Framework

Forecasting models
Judgement

Old realizations of
demand: y1, …, yt

Old realizations of
variables: x1, …, xt Forecasting Forecast for: yt+1
Process

Subjective opinions
about the future
Case 1: Forecasting the Trend Form
• A constant increase or decrease in demand
denotes
– Linear trend
– Model
• Forecast (t) = a + b * t
 Case 1: Forecasting the Trend Form …
Calculations
t D(t) t-Tavg D(t) - Davg (t-Tavg) * (D(t)-Davg) (t-tavg)^2 Forecast Abs error
1 328 -7.5 -78.81 591.09 56.25 326.40 1.60
2 310 -6.5 -96.81 629.28 42.25 337.12 27.12
3 355 -5.5 -51.81 284.97 30.25 347.84 7.16
4 362 -4.5 -44.81 201.66 20.25 358.56 3.44
5 375 -3.5 -31.81 111.34 12.25 369.29 5.71
6 380 -2.5 -26.81 67.03 6.25 380.01 0.01
7 408 -1.5 1.19 -1.78 2.25 390.73 17.27
8 415 -0.5 8.19 -4.09 0.25 401.45 13.55
9 417 0.5 10.19 5.09 0.25 412.17 4.83
10 412 1.5 5.19 7.78 2.25 422.90 10.90
11 429 2.5 22.19 55.47 6.25 433.62 4.62
12 434 3.5 27.19 95.16 12.25 444.34 10.34
13 449 4.5 42.19 189.84 20.25 455.06 6.06
14 471 5.5 64.19 353.03 30.25 465.78 5.22
15 475 6.5 68.19 443.22 42.25 476.51 1.51
16 489 7.5 82.19 616.41 56.25 487.23 1.77
Mean 8.5 406.8125 7.57
Sum 3645.50 340.00
b 10.72206
a 315.675
Different Estimates of Forecast Error
Absolute Square Absolute
Month Sales Forecast Error Deviation Error % Error
1 328 326 2 2 4 0.61%
2 310 337 -27 27 729 8.71%
3 355 348 7 7 49 1.97%
4 362 359 3 3 9 0.83%
5 375 369 6 6 36 1.60%
6 380 380 0 0 0 0.00%
7 408 391 17 17 289 4.17%
8 415 401 14 14 196 3.37%
9 417 412 5 5 25 1.20%
10 412 423 -11 11 121 2.67%
11 429 434 -5 5 25 1.17%
12 434 444 -10 10 100 2.30%
13 449 455 -6 6 36 1.34%
14 471 466 5 5 25 1.06%
15 475 476 -1 1 1 0.21%
16 489 487 2 2 4 0.41%
Mean     0.0625 7.5625 103.0625 1.98%
      ME MAD MSE MAPE
 Case 2: Forecasting Seasonality …
Some Equations
• Seasonality index of period i for block j
– S(i, j) = (di)/[(d1 + d2 + … + dp)/p]
• For each period i within block j
• Average seasonality index for period i within a
block
– S(i)
Case 2: Forecasting Seasonality … Analysis

Period Seasonal
within Index of
Time Block block Demand Average Period
1 1 6 0.03
2 2 55 0.23
1 239
3 3 249 1.04
4 4 646 2.70
5 1 24 0.12
6 2 73 0.36
2 201.5
7 3 140 0.69
8 4 569 2.82
9 1 12 0.06
10 2 28 0.14
3 201.75
11 3 136 0.67
12 4 631 3.13
Case 2: Forecasting Seasonality …Seasonality
Index Calculation

Average
Seasonality
Period/Block S(i, 1) S(i, 2) S(i, 3) Index
1 0.03 0.12 0.06 0.07
2 0.23 0.36 0.14 0.24
3 1.04 0.69 0.67 0.80
4 2.70 2.82 3.13 2.88
 Case 2: Forecasting Seasonality … De-
seasonalizing Demand
• De-seasonalized demand data (t)
– Demand(t)/S(t)
• Forecast (t)
– [Level (t)] * Seasonal index (t)
 Case 2: Forecasting Seasonality …
Calculations

