Supply Chain Management

Part 6
Demand Forecasting in the Supply Chain

1-1
The role of demand forecasting in supply chain

Demand forecasting in supply chain management refers to the

process of planning or predicting the demand of materials to
ensure you can deliver the right products and in the right quantities
to satisfy customer demand without creating a surplus.
Demand forecasting is the basis for all strategic and planning
decisions in a supply chain.
Accurate demand forecasting results in: customer satisfaction,
better allocation of resource, streamlining inventory, better sales
strategies, and better suppliers and purchase terms.

1-2
 Characteristics of forecasts

1- Forecasts are always inaccurate.

2- Long-term forecasts are usually less accurate than short-term

forecast.

3- Aggregate forecasts are usually more accurate than disaggregate

forecasts.

1-3
 Forecasting methods

Qualitative: primarily subjective; rely on judgment and

opinion.
Time Series: use historical demand only.
Causal: use the relationship between demand and
some other factor to develop forecast.
Simulation: imitate consumer choices that give rise to
demand. Can combine time series and causal methods.

1-4
 Time series forecasting

• A time series is simply a series of data points ordered in time.

In a time series, time is often the independent variable and the
goal is usually to make a forecast for the future. (example:
historical data on sales, inventory, costs, etc.)

1-5
 Elements of time series

• Trend: trend is the long term pattern of a time series. A trend can be
positive or negative (The growth or decline in Sales over time ).

1-6
 Elements of time series

• Seasonality: the repeating short-term cycle in the series. (Example.

You may have strong sales in the summer but weak sales in the spring
and fall ).

1-7
 Elements of time series

• Cyclic variation: a cyclic pattern exists when data exhibit rises and
falls that are not of fixed period.

1-8
 Elements of time series

• Irregular variation: Irregular variations are fluctuations in time

series that are short in duration, erratic in nature and follow no
regularity in the occurrence pattern.

1-9
 Naive forecast

• Naive forecast: nothing will change.

Future value = current value

Example: Last month you sold 250 computers, so you

predict that this month you’ll sell 250 computers again.

Good in very steady environment

1-
10
 Moving average

1-
11
Moving average

1-
12
 Moving average

• Question: What are the 3 week and 9 week moving average forecasts
for the following demand?

1-
13
Moving average

1-
14
Moving average

1-
15
 Weighted moving average

1-
16
 Weighted moving average

1-
17
 Simple exponential smoothing

• Simple exponential smoothing: based on the premise

that the most recent observations might have the highest
predictive value. Therefore, we should give more weight to the
more recent time periods when forecasting.
F t +1 = α D t + (1 – α) F t
where:
F t +1 forecast for next period
D t actual demand for present period
F t previously determined forecast for present period
α weighting factor, smoothing constant 1-
18
 Simple exponential smoothing

• 0.0 ≤ α ≤ 1.0
If α = 0.20, then Ft +1 = 0.20 Dt + 0.80 Ft
If α = 0, then F t +1 = 0 Dt + 1 Ft = Ft . Forecast does
not reflect recent data
If α = 1, then Ft +1 = 1 Dt + 0 Ft = Dt. Forecast based
only on most recent data

1-
19
 Simple exponential smoothing

Choosing appropriate values for α

➢ If actual demands are stable, use a small α to lessen effects
of short term changes in demand.
➢ If actual demands rapidly increase or decrease, use a large
α so forecasts keep pace with the changes in demand.
➢ α is usually determined by trial and error, with several
values tested on existing data or a portion of data (aim is
to minimize the average forecasting error.

