Information Technology Reference
In-Depth Information
stock whose performance is inverse (correlation value approaching -1) to a
target stock.
Symbol
XOM
AAPL
MSFT
BRK/A GE
WMT
GOOG CVX
IBM
PG
XOM
1.000
0.268
0.640
0.041
0.628
0.812
0.892
0.700
0.639
0.677
AAPL
0.268
1.000 - 0.405
0.501
0.101
0.107
0.301
0.751
0.843
0.283
MSFT
0.640 -0.405
1.000 -0.311
0.718
0.619
0.512
0.117 -0.075
0.476
BRK.A
0.041
0.501 - 0.311
1.000
0.164
0.234
0.097
0.299
0.380
0.344
GE
0.628
0.101
0.718
0.164
1.000
0.690
0.466
0.487
0.222
0.556
WMT
0.812
0.107
0.619
0.234
0.690
1.000
0.807
0.548
0.454
0.638
GOOG
0.892
0.301
0.512
0.097
0.466
0.807
1.000
0.667
0.688
0.672
CVX
0.700
0.751
0.117
0.299
0.487
0.548
0.667
1.000
0.867
0.436
IBM
0.639
0.843 - 0.075
0.380
0.222
0.454
0.688
0.867
1.000
0.516
1.000
PG
0.677
0.283
0.476
0.344
0.556
0.638
0.672
0.436
0.516
Fig. 2.7.
Left: stock price of Google and Exxon in 2010, with a correlation of .53,
i.e. moving together somewhat similarly. Right: 10 stocks showing 100
correlations as a 10x10 grid.
However, visualizing correlations as a grid becomes difficult as the
number of items increases. Grids quickly become large as the number of
correlations increases with the square of the number of items. Patterns are
more difficult to discern in large grids, and more space is required to show
larger grids. Visual scanning along rows and columns to associate
intersections with perimeter labels becomes an active cognitive task.
Interactions such as navigation, filtering, clustering, and sorting can become
additional user tasks requiring further cognitive planning and execution
effort. “It is difficult to see and scale this approach beyond 50 or so items,”
expressed a stock trader
[Str07]
.
Alternatively, a set of correlations can be represented as a graph of nodes
and links. The placement of the nodes is such that strong correlations (links)
are attracted to each other, and negative correlations repel each other. For
example, the 100 correlations for 10 stocks as shown in the right image of
Fig 2.7 may be represented as a graph shown in the right image of Fig 2.8.
The benefit of the force-directed graph visualization approach is that
items highly correlated intuitively cluster together, and items inversely
correlated tend to be far apart. But there are many challenges to this
approach. For example, distances between points are not Euclidian,
through training and/or interactive techniques can facilitate learning and
comprehension of the relative distances.
Another perceptual issue with force-directed 2D layouts is the relative
placement of items. For an item near the centre of the plot, the relationship
between immediate neighbours and distant items is clear. However, for an
item near the edge of the plot and an item on the opposite side, the
relationship can be ambiguous: the item on the opposite side could be
inversely correlated or it could be potentially highly correlated, but unable
to be placed close to the initial item due to the constraints of the other
Search WWH ::
Custom Search