Professional Documents
Culture Documents
I. Introduction
-- What is forecasting?
Scientific (educated) guess
Based on past data or experience
Rarely perfect
Group forecast more accurate
Shorter time horizon, more accurate
-- Why forecasting?
-- Forecasting time horizon
Short range usually less than 3 months
Medium range 3 months to 3 years
Long range over 3 years
-- Which forecasting method to use?
Time horizon
Costs vs. accuracy
Easy to understand?
-- Pattern of Data
Trend gradual upward or downward movement
Seasonality repeated pattern
Cycle background pattern over long period of time
Randomness unexplainable, unpredictable, unknown fluctuations
II. Qualitative Forecasting Methods
- no past data is available, for managerial decision, long term
forecast
Jury of executive opinion
The Delphi technique systematic survey of experts
Sales force composite
Consumer market survey
III. Quantitative Forecasting Methods
1) Smoothing Techniques:
- for data with no clear pattern, short term forecast
Naive: Ft+1 = Yt
1
72
2
82
3
85
4
90
5
?
best intercept: a y b x
where x and y are averages of xs and ys
-- Example continues:
X
y
xy
x2
* Excel commands:
calculating b:=slope(range of ys, range of xs)
calculating a:=intercept(range of ys, range of xs)
y2
-- Example continues:
Step 3. Find forecast interval, if necessary
Forecast interval = Fnext +- Z(1+ /2 *Sy,x
a y b xy
n2
-- Example continues:
* Excel commands:
calculating Sy,x:=steyx(range of ys, range of xs)
3) Causal Model by Linear Regression
- same set up as trend projection, except that x is an independent
variable (other than time), y is a dependent variable, medium term
-- Example: The sales manager of a large apartment rental complex feels
the demand for apartments may be related to the number of newspaper ads
placed during the previous month. She has collected the data shown
below.
Ads
15
9
40
20
25
25
15
35
Rental
6
4
16
6
13
9
10
16
If the number of ads placed in this month is 30, what would be her
estimate of rentals in the coming month?
-- Correlation coefficient r: measures the strength of linear
relationship between x and y
-1 r +1
r =
n x
n xy x y
( x ) 2 * n y 2 ( y ) 2
* Excel command:
calculating r:=correl(range of ys, range of xs)
-- Interpretation:
r > 0:
r = 0:
r < 0:
r2 = coefficient of determination: % of variation in the dependent
variable (y) is explained by regression equation (linear
relationship).
-- Example continued: How strong is the relationship between the ads
placed and the rentals?
2.
3.
Decompose the past data (filter out the seasonal influence from
original data)
Forecast trend pattern and seasonality pattern separately
Combine the forecasts using the multiplicative model: Yt = Tt *
St
season.
=
=
=
Sale
(Tt)
2001
(t)
5
6
7
8
Sale
(Tt)
2002
(t)
9
10
11
12
sale
(Tt)
T16 =
F15 =
F16 =
68
|Err|
-1.7
2.6
-0.1
-0.8
MAD
0
b=
5.7
Err Square |Err|/Actual
1.7
2.89
0.023611
2.6
6.76
0.031707
0.1
0.01
0.001176
0.8
0.64
0.008889
MSE
MAPE
1.3
2.575
0.016346