Docstoc

Hercules Inc. Historical Stock Prices - DOC - DOC

Document Sample
Hercules Inc. Historical Stock Prices - DOC - DOC Powered By Docstoc
					        Topic 12

Demand Forecasting in POM
                      FORECASTING
1. Forecasting vs. Prediction:

   Forecasting: Estimating Future by Casting Forward Past Data.

   Prediction: Estimating Future based on Subjective

     Considerations other than just Past Data.



2. Three Levels of Forecasting in Operations Management:

   Long Range Forecasting for Aggregate Demand.

   Intermediate Range Forecasting for Product Groups.

   Short Range Forecasting for Individual Item.



3. Forecasting Objectives:

   Forecasting in POM: Item Demand (Workload for Capacity

     Planning)

   Forecasting in Finance: Dollars Revenue (Cash Flow

     Requirement)

   Forecasting in Marketing: Unit of Sales (Selling Capability)


                                 Topic 12 - 1
              Forecasting in Operations
Forecast for                  Time horizon         MGT level
operations
Product & Process             Long range           Top
design
Capacity Requirement          Long/                Top / middle
Planning                      Intermediate

Aggregate Production          Intermediate         Middle
Planning
Production Scheduling Short                        Low



Impact of Inaccurate Demand Forecasting:

(Production Planning based on Demand Forecasting)

1. If Forecasting is consistently higher than Actual Demand:




2. If Forecasting is consistently lower than Actual Demand:




                             Topic 12 - 2
Forecasting Systems




      Topic 12 - 3
                         Forecasting Techniques


Qualitative Approach:
1. Delphi Methods: (Expert's Subjective Ratings)
2. Marketing Research and Analysis: (Customer Survey)
3. Historical Analogy: (Knowledge of Similar Products)
4. ...................


Quantitative Approach: (Two General Techniques)


A: Time Series Analysis:
1. Simple Moving Average
2. Weighted Moving Average
3. Exponential Smoothing
4. .................


B: Causal Relationship Models:
1. Regression Analysis
2. Econometric Models
3. ................




                                 Topic 12 - 4
       Selection of Forecasting Techniques
Principle of Forecasting:
* When Past Data are Known as Good Indicator for the Future.
* The Pattern of the Future can be recognized from Past Data.


Qualitative Techniques: (Subjective and Judgmental)
   Data Unavailable
   Unknown Pattern Change
Examples: Long Range Forecasting/Sales of New Product/...


Quantitative Techniques: (Objective and Quantitative)
A. Time Series Models: Used when Past Demand is A Good
Indicator for Future Demand. (e.g., Short Range Forecasting for
Scheduling)


B. Causal Relationship Models: Used when the Demand of an Item
is Dependent (Related to) on other Underlying Factors (not the Past
demand). (e.g., Short and Intermediate Range Forecasting of
Existing Products/Sales/Financial Data/....)




                              Topic 12 - 5
        Three Simple Time Series Models


1. Simple Moving Average: Given Number of Periods (n) to be

Averaged.




2. Weighted Moving Average: Given (n) and Weights (w i)




3. Simple Exponential Smoothing: Given α (smooth Constant).




                            Topic 12 - 6
Exercise #1

Assume that your stock of merchandise is maintained based on the
forecast demand. If the distributor’s sales personnel call on the first
day of each month, compute your forecast sales by each of the three
methods requested here.
                  June                July             August
Actual demand 140                     180              170


a) Using a simple three-month moving average, what is the forecast
for September?




b) Using a weighted moving average, what is the forecast for
September with weights of 0.20, 0.30, and 0.50 for June, July and
August, respectively?




c) Using simple exponential smoothing and assuming that the
forecast for June had been 130, calculate the forecast for September
with a smoothing constant alpha of 0.30.


                               Topic 12 - 7
          Forecasting Error Measurement
1. Forecasting Error in (t):

Et = (At - Ft)



2. Bias (Mean Error):

Bias = ∑ (Et)/n

     =∑ (At - Ft)/n

(Bias is a Measure of the Direction of Forecasting Error.)



3. MAD (Mean Absolute Deviation):

MAD = ∑|(At - Ft)|/n

(MAD is a Measure of the Size of Forecasting Error.)



4. Track Signal:

TS = Bias/MAD ( -1 ≤TS ≤+1)

(TS is a Measure of Forecasting Error in terms of both the

"Direction" and the "Size".)


