Chapter 1 Questions Answers by wxp19831

VIEWS: 11 PAGES: 5

									C1-5 (Chalmers)

What is a forecast? What are the characteristics of a good forecast?

“A forecast is a probabilistic estimate or description of a future value or condition,” and should
include a “mean, range, and probability estimate of that range.” Since a forecast is a
probabilistic estimate, it should be a range of values, as opposed to a single value. For example,
sales next week are expected to be $8000, with an 80% probability that sales will be between
$6000 and $10,000. (p. 9-10)

C1-7 (Chalmers)

List and define the common time series patterns described in the book. Also, define typical
causes of these time series patterns.

Random Patterns:
      Definition: Random time series have no seasonality or trend, but “are the result of many
      influences that act independently to yield nonsystematic and non-repeating patterns about
      some average value. A series that is completely random, will “have a constant mean and
      no systematic patterns.”
      Typical causes: As stated in the definition, random patterns “are the result of many
      influences that act independently.”
                                Random
    20

    15

  Y 10

     5

     0
         1             6               11              16
                                    Tim e



Trend Patterns:
       Definition: “A general increase or decrease in a time series that lasts approximately seven
       of more periods.” A trend can be linear, logarithmic, exponential, or some other
       nonlinear function.
       Typical causes: Population growth, early stages of the product life cycle, continuous
       economic growth etc.
                                Trend
      25

      20

      15
  Y
      10

      5

      0
           1          6               11              16
                                   Tim e



Seasonal Patterns:
      Definition: “Results from events that are periodic and recurrent.” Seasonal patterns need
      not be annual, they could be weekly, daily or even hourly
      Typical causes: Annual weather changes (seasons), holidays, promotions, pay periods
      (monthly social security checks) etc.


                               Seasonal
      20

      15

  Y 10

      5

      0
           1          6               11              16
                                   Tim e




Cyclical Patterns:
       Definition: Non-seasonal, and unequal fluctuations in a series. Cyclical patterns are
       difficult to forecast because the period of the peaks and troughs are unknown.
       Typical causes: The actual cause of these fluctuations is unknown, but some possible
       explanations include: population like cycles, product life cycles and long term weather
       conditions (droughts of floods damaging crops, leading to an economic down term).

Autocorrelated Patterns:
      Definition: “The value of a series in one time period is related to the value of itself in
      previous periods.” That is, the most recent actual value is the best forecast of the next
      value. Random walk series are highly autocorrelated.
      Typical causes: Any series where momentum is significant factor. The stock market is
      usually a random walk pattern. Causes may include customer preference or brand
      loyalty, that generate a slow change in the demand for an item.
      (p. 13-20)
C1-8 (Chalmers)

What is an outliner and why are they so important? How do outliers relate to planned and
unplanned events and interventions?

Outlier Defined: Atypical data values. Generally very large or very small values that for some
reason, are not typical. Any number of things can cause outliers. For example, power failures,
strikes, serious or atypical weather etc. properly identifying and removing outliers is extremely
important when forecasting. For example, Home Depot’s chainsaw sales following the big ice
storm, would be a significant outlier that should be removed.
Planned Events: Promotions are planned events, but if they are not properly documented, may
appear to be unexplained outliers. Properly documenting promotions (time and impact of the
promotion), not only helps to explain an outlier, but it also helps when determining the impact of
a similar promotion in the future.
Unplanned Events: Like planned events, unplanned events, such as a competitors promotion,
should also be properly documented.
Interventions: So, whether the intervention is a planned or unplanned event, it is important to be
able to track its impact.
(p. 19-20)

C1-10 (Chalmers)

Briefly explain the three general types of forecasting methods. Make up or relate to
examples that are different from those of this chapter.


