Basic Forecasting Methods
Basic Forecasting Methods
simple models
Outline
Basic forecasting models
– The basic ideas behind each model
– When each model may be appropriate
– Illustrate with examples
Forecast error measures
Automatic model selection
Adaptive smoothing methods
– (automatic alpha adaptation)
Ideas in model based forecasting techniques
– Regression
– Autocorrelation
– Prediction intervals
2
Basic Forecasting Models
3
M-Period Moving Average
m 1
Vt j
Pt 1(t) j 0
M
i.e. the average of the last M data points
Basically assumes a stable (trend free) series
How should we choose M?
– Advantages of large M?
– Advantages of large M?
Average age of data = M/2
4
Weighted Moving Averages
n
Pt 1(t) W V
t j t j
j 0
5
Simple Exponential
Smoothing
6
Two Views of Same Equation
7
Simple Exponential
Smoothing
Is appropriate when the underlying time
series behaves like a constant + Noise
– Xt = + N t
– Or when the mean is wandering around
– That is, for a quite stable process
Not appropriate when trends or seasonality
present
8
ES would work well here
4
2
0
-2
Demand
-4
-6
-8
-10
-12
-14
1
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
103
109
115
Period
9
Simple Exponential
Smoothing
We can show by recursive substitution
that ES can also be written as:
10
Weights on past data
0.7
0.6
0.5
0.4
Expo Smooth a=0.6
MoveAve(M=5)
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10
11
Simple Exponential
Smoothing
Ft+1(t) = At + (1-)At-1 + (1-)2At-2 + (1-)3At-3 +…..
12
Exponential Smoothing
Examples
13
Zero Mean White Noise
Series
0 Series
-1
-2
-3
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
14
15
-3
-2
-1
0
1
2
3
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
0.1
Series
16
-3
-2
-1
0
1
2
3
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
0.3
0.1
Series
Shifting Mean + Zero Mean White Noise
Series
0
Mean
-1
-2
-3
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
-4
17
18
-4
-3
-2
-1
0
1
2
3
4
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
0.1
Mean
Series
19
-4
-3
-2
-1
0
1
2
3
4
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
0.3
Mean
Series
Automatic selection of
20
RMSE vs Alpha
1.45
1.4
1.35
RMSE
1.3
1.25
1.2
1.15
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Alpha
21
Recommended Alpha
22
23
Time Series Value
-2
-1
0
1
2
-1.5
-0.5
0.5
1.5
1
6
11
16
21
26
31
36
41
46
51
Original Data
Period
56
61
66
71
76
81
86
91
96
Actual vs Forecast for
Various Alpha
1.5
0.5 Demand
Forecast
a=0.1
0
a=0.3
-0.5 a=0.9
-1
-1.5
-2
1
7
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
Period
24
Series and Forecast using Alpha=0.9
0.5
Forecast
-0.5
-1
-1.5
-2
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
Period
25
Series and Forecast using Alpha=0.9
1.5
0.5
Forecast
-0.5
-1
-1.5
-2
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
Period
26
Series and Forecast using Alpha=0.9
1.5
1
Forecast
0.5
-0.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
27
Series and Forecast using Alpha=0.9
1.5
1
Forecast
0.5
-0.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
28
Forecast RMSE vs Alpha
0.67
0.66
0.65
0.64
Forecast RMSE
0.63
0.62 Series1
0.61
0.6
0.59
0.58
0.57
0 0.2 0.4 0.6 0.8 1
Alpha
29
Exponential Smoothing on Lake Huron Level Data
Various Alphas
13
12
11
Forecast and actual
10 HuronLevel
a=0.1
9
a=0.3
8 a=0.9
5
1
7
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
Period
30
Forecast Errors for Lake Huron Data
Various Alphas
1
Forecast Error
a=0.1
0
a=0.9
-1
