Forecasting

1
Forecasting

• Today
• Basics about forecasting
• Forecasting system
• Forecasting methods
• Time series forecasting

2
Why learn about forecasting?

• Its a tricky business. Set the forecast too
  high, and you end up with warehouses full of
  inventory that can spoil, get damaged, marked
  down, or just thrown out. Set the forecast too
  low, and there are stock-outs at the retail
  level. Retailers may be losing as much as 11
  percent of revenues because of out-of-stock
  situations.
• Semiconductor manufacturing requires many
  hundreds of operations per product, a few hundred
  tool groups and weeks of manufacturing time.
  Tools cost up to 10 million, and the lead time
  for purchasing a tool may be up to a year.
  Capacity planners use long-term demand forecasts
  to determine the tools to purchase

3
Who uses forecasts of demand?

• Purchasing short-range to determine what should
  be purchased and when
• HR use short-range forecasts to determine
  workforce levels
• Operations uses demand forecasts for decisions
  such as
• short-range scheduling of workers
• short-range inventory decisions (production
  levels)
• medium-range production planning
• long-range capacity planning (facility and
  technology)
• long-range location planning
• supply chain management share with suppliers
• Finance use medium-range cash budgeting and
  long-range forecasts to project needs for capital
• Marketing develops demand forecasts that are
  used for many different types of planning uses
  forecasts to make decisions about promotions
  (medium-range) and new product development
  (long-range)
• MIS design and implement forecasting systems

4
Forecasting Horizons
Purchasing Job scheduling Work force levels Job
assignments
New product Capital expenditures Location or
expansion RD
Sales Production Cash budgeting
3 years
Now
12 months
3 months
5
Forecasting Demand (not sales)

• In a make-to-stock environment, customers
  expect stock to be available when they submit
  their purchase orders. Some customers, referred
  to as Just-in-Time (JIT) customers, will cancel
  any orders for items that are not immediately
  available. These canceled orders are not
  considered demand and can be overlooked by
  planners who base forecasts on demand. This lack
  of visibility results in low forecasts on already
  backordered items, making it difficult if not
  impossible to properly service all customers
  requesting those items.
• APICS Magazine, Jan 2007

6
Forecasting System

• System Design
• Determine use of forecast
• Select items to be forecasted
• Determine the time horizon
• Select forecasting method (model)

• System Use
• Gather data needed to make forecast
• Make the forecast
• Validate and implement results
• Monitor forecast accuracy

7
Realities of Forecasting

• Forecasts are seldom perfect
• Most techniques assume underlying stability in
  business environment
• By nature, some forecasts are more accurate than
  others
• Aggregate more accurate than for individual
  products
• Shorter more accurate than longer horizon
• Ideal to have
• Forecasts that are reliable and consistent
• Short forecasting horizon

8
Forecasting Approaches

• Quantitative Approaches
• Time series
• naïve
• averaging (moving average, weighted moving
  average, exponential smoothing)
• trend
• Associative models
• Useful if
• relevant data available
• underlying business environment will continue

• Qualitative Approaches
• jury of executive opinion
• Delphi method
• sales force composite
• consumer market survey
• Useful if
• quick turnaround needed
• relevant data not available
• very long horizon
• substantial changes
• new products

9
Time Series Forecasting

• Time series - sequence of evenly spaced (over
  time) data points

Trend? Seasonality? Cycles? Random variation?
10
Time Series Forecasting
What can we forecast? Trend trend projection
Seasonality predictable pattern, SI Cycles
NO Random variation cant be predicted but
use averaging to smooth out
short-term variations
Demand for riding lawn-mowers
summer 1
summer 2
summer 3
11
Example Time Series Data
12
For fun

• Form teams
• Count by 5 around the room
• On the next slide, click the number you counted

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Forecast Accuracy

• Forecast error
• how close an individual forecast is to actual
  demand that occurred
• Measures of forecast accuracy
• how well forecasting model predicts actual demand
  that occurred
• Measures
• Mean absolute deviation
• Mean squared error
• Mean absolute percent error

20
Forecast Accuracy

• Forecast with naïve approach

21
Forecast Accuracy

• Forecast error

22
Mean Absolute Deviation
Average of
the absolute value of the errors
Why use absolute values?
23
Mean Absolute Deviation
Average of
the absolute value of the errors
24
Forecast Accuracy