Univariate:
       Defined: “use the past, internal patterns in data to forecast the future.” Future values are
       a function of past values. Essentially all of the models we have learned to this point:
       smoothing, exponential smoothing, decomposition, linear trend and nonlinear trend, Box-
       Jenkins etc. Univariate methods are generally the most cost effective for short to medium
       term forecasts.
       Example: a simple weighted moving average to forecast December’s value.
       September = 450, October = 600, November = 575
       Yt = .5Yt-3 + .3Yt-2 +.2Yt-1
       Yt = .5(450) + .3(600) +.2(575)
       December forecast = 520
Multivariate:
       Defined: Multivariate, or “causal methods, make projections of the future by modeling
       the relationship between a series and other series.” That is, the dependent variable is a
       function of the independent predictor variables. Multivariate methods are generally more
       costly than univariate methods, and are generally not as accurate for short to medium-
       term forecasts.
       Example: Home mortgage rates may be some function of the Fed Funds Rate, and long-
       term bond rates.
       Mortgage rates = f(Fed Funds Rate, long-term bond rate)
Qualitative:
       Defined: “Are based on the judgment and opinions of others concerning future trends,
       tastes, and technological change.” Better for very long term forecasts, where univariate
       and multivariate ineffective, or when there is insufficient data. Qualitative methods are
       often used to estimate demand for new products (for which there is no historical data).
       Example: Qualitative forecasting is used to predict what the Fed is likely to do with
       interest rates. Several economists are polled, and predictions can be made based on their
       opinions.
       Fed’s next interest rate move = ∑(economists projections)/n
       Forecast rate move = ¼ + 0 + ¼ + ½ + ¼ + 0 + ¼ + ¼ + ½ + ¼ / 10 = ¼
       (p. 21-23)

C1-11 (Chalmers)
Describe the scientific method and how it relates to forecasting.

Generally, the scientific method has four steps:
1. Observation and description of a phenomenon or group of phenomena.
2. Formulation of a hypothesis to explain the phenomena.
3. Use of the hypothesis to predict the existence of other phenomena, or to predict quantitatively
the results of new observations.
4. Performance of experimental tests of the predictions by several independent experimenters
and properly performed experiments.
(http://teacher.nsrl.rochester.edu/phy_labs/AppendixE/AppendixE.html)

“The scientific method is the process by which scientists, collectively and over time, endeavor to
construct an accurate (that is, reliable, consistent and non-arbitrary) representation of the world.
The scientific method attempts to minimize the influence of bias or prejudice in the experimenter
when testing a hypothesis or a theory.” The seven steps in scientific method of forecasting are
an extension of the four general steps.

C1-11 (Chalmers)
Describe the seven steps of the forecasting process discussed in this chapter.

Since the forecasting process is a systematic, step-by-step process, and to avoid deterioration of
the meaning through paraphrasing, it seemed logical to directly quote the steps from the book
(page 26-27).

I. Problem definition--There is a need to solve a problem or explain some phenomenon; that is,
there is a need to plan or forecast some future event, for example, a product's demand.

II. Information search--This is the process of collecting information about the behavior of the
system in which the problem or phenomenon resides (i.e., what influences the time series). For
example, to understand the behavior of the series we need past data about sales of the product,
out-of-stock conditions, prices, sales of competitors' products, advertising expenditures, out-of-
stock conditions of competitors, and the number of customers.
III. Hypothesis/theory/model formulation--On the basis of the information and observations
realized in step II, a hypothesis or hypothetical model is formulated to describe the important
factors that influence the problem or phenomenon. For example, it may be hypothesized that
demand is seasonal and trending or that demand is a function of price, advertising, number of
competitors, and their prices.

IV. Experimental design--Using facts gathered in steps I, II, and II/ and statistical/mathematical
tools, experiments are designed to test the hypotheses and theories (e.g., fit a model to all data
except last year's, then see how well the model does in forecasting last year through this year).
This might be as simple as selecting a model from several models or designing an experiment
where data is collected and divided into two groups. The first group (called in-sample data) is
used in constructing the model; the second group (called out-of-sample data) is used to validate
the model in a simulated forecasting environment. Using out-of-sample data is an effective way
to judge the effectiveness of a model or theory; most would argue that this is an essential step
when sufficient data exists.

V. Execute the experiment--The experiment is designed and executed, then the results are
measured and collected (e.g., fit the model to the data before last year and see how well it
forecast last year through this year). Use the model fitted to in-sample data to forecast the out-of
sample data.

VI. Results analysis--The results of the experiment are analyzed in order to accept or reject the
hypothesis or model (e.g., calculate appropriate error measures and perform statistical
significance tests). Statistical diagnostic measures are used to judge the validity of the parts of
the model and its forecasts. The model is accepted, modified, or rejected. Depending on the
results achieved, several iterations may be made before converging on a best model. Finally,
because most forecasts support ongoing processes of planning, the seventh step addresses
ongoing use of the theory or model.

VII. Ongoing maintenance and verifications - This is the process of ensuring that the model or
theory is still valid and effective (e.g., diagnostic tools called tracking signals and other statistics
can be monitored to ensure model validity).

								
To top