-2
-3
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
Period
31
Forecast RMSE vs Alpha
for Lake Huron Data
1.1
1.05
1
0.95
0.9
RMSE
0.85
0.8
0.75
0.7
0.65
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Alpha
32
Monthly Furniture Demand vs Forecast
Various Alphas
160
140
120
100 Demand
Orders
a=0.1
80
a=0.3
60 a=0.9
40
20
0
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
Period
33
Monthly Furniture Demand Forecast Errors
Various Alphas
80
60
40
20
Forecast Error
a=0.1
0
a=0.3
-20
a=0.9
-40
-60
-80
-100
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
Period
34
Forecast RMSE vs Alpha
for Monthly Furniture Demand Data
45.6
40.6
35.6
30.6
RMSE
25.6
20.6
15.6
10.6
5.6
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Alpha
35
Exponential smoothing will
lag behind a trend
Suppose Xt=b0+ b1t
And St= (1- )St-1 + Xt
Can show that
E St E X t
(1
) b0
36
Exponential Smoothing on a Trend
12
10
8
Trend Data
6 0.2
0.5
4
0
1 2 3 4 5 6 7 8 9 10 11 12
Period
37
Double Exponential
Smoothing
Modifies exponential smoothing for following
a linear trend
Let St (1 )St1X t
38
Single and Double smoothed values
16
14
12
10
St Lags Trend Data
8 Single, a=0.5
Double, a=0.5
6
39
Double Smoothing
20
18
16
14
2St -St[2] doesn’t lag
12
Trend Data
10
2(S(t)-S2(t)
8
6
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Period
40
E St E X t
1 b1
E St
[2]
E St 1
b1
b1
E St E St
[2]
1
1
41
E X t E St
1
b1
1 [2]
E X t E St
E S E S
1 t t
E
X t 2E St E St
[2]
Xˆ t 2 St St[2]
42
Xˆ t 2St St[2]
Xˆ t Xˆ t bˆ1
[2]
[2]
Xˆ t 2St St St St
1
Xˆ t 2 St 1 St[2]
1 1
43
44
-1
0
1
2
3
4
5
6
1
11
16
21
Example
26
31
36
41
46
Trend
Series
51
56
61
66
71
76
81
86
91
96
101
6
3
=0.2
Trend
2 Series Data
Single Smoothing
1 Double smoothing
-1
1
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
45
6
4
Single Lags a trend
3
Trend
2 Series Data
Single Smoothing
1 Double smoothing
-1
1
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
46
6
4
Double Over-shoots a change
3 (must “re-learn” the slope)
Trend
2 Series Data
Single Smoothing
1 Double smoothing
-1
101
41
16
21
26
31
36
46
51
56
61
66
71
76
81
86
91
96
1
11
47
Holt-Winters Trend and
Seasonal Methods
“Exponential smoothing for data with trend
and/or seasonality”
– Two models, Multiplicative and Additive
Models contain estimates of trend and
seasonal components
Models “smooth”, i.e. place greater weight on
more recent data
48
Winters Multiplicative Model
Xt = (b1+b2t)ct + t
Where ct are seasonal terms and
L
ct L where L is the season length
t 1
Note that the amplitude depends on the level
of the series
Once we start smoothing, the seasonal
components may not add to L
49
Holt-Winters Trend Model
Xt = (b1+b2t) + t
Same except no seasonal effect
Works the same as the trend + season model
except simpler
50
Example:
1.5
X t 1 0.04t 0.5
1
51
Xt =(1 + 0.04t)(1.5,0.5,1)
4.5
3.5
(1+0.04t)
3
2.5
1.5
0.5
0
1
10
13
16
19
22
25
28
31
34
37
40
43
46
49
52
55
58
52
Xt =(1 + 0.04t)(1.5,0.5,1)
4.5
4
*150%
3.5
2.5
1.5
0.5
0
1
10
13
16
19
22
25
28
31
34
37
40
43
46
49
52
55
58
53
Xt =(1 + 0.04t)(1.5,0.5,1)
4.5
3.5
2.5
*50%
2
1.5
0.5
0
1
10
13
16
19
22
25
28
31
34
37
40
43
46
49
52
55
58
54
The seasonal terms average 100% (i.e. 1)
Thus summed over a season, the ct must add
to L
Each period we go up or down some
percentage of the current level value
The amplitude increasing with level seems to
occur frequently in practice
55
Recall Australian Red Wine Sales
Series
3000.
2500.
2000.
1500.
1000.
500.
