Академический Документы
Профессиональный Документы
Культура Документы
FORECASTING FUNDAMENTALS Forecast: A prediction, projection, or estimate of some future activity, event, or occurrence. Types of Forecasts - Economic forecasts o Predict a variety of economic indicators, like money supply, inflation rates, interest rates, etc. - Technological forecasts o Predict rates of technological progress and innovation. - Demand forecasts o Predict the future demand for a companys products or services. ince virtually all the operations management decisions !in "oth the strategic category and the tactical category# re$uire as input a good estimate of future demand, this is the type of forecasting that is emphasi%ed in our te&t"ook and in this course.
TYPES OF FORECASTING METHODS Q a!"tat"#e $et%o&s: These types of forecasting methods are "ased on judgments, opinions, intuition, emotions, or personal e&periences and are su"jective in nature. They do not rely on any rigorous mathematical computations. Q a't"tat"#e $et%o&s: These types of forecasting methods are "ased on mathematical !$uantitative# models, and are o"jective in nature. They rely heavily on mathematical computations.
Q a!"tat"#e Met%o&s
E)ec t"#e Op"'"o' Approach in 'hich a group of managers meet and collectively develop a forecast
Mar*et S r#ey Approach that uses intervie's and surveys to judge preferences of customer and to assess demand
Sa!es Force Co$pos"te Approach in 'hich each salesperson estimates sales in his or her region
De!p%" Met%o& Approach in 'hich consensus agreement is reached among a group of e&perts
Q a't"tat"#e Met%o&s
T"$e+Ser"es Mo&e!s Time series models look at past patterns of data and attempt to predict the future "ased upon the underlying patterns contained 'ithin those data.
Assoc"at"#e Mo&e!s Associative models !often called causal models# assume that the varia"le "eing forecasted is related to other varia"les in the environment. They try to project "ased upon those associations.
Mo&e! (a)ve
*ses an average of all past data as a forecast *ses an average of a specified num"er of the most recent o"servations, 'ith each o"servation receiving the same emphasis !'eight# *ses an average of a specified num"er of the most recent o"servations, 'ith each o"servation receiving a different emphasis !'eight# A 'eighted average procedure 'ith 'eights declining e&ponentially as data "ecome older Techni$ue that uses the least s$uares method to fit a straight line to the data A mechanism for adjusting the forecast to accommodate any seasonal patterns inherent in the data
E&ponential moothing
Trend Projection
easonal -nde&es
Time
Time
/ear 0
/ear 1
/ear 2
Time
Time
Time
-n this illustration 'e assume that each year !"eginning 'ith year 1# 'e made a forecast, then 'aited to see 'hat demand unfolded during the year. ,e then made a forecast for the su"se$uent year, and so on right through to the forecast for year 9. Actual Demand !At# 208 26: 27: 50: 5:8 56:
/ear 0 1 2 5 : 6 9
(otes There 'as no prior demand data on 'hich to "ase a forecast for period 0 3rom this point for'ard, these forecasts 'ere made on a year-"y-year "asis.
/ear 0
(otes This forecast 'as a guess at the "eginning. 3rom this point for'ard, these forecasts 'ere made on a year-"y-year "asis using a simple average approach.
1 2 5 : 6 9
/ear 0 1
(otes This forecast 'as a guess at the "eginning. This forecast 'as made using a na)ve approach. 3rom this point for'ard, these forecasts 'ere made on a year-"y-year "asis using a 1-yr moving average approach.
2 5 : 6 9
/ear 0 1 2
(otes This forecast 'as a guess at the "eginning. This forecast 'as made using a na)ve approach. This forecast 'as made using a na)ve approach. 3rom this point for'ard, these forecasts 'ere made on a year-"y-year "asis using a 2-yr moving average approach.
