The Time Data Preparation task analyzes the variable assigned to the time ID role to detect the time interval of the data. SAS assumes that all the values in the time ID variable are either date or datetime values and distinguishes between the values by their magnitude. This assumption fails if you have dates that extend beyond July 21, 2196, or datetimes before January 1, 1960.
For many businesses, their time series data is equally spaced, or any two consecutive indices have the same difference between the time intervals. The following table shows an equally spaced time series with a one-year interval.
Number of Sales
If the time interval cannot be detected from the variable that you assign, then you need to specify the interval and season length. For example, the following table shows an unequally spaced time series.
Number of Sales
Often the time interval cannot be detected with transactional data (time-stamped data that is recorded at no particular frequency). If this is the case, the task accumulates the data into observations that correspond to the interval that you specify. For nontransactional data, you might need to specify the interval and season length if there are numerous gaps (missing values) in the data. In this case, the task supplies the missing values. A validation routine checks the values of the time ID to determine whether they are spaced according to the interval that you specified.
The interval determines the frequency of the output. You can modify the time interval. You can change the interval from a higher frequency to a lower frequency or from a lower frequency to a higher frequency. Time intervals are specified in SAS by using character strings. Each of these strings is formed according to a set of rules that enables you to create an almost infinite set of attributes. For each time interval, you can specify the type (such as monthly or weekly), a multiplier, and a shift (the offset for the interval). You can specify a greater time interval than that found in the input data. A smaller interval should not be used, because a small interval generates a large number of observations.
Seasonal cycle length specifies the length of a season. This value is populated automatically if the task can determine the season length from the time ID variable. However, you can specify a season length other than the default if you want to model a cycle in the data. For example, your data might contain a 13-week cycle, so you need to specify a 13-week season length.
The examples in the following table show how the values that you specify for the interval, season length, multiplier, and shift work together.
Interval Name (in SAS Code Format)
YEAR2.7 specifies an interval of every two years. Because the value for the shift is 7, the first month in the year is July.
QTR3.2 specifies three-month intervals starting on April 1, July 1, October 1, and January 1.
First of each
MONTH2.2 specifies February–March, April–May, June–July, August–September, October–November, and December–January of the following year.
(1=Sunday . . .
WEEK6.3 specifies six-week intervals starting on Tuesdays.
DAY3 specifies three-day intervals starting on Sunday.
Here is an explanation of each of the user-supplied values:
is the name of the interval.
specifies the multiplier of the interval. This value can be any positive number. By default, the multiplier is 1. For example, YEAR2 indicates a two-year interval.
specifies the starting point for the interval. By default, this value is one. A value greater than 1 shifts the start to a later point within the interval. The unit for the shift depends on the interval. For example, YEAR.4 specifies a shift of four months, so the year is from April 1 through March 31 of the following year.
Time Series Exploration Task
About the Time Series Exploration Task
Example: Exploring the SASHELP.PRICEDATA Data Set
- In the Tasks section, expand the Forecasting folder and double-click Time Series Exploration. The user interface for the Time Series Exploration task opens.
- On the Data tab, select the SASHELP.PRICEDATAdata set.
- Assign columns to these roles and specify these options:
- To the Dependent variable role, assign the sale variable.
- Expand the Additional Roles heading. To the Time ID role, assign the date variable. From the Interval drop-down list, select Quarter.
- Under the Roles heading, locate the Transformations table. For the sale variable, find the Accumulation drop-down list, and select Sum as the accumulation method.
- Click the Analyses tab, and select these series plots:
- To run the task, click .
Assigning Data to Roles
specifies any explanatory, input, predictor, or causal factor variables. You can assign only numeric variables to this role.
specifies the column that contains the time ID values.
specifies the multiplier for the time interval. By default, the multiplier is 1. This value cannot be negative.
specifies the shift for the time interval. By default, the shift is 1. This value cannot be negative.
specifies the seasonality of the time interval. The default value depends on the time interval.
enables you to specify the seasonality of the data when you do not assign a time ID variable.
lists the variable or variables that you want to use as the classification (BY) variables.
Setting the Analyses Options
You can include these series plots in your results:
You can include these statistics in your results:
Perform autocorrelation analysis
specifies the plots to display in the results. By default, the results show the autocorrelation analysis panel. However, you can select whether to include these plots in the results as well:
specifies the lag values. By default, the number of lags is 0.
Note: To perform a cross-correlation analysis, you must assign a variable to the Independent variables role.
Perform cross-correlation analysis
Note: To perform a decomposition analysis, the seasonal cycle length must be greater than 1.
Perform decomposition analysis
specifies the plots to include in the results. By default, the decomposition panel is included. You can choose to include these plots as well:
Spectral Density Analysis
Spectral density estimate plot
specifies the minimum period to include in the spectral density plot. This value must be an integer greater than or equal to 0 and less than or equal to 32,767.
Adjust the series by its mean prior to the analysis
specifies whether the series should be adjusted by its mean before performing the Fourier decomposition.
specifies how the smoothing function is interpreted. You can choose from these options:
specifies the kernel function to use in the analysis. By default, no kernel function is specified. You can choose from these options:
specifies the exponent for the kernel function.
Perform augmented Dickey-Fuller test
specifies whether to perform an augmented Dickey-Fuller test.
specifies the augmenting order for the Dickey-Fuller test. This value must be an integer greater than or equal to 0 and less than or equal to 1,000.