Period Seasonal De-

within Index of Seasonlized
Time Block block Demand Average Period Demand Forecast Abs Error
1 1 6 0.03 88 14 8
2 2 55 0.23 226 52 3
1 239
3 3 249 1.04 310 171 78
4 4 646 2.70 224 613 33
5 1 24 0.12 353 14 10
6 2 73 0.36 300 52 21
2 201.5
7 3 140 0.69 174 171 31
8 4 569 2.82 197 613 44
9 1 12 0.06 177 14 2
10 2 28 0.14 115 52 24
3 201.75
11 3 136 0.67 169 171 35
12 4 631 3.13 219 613 18
Mean 213 26
 Case 3: Forecasting Combination of
Seasonality and Trend
• Methodology
– Forecasting model
• Demand (t) = (Level (t) + Trend parameter * t) * Seasonality
parameter (t) + random error
– Step 1: Determine the seasonality index for each time
period within a block (similar to Case 2)
– Step 2: De-seasonalize the demand data (similar to Case 2)
– Step 3: Determine the trend and level components for the
de-seasonalized data series (similar to Case 1)
– Step 4: Finalize the forecasting model
Case 3: Forecasting Combination of
Seasonality and Trend … Analysis

Period
within Average Seasonality
Time Block Block Demand Demand Index
1 1 45 0.18
2 2 335 1.34
1 250
3 3 520 2.08
4 4 100 0.40
5 1 70 0.23
6 2 370 1.23
2 300
7 3 590 1.97
8 4 170 0.57
9 1 100 0.22
10 2 585 1.30
3 450
11 3 830 1.84
12 4 285 0.63
13 1 100 0.17
14 2 725 1.26
4 573.75
15 3 1160 2.02
16 4 310 0.54
Case 3: Forecasting Combination of
Seasonality and Trend … Analysis

Average
Seasonality
Period/Block 1 2 3 4 Index
1 0.18 0.23 0.22 0.17 0.20
2 1.34 1.23 1.30 1.26 1.28
3 2.08 1.97 1.84 2.02 1.98
4 0.40 0.57 0.63 0.54 0.54
 Case 3: Forecasting Combination of
Seasonality and Trend … Analysis

De-
seasonlized
Time Demand t-Tavg D(t) - Davg (t-Tavg) * (D(t) - Davg) (t-tavg)^2
1 222 -7.5 -172 1290.14 56.25
2 261 -6.5 -133 867.28 42.25
3 263 -5.5 -131 722.82 30.25
4 187 -4.5 -207 933.27 20.25
5 346 -3.5 -49 169.89 12.25
6 288 -2.5 -106 265.44 6.25
7 298 -1.5 -96 144.05 2.25
8 318 -0.5 -77 38.29 0.25
9 494 0.5 100 49.82 0.25
10 456 1.5 61 91.86 2.25
11 420 2.5 25 63.21 6.25
12 533 3.5 138 484.23 12.25
13 494 4.5 100 448.37 20.25
14 565 5.5 170 936.40 30.25
15 586 6.5 192 1248.66 42.25
16 579 7.5 185 1388.05 56.25
Mean 8.5 394.28
Sum b 27 9141.77 340
a 165.74
 Case 3: Forecasting Combination of
Seasonality and Trend … Analysis

