Forecasting In Operations Management

Fore­cast­ing is the process of esti­mat­ing future demand, sales, needs or trends from past data and present con­di­tions, so that a busi­ness can plan ahead. A col­lege pre­dict­ing next year's admis­sions and a bak­ery esti­mat­ing how many cakes it will sell this week­end are both fore­cast­ing.

In oper­a­tions man­age­ment almost every plan starts with a fore­cast. It tells a busi­ness how much to pro­duce, how many work­ers are needed, how much raw mate­r­ial to buy and how much stock to keep. A poor fore­cast leads to over­pro­duc­tion and waste, or to stock-outs and lost cus­tomers; a good one low­ers cost and uncer­tainty across the whole organ­i­sa­tion.

Mean­ing and def­i­n­i­tion

Fore­cast­ing is the process of esti­mat­ing future events, espe­cially demand, for plan­ning and deci­sion-mak­ing pur­poses. A fore­cast is a state­ment about the future; it is not a plan or a tar­get. The plan (how much to make) is decided after the fore­cast (how much cus­tomers are likely to buy).

Every fore­cast has three parts: what is being fore­cast (units, rupees, hours), for which period (the fore­cast hori­zon), and how accu­rate it is likely to be.

Why fore­cast­ing is impor­tant

Fore­cast­ing sup­ports:

• pro­duc­tion plan­ning: decid­ing out­put lev­els;
• inven­tory con­trol: decid­ing how much stock to hold and when to reorder;
• man­power plan­ning: hir­ing, shifts and over­time;
• bud­get­ing: esti­mat­ing rev­enue and costs;
• capac­ity plan­ning: decid­ing whether to add machines, space or branches;
• reduc­ing short­age and wastage.

With­out fore­cast­ing, goods may be over­pro­duced, stock may run out, costs may increase and cus­tomer demand may not be met.

Types of fore­cast­ing by time hori­zon

Fore­casts are com­monly grouped by how far ahead they look. Many text­books add a medium-term cat­e­gory between the two clas­sic types.

1. Short-term fore­cast­ing

Cov­ers days, weeks or a few months (usu­ally up to about three months). It is used for daily pro­duc­tion, sched­ul­ing work­ers, order­ing raw mate­ri­als and set­ting job assign­ments. Exam­ple: a restau­rant pre­dicts week­end cus­tomer demand.

2. Medium-term fore­cast­ing

Cov­ers roughly three months to two years. It is used for sales plan­ning, pro­duc­tion plan­ning and bud­get­ing, and aggre­gate plan­ning of work­force and inven­tory. Exam­ple: a gar­ment maker esti­mates demand for the com­ing fes­ti­val and win­ter sea­sons.

3. Long-term fore­cast­ing

Cov­ers sev­eral years. It is used for expan­sion plans, new prod­uct deci­sions, invest­ment plan­ning, capac­ity increases and facil­ity loca­tion. Exam­ple: a school esti­mates stu­dent strength for the next three years before build­ing new class­rooms.

Short-term fore­casts are gen­er­ally more accu­rate than long-term ones, because fewer things can change in a short period.

Demand pat­terns

Before choos­ing a method, look at how past demand behaves. A demand series can con­tain:

• Level (aver­age): demand fluc­tu­ates around a con­stant mean.
• Trend: a steady rise or fall over time.
• Sea­son­al­ity: a repeat­ing pat­tern within a year (ice cream in sum­mer, sweets at Diwali).
• Cycles: longer up-and-down move­ments linked to the econ­omy.
• Ran­dom vari­a­tion: unex­plained noise that no method can pre­dict.

Meth­ods of fore­cast­ing

Meth­ods are divided into two broad groups: qual­i­ta­tive and quan­ti­ta­tive.

A. Qual­i­ta­tive meth­ods

These rely on judge­ment, opin­ion, expe­ri­ence and mar­ket knowl­edge. They are used when past data is lim­ited, for exam­ple for a new prod­uct, or when con­di­tions are chang­ing fast.