5 : 6 9
-n this illustration 'e assume that a 2-year 'eighted moving average is "eing used. ,e 'ill also assume that, in the a"sence of data at startup, 'e made a guess for the year 0 forecast !288#. Then, after year 0 elapsed, 'e used a na)ve method to make a forecast for year 1 !208# and year 2 !26:#. >eyond that point 'e had sufficient data to let our 2-year 'eighted moving average forecasts unfold throughout the years. The 'eights that 'ere to "e used are as follo's? +ost recent year, .:@ year prior to that, .2@ year prior to that, .1 Actual Demand !At# 208 26: 27: 50: 5:8 56:
/ear 0 1 2 5 : 6 9
(otes This forecast 'as a guess at the "eginning. This forecast 'as made using a na)ve approach. This forecast 'as made using a na)ve approach. 3rom this point for'ard, these forecasts 'ere made on a year-"y-year "asis using a 2-yr 'td. moving avg. approach.
,here is a smoothing coefficient 'hose value is "et'een 8 and 0. The e&ponential smoothing method only re$uires that you dig up t'o pieces of data to apply it !the most recent actual demand and the most recent forecast#. An attractive feature of this method is that forecasts made 'ith this model 'ill include a portion of every piece of historical demand. 3urthermore, there 'ill "e different 'eights placed on these historical demand values, 'ith older data receiving lo'er 'eights. At first glance this may not "e o"vious, ho'ever, this property is illustrated on the follo'ing page.
DEMONSTRATION: E3PONENTIAL SMOOTHING INCLUDES ALL PAST DATA (ote: the &athe&atical &ani%!lations in this box are not soðing o! $o!l" e)er ha)e to "o $hen a%%l ing ex%onential s&oothing* +ll o! nee" to !se is e,!ation 1 on the %re)io!s %age* -his "e&onstration is to con)ince the s#e%tics that $hen !sing e,!ation 1. all historical "ata $ill be incl!"e" in the forecast. an" the ol"er the "ata. the lo$er the $eight a%%lie" to that "ata* To make a forecast for ne&t period, 'e 'ould use the user friendly alternate e$uation 0? 3t A At-0 < !0-#3t-0 !e$uation 0#
,hen 'e made the forecast for the current period !3t-0#, it 'as made in the follo'ing fashion? 3t-0 A At-1 < !0-#3t-1 !e$uation 1#
-f 'e su"stitute e$uation 1 into e$uation 0 'e get the follo'ing? 3t A At-0 < !0-#DAt-1 < !0-#3t-1E ,hich can "e cleaned up to the follo'ing? 3t A At-0 < !0-#At-1 < !0-#13t-1 !e$uation 2# ,e could continue to play that game "y recogni%ing that 3t-1 A At-2 < !0-#3t-2 !e$uation 5# -f 'e su"stitute e$uation 5 into e$uation 2 'e get the follo'ing? 3t A At-0 < !0-#At-1 < !0-#1DAt-2 < !0-#3t-2E ,hich can "e cleaned up to the follo'ing? 3t A At-0 < !0-#At-1 < !0-#1At-2 < !0-#23t-2 -f you keep playing that game, you should recogni%e that 3t A At-0 < !0-#At-1 < !0-#1At-2 < !0-#2At-5 < !0-#5At-: < !0-#:At-6 4445 As you raise those decimal 'eights to higher and higher po'ers, the values get smaller and smaller.