• Step 1, 2 and 3 are complete

• Step 4:
Forecast(t) = (165.74 + 27 * t) * Seasonality Index(t)
 Case 3: Forecasting Combination of
Seasonality and Trend … Analysis
Period De-
within Seasonality seasonlized
Time Block Block Demand Average Demand Index Demand Forecast Abs Error
1 1 45 0.18 222 39 6
2 2 335 1.34 261 282 53
1 250
3 3 520 2.08 263 487 33
4 4 100 0.40 187 146 46
5 1 70 0.23 346 61 9
6 2 370 1.23 288 420 50
2 300
7 3 590 1.97 298 700 110
8 4 170 0.57 318 204 34
9 1 100 0.22 494 83 17
10 2 585 1.30 456 558 27
3 450
11 3 830 1.84 420 913 83
12 4 285 0.63 533 261 24
13 1 100 0.17 494 104 4
14 2 725 1.26 565 696 29
4 573.75
15 3 1160 2.02 586 1126 34
16 4 310 0.54 579 319 9
 Adaptive Methods
• Moving Average
• Simple Exponential Smoothening
• Holt’s Method
• Winter’s Method
 Basic Data
Month TV Sales CD Sales AC Sales
1 30 40 13
2 32 47 7
3 30 50 23
4 39 49 32
5 33 56 58
6 34 53 60
7 34 55 90
8 38 63 93
9 36 68 63
10 39 65 39
11 30 72 37
12 36 69 29
13 38 79 36
14 30 82 21
15 35 80 47
16 30 85 81
17 34 94 112
18 40 89 139
19 36 96 230
20 32 100 201
21 40 100 122
22 36 105 84
23 40 108 74
24 34 110 62
 Moving Average Method
• Time series that fluctuates about a constant
base level
• ft,1
Forecast for period t + 1made after observing xt
ft,1= Average of last N observations
 Moving Average Method Calculations
Month TV Sales MAF (N = 2) Error MAF (N = 3) Error MAF (N = 4) Error MAF (N = 5) Error MAF (N = 6) Error
1 30
2 32
3 30 31 1
4 39 31 8 30.67 8.33
5 33 34.5 1.5 33.67 0.67 32.75 0.25
6 34 36 2 34.00 0.00 33.50 0.50 32.80 1.20
7 34 33.5 0.5 35.33 1.33 34.00 0.00 33.60 0.40 33.00 1.00
8 38 34 4 33.67 4.33 35.00 3.00 34.00 4.00 33.67 4.33
9 36 36 0 35.33 0.67 34.75 1.25 35.60 0.40 34.67 1.33
10 39 37 2 36.00 3.00 35.50 3.50 35.00 4.00 35.67 3.33
11 30 37.5 7.5 37.67 7.67 36.75 6.75 36.20 6.20 35.67 5.67
12 36 34.5 1.5 35.00 1.00 35.75 0.25 35.40 0.60 35.17 0.83
13 38 33 5 35.00 3.00 35.25 2.75 35.80 2.20 35.50 2.50
14 30 37 7 34.67 4.67 35.75 5.75 35.80 5.80 36.17 6.17
15 35 34 1 34.67 0.33 33.50 1.50 34.60 0.40 34.83 0.17
16 30 32.5 2.5 34.33 4.33 34.75 4.75 33.80 3.80 34.67 4.67
17 34 32.5 1.5 31.67 2.33 33.25 0.75 33.80 0.20 33.17 0.83
18 40 32 8 33.00 7.00 32.25 7.75 33.40 6.60 33.83 6.17
19 36 37 1 34.67 1.33 34.75 1.25 33.80 2.20 34.50 1.50
20 32 38 6 36.67 4.67 35.00 3.00 35.00 3.00 34.17 2.17
21 40 34 6 36.00 4.00 35.50 4.50 34.40 5.60 34.50 5.50
22 36 36 0 36.00 0.00 37.00 1.00 36.40 0.40 35.33 0.67
23 40 38 2 36.00 4.00 36.00 4.00 36.80 3.20 36.33 3.67
24 34 38 4 38.67 4.67 37.00 3.00 36.80 2.80 37.33 3.33
Moving Average Method

N MAD
2 3.27
3 3.21
4 2.78
5 2.79
6 3.08
 Simple Exponential Smoothing
• A time series that fluctuates about a base level
• At = αxt + (1 – α) At-1
• A0 = 32
• α = 0.1
 Simple Exponential Smoothing
Calculations
Month TV Sa le s Fore ca st At et
1 30 32 31.8 2.00
2 32 31.8 31.82 0.20
3 30 31.82 31.64 1.82
4 39 31.64 32.37 7.36
5 33 32.37 32.44 0.63
6 34 32.44 32.59 1.56
7 34 32.59 32.73 1.41
8 38 32.73 33.26 5.27
9 36 33.26 33.53 2.74
10 39 33.53 34.08 5.47
11 30 34.08 33.67 4.08
12 36 33.67 33.91 2.33
13 38 33.91 34.32 4.09
14 30 34.32 33.88 4.32
15 35 33.88 34.00 1.12
16 30 34.00 33.60 4.00
17 34 33.60 33.64 0.40
18 40 33.64 34.27 6.36
19 36 34.27 34.45 1.73
20 32 34.45 34.20 2.45
21 40 34.20 34.78 5.80
22 36 34.78 34.90 1.22
23 40 34.90 35.41 5.10
24 34 35.41 35.27 1.41
Simple Exponential Smoothing

