K9Queen --
First, it's good to see you here again. Some of the issues with sales
forecasts, in particular seasonality and growth, are indicated in
other questions that you've posed. I'll try to provide an experienced
overview here, much of it from having run forecasting for a Fortune
500 personal computer company.
1. MOVING AVERAGES
The simplest moving average forecast is simply an average of previous
periods' demands (D) over a number of time periods (t):
F (t+1) = 1/n [Dt + D(t+1) + . . . + D(t+1-n)]
University of Colorado School of Business & Administration
"Demand Forecasting: Time Series Models" (Prof. Stephen R. Lawrence)
http://www-bus.colorado.edu/faculty/lawrence/Tools/FORECAST/6
In a market where sales growth is steady and seasonality is weak, it's
not a bad way to forecast. However a simple moving average doesn't
work well with seasonality -- so a major increase in sales during a
particular period would be under-forecast. If the 30-day Christmas
season were a major period for a business, there would be some of the
previous year's demand in the forecast -- but growth during the year
would have muted its effects.
Prof. Lawrence notes several other problems:
? older demand may be missing. If the economy has been weak, it will
result in under-forecasting activity during a recovery period. During
the last recession car makers correctly predicted that there was a
latent demand building that would be shown during a recovery. If
they'd dropped older numbers in the time series, they'd have been
un-prepared to meet demand.
? forecasting thousands of products may make this unworkable, as
demand may be highly-volatile by stock-keeping unit (SKU).
? an influx of new products for which there is no history creates major problems.
? all past observations are treated equally, so the effect of large
customers or contracts appearing or disappearing is not account for.
? Yet another common problem is failing to account for inventory
already in a customer's hands. One of the problems that chipmakers
are currently worried about is how quickly Chinese cellphone
manufacturers will work through their parts inventory:
CNN.com
"Texas Instruments Beats the Street" (July 21, 2003)
http://money.cnn.com/2003/07/21/technology/texasinstruments/
2. TREND FORECASTS
In companies experiencing steady growth -- say a restaurant chain
adding 5% new stores every quarter -- or in judging sales off an
installed base, more recent numbers can be given greater weight.
In the case of the restaurant chain, old stores will have a sales base
that rises to maturity and be both largest (per store) and steadiest.
Stores that were added within the past 12 months will still be
growing, as they rise to maturity. And brand-new openings will be
adding brand new sales -- making a trend forecast workable for each
category.
Yet another case would be parts sales, where the installed base is
growing. Imagine, for example, IBM's Personal Computer business with
growing sales of Thinkpad portable computers. For each model, the
most-important predictor of sales by the Maintenance & Repair
organization will be the number of systems now in the market.
3. RETAIL FORECASTS
The handling of seasonality is displayed very well in this presentation:
The University of Sheffield
"Forecasting" (Prof. Keith Ridgway, undated)
http://www.shef.ac.uk/uni/companies/msmu/forecasting1.htm
In order to determine sales trends, you'll need 4 quarters worth of
data, at a minimum. Several years are better, as seasonal impact will
vary slightly year-to-year.
Establish a 4-quarter moving average of sales (4Q-MA) which consists
of 2 previous quarter and 2 forward quarters. The moving average will
catch any trends. The numbers that Ridgway's example uses are:
MA Q1/1985: -- (there's not 2 earlier quarters)
MA Q2/1985: 270
MA Q3/1985: 292
MA Q4/1985: 305
MA Q1/1986: 307
Now we can look at seasonality by looking at changes from
quarter-to-quarter -- comparing the quarter's ACTUAL SALES to the mean
of moving averages:
Q3 SEASONALITY = Q3 Sales / (MA Q2 / MA Q3) * 0.5
Q3 SEASONALITY = 300 / (272 + 290) * 0.5 = 1.07
So, Q3 has a seasonality of 7% built-in.
4. REALTOR'S FORECAST
Much of the answer to this question depends on the specific market you
are serving as a Realtor (it's a trademark, you know:
http://www.realtor.com/Default.asp?poe=realtor). A Realtor largely
involved in the sales of new homes will have a much larger variable
for something like "number of new homes completed."
Nonetheless, we'll set up a model for Home Sales (HS) that's based on
New Homes (NH) and Existing Homes (EH). It assumes that we have a
stable sales history of NH and EH over many quarters. Also that
Interest rates (I) are a key variable for both classes, as virtually
all purchases are financed. For New Home (NH) sales, we're also
concerned about their average cost, so we'll insert a factor called
Delta (D), which is the difference between New Home Price (NHP) and
Average New Home Price (ANHP). Restated, D = (NHP - ANHP):
HS = A*I*NH + B*D*NH + C*I*EH
The variables here are all positive -- except for one instance. We
don't anticipate 0 sales in any category, even with high-priced homes.
Rather, we'd simply expect sales to drop as interest rates rise or
more expensive homes are marketed:
A: declines with interest rate increases, is above 1.0 if interest
rates decline. The factor may be fairly close to linear if homes are
financed at the 80-90% level.
B. declines with an increase in (NHP-ANHP) due increases in the base
cost. However, if NHP-ANHP is negative (meaning our houses are less
costly than the average), the sign on B would be negative. In this
case we'd see an increase in sales -- and the formula can't produce
negative house sales.
C. EH homes come from a large pool within the community. Of course if
our particular pool was more expensive, it would change the
regression, but we work hard to represent the entire community. As
with A, C will change inversely with interest rates. We've made this
factor separate because of difference in marketing of new homes and
financing options for NH vs. EH.
Google search strategy:
"sales forecasts" + issues
"sales forecasts" + "moving average" + issues
Best regards,
Omnivorous-GA |
,
0 Comments
)