Date 5^th August 2002-Monday Lecture –2

 

Forecasting Techniques:

 

Characteristics of forecast:

 

1. Usually they are wrong!
2. A good forecast should include a range
3. Aggregate forecasts tend to be more accurate as errors cancel out

 

 

Var (x1+x2+…+xN / N)          =          1 _* Var X

                     E (x1+x2+…xN / N)             n      EX

 

4. Larger the forecasting horizon, less accurate the forecast.
5. Forecasting techniques are not to be used in exclusion of known information

 

Types of forecast:

 

1. Subjective forecast: e.g. asking car dealers to give numbers for future car sales
• Prone to errors since bias may come in
• The dealer may have an incentive to give wrong numbers if his pay is linked to his meeting the target sales

 

Surveys:

·        It is easy to introduce bias

• Difficult to get representative sample
• Jury of executive authority

 

Delphi Method:

·        Send questionnaire to experts

·        Collate responses and when strong conflicts are encountered the process is again repeated until a consensus is obtained.

 

 

2. Quantitative forecast: using numerical techniques to do forecasting

·        Causal method

·        Extrapolative or time series based methods

 

 Examples of business decisions that rely on forecasting:

1.      Inventory control

2.      Upgrade capacity

3.      Cash flow management

4.      Cost and revenue forecasting – tax planning

5.      Shop floor activity planning

 

Classification based on the time period of forecast:

 

1.      Short term forecast: (daily – weekly)

·        High level of detail needed

·        Due to high frequency of use, method should be relatively inexpensive

·        Large number of items to be forecasted (e.g. for production of cars – the various subparts needed) therefore method should be such that it requires only small amount of data, so that the data as a aggregate is manageable.  

 

2.      Medium term forecast: (monthly – bi-monthly)

·        Forecasts are typically aggregated by product type

·        Details not needed

·        Extra cost and effort can be employed

 

3.      Long term forecasts:

·        Forecasts inherently have greater uncertainty – low accuracy

·        Subjective expert inputs needed i.e. advice of experts is to be sought

 

Quantitative techniques:

 

Common forecasting situations one might encounter:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

                                                                 

                                                      Ma_t = Dt+Dt-1+….+Dt-m

                                                                              m

                                                      Ft+1(forecast for time t+1) = Ma_t (average demand)

Another option could be to give weights to demand from different periods. ∑ w_i =1

Another method is to start with an initial guess and to keep updating the initial guesses using actual demand to improve the forecasting

 

S_t = S_t-1 + α*(D_t – S_t-1)

S_t = αD_t + α(1-α)(D_t-1 ) + α(1-α)² +…….

 

How to compare two forecasting techniques?

 

Consider 2 methods for forecasting, F1and F2 given the actual data D at times 1…n

• (D_t,F1) 1 ≤ t ≤n
• (D_t,F2) 1 ≤ t ≤n

 

Compare mean square errors: 1  ∑ e_t²  where e_t = D_t- F_t here larger errors are penalized

                                                n          more

Compare mean of errors: 1  ∑ e_t   here the sum may be zero hence this method of

     n    comparison but the advantage is that if sum> or< zero it         tells the side on which more errors tend to occur

 

Compare mean of modulus of error: 1  ∑ |e_t| x 100  mean absolute % error

                                                           n