Financial Visualization Case Study: Correlating Financial Timeseries and Discrete

Events to Support Investment Decisions

Eugene Sorenson1, Richard Brath2

1
Bloomberg L.P., 2 Oculus Info Inc.
{[email protected], [email protected]}

Abstract financial data and associated financial applications, such

In financial timeseries visual data analysis, there as derivative valuation, portfolio monitoring, sector rota-
are many potential data elements to display, including tion, etc.
continuous quantitative data and discrete event data. This paper describes one significant visualization
Bloomberg has designed and implemented a scalable project in 2006-2007 combining the visual depiction of
visual representation for the depiction of many discrete continuous timeseries data (e.g. stock prices) together
timestamped events in use by hundreds of thousands of with associated discrete event data (e.g. news, earnings
financial markets experts. This visualization enables a releases, announcements) as multi-variate glyphs:
single screen to visually organize a large volume of event
data; to facilitate inference through visual alignment of
related data; and to provide a workflow from the single
point of access to a wide variety of detailed information.
It allows an order of magnitude more event types to be
associated with timeseries data.

Keywords---Timeseries visualization, financial

visualization, multi-attribute gylphs.

1. Introduction Fig. 1. A financial price timeseries with discrete

events aligned along a common time axis provides
Participants in the capital markets are confronted visual access to possible causes of price movement
with the challenge of considering many discrete pieces of The prior technique displayed an alphanumeric label
information when analyzing a security and making an representing the event adjacent to a timeseries line. The
investment decision. For example, users of the Bloom- technique originally worked well when the event selec-
berg Professional® service have access to tens of thou- tion was limited to earnings reports and corporate ac-
sands of stocks, four million fixed income securities, tions. However, as we focused on adding many more
millions of indices and economic data series, both histor- events and event types it was clear this technique would
ic and real-time and across many variables. The volume not scale beyond 36 unique item types (i.e. limited to
of transactions and information continues to grow. Fur- alphanumeric characters) and could result in occlusion
thermore, the variety of data includes financial news, when there were many near simultaneous events:
analyst reports and audio/video segments from thousands
of sources as well as mandatory filings, government eco-
nomic data, management personnel, legal dockets, natu-
ral disasters, user-defined alerts and more. There is an
on-going need to design and develop effective user inter-
faces to enhance users’ abilities to exploit this data.
All this information is available interactively and a Fig. 2. Alphabetic characters superimposed over a
significant portion updates in real-time. Much of the val- price chart indicate events. The technique will not
ue of this data depends on speed, and visualization can scale to greater event density due to occlusion.
provide a key role to aid faster comprehension than tradi- Going forward, a more general technique was re-
tional rows and columns of data, or simple charts and quired that would scale to depict more event types,
graphs. Bloomberg has a number of different visualiza- greater volume of events, reduce occlusion and facilitate
tion projects to improve the comprehension of complex workflow.
2. User Needs • Workflow: The visualizations are intended to be
part of a broader workflow. They must co-exist
Line charts and other related high-frequency charts with other elements in the interface. Further, as
plotting price over time (e.g. candle charts) are very the visuals represent large amounts of data, they
heavily used in financial services. A user can view a need to aid navigation.
chart to validate an assumption (e.g. is the price still • Small-space: The visualizations need to be usable
trending); look for potential behaviors and patterns (has from full-screen down to a small size, including a
the price re-tested a high and failed indicating a new po- mandatory 480 x 320 size; and even smaller as
tential downtrend); etc. When there is an unexpected part within a user-configurable tiled-window
anomaly (e.g. unexpected price drop) the immediate desktop called Launchpad.
question is “why” and an explanation is sought. At this • Clear, Immediate Comprehension: All screens in
point, the viewer needs to understand what other events the system must provide immediate useful infor-
happened at that point in time, for example: mation. It was not an option to require one or two
• Was there a significant news story? clicks of the user to configure at start.
• Did a research or credit analyst change a rating? • Fast Start-up: Financial markets require speed
• Was there an unexpected result in newly released throughout the application. Screens in professional
economic data? systems must launch in a second, completely pop-
• Did the company have a significant action, such as ulated with data.
an acquisition or investor presentation? • High-Performance: The visualization needed to
More complex investment hypotheses emerge when update at sub-second intervals, ideally it should be
considering event patterns over time. For example: able to achieve 30 frames per second.
• Regular patterns between event occurrence and • Low CPU: A trader may have many different ap-
price movement, e.g. Is there a consistent price re- plications, all updating in real-time, running on a
sponse each time this company makes a public desktop with as many as 8 screens. The visualiza-