1-
20
 Simple exponential smoothing

Example. demands for an electrical component for the past

6 months are as follows.
Period 1 2 3 4 5 6
Demand 59 65 60 71 65 68

Obtain the forecast value of the demand for this product for
the seventh period by using the simple exponential
smoothing method and by selecting the appropriate alpha
from the numbers 0.2, 0.3, and 0.4. Assume the forecasted
demand for this product in the first period is 55.
1-
21
 Simple exponential smoothing

α = 0.2 α = 0.3 α = 0.4

Period Demand Forecast error Forecast error Forecast error

1 59 55 4 55 4 55 4

2 65 (0.2*59)+(0.8*55)= 55.8 9.2 56.2 8.8 56.6 8.4

3 60 =(0.2*65)+(0.8*55.8)=57.6 2.4 58.8 1.2 60 0

4 71 =(0.2*60)+(0.8*57.6)=58.1 12.9 59.2 11.8 60 11

5 65 =(0.2*71)+(0.8*58.1)=60.7 4.3 62.7 2.3 64.4 0.6

6 68 =(0.2*65)+(0.8*60.7)=61.6 6.4 63.4 4.6 64.4 3.4

1-
22
Simple exponential smoothing

1-
23
 Trend-adjusted Exponential Smoothing
(Holt’s Model)
• The trend-adjusted exponential smoothing (Holt’s model) is appropriate
when demand is assumed to have a trend in, but no seasonality.
In this case, in Period t, given estimates of level Lt and trend Tt, the forecast for
future periods is expressed as:
Ft = Lt-1 + Tt-1
Lt = αDt + (1- α)(Lt-1 + Tt-1)
Tt = β(Lt – Lt-1) + (1-β)Tt-1
where:
Lt: is estimates of level in period t
Tt: is estimates of trend in period t
1-
α and β are smoothing constants for level and trend, respectively (0< α and β<1) 24
 Trend-adjusted Exponential Smoothing
(Holt’s Model)
We obtain an initial estimate of level and trend by running a linear
regression between demand, Dt , and time, Period t, of the form
Dt = at + b

In this case, running a linear regression between demand and time

periods is appropriate because we have assumed that demand has a
trend but no seasonality.

The constant b measures the estimate of demand at Period t = 0 and is

our estimate of the initial level L0. The slope a measures the rate of
change in demand per period and is our initial estimate of the trend T0. 1-
25
 Trend-adjusted Exponential Smoothing
(Holt’s Model)
•

1-
26
 Trend-adjusted Exponential Smoothing
(Holt’s Model)
Example. An electronics manufacturer has seen
demand for its latest MP3 player increase over the
past six months. Observed demand (in thousands) has
been D1 = 8415, D2 = 8,732, D3 = 9014, D4 = 9,808,
D5 =10416, and D6 = 11,961. Forecast demand for
Period 7 using trend-adjusted exponential smoothing
with α = 0.1, β = 0.2.
The first step is to obtain initial estimates of level and
trend using linear regression.
1-
27
Trend-adjusted Exponential Smoothing
(Holt’s Model)

t D t2 tD •
1 8415 1 8415

2 8732 4 17464

3 9014 9 27042

4 9808 16 39232

5 10413 25 52065

6 11961 36 71766

Sum 58343 91 215984

1-
28
 Trend-adjusted Exponential Smoothing
(Holt’s Model)
L1 = αD1 + (1- α)(L0 + T0) = (0.1*8415) + 0.9*(7367 + 673) = 8078
T1 = β(L1 – L0) + (1-β)T0 = [0.2 * (8078 – 7367)] + (0.8*673) = 681
F2 = L1 + T1 = 8078 + 681 = 8759
Continuing in this manner, we obtain L2 = 8755, T2 = 680, L3 = 9393,
T3 = 672, L4 = 10039, T4 = 666, L5 = 10676, T5 = 661, L6 = 11399, T6
= 673.
This gives us a forecast for Period 7 of
F7 = L6 + T6 = 11399 + 673 = 12072

1-
29
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
• Trend- and seasonality-adjusted exponential smoothing (winter’s
model): This method is appropriate when the systematic component
of demand has a level, a trend, and a seasonal factor.
In Period t, given estimates of level, Lt, trend, Tt, and seasonal factors,
St, the forecast for future periods is given by
Ft = (Lt-1 + Tt-1)St
Lt = α(Dt/St) + (1- α)(Lt-1 + Tt-1)
Tt = β(Lt – Lt-1) + (1-β)Tt-1
St+p = γ (Dt/Lt) + (1- γ)St
where:
α, β, are γ smoothing constants for level, trend, and seasonal
factor, respectively. (0< <1) 1-
30
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
Steps of Winter’s model:
1- Deseasonalizing the demand data. Deseasonalized demand
represents the demand that would have been observed in the
absence of seasonal fluctuations.
2- Run linear regression based on deseasonalized demand to
estimate initial level and trend (same as Holt’s model).
3- Estimate seasonal factors.
4- Forecast

