Suchergebnisse

Applies time‐series forecasting, a traditional operations analysis
methodology, to develop a forecasting procedure and ordering policy for
a natural‐gas customer of Columbia Gas of Ohio, USA. Evaluates six
time‐series methods and four operating policies against four commonly
used measures of error and the cost consequences of error to the
customer. Demonstrates that time‐series forecasting and decision theory
developed by operations and applied in an actual industrial situation
can become a powerful marketing technique. Provides further insights
into evaluating forecasting models and ordering policies, demonstrating
that introducing optimal planned bias is a robust
decision‐making/forecasting approach within services. There are three
parts to the study. The first is a straightforward testing of
forecasting methods, using the forecasts as the natural‐gas ordering
policy. Results vary depending upon how well forecasts are fitted to the
data. For example, one inaccurate forecast with a poor fit incurs a
penalty cost of $179,270, while the best forecast results in a
penalty cost of $27,081. The second part evaluates two additional
complex ordering rules with the same forecasting methods, further
reducing the lowest cost to $17,709. The third part is a
technical analysis reflecting a redesign of the study, demonstrating the
difficulty of generalizing when characteristics of the underlying demand
change. Concludes that the best forecasting model/operating policy is to
use the very basic forecasting model of simple moving average (or the
equivalent, first‐order exponential smoothing) combined with an optimal
planned bias ordering policy, i.e. with the planned introduction of
bias.