Forecasting Demand for Services

1
(No Transcript)
2
Learning Objectives

• Recommend the appropriate forecasting model for a
  given situation.
• Conduct a Delphi forecasting exercise.
• Describe the features of exponential smoothing.
• Conduct time series forecasting using exponential
  smoothing with trend and seasonal adjustments.

3
DEMAND MANAGEMENT

• DEMAND MANAGEMENT
• marketing
• finance
• operations
• human resources

4
TYPES OF DEMAND

• Independent or dependent demand
• Demand for outputs or inputs
• Aggregate versus item demand

5
TIME DIMENSION

• short term- 15-30 days
• medium term -6-12 months
• long term - 10-20 years

6
LEAD TIME REQUIREMENTS

• Make to stock - short lead time
• Make parts-to-stock/assemble -to-order industry
• Make-to-order industry - long lead time

7
Data sources

• Marketing projections
• Economic projections
• Historical demand projections

8
Forecasting

• FORECASTING FOR SUPPORT SERVICES
• Hiring
• Layoffs and reassignments
• Training
• Payroll actions
• Union contract negotiations

9
continued

• FORECAST ERROR
• Et Dt - Ft
• Et error for period t
• Dt actual demand that occurred in period t
• Ft forecast for period t
• Period t depends on the purpose of the forecast

10
Cont

• MEAN ABSOLUTE DEVIATION (MAD) simplest way of
  calculating average error
• MAD S?Et?
• n

11
HISTORICAL DEMAND PROJECTIONS

• By time series we mean a series of demands over
  time. The main recognizable time-series
  components are
• Trend, or slope, defined as the positive or
  negative shift in series value over a certain
  time period
• Seasonality, usually occurring within one year
  and recurring annually
• Cyclical Pattern, also recurring, but usually
  spanning several years
• Random Events explained, such as effects of
  natural disasters or accidents
• Unexplained, for which no known cause exists

12
Forecasting Models

• Subjective Models Delphi Methods
• Causal Models Regression Models
• Time Series Models Moving Averages Exponential
  Smoothing

13
NAIVE METHOD OF FORECASTING

• use the most recent periods actual sales
• jury of executive opinion
• prompted by lack of good demand data

14
MULTIPERIOD PATTERN PROJECTION

• MEAN AND TREND
• used when the historical demand lacks trend and
  is not inherently seasonal

15
continued

• SEASONAL often an item showing a trend also has
  a history of demand seasonality, which calls for
  the seasonal index method of building seasonality
  into a demand forecast
• Seasonal Index example in handout
• Seasonally adjusted trends example in handout

16
PATTERNLESS PROJECTION

• These techniques make no inferences about past
  demand data but merely react to the most recent
  demands.
• These techniques moving average, exponential
  smoothing, and simulation typically produce a
  single value, which is the forecast for a single
  period into the future.

17
Moving Average

• It is the arithmetic mean of a given number of
  the most recent actual demands
• 3 period moving average - exhibit 4-13 (handout)
• Mean absolute deviation (MAD) - exhibit 4-13

18
EXPONENTIAL SMOOTHING
Most widely used quantitative forecasting
technique smoothes the historical demand time
series assigns different weight to each periods
data lower to points further away Ft1 Ft
a(Dt - Ft) Ft1 forecast for period t1 a
smoothing constant Dt actual demand that
occurred in period t Ft forecast for period t
19
continued

• next period forecast last period forecast
  a(last period demand - last period forecast)
• the future forecasts are being adjusted for the
  forecast error in the last period
• exhibit 4-16 (handout)
• small a means each successive forecast is close
  to its predecessor - stable demand
• large a means large up and down swings of actual
  demand - unstable demand

20
continued

• note - how the exponential smoothing extends back
  into the past indefinitely, that is, the
  adjustments made in the past are carried forward
  in a diminishing manner
• problem of startup forecast
• moving average and exponential smoothing are
  based on the assumption that past demand data is
  the best indicator of the future
• problem in exhibit 4.16 (handout)

21
ADAPTIVE SMOOTHING

• used as an extension of exponential smoothing
• forecasters may adjust the value of smoothing
  coefficient a if cumulative forecast error gets
  too large, thus adapting the forecasting model to
  changing conditions
• running sum of forecast error is used for
  signaling, whether a needs to be changed
• TRACKING SIGNAL RSFE
• MAD
• If RSFE is getting larger in the positive
  direction, implying, that actual demand is higher
  than the forecasted demand, then you want to
  increase the next period forecasted value. This
  can be done by increasing the value of a and
  vice versa.

22
FORECASTING BY SIMULATION

• using distributions of each variable, simulated
  runs are generated - suggesting the forecasted
  values.
• forecast error is calculated by subtracting the
  actual demand from the forecasted demand
• CORRELATION
• REGRESSION

23
QUESTIONS TO PONDER

• 1. What are the purposes of demand management?
• 2. What are the short, medium, and long term
  purposes of demand forecasting?
• 3. How is forecast error measured? What are the
  limitations of this measure?
• 4. What is a time series? What are its principle
  components?
• 5. How is one forecasting model compared with
  another in selecting a model for future use?
• 6. Make sure you know how to do the problems.