                               Topic 12 - 8
            Forecast Error Measurement

Forecast error is the difference between the forecast and actual
demand.
  1. Cumulative sum of forecast errors (CFE)
  2. Average forecast error equals (Bias = CFE/n)
  3. Mean squared error (MSE)
  4. Standard deviation (σ)
  5. Mean absolute deviation (MAD)
  6. Mean absolute percent error (MAPE)
  7. Tracking signal = CFE/MAD


Criteria for Evaluating Forecasting Models
   Minimize bias
   Minimize MAD or MSE
   Meet managerial expectations of changes
   Minimize the forecast error last period
   Use a holdout set (data from more recent periods) as a final
     test


Techniques for Improving Forecasting:
  1. Combination forecasts
  2. Focus forecasting
  3. Research indicates that:
      Simple techniques often perform as well or better than
       sophisticated procedures
      There is no one best forecasting technique for all products
       and/or services




                              Topic 12 - 9
                              Exercise #2

Month        1    2      3      4       5      6     7       8      9

Actual       26   32     39     40      38     47    50      59     56

Demand



a) What is the forecast for period 10 using the following models?

     i) Four period moving average

     ii) Four period weighted moving average with weights of 0.1,

     0.15, 0.25 and 0.5 for month 6, 7, 8 and 9 respectively

     iii) Simple exponential smoothing with alpha =0.2, assuming

     that the forecast for month 9 is 60.

b) If the forecasts for period 1 to period 9 were 30, 35, 42, 42, 40, 50,

53, 61, 59, respectively, determine the bias, MAD and tracking

signal. How would you improve the forecasting accuracy in this

situation?




                               Topic 12 - 10
                             Exercise #3


Assume that you are forecasting a weekly demand for Item A using

the simple exponential smoothing method. You are now at the end

of week 9. The forecasting for week 9 is 1008 units, and the actual

demand in week 9 is 1024 units. The smooth constant α is 0.2.



  a) Develop a forecast for week 10.

  b) At the end of week 10, you learned that the actual demand in

     week 10 was only 631 units. You use the last 4 periods in

     computing Bias and MAD, and the forecasting error in week 6

     was -90 units. The Bias and MAD calculated in week 9 were -

     50 and 60 units respectively. Compute the Tracking Signal for

     week 10.




                             Topic 12 - 11
     Elements of Time Series Analysis




Base Level:         Demand at Current Time.

Trend (Linear):     General Direction of Demand Growth

Seasonal Effects:   Pattern that Repeat Every Year

                    (Month/Quarter)

Cyclical Effects:   Pattern that Repeat other than a Year.

Random Errors:      Unpredictable Random Variations. 10.



                            Topic 12 - 12
         Advanced Time Series Models
         (Base, Trend, Seasonal and Cyclical Effects)

1) Base Level Model: (no Trend/Seasonal/Cyclical)
        Ft,k = Bt (forecast for the Kth period made at period t.

  Updating after At known:
      Bt = Bt-1 + α(At – Bt-1)

2) Base and Linear-Trend Model: (no Seasonal/Cyclical)
        Ft,k = Bt + K ∙ Tt

  Updating after At known:
      Bt = (Bt-1 + Tt-1) + α [At – (Bt-1 + Tt-1)]
      Tt = Tt-1 + β[(Bt - Bt-1) – Tt-1)]
      β is the Trend Smooth Constant (0 < β < 1)

3) Base and Seasonal Model: (no Trend/Cyclical)
        Ft,k = Bt ∙ SIt+k

  Updating after At known:
      Bt = Bt-1 + α(At/SIt+k – Bt-1)
      SI◦t+k = SI t+k + γ(At/Bt - SIt+k)
      γ is the Seasonal Smooth Constant (0 < γ < 1)

4) Base, Trend, and Seasonal Model: (no Cyclical)
        Ft,k = (Bt + k ∙ Tt) ∙ SIt+k

  Updating after At known:
      Bt = (Bt-1 + Tt-1) + α[At/SIt+k - (Bt-1 + Tt-1)]
      Tt = Tt-1 + β[(Bt - Bt-1) – Tt-1)]
      SI◦t+k = SI t+k + γ(At/Bt - SIt+k)
      (α, β, and γ; three Smooth Constants in this model)


                           Topic 12 - 13
          Regression Forecasting Models

In many cases, the Demand of an Item (dependent variables) is more

dependent upon other Leading Factors (independent variables) than

the Past Demand.



Regression Models are developed based on Least-Square method.



