sales using exponential smoothing
Sales forecasting is a difficult job for most products and services.
Business forecasting for markets where there are fixed demand levels or a predictable trend or predictable
competition are few and far between.
forecasting typically takes place where the demand for products and
services are not easily predictable – which means good business forecasting is needed in order to manage
production, stock levels and cash flow.
Many sales forecasts are prepared on the basis of past sales. Past events
(i.e. sales) are typically analysed in the following way:
Overall trends are identified
Seasonal or cyclical factors are identified that affect sales - for
example seasons or regular major events.
'One-off' events are isolated - for example hurricanes or elections
In-house activities are isolated - for example actual or intended
So, before beginning to use budgeting and forecasting software -
especially when using statistical forecasting methods and related sales forecasting software - a sales forecast is
Smoothing out the odd-ball factors that might bias sales forecasting
software systems - for example using moving averages)
Taking account of things like seasonal variations
Clarifying the impact of management actions - for example a marketing
Despite all of these preparatory techniques sales forecasting techniques
hardly ever produce an accurate sales forecast.
And unless people in every part of the company understand that forecasting
sales using tools like sales forecasting software is inherently inaccurate - but it is way better than
no sales forecast at all - people
begin to try and cover all of the bases to avoid uncertainty.
For example - unless people are operating within a rational sales
forecasting framework the Sales Department may 'tweak' their estimates so that they avoid running out of stock. But
of course if they do this, unless they are lucky, it means cash is
unwittingly poured into inventory. And production
costs are likely to rise - in order to meet higher demand people may have to work overtime, etc.
To avoid such problems, relying totally on statistical forecasting software
to estimate demand is not always the best option.
and Forecasting Software
A technique called Focus Forecasting has been successful for many
companies - in fact a client of mine was the first ever users of the software
in New Zealand and often described how it helped
transform how his company went about forecasting sales.
Focus Forecasting was developed about 25 years ago by Bernard T Smith. The
thing I personally like about it is the commonsense approach - plus I guess I am biased because of my personal
exposure via my friend in Auckland.
Forecasting sales with Focus Forecasting technique is pretty straightforward. You take historical sales figures (with distortions removed) and use a variety of sales forecasting
techniques (e.g. average, moving annual total, weighted averages, etc.) to predict the last 2-3 months of
actual sales. You then use the sales forecasting technique for your future sales forecast.
By forecasting sales for each product every month your overall business
forecasting is grounded in reality and pragmatism - major sales forecasting errors are limited to individual
products and overall your demand forecasting is likely to be better that creating budgets based on pure
statistical forecasting software.
Notwithstanding the above comments, forecasting sales using exponential
smoothing techniques has it's place.
The example problem below focuses on using these exponential smoothing techniques as a general
forecasting sales software solution.
But the same techniques could be used for pretty much and business
forecasting scenario - exponential smoothing is not just limited to forecasting sales.
Sales Forecasting using historical data to predict future sales -
software sales, medical sales, car sales, real estate sales, estate sales,
auto sales, home sales and so on.
smoothing is one technique which you can use to predict events in the future
by studying events in the past. By employing weighted averages to
"smooth" past values, it lets you forecast the value in the next
The basic model for exponential smoothing is:
Pt+1 = Pt + * ( Yt - Pt ) where
forecast value at period (t+1)
Pt = forecast value in period t
Yt = actual value at period t
(Alpha) = the smoothing constant
If Alpha is set to 1, the forecast for the next period is based entirely on
the actual value from the last period. If Alpha is set to 0, the actual
value from the last period is completely ignored. Since neither of these
cases will provide much insight into future data, we'll constrain Alpha to
be between .01 and .99.
In order to minimise costly overstocking and inventory holding, your retail
outlet needs useful forecasts of future sales.
For this simple exponential smoothing problem, you have sales data (in
$1,000's) for eight months. You need to find Alpha, the smoothing constant,
that minimises the sum of the error - which in this case is the difference
between the actual and forecast sales for each period.
The objective for this sales forecasting technique is to determine projected
sales and the Alpha smoothing constant while minimising the squared error.
Model | Back to Top | Next Model