Professional Documents
Culture Documents
Demand
Outline
⚫ Types of forecast
⚫ Types of forecast
⚫ Types of forecast
⚫ Optimistic
⚫ Pessimistic
Optimistic Forecast
It is considered better to be
Optimistic rather than Pessimistic when forecasting.
Pessimistic forecasts result in:
⚫ loss of customers due to inability to fill orders
⚫ Types of forecast
⚫ Quantitative methods
• based on data, statistics
• e.g. Causal methods, Extrapolative methods
Different Forecasting Approaches
Market Research:
⚫ Assess the demand for a new product by survey
Historical Comparisons:
⚫ Review available information about similar products or
processes
Delphi Method:
⚫ Pool separate knowledge of different experts wait for an
agreement to emerge
etc.
Qualitative Methods : Advantages &
drawbacks
⚫ Advantages :
• Take into account intangible factors.
• Used when there is little demand information (new product, new
market, etc.)
⚫ Drawbacks :
• Long delay of the process
• Bias/subjectivity
• High cost (ex. Experts consulting)
• Not enough precision
Different Forecasting Approaches
▪ There are methods that can be used to detect the outliers in a time series:
Confidence interval method, Student test, etc.
▪ Confidence interval method :
▪ Principle:
1 n
▪ Calculate the mean of the time series:
x = xi
n t =1
n
1
▪ Calculate the standard deviation: =
x
n − 1 t =1
( xi − x ) 2
▪ Calculate the confidence interval: IC = x 1.96
x
Demand 28 37 29 32 49 38 29 36 21 35 14 31
▪ Calculations :
• Mean = 31.58
Seasonnality
Cyclic
demand
Analysis of a temporal series:
variation and trend analysis
Month 1 2 3 4 5 6 7 8 9 10
Demand 1 19 20 20 22 17 22 20 21 21 18
Demand 2 11 12 15 17 18 21 23 24 28 31
Analysis of a temporal series:
Forecasting Performance
Month (t) 1 2 3 4 5 6 7 8 9 10
Forecast (Ft) 10 13 18 15 17 22 20 21 17 18
Actual Demand (Dt) 11 12 15 17 18 21 23 20 16 19
n
1
MAD = Dt − Ft
n t =1
⚫ Measures absolute error
⚫ Positive and negative errors thus do not cancel out
(as with MFE)
⚫ 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
Mean Absolute Percentage Error
(MAPE)
Sales
3 12.0 147 150
50
5 9.5 131
0
6 12.5 159 0 5 10 15 20
Advertising
7 14.5 160
8 11.0 124
Calculate the Sales Forecast
( y − y )( x − x )
t t
a= t =1
n
and b = y −ax
t
( x
t =1
− x ) 2
250
200
y = 8.78 x + 36.74
R2=0,8653
150
Sales
100
50
0
0 5 10 15
Advertising budget
Example Calculation
They do not take into account the external factors but they
look at a series of past values to predict what will happen
in the future
Demand 12 23 25 15 32 42 26 21 18
Forecast 12 23 25 15 32 42 26 21 18
Dt −1 + Dt −2 + Dt −3 + ..... + Dt −n
Ft =
n
Dt : Actual demand at time t
Ft : Forecast at time t
Moving Averages of order 3: Example
February 38
March 29
April 35
May 31
Calculate the forecasts in April, May and June by using the weighted
moving averages (over 3 periods) knowing that the weight of the
demand of the last month is twice important than the two previous
months ?
Example
Weighted Moving Averages
Exponential Smoothing
February 38
March 29
April 35 37.33
May 31 ?
June 30 ?
February 38
March 29
April 35 37.33
May 31 = 0.1×35 + (1-0.1)×37.33
= 37.09
June 30 ?
Exponentiel smoothing: Example
February 38
March 29
April 35 37.33
May 31 37.09
June 30 = 0.1×31 + (1-0.1)×37.09
= 36.49
Why is it called Exponential
Smoothing ?
weight
Decreasing weight given
to older observations
today
Limitations of the Single Exponential
Smoothing !!!
Calculate the Trend Adjusted Forecasts for the months Jan, Feb and Mar
Exponential smoothing with Trend (Holt's
Model): Example
⚫ Calculate in January:
BJan = (0.8) (DDec) + (1 – 0.8) (FDec)
Since: FDec = BDec + TDec , then: FDec = 1020 + 0 = 1020
⚫ Calculate in February:
BFeb = (0.8) (DJan) + (1 – 0.8) (FJan)
FJan = BJan + TJan = 1020 + 0 = 1020.
BFeb = (0.8) (940) + (0.2) (1020) = 956
⚫ One can use the same trend to estimate the forecasts after k months
(this assumes that the trend remains the same).
Ft+k = Ft + k Tt = Bt + (k+1)Tt
⚫ Adjusted Forecast for June (consider k = 3) :
FJuin = BMar + 4 * TMar = 920.8 + (4) * (-33.6) = 786.4
Month Demand Base Trend F
Dec 1020 1020 0 1020
Jan 940 1020 0 1020
Feb 920 956 -32 924
Mar 950 920.8 -33.6 887.2
Apr
May
June 786.4
Exponential Smoothing with Trend and
Seasonality (Holt-Winters Model)
Exponential Smoothing with Trend and
Seasonality (Holt-Winters Model)
St = Dt / D for t=1..m
• Initialization of the trend (Tt -1) and the base (Bt -1)
• Calculation of Bt , Tt , St
• Calculation of the adjusted forecast at time period t is :
Ft = ( B t + Tt )S t−m
Exponential Smoothing with Trend and
Seasonality (Holt-Winters Model): Example
Month Demand Base (B) Trend (T) Seasonality (S) Forecast (F)
1 990 ?
2 970 ?
3 1010 990 0 ?
4 1020 ? ? ? ?
5 980 ? ? ? ?
6 960 ? ? ? ?
7 1000 ? ? ? ?
Month Demand Base (B) Trend (T) Seasonality (S) Forecast (F)
1 990 1,000
2 970 0,980
3 1010 990 0 1,020
4 1020 1006 8 1,003 1014
5 980 1035,625 18,812 0,973 1033,135
6 960 979,363 -18,725 1,012 980,045
7 1000 957,996 -20,046 1,011 940,561
Example of a time series: Forecasting results
⚫ The data gathering does not require a high cost since the historical data
are often stored in the information systems
Drawbacks :
⚫ They don't take into account the new effects :
No
Rely on qualitative
methods
For mid term and short term decisions, most companies use a forecasting approach
that relies on quantitative methods (mainly extrapolation methods) adjusted by
knowledge of marketing/sales
Quantitative Qualitative
approaches approaches
Expert
Extrapolation Causal
methods judgment
methods
Marketing/Sales