announcement regarding earnings? tions could not consume more than 5% CPU at
• Event sequences and price movement, e.g. If a peak-load (e.g. start-up and fast moving markets)
particular analyst’s commentary on this stock has on lowest common denominator hardware.
the potential to impact the price, are there any • Rapid product cycle: At Bloomberg, new infor-
economic events that immediately precede and in- mation products are typically conceived, designed,
fluence that analyst. implemented, tested and deployed in months if not
The product team collected user feedback and en- weeks. The financial industry is competitive and
hancement requests through informal interviews with fast-paced: therefore, the product processes are
users and from comments collected by the support desk. similarly rapid [1].
This user feedback included:
• Overlapping alphabetic characters were difficult 3. Background and Related Work
to read.
• If characters overlapped the price line, critical Visualization is not new to professional users in the
price information could be obscured. financial markets. Bloomberg's solution needed to ac-
• Ability to access more detailed information, e.g. commodate the existing professional user community
via tooltips or workflow. and respect techniques they were already familiar with.
Meetings among different business groups indicated
potential demand for many different tasks and types of 3.1. Financial Visualization Literacy
events. For example:
• Monitoring: Traders were interested in monitoring Market professionals have long relied on high-
their trades in relationship to price movement in frequency charts of time series data with time plotted
order to better manage and assess their trading ac- horizontally and stock price (or other continuous varia-
tivity on a one day chart as well as the impact of ble) plotted vertically. This may be more simply de-
significant news or corporate actions, such as scribed as a “price chart”. The financial industry’s focus
earnings reports. on returns places greater demand on understanding a
• Exploration: Economic analysts were interested in security’s current price in the context of its history. As a
reviewing statistical agency announcements (e.g. result there is far greater use and familiarity with time
GDP, unemployment, factory orders) and associ- series charts versus other types of visual representations.
ated price movements to explore potential trends
over a recent period such as a year or two.
• Evaluation: Sales professionals were interested in
seeing their stock recommendations versus the ac-
tual market price over a period of a few months to
assess the quality of their forecasts.
In addition to the immediate need regarding the rep-
resentation, there were many additional constraints, e.g.: Fig. 3. Sample price chart over 10 days showing a
security’s price in the upper panel and trade volume • Details for each event can be elaborated in a side
in the lower panel. panel, referring to each character.
Some market professionals, such as technical ana- • In areas of density, the flag can be stacked to indi-
lysts, utilize many unique variants of price charts, includ- cate high density of events. Zooming in will show
ing open-high-low-close charts, candle stick charts, point each flag individually.
and figure, renko, market picture charts, etc. Also, there
is a long history of visual analytics with tight integration
between charts and analytical functions, such as averag-
es, ratios and spreads, with particular configurations
known by acronyms such as MACD (moving average
convergence/divergence), DMI (directional movement
indicator), RSI (relative strength indicator), etc. Fig. 6. Example chart similar to financial charts from
Google Finance , showing alphabetic flags in the
chart, and the corresponding news item in the
adjacent panel on the right (wireframe drawing based
on image from [2]).
The Google approach was considered very innova-
tive but possibly had limitations that may not have been
acceptable to the professional user:
• The approach requires a unique event label per
each event. This approach was either limited to 36
individual alphanumeric labels or required more
complex labels which would increase marker size
and potentially increase perceived clutter.
Fig. 4. Sample price chart over 6 months showing • The approach does not easily differentiate be-
common technical studies, including moving average
tween types of events. Google has addressed this
lines (yellow and green lines) and MACD lines (bot-
tom panel). somewhat in later versions by creating different
shapes and colors of markers for event types such
Thus there are established conventions for timeseries as dividends and stock splits.
representations within the professional financial markets. • The approach requires the viewer to shift gaze
Within this community, all users are familiar with the back and forth from the chart to the list, scanning
basic conventions and smaller groups of users have ac- the chart and scanning the list with each shift, im-
quired deep expertise in specialized charts and analyses. plying a potential perceptual inefficiency.
This last point is similar to Larkin and Simon’s [3]
3.2. Related Work conclusions that (1) visually grouping information to-
gether that is used together avoids the search effort and
The previous technique to represent events was to (2) removes the need to match symbolic labels, leading
display a single alphanumeric character as close as pos- to reductions in working memory and search effort.
sible to the line in the chart, without obscuring the line Another common approach is to organize and dis-
(figure 2). Many other financial applications used the play related information in a singular interface is the
similar techniques: portal. A portal simply puts all the different information
together across a series of separate view tiles with no
