FORECASTING
Chapter 17
FORECASTING 1
Forecasting Approaches
◆ Qualitative Methods:
– Jury of executive opinion
– Sales force composite
– Delphi method
– Consumer market survey
◆ Quantitative Methods:
Time series models: assume that the future
is a function of the past.
– Naive approach
– Moving averages
– Exponential smoothing
– Trend projection
– Seasonal Indexing
Associative models: find explanatory
factors.
– Linear regression
FORECASTING 2
Components of a Time Series
◆ Time Series: set of evenly spaced
numerical data , e.g., Weekly Sales, Quarterly
Earnings, Daily Major Market Indices.
* Trend (T): persistent, overall upward or
downward pattern.
* Seasonality (S) is the regular pattern of up &
down fluctuations which occurs every year.
* Cycles (C) are repeating up & down
movements in the data that occur every several
years. They are usually tied to the business
cycle.
* Random variations (R) are erratic,
unsystematic, ‘residual’ fluctuations due to
random variation or unforeseen events; they
follow no discernible pattern.
FORECASTING 3
Product Demand Charted over 4
Years with Trend and
Seasonality
Seasonal peaks Trend component
Demand for product or service
Actual
demand
line
Average
demand
over four
Rand years
om
variat
ion
Year Year Year Year
1 2 3 4
FORECASTING 4
Simple Moving Average
◆ Use the average of the most recent few
periods’ actual data as the forecast
value.
◆ When market demand is assumed to
stay fairly stable over time.
Simple moving average = (Sum of
demand in previous n periods) / n
Where n is the number of periods
included in the moving average.
FORECASTING 5
Simple Moving Average (cont.)
◆ Example:
W eek Actual 3-week
Sales MA
1 80 −
2 100 −
3 90 −
4 110
5 90
6 130
7
7 −
FORECASTING 6
Weighted Moving Average
◆ Assigns different weights to previous
actual demand values.
◆ If more recent data are weighted more
heavily, the forecast will be more
responsive to changes in the demand.
Weighted moving average =
∑ ( weighti )(demandi )
∑ ( weights)
FORECASTING 7
Weighted Moving Average
(cont.)
◆ Example: with (6, 3, 1) as weights,
using 6 for the most recent week.
Week Actual 3-week WMA
Sales weights=(6,3,1)
1 80 −
2 100 −
3 90 −
4 110
5 90
6 130
7
7 −
FORECASTING 8
Exponential Smoothing
◆ New forecast = last period’s forecast
+ a fraction of last period’s forecast error
Ft = Ft −1 + α (A t −1 − Ft −1 )
where,
Ft = new forecast
Ft-1 = previous forecast
α = smoothing constant (0 ≤ α ≤ 1)
A t −1 = previous period's actual demand
FORECASTING 9
Exponential Smoothing
(cont.)
◆ Example (α = 0.3)
W eek A ctual Exponential
Sales Sm oothing
1 80 −
2 100 −
3 90 −
4 110 90 (given)
5 90
6 130
7 −
7
FORECASTING 10
Error Measurement
◆ Using an error measurement, we can
select the best forecasting model from
several candidates.
◆ Overall forecast error of a model can be
measured in several ways:
Forecast error = A t − Ft
∑ forecast errors
MAD =
n
∑ ( forecast errors ) 2
MSE =
n
100∑ ( A t − Ft /A t )
MAPE = %
n
FORECASTING 11
Error Measurement
(cont.)
◆ Example (our SMA forecasts)
W eek A F A-F |A-F| (A-F)2 |A-F|/A
(in %)
4 110 90
5 90 100
6 130 96.7
sum − − −
MAD =
MSE =
MAPE =
FORECASTING 12
Exponential Smoothing
with Trend Adjustment
◆ Adding a trend component to the
simple exponential smoothing forecast.
Ft = α × At −1 + (1 − α ) × ( Ft −1 + Tt −1 )
Tt = β × ( Ft − Ft −1 ) + (1 − β ) × Tt −1
FITt = Ft + Tt
Ft = ES forecast in period t
Tt = trend value in period t
At = actual demand in period t
α = smoothing constant (0 ≤ α ≤ 1)
β = smoothing constant (0 ≤ β ≤ 1)
FORECASTING 13
Exponential Smoothing
with Trend Adjustment (cont.)
◆ Example
Let α = 0.7 , β = 0.4
Month A F T FIT
14 2
1 19 16
(given) (given)
2 24
3 30
FORECASTING 14
Trend Projections
◆ Fit a trend line to historical data and
project it into the future for medium-to-
long-range forecasts.
Trend Line : yˆ = a + bx
∑ ( xy ) − n( x )( y )
slope = b = 2 2
∑ ( x ) − n( x )
intercept = a = y − bx
x = index for periods, start with 1,
y = actual time series data,
x = average of the x values,
y = average of the y values,
yˆ = forecast values.
FORECASTING 15
Trend Projections (cont.)
◆ Example
year x y x2 xy
1 100
2 110
3 122
4 130
5 139
6 152
7 164
49
sum 917
FORECASTING 16
Trend Projections (cont.)
◆ Example
T r e n d P r o je c t i o n
200
180
160
140
120
100
80
1 2 3 4 5 6 7 8 9
FORECASTING 17
Trend Projections (cont.)
◆ Example
T r e n d P r o j e c t io n
200
180
160
140
120
100
80
1 2 3 4 5 6 7 8 9
FORECASTING 18
Seasonal Variations
◆ From year-to-year data find seasonal
indexes, and apply them to forecast
values.
(1) Find average yearly demand by period
(2) Find the average of the values
obtained in step (1) above
(3) Period-by-period, obtain
seasonal index = (1) / (2)
FORECASTING 19
Seasonal Variations (cont.)
◆Example
YR1 YR2 Avg Avg seasonal
Qtr sales sales YR1& quarterly index
YR2 demand
I 235 275
II 314 360
III 250 320
IV 230 250
sum − − − −
If our forecast for YR3 is 1600 units, what are
the quarterly forecasts?
YR3 Q I =
YR3 Q II =
YR3 Q III =
YR3 Q IV =
FORECASTING 20
Associative Model:
Simple Linear Regression
◆ Develop a linear relationship between
the independent variable (x) and the
dependent variable (y).
Regression Line : yˆ = a + bx
◆ Can set up confidence intervals around
the point estimate.
◆ Coefficient of Correlation, r, measures
the strength of the linear relationship
between x and y.
◆ Coefficient of determination, r2, tells us
the % of variation in y values that is
explained by the regression equation.
FORECASTING 21
Tracking Signal
◆ Measures how well the forecast is
predicting actual values.
◆ Needs to be re-computed at the end of
every period after the most current
actual data is gathered.
Tracking Signal
∑ (Ai − Fi )
RSFE i
= = , and
MAD MAD
∑ A i − Fi
MAD = i
n
FORECASTING 22
Plot of a Tracking Signal
Signal exceeded
limit
Tracking signal
Upper
+ control limit
0 Acceptable
MAD
range
-
Lower control
limit
Time
FORECASTING 23
Tracking Signal (cont.)
◆ Example
Month Ai Fi Ai - Fi |Ai - Fi|
1 71 78
2 80 75
3 101 83
4 84 84
5 60 88
6 73 85
sum − −
RSFE =
MAD =
Tracking Signal =
FORECASTING 24