Graphics Reference
In-Depth Information
information to find. he differences between the companies that went bankrupt and
the others that did not are more complicated than can be displayed in a scatterplot
of two variables.
Another alternative would be to use spinograms (as shown in Fig.
.
), but the
number of bankrupt companies is so small relative to the total number of companies
that little can be seen. Fitted smooths would be better, although they require more
computation.
he parallel coordinate plots employed here are a selection of many that might
have been shown. he choice and order of variables influences what might be seen.
he decision to discard some extreme outliers does too. he level of α-blending is
also an influential factor. In other words, a parallel coordinate plot is like any other
multivariate analysis in that the user has a lot of freedom of choice. Careful thought
helps, and so does statistical modeling. Having explored the data and built a model,
parallel coordinate plots can again be useful, this time as a way to understand the
model's relationship to the dataset.
Investigating Bigger Companies
3.8
Financial data for small companies is highly variable and could well be more unreli-
able than data for large companies, although this is di
cult to assess. Large compa-
nies are certainly different from small companies, and so studying them separately
makes sense. In one important way they do not differ: the bankruptcy rate for the
biggest
companies (eachwith Total Assets
),
.
%,isclosetotheratefor
therestofthedataset,
.
%fortheremaining
companies.Ontheotherhand,
a looser definition of big (Total Assets
) yields bankruptcy rates of
.
% for
the
“big” companies and
.
% for the rest, a significant difference. Given the
many different limits that might beused,awide range ofresultsis possible.However,
it is not modest variations in bankruptcy rates by company size that are of interest to
us; it is identifying which companies might go bankrupt.
Over fortyyears itwould bereasonable toexpectthat company sizehasincreased.
Curiously, for this dataset, the effect on logTA is negligible, as Fig.
.
shows.
Figure
.
showsboxplots ofthe original ratio variables forallof thedata with the
group of big companies highlighted. Only the first five financial ratios are shown, as
the distributions of the others are,as Fig.
.
shows, fartooskewed tobeinformative.
logTA is included to show the size distribution. Although the medians for the bigger
companies differ noticeably from the median ratios for all companies (for cash, in-
ventories and hence, obviously, current assets, they are lower, while the median ratio
is higher for the capital assets ratio), the distributions overlap substantially.
Amoreeffective wayoflooking at theratios istousespinograms. InFig.
.
there
are plots of the ratios Cash.TA, Inv.TA, Kap.TA,andIntg.TA with the big companies
selected. he proportion of big companies declines as the cash ratio increases; it also
declines as the inventory ratio increases, apart fromthe lowest group (Inv.TA
<
.
),