BIS Application Chapter two

1
BIS Application Chapter two

• Forecasting

2
Forecasting
Forecasting is the process of extrapolating the
past into the future Forecasting is something
that organization have to do if they are to plan
for future. Many forecasts attempt to use past
date in order to identify short, medium or long
term trends, and to use these patterns to project
the current position into the future. Backcasting
method of evaluating forecasting techniques by
applying them to historical data and comparing
the forecast to the actual data
3
Forecasting

• Why Forecasting?
• Characteristics of Forecasts
• Forecasts are usually wrong or seldom correct
• Aggregate forecasts are usually more accurate
• Less accurate further into the future
• Assumptions of Forecasting Models
• Information (data) about the past is available
• The pattern of the past will continue into the
  future.

4

• The forecasting approach to forecasting
• Starts with gathering and recording information
  about the situation
• Entering the data into the worksheet,
• Creating graphs
• The data and graphs are examined visually to get
  some understanding of the situation (judgmental
  phase)
• Developing hypotheses and models
• Trying alternative forecasting approaches and
  doing what if analysis to check if the resulting
  forecast fits the data

5
(No Transcript)
6
Forecasting Approaches
1- Qualitative Forecasting Forecasting
based on experience, judgment, and knowledge 2-
Quantitative Forecasting Forecasting based on
data and models
7
Forecasting Approaches

Judgmental/Qualitative
Time Series
Causal
Moving average
Regression
Curve fitting
Exponential smoothing
Econometric
Trend projection
Seasonal indexes
8
Quantitative Forecasting
Time Series Models
Sales1999 Sales1998 Sales1997
Year 2000 Sales
Time Series Model

• Casual Models

Price Population Advertising
Causal Model
Year 2000 Sales
9
Time series model

• Is based on the hypothesis that the future can be
  predicted by analyzing historical data samples.
  The time series model have the following type ,
  which can be classifies as shown below
• Forecasting directly from the data value (non
  seasonal)
• Moving average
• Exponential smoothing
• Forecasting by identifying patterns in the past
  data (seasonal) (Chapter 3)
• Trend projections
• Seasonal influences
• Cyclical influences

10
Time series model

• The Time series model can be also classified as
• Non-seasonal Model
• Trend
• Moving average
• Exponential smoothing
• Seasonal Model
• Seasonal Decomposition
• Cyclical influence

11
Causal Models (Chapter 3)
Causal forecasting seeks to identify specific
cause-effect relationships that will influence
the pattern of future data. Causes appear as
independent variables, and effects as dependent ,
response variables in forecasting
models. Independent variable Dependent,
response variable Price demand Decrease in
population decrease in demand Number of
teenager demand for jeans The issue is to
determine the approximate functional
relationships, the model, and the parameter of
the model that relate the input(independent) and
output(dependent) variables.
12
Causal Models (Chapter 3)

• Regression analysis
• Curve Fitting Simple Linear Regression
• One Independent Variable (X) is used to predict
  one Dependent Variable (Y) Y a b X
• Find the regression line with Excel
• Use Function
• a INTERCEPT(Y range X range)
• b SLOPE(Y range X range)
• Use Solver
• Use Excels Tools Data Analysis Regression

13
Causal Models (Chapter 3)
Curve Fitting Multiple Regression Two or more
independent variables are used to predict the
dependent variable Y b0 b1X1 b2X2
bpXp Use Excels Tools Data Analysis
Regression
14
Evaluation of Forecasting Model
To judge how well a forecasting model, or indeed
any forecast, fit the past observation , both
precision and bias must be considered. a-
Measuring the precision of a forecasting model
There are four possible measures used to
evaluate precision of forecasting systems, each
based on the error or deviation between the
forecasted and actual values Average of the
deviation, MAD, MAS, MAPE b - Measuring the bias
of a forecasting model The bias of a forecasting
model is examined on the basis of the spread of a
set of data which can be measured by its
variance, which depends on the sum of squares of
the differences between the values and their
mean. The more of the spread that is accounted
for by the fitted model , the more precise the
fit of the model to the data. R2 used only for
curve fitting model such as regression
15
Evaluation of Forecasting Model

• The arithmetic mean of the errors (the average
  deviation )
• n is the number of forecast errors
• Excel AVERAGE (error range)

16
Evaluation of Forecasting Model
Mean Absolute Deviation - MAD No direct Excel
function to calculate MAD
17
Evaluation of Forecasting Model

• Mean Square Error - MSE
• Excel SUMSQ(error range)/COUNT(error range)

18
Evaluation of Forecasting Model

• Mean Absolute Percentage Error - MAPE

19
Which of the measure of forecast accuracy should
be used?

• Straight average is not used because positive and
  negative deviations cancel out.
• The most popular measures are MAD and MSE.
• The problem with the MAD is that it varies
  according to how big the number are.
• MSE is preferred because it is supported by
  theory, and because of its computational
  efficiency.

