Glossary

accumulation

either of two processes that are used to convert a time series. (1) Accumulation converts a time series that has no fixed interval into a time series that does have a fixed interval (such as hourly or monthly). (2) Accumulation converts a time series that has a fixed interval into a time series with a lower frequency time interval (such as hourly into daily). Accumulation combines data within the same time interval into a summary value for that time period.

ACF plot

a plot of the autocorrelation coefficients across different values of time lags. This plot enables you to determine whether seasonality exists in the time series. Short form: ACF plot.

aggregation

the process of combining more than one time series to form a single series within the same time interval. For example, data can be combined into a total or average.

aggregation statistic

the mathematical operation used to combine forecasts across levels in the hierarchy. The reconciliation method that you choose determines the levels where the aggregation statistic is used.

autocorrelation

the internal correlation between observations in a time series. Autocorrelation coefficient values range from -1 to +1. When autocorrelation coefficient values are positive, deviations from the mean of one sign tend to be followed by deviations of the same sign. When they are negative, deviations of one sign tend to be followed by deviations of the opposite sign.

autocorrelation function plot

See ACF plot.

bottom-up method of reconciliation

a reconciliation method that uses the forecasts at the lowest level of the hierarchy to adjust forecasts for the higher levels in the hierarchy.

confidence limits

the upper and lower values of a (usually 95%) confidence interval. In repeated sampling, approximately (1-alpha) 100% of the resulting intervals would contain the true value of the parameter that the interval estimates (where alpha is the confidence level associated with the interval).

disaggregation method

a method that specifies how the forecasts in the lower level of the hierarchy are reconciled when the reconciliation method is top-down or middle-out. The disaggregation method can reconcile the forecasts in either of the following ways: (1) by using the proportion that each lower-level forecast contributes to the higher-level forecast; or (2) by splitting equally the difference between the higher-level forecast and the lower-level forecasts.

dummy variable

See indicator variable.

environment

a virtual container of run-time settings for SAS Forecast Server client sessions.

event

an incident that disrupts the normal flow of any process that generates the time series. Examples of events are holidays, retail promotions, and natural disasters.

event repository

a storage location that contains information about calendar events and includes a brief description of each event.

forecast

a numerical prediction of a future value for a time series.

holdout sample

the number of periods of the most recent data that should be excluded from the parameter estimation. The holdout sample can be used to evaluate the forecasting performance of a candidate model.

horizon

the number of intervals into the future, beyond a base date, for which analyses and predictions are made.

IACF plot

a plot of the inverse autocorrelation coefficients across different values of time lags. This plot is useful for detecting over-differencing in the model. Short form: IACF plot.

indicator variable

a numeric variable with a value of either 1 or 0 that is used to indicate whether or not unusual events occur. The variable takes the value of 1 during the event and 0 otherwise.

inverse autocorrelation

the autocorrelation of an autoregressive model remodeled as a moving average model.

inverse autocorrelation function plot

See IACF plot.

ISO year

a year as specified by ISO 8601, and represented by a minimum of four digits for the year (YYYY), followed by digits that represent the week number and the weekday number. For example, 2006-W52-7 (or 2006W527) is the Sunday of the 52nd week of 2006.

leap week

the 53rd week that can occur when an ISO year is measured in weeks. Normally, a year contains 52 weeks of seven days each, for a total of 364 days. However, some years can have an extra one or two days, for a total of 365 or 366 days. Approximately every six years, seven extra days have accumulated, and these extra days are called a leap week.

level shift

a persistent change in the magnitude of a time series curve.

locked override

a user-supplied value for a forecast that is honored when the hierarchy is reconciled. The final forecast value for that level is the value that you specified for the locked override. Locked overrides can generate override conflicts.

middle-out method of reconciliation

a reconciliation method that combines the bottom-up method above the reconciliation level and the top-down method below the reconciliation level. When a hierarchy has more than one middle level, you need to specify which of those levels to use as the reconciliation level. The forecasts at the reconciliation level are used to generate forecasts for both the higher and lower levels.

model selection criterion

the statistic of fit that is used for forecast model selection.

model selection list

a list of candidate model specifications. You can choose which model specification is best suited to forecast a particular time series.

override conflict

a condition that occurs when the value of one locked override is incompatible with the value of another locked override in the same branch of the hierarchy. Override conflicts that are not resolved prior to reconciliation can result in unreconciled nodes.

PACF plot

a plot of the partial autocorrelation coefficients across different values of time lags. This plot is useful for identifying the order of an autoregressive model. Short form: PACF.

partial autocorrelation

the internal correlation between observations in a time series that causes the effect of all intervening lags to be removed.

partial autocorrelation function plot

See PACF plot.

performance statistic

a statistic of fit that is used to determine how well a forecasting process is predicting the future.

project hierarchy

the order of the variables that you have assigned to the BY variables role. An example of a hierarchy is Region > Product Category > Product Line.

pulse

a temporary change in the magnitude of a time series curve. The magnitude returns to the former level immediately after the change.

ramp

a persistent change in the trend or slope of a time series curve.

reconciliation method

the method that specifies the level in the hierarchy where the process of reconciliation starts. The following reconciliation methods are available: bottom-up method, middle-out method, and top-down method.

report

output that is generated by running custom SAS code against the data in your project.

residual

the difference between an observed data value and its predicted value.

seasonal adjustment

a statistical technique designed to even out seasonal swings in time series data in order to analyze nonseasonal trends.

seasonality

a regular change in time series data values that occurs at the same point in each time cycle.

selection criterion

a statistical value that is used to evaluate how well a forecasting model $x+y=z$ performs by comparing the actual data to the predictions.

statistic of fit

a statistical value that is used to evaluate how well a forecasting model fits the historical series by comparing the actual data to the predicted values.

time series data

timestamped data collected over time at a particular frequency. Some examples of time series data are Web visits per hour, sales per month, and calls per day.

top-down method of reconciliation

a reconciliation method that uses the forecasts at the highest level of the hierarchy to adjust the forecasts for the lower levels.

transactional data

timestamped data collected over time at no particular frequency. Some examples of transactional data are point-of-sale data, inventory data, call center data, and trading data.

unlocked override

a user-supplied value for a forecast that acts as a guideline for the final forecast value. The final forecast for the level reflects the value of the unlocked override, but the final forecast and the unlocked override are often not identical. Because these overrides can be overridden when the hierarchy is reconciled, unlocked overrides do not generate override conflicts.

unreconciled node

a node in the hierarchy that cannot be reconciled. A node can be unreconciled if (1) the final forecast of an upper level is not equal to the aggregate of the final forecasts of the lower levels; or (2) a final forecast violates the limits that are implied by a locked override in one of the lower levels in the same branch of the hierarchy.

variable label

up to 256 characters of descriptive text that can be printed in the output by certain procedures instead of, or in addition to, the variable name.

white noise

a series of random fluctuations in the values of a data series. A white noise model has only a mean or constant parameter. A series is described as white noise if its spectral density function is constant.