Firms financial structure is commonly analyzed in terms of financial ratios
that measure the relative size of various accounts in the financial statements.
Some examples of well-known financial ratios are the acid test ratio, the inventory
to working capital ratio, the liability to asset ratio or the equity to debt ratio among others.
We focus on the use of those ratios for cluster analysis, that is, to classify firms
according to the similarity of the structure of their financial statements.
Financial ratios have some undesirable properties to be used on cluster analysis:
skewed distributions, asymmetry (A/B and B/A), outliers not reporting atypical behaviors,
redundancy (spurious correlations) and nonlinear relations between ratios.
Solutions to overcome those drawbacks have been proposed in literature,
but still have proved problematic.
In this work we put forward an alternative method for classifying firms
which aims at solving the above mentioned shortcomings and draws from the
Compositional Data analysis (CoDa) literature. CoDa analysis solves the major statistical
and methodological issues when relative magnitudes of positive components are of interest,
which is the case of financial ratios. Our approach is based on the use of existent cluster
methods on isometric log-ratio coordinates.
In this talk we present the most common financial ratios, their interest and use.
Then we focus on the problems on the use of those ratios on cluster analysis
and the solutions found in literature. Next we present the new methodology based on CoDa analysis
and finally we show an example of application and compare the results with the use of standard ratios.