1. Linear Regression Models: (Simple vs. Multiple)

Y = a + b1*X1 + b2*X2 + ...... + bn*Xn



     When n=1, it becomes the Simple Regression Model,

     Y = a + b*X (a: Intercept, b: Slope)




2. Non-Linear Regression Models: (no general model)

Y = a + b1*X1 + b2*X21




                             Topic 12 - 14
           Practical Forecasting Problems


1. Practical Forecasting Issues:
   Inaccuracy
   Inconsistency
   Cost and Accuracy Tradeoff (Simple model may perform
     better than complicated ones.)
   Data Unavailability
   Fitness and Predictability
- A model that best "fits" the past data may not be the best
"Predictive" one for the future, due to demand pattern changes.


2. New Direction in Forecasting:
   Integrated (Pyramid) Forecasting System: To Reduce
     Inconsistency.
   Combinational Forecasting Models: To Reduce Inaccuracy
     through:
        o Model Combinations
        o Result Combinations




                              Topic 12 - 15
 Criteria for Selecting a Forecasting Method
Cost and Accuracy
      A trade-off between cost and accuracy- more accuracy at a
       cost.
      High-accuracy with disadvantages of: need more data/data
       may be difficult to obtain/models are more costly to design,
       implement, and operate/take longer time to use.
      Low-Cost approaches- statistical models, historical
       analogies, executive-committee consensus
      High-Cost Approaches- complex econometric models,
       Delphi, and market research

Data Availability
   Is the data available/or be economically obtained?
   For a new product, a customer survey may not be practical.

Time Span
   What operations resource be forecasted and for what purpose?
   Short-term best be forecast with simple time series model.
   Long-term best be predicted with regression or similar models.

Nature of products and services
   Is the product/service high cost or high volume?
   Where is the product/service in its life cycle?
   Does the product/service have seasonal demand fluctuations?

Impulse response and noise dampening
   An appropriate balance must be achieved between:
   How responsive the model to change in the actual demand data
   Desire to suppress undesirable noise in the demand data.




                             Topic 12 - 16
                               Exercises
                  Demand Forecasting- Simple Regression
1. Finley Heaters Inc. is a mid sized manufacturer of residential water heaters. Sales have grown
during the last several years, and the company’s production capacity needs to be increased. The
company’s management wonders if national housing starts might be a good indicator of the
company’s sales.

Year                          National Housing Starts        Finley Heaters’ Annual
                              (millions)                     Sales (millions of dollars)
1                             6.2                            57
2                             5.1                            59
3                             6.5                            65
4                             7.9                            78
5                             6.3                            72
6                             7.4                            80
7                             7.0                            86

    a. Develop a simple linear regression analysis between Finley Heaters’ sales and national
       housing starts. Forecast Finley Heaters’ sales for the next two years. The National Home
       Builders Association estimates that national housing starts will be 7.1 million and 8.0
       million for the next two years.
    b. What percentage of variation in Finley Heaters’ sales is explained by national housing
       starts?
    c. Would you recommend that Finley Heaters management use forecast from Part a to plan
       facility expansion? Why or why not? What could be done to improve the forecast?

2. Chasewood Apartments is a 300/unit complex near Fairway University that attracts mostly
university students. Manager Joan Newman suspects that the number of units leased during each
semester is impacted by the number of students enrolled at the university. The university
enrollment and number of apartment units leased during the past eight semesters is:

Semester                      University Enrollment          Number of Units Leased
                              (thousands)
1                             7.2                            291
2                             6.3                            228
3                             6.7                            252
4                             7.0                            265
5                             6.9                            270
6                             6.4                            240
7                             7.1                            288
8                             6.7                            246




                                          Topic 12 - 17
    a. Use a simple regression analysis to develop a model to forecast the number of apartment
       units leased, based on university enrollment. If the enrollment for next semester is
       expected to be 6,600 students, forecast the number of apartment units that will be leased.
    b. What percent of variation in apartment units leased is explained by university
       enrollment?
    c. How useful do you think university enrollment is for forecasting the number of apartment
       units leased?




                  Demand Forecasting- Moving Averages
    3. IPC’s plant estimates weekly demand for its many materials held in inventory. One such
    part, the CTR 5922, is being studied. The most recent 12 weeks of demand for the CTR 5922
    are:

Week       Demand      Week       Demand      Week        Demand     Week       Demand
           (units)                (units)                 (units)               (units)
1          169         4          171         7           213        10         158
2          227         5          163         8           175        11         188
3          176         6          157         9           178        12         169

    a. Use the moving average method of short-range forecasting with an averaging period of
       three weeks to develop a forecast of the demand for the CRT 5922 component in week
       13.
    b. If a smoothing constant of 0.25 is used and the exponential smoothing forecast for week
       11 was 170.76 units, what is the exponential smoothing forecast for week 13?
    c. Which forecasting method is preferred- the AP=3 moving average method or the α=0.25
       exponential smoothing method? The criterion for choosing between the methods is mean
       absolute deviation (MAD) over the most recent nine weeks. Assume that the exponential
       smoothing forecast for week 3 is the same as the actual demand.




