ForecastIng
Pirzada
October 2, 2012
Statistics: Few terms
Dependent variable:
Independent variable:
Mean Absolute Deviation: The average forecast error
using absolute values of the error of each past forecast.
Statistics: Few more terms
Regression: Y = a + b x1
Multiple regression: Y = a + b x1 + b x2 + b x3 +
Least square fitting: Sum of the residuals is least
Time series analysis: A type of forecast in which
data relating to past demand are used to
predict future demand.
Linear Regression
Y a bx
Y= Value of dependent variable from
regression equation.
a : Y intercept
b: Slope of the line.
a y bx
xy n x. y
b
x 2 n x x
y : Average of all ys
x : Average of all xs
x: x value of each data point
y: y value of each data point
n: number od data points.
S yx
2
(
y
Y
)
i i
i 1
n2
Forecasts: Need and importance
Basis of corporate planning
Basis of budgetary planning
Operations planning
Capacity planning
Demand Management
Independent demand e.g. cars
Dependent demand e.g. Wheels of the cars
Influencing demand
Active role: Incentives, campaigns, price cuts etc.
Passive role: Simply accept because
Firm is running to full capacity,
market is fixed and static, funding issues etc.
Types of Forecasting
Qualitative
Time series analysis
Casual relationship
Simulation
See Exhibit 13.1, Page 540
Components of Demand
Average demand
Trends
linear
S-curve
asymptotic
exponential
Seasonal elements
Cyclic elements
Random variation
Auto correction
Qualitative Techniques -I
Grass Roots
- The person closest to customers knows future needs best.
Forecasts of bottom level +
safety stocks +
order size corrections
To next level
Qualitative Techniques-II
Panel Consensus
Panel of people from variety of position
Open meeting & free exchange
Historical Analogy
Modeling a generic product
Qualitative Techniques-III
Delphi Method
1.
2.
3.
4.
5.
Choose experts (variety of knowledgeable persons).
Obtain forecast through a questionnaire ((e-mails etc.).
Summarize the results/ redistribute with new questionnaire.
Summarize again develop new questions.
Repeat step 4.
Usually 3 rounds required!
Time Series Analysis
Selection of Method:
1.
2.
3.
4.
5.
Time horizon
Data availability
Accuracy required
Size of forecasting budget
Availability of qualified personnel
Simple Moving Average
At 1 At 2 At 3 ....At n
ft
n
Ft: Forecast for the coming period
n: No. of periods to be averaged
At-1: Actual occurrence in the past period
At-2, At-3, At-n : Actual occurrence in 2
periods ago, 3 periods ago, n periods ago,
Good for cases where demand is stable
and does not have seasonal characteristics
Weighted Moving Average
Allows placing of weights, e.g. most recent more weight
Ft w1 At 1 w2 At 2 w3 At 3 .... wn At n
n
w
i 1
W1: Weight to be given to the actual occurrence for the period t-1
W2: Weight to be given to the actual occurrence for the period t-2
Wn: Weight to be given to the actual occurrence for the period t-n
n: Total number of periods in the forecast.
Exponential Smoothening
Only 3 pieces of data required:
i. Most recent forecast (Ft-1)
ii. The actual demand (At-1)
iii. Smoothening constant ()
Ft Ft 1 ( At 1 Ft 1 )
Premise:
Most recent
occurrences
are most
indicative.
Here Ft is the exponentially smoothed forecast for period t.
Exponential Smoothening
Smoothening constant ()
The value is determined by nature of the product and a
judgment on response rate.
Smaller values
Larger values
Standard items
Fashion items
Stable demand
Growing demand
Small reaction rate
Higher reaction rate
More lagging*
Less lagging*
See Exhibit 13.7 page 550
Trend Effects in Exponential Smoothening
Forecast including trend =
Exponentially Smoothed forecast + Exponentially Smoothed Trend
: Smoothing constant
: Smoothing constant
FITt Ft Tt
Ft FITt 1 ( At 1 FITt 1 )
Tt Tt 1 ( Ft FITt 1 )
FITt: The forecast including trend for period t.
