lOMoARcPSD|4163173

Tutorial 5

Applied Financial Modelling (University of Wollongong)

StuDocu is not sponsored or endorsed by any college or university

Downloaded by Trung Kiên Tr?n ([email protected])
 lOMoARcPSD|4163173

Tutorial 5 (Week 6) hedonic

Question 1
Proc>Set sample> if – year=2002

1. OLS
Quick>Estimate Equation
Log(selling_price) c sfla beds baths stories vacant age
Sample should read 1 6660 if year=2002

Dependent Variable: LOG(SELLING_PRICE)

Method: Least Squares
Date: 09/24/20 Time: 16:08
Sample: 1 6660 IF YEAR=2002
Included observations: 1667

Variable Coefficient Std. Error t-Statistic Prob.

C 11.60161 0.027881 416.1186 0.0000

SFLA 0.000484 1.47E-05 32.97529 0.0000
BEDS -0.045381 0.009190 -4.938200 0.0000
BATHS 0.060182 0.013763 4.372722 0.0000
STORIES -0.048455 0.013434 -3.606956 0.0003
VACANT -0.027897 0.010421 -2.676928 0.0075
AGE -0.003772 0.000245 -15.38903 0.0000

R-squared 0.718174 Mean dependent var 12.15529

Adjusted R-squared 0.717156 S.D. dependent var 0.347565
S.E. of regression 0.184846 Akaike info criterion -0.534399
Sum squared resid 56.71885 Schwarz criterion -0.511645
Log likelihood 452.4215 Hannan-Quinn criter. -0.525967
F-statistic 705.0278 Durbin-Watson stat 1.380571
Prob(F-statistic) 0.000000

2. OLS with White’s Robust standard error

Quick>Estimate Equation
Log(selling_price) c sfla beds baths stories vacant age
Options> Heteroskedasticity Consistence covariance matrix> White

Dependent Variable: LOG(SELLING_PRICE)

Method: Least Squares
Date: 09/24/20 Time: 16:12
Sample: 1 6660 IF YEAR=2002
Included observations: 1667
White heteroskedasticity-consistent standard errors & covariance

Variable Coefficient Std. Error t-Statistic Prob.

C 11.60161 0.033151 349.9595 0.0000

SFLA 0.000484 1.60E-05 30.35459 0.0000
BEDS -0.045381 0.010111 -4.488072 0.0000
BATHS 0.060182 0.017219 3.495195 0.0005
STORIES -0.048455 0.012790 -3.788591 0.0002

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

VACANT -0.027897 0.010479 -2.662108 0.0078

AGE -0.003772 0.000329 -11.45822 0.0000

R-squared 0.718174 Mean dependent var 12.15529

Adjusted R-squared 0.717156 S.D. dependent var 0.347565
S.E. of regression 0.184846 Akaike info criterion -0.534399
Sum squared resid 56.71885 Schwarz criterion -0.511645
Log likelihood 452.4215 Hannan-Quinn criter. -0.525967
F-statistic 705.0278 Durbin-Watson stat 1.380571
Prob(F-statistic) 0.000000 Wald F-statistic 546.8534
Prob(Wald F-statistic) 0.000000
2 2
3. WLS (where t = ( SFLA t ))

Dependent Variable: LOG(SELLING_PRICE)

Method: Least Squares
Date: 09/24/20 Time: 16:14
Sample: 1 6660 IF YEAR=2002
Included observations: 1667
Weighting series: 1/SFLA^0.5
Weight type: Inverse standard deviation (EViews default scaling)
White heteroskedasticity-consistent standard errors & covariance

Variable Coefficient Std. Error t-Statistic Prob.