                                          Topic 12 - 18
4. The number of Texas tax auditors needed by the Internal Revenue Service varies from quarter
to quarter. The past 12 quarters of data are shown below:

Year                           Quarter                        Auditors
1                              1                              132
                               2                              139
                               3                              136
                               4                              140
2                              1                              134
                               2                              142
                               3                              140
                               4                              139
3                              1                              135
                               2                              137
                               3                              139
                               4                              141

    a. Use moving averages to forecast the number of auditors needed next quarter if AP=2,
       AP=4, and AP=6.
    b. Which of these forecasts exhibit the best forecast accuracy over the past six quarters of
       historical data based on mean absolute deviation?


                 Demand Forecast- Exponential Smoothing
5. A toy company buys large quantities of plastic pellets for use in the manufacture of its
products. Production manager Josh Kang wants to develop a forecasting system for plastic pellet
prices. The price per pound of plastic pellets has varied shown:

Month                  Plastic Pellets         Month                  Plastic Pellets
                       Price/ Pound                                   Price/ Pound
1                      $ 0.39                  9                      0.35
2                      0.41                    10                     0.38
3                      0.45                    11                     0.39
4                      0.44                    12                     0.43
5                      0.40                    13                     0.37
6                      0.41                    14                     0.38
7                      0.38                    15                     0.36
8                      0.36                    16                     0.39

    a. Use exponential smoothing to forecast monthly plastic pellet prices. Compute what the
       forecasts would have been for all the months of historical data for α=0.1, α=0.3, and
       α=0.5 if assumed forecast for all α’s in the first month is $0.39.
    b. Which alpha value results in the last mean absolute deviation for Months 7-16?
    c. Use the best alpha value from part b to compute the forecasted plastic pellets price for
       month 17.


                                          Topic 12 - 19
                     Demand Forecast- Seasonal Forecast
6. A computer manufacturer wants to develop next year’s quarterly forecasts of sales revenues
for its line of personal computers. The company believes that the most recent eight quarters of
sales should be representative for next year’s sales:

Year           Quarter         Sales          Year            Quarter         Sales
                               (millions of                                   (millions of
                               dollars)                                       dollars)
1              1               9.2            2               1               10.3
1              2               5.4            2               2               6.4
1              3               4.3            2               3               5.4
1              4               14.1           2               4               16.0

Use seasonalized time series regression analysis to develop a forecast of next year’s quarterly
sales revenue for the line of personal computers.




                                          Topic 12 - 20
                        Techniques to Support Better Forecasting
         Company leaders at a manufacturer of industrial fuel pumps decided to discontinue some of
their ~ products. They dismissed vendors and cancelled all existing agreements related to the phased-
out pumps. Not long after, a boom in small, diesel electricity generators caused the pumps to be back in
high demand. The manufacturer thus had to start from scratch in order to revive a product that
otherwise could have been a cash cow.

         Every operations management professional has a forecasting nightmare of his or her own
resembling this one. Market conditions change at an incredible pace, and the price and availability of
raw material often varies at a speed defying logic. A food shortage, a tsunami, terrorism, rising fuel
prices, and countless other events have the potential to completely alter the basic rules of business.
Even the smallest incidents can have a dramatic impact. Given these dynamic conditions, how is an
accurate forecast achieved?
         Normally associated with numbers and formulas, forecasting is a kind of magic box that uses
certain inputs to determine the products that the market expects. There are more than 100 different
quantitative forecasting methods available, which all begin with the simple assumption that the past
will repeat in the future.
   Time-series methods extrapolate existing trends and include seasonal and cyclical indices, if
necessary. They also assume that the trend, season, or cycle will have a predictable and similar effect
every time. Complex econometric and regression-based methods try to isolate the individual
components causing demand in order to create a forecasting model. But these models have an inherent
limitation in the number of factors they use because it is impossible to include all the key data. Moreover,
something that seems insignificant today all of a sudden may become a key driver.
   There is no doubt that forecasting is critically important; however, relying solely on these
numerical forecasting methods to drive business would be an exercise in corporate hara-kiri.

