GLOSSARY OF FORECASTING TERMS
ADDITIVE SEASONALITY:
When the seasonal effect on demand is not a function of
the size of demand.
For example, if demand is always 10 units lower than
normal in a given period, regardless of whether the normal size of demand
is 250 or 500, then additive seasonality is present.
BASE (OR HORIZONTAL, OR LEVEL) COMPONENT:
The part of the pattern in a time
series that is always present, even when no other part of the pattern is
present.
It can be thought of as the average of the time series, or
alternatively, as the level from which the time series (the historical
demand) grows or declines.
BIAS:
The average forecast error, allowing positive and negative forecast
errors to cancel each other out.
The optimal bias is zero, although a
bias of zero does not indicate perfect forecasting.
A positive bias
implies a tendency to underforecast, while a negative bias implies a
tendency to overforecast.
CAUSAL MODELS:
Techniques that derive forecasts by identifying the "causes" of
fluctuations in demand, such as the number of competitors, general
economic indicators, population growth, advertising expenses, prices, etc.
The most popular of these methods is multiple regression.
It should be
noted that the term "causal" is somewhat of a misnomer, since these models
only identify whether demand is related
to factors such as the ones listed
above.
They cannot actually prove that any factors cause demand to
fluctuate in any way.
CHANGE IN SEASONALITY:
A permanent change in the seasonal (i.e., yearly)
pattern in a time series.
Such a change is best identified by plotting
the residuals (forecast errors) that would result if your forecasting
model takes into account all relevant components (i.e., base, trend,
randomness) except seasonality (i.e, the forecast is unseasonalized).
By
closely examining the seasonal pattern that will show up in the residuals
from year to year, any significant changes can be identified.
To be
reasonably confident that a change is "permanent," the new pattern should
hold true for at least two years.
The response to a change in seasonality
is generally to ignore the full years of data before the change when
building the forecasting model.
CHANGE IN TREND:
A permanent change in the general movement upward or downward
in a time series.
Such a change is best identified by plotting the
residuals (forecast errors) that would result if your forecasting model
takes into account all relevant components (i.e., base, trend,
seasonality, randomness), or in other words, the forecast is trend and
seasonally adjusted.
A change in trend is generally identified as a
relatively sharp, one-time change of direction in the residuals.
If the
residuals change direction more than once over a period of several years,
and/or the change of direction(s) is(are) follow more of a rolling wave-
like curve, then it is possible that the time series has cyclicality, not
a change in trend.
The response to a change in trend is generally to
ignore the full years of data before the change when building the
forecasting model.

This
** preview**
has intentionally

**sections.**

*blurred***to view the full version.**

*Sign up*
This is the end of the preview.
Sign up
to
access the rest of the document.