• Jury of exec­u­tive opin­ion: senior man­agers pool their views.
• Sales force com­pos­ite: each sales­per­son esti­mates sales in their ter­ri­tory, and the esti­mates are added up.
• Mar­ket sur­vey (con­sumer sur­vey): cus­tomers are asked about their buy­ing inten­tions.
• Del­phi method: a panel of experts answers ques­tion­naires in sev­eral anony­mous rounds, with feed­back between rounds, until their views con­verge.
• Expert opin­ion: spe­cial­ists out­side the firm are con­sulted.

Exam­ple: a com­pany asks experts whether a new course will have enough demand.

B. Quan­ti­ta­tive meth­ods

These use numer­i­cal data and math­e­mat­i­cal tech­niques. They are used when reli­able past data is avail­able.

Time series meth­ods assume that the future will fol­low the pat­tern of the past:

• Naive method: next peri­od's fore­cast equals this peri­od's actual demand.
• Sim­ple mov­ing aver­age: the aver­age of the last $n$ peri­ods.
• Weighted mov­ing aver­age: recent peri­ods get larger weights.
• Expo­nen­tial smooth­ing: the new fore­cast is the old fore­cast plus a frac­tion of the last error.
• Trend pro­jec­tion: a straight line fit­ted to past data by least squares and extended for­ward.

Causal (asso­cia­tive) meth­ods link demand to other vari­ables, such as price, adver­tis­ing or pop­u­la­tion, usu­ally through lin­ear regres­sion.

Exam­ple: a store stud­ies the last 12 months' sales to esti­mate next mon­th's sales.

Key for­mu­las

Sim­ple mov­ing aver­age over $n$ peri­ods:

$$\displaystyle F_{t+1} = \frac{A_t + A_{t-1} + \cdots + A_{t-n+1}}{n}$$

Weighted mov­ing aver­age, with weights $w_i$ that add up to 1:

$$\displaystyle F_{t+1} = \sum w_i A_i$$

Sim­ple expo­nen­tial smooth­ing, with smooth­ing con­stant $\alpha$ between 0 and 1:

$$F_{t+1} = F_t + \alpha\,(A_t - F_t)$$

Mean absolute devi­a­tion (MAD), a com­mon accu­racy mea­sure:

$$\displaystyle MAD = \frac{\sum |A_t - F_t|}{n}$$

Here $A_t$ is actual demand and $F_t$ is the fore­cast for period $t$. A larger $\alpha$ makes the fore­cast respond faster to recent changes; a smaller $\alpha$ gives a smoother fore­cast.

Worked exam­ple: mov­ing aver­age and expo­nen­tial smooth­ing

Sup­pose a bak­ery recorded the fol­low­ing monthly cake sales and wants a fore­cast for Month 9.

Step 1: three-month mov­ing aver­age

The first fore­cast pos­si­ble is for Month 4, using Months 1 to 3:

$$\displaystyle F_4 = \frac{120 + 130 + 125}{3} = 125.00$$

Mov­ing the win­dow for­ward one month at a time:

• $F_5 = (130 + 125 + 140)/3 = 131.67$
• $F_6 = (125 + 140 + 150)/3 = 138.33$
• $F_7 = (140 + 150 + 145)/3 = 145.00$
• $F_8 = (150 + 145 + 160)/3 = 151.67$
• $F_9 = (145 + 160 + 165)/3 = 156.67$

The mov­ing aver­age fore­cast for Month 9 is about 157 cakes.

Step 2: weighted mov­ing aver­age

If the bak­ery gives weights of 0.5, 0.3 and 0.2 to the lat­est, sec­ond-lat­est and third-lat­est months:

$$F_9 = 0.5(165) + 0.3(160) + 0.2(145) = 82.5 + 48 + 29 = 159.5$$

The weighted fore­cast (about 160 cakes) is higher because it leans more on the ris­ing recent months.

