WORKSHOP 8 – HETEROSKEDASTICITY

Goals
By the end of this workshop, you should be able to do the following:

 Explain Heteroskedasticity
 Perform the Breusch-Pagan Test
 Perform the White Test

Tools 
You need to bring the following to class

 Download the dataset deal.dta

 Statistical Tables
 Calculator

Background
The data used in this exercise were obtain from televised episodes of Dear or No Deal. The
period covered is from 2 Febuary 2004 to 25 November 2005. The 232 games do not include
special prizes or ‘super cases’.
The game starts by randomly selecting one contestant among 156 people sitting in stands.
There are 26 briefcases each one having different prize money. The chosen contestant is
given a certain number of cases and guess the amount contained in each one. A ‘bank offer’
is made to the contestant when they have consecutively guessed the wrong amounts. A
‘Deal’ is made when the contestant accepts the offer. Otherwise, a ‘No Deal’ occurs. The
process is repeated with on the next round with a certain number of cases.

CONTINUE TO THE NEXT PAGE

200916/EFM/MEV&AE Page 1
Variable Definition
The variables in the data set are listed in the table below with their definition.
Variable Label
id episode
series series number
female =1 for female contestant
no_deal =1 for ‘No Deal’
own_case value of their own case
acc_rnd round when accepted offer
avg0 average value at beginning
avg_left average value of remaining cases
davg change in average value after last round
dmax change in maximum case value after last round
doff change in offer made after last round
last_off last offer made
max_left maximum value of remaining cases
min_left minimum value of remaining cases
num_left number of cases to be opened
winnings amount won

Group Work 
Complete the questions below with the other members with your group.
1. Write the econometric model where the amount won (winnings) is explained by the
change in the offer made after the last round (doff), the change in the average value
after the last round (davg) and the number of cases left to be opened (num_left).
Winning = B0 + B1doff + B2davg + B3num_left + u

2. With the other people in your group, discuss what the econometric model is trying to
determine.

3. Estimate the econometric model in STATA using the commands below along with the
other diagnostic tests.

regress winnings doff davg num_left

estat vif
estat ovtest
estat hettest
estat imtest, white
RESET test
200916/EFM/MEV&AE Page 2
 Compare to the significant level 1,10,5%
If probability small, reject H0 : model has no omitted variables

BP test:
If the probability small, reject H0 : constant variance  hesteroskedasticity

4. Report the results in an equation form.

Winning = 18026.95 + 0.77doff + 1.01davg + u
(18026.94). (0.133). (0.091)
5. Determine which of the independent variables have a relationship with winnings. Use
the significance level at 5%.

6. Determine whether there is heteroskedasticity based on the Breusch-Pagan Test and

the White Test in the output.

7. Explain the meaning of the outcome you obtained in part 5.

8. Re-estimate the model using heteroskedasticity-robust standard errors using the

commands below.

regress winnings doff davg num_left, robust

estat vif
estat ovtest
estat hettest
estat imtest, white

9. The adjusted R-Squared is not reported in STATA for the model. Obtain this figure
by using the formula below where n is the sample size and k the number of

200916/EFM/MEV&AE Page 3
 independent variables. The steps to obtain the Sum of Squares Residuals and the Sum
of Squares Total is outlined below the formula.

[ ] (∑ )
n
2
ui
i=1

2 n−k −1
R =1−

[ (∑ ( ) ]
n
2
y i− y )
i=1

n−1

n
Sum of Squared Residuals - ∑ u i
2

i=1

To obtain the sum of squared residuals, follow the steps below.

1. Type the STATA commands below.
regress winnings doff davg num_left, robust
predict winnings_hat, xb
generate uhat = winnings – winnings_hat
generate uhat2 = uhat * uhat
summarize uhat2
generate sse = r(sum)
format %100.0g sse

2. Follow the sequence of selection below. Look for the column sse and copy
the value on your paper.

Window > Data Editor

n
Sum of Squares Total - ∑ ( y i− y )
2

i=1

Follow the steps below to obtain the Sum of Squares Total.

1. Type the commands below in STATA.

summarize winnings
generate mean_winnings = r(mean)
generate distance = winnings – mean winnings
generate distance2 = distance * distance
summarize distance2
generate sst = r(sum)

200916/EFM/MEV&AE Page 4
 2. Follow the sequence of selection below. Look for the column sst and copy the
value on your paper.

Window > Data Editor

10. As you would have observed from the output in part 8 that it is not possible to obtain
the Breusch-Pagan Test. Type the commands below for the auxiliary regression.

regress winnings doff davg num_left, robust

predict uhat, residuals
generate uhat2 = uhat * uhat
regress uhat2 doff davg num_left

Conduct a joint hypothesis to test for Heteroskedasticity. Calculate the LM as shown

below and use the significance level of 5%.
2
LM =n ∙ Ru^ 2

11. Report the results in an equation from part 8. Include the adjusted R-square from part
9 and the LM from part 10.

12. Do the producers of this show have any control over the winnings by use of the
variables in part 8?

13. Reports your results on the table below. Test if the intercept and the independent
variables are significant at 10%, 5% and 1%. If the intercept or independent variable
and fill in the appropriate for each model. Model 1 refers to the model estimate and
the diagnostic tests in part 3 and Model 2 refers to the model estimate and diagnostic
tests in part 8, the R-squared from part 9 and the LM obtained in part 10. The
intercept has been filled in both models for you as an example.

Model 1 Model 2
Dependent Variable winnings winnings

18026.94*** 18026.94***
Intercept
(2365.663) (3275.655)

change in offer made after

last round (doff)

change in average value

200916/EFM/MEV&AE Page 5
 after last round (davg)

number of cases to be
opened (num_left)

Diagnostics
R-Square
Adjusted R-Square
F-Statistic F(3, 228) = 49.84***
Mean VIF
Ramsey Reset Test F(3, 225) = 90.96***
2
Breusch Pagan Test χ ( 1)=16.85 ***
2
White Test χ ( 9)=16.85 ***
Significant at 10% only (*)
Significant at 10% & 5% only (**)
Significant at 10%, 5% & 1% (***)

200916/EFM/MEV&AE Page 6