Principles of Forecasting

1
 Content
 Identify principles of forecasting.
 Explain the forecasting process steps
 Identify types of forecasting methods and their characteristics.
 Describe time series models and causal models.
 Generate forecasts for data with different patterns, such as level, trend,
and seasonality and cycles.
 Describe causal modeling using linear regression.

 Compute forecast accuracy.

 Explain the factors that should be considered when selecting a forecasting

model.

2
 What is Forecasting?
 Process of predicting a future event
 Underlying basis of all business decisions
 Production



Inventory
Personnel
??
 Facilities

3
 Why is forecasting important?
 Forecasting is one of the most important business functions because all
other business decisions are based on a forecast of the future.

 Decisions such as which markets to pursue, which products to produce,

how much inventory to carry, and how many people to hire all require a
forecast.

 Poor forecasting results in incorrect business decisions and leaves the

company unprepared to meet future demands.

 The consequences can be very costly in terms of lost sales and can even
force a company out of business.

4
 Forecast
 Forecasts are so important that companies are investing billions of dollars
in technologies that can help them better plan for the future.
For example: the ice-cream giant Ben & Jerry’s has invested in business
intelligence software that tracks the life of each pint of ice cream, from
ingredients to sale. Each pint is stamped with a tracking number that is
stored in an Oracle database. Then the company uses the information to
track trends, problems, and new business opportunities. They can track
such things as seeing if the ice-cream flavor Chocolate Chip Cookie Dough
is gaining on Cherry Garcia for the top sales spot, product sales by location,
and rates of change. This information is then used to more accurately
forecast product sales.
 Numerous other companies, such as Procter & Gamble, General Electric,
Lands’ End, Sears, and Burger King, are investing in the same type of
software in order to improve forecast accuracy.

5
 Forecasting Time Horizons
 Short-range forecast
 Up to 1 year, generally less than 3 months
 Purchasing, job scheduling, workforce levels, job assignments,
production levels
 Medium-range forecast
 3 months to 3 years
 Sales and production planning, budgeting
 Long-range forecast
 3+ years
 New product planning, facility location, research and
development

6
 Principles of Forecasting
 There are many types of forecasting models. They differ in their
degree of complexity, the amount of data they use, and the way they
generate the forecast.

However, some features are common to all forecasting models. They

include the following:
1. Forecasts are rarely perfect

2. Forecasts are more accurate for grouped data than for individual items

3. Forecast are more accurate for shorter than longer time periods

7
 Principles of Forecasting
1. Forecasts are rarely perfect
Forecasting the future involves uncertainty. Therefore, it is almost

impossible to make a perfect prediction. Forecasters know that they have

to live with a certain amount of error, which is the difference between
what is forecast and what actually happens.

The goal of forecasting is to generate good forecasts on the

average over time and to keep forecast errors as low as possible.

8
 Principles of Forecasting
2. Forecasts are more accurate for grouped data than for
individual items
When items are grouped together, their individual high and low values

can cancel each other out. The data for a group of items can be stable
even when individual items in the group are very unstable.
Consequently, one can obtain a higher degree of accuracy when
forecasting for a group of items rather than for individual items.

For example, you cannot expect the same degree of accuracy if you are
forecasting sales of long-sleeved hunter green polo shirts that you can
expect when forecasting sales of all polo shirts.

9
 Principles of Forecasting
3. Forecast are more accurate for shorter than longer time
periods
The shorter the time horizon of the forecast, the lower the degree of

uncertainty. Data do not change very much in the short run.

As the time horizon increases, however, there is a much greater

likelihood that changes in established patterns and relationships will
occur. Because of that, forecasters cannot expect the same degree of
forecast accuracy for a long-range forecast as for a short-range forecast.

For example, it is much harder to predict sales of a product two years

from now than to predict sales two weeks from now.