/ear 0
26:
280
27:
289.5
50:
206.06
5:8
216.855
56:
22;.5276
2:0.87:65
/ear 0
26:
281
27:
205.6
50:
228.6;
5:8
259.:55
56:
26;.82:1
2;9.51;06
/ear 0
26:
285
27:
21;.5
50:
2::.85
5:8
297.815
56:
589.5055
528.55;65
TREND PRO:ECTION
Tre'& pro;ect"o' $et%o&: This method is a version of the linear regression techni$ue. -t attempts to dra' a straight line through the historical data points in a fashion that comes as close to the points as possi"le. !Technically, the approach attempts to reduce the vertical deviations of the points from the trend line, and does this "y minimi%ing the s$uared values of the deviations of the points from the line#. *ltimately, the statistical formulas compute a slope for the trend line !"# and the point 'here the line crosses the y-a&is !a#. This results in the straight line e$uation / A a < "F ,here F represents the values on the hori%ontal a&is !time#, and / represents the values on the vertical a&is !demand#. 3or the demonstration data, computations for " and a reveal the follo'ing /(0-1: 2 $ill not re,!ire o! to &a#e the statistical calc!lations for b an" a3 these $o!l" be gi)en to o!* 4o$e)er. o! "o nee" to #no$ $hat to "o $ith these )al!es $hen gi)en to o!*' " A 28 a A 17: / A 17: < 28F This e$uation can "e used to forecast for any year into the future. 3or e&le? /ear 9? 3orecast A 17: < 28!9# A :8: /ear ;? 3orecast A 17: < 28!;# A :2: /ear 7? 3orecast A 17: < 28!7# A :6: /ear 08? 3orecast A 17: < 28!08# A :7:
+echanisms for dealing 'ith seasonality are illustrated over the ne&t several pages.
75 002 008 028 00; 058 015 056 02: 060 027 061 !75<008< !002<028< 00;<015< 058<056< 02:<027# 060<061# L 6 A 018 L 6 A 051
(e&t, note that the total demand over these si& years of history 'as 1588 !i.e., 208 < 26: < 27: < 50: < 5:8 < 56:#, and if this total demand of 1588 had "een evenly spread over each of the 15 $uarters in this si& year period, the average $uarterly demand 'ould have "een 088 units. Another 'ay to look at this is the average of the $uarterly averages is 088 units, i.e. !;8 < 018 < 051 < :;#M5 A 088 units. >ut, the num"ers a"ove indicate that the demand 'asnt evenly distri"uted over each $uarter. -n 4uarter 0 the average demand 'as considera"ly "elo' 088 !it averaged ;8 in 4uarter 0#. -n 4uarters 1 and 2 the average demand 'as considera"ly a"ove 088 !'ith averages of 018 and 051, respectively#. 3inally, in 4uarter 5 the average demand 'as "elo' 088 !it averaged :; in 4uarter 5#. ,e can calculate a seasonal inde& for each $uarter "y dividing the average $uarterly demand "y the 088 that 'ould have occurred if all the demand had "een evenly distri"uted across the $uarters. This 'ould result in the follo'ing alternate seasonal inde& values? /ear easonal -nde& 40 ;8M088 A .;8 41 018M088 A 0.18 42 051M088 A 0.51 45 :;M088 A .:;
A $uick check of these alternate seasonal inde& values reveals that they average out to 0.8 !as they should#. !.;8 < 0.18 < 0.51 < .:;#M5 A 0.888
-f these annual forecasts 'ere evenly distri"uted over each year, the $uarterly forecasts 'ould look like the follo'ing? Annual 3orecast :8: :2: :6: :7:
/ear 9 ; 7 08
.o'ever, seasonality in the past demand suggests that these forecasts should not "e evenly distri"uted over each $uarter. ,e must take these even splits and multiply them "y the seasonal inde& ! .-.# values to get a more reasona"le set of $uarterly forecasts. The results of these calculations are sho'n "elo'. .-. /ear 9 ; 7 08 .;8 40 080.888 089.888 002.888 007.888 0.18 41 0:0.:88 068.:88 067.:88 09;.:88 0.51 42 097.19: 0;7.71: 188.:9: 100.11: .:; 45 92.11: 99.:9: ;0.71: ;6.19: Annual 3orecast :8: :2: :6: :7:
-f you check these final splits, you 'ill see that the sum of the $uarterly forecasts for a particular year 'ill e$ual the total annual forecast for that year !sometimes there might "e a slight rounding discrepancy#.
