Supply Chain Management Lecture 13
Dec 21, 2015
Outline
• Today– Chapter 7
• Thursday– Network design simulation assignment– Chapter 8
• Friday– Homework 3 due before 5:00pm
Outline
• February 23 (Today)– Chapter 7
• February 25– Network design simulation description– Chapter 8– Homework 4 (short)
• March 2– Chapter 8, 9– Network design simulation due before 5:00pm
• March 4– Simulation results– Midterm overview– Homework 4 due
• March 9– Midterm
Summary: Static Forecasting Method
1. Estimate level and trend• Deseasonalize the demand data• Estimate level L and trend T using linear regression
• Obtain deasonalized demand Dt
2. Estimate seasonal factors• Estimate seasonal factors for each period St = Dt /Dt
• Obtain seasonal factors Si = AVG(St) such that t is the same season as i
3. Forecast• Forecast for future periods is
• Ft+n = (L + nT)*St+n
0
500
1000
1500
2000
2500
3000
3500
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Quarter
Dem
and
Forecast Ft+n = (L + nT)St+n
Ethical Dilemma?
In 2009, the board of regents for all public higher education in a large Midwestern state hired a consultant to develop a series of
enrollment forecasting models, one for each college. These models used historical data and exponential smoothing to forecast the following year’s enrollments. Each college’s budget was set by
the board based on the model, which included a smoothing constant () for each school. The head of the board personally selected each smoothing constant based on “gut reactions and
political acumen.”
How can this model be abused?
What can be done to remove any biases?
Can a regression model be used to bias results?
Forecast Forecast error
Time Series Forecasting
Observed demand =
Systematic component + Random component
L Level (current deseasonalized demand)T Trend (growth or decline in demand)S Seasonality (predictable seasonal fluctuation)
The goal of any forecasting method is to predict the systematic component (Forecast) of demand and measure the size and
variability of the random component (Forecast error)
1) Characteristics of Forecasts
• Forecasts are always wrong!– Forecasts should include an expected value and a
measure of error (or demand uncertainty)• Forecast 1: sales are expected to range between 100
and 1,900 units• Forecast 2: sales are expected to range between 900
and 1,100 units
Examples
8000
9000
10000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
50000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
800000
900000
1000000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
Measures of Forecast Error
Measure Description
Error
Absolute Error
Forecast – Actual Demand
Absolute deviation
Mean Squared Error (MSE) Squared deviation of forecast from demand
Mean Absolute Deviation (MAD)
Absolute deviation of forecast from demand
Mean Absolute Percentage Error (MAPE)
Absolute deviation of forecast from demand as a percentage of the demand
Tracking signal (TS) Ratio of bias and MAD
Forecast Error
• Error (E)
• Measures the difference between the forecast and the actual demand in period t
• Want error to be relatively small
Et = Ft – Dt
Forecast Error
-100000
-75000
-50000
-25000
0
25000
50000
75000
100000
1 2 3 4 5 6 7 8 9 10 11 12
Et
-5000
-4000
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
1 2 3 4 5 6 7 8 9 10 11 12
Et
-500
-400
-300
-200
-100
0
100
200
300
400
500
1 2 3 4 5 6 7 8 9 10 11 12
Et
-5000
-4000
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
1 2 3 4 5 6 7 8 9 10 11 12
Et
Forecast Error
• Bias
• Measures the bias in the forecast error• Want bias to be as close to zero as possible
– A large positive (negative) bias means that the forecast is overshooting (undershooting) the actual observations
– Zero bias does not imply that the forecast is perfect (no error) -- only that the mean of the forecast is “on target”
biast = ∑n∑t=1 Et
Forecast Error
8000
9000
10000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
50000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
Bias-19000120007300014000
-15000-400067000-2000
-51000-1000011000
-78000
Bias-300-900
-1800-3000-4500-6300-8400
-10800-13500-16500-19800-23400
Bias-200-500-200-500-300-100
0100
-100-400-90
-390
Bias912.61
1091.151350.811386.801109.80
-2332.49648.46435.64
-754.752789.40
-1361.73-920.13
Undershooting
Forecast mean “on target” but not perfect
800000
900000
1000000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
Forecast Error