10
 Forecasting Steps
Regardless of what forecasting method is used, there are some
basic steps that should be followed when making a forecast:

Decide what needs to be forecast

1.
 Level of detail, units of analysis & time horizon required

Evaluate and analyze appropriate data

2.
 Identify needed data & whether it’s available

Select and test the forecasting model

3.
 Cost, ease of use & accuracy
4.Generate the forecast
5.Monitor forecast accuracy over time
11
12
 Types of Forecasting Models
Forecasting methods can be classified into 2 groups: qualitative and
quantitative.
1) Qualitative methods – judgemental methods
 Forecasts generated subjectively by the forecaster

When you decide, based on your intuition, that a particular team is going
to win a baseball game, you are making a qualitative forecast. Because
qualitative methods are made by people, they are often biased.
These biases can be related to personal motivation (“They are going to set

my budget based on my forecast, so I’d better predict high.”), mood (“I

feel lucky today!”), or conviction (“That pitcher can strike anybody out!”).
 Educated guesses

13
 Types of Forecasting Models
2) Quantitative methods – based on mathematical modeling:
 Forecasts generated through mathematical modeling

The same model will generate the exact same forecast from the same set
of data every time.

These methods are also objective. They do not suffer from the biases found
in qualitative forecasting. Finally, these methods can consider a lot of
information at one time.

Because people have limited information-processing abilities and can easily

experience information overload, they cannot compete with mathematically
generated forecasts in this area.

14
 Types of Forecasting Methods
Summary of types of forecasting methods:

15
 Qualitative Methods
1. Executive opinion: Forecasting method in which a group of
managers collectively develop a forecast.

2. Market research: Approach to forecasting that relies on surveys

and interviews to determine customer preferences.

3. Delphi method: Approach to forecasting in which a forecast is

the
product of a consensus among a group of experts.

16
 Although managers can bring good insights to the forecast, this method has a
number of disadvantages. Often the opinion of one person can dominate the
forecast if that person has more power than the other members of the group
or is very domineering. Think about times when you were part of a group for a
course or for your job. Chances are that you experienced situations in which
one person’s views dominated.
 Market research can be a good determinant of customer preferences.
However, it has a number of shortcomings. One of the most common has to
do with how the survey questions are designed.
 For example, a market research firm may call and ask you to identify which of
the following is your favorite hobby: gardening, working on cars, cooking, or
playing sports. But maybe none of these is your favorite because you prefer
playing the piano or fishing, and these options are not included. This question
is poorly designed because it forces you to pick a category that you really
don’t fit in, which can lead to misinterpretation of the survey results.

17
 Delphi method
Iterative group process, continues
until consensus is reached

3 types of participants:
 Decision makers
 Staff
 Respondents

18
Delphi method

Although it takes a large amount of time,

it has been shown to be an excellent
method for forecasting long-range product
demand, technological change, and
scientific advances in medicine. For
example, if you
wished to predict the timing for an AIDS
vaccine or a cure for cancer, you would
probably use this technique. 19
Qualitative Methods

20
 Quantitative Methods
 Quantitative methods are different from qualitative ones because they are
based on mathematics.
 A time series is a sequence of data points, typically consisting of successive
measurements made over a time interval.
1. Time Series Models
 Assumes information needed to generate a forecast is contained in a time series

of data
 Assumes the future will follow same patterns as the past

 For example, if you were forecasting quarterly corporate sales and

had collected 5 years of quarterly sales data, you would have a time
series.
 Time series analysis assumes that we can generate a forecast
based on patterns in the data. As a forecaster, you would look for patterns such
as trend, seasonality, and cycle and use that information to generate a
forecast.
21
 Quantitative Methods
 Quantitative methods are different from qualitative ones because they
are based on mathematics.
2. Causal Models or Associative Models
 Explores cause-and-effect relationships

 Uses leading indicators to predict the future

They assume that the variable we wish to forecast is somehow related to other
variables in the environment.
The forecaster’s job is to discover how these variables are related in
mathematical terms and use that information to forecast the future

