Estimation of Missing Observations in the Generalized Linear Regression Model by Means of Dummy Variables

Given the multiple linear regression model with a nonscalar disturbance covariance matrix, missing observations pertaining to the dependent variable are shown to be estimated by Generalized Least Squares with dummy variables added to the explanatory variables set for each missing observation. This result is applied to heteroscedasticity, autocorrelation and to the generation of quarterly or monthly observations for stock variables.