Module 2 – Forecasting

1. What is forecasting?
Forecasting is defined as estimating the future value that a parameter will take. Most
scientific forecasting methods forecast the future value using past data. In
Operations Management forecasting is used extensively to estimate future demand
of product(s)

2. What is time series forecasting?

Time series forecasting uses past data to estimate the future values. Here the
performance with respect to time is considered. Time is the independent variable.

3. Mention some simple forecasting models for time series data?

Some simple forecasting models using time series data are simple average, moving
average and simple exponential smoothing.

4. What is moving average?

Moving average is a simple time series forecasting model based on averages of a
chosen number of periods. It is used to forecast a constant model or level data.

5. Write the basic equation for simple exponential smoothing?

𝐹𝑡+1 = 𝛼𝐷𝑡 + (1 − 𝛼)𝐹𝑡 . This equation can be used when simple exponential
smoothing is used as a forecasting model. Here F t represents the forecast for period
t, D t is the known demand for period t and α is the smoothing constant. The basic
equation for simple exponential smoothing is 𝑌�𝑡 = 𝛼𝑌𝑡 + (1 − 𝛼)𝑌�𝑡−1 . Here 𝑌�𝑡 is the
smoothed value of the data including Y t the most recent data.

6. Is exponential smoothing a form of weighted average? How?

Simple exponential smoothing can be seen as a form of weighted moving average.
Expanding the general equation Ft +1= α Dt + (1 − α ) Ft , we get
Ft +1 = α Dt + α (1 − α ) Dt −1 + α (1 − α ) Dt − 2 + ... + α (1 − α ) D1 + (1 − α ) F1
2 t −1 t

As t is large and tends to infinity, the term (1- α)t tends to zero. The rest of the terms
are all terms involving D j . It can be seen that the F t+1 value is a weighted average of
the terms D t to D 1 with weights α , α (1 − α ) , α (1 − α ) ,... + α (1 − α ) . If 0 ≤ α ≤ 1,
2 t −1

each weight is smaller than 1 and is decreasing. The highest weight is given to the
most recent point and the weights progressively decrease by a factor (1 – α) as the
data gets older. As t tends to infinity, the weights are α , α (1 − α ) , α (1 − α ) ,... This is
2

an infinite geometric series whose first term is α and the common term is (1 – α).
α
The sum of all the terms of the progression is = 1.
(1 − (1 − α ))

7. What are the implications of using small α?

A small value of α implies that initial weight given to the recent data is small and the
subsequent weights are smaller. This means that more terms contribute to the
forecast. This also means that more weight is given to the forecast than to the
demand.
8. In the equation Y = a + bt + ε, what does ε represent? What can you say about the
mean and variance of ε?
The symbol ε represents the error term. It is assumed to be normally distributed
with mean = 0 and with small variance. This means that the errors are expected to
cancel out each other. The error term is also expected to be small.

9. What do a and b represent in the equation Y = a + bt + ε?

In the linear equation Y = a + bt + ε, b is the slope and a is the y intercept – the point
in which the line touches the y axis.

10. Write the equations for Holt’s model?

The basic equation for Holt’s model is F t+1 = a t + b t . Here a t is called the level which
represents the smoothed value up to and including the last data. The slope of the
line is given by b t and therefore the forecast for the next period F t+1 = a t + b t . The
values of a t and b t are updated using a t = α D t + (1 – α)(a t-1 + b t-1 ) and b t = β (a t – a t-
1 ) + (1 – β)b t-1 .

11. How is the Holt’s model different from the linear regression model?
Holt’s model is different from linear regression because it computes different values
of the slope and intercept at different points using simple exponential smoothing

12. Write the equations for the Winter’s model?

F t+1 = (a t + b t )C t+1 where a t and b t are the level and trend as described in the Holt’s
model. C t+1 is the seasonality index for the period that we are forecasting. The
equations are
a t+1 = α(D t+1 /C t+1 ) + (1-α)(a t + b t )
b t+1 = β(a t+1 - a t ) + (1 - β)b t
C t+p+1 = γ(D t+1 /a t+1 ) + (1 - γ)C t+1

13. What is seasonality index and how is it calculated?

Seasonality index captures the effect of the season on the data. It can be defined as
S i = D i /Average. For example, if there are 4 seasons, we can compute the average of
the demands of four seasons. The demand in a period divided by the average gives
the seasonality index for the period. It is important to know the number of periods
that constitute a season.

