Quantitative

Business
Analysis
Chapter 10: Forecasting
Forecasting
Forecasting is the process of making
predictions or estimates about future
events or conditions based on historical
data, trends, and patterns.
 Use of Forecasting
● It is used in many different fields, including finance, economics,
marketing, supply chain management, and more.

● The primary purpose of forecasting is to help organizations

make informed decisions by providing them with insights into
future outcomes or trends.

● Forecasting can help organizations plan for future demand,

allocate resources effectively, and identify potential risks or
opportunities.

● It can also be used to develop sales forecasts, production

forecasts, budget forecasts, and workforce forecast
 Types of Forecasting Methods
Counting Method
• Based on simple counting of observations or events
• Useful when the occurrence of events is the primary focus
• Can be used to forecast things like customer traffic or equipment failure rates
• May not account for complex relationships or patterns in the data

Time Series Forecasting Methods

• Based on past data
• Assumes past patterns and trends will continue into the future
• Can be simple or sophisticated
• Can be automated

Judgmental Forecasting Methods

• Based on expert opinions and subjective judgments
• Useful when historical data is scarce or unreliable
• Can incorporate qualitative factors that may not be captured in the data
• Can be prone to bias and subjectivity

Causal Forecasting Methods

• Based on identifying cause-and-effect relationships between variables
• Useful for forecasting changes in response to specific interventions or events
• Requires accurate measurement and understanding of the underlying causal factors
• Can be complex and computationally intensive
 Factors Influencing Choice of Forecasting Methods
Availability of Historical Data
• The amount and quality of historical data can significantly impact the choice of forecasting method
• Some methods rely heavily on historical data, such as time series analysis, and may not be suitable if data is scarce or
of poor quality
• If historical data is incomplete or inconsistent, it may be necessary to use other methods, such as expert judgment or
causal analysis

Availability of Money and Time

• The availability of resources, including money and time, can also influence the choice of forecasting method
• Some methods may require specialized software, training, or expertise, which may be costly or time-consuming to
acquire
• In some cases, it may be necessary to use simpler, less resource-intensive methods, such as moving averages or
exponential smoothing, due to budget or time constraints

Accuracy of Data
• The accuracy and quality of data used in forecasting can affect the choice of method
• If the data is unreliable or inconsistent, it may be necessary to use more robust methods, such as causal analysis or
judgmental forecasting
• It may also be necessary to clean and preprocess the data before using certain methods, which can require additional
resources and expertise
 Moving Averages

To overcome the deficiency of using a simple

average, the moving average technique
generates the next period's forecast by
averaging the actual demand for only the last
n (n is often in the range of 4 to 7) time
periods, rather than for all the periods such as
in a simple average.
TABLE 10.1 Patient Demand for the Lakeland City Health Center
Year Quarter Period Number Number of Patient Visits
1990 1 1 3,500
2 2 8,000
3 3 5,500
4 4 10,000
1991 1 5 4,500
2 6 6,000
3 7 3,000
4 8 5,500
1992 1 9 5,000
2 10 9,500
3 11 7,500
4 12 15,000
1993 1 13 13,500
2 14 17,500
3 15 ?
 Moving Averages

Number of visits
(1000)
 Moving Averages
Formula of moving average:

Ft+1= 1/n Σ t Di
i=(t-n+1)
Where:
t= Period number for the current period =14
F t+1 = Forecast of demand for the next period =F 15

Di = Actual demand in period =D11, D12, D13, D14

n = Number of periods of demand to be included (known as the "order" of the

moving average) =4
So,
F14+1= 1/4 Σ 14 D1
i= (14-4+1)

F15= (D11+ D12 + D13 +D14)/4

F15=(7500+15000+13500+17500)/4
=13,375
Weighted Moving
Averages
Weighted moving average (WMA) is a
forecasting method that assigns different
weights to each data point in a time series
before calculating the moving average. This
allows more recent data points to have a
greater influence on the forecast than older
data points.
 Calculating WMA
• First determine the weights to be used for each data point

• In a four-period moving average, we may use the weights of 0.1, 0.2, 0.3, and 0.4,
respectively, for the oldest to the most recent data point

• These weights should add up to 1 to ensure that the moving average is properly
normalized
• Once the weights are determined, calculate the WMA forecast for a given period by
multiplying each data point by its corresponding weight

