Forecasting

1
Forecasting

• Anticipating service requirements over the short
  term
• Operational purposes
• Planning and scheduling resources
• Examples
• Manufacturing and distributing printers at HP
• Staffing levels (NS)
• etc.

2
Objectives

• What is forecasting
• What are the issues
• What are the tools

3
Forecasting

• Developing predictions or estimates of future
  values
• Demand volume
• Price levels
• Lead times
• Resource availability
• ...

4
Taco Bell
Feed the dog

• Labor is 30 of revenue
• Make to order environment
• Significant seasonality
• 52 of days sales during lunch
• 25 of days sales during busiest hour
• Balance staff with demand

5
Value Meals

• Drove demand
• Forecasting system in each store
• forecasts arrivals within 15 minute intervals
• Simulation system
• predicts congestion and lost sales
• Optimization system
• Finds the minimum cost allocation of workers

6
Forecasting System

• Customer arrivals by 15-minute interval of day
  (e.g., 1115-1130 am Friday)
• Fed by in-store computer system
• 6-week moving average
• Estimated savings Over 40 Million in 3 years.

7
Independent vs Dependent

• Independent
• Exogenously controlled
• Subject to random or unpredictable changes
• What we forecast
• Dependent or Derived
• Calculated or derived from other sources
• Bill of Materials
• Related activity like packaging

8
Phenomena To Capture

• Randomness
• Trend
• Linear
• Exponential
• Seasonality

9
Randomness
10
Linear Trend
11
Exponential Trend
12
Seasonality
13
Seasonality
14
Forecasting Methods

• Qualitative or Judgemental
• Ask people who ought to know
• Historical Projection or Extrapolation
• Moving Averages
• Exponential Smoothing
• Regression based methods
• Neural Networks
• Econometric or Causal
• Regression
• Simulation

15
Moving Averages

• Simple, ubiquitous
• Reduce random noise
• One Extreme
• Predict next period This period
• Another Extreme
• Predict next period Long run average
• Intermediate View K period moving avg.
• Predict next period Average of last K periods

16
The Dow
17
Questions

• Whats the forecast for 6 months out?
• Will a shorter span always be better?
• Is moving average a good method here?

• Which will handle this better?

18
Which is better?

• Average Error
• 3-month 1490
• 6-month 619

19
Exponential Smoothing

• Moving Averages
• Equal weight to older observations
• Exponential Smoothing
• More weight to more recent observations
• Forecast for next period is a weighted average of
  
• Observation for this period
• Forecast for this period
• AlphaObservation (1-Alpha)Past Forecast

20
Initial Values

• First Observation
• Average of previous observations
• etc.

21
Which is which?
Alpha .01 or Alpha .2
22
Modeling Trend

• Holts Method
• Forecast is weighted combination of
• Current Observation
• Current Forecast for Next Period
• Forecast Trend as weighted combination of
• Current Trend in Forecasts
• (our estimate of trend)
• Current Forecast for Trend
• (differences in successive forecasts)
• Why this way?

23
With and Without Trend
24
Intels Trend
25
Exponential Growth

• Forecasted Sales ?e?t
• Natural Log of Forecasted Sales
• Ln? ?t
• Thats linear growth
• Take Natural Log of observations
• Forecast Natural Log of Sales
• Convert back to Forecast of Sales

26
Seasonality

• Deseasonalize the data
• Forecast
• Seasonalize the forecast
• Seasonal Factors
• Ratio of Actual to Average
• Updated with Exp. Smooth.
• Weighted combination of
• Actual/Deseasonalized Forecast
• Current Forecast of seasonal factor

27
Seasonality

• Deseasonalized Forecast
• Alpha(Actual/Seasonal Factor)
• (1-Alpha)(Past Deseasonalized Forecast)
• Seasonalized Forecast
• Deseasonalized Forecast Seasonal Factor
• Updating the seasonal factors
• Gamma (Actual/Deseasonalized Forecast)
• (1- Gamma) Previous estimate of seasonal factor
  

28
Initialization and Factors

• Level
• Trend
• Seasonality

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
Regression-Based Models

• Time Series
• Find best fit of proposed model to past data
• Project that fit forward
• Econometric
• Find exogenous factors driving value
• Weather
• Economic factors
• Rainfall
• Develop formula for (future) values based on
  these factors

34
Example

• Locating a new retail store
• Build a model of sales volume (profitability)
  based on existing stores
• Population
• Wealth
• Competitors
• Access
• Predict sales for new store with this model

35
Forecast Error

• Building a Forecast
• Fit to historical data
• Project future data
• Forecast Error
• How well does model fit historical data
• Do we need to tune or refine the model
• Can we offer confidence intervals about our
  predictions

36
Measuring Forecast Error

• MAD or MAE
• average of the absolute errors
• RMSE (root mean square error)
• Square root of average squared error
• Sample std deviation
• Differs by 1 degree of freedom (N-1)
• MAPE (mean absolute percentage error)
• Average absolute ratio of error to actual

37
Issues

• Forecasting is a necessary evil, try to reduce
  the need for it.
• Complexity costs money, does it provide better
  forecasts?
• Aggregation provides accuracy, but precludes
  local information
• Forecast the right thing

38
What about HP?

• Why are forecasts bad in Europe?