Vous êtes sur la page 1sur 11

Case Study: Predicting U.S.

Saving Behavior
Group 3

Statistical Properties
5%
tLags
10% level
Prob.
level
Statistic
(BIC)
-3.46446 2.87643 -2.574788 -1.286847 0.6356
2
5
Log(RC)
-3.46446 2.87643 -2.574788 -2.213084 0.2024
0
5
Log(RDY)
-3.46446 2.87643 -2.574788 -0.242058 0.9295
0
5
Log(Net Worth)
-2.572754
-0.88207
0
For eachPrices)
variable, we3.455786
cannot reject
the
hipothesis
of a unit0.7929
root at
2.87263
Log(Equity
conventional levels of significance.2.89395 -2.584126 -0.928564 0.7749
5
3.504727
6
Log(House Prices)
2.87659 -2.574874 -2.478584 0.1223
1
3.464827
5
Prime Rate
Dickey Fuller Tests. H0: Series has unit root
Dickey Fuller Tests

1% level

Significance of variables: Long


Run

Log(RC)

Adjusted R-squared

S.E. of regression

Coefficie Std.
tProb.
nt
Error Statistic
0.9129 0.0282 32.3893 0.0000

LOG(RDY)
LOG(NET_WORT
0.0532
H)
PRIME_RATE
-0.0047
C
0.1737
R-squared

0.0116

4.5720

0.0000

0.0005
0.1316

-9.3305
1.3198

0.0000
0.1885

Mean dependent var


8.356899
S.D. dependent
0.999239 var

0.472425
Sum squared
0.013032 resid

0.03176
Long-run
0.435088 variance

0.000461
Long-run covariance estimate (Bartlett kernel, Newey-West fixed bandwidth = 5)

0.999251

All variables have the correct sign.

Durbin-Watson stat
All variables are statistically different from zero.

Estimation method is Fully Modified Least Squares

Significance of variables: Long


Run

Dickey Fuller Test

1% level

Residuals from log run


Cointegrating Equation

-3.465

5% 10% level
tlevel
Statistic
-2.877
-2.575
-4.767

Prob.
0.000

Lags
(BIC)
0.000

Dickey Fuller Tests. H0: Series has unit root

Residuals from the long run equation are stationary

Short Run Equation


Log(RC)
C
Coint Eq ( )
DLOG(RC(-1),0,4)
DLOG(RC(-3),0,4)
DLOG(RC(-4),0,4)
DLOG(RDY,0,4)
D(PRIME_RATE(-1),0,4)
DLOG(EQUITY_PRICES,0,4)
DLOG(HOUSE_PRICES(1),0,4)
R-squared
Adjusted R-squared
S.E. of regression

Coefficie
nt
-0.002
-0.111
0.578
0.424
-0.274
0.193
-0.001
0.016

Std.
Error
0.002
0.064
0.087
0.100
0.093
0.051
0.000
0.004

tStatistic
-0.647
-1.737
6.646
4.226
-2.952
3.785
-2.130
3.820

0.0592

0.0251

2.3560

0.863459

Mean dependent var


S.D.
0.847854 dependent var

0.004728 Akaike info criterion


Schwarz
0.001565 criterion

315.6614 Hannan-Quinn criter.


55.33325 Durbin-Watson stat
0

Prob.
0.520
0.087
0.000
0.000
0.004
0.000
0.037
0.000
0.0213
0.032149
0.012122
-7.763579

Sum variables
squared resid
-7.493642
All
are statistically significant.
Log likelihood
-7.655434
F-statistic
Log
of 4th difference has an easy interpretation: Yearly Growth Rates1.857313
Prob(F-statistic)

Criteria like R2 indicate a good fit.


Variables were selected using a general to particular approach.

Short Run Equation: Actual vs.


Fitted
.06
.04
.02
.015
.00

.010
.005

-.02

.000
-.005
-.010
-.015
1988

1990

1992

1994

1996

1998

Residual

2000

Actual

2002

2004

2006

Fitted

Residuals exhibit no autocorrelation:


Sample Errors
AR(1)

Coefficien Std.
t-Statistic Prob.
t
Error
0.0706
0.1136
0.6212
0.5363

Short Run Equation:


Diagnostics
.015
.010
.005
.000
-.005
-.010
-.015
1990

1992

1994

1996

1998

2000

Recursive Residuals

2002
2 S.E.

2004

2006

Short Run Equation:


Diagnostics
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
1990

1992

1994

1996

1998

CUSUM of Squares

2000

2002

2004

5% Significance

2006

Out of Sample Forecasts


8
7
6
5
4
3
2
1
I

II

III

2007

IV

II

III

2008

IV

II

III

2009

In blue: Actual Savings rate


In red: Actual Interest & transfers income
In green: Forecasted Interest & transfers income

IV

II

III

2010

IV

Out of Sample Forecasts


Forecast Performance
RMSE
Theil 1
Theil 2
Bias in the forecast
Proportion due to variance
Proportion due to covariance
Mean forecast error
Mean percentage error
Mean absolute error
Mean absolute percentage forecast error

Actual
Interest &
Transfer
Expenditur
e

Forecasted
Interest &
Transfer
Expenditure

0.771
0.075
0.496
0.512
0.021
0.467
0.552
11.080
0.644
12.720

1.072
0.109
0.660
0.782
0.003
0.215
0.948
18.135
0.948
18.135

Transitory Nature of Savings Rate Upswing


8
7
6

5
4
3
2
I

II

III

IV

2008

II

III

IV

2009
SRATE_SR

SRATE_LR

II

III

IV

2010
SRATE

Forecast from the Long Run Equation (in red) shows that the recent
upswing comes from a transitory shock in the US Economy

Vous aimerez peut-être aussi