/ear 0 1 2 5 : 6
40 61 92 97 ;2 ;7 75
45 50 :1 :; 61 6: 98
To illustrate, - have used the linear trend line method on the $uarter 0 strip of data, 'hich 'ould result in the follo'ing trend line? / A :;.; < 6.8:90F 3or year 9, F A 9, so the resulting 40 forecast for year 9 'ould "e 080.188 ,e could do the same thing 'ith the 41, 42, and 45 strips of data. 3or each strip 'e 'ould compute the trend line e$uation and use it to project that $uarters year 9 demand. Those results are summari%ed here? 41 trend line? / A ;7.5 < ;.9517F@ /ear 9 41 forecast 'ould "e 0:8.688 42 trend line? / A 089.6 < 7.;1;6F@ /ear 9 42 forecast 'ould "e 096.588 45 trend line? / A 27.1 < :.2905F@ /ear 9 45 forecast 'ould "e 96.;88 Total forecast for year 9 A 080.188 < 0:8.688 < 096.588 < 96.;88 A :8:.888 These $uarterly forecasts are in the same "allpark as those made 'ith the seasonal inde& values earlier. They differ a "it, "ut 'e cannot say one is correct and one is incorrect. They are just slightly different predictions of 'hat is going to happen in the future. They do provide a total annual forecast that is e$ual to the trend projection forecast made for year 9. !Dont e&pect this to occur on every occasion, "ut since it corro"orates results o"tained 'ith a different method, it does give us confidence in the forecasts 'e have made.#
-f 'e attempted to perform a time series analysis on demand, the results 'ould not make much sense, for a $uick plot of demand vs. time suggests that there is no apparent pattern relationship here, as seen "elo'.
-f you plot the relationship "et'een demand and the num"er of construction permits, a pattern emerges that makes more sense. -t seems to indicate that demand for this product is lo'er 'hen fe'er construction permits are issued, and higher 'hen more permits are issued. Therefore, regression 'ill "e used to esta"lish a relationship "et'een the dependent varia"le !demand# and the independent varia"le !construction permits#.
The independent varia"le !F# is the num"er of construction permits. The dependent varia"le !/# is the demand for dry'all. Application of regression formulas yields the follo'ing forecasting model? / A 1:8 < 0:8F -f the company plans finds from pu"lic records that 2:8 construction permits have "een issued for the year 1801, then a reasona"le estimate of dry'all demand for 1801 'ould "e? / A 1:8 < 0:8!2:8# A 1:8 < :1,:88 A :1,9:8 !'hich means ne&t years forecasted demand is :1,9:8 sheets of dry'all#
.ypothetical 3orecasts Actual +ade ,ith Demand +ethod 0 /ear At 3t 0 208 20: 1 26: 29: 2 27: 278 5 50: 58: : 5:8 52: 6 56: 5;8 Accumulated 3orecast Errors +ean 3orecast Error, +3E
.ypothetical 3orecasts +ade ,ith +ethod 1 3t 298 5:: 28: :2: 278 25:
>ased on the accumulated forecast errors over time, the t'o methods look e$ually good. >ut, most o"servers 'ould judge that +ethod 0 is generating "etter forecasts than +ethod 1 !i.e., smaller misses#.
.ypothetical 3orecasting +ethod 0 Actual Demand /ear At 0 208 1 26: 2 27: 5 50: : 5:8 6 56: 3orecast 3t 20: 29: 278 58: 52: 5;8 3orecast Error At - 3t -: -08 : 08 0: -0: A"solute Deviation OAt - 3tO : 08 : 08 0: 0: 68 68M6A08
.ypothetical 3orecasting +ethod 1 3orecast 3t 298 5:: 28: :2: 278 25: 3orecast Error At - 3t -68 -78 78 -018 68 018 A"solute Deviation OAt - 3tO 68 78 78 018 68 018 :58 :58M6A78
The smaller misses of +ethod 0 has "een formali%ed 'ith the calculation of the +AD. +ethod 0 seems to have provided more accurate forecasts over this si& year hori%on, as evidenced "y its considera"ly smaller +AD.