Alpha MAD
0.05 3.20
0.1 3.04
0.15 2.94
0.2 2.89
0.25 2.88
0.3 2.90
0.35 2.94
0.4 2.98
0.45 3.05
0.5 3.14
 Holt’s Method: Exponential Smoothing with
Trend
• Base Level at the end of tth period = Lt
• The per-period trend at the end of tth period = Tt
– For e.g., if L20 = 20 and T20 = 2. Implication?
• Lt = αxt + (1 – α) (Lt-1 + Tt-1)
• Tt = β(Lt - Lt-1 ) + (1 – β) Tt-1
• Ft,k = Lt + k * Tt
• T0 = average monthly increase in the time series
during the previous year
• L0 = Last month’s observation
 Holt’s Method Calculations
• Let, the CD sales during each of the last 12
months are given by 4, 6, 8, 10, 14, 18, 20, 22,
24, 28, 31, 34.
• T0 = [(6 - 4) + (8 – 6) + (10 – 8) + … + (34 – 31)]/11 = 2.73
• L0 = 34
• α = 0.3
• β = 0.1
 Holt’s Method Calculations
ft-1,1
(= Lt-1+Tt- e t (xt - ft-
Month CD Sale s Lt Tt 1) 1,1)
1 40 37.71 2.83 36.73 3.27
2 47 42.47 3.02 40.53 6.47
3 50 46.85 3.15 45.49 4.51
4 49 49.70 3.12 50.00 1.00
5 56 53.78 3.22 52.82 3.18
6 53 55.80 3.10 57.00 4.00
7 55 57.73 2.98 58.90 3.90
8 63 61.40 3.05 60.71 2.29
9 68 65.51 3.16 64.45 3.55
10 65 67.57 3.05 68.67 3.67
11 72 71.03 3.09 70.62 1.38
12 69 72.59 2.94 74.12 5.12
13 79 76.57 3.04 75.52 3.48
14 82 80.32 3.11 79.61 2.39
15 80 82.40 3.01 83.44 3.44
16 85 85.29 3.00 85.41 0.41
17 94 90.00 3.17 88.29 5.71
18 89 91.92 3.04 93.17 4.17
19 96 95.27 3.07 94.96 1.04
20 100 98.84 3.12 98.35 1.65
21 100 101.38 3.06 101.97 1.97
22 105 104.61 3.08 104.44 0.56
23 108 107.78 3.09 107.69 0.31
24 110 110.61 3.06 110.87 0.87
 Choice of α & β
Beta
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.1 2.86 2.80 2.74 2.70 2.75 2.80 2.83 2.86 2.92
0.2 2.77 2.73 2.76 2.79 2.84 2.92 2.97 2.98 2.99
0.3 2.85 2.87 2.91 2.95 2.99 3.04 3.13 3.22 3.29
Alpha 0.4 2.96 3.00 3.05 3.11 3.17 3.23 3.30 3.35 3.39
0.5 3.07 3.13 3.19 3.25 3.31 3.36 3.43 3.50 3.58
0.6 3.19 3.26 3.33 3.40 3.48 3.59 3.69 3.78 3.88
0.7 3.32 3.39 3.47 3.60 3.72 3.84 3.95 4.07 4.21
0.8 3.43 3.53 3.67 3.81 3.94 4.07 4.27 4.47 4.68
0.9 3.54 3.69 3.85 4.02 4.23 4.46 4.69 4.95 5.25
 Winter’s Method
• c = number of periods in the length of the seasonal
pattern
c = 12 for monthly data.
• st = seasonal multiplicative factor for month t,
obtained after observing xt.
 Winter’s Method
• Lt = α(xt/st-c) + (1 – α) (Lt-1 + Tt-1)
• Tt = β(Lt - Lt-1 ) + (1 – β) Tt-1
• st = γ (xt/Lt) + (1 – γ) st-c
• ft,k = (Lt + k * Tt) * St+k-c
 Initialization of Winter’s Method
• L0 = estimate of base at beginning of month 1
• T0 = estimate of trend at beginning of month 1
• S-11 = estimate of January seasonal factor at the
beginning of month 1
• S-10 = estimate of February seasonal factor at the
beginning of month 1
• .
• S0 = estimate of December seasonal factor at the
beginning of month 1.
 Initialization of Winter’s Method
• Variety of methods are available to estimate
above parameters
• A simple approach:
– Suppose we have two years of data
• Year –2: 4, 3, 10, 14, 25, 26, 38, 40, 28, 17, 16, 13
• Year –1: 9, 6, 18, 27, 48, 50, 75, 77, 52, 33, 31, 24
• Total sales during year –2 = 234
• Total sales during year –1 = 450
 Initialization of Winter’s Method
• T0 = [(Avg. monthly sales during year –1) – (Avg.
monthly sales during year –2)]/12
– T0 = 1.5
• L0 = Avg. monthly demand during year –1.
– Further correction
• This estimates the base at the middle of the year –1
– Month 6.5 of year –1
» Hence, to bring this estimate to the end of the year
» Add, (12 – 6.5)T0 = 5.5T0
» L0 = 37.5 + 5.5(1.5) = 45.75
 Initialization of Winter’s Method
• To estimate the seasonality factor for a given
month (say, January = s-11)
– we take an estimate of January seasonality of year
–2 and year –1 and average them.
– In year –2, average monthly demand = 19.5
– In January of year –2, number of ACs sold = 4
• Therefore, s-11 = [(4/19.5) + (9/37.5)]/2 = 0.22
 Initialization of Winter’s Method
• s-10 = 0.16, s-9 = 0.50, s-8 = 0.72, s-7 = 1.28, s-6 = 1.33, s-5 = 1.97, s-
4 = 2.05, s-3 = 1.41, s-2 = 0.88, s-1 = 0.82, s0 = 0.65

