# BA544FORECASTINGW05doc - BA 544 - FORECASTING IN SUPPLY CHAINS

Document Sample

```					BA 544 - FORECASTING IN SUPPLY CHAIN MANAGEMENT

Dear Class:

Here are the types of forecasting questions that I might ask. If you study and know all of these, you
should be in great shape for the forecasting section. I will limit my questions to the following, 1) to 4). If
you do not see something below, then it will not be on the test. For example, the compete examples are
not included here and therefore will not be on the test, even though I view those as very important, state
of art concepts (however, they are very easily learned concepts if you understand the following
information).

Best of Luck,

Prof. Steve De Lurgio

The types of forecasting questions will be based on the following (the selected question will test over the
concepts developed below).

1) A simple seasonal exponential smoothing question that requires you to calculate and interpret:

MAPE
RSE = SEE = square root of MSE
Mean Error =Bias
TS
This is very much like the example from the past test.

This will require the computational part to be short, such as two to four period long at most.

2) Output of a Holt-Winters presentation where you to interpret the fitted versus forecasted values. Refer
to the spreadsheet we used in class.

http://forecast.umkc.edu/ftppub/ba544/FORECASTING/HOLT-WINTERS.xls

I might include output from this spreadsheet on the test and then ask you to interpret it. I might include
every worksheet from this spreadsheet (i.e., from the workbook).

3) Given the following information, answer the next three questions. All calculations are based on
periods t = 22 through t = 26. Only one statistic is to be calculated for this period.
t FORECAST ACTUAL
22 1005                  1000
23 1015                  1010
24 1025                  1025
25 1045                  1035
26 1040                  1055

a) What is the mean forecast error for this model?
b) Explain its meaning using a common English definition.
c) How can a manager use this statistic in estimating demand for the next four weeks (weeks 27 to 30)?
d) What is the MAD for this model?
e) Explain its meaning using a common English definition.
f) How can a manager use this statistic in estimating demand for the next four weeks (weeks 27 to 30)?
g) What is the tracking signal for this model?
h) How can a manager use this statistic?
i) What is the mean absolute percent forecast error?
j) Explain its meaning using a common English definition.
k) What conclusions do you make about the forecasting model based on these statistics?

4) Knowing how to apply the following:

Holt's exponential smoothing method. An extension of single exponential smoothing that allows for
trends in the data. It uses two smoothing parameters, one of which is used to add a trend adjustment to the
single smoothed average value. The smoothing constants are called alpha and beta which are not shown
below nor are you responsible for knowing the smoothing equations. The smoothing constants alpha and
beta are used to arrive at the current values of S(t) and b(t).

F(t) = S(t) + b(t)                                                 [1]

F(t+m) = S(t) + m*b(t)                                             [2]

Holt-Winters' exponential smoothing method. Winters extended Holt's exponential smoothing method
by including an extra equation that is used to adjust the forecast to reflect seasonality. This form of
exponential smoothing can model data series that include both random, trend, and seasonal elements. It
uses three smoothing constants, alpha, beta, and gamma that are used to arrive at the current estimates of
level, trend, and seasonality.

F(t) = (S(t) + b(t))*I(t-12)                                       [3]

F(t+m) = (S(t) + m*b(t))*I(t-12+m)                                 [4]

What are the definitions of F(t), F(t+m), m, S(t), b(t), and I(t-12+m)

Which terms are in the above relationships for the following models?
Smoothed, No Trend, No Seasonality Model
Smoothed, Trend, No Seasonality Model
Smoothed, No Trend, Seasonality Model
Smoothed, Trend, Seasonality Model

Simple Exponential Smoothing

With simple exponential smoothing, explain the meaning of zero smoothing and infinite smoothing. Be
sure to explain in the context of:

F(t) = alpha*A(t-1) + (1-alpha)F(t-1)                              [5]

Explain the function of the smoothing constant in the above relationship.

Understanding simple seasonal exponential smoothing:

F(t) = alpha*A(t-12) + (1-alpha)*F(t-12)                           [6]

You have all of these equations ([1] to [6]) on your formula sheet.
Dear Class:

I have put this information together to assist you in studying forecasting in the context of supply
our optional study review class before the midterm exam.

Powerpoint Lecture Notes

The following powerpoint lecture notes are important for you to study. I have edited that document
eliminating any slides that are not important this semester. You will find relevant formulas in this
document. Rather than printing this out, simply take your old copies of these slides and delete slides that
are not important.
Study http://forecast.umkc.edu/ftppub/ba544/PPT/CCH07D.ppt through slide 29. In addition to the
above, I have included a glossary of important terms and concepts that should be studied. This can be
found at:
http://forecast.umkc.edu/ftppub/ba544/forecastingglossary.doc

Included in the concepts of the above slides and glossary are computational procedures. The
computational methods and formulas you are responsible for are:
Simple moving average method.
Simple exponential smoothing.
See http://forecast.umkc.edu/ftppub/ba544/xls/C4P2.xls for examples.
Seasonal simple exponential smoothing, see below.
Know how to apply the Holt-Winter’s formulas shown below in equations [1] through [6].

In addition, study the methods developed in our forecasting competition that can be found at the
following url:
http://forecast.umkc.edu/ftppub/ba544/xls/COMPETEDATA.xls
MAPE
MSE
RSE = SEE = square root of MSE
Bias
MPE
TS
The definitions to the above terms can be found at either the powerpoint presentation or xls spreadsheets
given above.