• Then we have to sum up the results

 Calculating WMA

For instance, if we want to forecast the value for period 15 using the weights mentioned
above, we can use the formula:

F15 = .10D11 + 20D12 + .30D13 + .40D14

= .10(7,500) + 20(15,000) + .30(13,500) + .40(17,500)
= 14,800

This compares to the equally weighted moving averages forecast of 13375. We can
adjust the weights and the number of periods used in the moving average to suit our
needs.
Simple Exponential
Smoothing
Simple exponential smoothing is a popular
forecasting method that works by assigning
exponentially decreasing weights to past
observations based on their age. The most recent
observation is given the highest weight, and the
weights decrease exponentially for each
previous observation. The sum of the weights is
equal to 1.
 Simple Exponential Smoothing
The formula for simple exponential smoothing is:

Ft+1 = α(Dt) + (1 - α)Ft

Where:
Ft+1 is the forecast for the next period
α is the smoothing factor (0 ≤ α ≤ 1)
Dt is the actual value of the time series for the current period
Ft is the forecast for the current period

The value of α determines the degree of smoothing applied to the data. A larger value of α gives
more weight to recent data and produces a forecast that is more responsive to changes in the time
series. On the other hand, a smaller value of α gives more weight to historical data and produces a
forecast that is smoother and more stable.
Forecasting Error
Forecasting error is the difference between the
actual value and the forecasted value for a
given period in a time series. It is used to
evaluate the accuracy of a forecasting model
and to identify areas where improvements can
be made.

There are two other commonly used measures:

• Mean Absolute Deviation (MAD)
• Bias
Mean Absolute Deviation
(MAD)
MAD measures the average magnitude of the
errors in the forecast. The formula for MAD:

MAD = (1/n) * Σ|Error|

where n is the number of periods in the time

series, and Σ|Error| represents the sum of the
absolute values of the errors.
MAD is a useful measure of the accuracy of
the forecast, as it provides a sense of the
typical size of the errors made by the model.
Bias
Bias measures the tendency of the model to
over- or under-forecast the time series. It is
calculated as the average of the forecast errors,
and can be positive or negative depending on
whether the model tends to over- or under-
forecast. The formula for Bias:

Bias = (1/n) * Σ(Error)

where n is the number of periods in the time

series, and Σ(Error) represents the sum of the
forecast errors.
Bias
Example Month Method A

January 102

February 107
MAD (A) = (|102 - 103| + |107 - 106| +
|106 – 106| + |113 – 111 |) / 4 March 106
=1
Bias (A) = [(102 - 103) + . . .] / 4 = 0.5. April 113
Causal Method
While the trend extension method assumes
that time reflects all factors, the causal
method (also called the regression method)
seeks to establish direct relationships between
fertilizer demand and factors influencing it.

Types of Causal Forecasting Method:

• Linear (Least Squares) Regression
• Multiple Regression
Linear Regression
Least squares regression is used to estimate a
trend-predicting equation. The simple linear
regression equation is:

Y₁ = a + bx

In the case of housing starts, it would be

interpreted as:
Y₁ = Predicted number of new housing starts
a = The Y-axis intercept
B = The slope =1
X = Number of new marriages in the previous
quarter
 Multiple Regression
This linear regression methodology can be extended
to situations in which more than one explanatory
variable is used, called multiple regression, to
explain the behavior of one dependent variable Y.

Y’x1x2 = a + b1X1 + b2X2

Y’x1x2 = The predicted number of housing starts based on new

marriages and the employment rate
a = The Y-axis intercept
b, and b₂ The slopes (rates of change in Y') with respect to X 1
and X2
X1 and X2 = The number of marriages and the employment
rate, respectively, in the previous quarter
 Decomposition method
The decomposition model assumes that sales are
affected by four factors:

● Trend T
● Seasonal variation S
● Cyclical variation C
● Random variation R

The forecast is made by considering each of these

components separately and then combining them
together.
 Decomposition Procedure
Step 1 Step 2 Step 3
Collection and Graphing the Time Measure Trend and
Adjustment of data Series Cycle (T x C)

Step 4 Step 5 Step 6

Seperating the Computing the Measuring Combined
Effect of the Trend Cycle Effects of Seasonal &
Random Variation
 Decomposition Procedure

Step 7 Step 8 Step 9

Compute the Seasonal Measuring the Develop a Forecast

Variations Random Variations
 Thank
You!
Group 1
190308, 190312, 190318, 190319, 190335