.ypothetical 3orecasting +ethod 0 Actual Demand /ear At 0 208 1 26: 2 27: 5 50: : 5:8 6 56: 3orecast 3t 20: 29: 278 58: 52: 5;8 3orecast Error At - 3t -: -08 : 08 0: -0: $uared Error !At - 3t#1 1: 088 1: 088 11: 11: 988 988M6 A 006.69
.ypothetical 3orecasting +ethod 1 3orecast 3t 298 5:: 28: :2: 278 25: 3orecast Error At - 3t -68 -78 78 -018 68 018 $uared Error !At - 3t#1 2688 ;088 ;088 05588 2688 05588 :1188 :1188M6 A ;988
+ethod 0 seems to have provided more accurate forecasts over this si& year hori%on, as evidenced "y its considera"ly smaller + E. The 4uestion often arises as to 'hy one 'ould use the more cum"ersome + E 'hen the +AD calculations are a "it simpler !you dont have to s$uare the deviations#. +AD does have the advantage of simpler calculations. .o'ever, there is a "enefit to the + E method. ince this method s$uares the error term, large errors tend to "e magnified. Konse$uently, + E places a higher penalty on large errors. This can "e useful in situations 'here small forecast errors dont cause much of a pro"lem, "ut large errors can "e devastating.
.ypothetical 3orecasting +ethod 0 Actual Demand /ear At 0 208 1 26: 2 27: 5 50: : 5:8 6 56: 3orecast 3t 20: 29: 278 58: 52: 5;8 3orecast Error At - 3t -: -08 : 08 0: -0: A"solute Q Error
088OAt - 3tOMAt
.ypothetical 3orecasting +ethod 1 3orecast 3t 298 5:: 28: :2: 278 25: 3orecast Error At - 3t -68 -78 78 -018 68 018 A"solute Q Error
088OAt - 3tOMAt
+ethod 0seems to have provided more accurate forecasts over this si& year hori%on, as evidenced "y the fact that the percentages "y 'hich the forecasts miss the actual demand are smaller 'ith +ethod 0 !i.e., smaller +APE#.
/ear 0 1 2 5 : 6
A"s. Q Error OAt-3tOMAt 6.5:Q :.5;Q :.86Q 5.;1Q 5.55Q 5.28Q 28.::Q +APEA 28.::M6 :.87Q
/ou can o"serve that for each of these forecasting methods, the same +3E resulted and the same +AD resulted. ,ith these t'o measures, 'e 'ould have no "asis for claiming that one of these forecasting methods 'as more accurate than the other. ,ith several measures of accuracy to consider, 'e can look at all the data in an attempt to determine the "etter forecasting method to use. -nterpretation of these results 'ill "e impacted "y the "iases of the decision maker and the parameters of the decision situation. 3or e&le, one o"server could look at the forecasts 'ith method A and note that they 'ere pretty consistent in that they 'ere al'ays missing "y a modest amount !in this case, missing "y 18 units each year#. .o'ever, forecasting method > 'as very good in some years, and e&tremely "ad in some years !missing "y 68 units in years : and 6#. That o"servation might cause this individual to prefer the accuracy and consistency of forecasting method A. This causal o"servation is formali%ed in the calculation of the + E. 3orecasting method A has a considera"ly lo'er + E than forecasting method >. The s$uaring magnified those "ig misses that 'ere o"served 'ith forecasting method >. .o'ever, another individual might vie' these results and have a preference for method >, for the si%es of the misses relative to the si%es of the actual demand are smaller than for method A, as indicated "y the +APE calculations.
-llustration of the computation of tracking signals to accompany a progression of hypothetical forecasts made over time some hypothetical forecasting method. !These forecasts 'ere not made 'ith any of the forecasting methods 'e illustrated C the forecasts 'ere contrived to keep the num"ers managea"le.# /ear 0 1 2 5 : 6 At 208 26: 27: 50: 5:8 56: 3t 288 290 2;9 520 568 560 At - 3t 08 -6 ; -06 -08 5 Kum. Error 08 5 01 -5 -05 -08 OAt - 3tO 08 6 ; 06 08 5 Total OAt - 3tO 08 06 15 58 :8 :5 +AD 08 ; ; 08 08 7 T. . <0.88 <.:8 <0.:8 C.58 C0.58 C0.00
Reep in mind that each line in the a"ove ta"le 'ould have "een calculated in successive years. At the end of each year 'e can look "ack at the most recent year and compare the forecast 'e made 'ith the actual demand that occurred. The ne&t several pages sho' ho' these calculations 'ould have unfolded through the years, and ho' they 'ould have "een plotted on a graph to determine 'hether our forecasting method still appeared to "e 'orking 'ell.
