Page 8 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.72, # 2, 2015, pp. 4-23

assumptions on the behavior of the parameter in (3). First, the factorial zero-
conditional mean rule, i.e. . Second, the static zero mean assumption:
. Finally, zero correlation across the factor parameters : . Under

the established constraints on , component is the factor loading matrix while the
solution to (3) is the dimension-reduced factor.
The factor analysis procedure will produce a set of result tables, from which
the primary ones we will now briefly discuss one by one. First, the communalities
matrix, which will not be reported to preserve space, produces the coefficients of
across-industrial correlation. It is believed that any post-extraction communality
coefficient of above 0.8 can be considered as solid and sufficient. The so-called
Kaiser method of sampling adequacy will also be presented as part of our factor
analysis exploration stage. The adequacy test shows if our sample is suitable for the
factor analysis approach in the first place. Any Kaiser adequacy coefficient of above
0.7 indicates a positive response.
Following the preliminary assessment, the principal components method with
correlation matrices is chosen as the method of factor extraction. We extract only the
factors with an eigenvalue greater than unity, which is a standard rule in literature.
This shows that the percentage of variation in our parameters is better explained
after the factor is introduced; if the eigenvalue is smaller than unity then the model
is better off without dimension reduction. The maximum number of iterations is set
at 1000, after which the procedure selects the optimal quantity of common factors
(in our case, for example, 9 underlying factors were established). In theory, it is
possible to parsimoniously select the number of factors by the author himself and
force the procedure to load the observables on the imposed quantity of unobservable
factors. However, we leave such experimentations for future research and resort to
the rule-based selection procedure for now.
After obtaining the first baseline results, it is recommended to perform a rotation
on the parameter matrix. We rotate the factor solution with the oblique varimax rotation
method with Kaiser normalization. Matrix rotations straighten and improve factor
loadings for interpretation purposes as well as for more precise arbitrary labeling.
Oblique rotation is designed specifically for potentially cross-correlated variables, which
is indeed the case in our model (under our assumption that industries are
interconnected). Again, 1000 rounds of iterations was chosen as the maximum for

8