Predictive Planning Forecasting and

Statistical Descriptions

The topics in this section are for those who want to know more about the forecasting
methods and error measures used in Predictive Planning.

Classic Time-series Forecasting

Two primary techniques of classic time-series forecasting are used in Predictive
Planning:

• Classic Nonseasonal Forecasting Methods — Estimate a trend by removing

extreme data and reducing data randomness

• Classic Seasonal Forecasting Methods — Combine forecasting data with an

adjustment for seasonal behavior
For information about autoregressive integrated moving average (ARIMA) time-series
forecasting, see ARIMA Time-series Forecasting Methods.

Classic Nonseasonal Forecasting Methods

Nonseasonal methods attempt to forecast by removing extreme changes in past data
where repeating cycles of data values are not present.

Single Moving Average (SMA)

Smooths historical data by averaging the last several periods and projecting the last
average value forward.
This method is best for volatile data with no trend or seasonality. It results in a
straight, flat-line forecast.

Figure B-1 Typical Single Moving Average Data, Fit, and Forecast Line

Predictive Planning Forecasting and Statistical Descriptions B-1

Classic Time-series Forecasting

Double Moving Average (DMA)

Applies the moving average technique twice, once to the original data and then to the
resulting single moving average data. This method then uses both sets of smoothed
data to project forward.
This method is best for historical data with a trend but no seasonality. It results in a
straight, sloped-line forecast.

Figure B-2 Typical Double Moving Average Data, Fit, and Forecast Line

Single Exponential Smoothing (SES)

Weights all of the past data with exponentially decreasing weights going into the past.
In other words, usually the more recent data has greater weight. Weighting in this
way largely overcomes the limitations of moving averages or percentage change
methods.
This method, which results in a straight, flat-line forecast is best for volatile data with
no trend or seasonality.

Figure B-3 Typical Single Exponential Smoothing Data, Fit, and Forecast Line

Double Exponential Smoothing (DES)

Applies SES twice, once to the original data and then to the resulting SES data.
Predictive Planning uses Holt’s method for double exponential smoothing, which can
use a different parameter for the second application of the SES equation.
This method is best for data with a trend but no seasonality. It results in a straight,
sloped-line forecast.

B-2 Working with Predictive Planning in Smart View

 Classic Time-series Forecasting

Figure B-4 Typical Double Exponential Smoothing Data, Fit, and Forecast Line

Damped Trend Smoothing (DTS) Nonseasonal Method

Applies exponential smoothing twice, similar to double exponential smoothing.
However, the trend component curve is damped (flattens over time) instead of being
linear. This method is best for data with a trend but no seasonality.

Figure B-5 Typical Damped Trend Smoothing Data, Fit, and Forecast Line

Classic Nonseasonal Forecasting Method Parameters

The classic nonseasonal methods use several forecasting parameters. For the moving
average methods, the formulas use one parameter, period. When performing a
moving average, Predictive Planning averages over a number of periods. For single
moving average, the number of periods can be any whole number between 1 and half
the number of data points. For double moving average, the number of periods can be
any whole number between 2 and one-third the number of data points.
Single exponential smoothing has one parameter: alpha. Alpha (a) is the smoothing
constant. The value of alpha can be any number between 0 and 1, not inclusive.
Double exponential smoothing has two parameters: alpha and beta. Alpha is the same
smoothing constant as described above for single exponential smoothing. Beta (b) is
also a smoothing constant exactly like alpha except that it is used during second
smoothing. The value of beta can be any number between 0 and 1, not inclusive.
Damped trend smoothing has three parameters: alpha, beta, and phi (all between 0
and 1, not inclusive).

Predictive Planning Forecasting and Statistical Descriptions B-3

Classic Time-series Forecasting

Classic Seasonal Forecasting Methods

Seasonal forecasting methods extend the nonseasonal forecasting methods by adding
an additional component to capture the seasonal behavior of the data.

Seasonal Additive
Calculates a seasonal index for historical data that does not have a trend. The method
produces exponentially smoothed values for the level of the forecast and the seasonal
adjustment to the forecast. The seasonal adjustment is added to the forecasted level,
producing the seasonal additive forecast.
This method is best for data without trend but with seasonality that does not increase
over time. It results in a curved forecast that reproduces the seasonal changes in the
data.

Figure B-6 Typical Seasonal Additive Data, Fit, and Forecast Curve without Trend

Seasonal Multiplicative
Calculates a seasonal index for historical data that does not have a trend. The method
produces exponentially smoothed values for the level of the forecast and the seasonal
adjustment to the forecast. The seasonal adjustment is multiplied by the forecasted
level, producing the seasonal multiplicative forecast.
This method is best for data without trend but with seasonality that increases or
decreases over time. It results in a curved forecast that reproduces the seasonal
changes in the data.

Figure B-7 Typical Seasonal Multiplicative Data, Fit, and Forecast Curve without
Trend

B-4 Working with Predictive Planning in Smart View

 Classic Time-series Forecasting

Holt-Winters’ Additive
Is an extension of Holt's exponential smoothing that captures seasonality. This method
produces exponentially smoothed values for the level of the forecast, the trend of the
forecast, and the seasonal adjustment to the forecast. This seasonal additive method
adds the seasonality factor to the trended forecast, producing the Holt-Winters’
additive forecast.
This method is best for data with trend and seasonality that does not increase over
time. It results in a curved forecast that shows the seasonal changes in the data.