• Absolute deviation (A)
• Measures the absolute value of error in period t• Want absolute deviation to be relatively small
At = |Et|
Forecast Error
• Mean absolute deviation (MAD)
• Measures absolute error• Positive and negative errors do not cancel out (as
with bias)• Want MAD to be as small as possible
– No way to know if MAD error is large or small in relation to the actual data
∑n1n
MADn = ∑t=1 At
= 1.25*MAD
Forecast ErrorMAD
190002500037000425003980035000401434375044333440004190945833
MAD300450600750900
1050120013501500165018001950
MAD200250267275260250229213211220228234
MAD913546450347333851
115510371054130315621469
8000
9000
10000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
50000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
800000
900000
1000000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
Not all that large relative to data
Forecast Error
• Tracking signal (TS)
• Want tracking signal to stay within (–6, +6)– If at any period the tracking signal is outside the range
(–6, 6) then the forecast is biased
TSt = biast / MADt
Forecast Error
Biased (underforecasting)
TS-1.000.481.970.33
-0.38-0.111.67
-0.05-1.15-0.230.26
-1.70
TS-1.00-2.00-3.00-4.00-5.00-6.00-7.00-8.00-9.00
-10.00-11.00-12.00
TS-1.00-2.00-0.75-1.82-1.15-0.400.000.47
-0.47-1.82-0.39-1.67
TS1.002.003.004.003.34
-2.740.560.42
-0.722.14
-0.87-0.63
0
10000
20000
30000
40000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
1 2 3 4 5 6 7 8 9 10 11 12
Tracking Signal
-15.00
-10.00
-5.00
0.00
5.00
10.00
15.00
1 2 3 4 5 6 7 8 9 10 11 12
Tracking Signal
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
1 2 3 4 5 6 7 8 9 10 11 12
Tracking Signal
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
1 2 3 4 5 6 7 8 9 10 11 12
Tracking Signal
Forecast Error
• Mean absolute percentage error (MAPE)
• Same as MAD, except ...• Measures absolute deviation as a percentage of
actual demand• Want MAPE to be less than 10 (though values
under 30 are common)
Et
Dt100∑n∑t=1
n
MAPEn =
Forecast Error
MAPE2.112.884.404.874.533.994.674.995.025.014.785.14
MAPE3.755.216.477.588.579.45
10.2410.9611.6212.2212.7813.29
MAPE2.222.883.143.152.962.852.622.422.392.512.612.65
MAPE11.416.394.643.503.365.986.986.186.598.669.058.39
8000
9000
10000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
0
10000
20000
30000
40000
50000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
800000
900000
1000000
1 2 3 4 5 6 7 8 9 10 11 12
Demand
Forecast
Smallest absolute deviation relative to
demand
MAPE < 10 is considered very good
Forecast Error
• Mean squared error (MSE)
• Measures squared forecast error • Recognizes that large errors are
disproportionately more “expensive” than small errors
• Not as easily interpreted as MAD, MAPE -- not as intuitive
∑n Et2MSEn = ∑t=1
1n
VAR = MSE
Measures of Forecast Error
Measure Description
Error
Absolute Error
Et = Ft – Dt
At = |Et|
Mean Squared Error (MSE) MSEn = ∑t=1Et2
Mean Absolute Deviation (MAD)
MADn = ∑t=1At
Mean Absolute Percentage Error (MAPE)
MAPEn =
Tracking signal (TS) TSt = biast / MADt
∑n1n
∑n1n
Et
Dt100∑n∑t=1
n
Summary
1. What information does the bias and TS provide to a manager?
• The bias and TS are used to estimate if the forecast consistently over- or underforecasts
2. What information does the MSE and MAD provide to a manager?
• MSE estimates the variance of the forecast error• VAR(Forecast Error) = MSEn
• MAD estimates the standard deviation of the forecast error• STDEV(Forecast Error) = 1.25 MADn
Forecast Error in Excel
• Calculate absolute error At
=ABS(Et)• Calculate mean absolute deviation MADn
=SUM(A1:An)/n=AVERAGE(A1:An)
• Calculate mean absolute percentage error MAPEn
=AVERAGE(…)• Calculate tracking signal TSt
=biast / MADt
• Calculate mean squared error MSEn
=SUMSQ(E1:En)/n
Forecast Error in Excel
Bias
Bias
bias_t=SUM($D$4:D4)=SUM($D$4:D5)=SUM($D$4:D6)=SUM($D$4:D7)
biasn = ∑n∑t=1 Et
Forecast Error in Excel
Mean Absolute Deviation
MeanAbs Error
MAD_t=AVERAGE($F$4:F4)=AVERAGE($F$4:F5)=AVERAGE($F$4:F6)=AVERAGE($F$4:F7)
∑n1n
MADn = ∑t=1 At
Forecast Error in Excel
Tracking Signal
TrackingSignalTS_t
=E4/G4=E5/G5=E6/G6=E7/G7
TSt = biast / MADt
Forecast Error in Excel
|%Error|
|%Error|t =
|%Error|
|%Error|=ABS(D4/B4)*100=ABS(D5/B5)*100=ABS(D6/B6)*100=ABS(D7/B7)*100
Et
Dt100
Forecast Error in Excel
Mean Absolute Percentage Error
Mean|%Error|MAPE_t
=AVERAGE($I$4:I4)=AVERAGE($I$4:I5)=AVERAGE($I$4:I6)=AVERAGE($I$4:I7)
|%Error|tnMAPEn =
∑n∑t=1