•Kateryna Kyzyma

•Daria Troian
•Adriano Aucello
•Abedelazez Safi
 THE PRESENTATION WILL LOOK AT
 TWO WAYS TO CONSIDER THE DATA
 INITIAL DATA PLOTS
 DIFFERENCIATING
 SARIMA PREDICTION
 HOLT-WINTER PREDICTION
 CONCLUSION ON BETTER PREDICTION
 INITIAL DATA PLOT

1e+06
8e+05
Series 1

6e+05
4e+05

1980 1985 1990 1995

Time
 HISTORICAL
BACKGROUND
 CORRELOGRAM
Series 1

1.0
0.8
0.6
ACF

0.4
0.2
0.0

0.0 0.5 1.0 1.5

Lag
 PARTIAL
AUTOCORRELATION
FUNCTION
Series xt

1 .0
0 .5
P artial A C F

0 .0
-0.5

0.5 1.0 1.5

Lag
REGULAR DIFFERENTIATON
 REGULAR
DIFFERENTIATING

100000
50000
Series 1

0
-50000

1980 1985 1990 1995

Time
 STANDARD
DIFFERENTIATING
Series 1

1.0
0.8
0.6
ACF

0.4
0.2
0.0
-0.2

0.0 0.5 1.0 1.5

Lag
 STANDARD
DIFFERENTIATING
Series residui

1.0
0.8
0.6

Seasonal
ACF

adjustment
0.4
0.2
0.0

0.0 0.5 1.0 1.5

Lag
SEASONAL DIFFERNTIATING
 SEASONAL RANDOM WALK
MODEL
5e+04
0e+00
residui

Series residui
-5e+04

1.0
0.8
-1e+05

0.6
0.4
1980 1985 1990 1995

ACF
Time

0.2
0.0
-0.2
-0.4

0.0 0.5 1.0 1.5

Lag
 EXPONENTIAL SMOOTHING
MODEL
60000
40000
20000
residui

Series residui
0
-20000

1.0
-60000 -40000

0.8
0.6
1980 1985 1990 1995

ACF
Time

0.4
0.2
0.0

0.0 0.5 1.0 1.5

Lag
SARIMA PREDICTION
 PAST VALUES +PREDICTED
VALUES

1e+06
8e+05
Series 1

6e+05
4e+05

1980 1985 1990 1995

Time
HOLT-WINTERS
HOLT –WINTERS FILTERING
FOR 6 MONTH
Holt-Winters filtering

1e+06
8e+05
Observed / Fitted

6e+05
4e+05

1980 1985 1990 1995

Time
 season trend level xhat
0.90 1.00 1.10 -10000 0 10000 4e+05 7e+05 1e+06
4e+05 8e+05

1980
1985

Time
fitted(m)

1990
1995
HOLT –WINTERS
PROCEDURE
 PREDICTED VALUES

 fit upr lwr

Sep 1995 731728.2 773668.6 689787.8

Oct 1995 681084.8 734762.3 627407.4

Nov 1995 675398.6 742190.4 608606.9

Dec 1995 722942.8 807768.6 638117.1

Jan 1996 763692.3 867848.4 659536.3

Feb 1996 793376.2 917302.0 669450.3

 TAKING THE LAST
VARIABLES

13.8
13.7
Series 1

13.6
13.5

1991 1992 1993 1994 1995

Time
 CORRELOGRAM
Series 1

1.0
0.8
0.6
0.4
ACF

0.2
0.0
-0.2

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
 PARTIAL
AUTOCORRELATION
Series xt

0.8
0.6
0.4
Partial ACF

0.2
0.0
-0.2

0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
 STANDARD
DIFFENTIATION

0.10
0.05
Series 1

0.00
-0.05
-0.10

1991 1992 1993 1994 1995

Time
 STANDARD
DIFFERENTIATION

Series 1

1.0
0.8
12

0.6
0.4
ACF

0.2
0.0
-0.2

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
SEASONAL DIFFERENTIATION
 SEASONAL
DIFFERENTIATION

0.05
0.00
Series 1

-0.05
-0.10

1992.0 1992.5 1993.0 1993.5 1994.0 1994.5 1995.0

Time
 SEASONAL
DIFFERENTIATION
Series 1

1.0
0.8
0.6
0.4
ACF

0.2
0.0
-0.2
-0.4

0.0 0.2 0.4 0.6 0.8 1.0 1.2

Lag
 SEASONAL RANDOM WALK
MODEL
5e+04
0e+00
residui

Series residui
-5e+04

1.0
0.8
-1e+05

0.6
0.4
1991 1992 1993 1994 1995

ACF

0.2
Time

0.0
-0.2
-0.4

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
 SARIMA(1,1,0)X(1,1,0)

5e+04
0e+00
residui

Series residui
-5e+04

1.0
-1e+05

0.8
0.6
1991 1992 1993 1994 1995

0.4
ACF
Time

0.2
0.0
-0.2

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
 SEASONAL DIFFERENTIATION
WITH SARIMA(0,1,1)X(0,1,1)

Series residui
5e+04

1.0
0e+00

0.8
residui

0.6
-5e+04

0.4
ACF
-1e+05

0.2
0.0

1991 1992 1993 1994 1995

Time
-0.2

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Lag
SARIMA PREDICTION
 SARIMA(0,1,1)X(0,1,1)
1e+06
9e+05
8e+05
Series 1

7e+05
6e+05
5e+05
4e+05

1991 1992 1993 1994 1995 1996 1997 1998

Time
HOLT-WINTER PROCEDURE
HOLT-WINTER PROCEDURE

Holt-Winters filtering

1e+06
9e+05
Observed / Fitted

8e+05
7e+05

1992.0 1992.5 1993.0 1993.5 1994.0 1994.5 1995.0

Time
HOLT-WINTER PROCEDURE
fitted(m)

9e+05
xhat
7e+05
850000
level
700000
-5000 0 5000
trend
50000
season
0
-50000

1992.0 1992.5 1993.0 1993.5 1994.0 1994.5 1995.0

Time
 PREDICTED VALUES
fit lwr uppr
Jun 1995 705292.2 785938.4 624646.03
Jul 1995 689001.1 805791.5 572210.82
Aug 1995 672864.2 819280.9 526447.55
Sep 1995 716439.8 889433.2 543446.38
Oct 1995 638669.5 836497.6 440841.44
Nov 1995 643078.4 864652.0 421504.81
Dec 1995 731008.1 975615.6 486400.71
Jan 1996 753654.5 1020823.2 486485.91
Feb 1996 777571.8 1066990.2 488153.32
 SUMMARY

Last observed values: Whole time series:

 Coefficients:  Coefficients:

ma1 sma1 ma1 sma1

0.4600 0.1709 -0.0119 -0.8163

s.e. 0.1539 0.2504 s.e. 0.0634 0.0634

Despite the positive result of residual analysis of last values the

standards error indicates the suitability of whole time series
consideration to produce the better forecast
 Series 1

4e+05 6e+05 8e+05 1e+06

1980
1985
1990
1995
Historical reasons:
 Economical recessions-1980s,1990s
 10 years cycle of unemployment(peak in 2000
proves the cyclicality )
 Baby boom late 1970s-women are on child care
holidays.