C 11.53225 0.034683 332.5067 0.0000

SFLA 0.000491 1.65E-05 29.70248 0.0000
BEDS -0.025574 0.010793 -2.369476 0.0179
BATHS 0.059369 0.017928 3.311566 0.0009
STORIES -0.052919 0.012379 -4.275083 0.0000
VACANT -0.015906 0.011169 -1.424166 0.1546
AGE -0.003765 0.000335 -11.22514 0.0000

Weighted Statistics

R-squared 0.704762 Mean dependent var 12.11001

Adjusted R-
squared 0.703695 S.D. dependent var 1.673945
S.E. of regression 0.189159 Akaike info criterion -0.488268
Sum squared
resid 59.39666 Schwarz criterion -0.465513
Log likelihood 413.9710 Hannan-Quinn criter. -0.479836
F-statistic 660.4309 Durbin-Watson stat 1.375269
Prob(F-statistic) 0.000000 Weighted mean dep. 12.06588
Wald F-statistic 528.1804 Prob(Wald F-statistic) 0.000000

Unweighted Statistics

R-squared 0.715872 Mean dependent var 12.15529

Adjusted R-
squared 0.714845 S.D. dependent var 0.347565
S.E. of regression 0.185599 Sum squared resid 57.18212
Durbin-Watson
stat 1.399464

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

2
4. WLS (where t = 2
( SFLA 2t )

Dependent Variable: LOG(SELLING_PRICE)

Method: Least Squares
Date: 09/24/20 Time: 16:20
Sample: 1 6660 IF YEAR=2002
Included observations: 1667
Weighting series: 1/SFLA
Weight type: Inverse standard deviation (EViews default scaling)

Variable Coefficient Std. Error t-Statistic Prob.

C 11.46914 0.030805 372.3140 0.0000

SFLA 0.000506 1.91E-05 26.52771 0.0000
BEDS -0.008853 0.010076 -0.878646 0.3797
BATHS 0.055515 0.014779 3.756212 0.0002
STORIES -0.056706 0.018223 -3.111841 0.0019
VACANT -0.004005 0.010493 -0.381655 0.7028
AGE -0.003741 0.000256 -14.62038 0.0000

Weighted Statistics

R-squared 0.681527 Mean dependent var 12.06588

Adjusted R-squared 0.680375 S.D. dependent var 3.596491
S.E. of regression 0.199457 Akaike info criterion -0.382246
Sum squared resid 66.03995 Schwarz criterion -0.359491
Log likelihood 325.6017 Hannan-Quinn criter. -0.373814
F-statistic 592.0610 Durbin-Watson stat 1.384151
Prob(F-statistic) 0.000000 Weighted mean dep. 11.98264

Unweighted Statistics

R-squared 0.709092 Mean dependent var 12.15529

Adjusted R-squared 0.708041 S.D. dependent var 0.347565
S.E. of regression 0.187801 Sum squared resid 58.54671
Durbin-Watson stat 1.427200

Question 2
The R2 is the highest for OLS and OLS with White’s Robust standard errors.
We would choose OLS with White’s Robust standard errors.
Since there is heteroskedasticity, the estimators from the OLS would not be BLUE. The
estimates are unbiased. In addition, the standard errors could be too large or too small. If the
standard errors are not correct, then the conclusion may be incorrect as well.

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

Thus, it would be better to choose OLS with White’s Robust standard errors as the standard
errors have been ‘corrected’.

Question 3
i. Plot residuals over time; is there any evidence of autocorrelation?

Dependent Variable: RP
Method: Least Squares
Date: 09/24/20 Time: 16:36
Sample (adjusted): 6/01/1984 9/01/2003
Included observations: 77 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C -0.009274 0.001167 -7.944185 0.0000

RASX200A 0.019926 0.016233 1.227543 0.2237
RDIVY -0.010793 0.007526 -1.434087 0.1559
RAPS -0.020642 0.026227 -0.787040 0.4339
RROI 0.971941 0.012674 76.69009 0.0000
REI -0.002595 0.007707 -0.336723 0.7373

R-squared 0.994817 Mean dependent var 0.020597

Adjusted R-squared 0.994452 S.D. dependent var 0.128986
S.E. of regression 0.009607 Akaike info criterion -6.377872
Sum squared resid 0.006553 Schwarz criterion -6.195238
Log likelihood 251.5481 Hannan-Quinn criter. -6.304820
F-statistic 2725.664 Durbin-Watson stat 3.253527
Prob(F-statistic) 0.000000

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

.4

.2

.0

.02 -.2

.01 -.4

.00

-.01

-.02
1984 1986 1988 1990 1992 1994 1996 1998 2000 2002

Residual Actual Fitted

Based on the graph, there is strong evidence of negative correlation.

ii. Test for autocorrelation, functional form and normality.