14. Mention some measures of goodness of forecasts?

Some measures of goodness of forecasts include, mean squared deviation, mean
absolute deviation, mean percentage deviation etc.

15. What is a causal model?

In a causal model, there is an independent variable or a causal variable that impacts
the dependent variable (demand).

16. Write equations for causal model?

 The equation for a causal model is Y = a + bX where X is the independent (causal)
variable and Y is the dependent variable. This is a linear model. Other models exist.

Problems

1. Given the data 92, 93, 92, 91, 93, 94, 92 find the forecast for the eighth period using
simple average, weighted average (weight of 1 for the first four periods and 2 for the
remaining three), 3 period moving average?
Simple average = (92 + 93 + 92 + 91 + 93 + 94 + 92)/7 = 92.28
Weighted moving average = [ (92+ 93 + 92 + 91) + 2(93 + 94 + 92)]/10 = 92.6
Three period moving average = (93 + 94 + 92)/3 = 93

2. Given the data 92, 93, 92, 91, 93, 94, 92 find the forecast for the eighth period using
simple exponential smoothing? Use α = 0.3 and initial forecast using simple average?
Simple average = 92.28; F 1 = 92.28, α = 0.3
F t+1 = αD t + (1- α)F t ; F 2 = 92.196, F 3 = 92.44, F 4 = 92.31, F 5 = 91.91, F 6 = 92.24, F 7 =
92.77, F 8 = 92.54

3. Given the data 63, 64, 66, 67, 67, 69, 71, 72 find the forecast for the eighth period
using simple average, and 3 period moving average? Is it a good forecast? Why or
why not?
Simple average = 67.375, three period moving average forecast = (69 + 71 + 72)/3 =
70.666. Both are not good forecasts because the data shows increasing trend while
the forecasting models used are for constant (level) data and indicate a central value
(average).

4. Given the data 63, 64, 66, 67, 67, 69, 71, 72 find the forecast for the ninth period
using simple exponential smoothing? Use α = 0.3 and initial forecast using simple
average. Is it a good forecast? Why or why not?
F 1 = average = 67.375, F t+1 = αD t + (1- α)F t ; F 2 = 66.06, F 3 = 65.44, F 4 = 65.61, F 5 =
66.03, F 6 = 66.32, F 7 = 67.12, F 8 = 68.27, F 9 = 69.40 . The forecast is not good
because the data shows increasing trend while simple exponential smoothing is to
be used for constant (level) data and indicates a central value (average).

5. Derive the expression for a and b in the equation Y = a + bt?

n
We wish to derive yt =a + bt + ε and find a and b such that ∑( y − a − bt ) is minimized.
2
t
t =1
Partially differentiating the residue with respect to a and b and setting the first derivative to

zero, we get the equations ∑ y= t na + b∑ t and ∑

= tyt a ∑ t + b∑ t 2 . Here a and b

are unknowns and the other terms can be computed. Solving these equations we get the
values of a and b.

6. Given the data 63, 64, 66, 67, 67, 69, 71, 72 find the forecast for the ninth period
using linear regression?
 We compute ΣY = 539, Σt = 36, Σt2 = 204 and ΣYt = 2479. The equations are 539 = 8a
+ 36b and 2479 = 36a + 204b. Solving, we get b = 1.274 and a = 61.64. F 9 = a + 8b =
73.11

7. Given the data 63, 64, 66, 67, 67, 69, 71, 72 find the forecast for the ninth period
using Holt’s model? Use α = β = 0.2.
The equations for Holt’s model have been given earlier. Using these equations, we
get F 2 = 64.29, F 3 = 65.51, F 4 = 66.91, F 5 = 68.23, F 6 = 69.23, F 7 = 70.42, F 8 = 71.80
and F 9 = 73.1

8. Data for four quarters for three years is 81, 62, 76, 55, 85, 65, 79, 60, 90, 69, 84, 64.
Find the forecast for the next four periods using a simple seasonality model
computing seasonality indices?
We assume that there are 4 seasons and data for three years (say). The total
demands for 3 years are 274, 289 and 307. The forecasted total demand is 323 and
per season it is 80.75. The seasonality indices (average) are 1.18, 0.9, 1.09 and 0.82.
The forecasted values are 95.3, 72.68, 88, 67.