You are not responsible for understanding the way the smoothing constants are determined for Holt-
Winter’s Exponential smoothing. However, you are responsible for understanding how to apply formulas
[1] to [6]:

Holt's exponential smoothing method. An extension of single exponential smoothing that allows for
trends in the data. It uses two smoothing parameters, one of which is used to add a trend adjustment to the
single smoothed average value. The smoothing constants are called alpha and beta which are not shown
below nor are you responsible for knowing the smoothing equations. The smoothing constants alpha and
beta are used to arrive at the current values of S(t) and b(t).

F(t) = S(t) + b(t)                                                [1]
F(t+m) = S(t) + m*b(t)                                     [2]

Holt-Winters' exponential smoothing method. Winters extended Holt's exponential smoothing method
by including an extra equation that is used to adjust the forecast to reflect seasonality. This form of
exponential smoothing can model data series that include both random, trend, and seasonal elements. It
uses three smoothing constants, alpha, beta, and gamma that are used to arrive at the current estimates of
level, trend, and seasonality.

F(t) = (S(t) + b(t))*I(t-12)                                       [3]

F(t+m) = (S(t) + m*b(t))*I(t-12+m)                                 [4]

What are the definitions of F(t), F(t+m), m, S(t), b(t), and I(t-12+m)

Which terms are in the above relationships for the following models?
Smoothed, No Trend, No Seasonality Model
Smoothed, Trend, No Seasonality Model
Smoothed, No Trend, Seasonality Model
Smoothed, Trend, Seasonality Model

Simple Exponential Smoothing

With simple exponential smoothing, explain the meaning of zero smoothing and infinite smoothing. Be
sure to explain in the context of:

F(t) = alpha*A(t-1) + (1-alpha)F(t-1)                              [5]

Explain the function of the smoothing constant in the above relationship.

Understanding simple seasonal exponential smoothing:

F(t) = alpha*A(t-12) + (1-alpha)*F(t-12)                           [6]

You will have all of these equations ([1] to [6]) on your formula sheet.

Important forecasting principles:
The forecast will be wrong, the best we can hope for is to significantly reduce the error in forecasting.
There is not one best method that can be applied to all time series.
A good method must be able to model all of the patterns in the data.
The statistical definition of a best model includes the lowest SEE in forecasting the future.
Good systems employ multiple methods.
Good systems choose the best method for a series based on out-of-sample forecasting accuracy, not based
on the fit of the model. (We will develop this concept more in class.)
A good system will facilitate a management by exception using tracking signals and error filters.
The ability to handle outliers and promotional influences is critical in a modern forecasting package.
At a minimum a forecasting method should be able to model trend, seasonal, promotional, and outlier
patterns.
Many studies have verified that forecasts based on the average of two or more independent methods will,
on average outperform the forecast of a single method.

You should be able to illustrates simple moving averages versus exponential smoothing and the use of
tracking signals and other diagnostic statistics. http://forecast.umkc.edu/ftppub/ba544/xls/C4P2.xls. You
should understand each of these concepts.

You should understand the forecasting competition time series and the simple and complex models that
we used to model those time series found at: http://forecast.umkc.edu/ftppub/ba544/compete.doc

The following spreadsheet illustrates simple exponential smoothing, Holt’s Two Parameter Exponential
Smoothing, and Holt-Winter’s Three Parameter Exponential Smoothing:

http://forecast.umkc.edu/ftppub/ba544/FORECASTING/HOLT-WINTERS.xls

In addition, you should understand the manner in which seasonal indexes are calculated in the above
spreadsheet along with an understanding of and ability to apply tracking signals.

Identify the dominant patterns of SERIESA.XLS, SERIESB.XLS, SERIESC.XLS, SERIESD.XLS,
SERIESE.XLS AND SERIESF.XLS and illustrate the best forecasting formula from equations [1] to [6]
above being sure to include only those terms in the Winters relationship that are functional in these time
series.

Using the data and forecasting models that were developed as the “correct solutions” for the competition
data, complete the following task for each of the five time series (items 1 to 5):

a) In common English terms explain each model.
b) Forecast twelve months into the future, that is for periods 25 to 36. There are planned promotions for
both items 3 and 5 for periods 29 and 33.
c) For periods 1 to 24 calculate the fit statistics of mean error, standard error of estimate, and Mean
Absolute Percent Error.
d) Interpret each of these error statistics for each of the models. (Our handout material and pages 190 to
192 of the Chopra textbook discusses these error statistics.)

To make this task easier, I have provided you with a spreadsheet that has all of the data in a format for
http://forecast.umkc.edu/ftppub/ba544/FORECASTING/COMPETEDATA.xls

As you complete a) to d) please study the handout material and/or refer to the following glossary for
definitions and help: I have put together a forecasting glossary for your use, study this as part of your
assignment, not every term but those that relate to what we have done in class and homework:
http://forecast.umkc.edu/ftppub/ba544/FORECASTING/forecastingglossary.doc

Finally, do not forget the most recent forecasting assignment and explanations found at:

http://forecast.umkc.edu/ftppub/ba544/FEB21STASSIGNMENT.doc

Formula Sheet:
I am in the process of revising the formula sheet for our first test.

```
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
 views: 17 posted: 5/14/2010 language: English pages: 5