Autocorrelation (BG test)

Breusch-Godfrey Serial Correlation LM Test:

F-statistic 48.93562 Prob. F(1,70) 0.0000

Obs*R-squared 31.68137 Prob. Chi-Square(1) 0.0000

Test Equation:
Dependent Variable: RESID
Method: Least Squares
Date: 09/24/20 Time: 16:43
Sample: 6/01/1984 9/01/2003
Included observations: 77
Presample and interior missing value lagged residuals set to zero.

Variable Coefficient Std. Error t-Statistic Prob.

C 2.07E-05 0.000902 0.022968 0.9817

RASX200A -0.017754 0.012796 -1.387486 0.1697
RDIVY 0.001532 0.005819 0.263328 0.7931
RAPS 0.029444 0.020696 1.422686 0.1593
RROI 0.008417 0.009866 0.853169 0.3965
REI 0.000232 0.005955 0.038897 0.9691
RESID(-1) -0.669456 0.095699 -6.995400 0.0000

R-squared 0.411446 Mean dependent var 6.67E-18

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

Adjusted R-squared 0.360999 S.D. dependent var 0.009286

S.E. of regression 0.007423 Akaike info criterion -6.881985
Sum squared resid 0.003857 Schwarz criterion -6.668912
Log likelihood 271.9564 Hannan-Quinn criter. -6.796757
F-statistic 8.155937 Durbin-Watson stat 1.734920
Prob(F-statistic) 0.000001

Hypothesis:
H0: No autocorrelation
H1: Autocorrelation
P-value is less than 0.05. Reject H0.
There is autocorrelation present. At the 5% level of significance we can conclude that
autocorrelation exists in the model.

Functional Form (Ramsey RESET test)

Ramsey RESET Test

Equation: UNTITLED
Specification: RP C RASX200A RDIVY RAPS RROI REI
Omitted Variables: Squares of fitted values

Value df Probability
t-statistic 0.853053 70 0.3965
F-statistic 0.727699 (1, 70) 0.3965
Likelihood ratio 0.796337 1 0.3722

F-test summary:
Mean
Sum of Sq. df Squares
Test SSR 6.74E-05 1 6.74E-05
Restricted SSR 0.006553 71 9.23E-05
Unrestricted SSR 0.006486 70 9.27E-05

LR test summary:
Value df
Restricted LogL 251.5481 71
Unrestricted LogL 251.9462 70

Unrestricted Test Equation:

Dependent Variable: RP
Method: Least Squares
Date: 09/24/20 Time: 16:48
Sample: 6/01/1984 9/01/2003

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

Included observations: 77

Variable Coefficient Std. Error t-Statistic Prob.

C -0.009947 0.001411 -7.050951 0.0000

RASX200A 0.023950 0.016934 1.414326 0.1617
RDIVY -0.011204 0.007556 -1.482748 0.1426
RAPS -0.022137 0.026336 -0.840581 0.4034
RROI 0.970216 0.012858 75.45700 0.0000
REI -0.002614 0.007722 -0.338521 0.7360
FITTED^2 0.036784 0.043121 0.853053 0.3965

R-squared 0.994871 Mean dependent var 0.020597

Adjusted R-squared 0.994431 S.D. dependent var 0.128986
S.E. of regression 0.009626 Akaike info criterion -6.362240
Sum squared resid 0.006486 Schwarz criterion -6.149167
Log likelihood 251.9462 Hannan-Quinn criter. -6.277012
F-statistic 2262.797 Durbin-Watson stat 3.304376
Prob(F-statistic) 0.000000

Hypothesis:
H0: Correct functional form and no omitted variables assume
H1: Either incorrect functional form or omitted variables or both
P-value (0.3965) is more than 0.05. Do not reject H0.
At the 5% significant we cannot conclude that there are omitted variables or the model has
incorrect functional form.

