Page 12 - Azerbaijan State University of Economics

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

remaining factors, we rotate our industry matrix using the oblique varimax method
and obtain the rotated sums of the squared loadings. We highlight that the total
variance explained by the 9 factors remains unchanged, but the first factor‟s role has
declined by 10%, thus raising the relevance of the other factors. Which is precisely
what we wanted. We now investigate each industry‟s loading score on each of the 9
extracted factors in order to deduce their most intuitive labeling.
We will only report the factor loading estimates for the rotated matrix case, since
this is more correct both for the technical and the intuitive reasons outlined earlier in the
paper. Table 3 reports the rotated matrix‟s factor scores for each industry. Under the
oblique varimax rotation and the principal components method of extraction, matrix
rotation convergence was achieved after only 28 iterations. We first note that factor
belongingness is not restrictive, meaning that certain industries can load on more than
just one factor with equal degrees of score strength. We can also clearly notice that the
first factor is loaded on by almost all industries, whereas factor 9 is the least responsive.
Intermediary factors are all moderately influential. We are sticking to the eigenvalue
selection rule and will not act by discretion and drop any of the least powerful factors,
although such decision would have been justified.
Based on the factor score matrix we will now assign arbitrary factor-specific labels
and thus complete the dimension reduction procedure (Table 4). Given the universal
loading of basically every industry on factor 1, we label it simply as the “All Industries”
factor. Careful scrutiny of the rotated factor scores have led us to assign the following
names to the remaining 8 factors: “Non-Heavy Industries”, “Communication and
Utilities”, “Textiles and Light Equipment”, “Machinery, Vehicles, and Related Tools”,
“Heavy Metals and Inorganic Chemicals”, “Storage and Infrastructure”, “Agriculture and
Organic Chemicals”, and “Mineral and Quarrying Goods”. We emphasize that in no way
are our labels final and undisputable. It may well be that an attentive reader or future
studies would detect an even better and more intuitive labeling strategy. However, this is
the best we can offer, and we believe that the labels are broad and yet specific enough for
implications and conclusions. We therefore save our extracted 9 factors as separate series
and use them for the purpose of our balance of trade regression.
We continue the representation of results with the second and final phase of our
empirical strategy: an ARDL analysis of the effect of exchange rate shocks on the
common factors. Although the method does not require it, we still report the unit root
test results in Table 5. Some of the variables possess a unit root, while all of the

12