 The foundations:

    Over the years, forecasting has evolved from a set of principles to a set of tools. While principles
are generic and do not change, tools are prone to inaccuracy and, hence, create a negative impression.
    Something obviously has to be done. One idea is to create better forecasting models that are
monitored and improved in real time. Advanced software tools have at least provided the ability to
change formulas globally in the fraction of a second. But, given the nature of abrupt changes, creating
an accurate model seems even more difficult. Investing efforts and resources into seemingly better
tools also would lead to destruction-albeit more slowly. The point is: When the path is wrong, a
change in walking shoes does not improve anything.
    Operations management professionals must alter the way forecasting is used, not the way it is done.
Better processes are required, and they must be resistant to inaccurate forecasts. The ideal situation would
be to have processes that are responsive to customer needs and do not require a forecast to function.
Following are some ways to begin.
    Recognize the change. Operations management professionals have to appreciate the
variability in a situation. They must bear in mind that sporadic events can occur and change their
businesses-and these people must understand that such events cannot be forecast.
    Institute flexibility. Manufacturing facilities, vendors, product design, and other key elements
should be developed with an eye for flexibility. Vendors must be able to respond to a change of scale
and scope without a major impact on their pricing. Plants should be able to produce multiple products.
Employees and managers must have the necessary skill and-more importantly-the right attitude to be able
to change their responsibilities according to current requirements. The benefits of instituting flexibility
almost always outweigh the costs.



                                              Topic 12 - 21
  Standardize products. Standardization is not the antonym of flexibility. It essentially means that
the changes to a base product must be incremental. Customers expect multiple, unique products, and a
company must be able to change the volume of these products as necessary.
  Segment products. Not all products change regularly. Unless a firm is in a very dynamic market,
there always will be products that are more stable than others. Putting these items aside and setting a
standard schedule for them reduces complexity. For example, if 70 percent of output can be
predicted with 95 percent accuracy, that means 25 percent of the errors in the remaining 30 percent
of the output would have a less significant impact.
    Postpone. This is more important-and more possible than ever before. Technology enables us to do
final assembly much closer to the customer and without any major increase in cost. Forecasting should
be moved upstream, as well. If company leaders at the fuel pump manufacturer from the beginning of
this article had kept basic product design constant and limited variation to modules, then the firm could
have absorbed market fluctuations much more easily.
    Make "small" work. The shorter the term, the better the forecast. As such, every aspect of business
must be rewired to enable small batch sizes. Smaller vehicles can be used to transport material more
frequently. Delivering or picking up goods from multiple sites and then returning to the original
location with them should be considered. Rather than waiting for a batch of goods, work must be able
to proceed on individual pieces; and, wherever small batch sizes are not economical, the processes
should be changed to make them work. Keep in mind: Making small work means that goods will reach
the market much faster. So, if the cumulative lead time is 30 days, the market has to be approximated
by at least 30 days. If small batches can reduce this time to 10 days, the market has to be pre-empted by
only 10 days.
    Increase the speed of information transfer. Accurate and fast information is the lifeline of a
strong business. Data on actual customer usage should be tracked whenever possible, and technology
tools should be used to enable information entry once, at the point of occurrence. Operating with high
speed and accuracy ensures the forecasting model will work on current information. Plus, it becomes
much more likely that employees can react accurately.
    Rework business rules. Given that the stakes of the game have changed, the rules need to be
altered, as well. For example, an after-sales service firm stored fast-moving components at all its
regional depots. Company leaders decided to shift all the slow-moving spare parts to a central
warehouse. Then, they would be flown to the different regions on an as-needed basis. Instead of
forecasting all the components at each location, a few components now are stored and forecast
only at the central location, significantly increasing accuracy.
    Monitor international politics. Buying and selling internationally involves a lot of risk.
Politics, social issues, and economics have to be monitored to create scenarios of possible impact.
Forecasting never can predict such events, and operations management professionals must
recognize this and buffer the forecasts when necessary.
    Elevate forecasting. Forecasting cannot be merely an operational tool. Senior managers must
recognize the limitations of the process and lead the necessary changes in design, manufacturing,
and distribution. Results of forecasting processes must be monitored at the highest level in order to
develop other processes and reduce dependence on forecasts.
    Recognize the goal. Firms are not in business to make accurate forecasts; they're in business to
make more money. Forecasting is merely a tool that helps along the way.
    Forecasting is bound to be inaccurate, but it is nonetheless necessary for a firm's survival-especially
when an organization has multiplying product variants in a dynamic global marketplace. Software
solutions offer some assistance, but the current tools never will be 100 percent effective. Thus,
operations management professionals must refocus their efforts to create processes that can deliver
standard output with forecasts of limited accuracy.

Question: Summarize what you have learned from reading this article.



                                              Topic 12 - 22
                             OUTLOOK-Warm and Sunny

                                                              By Nada R. Sanders, PhD.