56
Smoothing
57
Step 1. Update the Permanent Component
aˆ1(T )
T
cˆ (T L)
1 2
T
58
Step 1. Update the Permanent Component
aˆ1(T )
T
cˆ (T L)
1 2
T
59
Step 1. Update the Permanent Component
aˆ1(T )
T
cˆ (T L)
1 2
T
60
Step 1. Update the Permanent Component
aˆ1(T )
T
cˆ (T L)
1 2
T
aˆ1(T )
T
cˆ (T L)
1 2
T
62
Step 2. Update the Trend Component
63
Step 2. Update the Trend Component
“observed” slope
64
Step 2. Update the Trend Component
65
Step 3. Update the Seasonal Component
for this period
V
cˆT (T ) T (1 )cˆT (T L)
aˆ (T )
1
66
To forecast ahead at time T
use current values of a, b, and c
VˆT (T ) aˆ (T ) bˆ (T ) cˆT (T L)
1 2
67
To forecast ahead at time T
use current values of a, b, and c
VˆT (T ) aˆ (T ) bˆ (T ) cˆT (T L)
1 2
68
To forecast ahead at time T
use current values of a, b, and c
VˆT (T ) aˆ (T ) bˆ (T ) cˆT (T L)
1 2
69
Winters Additive Method
Xt = b1+ b2t + ct + t
Where ct are seasonal terms and
L
ct 0 where L is the season length
t 1
Similar to previous model except we
“smooth” estimates of b1, b2, and the ct
70
Croston’s Method
71
Demand Distribution
0.8
0.7
0.6
0.5
Probability
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9
Demand
72
Typical situation
73
Example
Demand Prob Demand each
0 0.85 period follows a
1 0.1275 distribution that
2 0.0191 is usually zero
3 0.0029
4 0.0004
5 0.00006
6 0.00001
7 0.000002
74
75
Demand
0.5
1.5
2.5
3.5
0
1
2
3
1
14
27
40
53
66
79
92
Example
105
118
131
144
157
170
183
196
Period
209
222
235
An intermittent Demand Series
248
261
274
287
300
313
326
339
352
365
378
391
Example
0.9
0.8
0.7
0.6
Demand
0.5
0.4
0.3
0.2
0.1
0
1
19
37
55
73
91
109
127
145
163
181
199
217
235
253
271
289
307
325
343
361
379
397
Period
76
Using Exponential Smoothing:
77
Croston’s Method
Separately Tracks
– Time between (non-zero) demands
– Demand size when not zero
Smoothes both time between and demand
size
Combines both for forecasting
Demand Size
Forecast =
Time between demands
78
Define terms
79
Forecast Update
q=q+1
– Keep counting time since last order
80
Forecast Update
– Z(t)=Z(t-1) + (V(t)-Z(t-1))
– X(t)=X(t-1) + (q - X(t-1))
– q=1
81
Forecast Update
– Z(t)=Z(t-1) + (V(t)-Z(t-1))
– X(t)=X(t-1) + (q - X(t-1))
– q=1 Latest
order size
Update Size of order via smoothing
82
Forecast Update
– Z(t)=Z(t-1) + (V(t)-Z(t-1))
– X(t)=X(t-1) + (q - X(t-1))
– q=1 Latest time
between orders
Update size of order via smoothing
Update time between orders via smoothing
83
Forecast Update
– Z(t)=Z(t-1) + (V(t)-Z(t-1))
– X(t)=X(t-1) + (q - X(t-1))
– q=1 Reset
counter
Update size of order via smoothing
Update time between orders via smoothing
Reset counter of time between orders
84
Forecast
85
Recall example
0.5
0.4
0.3
0.2
0.1
0
1
12
23
34
45
56
67
78
89
100
111
122
133
144
155
166
177
188
199
210
221
232
243
254
265
276
287
298
309
320
331
342
353
364
375
386
397
Period
86
Recall example
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
12
23
34
45
56
67
78
89
100
111
122
133
144
155
166
177
188
199
210
221
232
243
254
265
276
287
298
309
320
331
342
353
364
375
386
397
87
What is it forecasting?
0.8
0.7
0.6
True average demand per period=0.176
0.5
0.4
0.3
0.2
0.1
0
1
12
23
34
45
56
67
78
89
100
111
122
133
144
155
166
177
188
199
210
221
232
243
254
265
276
287
298
309
320
331
342
353
364
375
386
397
88
Behavior
89
Croston’s Method
90
Counter Example
91
Better Examples
92
Is demand Independent?