• Observation and a check!

– Sum of initial seasonal factor estimates should
average to 1
• At the beginning of month 1, our forecast for
month 1 AC sales is
– f0,1 = (L0 + T0)s0+1-12 = (45.75 + 1.5)0.22 = 10.40
 Initialization of Winter’s Method
• At the beginning of month 1, our forecast for
month 7 AC sales is
– f0,7 = (L0 + 7T0)s0+7-12 = (45.75 + 7 *1.5) 1.97 = 110.81
• For
– α = 0.5
– β = 0.4
– γ = 0.6
 Computation of L1, T1 and S1
• L1 = 0.5 (x1/s-11) + 0.5 (L0 + T0)
= 0.5 (13/0.22) + 0.5 (45.75 + 1.5) = 52.83
• T1 = 0.4 (L1 – L0) + 0.6 T0
= 0.4 (52.83 – 45.75) + 0.6 (1.5) = 3.73
• s1 = 0.6 (x1/L1) + 0.4 (s-11)
= 0.6 (13/52.83) + 0.4 (0.22) = 0.24
Thus, at the end of month 1, our forecast for (say) month 7 AC
sales
f1,6 = (L1 + 6T1)s1+6-12 = (52.83 + 6(3.73)) 1.97 = 150.49
Winter’s Method for Air-Conditioners

Month Sales Lt Tt st ft-1,1 Error

1 13 52.83 3.73 0.24 10.52 2.48
2 7 50.58 1.34 0.15 8.88 1.88
3 23 49.13 0.22 0.48 25.78 2.78
4 32 46.93 -0.75 0.70 35.48 3.48
5 58 45.73 -0.93 1.27 59.16 1.16
6 60 44.90 -0.89 1.34 59.74 0.26
7 90 44.80 -0.57 2.00 86.90 3.10
8 93 44.77 -0.36 2.07 90.76 2.24
9 63 44.53 -0.31 1.41 62.68 0.32
10 39 44.37 -0.25 0.88 38.73 0.27
11 37 44.52 -0.09 0.83 36.34 0.66
12 29 44.41 -0.10 0.65 29.03 0.03
 Two Strategies
• Demand Pull
– The retailer orders every week throughout the
year
• Forward-Buying
– The retailer orders only twice per year during the
trade promotion
 Costs of Two Strategies
• Demand Pull
– During the two weeks of promotion, the avg. inv. is
$ 2760 (= 150 * $18.4)
– During the remaining 50 weeks of the year, the avg. inv. is
$ 3000 (= 150 * $ 20)
– The weighted avg. inv. = $ 2991
– The purchased cost during the year
$ 103680 (= $ 20 * 100 * 50 + $ 18.40 * 100 * 2)
– The demand pull strategy total cost
$ 1,04,398 (= 718 + 103680)
 Costs Comparison

