y = (Z'(-I IT)Z) (Z'( l IT) P). (72)


Although this estimator is consistent, asymptotically normal and effi-
cient (Kmenta, 1971), it is not feasible in that n is unknown. The
covariance matrix can, however, be estimated as follows. Let eit be the

estimated residuals from the ordinary least squares regression of the
price equation for the it state alone. The aij can then be estimated

by

 1 ^ ^
 _i T E eit ejt. (73)
 ij T jt


By replacing o.. in equation (71) with .ij, the two-stage Aitken's
estimator (TSAE) for y


 S= (Z'(n1 IT)Z)- (C'("-1 IT)P) (74)


 is obtained. Zeller (1962) has shown that y has the same asymptotic
 properties as y given in equation (72).