FinanceVis - Design and Evaluation of Visualization Methods to Compare Multivariate Time Series

Team
  • Christian Kainz,Vienna University of Technology, Institute of Software Technology and Interactive System, Vienna, Austria
  • Rui Ma,Vienna University of Technology, Institute of Software Technology and Interactive System, Vienna, Austria
  • Wolfgang Aigner, Vienna University of Technology, Institute of Software Technology and Interactive System, Vienna, Austria
Status
finished

The comparison of several stocks against each other is an important task in financial analysis. The data is usually compared by juxtaposition or by superimposition. But even a large enough difference between the price ranges can make a simple comparison of the data difficult.

Stock market data is not limited to stocks. It also includes stock (market) indices such as the Dow Jones and NASDAQ. Stock indices have their own proprietary unit, which is different to the unit of a stock. Time series which have different units are also called heterogeneous data.

To compare heterogeneous data by superimposition multiple y-axes are often used. But in most cases the arrangement of the axes makes comparisons between different variables completely arbitrary and comparisons are often misleading. Also intersections and spatial distances are depending on the scale and may be misleading as well. The visualization pioneer Jacques Bertin has studied the problem and has also worked out some suggestions. One suggestion is to index the values, which transforms all data into values which reflect the percent change compared to an indexing point.

To evaluate the indexing method a prototype application in Java was developed. The prototype offers multiple visualization methods to compare data. The prototype application can display the following stock charts:

  • Line chart (juxtaposition, superimposition and indexing)
  • OHLC chart
  • Candlestick chart

It also features switching between linear and logarithmic scale and common user interactions such as zooming and panning. The prototype also has a built-in evaluation mode, which was very helpful to measure the task time and task correctness for a pre-defined set of user tasks.

The requirements of the prototype were gathered from scientific literature, stock market web applications and from an user & task study. Six domain experts in the field of stock market analysis were interviewed in this study.

The prototype was then used to conduct a user study in order to get feedback from five domain experts about the implemeted visualizations. A comparative study with 24 test persons helped to find differences between visualization methods. Linear and logarithmic scales and comparisons by juxtaposition, superimposition and indexing were tested for significant differences in task completion times and error rates.