FORECASTING

Objectives of the Topic

When we are through with this Topic, Students should be able to:
 Describe the importance of forecasting.
 Explain various components of a time series.
 Choose an appropriate forecasting model.
 Perform regression analysis.
 Identify cause–effect relationships.
 Analyse and evaluate forecasting errors.
 Pool information for multiple forecasts
Introduction
Every business is created to serve particular needs of the society at some future time that are never
known with certainty. Forecasting is used to foretell various circumstances of that future as it relates
to the needs to be served. It predicts the existence of takers for its products in that f uture, which
starts a minute hence and proceeds into the far horizon, and every other circumstances that relates
to its capacity to continuously serve those needs. Most often, such circumstances are related and
dependent on a major factor. For example, it is from a forecast of future demands that materials,
human resource and financing needs can be determined. A forecast of sales is to foretell demand
volume even though the probabilities have never been formally studied. There is often some
empirical basis for estimating what is likely to happen in the future. Marketing models for predicting
sales deal with levels of uncertainty that make forecasts of demand volumes, market shares, and
revenues difficult, but not irrational. One of the best sources of information about the future is the
past. For new products, there is no past, and so other methods can be tried.

Forecasting

Qualitative Quantitative
Methods Methods

Executive Market Grassroot

Delphi Time Series Causal
Judgement Research Approach

Surveying Moving
customers Averages

Testing Product Weighted

performance Moving Average

Exponential
Focus Group
Smoothing

Historical
Trend Analysis
Analogy
Methods of Forecasting
Qualitative Techniques
Grass Root method
This method involves talking to sales people to gauge the future. The method is also referred to as
the Naïve method. The knowledge of the Sales people on Market trends and customer preferences is
tapped to gain an indication of the future markets.

Market Research
In Marketing Research, there are Four methods to select from. The first involves Surveying
Customers. This involves asking customers about their opinion on the product we wish to estimate.
The second, involves testing the performance of the product in the market place. Third involves the
use of a focus group to discuss the product in order to come up with an estimate of its performance
in the future. Finally, there is Historical analogy in which the future estimates are deduced from
other products with similar salient property(ies).

Executive Judgement
In this method the judgement of managers is sought in order to come up with estimates of a
product’s future performances.

Delphi
This involves a Delphi manager who is in contact with a number of some other experts in particular
fields. The manager may or may not start the experts up on an estimation of performance on a
particular product by showing them some estimates from another methods and seek for their
opinions. She then pools the information for moderation and then resends the new position with all
the attached reasons to the experts to seek for their second opinion after which the manager draws
a conclusion.

Quantitative Methods
Time Series Analyses
A time series is a stream of data that represents past measurements. Each event (for example
observation of demand) is time-tagged so that it is known where it is located in the series of data.
The time series consists of data recorded at different time periods such as weekly or daily for the
variable, which could be units produced or demands received. Forecasters attempt to predict the
next value or set of values that will occur at a future time. In time-series analysis, external causes are
not brought into the picture. The pattern of the series is considered to be time-dependent.
Extrapolation is the process of moving from observed data (past and present) to the unknown values
of future points. The extrapolation of time series is one of the main functions of forecasting. The
data in time series may consist of several different kinds of variations. Important among them are
random variations, an increasing or decreasing trend, and seasonal variations. Random variations
occur because the demand is seldom constant at a given level. Minor variations do occur from one
period to another. In many instances, there are no specific assignable causes for these random
variations. The random variations are a result of the economic environment and the marketplace
within which an organization is operating. If the time series shows an increasing linear trend there is a
constant rate of change (increasing) with the increase in time. Linear decreases can also occur. Nonlinear
trend lines occur where the rate of change is geometric. In such cases, it is possible to extrapolate
the curve by eye, but the rate of change can also be calculated, and projections can be made
mathematically. The time series may also exhibit seasonal (or cyclical) variations.
Time-series analysis is an analysis of any variable classified by time, in which the values of the
variable are functions of the time periods. Time-series analysis deals with using knowledge about
cycles, trends, and averages to forecast future events. It is always useful to try to find causal links
between well-known cycles, such as the seasons, and demand patterns that are being tracked.

Moving Average (M. A)

The moving average (MA) method supplies a forecast of future values based on recent past history.
The latest n consecutive values, which are observations of actual events such as daily, weekly,
monthly, or yearly demand, are used in making a forecast. These data are recorded and must be
updated to maintain the most recent n values. With the passing of each time period, the most recent
value is stored and the value of the earliest period is dropped off. For example, if you want to
forecast demand for April using n = 3, then the demands for last three months, that is, January,
February, and March, are needed.

