Information Technology Reference
In-Depth Information
Hibon, 1979, 2000). The M3-Competition is a kind of a tournament set up to evaluate the performance
of various forecasting techniques. It features various academic and commercial forecasting methods,
including Naïve/simple, explicit trend models, decomposition, variations of the general ARMA (auto-
regressive moving average) model, expert systems and neural networks. The results, even with a new
and much enlarged set of data demonstrated, that simple methods developed by practicing forecasters
do as well, or in many cases better, than more sophisticated ones (Makridakis & Hibon, 2000).
Even though there is an array of forecasting methods that perform well in certain situations and
simple traditional forecasting techniques seem to outperform more complex ones, there is still no rec-
ommendation for the best forecasting technique to use that would result in the most accurate forecast.
f orecasting Methods used by Practitioners
Jain (2004c) notes that in general there are three types of forecasting models commonly used in the
industry, including: Time-Series (71%), Cause-and-Effect (19%), and Judgmental (10%) categories.
Within the Time-Series category, the most common methods are Averages and Simple Trend (65%);
Exponential-Smoothing (24%); Box-Jenkins (6%); and Decomposition methods (5%). The Cause-and-
Effect modeling is most often executed with Regression analysis (74%), Econometric models (21%), and
Neural Networks (5%). Furthermore, Jain (2004b) indicates that when these models are implemented
using computer software; 63% of the forecasting market share is held by Microsoft Excel. Among the
stand-alone forecasting software packages, John Gault has a large part of the market (46%) followed by
SAS (28%). Regarding the integrated software solutions, SAP (23%) and Manugistics (16%) dominate
market share. From these numbers we see that most forecasting is still done with time-series analysis in
spreadsheet software and that a minority of businesses uses integrated solutions that include forecasting
solutions. The average forecasting error across all industries for 1 month ahead is 26%, for 2 months
and 1 quarter ahead is 29% and for 1 year ahead is 30% (Jain, 2004a). This indicates that there can be
significant competitive advantages gained from improved forecast accuracy.
On the other hand, Wisner and Stanley (1994) reported that 39% of their respondents rarely or never
used forecasts, which may indicate lack of data available for forecasting, lack of simple but effective
forecasting techniques, or lack of skills and resources needed to perform forecasting. Additionally, 73%
of respondents indicated that they have been using purchasing-forecasting for less then 10 years. With
respect to actual forecasting, 67% indicated that they generated their forecasts manually and about half
attempted to change the forecast parameters to increase forecast accuracy. Although this adjustment
may lead to better forecasts, it is important to note that tuning forecasts may result in high accuracy on
the known existing data set while decreasing the accuracy of future forecasts. These adjustments can
potentially lead to over-fitting of the forecasting to past data, which can decrease forecasting accuracy.
This is not only detrimental to the individual supply chain member, but also to other members since the
increased forecasting errors drastically increase the overall supply chain demand distortion in a cascad-
ing fashion. The top quantitative forecasting techniques are simple moving average (62%), weighted
moving average (46%), exponential smoothing (45%), exponential smoothing using trend and seasonal
enhancements (39%), simulation model (22%), regression model (21%), econometric model (19%) and
Box-Jenkins model (14%) (Wisner & Stanley, 1994). It seems that, on average, purchasing managers rely
on manual and simple forecasting techniques and that there is a lack of simple but powerful forecasting
techniques that can be effectively adopted and relied on by practitioners.
Search WWH ::




Custom Search