Student Seminar Series - September 30, 2005
University of Minnesota
School of Statistics
College of Liberal Arts
Methods
for Unconstraining Right-Censored Demand Using the EM Algorithm and
Other Techniques
Melissa Skluzacek
Friday, September 30, 2005
9:00 AM, 127
Ford Hall
Minneapolis, East Bank Campus
Refreshments at 8:45 AM
300 Ford Hall
Abstract
Accurate forecasting is a crucial part of the airline revenue
management process. True demand is not always captured as a result of
the optimization process; so many historical observations are
right-censored. Using censored data in the forecasting models leads to
a forecast that is negatively biased. Therefore, airlines seek to
“unconstrain” the historical censored demand prior to the forecasting
process.
There are four major techniques for handling censored data:
ignore, discard, impute, and statistically estimate using the EM
Algorithm. These methods are investigated using simulation studies on
both the historical demand and constraining levels. When exploring the
four different methods on the simulated data, statistical estimation
via the EM Algorithm is superior to all other methods in reducing
negative bias regardless of demand level, number of historical
observations, or censoring level. Additionally, the number of
iterations needed for convergence does not appear to be problematic.