Normality (LB test)

12
Series: Residuals
Sample 6/01/1984 9/01/2003
10
Observations 77

8 Mean 6.67e-18
Median -0.001196
Maximum 0.012183
6
Minimum -0.018640
Std. Dev. 0.009286
4 Skewness -0.364454
Kurtosis 1.878373
2 Jarque-Bera 5.740842
Probability 0.056675
0
-0.015 -0.010 -0.005 0.000 0.005 0.010

Hypothesis:
H 0 : Normal distribution

H 1 : Non-normal distribution

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

P-value (0.056675) is more than 0.05. Do not reject H0 .

At the 5% level of significance, we can conclude that the model is normally distributed.

iii. Does this model suffer from first order (lag=1) autocorrelation? What are the
consequences of having autocorrelated residuals? Re-estimate the model
accounting for the first order autocorrelation present using non-linear least
squares. Has the problem been solved?
rp c rasx200a rdivy raps rroi rei

Breusch-Godfrey Serial Correlation LM Test:

Null hypothesis: No serial correlation at up to 1 lag

F-statistic 48.93562 Prob. F(1,70) 0.0000

Obs*R-squared 31.68137 Prob. Chi-Square(1) 0.0000

Test Equation:
Dependent Variable: RESID
Method: Least Squares
Date: 09/24/20 Time: 13:20
Sample: 6/01/1984 9/01/2003
Included observations: 77
Presample and interior missing value lagged residuals set to zero.

Variable Coefficient Std. Error t-Statistic Prob.

C 2.07E-05 0.000902 0.022968 0.9817

RASX200A -0.017754 0.012796 -1.387486 0.1697
RDIVY 0.001532 0.005819 0.263328 0.7931
RAPS 0.029444 0.020696 1.422686 0.1593
RROI 0.008417 0.009866 0.853169 0.3965
REI 0.000232 0.005955 0.038897 0.9691
RESID(-1) -0.669456 0.095699 -6.995400 0.0000

R-squared 0.411446 Mean dependent var 3.60E-19

Adjusted R-squared 0.360999 S.D. dependent var 0.009286
S.E. of regression 0.007423 Akaike info criterion -6.881985
Sum squared resid 0.003857 Schwarz criterion -6.668912
Log likelihood 271.9564 Hannan-Quinn criter. -6.796757
F-statistic 8.155937 Durbin-Watson stat 1.734920
Prob(F-statistic) 0.000001

Hypothesis (BG test):

H0: No autocorrelation
H1: Autocorrelation
P-value (0.0000) is less than 0.05. Reject H0.
There is autocorrelation (first order).

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

The consequence of having autocorrelated residuals is the estimates will not be BLUE but
they are still unbiased.
Standard error might be incorrect and any inferences (conclusion) might be misleading.
For positively correlated residuals, the R 2 might be higher than it should be (inflated relative
to the actual value).

Re-estimated model (accounting for autocorrelation)

Dependent Variable: RP
Method: ARMA Maximum Likelihood (OPG - BHHH)
Date: 09/24/20 Time: 13:29
Sample: 6/01/1984 9/01/2003
Included observations: 77
Convergence achieved after 6 iterations
Coefficient covariance computed using outer product of gradients

Variable Coefficient Std. Error t-Statistic Prob.

C -0.009133 0.000560 -16.32273 0.0000

RASX200A -0.013572 0.009250 -1.467325 0.1468
RDIVY -0.009894 0.003547 -2.789080 0.0068
RAPS -0.003955 0.016278 -0.242974 0.8087
RROI 0.996454 0.007167 139.0425 0.0000
REI -0.003736 0.003917 -0.953782 0.3435
AR(1) -0.728735 0.098028 -7.433987 0.0000
SIGMASQ 4.25E-05 6.95E-06 6.110287 0.0000

R-squared 0.997413 Mean dependent var 0.020597

Adjusted R-squared 0.997151 S.D. dependent var 0.128986
S.E. of regression 0.006885 Akaike info criterion -7.005570
Sum squared resid 0.003271 Schwarz criterion -6.762058
Log likelihood 277.7144 Hannan-Quinn criter. -6.908167
F-statistic 3801.082 Durbin-Watson stat 1.771440
Prob(F-statistic) 0.000000

