Forecasting

Demand Estimation ..

There is only one method....





























GUESS !






















See also the article "The Future of Spending" at American Demographics. January 1995


What is the purpose of forecasting?

  1. To Plan for long term

  2. To Plan for short term

What do we want to forecast?

Just about anything.
Number of customers
Unit Sales
Dollar Sales
Market Share (you should know what this is by now.)
Rate of adoption

Educated guessing

With any forecasting, the important thing to remember is that no one forecasting method is perfect, and therefore you should always use more than one forecasting method and compare their results.

Do not be lulled into thinking that just because this method has worked pretty well for a few periods that it will continue to work well. It may assume a state of the world that hasn't changed during the time that it has worked, but which will not be reflected in forecasts if it does change.

A second thing to remember about forecasts is that all forecasts should be subjected to a sensitivity analysis where changes in underlying assumptions are made and the resulting changes in forecasts compared.

Note that Sales forecasts are the base number used for budgeting by accounting. Accountants should remember that these things are at best good guesses, and should do sensitivity analysis on their budgets based on different actual sales results.

Educated guessing -- Heuristic methods

(rules of thumb)
For total demand
Total $ales = #buyers X quan X price

Chain ratio method (adjusting percent)

For areas (if whole country, total)

Market Buildup (often called SIC)

Examples of NAISC (SIC)Codes   
Product Names: Apparel and similar products(230000);
Men's and boys' jeans(232549);
Women's, misses' and juniors' jeans and pants(233952);
Apparel and accessory stores(560000)

Market Factor Index Methods (e.g. Buying Power Index or BPI from Sales and Marketing Management Magazine)

Chain ratio method a heuristic

Example of Chain ratio method
(remember it is only as good as its links)

Estimating the number of

potential female romantic partners for a Freshmen male

last updated Fall 2001




















12,409 students at WWU
x .56 female










leaves 6,969 females
x .8 not married










leaves 5,575 females eligible for marriage










x .5 already attached at any given time
leaves

2,788 potential romantic partners












x .96 potentially interested
leaves

2,676 potential romantic partners







But!!!












Only .01 of those are actually interested in freshmen males





leaves 26 or 27 Western women actually interested in freshmen males

(this might call for a sales philosophy --- marketing)

Market factor index methods

Any method that makes some kind of index by combining various market place information.

BPI Buying Power Index

(one well known example)
Published annually by Sales and Marketing Management Magazine For any given metropolitan area, BPI is calculated as a weighted sum of three factors:

BPI is a Percentage of national buying power =

.5 x percent nat. effective buying income in the area
+
.3 x percent nat. retail sales in the area
+
.2 x percent nat. population in the area

Caveats:
applies only to retail operations and mass market consumer goods
assumes potential sales are a function of purchasing power
assumes that people purchase the goods in question in the same ratio everywhere

Subjective methods

Jury of Experts
Leading indicators
Salesforce Composite (popular)
   - salesmen have other motives
   - may not know
Buyer intention
   - intentions don't always lead to sales

Leading Indicators
-- changes in these happen first in a dynamic system

An undervalued leading indicator is the Consumer Confidence Index. published monthly by the Conference Board

Leading Indicators can predict turns in a series.

Every issue of Business Week has a listing of leading indicators.

Objective (aka Quantitative) methods of Forecasting

Theory based multiple regression
fit a model with theoretically relevant variables (cross sectional data - one point in time)

Time series based methods

A time series consists of the combination of the following components:
Correction for seasonality
Nearly all businesses are seasonal.
Therefore, All forecasts should be made with adjustments for seasonality. There is more than one way to approach seasonality, but the most robust is outlined below.
1) Deseasonalize the data.
  • Makes the lumps smooth out
  • Preserves any trend changes
  • Each de-seasonalized point is the ratio (actual data / index)
  •    where the index is (average of this particular period / grand average of all relevant periods)

2) Make a forecast in deseasonalized terms
Choose one or more methods.

3) Reseasonalize the forecast to an actual expected number given the period.


Naive methods
    same value as last time
    same change as last change ( Naive or simple Trend)
    change can be either a percentage change, or an absolute one

Moving averages
An average of a fixed number of prior periods.
To forecast more than one period into the future, use forecasts for interim periods as data. [n.b. always use the deseasonalized data until the final reseasonalizing step.]

Exponential smoothing -- weighted moving averages (Operations Management 360)
Box-Jenkins -- input output models
Time-series (econometric) multiple regression. (Econ 309)
Factors to consider in choosing forecasting method.
-time available
-cost
-pattern of the data
-accuracy
-ability to predict turning points
-ease of intrepretation

Remember:
  1. Always use more than one forecasting method.
  2. Always do a sensitivity analysis that tests the effects of changes in the major assumptions of your method.
  3. Try to forecast within a range so contingency plans can be made.