22
Time Series Components

Trend Cyclical

Seasonal Random

4 - 23
 Time Series Models
 There are 4 time series models
 Level (long-term average) – data fluctuates around a constant mean
 Trend – data exhibits an increasing or decreasing pattern
 Seasonality – any pattern that regularly repeats itself and is of a
constant length (weekly, monthly or yearly)
 Cycle – patterns created by economic fluctuations
Random variation is unexplained variation that cannot be predicted. So if we
look at any time series, we can see that it is composed of the following:

24
 Trend Component
 Persistent, overall upward or downward pattern

 Changes due to population, technology, age, culture, etc.

 Typically several years duration

 Seasonal Component
 Regular pattern of up and down fluctuations
 Due to weather, customs, etc.
 Occurs within a single year

Number of
Period Length Seasons
Week Day 7
Month Week 4-4.5
Month Day 28-31
Year Quarter 4
Year Month 12
Year Week 52
 Cyclical Component
 Repeating up and down movements

 Affected by business cycle, political, and economic factors

 Multiple years duration

0 5 10 15 20
 Random Component

 Erratic, unsystematic, ‘residual’ fluctuations

 Due to random variation or unforeseen events

 Short duration
and nonrepeating

M T W T F
29
 Causal (associate) models
 Causal models, sometimes called associative models, use a very different
logic to generate a forecast.
 They assume that the variable we wish to forecast is somehow related to
other variables in the environment. The forecaster’s job is to discover how
these variables are related in mathematical terms and use that information to
forecast the future.
 Time series models are generally easier to use than causal models. Causal
models can be very complex, especially if they consider relationships among
many variables.

However, time series models can often be just as accurate and have the
advantage of simplicity.
 They are easy to use and can generate a forecast more quickly than causal
models, which require model building.

30
31
 Time Series Models
 Naïve: Good for level patterns

Ft 1  At
where Ft+1 = Forecast for period t + 1,
A t = Actual value for period t.

 Assumes demand in next period is the same as demand in most

recent period
 If January sales were 68, then February sales will be 68
 Sometimes cost effective and efficient
 Can be good starting point

32
Naïve Example

33
 Time Series Models
 Simple Mean or Average
One of the simplest averaging models is the simple mean or
average. Here the forecast is made by simply taking an average
of all data:

34
Simple Mean or Average Example

35
 Time Series Models
Moving Average:
where n = number of periods
The average value over a set time period

(e.g.: the last four weeks)

Each new forecast drops the oldest data point & adds a new
observation

 More responsive to a trend but still lags behind actual data

Ft 1   A t / n

36
 Moving Average Example
 Sales forecasts for a product are made using a three-period moving
average. Given the following sales figures for January, February,
and March, make a forecast for April, May and June.

Before You Begin: Remember that to use a three-period moving average, you have to
compute the average of the latest three observations.
As new data become available, you drop off the oldest data,
always averaging the latest three observations.

37
38
Comparing the three-period and five-period moving average forecasts, we can see
that the three-period moving average forecasts are more responsive to the period-
to period changes in the actual data—they follow the data more closely. The reason
is that the smaller the number of observations in the moving average, the more
responsive the forecast is to changes in demand. However, the forecast is also more
subject to the random changes in the data. If the data contain a lot of randomness,
high responsiveness could lead to greater errors. On the other hand, the larger the
number of observations in the moving average, the less responsive the forecast is
to changes in the demand, but also to the randomness. These forecasts are more
stable. One is not better than the other. Selection of the number of observations in
the moving average should be based on the characteristics of the data. 39
 Just like the mean, the moving average is good only for a level
pattern.

 The problem is that the forecasts are trailing behind the

actual data. We say that they are “lagging” the data.