explicit linkages across views. Finding related data at a
particular time point requires the user to exhaustively
search through all the portal tiles (often more tiles than
can be arranged in a single display screen) and perform
mental arithmetic to evaluate whether the timestamp
associated with a piece of information is in relevant
Fig. 5. Example chart showing a single character per timeframe or not before evaluating the information.
event (e.g. E for Earnings). The seemingly random
placement of the event icons throughout the plot area 4. Design, Implementation and Feedback
increases the sense of clutter and could reduce
speed of comprehension of the timeseries price line. The project went through an initial design phase
(wireframe drawing based on chart from wsj.com). with feedback primarily from internal stakeholders and
Another approach was popularized by the stock was followed by an implementation phase.
charts in Google Finance that launched in early 2006 [2]:
• Each event can have a unique alphanumeric char- 4.1. Requirements and Design
acter. The character is attached via a leader line to
the chart, as if a flag. The approach consisted first of assessing all the po-
tential event data items that needed to be addressed. Ini-
tially, 4 different categories of events were required with and it worked well when events were few in number (e.g.
approximately 25 different unique types of events. The 5-25 events).
design approach also needed to be flexible for future Horizontal time-aligned panels (approach B) are not
extensibility, on the assumption that users would widely new to this community. Analytic functions of continuous
adopt the newer design. (This was not a given). data based on mathematical algorithms used to detrend
In assessing the requirements, we considered that the data such as MACD, DMI and RSI are typically dis-
our design approach needed to fulfill Jacques Bertin’s [4] played in adjacent time-aligned panels immediately be-
three visualization functions: low the key price timeseries. What is different is the
1. Recording information i.e. a visual representa- placement of discrete data, i.e. event data, in multiple
tion that acts as an inventory, where everything adjacent aligned panels. This was not common in the
can be collected in one place, e.g. a road map. financial timeseries analysis community. Yet, there is
2. Processing information i.e. an environment for historic precedent for placing discrete events along a
visual analysis facilitating visual inferences common time axis with discrete horizontal bands, as seen
through perceptual mechanisms, e.g. a scatterplot in various old and recent charts e.g. [5,6,7,8]:
with color-coded points; and interactive tech-
niques such as filtering.
3. Communicating information i.e. a visual to
convey a message, e.g. an infographic or narrative
visualization.
We felt fulfilling Bertin’s three functions required
first and foremost a successful execution of the inventory
of all timeseries information in a single graphic. In ef-
fect, the goal was to create the graphical equivalent of a
roadmap for timeseries data. The visualization needed to
create a frame of reference where any kind of event data
could be placed and a user would know what the event
was and what other events were associated with it of the
same type or at the same time.
In addition to Bertin’s functions, we required a visu-
al solution that differentiated many event types in a com- Fig. 7. Example showing bands of continuous data
pact size independent of language to meet the needs of a above (bars and lines) and discrete events below
(text) aligned to a common horizontal time axis [5].
global user community.
Early conceptual sketches focused on two different
design approaches:
A. Icons close to the line. Event icons are placed
along the price line in the chart, consistent with
the then current approach. In addition, leaderlines
and logic are used to lessen issues with occlusion
and address data density. To differentiate between
event categories and event types, glyphs would
need to be designed to differentiate between cate-
gories and between types within a category.
B. Markers in time aligned horizontal panels.
Simple event markers are located in parallel pan-
els immediately below the price chart, and
aligned to a common time axis. In this case, the
icons are simply markers, as labels indicating the
categories of events can be located within each Fig 8. Example showing continuous data below and
horizontal panel. discrete categorical events above, all aligned to a
common time axis shown at the top of the chart [6].
C. Other approaches. Some other visualization ap-
proaches were considered that did not include a Rather than choose only one or the other, both ideas
horizontal linear time axis. However, it became a were blended together on the assumption that some users
requirement to use a familiar horizontal linear would prefer to have events located close to the price
time axis as the professional users were extremely line when events were few (and as occurred in their ex-
familiar with horizontal time oriented charts. isting interface); but there would inevitably be situations
Approach B, horizontal time-aligned panels, was where the main chart would be overwhelmed with too
appealing, as it seemed highly scalable and fit with the many events interfering with the users’ required need to
objective of creating the timeseries roadmap. visually parse the price line for patterns. The resulting
Approach A, icons along the line, was also appeal- design allowed users to move items between locations
ing in that users were already familiar with the technique with either drag and drop or via a traditional dialog box.
 As intended, the visual alignment of events allowed
for quick decoding of significant events in relationship to
stock price movements. Some examples:

