This page was last edited on 12/14/99 by Malcolm R Forster

Time series data consists of a series of values of a variable. The example (which is straight from Pandit and Wu (1983)) consists of 161 values of a quantity measured once every 20 minutes.  The data is plotted in the background of this page.   Time is along the x-axis and the value of the variable is on the vertical y-axis.   If the problem were treated as a simple curve-fitting problem, then one would try to represent this quantity as a function of time.  That function would be a smoothish curve that would follow the trend of the data.  The curve would then be used to predict the new data.  How effective is this for the purpose of prediction?
      The alternative times series model does something different.  It looks for a dependence of the value of the quantity at time t on the value of the quantity at previous times.  These models are curve-fitting models as well, albeit between different variables.  Times series models can fit better and be simpler at the same time.  This note explains that in terms of the example.

Note: You need Adobe Acrobat Reader 3.0, or later, to read and print this article.  It is free.

Version 1, 17 KB (just 2 pages)

As the conference approaches, I will expand the abstract and post it here.

Pandit, Sudhakar M. and Wu, Shien-Ming (1983): Times Series and Systems Analysis with Applications, John Wiley and Sons, New York.

N/A