20
Which of the measure of forecast accuracy should
be used?

• The ratio of MAD or MSE to the average demand
  which describes the relative percentage of error,
  may be used
• MAPE is not often used.
• In general, the lower the error measure (BIAS,
  MAD, MSE) or the higher the R2, the better the
  forecasting model

21
Good Fit Bad Forecast
As its discussed that neither MAD nor MSE gives
an accurate indication of validity of forecast.
Thus, judgment must be used. Raw data sample
should always be subjected to managerial
judgment, and analyzed and adjusted before formal
quantitative techniques can be applied.
22
a- Dirty Data
Outlier may result from simple data entry
errors, or they may be correct but atypical
observed values (ex can occur in time periods
when the product was just introduced or about to
be phased out). So experienced analyst are well
aware that raw data sample may not be clear.
Demand data with an outlier
P
23
b- Causal data adjustment

• Before quantitative analysis is performed, the
  historical data sample needs to be examined from
  the point of view of cause-and-effect
  relationships.
• A multitude of causes may affect the patterns in
  data sample
• The data sample before a particular year may not
  be applicable because
• - Economic conditions have changed
• - The product line was changed
• Data for a particular year may not be applicable
  because
• - There was an extraordinary marketing effort
• - A natural disaster prevented demand from
  occurring

24
c- Illusory (misleading) patterns
The meaning of a good fit is subject to
interpretation, so before a forecast is accepted
for action, quantitative techniques must be
augmented by such judgmental approaches as
decision conferencing and expert consultations.
25

• To prepare a valid forecast, the following
  factors that influence the forecasting model must
  be examine
• Company actions
• Competitors actions
• Industry demand
• Market share
• Company sales
• Company costs
• Environmental factors

26
Time series forecasting model
27
Time Series Model Building

• Historical data collection
• Data plotting (time series plot)
• Forecasting model building
• Evaluation and selection of model
• Forecasting with the final selected model

28
Components of A Time Series

• Trend long term overall up or down movement
• Seasonality periodic pattern repeating every
  year
• Cycles up down movement repeating over long
  time frame
• Random Variations random movements follow no
  pattern

29
Components of A Time Series
Cycle
Trend
Random movement
Time
Time
Seasonal pattern
Trend with seasonal pattern
Demand
Time
Time
30
First Forecasting directly from the data value
Moving average

• the forecast is the mean of the last n
  observation. The choice of n is up to the manager
  making the forecast
• If n is too large then the forecast is slow to
  respond to change
• If n is too small then the forecast will be
  over-influenced by chance variations

31
First Forecasting directly from the data value
Moving average

• This approach is considered as a quick and
  dirty approach for forecasting
• This approach can be used where a large number of
  forecasting needed to be made quickly, for
  example in a stock control system where next
  weeks demand for every item needs to be forecast

32
Demand
Forecast
33
Longer-period moving averages (larger n) react
to actual changes more slowly
34
First Forecasting directly from the data value
Exponentional smoothing

• it gives weight to all past observations, in such
  a way that the most resent observation has the
  most influence on the forecast, and older
  observation always has less influence than the
  more recent one.
• It is only necessary to store two values (the
  last actual observation and the last forecast,
  plus the value of the smoothing constant) in
  order to make the next periods forecast.
• Smoothing constant (?) the proportion of the
  different between the actual value and the
  forecast.
• F2 ?D1 (1- ?)F1

35
First Forecasting directly from the data value
Exponentional smoothing

• Alpha (smoothing constant) must set between 0 and
  1. Normally the value of the smoothing constant
  is chosen to lie in the range 0.1 to 0.3.
• Typically, a value closer to 0 is used for demand
  that is changing slowly, and a value closer to 1
  for demand that is changing more rapidly.
• There is no way to calculate F1 because each
  forecast is based on the previous forecasts.

36
First Forecasting directly from the data value
Exponential smoothing

• How to select smoothing constant ?
• Sensitivity analysis is an analysis used to test
  how sensitive the forecast is to the change in
  alpha or smoothing constant.
• A general rule for selecting alpha is to perform
  scenario analysis and pick the value that
  produces a reasonable value for the MAD and a
  forecast that is reasonably close to the actual
  demand.

37
Trend-Adjusted Exponential Smoothing
With trend-adjusted exponential smoothing, the
trend is calculated and included in the forecast.
This allows the forecast to be smoothed without
losing the trend. Trend-adjusted exponential
smoothing requires two parameters the alpha
value used by exponential smoothing and beta
value used to control how the trend component
enters the model. Both values must be between 0
and 1. The formula to calculate the forecast
component is F2 FiT1 ?(D1-FiT1) The formula
to calculate the trend component is T2 T1 ?
? (D1-FiT1)
38
Optimizing Trend-Adjust Exponential Smoothing
Optimizing alpha and beta with trend-adjusted
exponential smoothing has a marginal impact. To
find the optimum value for alpha and beta First
the original value of alpha and beta will be used
in the forecasting model. Once the spreadsheet is
ready, Solver is used to vary alpha and beta in
order to minimize the MAD.