Figure B-8 Typical Holt-Winters’ Additive Data, Fit, and Forecast Curve

Holt-Winters’ Multiplicative
Is similar to the Holt-Winters’ additive method. Holt-Winters’ Multiplicative method
also calculates exponentially smoothed values for level, trend, and seasonal
adjustment to the forecast. This seasonal multiplicative method multiplies the trended
forecast by the seasonality, producing the Holt-Winters’ multiplicative forecast.
This method is best for data with trend and with seasonality that increases over time.
It results in a curved forecast that reproduces the seasonal changes in the data.

Figure B-9 Typical Holt-Winters’ Multiplicative Data, Fit, and Forecast Curve

Damped Trend Additive Seasonal Method

Separates a data series into seasonality, damped trend, and level; projects each
forward; and reassembles them into a forecast in an additive manner.
This method is best for data with a trend and with seasonality. It results in a curved
forecast that flattens over time and reproduces the seasonal cycles.

Predictive Planning Forecasting and Statistical Descriptions B-5

Classic Time-series Forecasting

Figure B-10 Typical Damped Trend Additive Data, Fit, and Forecast Curve

Damped Trend Multiplicative Seasonal Method

Separates a data series into seasonality, damped trend, and level; projects each
forward; and reassembles them into a forecast in a multiplicative manner.
This method is best for data with a trend and with seasonality. It results in a curved
forecast that flattens over time and reproduces the seasonal cycles.

Figure B-11 Typical Damped Trend Multiplicative Data, Fit, and Forecast Curve

Classic Seasonal Forecasting Method Parameters

The seasonal forecast methods use the following parameters:

• alpha (α) — Smoothing parameter for the level component of the forecast. The
value of alpha can be any number between 0 and 1, not inclusive.

• beta (β) — Smoothing parameter for the trend component of the forecast. The value
of beta can be any number between 0 and 1, not inclusive.

• gamma (γ) — Smoothing parameter for the seasonality component of the forecast.
The value of gamma can be any number between 0 and 1, not inclusive.

• phi (Φ) — Damping parameter; any number between 0 and 1, not inclusive.
Each seasonal forecasting method uses some or all of these parameters, depending on
the forecasting method. For example, the seasonal additive forecasting method does
not account for trend, so it does not use the beta parameter.
The damped trend methods use phi in addition to the other three.

B-6 Working with Predictive Planning in Smart View

 ARIMA Time-series Forecasting Methods

ARIMA Time-series Forecasting Methods

Autoregressive integrated moving average (ARIMA) forecasting methods were
popularized by G. E. P. Box and G. M. Jenkins in the 1970s. These techniques, often
called the Box-Jenkins forecasting methodology, have the following steps:

1. Model identification and selection

2. Estimation of autoregressive (AR), integration or differencing (I), and moving

average (MA) parameters

3. Model checking
ARIMA is a univariate process. Current values of a data series are correlated with past
values in the same series to produce the AR component, also known as p. Current
values of a random error term are correlated with past values to produce the MA
component, q. Mean and variance values of current and past data are assumed to be
stationary, unchanged over time. If necessary, an I component (symbolized by d) is
added to correct for a lack of stationarity through differencing.
In a nonseasonal ARIMA(p,d,q) model, p indicates the number or order of AR terms, d
indicates the number or order of differences, and q indicates the number or order of
MA terms. The p, d, and q parameters are integers equal to or greater than 0.
Cyclical or seasonal data values are indicated by a seasonal ARIMA model of the
format:
SARIMA(p,d,q)(P,D,Q)(t)
The second group of parameters in parentheses are the seasonal values. Seasonal
ARIMA models consider the number of time periods in a cycle. For a year, the number
of time periods (t) is 12.

Note:

In Predictive Planning charts, tables, and reports, seasonal ARIMA models do

not include the (t) component, although it is still used in calculations.
Predictive Planning ARIMA models do not fit to constant datasets or datasets
that can be transformed to constant datasets by nonseasonal or seasonal
differencing. Because of that feature, all constant series, or series with absolute
regularity such as data representing a straight line or a saw-tooth plot, do not
return an ARIMA model fit.

Time-series Forecasting Error Measures

One component of every time-series forecast is the data’s random error that is not
explained by the forecast formula or by the trend and seasonal patterns. The error is
measured by fitting points for the time periods with historical data and then
comparing the fitted points to the historical data.

RMSE
RMSE (root mean squared error) is an absolute error measure that squares the
deviations to keep the positive and negative deviations from cancelling out one

Predictive Planning Forecasting and Statistical Descriptions B-7

Forecasting Method Selection

another. This measure also tends to exaggerate large errors, which can help eliminate
methods with large errors.

Forecasting Method Selection

All of the nonseasonal forecasting methods and the ARIMA method are run against
the data.
If the data is detected as being seasonal, the seasonal forecasting methods are run
against the data.
The forecasting method with the lowest error measure (for example, RMSE) is used to
forecast the data.

B-8 Working with Predictive Planning in Smart View