,e no' "egin illustrating the computation and plotting of tracking signals to accompany the progression of forecasts made over time 'ith hypothetical forecasting +ethod 0. -n this illustration 'e 'ill assume that the upper limit has "een set at a value of 2, and the lo'er limit has "een set at a value of -2. -n practice these limits may "e higher or lo'er than these values, and they do not necessarily need to have the same numerical value. The values for these limits are largely a function of ho' costly or disruptive inaccurate forecasts are. As 'e run through time, assume that the forecast made for year 0 'as 288, and the su"se$uent demand that occurred in year 0 'as 208. The tracking signal calculated and plotted after year 0 'ould "e as follo's? Kum. Error 08 Total OAt - 3tO 08
/ear 0
At 208
3t 288
At - 3t 08
OAt - 3tO 08
+AD 08M0 A 08
T. . 08M08 A <0.88
Tracking ignal
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
Kontinuing 'ith our movement through time, assume that the forecast made for year 1 'as 290, and the su"se$uent demand that occurred in year 1 'as 26:. The tracking signal calculated and plotted after year 1 'ould "e as follo's? Kum. Error 08
!08#<!-6#
/ear 0 1
At 208 26:
3t 288 290
At - 3t 08 -6
OAt - 3tO 08 6
06
Tracking ignal
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
Kontinuing 'ith our movement through time, assume that the forecast made for year 2 'as 2;9, and the su"se$uent demand that occurred in year 2 'as 27:. The tracking signal calculated and plotted after year 2 'ould "e as follo's? Kum. Error 08
!08#<!-6#
/ear 0 1 2
At - 3t 08 -6 ;
OAt - 3tO 08 6 ;
5
!08#<!-6# <!;#
06
!08#<!6# <!;#
01
15
Tracking ignal
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
Kontinuing 'ith our movement through time, assume that the forecast made for year 5 'as 520, and the su"se$uent demand that occurred in year 5 'as 50:. The tracking signal calculated and plotted after year 5 'ould "e as follo's? Kum. Error 08
!08#<!-6#
/ear 0 1 2 5
At - 3t 08 -6 ; -06
OAt - 3tO 08 6 ; 06
5
!08#<!-6# <!;#
06
!08#<!6# <!;#
01
!08#<!-6# <!;#<!-06#
15
!08#<!6# <!;#<!06#
-5
58
Tracking ignal
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
Kontinuing 'ith our movement through time, assume that the forecast made for year : 'as 568, and the su"se$uent demand that occurred in year : 'as 5:8. The tracking signal calculated and plotted after year : 'ould "e as follo's? Kum. Error 08
!08#<!-6#
/ear 0 1 2 5 :
At - 3t 08 -6 ; -06 -08
OAt - 3tO 08 6 ; 06 08
T. . 08M08 A <0.88 5M; A <.:8 01M; A <0.:8 -5M08 A C.58 -05M08 A C0.58
5
!08#<!-6# <!;#
06
!08#<!6# <!;#
01
!08#<!-6# <!;#<!-06#
15
!08#<!6# <!;#<!06#
-5
!08#<!-6# <!;#<!-06# <!-08#
58
!08#<!6# <!;#<!06# <!08#
-05
Tracking ignal
:8
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
Kontinuing 'ith our movement through time, assume that the forecast made for year 6 'as 560, and the su"se$uent demand that occurred in year 6 'as 56:. The tracking signal calculated and plotted after year 6 'ould "e as follo's? Kum. Error 08
!08#<!-6#
/ear 0 1 2 5 : 6
At - 3t 08 -6 ; -06 -08 5
OAt - 3tO 08 6 ; 06 08 5
T. . 08M08 A <0.88 5M; A <.:8 01M; A <0.:8 -5M08 A C.58 -05M08 A C0.58 -08M7 A C0.00
5
!08#<!-6# <!;#
06
!08#<!6# <!;#
01
!08#<!-6# <!;#<!-06#
15
!08#<!6# <!;#<!06#
-5
!08#<!-6# <!;#<!-06# <!-08#
58
!08#<!6# <!;#<!06# <!08#
-05
!08#<!-6# <!;#<!-06# <!-08#<!5#
:8
!08#<!6# <!;#<!06# <!08#<!5#
-08
Tracking ignal
:5
:5M6 A 7
5 2 1 0
0 1 2 5 : 6 9 ; /ear
*pper Bimit
