8.253 2.737 7.591 III 4.448 IV 12.236 IV 13.143 8.097 3.200 II 1.648 7.640 IV 8.658 III 2.491 3.120 4.316 18000 Black Plastic Demand 16000 Clear Plastic Demand 14000 12000 Demand 10000 8000 6000 4000 2000 0 Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 FIGURE 7-12 Plot of Quarterly Demand for Clear and Black Plastic Containers .514 3.486 II 4.412 4.737 4 I 5.728 II 3.384 2 I 3.766 13.654 II 2.556 6.673 III 2.953 3 I 4.056 1.035 3.382 13.420 IV 7.