Ft: Forecast for the coming period
Tt: The exponentially smoothed trend for period t.
Forecast Errors
Types and Sources
The confidence band of regression line
does nor work satisfactorily for future values
Bias Errors: making a consistent mistake: Because of :
Failure to include right variables
Using wrong relationships
Wrong trend lines
Undetected trends
Random Errors: Errors that cannot be explained.
Measurement of Errors
Degree of error measured as standard deviation, variance, mean
absolute deviation (MAD)
n
At Ft
MAD i 1
n
1 MAD = 0.8 Standard deviation
3.75 MAD = 3 Standard deviation
An exponentially smoothed MAD is used as error range for next
periods forecast.
Tracking Signal- Measure of Error
Tracking signal is a measurement that indicated whether the
forecast averages keep in pace with any genuine upward or
downward changes in demand. As used in forecasting, the
tracking signal is the number of mean absolute deviations that
the forecast value is above or below the actual occurrence.
RSFE: Running sum of forecast errors
MAD: the average of absolute forecast errors.
RSFE
TS
MAD
Follow exhibit 13.9 1nd 13.10, page 555
Linear Regression Analysis
Discussed earlier. The formulae including
those of standard error were also given.
Follow exhibit 13.9 1nd 13.10, page 555
Decomposition of Time Series
Time series is a chronologically ordered data set.
Trend, seasonal, cyclic, autocorrelation, random.
Decomposition: Identifying and separating into components
Easy to identify
Trend, Seasonal
Difficult to identify
Cyclic, autocorrelation, random
Seasonal Variation
Additive seasonal variation
Forecast including trend and seasonal = Trend + seasonal factor
Multiplicative seasonal variation
Forecast including trend and seasonal = Trend x seasonal factor
See Exhibit 13.15 B, Page 561
Seasonal Factor (or Index)
Seasonal Factor (or Index)
Amount of correction needed in a time series to adjust for the
season of the year.
Let us follow examples 13.3 and 13.13.14 page 562
Forecast including trend and seasonal
factor for period t (FITSt):
FITSt = Trend . seasonal factor
Decomposing Time Series
Decomposing using Least Square Regression
I.
Decompose the time series into its components
a.
b.
c.
Find seasonal components
Deseasonalize the demand
Find Trend component
II. Forecast future values of each component
a.
b.
Project trend component into the future
Multiply trend component by seasonal component.
Scheme for Decomposition
Step 1: Determine the seasonal factor
Step 2: Deseasonalize the original data
Step 3: Develop regression line for deseasonalized data
Step 4: Project for the period of forecast
Step 5: Create final forecast using seasonal factor.
Casual Relationship Forecast
Where one occurrence causes another.
1. Find causes
2. Develop a relation ship
3. Make forecast
Follow Example 13.5 page 567
Multiple Regression Analysis
S B Bm ( M ) Bh ( H ) Bi ( I ) Bt (T )
S= Gross sale for the year
B= Base sales (starting point)
M= Marriages during the year
H= Housing starts during the year
I= Annual disposable personal income
T= Time trend
Focused Forecasting
1.
2.
3.
4.
5.
Whatever we sold in the past three month is what we will
probably sell in the next three months.
What we sold in the same three month period last year, we will
probably sell in that three month period this year. (This would
account for seasonal effects.)
We will probably sell ten percent more in the next three months
than we sold in the past three months.
We will probably sell fifty percent more over the next three
months than we did for the same three months of last year.
Whatever percentage chance we had for the past three months
this year compared to the same three months last year will
probably be the same percentage change that we will have for the
next three months of this year.
Web-Based Forecasting
Collaborative Planning, Forecasting and Replenishment
(CPFR)
Web-based tool for trading partners (food, apparel etc.)
Step 1: Creation of a front-end partnership agreement.
Step 2: Joint business planning.
Step 3: Development of demand forecast.
Step 4: Sharing forecasts.
Step 5: Inventory replenishment.
Home Work 3
Due Date April 11, 2010
Problem Nos. 4, 5, 6