4.2.1. Single Price Move and Corresponding

Event. The stock Motorola Mobility (MMI) had a huge
leap in stock price on August 15, 2011. Visually scan-
ning vertically at this date reveals many news stories
(blue 1,2,5 icons, e.g. ), an acquisition (light blue A
), and many analyst reports (white diamonds and blue
down arrow ). The catalyst for all the activity is likely
the acquisition event. This can be confirmed via tooltips
and drill-throughs, which show that Google announced
Fig. 9. Early design for the depiction of events along their intent to acquire Motorola Mobility for $10B.
price line in chart and in separate panels below.
In order to maintain a consistent representation of
events in both the chart and the aligned panels, a single
design was required. As these representations need to be
compact, a multivariate icon approach was used. The
glyphs needed to:
• Differentiate between categories, such as news or
stock trade.
• Differentiate within a category, such as a popular
news story or an earnings release.
• Indicate potential price impact, for example, an Fig. 10. Events associated with Motorola Mobility in
insider selling stock in a company could be per- 2011. The sharp price increase in mid-August is heav-
ceived as a negative influence on the stock price. ily covered in news stories (blue icons above the
The solution was a series of pictographic glyphs line), the trigger event being the acquisition (vertical-
(e.g. [9]) as opposed to abstract glyphs (e.g. [10,11, ly aligned A icon) of Motorola Mobility by Google.
12,13,14]), each glyph indicating multiple data attributes: 4.2.2. Multiple Price Drops and Corresponding
• Broad outline shape indicates the category, e.g. Events. Research In Motion (RIM/BB) stock price has
square for news, chevron for trades. dropped throughout 2011. In particular there are a num-
• Icon inside the shape to indicate the specific type, ber of sharp drops. Three sharp drops correspond to earn-
e.g. big letter E for earnings release, a pictograph ings events (green and red E icons, , directly along
of a human figure for insider trades. the price line), i.e. corporate filings of quarterly earnings
• Color of the broad shape indicates impact, e.g. red results, which in these three instances, the market reacted
for a stock sell, green for a buy, blue for neutral. negatively towards. There is one more sharp drop in
• Size was considered, but not used. price on April 28. Scanning vertically reveals a GU
(guidance update ), a number of analyst revisions (var-
4.2. Implementation and Examples ious H and S down arrows and white diamonds ) and
a negative GDP report (red flag in lowest panel ). Fig-
As design and implementation progressed, internal
ure 1 provides a large image. In this case, the trigger
product reviews by different product areas revealed
event was likely the guidance update (GU ), which is a
strong positive response to the design with many sugges-
corporate announcement forecasting future earnings,
tions for additional event types, including events for
again resulting in negative market response. This align-
commodities, messaging, automated patterns, rating
ment can be confirmed with a mouse over interaction
agency announcements, etc. Internal design critiques
which overlays a visual leader line connecting the icon to
provided incremental feedback to refine the design, e.g.
the price line and additional details in a tooltip (fig. 11).
glyph sizing based on data attributes was deferred as
there were unresolved design issues and sizing could be
added in later. Design critiques and reviews by subject
matter experts highlighted issues with co-incident events,
a need to contract multiple tracks together, leader lines,
labels, magnification, workflow, event location, etc.
The first version was targeted for users interested in
individual stocks. It was made available as one button on
a toolbar of seven buttons immediately above the chart.
Within twelve weeks there were thousands of users that
had organically discovered the new functionality, were Fig. 11. Events associated with RIM in 2011. The
using it many times per day, and usage was increasing. stock has a number of sharp drops, each of which
occurs when the company speaks about future earn- sor shape implied additional functionality to encourage
ing prospects, indicated as either E icons (earnings clicks or click-drag actions. Click-through workflows
releases) or GU icon (guidance update). were provided on every event type to provide paths to
4.2.3. Event Patterns and Corresponding Price other functionality in the system.
Changes. Scanning horizontally for patterns and anoma- Feedback came through direct meetings with users,
lies within an event type can be used for analysis. For seminars and customer support. The most common re-
example, Bank of America had a sequence of green and quest was related to event coverage. Since the new func-
yellow earnings reports (no negative surprises) shown in tionality increased legibility and offered more space for
the upper line of E markers in events, different users asked for additional events of in-
2006 with a single negative report in late 2007 (near the terest: such as equity offerings, investor presentations,
