Ceres Logo
Ceres Analytics Platform

Cluster Analysis OS:  Part 2 (Illustrate Solution)

Start: 1.

In the "Cluster Analysis" group, "OS Cluster" tab, choose "Ward's (R) Part 2 (Show Chosen)" from the menu.

pic_13

2.

You'll be asked for the preferred number of clusters:

pic_2

3.

To prepare for possible export to PowerPoint, you can choose the height of the graphic plots:

pic_3

4.

For the profile histograms, you can choose how fine (or chunky) you'd like the display. 
Here, we'll choose a number in the middle, spreading the observations in each histogram among 13 bars:

pic_16a


5.


The first of two charts shows the profile histograms.

In this example, for simplicity, we worked with only two measures:
  • $  absolute price change
  • % proportionate price change
Not surprisingly, the profiles contrast the cluster compositions as follows:

Cluster#
$ Absolute Change Skewed...
% Proportionate Change Skewed...
1
Low
High
2
High
High
3
Low
Low


Also, after the charts are pulled into Excel, you'll be told where the data with the cluster indicator are saved.

Note: again, for larger-scale applications, where the data are unsuitable for Excel, we point the platform to the relevant data source
    and report accordingly.

pic_17a


6.


The cohesion and dispersion of the clusters is illustrated in two-dimensional space through R's Multi-Dimensional Scaling algorithm:

pic_18


Conclusion


We can see that the smallest cluster, #1, is quite compact.  A number of outliers distort clusters 2 and 3:
   these must be the few observations that were grouped into the tiny clusters when we looked at solutions from 4 to 6 clusters.

Of course, given that we used only two variables for clustering, the concept of a two-dimensional illustration is not as impressive as it would be if we'd clustered with more variables.

Nonetheless, Cluster 1 merits investigation.  The implication is that Cluster 1 contains lower-priced stocks, where a penny change represents a large percentage.  Perhaps Cluster 1 contains lower-priced stocks, of which the price is so low that any change in $ price is necessarily a large percent.

Re-running the cluster analysis with more variables will, of course, give a different solution.
 
Logically speaking, this cluster analysis tells us about stock price volatility in both absolute and relative senses.


© 2012 Ceres Anaytics, LLC All rights reserved