 This is what happens when you apply a model that is good only
for a level pattern to data that have a trend. You will not obtain
a good forecast

40
 Time Series Models
 Weighted Moving Average: Ft 1   C t A t

A time-series method in which each historical demand in the average

can have its own weight;

the sum of the weights equals 1.0

Differs from the simple moving average that weighs all periods
equally - more responsive to trends

41
Weighted Moving Average Example

42
 Time Series Models
Exponential Smoothing: Ft 1  αA t   1  α  Ft
Most frequently used time series method because of ease of use
and minimal amount of data needed
Need just three pieces of data to start:

 Last period’s forecast (F )
t

 Last periods actual value (At)

 Select value of smoothing coefficient,  ,between 0 and 1.0
If no last period forecast is available, average the last few periods or

use naive method

Higher values may place too much weight on last period’s random
variation

43
Exponential Smoothing Example

Ft 1  αA t   1  α  Ft
44
 Trend-adjusted exponential smoothing

 Exponential smoothing model that is suited to data that

exhibit a trend.

45
 smoothing

46
 Selecting 
 Selecting that depending on which value you select for , you can place more
weight on either the current period’s actual or the current period’s forecast.

 In this manner the forecast can depend more heavily either on what happened
most recently or on the current period’s forecast. Values of  that are low—say,
0.1 or 0.2—generate forecasts that are very stable because the model does not
place much weight on the current period’s actual demand.

 Values of that are high, such as 0.7 or 0.8, place a lot of weight on the current
period’s actual demand and can be influenced by random variations in the data.

 Thus, how is selected is very important in getting a good forecast.

47
 Linear Trend Line
 Linear trend line is a time series
technique that computes a
forecast with trend by drawing a
straight line through a set of
data.

 This approach is a version of the

linear regression technique,
covered later in this chapter, and
is useful for computing a forecast
when data display a clear trend
over time.

 The method is simple, easy to

use, and easy to understand.
48
49
 Forecasting Seasonality

 Recall that any regularly repeating pattern is a seasonal pattern.

 We are all familiar with quarterly and monthly seasonal patterns.

 Whether your university is on a quarter or semester plan, you can see that
enrollment varies between quarters or semesters in a fairly predictable way.

 For example, enrollment is usually much higher in the fall than in the
summer.

 Other examples of seasonality include sales of turkeys before Thanksgiving

or ham before Easter, sales of greeting cards, hotel registrations, and sales
of gardening tools.

50
 Forecasting Seasonality
 Step 1 Calculate the Average Demand for Each Quarter or “Season.” This is done by
dividing the total annual demand by 4 (the number of seasons per year).

 Step 2 Compute a Seasonal Index for Every Season of Every Year for Which You
Have Data. This is done by dividing the actual demand for each season by the average
demand per season (computed in Step 1).

 Step 3 Calculate the Average Seasonal Index for Each Season. For each season,
compute the average seasonal index by adding up the seasonal index values for that season
and dividing by the number of years.

 Step 4 Calculate the Average Demand per Season for Next Year. This could be done by
using any of the methods used to compute annual demand. Then we would divide that by the
number of seasons to determine the average demand per season for next year.

 Step 5 Multiply Next Year’s Average Seasonal Demand by Each Seasonal Index. This
will produce a forecast for each season of next year. 51
 Forecasting Seasonality Example
 U-R-Smart University wants to develop forecasts for next year’s
quarterly enrollment. It has collected quarterly enrollments for the
past two years. It has also forecast total annual enrollment for next
year to be 90,000 students. What is the forecast for each quarter of
next year?

52
Forecasting Seasonality Example

53
 Time Series: Linear
Regression
A time series technique that computes a forecast with trend by drawing
a straight line through a set of data using this formula:

Y = a + bx where

Y = forecast for period X

X = the number of time periods from X = 0
a = value of y at X = 0 (Y intercept)
b = slope of the line

54
Linear Regression
 Identify dependent (y) and
independent (x) variables

b
 XY   X  Y   Solve for the slope of the line
 X 2    X  X 

b
 XY  n XY

 X  nX 2 2

 Solve for the y intercept

a  Y  bX
 Develop your equation for the
trend line

Y=a + bX 55
sales are the dependent variable
(Y )

advertising dollars are the

independent variable (X).

