* Use HPRICE1.dta
regress price lotsize sqrft bdrms
predict uhat, residual
generate uhat2 = uhat^2
generate lotsizesq = lotsize^2
generate sqrftsq = sqrft^2
generate bdrmssq = bdrms^2
generate lotsize_sqrft = lotsize*sqrft
generate lotsize_bdrms = lotsize*bdrms
generate sqrft_bdrms = sqrft*bdrms
* White's test for heteroskedasticity with squared and cross-product terms (F-test)
regress uhat2 lotsize sqrft bdrms lotsizesq sqrftsq bdrmssq lotsize_sqrft lotsize_bdrms sqrft_bdrms
* White's test on log(price) equation (F-test)
regress lprice lotsize sqrft bdrms
predict uhat_2, residual
generate uhat2_2 = uhat_2^2
regress uhat2_2 lotsize sqrft bdrms lotsizesq sqrftsq bdrmssq lotsize_sqrft lotsize_bdrms sqrft_bdrms
* The log transformation of price solved the heteroskedasticity problem!
*
* Now let us redo the above tests using the Special Case of White's test (p.253)
regress price lotsize sqrft bdrms
predict pricehat
generate pricehat2 = pricehat^2
* Test of price equation (F-test)
regress uhat2 pricehat pricehat2
*
regress lprice lotsize sqrft bdrms
predict logpricehat
generate logpricehat2 = logpricehat^2 
* Test of the log(price) equation (F-test)
regress uhat2_2 logpricehat logpricehat2
*  NOTE: The LM form of the above heteroskedasticity tests is n*R2
*  which is distributed as a Chi-square distribution with J degrees of freedeom
* To get p-values for the chi-square distribution use:
* display chi2tail(df,x)
*
* Now lets us the estat hettest option to test heteroskedasticity
* for the price and lprice equations
regress price lotsize sqrft bdrms
estat hettest
regress lprice lotsize sqrft bdrms
estat hettest
* Again the log(price) transformation solved the heteroskedasticity problem