Getting the best forecast by combining judgmental and statistical methods
         Accurate forecasting always has been a critical organizational capability for effective
business planning. Good forecasts are essential for identifying and new market opportunities,
anticipating future demands, effectively scheduling production, and reducing inventories.
         Over the past few years, however, the role of forecasting has become especially
significant due to more competitive market pressures. Information technology has enabled
forecasts to drive entire supply chains and enterprise resources planning systems.
Simultaneously, global competition has created an environment characterized by uncertainty,
rapidly shifting markets, and compressed cycle times. Customers are demanding increasingly
shorter response times, improved quality, and greater product choice. The result has been a
sharp rise in forecasting's complexity and historical data that are often of limited value for
predicting the future.
         Relying on statistical forecasts alone can be ineffective in this highly complex
environment. Consider the case of Nike's $400 million failure in 2000 with demand
forecasting software. According to the July 15, 2003, issue of CIO magazine, nine months
after implementing a much publicized i2 system, Nike leaders acknowledged that they would
be taking a major inventory write-off due to inaccurate forecasts from the automated system.
Nike had entirely too much inventory of slow-moving items and a major shortage of popular
sellers.
         The problem, as it turned out, was that Nike executives relied exclusively on
automated forecasts without any judgmental checks, and the automated forecasts simply were
not accurate enough. Unfortunately, Nike's experience with automated statistical forecasts is
not an isolated case.
         When making forecasts, managers can choose from either judgmental forecasting
methods, which are based on opinions, or statistical forecasting methods, based on
mathematical modeling. Each category has unique strengths and weaknesses.(See Figure 1.)
The best forecasting approach is one that leverages the strengths of both methods.
Increasingly, this is something that managers find to be effective, and it is supported by
numerous research studies. However, combining judgmental and statistical forecasting
requires following well-established rules.

Benefits and drawbacks

       Judgmental and statistical forecasting methods each bring valuable information to
the forecasting process. Practitioners often have up-to-date knowledge of changes and
events occurring in their environment that can influence the forecast. This information
often is last-minute and cannot easily be incorporated into the statistical forecast. While
causal models, such as regression, can be developed to capture this human judgment, such
refinements are impractical for making inventory replenishment decisions when there are
thousands of different items to be controlled.



                                         Topic 12 - 23
         Judgmental forecasts also have the advantage of being able to incorporate "soft" or
"inside" information that can be helpful predictor is in changing environments. These types
of information might include a rumor about a competitor launching a new product or an
impending labor strike.
        Judgmental forecasts, however, often are inaccurate due to limitations in human
cognitive ability. People naturally have a limited attention span and can process only a
restricted amount of information at a time. Similarly, research has documented that
judgmental forecasts are influenced by short-term memory and the inability of forecasters
to understand causal relationships. In addition, judgmental forecasts can be biased because
they are subjective. Biases include optimism, wishful thinking, lack of consistency, and
political manipulation. For example, consider the manager who is optimistic one day after a
large sale or highly pessimistic another day following a sales slump. Such events often lead
people to inadvertently bias their forecasts, which can result in degradation of accuracy.
        Unlike judgmental-also called managerial-forecasts, statistical forecasts are based on ,
mathematical principles and are typically generated by any one of the many software
packages available. Statistical forecasts are consistent, objective, and unbiased, always
producing the same forecast for the same data set. Also, statistical forecasts can process large
amounts of data at one time. This is particularly effective for generating forecasts for a large
number of stock keeping units (SKUs), when managerial involvement would be time-
consuming and costly.
        However, statistical models are only as good as the data upon which they are based.
When changes occur in the data that are not incorporated in the model, the generated
forecasts can't be accurate. Statistical forecasts also are based on historical data and are
ineffective when market conditions change, such as a new competitor entering the
marketplace or a snow storm delaying a shipment.




                                         Topic 12 - 24
A winning combination
        More and more, successful forecasting uses composite methodologies. Judgmental
and statistical forecasts can '     be combined in different ways in order to take advan-
tage of their individual strengths. One way is to take a mathematical average of both
methods to generate a final forecast.
        By far the most popular method in practice, however, is managerial adjustment of
statistical forecasts. This method requires managers to adjust the statistical fore cast up or
down based on their opinions, often called a "managerial override."
Managerially adjusting statistical forecasts often can improve forecast accuracy by
including information not available to the statistical model. However, if performed
incorrectly, adjustments can cause inaccuracy due to inherent human biases. Managers
should follow established rules for effectively adjusting statistical forecasts.

 Rule following
       In order to gain the best from both forecasting methods, it is critical to understand
rules on how and when statistical forecasts should be adjusted.