Step 3: expo­nen­tial smooth­ing with alpha = 0.3

Assume the fore­cast for Month 1 was 120. Then:

• $F_2 = 120 + 0.3(120 - 120) = 120.00$
• $F_3 = 120 + 0.3(130 - 120) = 123.00$
• $F_4 = 123 + 0.3(125 - 123) = 123.60$
• $F_5 = 123.60 + 0.3(140 - 123.60) = 128.52$
• $F_6 = 128.52 + 0.3(150 - 128.52) = 134.96$
• $F_7 = 134.96 + 0.3(145 - 134.96) = 137.97$
• $F_8 = 137.97 + 0.3(160 - 137.97) = 144.58$
• $F_9 = 144.58 + 0.3(165 - 144.58) = 150.71$

The expo­nen­tial smooth­ing fore­cast for Month 9 is about 151 cakes.

Step 4: com­pare accu­racy using MAD (Months 4 to 8)

$$\displaystyle MAD_{MA} = \frac{68.33}{5} = 13.67 \qquad MAD_{ES} = \frac{90.37}{5} = 18.07$$

Inter­pre­ta­tion: the mov­ing aver­age has the lower MAD, so it was more accu­rate for this data. Both meth­ods lag behind a ris­ing trend, and expo­nen­tial smooth­ing with a small $\alpha$ of 0.3 lags the most. For data with a clear trend, a larger $\alpha$ or a trend-adjusted method would usu­ally do bet­ter.

Steps in the fore­cast­ing process

1. Decide the pur­pose of the fore­cast and how it will be used.
2. Select the items to be fore­cast and the time hori­zon.
3. Col­lect and clean the rel­e­vant data.
4. Exam­ine the data for level, trend and sea­sonal pat­terns.
5. Select a suit­able fore­cast­ing method.
6. Pre­pare the fore­cast.
7. Mon­i­tor accu­racy by com­par­ing fore­casts with actual results, and revise the method when errors grow.

Qual­i­ties of a good fore­cast

A good fore­cast should be:

• sim­ple and easy to under­stand;
• timely, avail­able before deci­sions must be made;
• rea­son­ably accu­rate, with its likely error stated;
• eco­nom­i­cal, cost­ing less than the ben­e­fit it gives;
• reli­able and con­sis­tent over time;
• expressed in mean­ing­ful units (cakes, hours, rupees).

Fac­tors affect­ing fore­cast­ing

• mar­ket demand and cus­tomer pref­er­ences;
• com­pe­ti­tion and com­peti­tors' pric­ing;
• sea­sonal changes and fes­ti­vals;
• gov­ern­ment pol­icy and taxes;
• gen­eral eco­nomic con­di­tions;
• tech­nol­ogy changes;
• the fir­m's own pro­mo­tions and prices.

A real-life illus­tra­tion: an ice cream shop

In sum­mer, demand for ice cream is high; in the rainy sea­son it is lower. The shop fore­casts demand for each sea­son and uses that fore­cast to decide how much stock to keep, how many work­ers are needed and how much milk and sugar to buy. Recog­nis­ing the sea­sonal pat­tern is what makes the fore­cast use­ful.

Advan­tages and lim­i­ta­tions

Fore­cast­ing is help­ful, but it is never per­fect. Good prac­tice is to state the expected error and to keep mon­i­tor­ing.

Exam answer for­mats

5-mark answer. Fore­cast­ing is the process of esti­mat­ing future demand or future events for plan­ning pur­poses. It is impor­tant in oper­a­tions man­age­ment because it helps in pro­duc­tion plan­ning, inven­tory con­trol, man­power plan­ning and bud­get­ing. Fore­casts may be short-term, medium-term or long-term. The main meth­ods are qual­i­ta­tive (such as the Del­phi method and mar­ket sur­veys) and quan­ti­ta­tive (such as mov­ing aver­ages and expo­nen­tial smooth­ing). Good fore­cast­ing reduces uncer­tainty and improves deci­sion-mak­ing.

