Correcting for autocorrelation:

After testing for autocorrelation you find  that you cannot accept  the null hypothesis of no autocorrelation. What is  the next sourse of actiona It would be very simple  if we  knew the actual value of ρ.  If the model in holds for time t  then it also holds for time (t-  1).  Therefore,  

Multiplying (7.19)  throughout by  p  and subtracting from  (7.3) we get

where by definition. Note elat  the first observation is lost here since a lag has been introduced in the model  and therefore you are left with t-2,3,,..,T.

This transformation is known as  the Cochrune-Orcutt (1949) transformation and it reduces the error terms to classical errors, that is, since E, is distributed identically and independently, all the OLS assumptions are valid for this nzw regression with,transformed  variables. Therefore,  the OLS estimator has all  the optimal properties. that is, it is  the best linear unbiased estimator or BLUE  However, in most practical applications p  is not known and has to be estimated.