|
FORKAF was developed at the Institute
for Reliability and Risk Analysis, and forms part of a suite of
programs designed to work in the Microsoft Windows environment.
OVERVIEW
The program is concerned with forecasting time series, and uses the theory
of dynamic linear models. The novelty of the forecasting technique is that
in forecasting one data series, information from related series is used to
influence the forecasts. Hence, if one has multiple series with a common
trend, the forecasts given by this software will be based on the data from
all series. This is a significant advantage over both standard dynamic
linear model techniques and other more primitive forecasting methods, where
similar data sets are treated independently. The models available in the
software are the linear growth model, the growth model with bends, and the
S-shaped model. These models cover a wide range of different modelling
needs, and are particularly appropriate to the forecasting of warranty
claims. However, they are not restricted to warranties - they can be used
for any application that involves many data sets with a similar trend.
DATA INPUT
|
|
|
The
software has an input screen, for entering data from the multiple time
series. Once entered the data can be saved for future analysis. Once a
data set has been input, there are a number of analysis options available
to the user. First, one must choose a series to forecast, and specify how
many forecasts to make. Then, one may specify which of the other series
to use as leading indicator series (series that influence the series of
interest) and specify how much these indicator series should affect our
forecasts. Finally, one must specify the amount of variation in the data.
In case the user is unsure of what type of analysis to perform, the
software automatically defaults to the most usual set-up.
|
RESULTS
|
|
|
Once an analysis has been specified, the program will calculate the
forecasts for the specified period. Once this is done, the user is faced
with a number of display options for the generated analysis. First, the
forecasts are available in tabular format, with 95%, 90%, 75% and 50%
prediction intervals, plus estimates of forecast mode and variance.
Alternatively the user may view the forecasts in a graphical form, in the
form of a Box-Whiskers plot. By clicking the mouse on an individual bar
of the plot, graph is replaced by the predictive distribution of the
forecast of interest. The user may then return to the original plot
either by selecting the option on the menu-bar or by use of a pop-up
menu. This pop-up menu can also be used to cycle through forecast values,
to get an idea of how forecasts evolve with time.
|
OTHER FEATURES
The program has all the usual features of a Windows-based program,
including full file management facilities and the facility to print the
results either in tabular or graphical form.
REFERENCE:
Chen. J., Lynn. N., and Singpurwalla N. D. (1996). 'Forcasting Warrenty
Claims.' In Product Warrenty Handbook; W. R. Blishke and P.
Murthy, eds. Marcel Dekker, Inc. New York, 803-816.
|