Demand Pull Forward Buying

Annual Purchase (Units) 5200 5200
Average Inventory (Units) 150 1400
Average Inventory (Monetory Value) 2991 25760
Holding Cost 718 6182
Units Purchased at Regular Price 5000 0
Units Purchased at Discount Price 200 5200
Total Purchase Cost 103680 95680
Total Holding plus Procurement Cost 104398 101862
Week Retailer Manufacturer
Supply Stock at
Shipment Stock at the Order Placed received Order the end Order
received from end of the on from from Order of the placed
Manufacturer Demand week Backlog Manufacturer plant retailer Shipped week on plant Backlog
1 100 100 200 0 100 100 100 100 200 100 0
2 100 100 200 0 100 100 100 100 200 100 0
3 100 200 100 0 500 100 100 100 200 100 0
4 100 100 100 0 0 100 500 300 0 1700 200
5 300 100 300 0 0 100 0 100 0 0 100
6 100 100 300 0 0 1700 0 100 1600 0 0
7 100 100 300 0 100 0 0 0 1600 0 0
8 0 100 200 0 100 0 100 100 1500 0 0
9 100 100 200 0 100 0 100 100 1400 0 0
10 100 100 200 0 100 0 100 100 1300 0 0
 An Illustration
• P&G, the maker of Puffs tissues, traditionally sells these tissues for
$9.40 per case
– where a case contains 8 boxes
• A retailer’s average weekly demand is 25 cases of particular Puffs SKU
(color, scent etc.)
• P& G has decided to change its pricing strategy by offering two
different plans
• With one plan, the retailer can purchase that SKU for the every-day-
low-wholesale price of $9.25 per case
• With the other plan, P&G charges the regular price of $9.40 per case
throughout most of the year
– but purchases made for a single delivery at the start of each quarter are
given a 5% discount
• The retailer receives weekly shipments with a 1-week lead time
between ordering and delivery
 Analysis
• 1st Question
– Orders are made every week!
In this case “sawtooths” from a high of 2-weeks’ worth of inventory down to 1-
week, with an average of 1.5 weeks
– On average, the inventory value?
• 37.5 * 9.25 = $346.9
– The holding cost per year
• 52 * 0.4% = 20.8%
– Hence, the inventory holding cost with the 1st plan
• 20.8% * $346.9 = $72
– Purchase cost
• 52 * 25 * $9.25 = $12,025
• Total?
$12,025 + $72 = $12,097
 Analysis
• 2nd Question
– 4 Orders are made each year!
325/2 + 25 = 187.5
– On average, the inventory value?
• 187.5 * 9.40 * 0.95 = $1,674
– The holding cost per year
• 52 * 0.4% = 20.8%
– Hence, the inventory holding cost with the 1st plan
• 20.8% * $1674 = $348
– Purchase cost
• 52 * 25 * $8.93 = $11,609
• Total?
$11,609 + $348 = $11,957
 Production of Pasta

Slope = P - D Slope = - D
Q(1-D/P)

Q/P
Run Time
Q/D
Production Interval
 Analysis
• 3rd Question
– P&G will prefer 3rd plan as long as
• Price is higher than in the 2nd plan, i.e., $8.93
– But the retailer needs a low enough price so that its
total cost with the 3rd plan
• Is not greater than in the 2nd plan
– i.e., $11,957

– In 1st question, we determined that the annual holding cost with a weekly
ordering plan is $72
• If we lower the price, the annual holding cost will be a bit lower
– But $72 is a conservative approximation of the holding cost

– So the retailer’s purchase cost should not exceed $11,957 - $72 = $11,885
– Total Purchasing Quantity?
• 25 * 52 = 1300 units
• So, if the price is $11,885/1300 = $9.14, then the retailer will be slightly better offf
– Relative to the second plan
• And P&G is much better off
– Revenue of $12,012 instead of $11,885
A Numerical Illustration

