Scribd
Upload
56 views2 pages

Practical 1 - Heteroskedasticity & Autocorrelation

The document provides instructions for completing a practical assignment involving analyzing heteroskedasticity and autocorrelation. Students are asked to estimate several regression models using different datasets and testing for heteroskedasticity and autocorrelation, and comparing the results.

Uploaded by

yandhias
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
56 views2 pages

Practical 1 - Heteroskedasticity & Autocorrelation

The document provides instructions for completing a practical assignment involving analyzing heteroskedasticity and autocorrelation. Students are asked to estimate several regression models using different datasets and testing for heteroskedasticity and autocorrelation, and comparing the results.

Uploaded by

yandhias
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Practical 1 – Heteroskedasticity & Autocorrelation

• Work precisely and read carefully.


• Answer all questions, and only the questions.
• When you preform a hypothesis test provide the null hypothesis, the value of relevant
test statistic, and the conclusion.
• The datasets can be found on the Blackboard.

1 PART I: Heteroskedasticity
For the first part of this practical we will use the house prices dataset. It contains 88
observations on the following variables:

price : selling price of a house in thousands of $


bdrms : number of bedrooms in the house
lotsize : size of lot in square feet
sqrtf : size of house in square feet

Consider the following model:


pricei = β1 + β2 lotsizei + β3 sqrtf i + β4 bdrmsi + εi (1)
1. Use the regress command1 , to run OLS on model (1). What is the elasticity of price
with respect to the size of the house, evaluated at the sample mean of the data? Do
not report the estimation results, but do report the estimated elasticity.
2. Get the heteroskedasticity robust standard errors (use the robust option after the
regress command), and report the regression results2 . Also test for heteroskedasticity
using White’s test3 . State the null hypothesis and report the value of the test statistic
and the conclusion of the test.
3. Perform a Breusch-Pagan test for heteroskedasticity4 using the specification:
σi2 = σ 2 exp {α1 lotsizei + α2 sqrtf i + α3 bdrmsi }
Note that Stata uses the exponential specification by default. (You will need to run
the model without the robust option to be able to do this.) State the null hypothesis
and report the value of the test statistic and the conclusion.

Now consider the same model in logarithms, except for the number of bedrooms:
log (pricei ) = γ1 + γ2 log (lotsizei ) + γ3 log (sqrtf i ) + γ4 bdrmsi + εi (2)
1 Statistics→Linear models and related→Linear regression
2 Statistics→Linear models and related→Linear regression: SE/Robust
3 Statistics→Postestimation→Reports and statistics: Information matrix test
4 Statistics→Postestimation→Reports and statistics: Tests for heteroskedasticity

1
4. Run the model in logarithms. Report the estimated elasticity of price with respect to
size of the house. Use both White’s and Breusch-Pagan’s tests for heteroskedasticity.
What do the two tests suggest with respect to heteroskedasticity? Why do the two
tests lead to different conclusions?
5. Use a non-nested test to decide whether you should use the model in levels (model (1))
or in logarithms (model (2)). What is the value of the test statistic and what model
would you use?

2 PART II: Autocorrelation


For the second part of this practical we will use the phillips curve dataset. It contains 49
annual observations on:

year : 1948 - 1996


unem : civilian unemployment rate
inf : CPI inflation rate

Phillip’s (1958) model suggests that there is a negative effect of unemployment on infla-
tion:
inf t = β1 + β2 unemt + εt (3)
with β2 < 0.

1. Use the command tsset year to tell Stata that you are using a time-series dataset,
where the time-indexing variable is year.
2. Estimate model (3) by OLS. Do you reject the hypothesis of the Phillips curve? Test
the hypothesis of absence of autocorrelation using a Durbin-Watson test5 . Test for
absence of autocorrelation of order 3, 2, and 1, using Breusch-Godfrey6 tests.
3. Use the newey command to get heteroskedasticity and autocorrelation robust standard
errors7 . Truncate the lags needed for the variance matrix estimation to 5 (use the
option lag(5)). Report the results.
4. Use the prais command8 to estimate the model using the Prais-Winston FGLS esti-
mator. Use the corc option to get the Cochrane-Orcutt estimator. For both estimators
report the results. Did running FGLS correct for autocorrelation? Support your answer
with a formal test.
5. Most likely there is a misspecification issue in the model. The inflation rate does not
depend only on the rate of unemployment, but also on the stage of the business cycle.
To account for the effect of business cycles, run OLS on the extended model9 :

inf t = γ1 + γ2 unemt + ρinf t−1 + εt (4)

and report the results. Did including lagged inflation rate as an independent variable
correct for autocorrelation? Support your answer with a formal test (state the null, the
value of the test statistic, and the conclusion).

5 Statistics→Postestimation→Reports and statistics: Durbin-Watson d statistic


6 Statistics→Postestimation→Reports and statistics: Breusch-Godfrey test
7 Statistics→Time series→Regression with Newey-West std. errors
8 Statistics→Time series→Prais-Winsten regression
9 You can use the syntax “regress inf unem L.inf” to include the first lag of inflation as an independent

variable.

You might also like