Example: Consider the demand (sales) data below for 10 periods (months in this example). The
forecast for months 4–10 are given in the table. The numbers used in the formula for each month
are given in the calculation column.

Table 1 An Example of the MA Method (n = 3)

Month Sales Forecast Calculation

1 100

2 80

3 90

4 110 90.00 = (100 + 80 + 90)/3

5 100 93.33 = (80 + 90 + 110)/3

6 110 100.00 = (90 + 110 + 100)/3

7 95 106.67 = (110 + 100 + 110)/3

8 115 101.67 = (100 + 110 + 95)/3

9 120 106.67 = (110 + 95 + 115)/3

10 90 110.00 = (95 + 115 + 120)/3

11 105 108.33 = (115 + 120 + 90)/3

12 110 105.00 = (120 + 90 + 105)/3

Classwork 1; Using a graph sheet, plot the data obtained from the table above for the actual and a 3-
months and 5-months moving averages. Discuss the behaviours of the moving averages curves.

Weighted Moving Average (WMA)

A way to make forecasts more responsive to the most recent actual occurrences (demand) is to use
the weighted moving average (WMA) method. Just like the MA method, the most recent n periods
are used in forecasting. However, each period is assigned a weight between 0 and 1. The total of all
weights adds up to 1. The highest weight is assigned to the most recent period and then the weights
are assigned to the previous periods in the descending order of magnitude.
Classwork 2: Consider the data given. Suppose the number of periods used in forecasting n = 3 and
the weights are 0.2, 0.3, and 0.5. The highest weight (in this case 0.5) is assigned to the most recent
period. For example, the forecast for period 4 will be calculated by using the weight 0.5 for period 3,
0.3 for period 2, and 0.2 for period 1. The forecast for period 4 will then be

Forecast (4) = ((0.2 * demandp1) + (0.3 * demandp2) + (0.5 * demandp3))/3

= 0.2 * 100 + 0.3 * 80 + 0.5 * 90 = 89.

Construct the table of forecasts and plot the graph. Comment on the qualities of the two forecast
methods.

Exponential Smoothing
The exponential smoothing (ES) method, like the WMA method, calculates an average demand
(forecast). ES methodology remembers the last estimate of the average value of demand and
combines it with the most recent observed, actual value to form a new estimated average. ES
forecasts the demand for a given period t by combining the forecast of the previous period (t − 1)
and the actual demand of the previous period (t − 1). The actual demand for the previous period is
given a weight of α and the forecast of the prior period is given a weight of (1 − α), where α is a
smoothing constant whose value lies between 0 and 1. The equation for the forecast for period t is

Forecast (t) = α * Actual demand (t - 1) + (1 − α) * Forecast (t − 1),

and 0 ≤ α ≥ 1.

Rearrange to

Forecast (t) = Forecast (t – 1) + [α * (Actual Demand (t – 1) – Forecast (t – 1)],

and 0 ≤ α ≥ 1.

The forecast for the first period is generally set equal to the actual demand in that period to get the
forecasting process started. In this way, the forecast for the second period of our data, using alpha
of 0.2, will be;

F2 = 100 + 0.2(100 – 100) =100.

Classwork 3; Determine the remaining forecasts from our data and plot the graph. Discuss the
differences in shapes of our three plots.

Exponential Smoothing can be more effective because only one weight, alpha (α), has to be chosen.
This makes it easier to experiment with past data to see which value of α provides the least forecast
error. Often, the use of the ES method for forecasting is preferred to the WMA method. The
response rate of the ES model is a function of the α value that is used. The response rate of a
forecasting system is the speed with which the forecasting system makes adjustments to the
forecasts if the data in the series show an upward or downward trend. Small values of α (which will
be analogous to large values of n in MA) are used for stable systems where there is, at most, a
minimum amount of random fluctuation. Large values of α are used for changing and evolving
systems where much reliance is placed on the last observation. New products, as they move through
their life-cycle stages, start with a large α value, which gradually diminishes as the product enters its
maturation stage. The ES method has a clear advantage over the MA method in terms of the amount
of memory required for storing the data. In addition to the smoothing constant α, at any time we
need to store only the exponentially smoothed average and the actual demand for the previous
period whereas for the MA method, actual demands for the last n periods have to be stored.