• Forecast error
• how close an individual forecast is to actual
  demand that occurred
• Measures of forecast accuracy
• how well forecasting model predicts actual demand
  that occurred
• How used?
• Monitor forecasts
• Select forecasting model

25
Forecast Accuracy

• Forecast error
• how close an individual forecast is to actual
  demand that occurred
• Measures of forecast accuracy
• how well forecasting model predicts actual demand
  that occurred
• How used?
• Monitor forecasts
• Select forecasting model

26
Selecting a quantitative forecasting technique

• Using past actual demand data, develop forecasts
  in the past using alternative models to be
  evaluated
• Determine the forecast accuracy of each model
  (MAD)
• Select the forecasting model that is most
  accurate (based on MAD)

27
(No Transcript)
28
Trend Projections

• Use if relatively strong positive or negative
  trend
• Fit line to historical data (prior actual demand)
  and projects that line into the future
• Simple linear regression
• Fit line to data to minimize squared error
• Variables
• y dependent variable
• x independent variable
• Only one independent variable

29
Can you see a line?
30
Trend Projection

• Regression analysis in Excel
• Tools Data Analysis Regression
• Indicate data to be used in regression example
• Highlight dependent variable data
• Highlight independent variable data

31
Excel Regression Output
32
Trend Projections
33
Excel Regression Output
34
Fit of Regression Model

• Coefficient of Correlation (r) -1 lt r lt 1
• Measure of strength of relationship between
  variables

r closer to 1
r closer to -1
r about 0
35
Fit of Regression Model

• Coefficient of Correlation (r) -1 lt r lt 1
• Measure of strength of relationship between
  variables
• Coefficient of determination (r2) 0 lt r2 lt 1
• Percentage of variation in dependent variable
  explained by the independent variable(s)

36
Seasonality What we do (in general)

• Seasonality recurring pattern of ? and ? in
  demand depending on season
• You can predict when forecast should be higher or
  lower than average
• Find average (with averaging approach or trend)
  and adjust up or down, based on which season

Here season quarter
Others season Month Week
Day Hour
37
Adjusting for Seasonality

• De-trend the data (to avoid distortion of trend)

Demand for riding lawn-mowers
summer 1
summer 2
summer 3
38
Adjusting for Seasonality

1 cycle
Demand for riding lawn-mowers
1 season
summer 1
summer 2
summer 3
39
Adjusting for Seasonality

• Calculate the average for each season
• Calculate the overall average
• Compute seasonal index for each season

Demand for riding lawn-mowers
summer 1
summer 2
summer 3
40
Adjusting for Seasonality

• Calculate the average for each season
• Calculate the overall average
• Compute seasonal index for each season

Demand for riding lawn-mowers
summer 1
summer 2
summer 3
41
Adjusting for Seasonality

• Forecast the average
• With trend - trend projection
• Without trend - Estimate total demand
• for next year and divide by of seasons

Demand for riding lawn-mowers
y a bt
summer 1
summer 2
summer 3
summer 4
42
Adjusting for Seasonality
Compute the seasonally adjusted trend projection
the final forecast Ft average SI Ft a
bt SI
Demand for riding lawn-mowers
summer 1
summer 2
summer 3
summer 4
43
(No Transcript)
44
Associative Model

• Y a b1X1 b2X2
• ex hotel occupancy f(gas price relative to
  airline price, economy, advertisements)
• Compared to time series
• Dependent variable demand
• Independent variable(s) ???

45
Associative Model Example
Apts Leased 6 9 8 16 13 6 10 6
Ads 15 25 22 35 25 9 15 20
Period 1 2 3 4 5 6 7 8

• Sales manager of a large apartment rental complex
  wants to predict demand for apartments. What
  could be used as the independent variable?
• time
• of ads placed in previous month

46
(No Transcript)
47
(No Transcript)
48
Monitoring Forecasts

• Why is monitoring of forecasts needed?
• Changes in the environment such as
• Inaccurate forecasts can be costly
• When is it done?
• After each actual demand is known
• How is it done?
• Tracking signal