Graphics Reference
In-Depth Information
high price velocities on day
and high prices on day
). We are currently work-
ing on developing a facility for “curve-matching" that is more automated. For
instance, consider the case of forecasting the closing price of a seven-day auc-
tion that is scheduled to close on a Sunday, with an opening price of $
.
, and
that has displayed very low dynamics so far. Let us assume that we observe this
auction until day
(
% of the auction duration). Figure
.
illustrates a selec-
tion of auctions that all have similar attributes tothe above auction (all are seven
days long, have an opening price of less than $
, and close on a weekend), and
also have similar curve structures during the first six days of the auction (low
velocities, as shown by the filtering box placed on the velocity curves).
similar set:
Make a forecast based on the We then use the selected “similar" set of
curvestomakeapredictionfortime t byexamining their riverplots. hemedian
at time t is then the forecast of interest, and the quartiles at that time can serve
as a confidence interval. Although this is a very crude method, it is similar in
concepttocollaborativefiltering.hekeyistohavealargeenoughdataset,sothat
the “similar" subset is large enough. To continue our illustration, Fig.
.
shows
the riverplot of the subset of “similar" auctions. he forecasted closing price is
then the median of the closing prices of the subset of auctions, and we can learn
about the variability in these values from the percentile curves on the riverplot.
he forecasting module is still under development, with the goal being a more auto-
mated process. However, the underlying concept is that interactive visualization can
support more advanced operations (including forecasting) than static visualization.
Figure
.
.
Filtering the data to find a set of “similar" auctions to an ongoing open auction