Inverted AR Roots -.73

iv. Estimate a dynamic version of the model including a lag of all the independent
variables as well as a lag of the dependent variable. Does this solve the autocorrelation?
rp c rasx200a rdivy raps rroi rei rp(-1) rasx200a(-1) rdivy(-1) raps(-1) rroi(-1) rei(-1)
Dynamic model with lags

Dependent Variable: RP
Method: Least Squares
Date: 09/24/20 Time: 13:36
Sample (adjusted): 3/01/1985 9/01/2003
Included observations: 75 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C -0.014945 0.001124 -13.29050 0.0000

RASX200A -0.003433 0.011741 -0.292351 0.7710

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

RDIVY -0.011848 0.005099 -2.323686 0.0234

RAPS 0.004710 0.019235 0.244872 0.8074
RROI 0.980422 0.008648 113.3678 0.0000
REI -0.005008 0.005229 -0.957692 0.3419
RP(-1) -0.667295 0.082957 -8.043832 0.0000
RASX200A(-1) -0.031182 0.011175 -2.790191 0.0070
RDIVY(-1) -0.008313 0.005528 -1.503678 0.1377
RAPS(-1) -0.017522 0.018749 -0.934544 0.3536
RROI(-1) 0.684657 0.080535 8.501349 0.0000
REI(-1) -0.003715 0.005234 -0.709752 0.4805

R-squared 0.997954 Mean dependent var 0.019883

Adjusted R-squared 0.997597 S.D. dependent var 0.130569
S.E. of regression 0.006400 Akaike info criterion -7.119271
Sum squared resid 0.002581 Schwarz criterion -6.748473
Log likelihood 278.9727 Hannan-Quinn criter. -6.971215
F-statistic 2793.961 Durbin-Watson stat 1.777711
Prob(F-statistic) 0.000000

Perform BG test to find out if autocorrelation is still present.

Breusch-Godfrey Serial Correlation LM Test:

Null hypothesis: No serial correlation at up to 1 lag

F-statistic 1.256475 Prob. F(1,62) 0.2666

Obs*R-squared 1.489739 Prob. Chi-Square(1) 0.2223

Test Equation:
Dependent Variable: RESID
Method: Least Squares
Date: 09/24/20 Time: 13:37
Sample: 3/01/1985 9/01/2003
Included observations: 75
Presample missing value lagged residuals set to zero.

Variable Coefficient Std. Error t-Statistic Prob.

C -0.000684 0.001277 -0.535490 0.5942

RASX200A -7.87E-05 0.011718 -0.006716 0.9947
RDIVY -0.000451 0.005104 -0.088344 0.9299
RAPS -0.001679 0.019255 -0.087207 0.9308
RROI 5.85E-05 0.008631 0.006778 0.9946
REI -0.001309 0.005347 -0.244718 0.8075
RP(-1) -0.073308 0.105504 -0.694836 0.4898
RASX200A(-1) 0.001189 0.011203 0.106175 0.9158
RDIVY(-1) -0.000557 0.005539 -0.100548 0.9202
RAPS(-1) -0.000754 0.018723 -0.040285 0.9680
RROI(-1) 0.071768 0.102756 0.698427 0.4875
REI(-1) -8.18E-05 0.005224 -0.015658 0.9876
RESID(-1) 0.187380 0.167165 1.120926 0.2666

R-squared 0.019863 Mean dependent var -1.20E-17

Adjusted R-squared -0.169841 S.D. dependent var 0.005906
S.E. of regression 0.006387 Akaike info criterion -7.112667
Sum squared resid 0.002530 Schwarz criterion -6.710969
Log likelihood 279.7250 Hannan-Quinn criter. -6.952274
F-statistic 0.104706 Durbin-Watson stat 1.991803
Prob(F-statistic) 0.999928

Hypothesis (BG test):

Downloaded by Trung Kiên Tr?n ([email protected])

 lOMoARcPSD|4163173

H0: No autocorrelation
H1: Autocorrelation
P-value (0.2223) is more than 0.05. Do not reject H0.
There is no autocorrelation (first order).

Downloaded by Trung Kiên Tr?n ([email protected])