9. Data for four quarters for three years is 81, 62, 76, 55, 85, 65, 79, 60, 90, 69, 84, 64.
Find the forecast for the next four periods using Winter’s model. α = β = 0.2, γ = 0.3.
The equations for Winters model are given: F t+1 = (a t + b t )C t+1 where a t and b t are the
level and trend as described in the Holt’s model. C t+1 is the seasonality index for the period
that we are forecasting. The equations are a t+1 = α(D t+1 /C t+1 ) + (1-α)(a t + b t ); b t+1 = β(a t+1 - a t )
+ (1 - β)b t ; C t+p+1 = γ(D t+1 /a t+1 ) + (1 - γ)C t+1 . The computations give C 13 = 0.3, C 14 = 0.225, C 15
= 0.275, c 16 = 0.2, a 12 = 313.9, b 12 = 4.22, F 13 = 95.4. Using F 13 = D 13 , we get F 14 =71, F 15 =
89.5, F 16 = 66

10. Students believe that the salary they can expect during a placement process is
related to their academic performance. The CGPA (indicator of performance) and the
salary obtained by six students are (7, 6), (6.8, 5.8), (7.5, 6.5), (8, 7), (8.2, 7.5) and
(8.6, 8). Find the salary that a student with CGPA 8.7 can expect?
We build a causal model of the form Y = a + bX where Y is the salary and X is the
CGPA. The equations are ΣY = na + bΣX; ΣXY = aΣX + bΣX2. Computing, we get ΣX =
46.1, ΣY = 40.8, ΣXY = 316.49 and ΣX2 = 281.14. Solving, we get a = 7.11 and b = -0.04
For X = 8.7, Y = 6.76.

11. Given the data 92, 93, 92, 91, 93, 94, 92 find the forecast for the eighth period using
simple average. Compute the mean absolute deviation?
∑n Di − Fi
F 8 = 92.43. Mean absolute deviation = MAD = . Computing, we get MAD =
n
(.43 + 0.57 + 0.43 + 1.43 + 0.57 + 1.57 + 0.43)/7 = 0.776

12. Given the data 83, 87, 90, 92, 96, 99, find the forecast for the eighth period using
linear regression?
We compute ΣY = 547, Σt = 21, Σt2 = 91 and ΣYt = 1969. The equations are 547 = 6a +
21b and 1969 = 21a + 91b. Solving, we get b = 3.11 and a = 80.27. F 8 = a + 7b = 102
13. Given the data 83, 87, 90, 92, 96, 99 find the forecast for the eighth period using
Holt’s model? Use α = β = 0.2.
The equations for Holt’s model are given earlier. Performing the computations, we
get F 7 = a 6 + b 6 = 99.14 + 3.21 = 102.35

14. Given the data 92, 93, 92, 91, 93, 94, 92 find the forecast for the eighth period using
simple exponential smoothing? Use α = 0.2 α = 0.7 and initial forecast using simple
average? Explain the effect of α on the contribution of the data for the various
periods
At α = 0.2, F 7 = 92.63. Since we are building a constant model, F 8 = 92.63. At α = 0.7,
F 7 = 93.55. Since we are building a constant model, F 8 = 93.55. Lower α gives more
weight to forecast while high α gives more weight to demand.

15. Consider the data 92, 93, 92, 91, 93, 94, 92. Find the mean absolute deviation for the
forecast using simple exponential smoothing α = 0.2.
The forecasts are F 1 = 92.43, F 2 = 92.34, F 3 = 92.47, F 4 = 92.38, F 5 = 92.43, F 6 = 92.43,
F 7 = 92.43. MAD = (0.43 + 0.66 + 0.47 + 1.38 + 0.9 + 1.71 + 0.63)/7 = 0.883.