Two-phase Gauss-Seidel solution of the model 

For data years the model is solved by Gauss-Seidel iterative method using data values as starting values and an 1 per cent maximum deviation from previous iteration as the convergency criterium as well as 30 iteration maximum limit of iterations. In all solutions at least the last data year is also solved iteratively. For forecasting years, beyond the data, the previous year solution values, are used as the starting values. 

After the normal Gauss-Seidel solution over the entire solution period a level correction on the basis of last year solution values compared to the same year observed values is performed. This proportional level correction with the coefficients of the last data year is undertaken also for all years beyond the data years. 

After this level correction still another Gauss-Seidel solution is performed, but for all definitional equations only, holding the behavioural equations, those with estimated parametres, in their corrected values. 

With the above criteria the number of iterations needed for a solution varies between 10 and 15. Of these 3 to 5 are used in the post-correction iterations.

Variable notation in the TajkaPM procedure

X[i,j]	exogenous variable, observation or assumed value [i year, j variable number]
Y[i-t,j]	endogenous variable observation [t lag; minimum 1, maximum 2, j variable number]
Z[i,j]	endogenous variable calculated value
ZL1[j]	endogenous variable starting value at the beginning of iteration;
	first iteration: observed value Y[i,j], later: calculated value Z[i,j]
	first iteration in case of forecast: previous year final calculated value Z[i-1,j]
	subsequent iterations: calculated value Z[i,j]