Demand Management and Forecasting in a Supply Chain

1
Demand Management and Forecasting in a Supply
Chain

• Demand forecasting is magic or art, and leave
  everything an chance !?

2
Learning Objectives

• The Role of Forecasting
• Identify components of a demand forecast
• Time series forecasting
• Estimate forecast error

3
The Role of Forecasting

• Basis for all strategy and planning
• -- pull process level of capacity
• -- push process level of production
• Strategy long-term (ys)
• Aggregate Planning medium-term (mo)

4
Characteristics of Forecast

• Forecasts are always wrong. Should include
  expected value and a measure of error (standard
  deviation).
• --- same E(X), but different SD(X)
• Long-term forecasts are less accurate than
  short-term forecasts uncertainty
• Aggregate forecasts are more accurate than
  disaggregate forecasts
• --- forecast GDP vs. profit of a company

5
When to Make Forecast?

• It depends on the supply chains response time.
• For example, if a mfger knows it takes the supply
  chain 6 months to respond to an order, he must
  make a forecast 6 months before the product is
  needed.

6
What Factors Need to be Considered?

• Past demand pattern (seasonal change)
• Planned ads or marketing efforts
• Planned price discounts
• Actions competitors have taken
• State of the economy

7
Forecasting Methods

• Qualitative subjective, human judgment and
  opinion, little historical data.
• --- New industry computer, Internet
• Time Series past demand history is a good
  indicator of future demand. One of the most
  common and effective methods.
• Static
• Adaptive

8
Forecasting Methods (contd.)

• Causal demand is correlated with some factors,
  like price, location, promotion, quality. Run
  regression.
• Simulation what if in different scenarios?
• --- Airlines simulate customer buying behavior to
  forecast demand for higher-fare seats when there
  are no seats available at the lower fares.

9
Time Series Forecasting (TSF)

• Components of an observation
• Systematic component Random component

10
Forecasting methods

• Static estimate level, trend, seasonality of the
  systematic component of demand once and then do
  not update these estimates even as new demands
  are observed.
• Adaptive update the estimates of level, trend,
  seasonality of systematic component of demand
  after each demand is observed
• Moving average (level)
• Simple exponential smoothing (level)
• Holts model (level, trend)
• Winters model (level, trend and seasonality)

11
Static Forecasting

• Deseasonalize demand (centered moving average)
• Run regression to estimate level and trend
• Estimate seasonal index (factors)
• Make forecast
• FtnL(tn)TStn
• t is the current period that you make
  forecast
• tn is the period you are forecasting

12
Static Forecasting (contd.)

• Given 5 year quarterly data of the demand of
  product A, make forecast of the demand for the
  next four quarters using static methods.

• Example 1
• Note refer to static forecasting.xls in the
  folder forecasting to locate the excel
  spreadsheet.

13
Adaptive Forecasting

• Moving Average
• Simple exponential smoothing
• Trend-corrected exponential smoothing (Holts
  model)
• Trend- and seasonality-corrected exponential
  smoothing (Winters model)

14
Adaptive Forecasting (contd.)

• Moving Average
• Systematic component of demand level
• Lt(DtDt-1.Dt-N1)/N
• FtlLt
• Example 2 (refer to moving average.xls in the
  folder forecasting to locate the excel
  spreadsheet.)

15
Adaptive Forecasting (contd.)

• Simple exponential smoothing
• Systematic component of demand level
• --- initial level L0avg (demand in all periods)
• --- given an alpha (a), new level
• Lt1aDt1(1-a)Lt
• --- forecast demand FtlLt
• Example 3 (refer to simple exponential
  smoothing.xls in the folder forecasting to
  locate the excel spreadsheet.)

16
Adaptive Forecasting (contd.)

• Holts Model (trend-corrected ES)
• Systematic component of demand
• level trend
• --- Initial level and trend are given by
  regression
• --- given alpha (a) and beta (b),
• Lt1aDt1(1-a)(LtTt)
• Tt1b(Lt1-Lt)(1-b)Tt
• --- forecast demand FtnLtnTt
• Example 4 (refer to Holts model.xls in the
  folder forecasting to locate the excel
  spreadsheet.)

17
Adaptive Forecasting (contd.)

• Winters model (trend- and seasonality- corrected
  exponential smoothing)
• Example 5 (refer to Winters model.xls in the
  folder forecasting to locate the excel
  spreadsheet.)
• It is a little bit complicated. We can discuss in
  my office hour if anybody has interest.

18
Error measures

• What is error?
• -- The difference b/w forecasting and real demand
• Et Ft - Dt
• How to measure error?
• -- Mean Absolute Deviation (MAD)
• -- Mean Squared Error (MSE)

19
Error Measurement (contd.)

• -- Mean Absolute Percentage Error (MAPE)
• -- Bias
• -- Tracking Signal bias/MAD
• Use these error measures to see which model is
  more accurate.

20
Summary

• Why we need forecasting?
• Time series forecasting
• What is the next step after forecasting?
• Questions?