Only practitioners with domain knowledge should adjust statistical forecasts.
        Domain knowledge is information managers gain through on-the-job experience.
Becoming familiar with their environment, managers learn many cause-and-effect
relationships and environmental cues. Research studies repeatedly have found that
judgmental adjustment is more likely to improve accuracy when the adjustment is based on
domain knowledge.

        Managers with domain knowledge understand which cues in the environment are
significant and which are unimportant. Specific information available in the forecast
environment is called contextual information. Examples of contextual information include
a price increase, an impending strike, or new policies that may affect the forecast. Domain
knowledge enables managers to evaluate the importance of specific contextual information.
If this information is not contained in the statistical forecasting model, the manager can
incorporate the information by adjusting the statistical forecasts.


Adjust statistical forecasts when there are known changes in the environment.
        To be useful, judgment should incorporate information that is not captured by the
statistical forecast. For example, statistical forecasts should account for discon tinuities or
pattern changes in the data. Judgmental adjustment of the statistical forecast can improve
accuracy if the forecaster can identify these patterns in the data and incorporate this
information in the adjustment.
        The adjustment should compensate for specific events not captured by the statistical
model or not yet observed in the time series. These events can include machines down due
to repair, advertising campaigns, and labor strikes.

Structure the judgmental adjustment process.

One of the disadvantages of human judgment is the limited ability to consider and process
large amounts of information. Structuring available data through a procedure or decision -


                                        Topic 12 - 25
support system is helpful in forecasting and decision making. Judgmental adjustment has
been shown to lead to greater improvements in accuracy if the process is structured, as
opposed to an ad hoc adjustment. Structuring can be complex, such as that generated by a
computer-aided decision support system, or as simple as using a paper and pencil. The
important thing is that it is used.

Document all judgmental adjustments made and measure forecast accuracy
       Like all forecasts, judgmentally adjusted forecasts need to be measured using forecast
accuracy measures. In addition, records should be kept of all adjustments made and the
reasons for the adjustment. Over time, managers can evaluate what types of adjustments led
to the greatest improvement and which adjustments were ineffective - enabling future
improvement. This process can have a powerful effect on improving forecast accuracy.
However, to succeed, accurate records must be kept over time. Also, an important aspect of
accuracy is ensuring that the numbers and the arithmetic are correct.

Current conditions

        The rules provided raise a number of implications for managers. First, managers
must decide what type of forecasting procedure to use and whether to adjust a statistical
forecast. This depends on the amount and type of available information. A sufficient
amount of quantifiable data need to exist in order to apply a statistical forecast. Without
such data, forecasts may need to be based solely on judgment. This may be the case for
forecasting demand for a brand-new product or for long range strategic forecasting
decisions where information is hard to quantify.
        When good, quantifiable, historical data are available, reliance should be placed on
statistical forecasts. Only when forecasters know of events and information that influence
the forecast should judgment be used to adjust it.        Judgmental       adjustments     only
should be made when there is contextual information in the presence of domain knowledge,
forecasting process structure, and adequate feedback. For example, in studies where
managers’ judgmental forecasts significantly outperformed statistical forecasts, the
managers in question had been responsible for generating forecasts over a period of years.
They kept records of their forecasts and received regular feedback. Repetition and good
feedback over a long period of time enabled these managers to develop expertise; thereby,
giving them the tools to create more effective forecasts.
        Many forecasting situations do not provide this advantage. For example, in
inventory management, managers may be responsible for hundreds of different SKUs.
There often are changes in product mix, customer mix, and markets. In such environme nts,
achieving high levels of knowledge and familiarity may be impossible.
        Another issue to consider is the cost of managerial involvement. Though there are
benefits with judgmental adjustment, are these benefits significant in terms of cost? The
cost of managerial involvement can be high when making forecast adjustments and
developing domain knowledge. The benefits of a more refined forecast need to make financial
sense. Managers should be selective as to which forecasts they adjust. In addition, focusing on
a smaller number of forecasts will enable a manager to achieve greater accuracy.

Questions: Based on this article, why the combining forecasting approach may give
you the best results?