The different values of α used give different forecast just as different forecasts are obtained by
changing the values of n in the MA method. A small value of α makes minor adjustments to the
forecast results as discussed above—giving smoother forecasts. A smooth forecast is obtained in the
MA method if n is large. Therefore, smaller values of α and large values of n tend to give similar
forecasts. Similarly, large values of α and small values of n tend to give similar results. In fact, when α
= 1, or n = 1, the forecast in a given period is equal to the demand in the previous period and the
forecasting system is extremely responsive to changes in the demand. In this case, the forecast
fluctuates as actual demand fluctuates. However, forecasts are lagging behind the demand by one
period the actual demand for the previous period. For the MA method, actual demands for the last n
periods have to be stored.

Trend Analyses
If the time series exhibits seasonality ln an increasing or decreasing trend, then the techniques
discussed above (MA, WMA, and ES) may not be appropriate for making a forecast. We perform a
trend analysis to make a forecast in this case. The trend analyst will fit a trend line, either by eye or
by using a formula, through the data. This line, or curve, fitted can then be extrapolated to make
forecast. The equation of the trend fitting line is,

Y = a + bX,

where Y is the demand forecast and X is the time period. X is the independent variable and Y is the
dependent variable since the demand depends on the time period. As X increases, Y increases in an
increasing trend. Y will decrease as X increases in a decreasing trend. In the trend line equation, a is
the intercept on the Y-axis which is the value of the demand (variable Y) when X = 0. The slope of the
line is represented by b which gives the change in the value of demand (variable Y) for a unit change
in the value of X. That is, b is the amount by which demand will change if the time period changes by
1. The intercept (a) and the slope (b) can be calculated from

a=Y −b X

b=
∑ xy−n x y
∑ x 2−n x 2
Or by the use of Microsoft excel or similar software.

Classwork 4; Plot the graph of the series below and determine the next two values of the forecasts.

x 1 2 3 4 5 6 7 8 9 10
y 9 15 32 48 52 60 39 65 90 93
Forecasting Errors
It is very rare to get 100% accurate forecasts. There are always some errors in forecasting. A good
forecasting technique tries to minimize the errors. The forecasting errors are computed by
comparing the actual demand for time period t with the forecast for that same period, that is,
Error (t) = Demand (t) − Forecast (t).
Two kinds of forecasting errors can occur.
First, actual demand is greater than the forecast. This is a forecasting underestimate. By convention,
when actual demand is greater than forecast demand, the error term is positive.
Second, actual demand is less than the forecast. This is a forecasting overestimate and in this case
the error is negative.
To choose a forecasting method, it is necessary to be able to compare the errors that each method
generates for particular circumstances. There are several different ways of measuring errors. The
choice is dependent on the situation and what kind of comparison is wanted. Here, we would
consider simple deviation (D), Absolute Deviations (AD) and Mean Absolute Deviations (MAD).
Deviations are the error terms away from the actual. Absolute measures mean that positive and
negative errors are treated the same way. The reason for doing this is to prevent positive and
negative errors from cancelling each other out. To calculate MAD, take the sum of the absolute
measures of the errors and divide that sum by the number of observations. MAD treats all errors
linearly and is a more conservative measure.

Example: Consider the data given. The demand and forecast are given for 10 periods in columns 2
and 3. Column 4 gives the error for each period, and column 5 gives the absolute error. The sum of
the absolute errors for 10 periods is 171. Therefore, MAD = 171/10 = 17.10.

Period Actual Forecast Error Absolute

Demand Error
1 212 206 6 6
2 224 207 17 17
3 220 210 10 10
4 211 212 -1 1
5 198 205 -7 7
6 236 209 27 27
7 219 224 -5 5
8 296 238 58 58
9 280 249 31 31
10 252 261 -9 9
Total 127 171
To select a particular forecasting method, say ES, we will calculate the values of MAD by using
different values of α. The value of α that minimizes MAD will be selected. Similarly, applied to the
MA method, we will calculate MAD for different values of n. The n that minimizes MAD will be
selected. A similar procedure will be used with the WMA method to find the best combination of
weights.
Causal Analyses
There are various relationships that can exist between two or more products. Specifically, in Causal
analyses, how the factors of a product or of other products determine the performance of another
factor of interest to us is analysed. For example, a Causal analyst maybe interested in how such
factors as, advertisement spend on a product, price of the product and or price of a substitute or
even complimentary product may affect the demand of her product of interest. Regression analysis
essentially fits a line, referred to as the regression line, between observations of the independent
and dependent variables.

Classwork 5; Given the table below of actual relationship between the Advertising spending of a
company and its sales, what would be its next period Forecast of Sales if N1,750 is earmarked for
Advertisement.