price peak), followed by a succession of negative reports extreme weather, broader country coverage, etc. More
through 2008 . The relationship between earn- than 200 unique events are now supported.
ing report color and stock price is strong: Another common request was configurability. For
example, unemployment reports were colored based on
the change in the rate from the prior period, however,
some users were more interested in the difference be-
tween the forecast and reported unemployment rates.
One unforeseen request was the need of technical
analysts to use events to “paint bars”, that is, to color the
timeseries data (e.g. price line) based on events that trig-
gered new states. This allowed expert users to visually
read the shape primary price line without shifting focus
onto the secondary icons; while at the same time having
close visual access to the event glyphs if needed.
Fig. 12. Bank of America earnings (E) and insider One problem expressed by users is related to filter-
transaction events (pictograph) shown in two rows of ing of events. Depending on the security, the time range
colored icons at the bottom of the chart. The shift displayed in the chart, the types of events selected and
from green and yellow earnings to red earnings cor-
other factors, thousands of events could be displayable.
responds to a significant decrease in price. The peri-
od of solid green insider transaction events through- To display the most salient events requires filtering tech-
out 2009 correspond with a period of price increases. niques, but the criteria for filtering is different for each
type of event and remains an on-going challenge. Alter-
The lower panel of events, Insider Transactions, is natively, a different representation was created for ex-
generally mixed between red, yellow and green tremely high frequency events, such as trades.
, except for strong continuous green pat- Many other incremental refinements were driven
tern throughout most of 2009 . Green in- from user feedback, such as adjustment of icon sizes,
sider transactions indicate corporate executives buying improvements to leader lines, performance improve-
stock in their own company. This could be interpreted as ments, etc.
a signal of confidence from the bank executives to buy The use of icons indicating multiple attributes was
their own stock. both a design and usability challenge. As a design chal-
lenge, it was difficult to create different unique picto-
4.3. Challenges and User Feedback graphs, as well as a broad range of background shapes.
The background shape had to provide a large area for a
There were numerous implementation challenges fit- foreground pictograph, making convex shapes inappro-
ting the new techniques within the existing system. All priate (e.g. stars) and inspired further investigation into
the implementation constraints (e.g. low CPU utilization, the potential different visual attributes associated with
high performance, fast start-up) were eventually shapes (e.g. [15, 16]).
achieved. There was also the challenge of the user ability to
Deploying the new visualization to users had various decipher the icons. A tooltip provided immediate feed-
issues. One issue was discoverability of the new func- back as to the event type. However, one early icon was
tionality and how to communicate this from the face of misinterpreted as a rude gesture and was quickly re-
the chart. The initial users of the new function were pre- placed with a different icon.
sented with a new look for the events (i.e. icons instead
of floating alphanumeric characters) and still located in
5. Results
the plot area of the main chart. If the user explored the
user interface that configured the event functionality, the Internal studies show that a significant portion of us-
user saw their familiar list of events, and the ability to ers have adopted the use of events since release in 2007.
place those events either on the chart or on a separate Another indicator of success is the growth in the ar-
panel. Users discovered the new functionality through ray of events which have been added by different
the use of their familiar interface. Rich tooltips provided Bloomberg business units seeking to expose more con-
greater detail than previously available. Changes in cur- tent in the chart. A recent count indicates that there are
now more than 200 unique event types available, show- 6. Future Work
ing that the original design goal of efficiency and exten-
sibility has been achieved and showing that there is de- The technique has scaled to 200 unique event types
mand from various user communities for additional data and there remain on-going challenges (e.g. designing
events not originally considered. unique, quickly understood icons) and new challenges
The approach led to several benefits for end-users: (e.g. browsing, finding and managing event types of in-
Information Inventory: The original concept pro- terest). Other applications areas for the visualization of
vides “a place for everything and everything in its event data are also under consideration.
place.” Similar to a roadmap, it provides a repeatable, The strategy and design objectives used here to