                                        Topic 12 - 26
                        Review Questions for Topics 11 and 12:

       Be prepared to discuss the following cases:
       (a) Common Sense for Supply Chain (Supplement, p.11-21 to 11-22)
       (b) Techniques to Support Better Forecasting (Supplement, p.12-21 to 12-22)
       (c) Outlook – Warm and Sunny (Supplement, p.12-23 to 12-26)
       (d) The Yankee Fork and Hoe Company (Text, p.498-499)


(1) Differentiate between qualitative and quantitative forecasting techniques. Discuss - under
   what conditions these two techniques will be preferred to use in practice.
(2) Be prepared to compute a forecast using a simple moving average, a weighted moving
   average, and exponential smoothing.
(3) What is the “principle of forecasting”? Brief explain.
(4) Why are there different considerations in selecting a forecasting model regarding three
   different forecasting horizons: short-term, medium-term, and long-term.
(5) How is the mean absolute deviation (MAD) of a forecast series computed? Why is it
   computed?
(7) What is the impact of using a large (or a small value of)  in computing an exponentially
   weighted forecast?
(8) What is the impact of using a large number of period (or a small number of) (n) in
   computing a simple (or weighted) moving average forecast?
(9) Generally, how are seasonal effects included in exponential smoothing?
10) Explain the differences between "bias", MAD, “tracking signal” and "random errors" in
   forecasting.
11) What is the primary difference between a causal model and a time series model for
   forecasting?
12) Differentiate between the projection type of forecasting and the predictive type of
   forecasting.
13) Explain the purpose of the tracking signal used to monitor the performance of a forecasting
   system.




                                          Topic 12 - 27
Use <POM-Window> for Demand Forecasting Problems
Step-1: Start <POM-Windows> - click <Forecasting> from the [Module] menu.

Step-2: Under <File>, Click <New> - you will see a downward menu – listing three available
   forecasting methods:       1. Time Series Analysis
                                     2. Least Squares – Simple and Multiple Regression
                                     3. Regression Projector.
   Now, selecting the method for the problem to be solved now.
   (Note: The 3rd choice is not covered in this class. Choose either (1) or (2) depending on the
   problem. More specifically, all Moving Average methods, Exponential Smoothing, Seasonal
   Decomposition, etc. are in the <Time Series> group, and Simple or Multiple regression
   models should be self-indicative from the given information.)

Step-3: Now you well see data input screen <Creating a new data>: On this screen,
        Type in: <Problem Name> in the [Title] box (or any name you like)
                   < X > in the [No. of Past Periods] box (here is the given number of past
                 period demand information, i.e., 12, or 10.
                         Selecting a name for each row data [like, 1, or A, or a, ….]
       Then, Click on [OK], you will see data input matrix screen.

Step-4: On <Data input matrix screen>:

    For all <Time Series> models, there will be one column data [y – demand] to be inputted:
       Under [demand - y], typing all past demand data into this column (e.g., 27, 35, …)
       Click [Method] (left – top), select the model you want to use (e.g., moving average.)
        - For [Moving Average], you need now input [# Period to Average] (e.g., 3, or 4)
        - For [Weighted Moving Average], you need input both [# Period to Average] and the
          actual weights to be used (the total of all weights must equal to 100%).
        - For [Exponential Smoothing], you need input [alpha value] and the forecasted value
          for the first period [under Forecast column] – if no specific information provided,
          using the given actual demand number of the first Period as the forecasted value for
          the first period. (That is, assuming for Period-1, your forecasting = actual demand)
        - For [Seasonal Decomposition], you need input [# Seasons] (e.g., 4, or 124) and
          selecting one of [Basis for Smoothing] – selecting [Centered Moving Average] method
          for your homework.

   For <Simple and Multiple Regression> models, there will be at least two column data (for
   Simple Regression) – including [depend - Y ] and [X1] - or more column data (for Multiple
   Regression) to be inputted:
   - Under [depend - Y], typing all past demand data into this column – this is what to be
     forecasted from [Independent – X].
   - Under [X1 – independent variables], typing all past data of this independent variable into
     this column – this is the variable upon which the forecasted values of the demand will be
     made.



                                         Topic 12 - 28
Step-5: Now, click on <Solve> - you will see <Output> screen.
        On <Output> screen, you will see three sections –
        1) [Forecast Results] – including [Error Measures] and [Forecast for next Period] – you
           need to print out this section for your homework.
           For [Regression Models] – this section will include: <Error Measures – like Bias and
           MAD>, <Regression Line – the regression equation, both (a) and (b) for [Y = a +
           bX], and [Statistics – both < r > and < R2> are provided].
        2) [Details and error analysis] – ignoring this section – never print out this section.
        3) [Graph] –you can print out this section to see the graphical results and hand-in for
           your homework.

       On this screen, click [Results] click on [Print], you will see your Printer – starting to
       work.


Step-6: For certain problems, you need to write down (or write out) the required answers from
        your printouts – based on your understanding and interpretation about your printouts.




                                          Topic 12 - 29

				
DOCUMENT INFO
Description: Hercules Inc. Historical Stock Prices document sample