Month Ad Bug Sales

(N.000) units
(.000)
1 2.5 264
2 1.3 116
3 1.4 165
4 1.0 101
5 2.0 209

First step, we need our slope and intercept

b=
∑ xy−n x y
∑ x 2−n x 2
Month Advert Sales xy
Budget units
(N.000) (.000) x
2
(x) (y)
1 2.5 264 660 6.25
2 1.3 116 150.8 1.69
3 1.4 165 231 1.96
4 1 101 101 1
5 2 209 418 4
Sums 8.2 855 1560.8 14.9
Means= 1.64 171

Therefore,

b = 1560.8 – 5*1.64*171/14.9 – 5*1.64*1.64

= 158.6/1.452

= 109.229

And since

a=Y −b X

= 171 – 109.229* 1.64 = -8.136

Remember Y = a + bx and given Advert budget of N1,750 which in (.000) = 1.75

So, Y = -8.136 + 109.229*1.75

= 183.015 = 183,015 Units.

Forecasts Accuracy
1. Correlation
2. Coefficient of Determination
3. Standard Error of the Estimate

Correlation is the measures the strength of the relationship that exist between the dependent and
independent variables. It ranges between -1 through 0 to +1. It is calculated as;
n
r =∑ ¿ ¿ ¿
1

Month Ad Bug Sales units xy xi - ẋ yi - ẏ (xi - ẋ) (xi - ẋ)sqr (yi - ẏ)sqr

(N.000) (.000) (yi - ẏ)
2
x
1 2.5 264 660 6.25 0.86 93 79.98 0.7396 8649
2 1.3 116 150.8 1.69 -0.34 -55 18.7 0.1156 3025
3 1.4 165 231 1.96 -0.24 -6 1.44 0.0576 36
4 1 101 101 1 -0.64 -70 44.8 0.4096 4900
5 2 209 418 4 0.36 38 13.68 0.1296 1444
8.2 855 1560.8 14.9 158.6 1.452 18054
Means= 1.64 171

That is

r = 158.6/√ 1.452∗18054

= 158.6/161.9 = 0.98

This value of 0.98, which is very close to +1, indicates a strong positive relationship between the
independent variable, the advertising spends, and the dependent variable, the volume of sales.

The Coefficient of Determination is the square of the correlation and ranges from 0 to 1. The closer
the figure found to the value of 1 the stronger the relationship between variables.

So, for us, the r2 = 0.98*0.98 = 0.9604

Or calculated by using the formula

r2 = ∑ ¿¿ ¿
Where yˋ1 = a + bx1 i.e = - 8.136 + 109.229*2.5 =264.9365

Which we can extend our table above with

 yˋ

264.4965 -0.9365 0.877

133.8617 -17.8617 319.04
144.7846 20.2154 408.662
104.094 -0.093 0.008649
210.322 -1.322 1.747684
730.3353
Hence r2 = 18054 – 730.335/ 18054

= 17323.665/18054 = 0.96.

Standard Error of the Estimate measures the mean absolute error per period. The lower the value
calculated the better. It is calculated as

∑ yi− yˋ ¿
σxy = √ n−2
2

= √ ❑730.335/n – 2 = √ 243.445 = 15.6.

So the three measure agree that there is goodness of fit between our independent and dependent
variables.

Microsoft excel output

SUMMARY OUTPUT
Regression Statistics
Multiple R 0.979565
R Square 0.959547
Adjusted R 0.946063
Square
Standard 15.60274
Error
Observation 5
s

ANOVA
df SS MS F Significanc
eF
Regression 1 17323.6 17323.6 71.1603 0.003496
6 6 8
Residual 3 730.336 243.445
1 4
Total 4 18054

Coefficient Standard t Stat P-value Lower 95% Upper Lower Upper

s Error 95% 95.0% 95.0%
Intercept -8.13499 22.3524 -0.36394 0.74003 -79.2705 63.0005 -79.2705 63.0005
7 9 6 6
X Variable 1 109.2287 12.9484 8.43566 0.00349 68.02094 150.436 68.0209 150.436
4 1 6 4 4 4

RESIDUAL OUTPUT

Observation Predicted Y Residual Standard

s Residual
s
1 264.9366 -0.93664 -0.06932
2 133.8623 -17.8623 -1.32192
3 144.7851 20.2148 1.49602
8 8
4 101.0937 -0.09366 -0.00693
5 210.3223 -1.32231 -0.09786