consistent framework for organizing information. Like a achieve Bertin’s functions through organizing an inven-
roadmap, it provides ways to understand different classes tory of information in a roadmap layout to support ana-
of local information through the use of data dense lytic processes is being used to solve other information
glyphs. And, like interactive maps, it provides means to challenges at Bloomberg.
add and remove different groups of data from the display
in an expected manner. References
Visual Correlation. By visually organizing data
vertically along a common time axis, the user can visual- [1] M. Bloomberg. Bloomberg by Bloomberg. Wiley, New
ly infer relationships at points in time by scanning verti- York. 2001.
cally. For example, for a given price anomaly, the viewer [2] M. Arrington. “Google Finance – Lots of Flash & Ajax,”
TechCrunch, http://techcrunch.com/2006/03/21/google-
can scan vertically to see events occurring at the same finance-lots-of-flash-ajax 2006.
time and infer a possible relationship, such as a trigger [3] J. Larkin, H. Simon. “Why a Diagram is (Sometimes)
event. By organizing groups of events into panels, pat- Worth Ten Thousand Words.” In Cognitive Science, vol.
terns within one type of event can be inferred by scan- 11, no. 1, pp. 65-99, Jan. 1987.
ning horizontally. Horizontally aligned labels reduce the [4] J. Bertin. Semiologie Graphique, Gauthier-Villars, Paris
requirement to memorize event types and the need to rely 1967.
on tooltips. The use of connotative coloring of icons aids [5] A. Grafton, D. Rosenberg. Cartographies of Time: A His-
tory of the Timeline. Princeton Architectural Press. 2010.
visual screening for particular events (e.g. looking for [6] W. C. Brinton. Graphic Methods for Presenting Facts.
red events associated with a price drop). Engineering Magazine Company, New York, 1919.
Data Density. The original need was to increase the [7] S. B. Cousins and M. G. Kahn, "The visual display of
data density. Previously, only a few dozen events could temporal information," Artificial Intelligence in Medicine,
be clearly depicted along a timeline. The new technique vol. 3, pp. 341-357, 1991.
provides compact icons that can be densely packed and is [8] C. Plaisant, et al., "LifeLines: Visualizing Personal Histo-
ries," in ACM CHI'96 Conference, 1996, pp. 221-227.
routinely used to show effectively anywhere from a few
[9] S. Rose. Social Stratification in the United States. The
events to a few hundred events. New Press. 2007.
Workflow. Furthermore, the technique provides an [10] A. Barr. Superquadrics and angle-preserving transfor-
effective workflow that provides a single consistent con- mations. IEEE Computer Graphics and Applications 1, 1
text (the visualization) and details as needed (through (Jan. 1981), 11–23.
both tooltips and a click-through pop-up). While not a [11] H. Chernoff. The use of faces to represent points in k-
“focus + context” display (e.g. [17]), the approach ena- dimensional space graphically. J. American Statistical As-
sociation 68, 342 (1973), pp. 361–368.
bles the users to easily move back and forth from a [12] R. Rohrer, D. Ebert, J. Sibert. The shape of Shakespeare:
common context to detail data. This enhances discovera- visualizing text using implicit surfaces. Proceedings Visu-
bility of functionality associated with any event type. alization 98 (Research Triangle Park, NC). Los Alamitos,
Adoption. The initial view of the new system did CA: IEEE Computer Society Press, 1998; 121 – 129.
not require extra cognitive effort by the users. The icons [13] F. Post, T. Walsum, F. Post. Iconic techniques for feature
utilized the same characters and colors for events that the visualization. Proceedings Visualization 95 (Atlanta, GA).
user were familiar with. Also the events were located Los Alamitos, CA: IEEE Computer Society Press, 1995;
288 – 295.
adjacent to the price line, with the addition of subtle [14] M. Ward. A Taxonomy of glyph placement strategies for
leader lines to aid the viewer’s ability to discriminate the multidimensional data visualization. Journal of Infor-
time of the event in relation to the price line. Through a mation Visualization 1 (3-4). 2004. pp. 194-210.
revised dialog box for managing events, viewers were [15] R. Brath. “The Many Dimensions of Shape.” IV09 Open-
exposed to the new option of aligned panels. The new ing Keynote, www.oculusinfo.com/expertise.html
functionality was steadily adopted by users as they dis- [16] R. Brath. “Multiple Shape Attributes in Information Visu-
alization: Guidance from Prior Art and Experiments.”
covered the additional functionality in their current
Proc. Information Visualization 2010 (IV10),14 th Int’l
workflow and did not require any specialized training or Conf. pp. 433-438, 2010.
documentation. [17] B. Spence and M. Apperley. “A Professional's Interface
Finally, another important result is that this creation Using the Bifocal Display.” In: Proceedings of the 1981
of a unique, efficient, easy-to-use visualization provides Office Automation Conference 1981. pp. 313-315, 1981.
unique benefits only available at Bloomberg. This pro-
vides a competitive business advantage in the growth of
new customers and retention of existing customers.