Temple University

Department of Economics

 

Heteroscedasticity

 

1.  The file FOODUS.TXT contains observations on food expenditure (y_t) income (x_t) and number of persons in each household (n_t) from a random sample of 38 households in a large U.S. city. Food expenditure and income are measured in terms of thousands of dollars. Consider the statistical model

where the e_t are independent normal random errors with zero mean.

(a)            Estimate the unknown coefficients in the model.

(b)            Plot the least squares residuals in (a) against income and against the number of persons.  Do these plots suggest heteroscedasticity?

(c)             Use a Godlfeld-Quandt test to test for heteroscedasticity with the observations ordered by income and then ordered by number of persons.

(d)            Find generalized least squares estimates of the above model under the assumption that Compare the estimates with those obtained using least squares.  Does correcting for heteroscedasticityappear to have improved the precision of estimation.

(e)            At the 5% level test that the model is heteroscedastic using White’s test.

(f)              Test the hypothesis that the model is a heteroscedastic function of the number of people and income using the Breusch-Pagan test.

(g)            Estimate the model under the assumption that the error variance is given by. 
Use maximum likelihood to estimate all of the model parameters.