Linear Regression Model: Relaxing the Classical Assumptions

The classical assumptions of homoskedasticity and non-autocorrelation of the distribution of random error in a linear regression model are essential for OLS estimates to follow BLUE property. Homoscedasticity means constant variance and non-autocorrelation means random errors are not correlated. These assumptions, however, are not valid in most of the sample data and the problems of heteroscedasticity and autocorrelation appear in a linear regression model. Heteroskedasticity is a problem mainly for cross section data. The problem of autocorrelation arises when errors are serially correlated. This problem is usually found in time-series data. In time series, autocorrelation is the correlation of a variable with lags of itself. Presence of autocorrelation implies that current error can remember its past values. This chapter discuses different aspects of these problems.