CHAPTER 15:  FORECASTING

Forecast - extrapolation of the past into the future, generally based on objective computation using data.

Prediction - conjecture or use of subjective knowledge, or experience, to arrive at an estimate of the future.

Why forecast?  Major planning premise for all kinds of plans.

If you cannot alter demand to fit your supply, then you must adjust your supply to fit the demand—you hope!  To take a peek at demand, we must forecast.

Types of forecasts:

a.  Technological Forecast - estimate rate of technological progress.

b.  Economic - expected future business (economic) conditions for         world/country/state/industry/locality/the firm/etc...

c.  Demand forecast - expected demand for a firm's output in some future period.

Some Factors That Influence Demand:

Internal (we control)

1.  Promotion

2.  Quality

3.  Product attributes

4.  Reputation

External (little or no control)

1.  economic conditions

2.  competitors actions

3.  changes in society that influence tastes and preferences

4.  legal environment

5.  catastrophes

6.  random variation

5 step process in selecting model for use:

1.  determine forecasting objectives

2.  select a model

3.  assess proposed model

4.  apply the model

5.  control (monitor performance of model in meeting objectives)

            1.  Determine objectives:

For dependent or independent demand?

For what time frame?

LR<--MR-->SR plans

Desired level of accuracy?

Related question: to what extent is the forecast biased? 

 2.  Evaluate model:  Accuracy is important criteria, but so too can be things like ease of use.  Error can be random or bias.  We can compensate for bias, but not for random variation—after all it is random!  We measure forecast performance using measures like RSFE, MSE, MAD (most commonly used).

QUALITATIVE    

 

a.  delphi

 

b.  sales force estimation (grass roots)

 

c.  customer surveys

 

d.  executive opinions / panel consensus

 

          e.  outside opinions (think tanks, consulting groups, government agencies)

QUANTITATIVE

 

1.  intrinsic

    (time series)

                 A.  smoothing models

 

                    1.  simple moving average

 

                    2.  weighted moving average

 

                    3.  exponential smoothing

 

                    4.  naive

 

                  B.  decomposition models (time series decomposition)

 

             2.  extrinsic model (associative/causal models)

 

                  A.  simple regression

 

                  B.  multiple regression

3.  Select a model to use.

4.  Use the model to develop some forecast.

 

5.  Periodically evaluate the model.

        About Time Series Decomposition

1.   Break Down a time series into its component parts:

     Trend

     Cyclical

     Seasonal

     Random (Irregular)

2.   then describe them mathematically

3.   then putting them back together in developing a forecast for some future period(s).

Major assumptions:

1.  components are independent of each other

2.  forces shaping past patterns will shape future patterns in a predictable fashion.

                 Note:  Projections further into the future are more uncertain.