Number of Days until

SKU Days of the next scheduled
Number Stock replenishment
1 0 2
2 1 2
3 2 2
4 3 1
5 3.5 2
6 1.6 2
 Further …

Number of Days until the next Days of

SKU Number Days of Stock scheduled replenishment Stock/Column 3
1 0 2 0
2 1 2 0.5
3 2 2 1
4 3 1 3
5 3.5 2 1.75
6 1.6 2 0.8
 Connecting to Truck Load

Number of Days until the next Days of

SKU Number Days of Stock scheduled replenishment Stock/Column 3
1 2 2 1
2 2 2 1
3 2 2 1
4 3 1 3
5 3.5 2 1.75
6 2.0 2 1
 From the Last Class …
• Introduction to
– Performance measures
• Evaluation
Expected
Fill rate
demand, m

If Normal Expected
demand, s sales

Loss function Expected

table lost sales

Order quantity,
Q, and, if Exp. left over Expected
Normal demand, inventory profit
z = (Q In-stock
– m)/s probability

Distribution Stockout
function table probability

Price, cost,
salvage value
 Performance Measures
• Expected lost sales σ× L(Z)

• Expected sales Expected Demand µ

• Expected leftover inventory Q
• Expected profit
• Fill rate
• Expected Sales/µ
In-stock probability/Stockout probability
– Also known as service level
 Summary of Performance Measures

Order Quantity 4095 Units F(Q) = Cu/(Cu + Co)

Expected Demand 3192 Units
µ
Std. Dev. Of Demand 1181 σ
Expected Lost Sales 150 Units σ * L(Z)
Expected Sales 3042 Units µ - Expected Lost Sales
Expected Leftover Inventory 1059 Units Q – Expected Sales
Expected Revenue 642870 $ Price * Expected Sales + Salvage Value * Expected Leftover Inventory
Expected Profit 191760 $ Cu * Expected Sales – Co * Expected Leftover Inventory
Mismatch Cost 31680 Cu * Expected Lost Sales + Co * Expected Leftover Inventory
 Value of Perfect Information
• Assume company XYZ had the opportunity
to purchase a magic crystal ball
• Which can reveal exact demand for the entire
season
• Crystal ball doesn’t change demand, increases
selling price or decrease the production cost
– XYZ will then order according to this demand quantity
• Mismatch cost will be zero
• Therefore
• The difference between company profit with the
crystal ball and without it
• must be equal to the mismatch cost
 Further …
• With Crystal ball
Maximum profit?
• (p – c) * mu
$223,440
• Mismatch cost = Maximum Profit – Expected Profit
Cost to purchase crystal ball
• Minimizing mismatch cost
Is equivalent to maximizing expected profit
 Observation
• Mismatch cost
– Increases as demand variability increases
• Demand variability measure: CV
– Why CV is important measure?
 Measure of Demand Variability
• CV
Probability (Demand is Probability (Demand
Coefficient of less than 75%) of the is within 25% of the
Variation Forecast) Forecast)
0.1 0.62% 98.76%
0.25 15.87% 68.27%
0.5 30.85% 38.29%
0.75 36.94% 26.11%
1 40.13% 19.74%
1.5 43.38% 13.24%
2 45.03% 9.95%
3 46.68% 6.64%
 Another Observation
• Mismatch cost
• Increases as critical ratio becomes smaller
– Observation
» The first term in the derivation
» Ratio of std. normal density function to the std. normal
distribution function
» It depends on z and z depends on critical ratio
» Higher the z implies CR to be high but ratio becomes smaller
» Which implies for higher critical ratio, the mismatch cost is
lower
 An Example
• Lets, Say XYZ can send a second order
• TEC will charge 20% premium
• Q?
3,263
• Expected second order quantity?
437
• Expected profit
maximum profit – [co * Expected leftover inventory + cu * expected second
order quantity] ?
 Another Approach to Estimate Expected
Profit
• Estimate Q1
• Estimate Expected Lost Sales based on Q1
– This becomes equal to Q2
• Estimate Expected Sales from Q1
• Estimate Expected Leftover Inventory from Q1
• Expected Profit
= Price * Expected Sales from Q1
+ Salvage Value * Expected Leftover Inventory from Q1
– Cost of ordering Q1 – Cost of ordering Q2 + Price * Q2