1-
31
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
•

1-
32
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
•

1-
33
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
• Example. Quarterly retail demand data for the past three years are
shown in bellow table. Using Winter’s model, forecast demand for
period 1 and 2 with α=0.1 , β=0.2, γ=0.1

1-
34
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)

Given that we are measuring demand on a quarterly basis, the

periodicity for the demand is p =4
1-
35
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
•

1-
36
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)

1-
37
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
•

1-
38
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)

1-
39
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
•

1-
40
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
So far we obtained initial estimates of level, trend and seasonal factors as:
L0 = 18439, T0 = 524, S1 = 0.47, S2 = 0.68, S3 = 1.17, S4 = 1.67
Ft = (Lt-1 + Tt-1)St
F1 = (L0 + T0)S1 = (18439 + 524) * 0.47 = 8913
Lt = α(Dt/St) + (1- α)(Lt-1 + Tt-1)
L1 = α(D1/S1) + (1- α)(L0 + T0) = [0.1 * (8000/0.47)] + [0.9 * (18439 + 524]
= 18769
Tt = β(Lt – Lt-1) + (1-β)Tt-1
T1 = β(L1 – L0) + (1-β)T0 = [0.2 *(18769 – 18439)]+ (0.8 * 524) = 485
1-
41
 Trend- and seasonality-adjusted
Exponential Smoothing (Winter’s Model)
So far we obtained initial estimates of level, trend and seasonal factors as:
St+p = γ (Dt/Lt) + (1- γ)St
S5 = γ (D1/L1) + (1- γ)S1 = [0.1 * (8000/18769)] + (0.9 * .47) = 0.47

F2 = (L1 + T1)S2 = (18769 + 485) * 0.68 = 13093

1-
42
 Static Method

A static method assumes that the estimates of level, trend, and seasonality
do not vary as new demand is observed. In this case, we estimate each of
these parameters based on historical data and then use the same values for
all future forecasts.

Ft = (L0 + tT0) St

Note: L0 (estimate of level at t=0), T0 (estimate of trend at t=0), and St

(overall seasonal factor at period t) are calculated same as Winter’s model.

1-
43
 Static Method

Example. Consider the previous example and forecast for the next four
periods using the static forecasting method.
we obtained initial estimates of level, trend and seasonal factors as in the
previous example:
L0 = 18439, T0 = 524, S1 = 0.47, S2 = 0.68, S3 = 1.17, S4 = 1.67
F13 = (L0 + 13T0)S13 = (18439 + 13 * 524) 0.47 = 11868
F14 = (L0 + 14T0)S14 = (18439 + 14 * 524) 0.68 = 17527
F15 = (L0 + 15T0)S15 = (18439 + 15 * 524) 1.17 = 30770
F16 = (L0 + 16T0)S16 = (18439 + 16 * 524) 1.67 = 44794
1-
44
 Linear Regression

1-
45
 Linear Regression

1-
46
 Linear Regression

Example. A company based on its research, has concluded that the amount
of product sales is directly related to the amount of advertising. The data of
the last 4 years are as follows:
Advertising cost (x) Sales amount (y)

32 130
52 151
50 150
55 158
If the company allocates $ 53 million for advertising for the next year,
forecast the amount of sales by using linear regression method. What is
1-
your idea about the accuracy of this model? 47
 Linear Regression

• Advertising cost (x) Sales amount (y) x^2 xy

32 130 1024 4160
52 151 2704 7852
50 150 2500 7500
55 158 3025 8690
Sum 189 589 9253 28202

1-
48
 Linear Regression

1-
49
 Linear Regression

1-
50
 Measures of forecast error

1-
51
 Measures of forecast error

1-
52
 Measures of forecast error

1-
53
Forecasting Process

1-
54