2-mark answer. Fore­cast­ing is the process of pre­dict­ing future demand or future events for plan­ning and deci­sion-mak­ing.

Mem­ory line. Fore­cast­ing means look­ing ahead so that plan­ning is bet­ter.

Key terms

Fore­cast

An esti­mate of a future value, such as demand, for a stated period.

Fore­cast hori­zon

The length of time into the future that a fore­cast cov­ers.

Time series

A set of obser­va­tions recorded at reg­u­lar time inter­vals.

Mov­ing aver­age

The aver­age of the most recent $n$ peri­ods, used as the next fore­cast.

Expo­nen­tial smooth­ing

A method that updates the last fore­cast by a frac­tion $\alpha$ of the last error.

Smooth­ing con­stant

The weight $\alpha$, between 0 and 1, given to the most recent error.

Del­phi method

A qual­i­ta­tive method using anony­mous, repeated rounds of expert opin­ion.

Mean absolute devi­a­tion

The aver­age of the absolute fore­cast errors, used to mea­sure accu­racy.

Sea­son­al­ity

A demand pat­tern that repeats at reg­u­lar inter­vals within a year.

Com­mon ques­tions

What is the dif­fer­ence between qual­i­ta­tive and quan­ti­ta­tive fore­cast­ing?

Qual­i­ta­tive meth­ods rely on judge­ment and opin­ion and are used when data is scarce, such as for a new prod­uct. Quan­ti­ta­tive meth­ods use past numer­i­cal data and math­e­mat­i­cal mod­els and are used when reli­able his­tory exists.

Which value of alpha should be used in expo­nen­tial smooth­ing?

A higher $\alpha$ (for exam­ple 0.5 or more) reacts quickly to recent changes but also to ran­dom noise. A lower $\alpha$ (0.1 to 0.3) gives a smoother, more sta­ble fore­cast. The best value is usu­ally the one that gives the low­est error, such as MAD, on past data.

Why do mov­ing aver­ages lag behind a trend?

A mov­ing aver­age gives equal weight to older peri­ods, so when demand is ris­ing steadily the aver­age is always pulled down by the older, lower val­ues. Weighted aver­ages and trend meth­ods reduce this lag.

What is the dif­fer­ence between a fore­cast and a plan?

A fore­cast esti­mates what cus­tomers are likely to demand. A plan decides what the firm will actu­ally do, such as how much to pro­duce, tak­ing the fore­cast, capac­ity and costs into account.

How is fore­cast accu­racy mea­sured?

By com­par­ing fore­casts with actual results. Com­mon mea­sures are mean absolute devi­a­tion (MAD), mean squared error (MSE) and mean absolute per­cent­age error (MAPE). A lower value means a more accu­rate method.

Ref­er­ences

1. Heizer, J., Ren­der, B. and Mun­son, C. Oper­a­tions Man­age­ment: Sus­tain­abil­ity and Sup­ply Chain Man­age­ment. Pear­son.
2. Steven­son, W. J. Oper­a­tions Man­age­ment. McGraw-Hill Edu­ca­tion.
3. Chase, R. B. and Jacobs, F. R. Oper­a­tions and Sup­ply Chain Man­age­ment. McGraw-Hill Edu­ca­tion.
4. Brown, R. G. (1959) Sta­tis­ti­cal Fore­cast­ing for Inven­tory Con­trol. McGraw-Hill.
5. Chary, S. N. Pro­duc­tion and Oper­a­tions Man­age­ment. McGraw-Hill Edu­ca­tion (India).

Related read­ing

• Types of Pro­duc­tion Sys­tems in Oper­a­tions Man­age­ment
• Capac­ity Plan­ning in Oper­a­tions Man­age­ment
• Loca­tion Analy­sis in Oper­a­tions Man­age­ment