-0 -1 -2 -5
Bo'er Bimit
The cumulative error can "e either positive or negative, since each forecast error !At - 3t# can "e either positive or negative. -f you forecast too lo' in any period !i.e., the forecast is "elo' the demand#, the forecast error 'ill "e positive. -f you forecast too high in any period !i.e., the forecast is a"ove the demand#, the forecast error 'ill "e negative#. The +AD 'ill al'ays "e positiveS Konse$uently, the tracking signal could end up "eing either a positive num"er or a negative num"er. 2%at to 6atc% for -f the tracking signal plots outside the accepta"le range !i.e., a"ove the upper limit or "elo' the lo'er limit#, that is an indication that things are no longer going 'ell in our forecasting, and 'e should re-e&amine our method. .o'ever, even 'hen the tracking signal plots "et'een the upper limit and lo'er limit, this is not al'ays an indication that things are going 'ell in our forecasting. Konsider the follo'ing? ,hen 'e forecast, 'e e&pect to miss !i.e., make forecast errors#. -f 'e are un"iased in our forecasting approach, 'e should e&pect our forecast to "e too high on some occasions, and too lo' on other occasions. *ltimately, if 'e are making reasona"ly accurate forecasts the cumulative error should fluctuate "et'een positive and negative values, al'ays hovering around %ero. uppose the T is consistently plotting in the positive range. This 'ould "e an indication that 'e are consistently incurring a lot of positive forecast errors !i.e., forecasting too lo'#. This 'ould "e an indication that "ias has crept into our forecasting approach !i.e., a "ias to'ard forecasting too lo'#, and 'e should re-e&amine our forecasting approach. uppose the T is consistently plotting in the negative range. This 'ould "e an indication that 'e are consistently incurring a lot of negative forecast errors !i.e., forecasting too high#. This 'ould "e an indication that "ias has crept into our forecasting approach !i.e., a "ias to'ard forecasting too high#, and 'e should re-e&amine our forecasting approach. 2%at are reaso'a,!e trac*"'1 s"1'a! !"$"ts
.o' tight or ho' loose the tracking signal limits are set is a function of the conse$uences of forecast errors. The upper limit and the lo'er limit do not have to "e the same distance from the %ero mark. uppose that it is not a "ig deal if 'e forecast too high !negative forecast error# and make too much of a product. ,e can hold the e&cess in inventory and sell it at a later date. .o'ever, if 'e forecast too lo' !positive forecast error# 'e 'ill not have enough product to satisfy customer demand, and 'e are likely to lose customers to our competitors. -n such a case, 'e 'ould pro"a"ly have a tight range on the positive side of our tracking signal graph, and a relatively loose range on the negative side of our tracking signal graph. Alternatively, suppose that if 'e forecast too high !negative forecast error# and make too much of a product, the cost conse$uences are severe, for this product has a short shelf life or "ecomes o"solete $uickly, and e&cess inventory 'ill $uickly "ecome 'orthless. .o'ever, if 'e forecast too lo' !positive forecast error# and do not have enough product to satisfy customer demand it is no "ig deal, for customers are 'illing to 'ait for later deliveries. -n such a case, 'e 'ould pro"a"ly have a tight range on the negative side of our tracking signal graph, and a relatively loose range on the positive side of our tracking signal graph.