Time Series - Using

Using Time Series XLMiner™:

In XLMiner™ select Time Series. A list of option appears in the menu.

Partition Data :

Select Time Series --> Partition Data

Time variable : We can select a time variable from the available variables. If we do not select a time variable, XLMiner™ assigns one to the partitioned data.

Variables in the partitioned data: Select one or more variables from the Variables box by clicking on the corresponding selection button.

Specify Partitioning options : We have the convenience of partitioning as per the desired percentage or number of records.

Specify percentages for partitioning : For Automatic option XLMiner™ assumes 60% of the records in Training set and shows those percentages or #records in the box. If we decide to specify the values, we might enter the desired percentage or #records in the training set and XLMiner™ calculates the remaining values.

ARIMA :

Time Variable : We select the time variable here which influences the values of other variables in the series.

Do not fit constant term : Check this if we do not want to have the constant term in the series.

Fit seasonal model : We can check this if we want to specify seasonal model. The seasonal parameters become active only if we check this.

Period : When data is expected to show seasonality, the patterns repeat after a certain period of time. We can check the seasonality by specifying the period. In the data set shown in the dialog above, we enter period to be 12 if we assume that the seasonality period of the series is 12 months. We may change it later depending on the output.

Parameters : We can identify the values of these parameters on the basis of results of the exploratory techniques.

Forecast :

Produce forecasts : XLMiner™ displays the desired number of forecasts if this option is checked. For partitioned data, it displays the forecasts on validation data.

Report confidence intervals for forecasts : If checked, XLMiner™ asks for the required confidence level. (The default is 95%). It then displays the lower and upper values of the range in which the forecast can be with the given confidence interval.

ACF :

Lags : When we specify a number for the lags, XLMiner™ displays the ACF output for all lags between 0 and the specified number.

Plot ACF chart : On checking this, ACF plots for the selected time variable are displayed.

PACF :

The PACF dialog and options are same as the ACF dialog, excepting the lags. In PACF dialog we have to enter the minimum and maximum #lags.

Smoothing :

Exponential smoothing :

Select Time Series --> Smoothing --> Exponential...

Optimize : On checking this option, XLMiner™ works with the basic principles and finds that optimal value of the weight parameter Alpha for which the FMSE is the lowest. If, on working like this the value of Alpha it arrives at is not in the range then it prompts the user that it is unable to optimize for that data.

Give forecast : On checking this XLMiner™ displays the forecast. The number of forecasts is decided by the value you enter in front of #Forecasts.

Double exponential smoothing :

Select Time Series --> Smoothing --> Double Exponential...

As you know, in Double Exponential smoothing we have to enter the value of Beta as well.

Moving average smoothing :

Interval : This number decides how many earlier terms will be involved while calculating the average in the current term.

Holt Winters' Smoothing :

Period : Enter the period ie. the season length.

Update estimate each time : If checked, XLMiner™ updates the estimates before using them to find the next term in the forecast.

The parameters to be entered for the Holt Winter » Additive.. are similar, though the outputs will differ. When the data do not show a particular trend, we can use Holt Winter No Trend..