We assume that there is a

relationship between these two
variables.

56
 Correlation coefficient:
Statistic that measures the direction and strength of the linear relationship
between two variables

We can compute the correlation coefficient and evaluate the strength of the linear relationship between sales
and advertising dollars. 57
 Measuring Forecast Error
 Forecasts are never perfect
 Need to know how much we should rely on our chosen forecasting
method

 Measuring forecast error:

E t  A t  Ft
 Note that over-forecasts = negative errors and
under-forecasts = positive errors

58
 Measuring Forecasting Accuracy
 Mean Absolute Deviation (MAD) MAD   actual  forecast
 measures the total error in a n
forecast without regard to sign

 Mean Square Error (MSE)

 Penalizes larger errors
  actual - forecast  2

MSE 
n

59
 Standard Parts Corporation is comparing the accuracy of two methods that it
has used to forecast sales of its popular valve. Forecasts using method A and
method B are shown against the actual values for January through May.
Which method provided better forecast accuracy?

Of the two methods, method A produced a lower MAD and a lower MSE, which means
60
that it provides better forecast accuracy.
 Selecting the Right Forecasting Model
1. The amount & type of available data
 Some methods require more data than others
2. Degree of accuracy required
 Increasing accuracy means more data
3. Length of forecast horizon
 Different models for 3 month vs. 10 years
4. Presence of data patterns
 Lagging will occur when a forecasting model meant for a level
pattern is applied with a trend

61
 Forecasting Software
 Spreadsheets
 Microsoft Excel, Quattro Pro, Lotus 1-2-3

 Limited statistical analysis of forecast data

 Statistical packages
 SPSS, SAS, NCSS, Minitab

 Forecasting plus statistical and graphics

 Specialty forecasting packages

 Forecast Master, Forecast Pro, Autobox, SCA

62
 Chapter Highlights
 Three basic principles of forecasting are:
 forecasts are rarely perfect;
 forecasts are more accurate for groups or families of items rather than for
individual items; and
 forecasts are more accurate for shorter than longer time horizons.
 The forecasting process involves five steps:
 decide what to forecast;

 evaluate and analyze appropriate data;

 select and test a forecasting model;

 generate the forecast; and

 monitor forecast accuracy.

63
 Chapter Highlights
 Forecasting methods can be classified into two groups:
 qualitative and quantitative.
 Qualitative forecasting methods generate a forecast based on the subjective
opinion of the forecaster.
 Some examples of qualitative methods include executive opinion, market
research, and the Delphi method.
 Quantitative forecasting methods are based on mathematical modeling. They
can be divided into two categories: time series models and causal models.
 Time series models are based on the assumption that all the information
needed for forecasting is contained in the time series of data.
 Causal models assume that the variable being forecast is related to other
variables in the environment.

64
 Chapter Highlights
 There are four basic patterns of data:
 level, trend, seasonality, and cycles.
 In addition, data usually contain random variation.
 Some forecasting models that can be used to forecast the level of a
time series are naïve, simple mean, simple moving average, weighted
moving average, and exponential smoothing.
 Separate models are used to forecast trend, such as trend-adjusted
exponential smoothing. Forecasting seasonality requires a procedure in
which we compute a seasonal index, the percentage by which each
season is above or below the mean.

65
 Chapter Highlights
 A simple causal model is linear regression, in which a straight-line
relationship is modeled between the variable we are forecasting and
another variable in the environment.
 The correlation coefficient is used to measure the strength of the linear
relationship between these two variables.
 Three useful measures of forecast accuracy are mean absolute deviation
(MAD), mean square error (MSE)
 There are four factors to consider when selecting a forecasting model:
 the amount and type of data available,
 the degree of accuracy required,
 the length of forecast horizon, and
 patterns present in the data.

66