93
Theoretical behavior
12
10
8
Fequency
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Time Between
94
In our example:
14
12
10
Frequency
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Time Between
95
Comparison
14
12
10
Frequency
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Time Between
96
Counterexample
12
10
8
Frequency
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time Between
97
Counterexample
98
Error Measures
Errors: The difference between actual and
predicted (one period earlier)
et = Vt – Pt(t-1)
– et =can be positive or negative
Absolute error |et|
– Always positive
Squared Error et2
– Always positive
The percentage error PEt = 100et / Vt
– Can be positive or negative
99
Bias and error magnitude
100
Error Measures
Look at errors over time
Cumulative measures summed or averaged
over all data
– Error Total (ET)
– Mean Percentage Error (MPE)
– Mean Absolute Percentage Error (MAPE)
– Mean Squared Error (MSE)
– Root Mean Squared Error (RMSE)
Smoothed measures reflects errors in the
recent past
– Mean Absolute Deviation (MAD)
101
Error Measures Measure Bias
Look at errors over time
Cumulative measures summed or averaged
over all data
– Error Total (ET)
– Mean Percentage Error (MPE)
– Mean Absolute Percentage Error (MAPE)
– Mean Squared Error (MSE)
– Root Mean Squared Error (RMSE)
Smoothed measures reflects errors in the
recent past
– Mean Absolute Deviation (MAD)
102
Error Measures Measure error
Look at errors over time magnitude
Cumulative measures summed or averaged
over all data
– Error Total (ET)
– Mean Percentage Error (MPE)
– Mean Absolute Percentage Error (MAPE)
– Mean Squared Error (MSE)
– Root Mean Squared Error (RMSE)
Smoothed measures reflects errors in the
recent past
– Mean Absolute Deviation (MAD)
103
Error Total
Sum of all errors
n
ET e
t
t 1
Uses raw (positive or negative) errors
ET can be positive or negative
Measures bias in the forecast
Should stay close to zero as we saw in last
presentation
104
MPE
1 n
MPE PE
nt 1 t
105
MSE
1 n
MSE e 2
nt 1 t
Always positive
Measures “magnitude” of errors
Units are “demand units squared”
106
RMSE
Square root of MSE
1 n 2
RMSE et
n t 1
Always positive
Measures “magnitude” of errors
Units are “demand units”
Standard deviation of forecast errors
107
MAPE
1 n
MAPE PE
nt 1 t
Always positive
Measures magnitude of errors
Units are “percentage”
108
Mean Absolute Deviation
109
Percentage or Actual units
110
Squared or Absolute Errors
111
Ex-Post Forecast Errors
Given
– A forecasting method
– Historical data
Calculate (some) error measure using the
historical data
Some data required to initialize forecasting
method.
Rest of data (if enough) used to calculate ex-
post forecast errors and measure
112
Automatic Model Selection
113
Automatic Adaptation
114
Tracking Signal (TS)
ETt
TSt
MADt
115
Adaptation
t 0.2 TS
t 1
t t 1
t 0.8 0.2TSt
t 1
subject to 0.05 0.9
If TS increases, bias is increasing, thus
increase
I don’t like these methods due to instability
116
Model Based Methods
117
Univariate Time Series
Models Based on
Decomposition
Vt = the time series to forecast
Vt = T t + St + N t
Where
– Tt is a deterministic trend component
– St is a deterministic seasonal/periodic
component
– Nt is a random noise component
118
Raw Material Price
3.8
3.6
3.4
3.2
Price ($/Unit)
3
Price
2.8
(Vt)=0.257
2.6
2.4
2.2
2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Period
119
Raw Material Price
3.8
3.6
3.4
3.2
Price ($/Unit)
3
Price
2.8
2.6
2.4
2.2
2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Period
120
Simple Linear Regression
Model:
Vt=2.877174+0.020726t
SUMMARY OUTPUT
Regression Statistics
Multiple R 0.569724
R Square 0.324585
Adjusted R Square
0.293884
Standard Error0.21616
Observations 24
ANOVA
df SS MS F Significance F
Regression 1 0.494006 0.494006 10.57257 0.003659
Residual 22 1.027956 0.046725
Total 23 1.521963
Coefficients
Standard Error t Stat P-value Lower 95% Upper 95%Lower 95.0%
Upper 95.0%
Intercept 2.877174 0.091079 31.58978 7.99E-20 2.688287 3.066061 2.688287 3.066061
X Variable 1 0.020726 0.006374 3.251549 0.003659 0.007507 0.033945 0.007507 0.033945
121
Use Model to Forecast into the Future
3.8
3.6
3.4
3.2
3
Price
Price
2.8 Forecast
2.6
2.4
2.2
2
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
Period
122
Residuals = Actual-Predicted
et = Vt-(2.877174+0.020726t)
(et)=0.211
Residuals After Regression
0.4
0.3
0.2
0.1
Residuals
0 Residuals
-0.1
-0.2
-0.3
-0.4
1
11
13
15
17
19
21
23
Period
123
Simple Seasonal Model
124
Residuals Season Residuals Season Season Averages
0.1521 1 0.1521 1
-0.24862609 2 0.27992173 1 Sorted
0.03064782 3 0.22774346 1 by season
0.27992173 1 0.19556519 1
-0.21080436 2 0.18338692 1
-0.07153045 3 0.28120865 1
0.22774346 1 0.33903038 1
-0.28298263 2 0.34685211 1 0.250726055
0.03629128 3 -0.24862609 2
0.19556519 1 -0.21080436 2
-0.1951609 2 -0.28298263 2
0.00411301 3 -0.1951609 2 Season
0.18338692 1 -0.28733917 2
-0.28733917 2 -0.20951744 2
averages
-0.00806526 3 -0.24169571 2
0.28120865 1 -0.26387398 2 -0.242500035
-0.20951744 2 0.03064782 3
-0.00024353 3 -0.07153045 3
0.33903038 1 0.03629128 3
-0.24169571 2 0.00411301 3
-0.0424218 3 -0.00806526 3
0.34685211 1 -0.00024353 3
-0.26387398 2 -0.0424218 3
-0.01460007 3 -0.01460007 3 -0.008226125
125
Trend + Seasonal Model
Vt=2.877174+0.020726t + Smod(t,3)
Where
– S1 = 0.250726055
– S2 = -0.242500035
– S3 = -0.008226125
126
Actual vs Forecast (Trend + Seasonal Model)
3.8
3.6
3.4
3.2
Price
Price
3
Forecast
2.8
2.6
2.4
2.2
2
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
Period
127
et = Vt - (2.877174 + 0.020726t + Smod(t,3))
(et)=0.145
Residuals from Trend+Season
0.15
0.1
0.05
Residuals
0 Residuals2
-0.05
-0.1
-0.15
1
11
13
15
17
19
21
23
Period
128
Can use other trend models
129
130
-3
-2
-1
0
1
2
3
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
Signal=COS(2*PI*t/12)
1
Tim e t
Series1
0
Signal
-1
-2
-3
1
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
S(t)
131
Model: Vt = Tt + St + Nt
132
Zero Mean, and Aperiodic:
ˆ
Is our best forecast t 1 0 ?
N
Demand
2
Demand
-2
-4
-6
1
14
27
40
53
66
79
92
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
Period
133
AR(1) Model
Nˆ 0.9Nt
t 1
2
Nˆ 0.9Nˆ 0.9 Nt
t 2 t 1
134
AR(1): Actual vs 1-Step Ahead Forecast
2
Demand
Actual
0
Forecast
-2
-4
-6
17
33
49
65
81
97
1
113
129
145
161
177
193
209
225
241
257
273
289
Period
135
Forecasting N Steps Ahead
2
Demand
Actual
0
Forecast
-2
-4
-6
1
115
134
153
172
191
210
229
248
267
286
305
324
343
20
39
58
77
96
Period
136
Time Series Models
Nˆ 0
t 1
137
Sample Autocorrelation Function
.80 .80
.60 .60
.40 .40
.20 .20
.00 .00
-.20 -.20
-.40 -.40
-.60 -.60
-.80 -.80
-1.00 -1.00
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
138
Back to our Demand Data
0.15
0.1
0.05
Residuals
0 Residuals2
-0.05
-0.1
-0.15
1
11
13
15
17
19
21
23
Period
139
No Apparent Significant
Autocorrelation
Sample ACF Sample PACF
1.00 1.00
.80 .80
.60 .60
.40 .40
.20 .20
.00 .00
-.20 -.20
-.40 -.40
-.60 -.60
-.80 -.80
-1.00 -1.00
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
140
Multiple Linear Regression
V= 0+ 1 X1 + 2 X2 +….+ p Xp +
Where
– V is the “independent variable” you want
to predict
– The Xi‘s are the dependent variables you
want to use for prediction (known)
Model is linear in the i‘s
141
Examples of MLR in
Forecasting
Vt= 0+ 1t + 2t2 + 3Sin(2t/k) + 4ekt
– i.e a trend model, a function of t
Vt= 0+ 1X1t + 2X2t
– Where X1t and X2t are leading indicators
Vt= 0+ 1Vt-1+ 2Vt-2 + 12Vt-12 +13Vt-13
– An Autoregressive model
142
Example: Sales and Leading Indicator
Series 1 Series 2
14.00
260.
13.50
13.00 250.
12.50
240.
12.00
230.
11.50
220.
11.00
10.50 210.
10.00
200.
9.50
143
Example: Sales and Leading Indicator
Series 1 Series 2
14.00
260.
13.50
13.00 250.
12.50
240.
12.00
Sales(t) = -3.93+0.83Sales(t-3)
230.
10.50 210.
10.00
200.
9.50
144