Time Series Analysis (TimeSID)

41 Time Series Analysis (TimeSID)

41.1 Overview

TimeSID is a component of WinIDAMS for time series analysis. It uses IDAMS datasets as input where the dictionary and data files must have the same name with extensions .dic and .dat respectively.

Only one dataset can be used at a time, i.e. opening of another dataset automatically closes the one being used.

41.2 Preparation of Analysis

Selection of data. Use the menu command File/Open or click the toolbar button Open. Then, in the Open dialogue box, select your file. Setting "Files of type:" to "IDAMS Data File (*.dat)" displays only IDAMS data files.

Selection of series. You are also asked to specify the series (variables) you want to analyse. Numeric variables can be selected from the "Accessible series" list and moved to the "Selected series" area. Moving variables between the lists can be done by clicking the buttons >, < (move only highlighted variables), >>, << (move all variables). Note that alphabetic variables are not available here.

Missing data treatment. Missing data values are excluded from transformations of series; they are also excluded from calculation of statistics and autocorrelations. For the other analysis, missing data values are replaced by the overall mean.

41.3 TimeSID Main Window

After selection of variables and a click on OK, the TimeSID Main window displays the graphic of the first series from the list of selected series. The series can be manipulated and analysed using various options and commands in the menus and/or equivalent toolbar icons.

41.3.1 Menu bar and Toolbar

File

Open: Calls the dialogue box to select a new dataset for analysis.
Close: Closes all windows for the current analysis.
Save As: Calls the dialogue box to save the contents of the active pane/window. Graphical images are saved in Windows Bitmap format (*.bmp). Data table and tables with statistics are saved in text format.
Print: Calls the dialogue box to print the contents of the active pane/window.
Print Preview: Displays a print preview of the contents of the active pane/window.
Print Setup: Calls the dialogue box for modifying printing and printer options.
Exit: Terminates the TimeSID session.

The menu can also contain the list of recently opened files, i.e. files used in previous TimeSID sessions.

Edit

The menu has only one command, Copy, to copy the contents of the active pane/window to the Clipboard.

View

Toolbar: Displays/hides toolbar.
Status Bar: Displays/hides status bar.
OX Scale: Displays/hides OX scale for the time series.
Font for Scales: Calls the dialogue box to select the font for scales.
Basic Colors: Calls the dialogue box to select colours for the margin and background.

Window

Data Table: Calls the window with the data table. Columns of the data table are the analyzed time series (including transformation results).

Besides Data Table, the menu contains the list of opened windows and Windows options for arranging them.

Help

WinIDAMS Manual: Provides access to the WinIDAMS Reference Manual.
About TimeSID: Displays information about the version and copyright of TimeSID and a link for accessing the IDAMS Web page at UNESCO Headquarters.

The two other menus, Transformations and Analysis, are described in details in sections "Transformation of Time Series" and "Analysis of Time Series" below.

Toolbar icons

There are 9 active buttons in the toolbar providing direct access to the same commands/options as the corresponding menu items. They are listed here as they appear from the left to the right.

Open	Histograms, basic statistical characteristics
Copy	Auto-, cross-correlation
Print	Auto-regression
Basic colors	Display information about TimeSID
Font for scales

41.3.2 The Time Series Window

The time series window is divided into 3 panes: the left one is for changing the window properties and for selecting series (variables), the right upper is for displaying several time series and the right lower is for displaying the current series.

Changing the pane appearance. The two panes for displaying time series are synchronized and they can be changed using the controls provided in the left pane. By default, the right upper pane is empty and its size is reduced. The right lower pane displays the current series, keeping scroll bar and scales visible. The size of either pane can be changed using the mouse, and the OX scale can be hidden/displayed using the OX Scale command of the menu View. Moreover, presentation of graphics can be modified as follows:

regulation of graphic compression degree - use the buttons under Compression of OX,
colours for background and margins - use the Colors button or View/Basic Colors command,
font for scales - use the Scale Font button or View/Font for Scales command.

Changing time series name. Select the required time series, click its name with the right mouse button and select the Change name option. The active window presents the name for modification. Note that these modifications are temporary and they are kept only during the current session.

Selecting time series for display. A list of analysed time series is provided in the left pane. By double clicking a variable in the list, you can choose the shape and colour of the line for projection. After OK, the corresponding graphic is displayed in the upper pane. This operation can be repeated for different variables and thus you can get several graphics displayed simultaneously in the upper pane. The right lower pane always displays the current series.

Deleting time series from analysis. Select the required time series, click its name with the right mouse button and select the Delete series option.

41.4 Transformation of Time Series

Time series data can be transformed by calculating differences, smoothing, trend suppression, using a number of functions, etc. The menu Transformations contains commands for creating new time series based on values of selected series. Note that variables displayed for selection are renumbered sequentially starting from zero (0).

Average: creates a new time series as an average of the specified series. Series to be taken for calculation are selected in the dialogue box "Selection of series" (see section "Preparation of Analysis").
Paired arithmetic: creates a set of time series by performing arithmetic operations on pairs of time series specified in the dialogue box (each series specified in the first argument list with the second argument).
Differences, MA, ROC: creates a set of time series based on transformations (sequential differences, uncentered moving average, rate of change) of the series specified in the dialogue box. Parameters specific for each transformation as well as the type of ROC transformation are set in the same dialogue box.

41.5 Analysis of Time Series

Analysis features are activated through commands in the menu Analysis.

Statistics: creates the table with mean, standard deviation, minimum and maximum values as well as the table with statistics for testing the hypothesis "randomness versus trend" for the selected time series. It also displays a histogram for this series.
Auto-, cross-correlations: creates a new window with a set of cells containing graphs of auto- and cross-correlations for the set of specified time series.
Trend (parametric): creates a new time series as the estimation of a parametric trend model for the specified time series. The trend model and the series are selected in a dialogue box.
Autoregression: estimates the parameters of an auto-regression model for short-term prediction for the specified time series.
Spectrum: (spectral analysis) creates a table of spectrum values (frequency, period, density), graph of spectrum estimation, and for DFT spectrum, graph of deviations of the cumulative spectrum from the cumulative "white-noise" spectrum. It can use the fast discrete Fourier transformation (DFT) and/or maximal entropy (MENT) method for the spectrum density estimation. In the DFT procedure, two windows are used to get the improved estimation of spectral density: Welch data window in the time domain and a polynomial smoothing in the frequency domain.
Cross-spectrum: analyses a pair of stationary time series. It provides the values of cross-spectrum power, phase and coherency function as well as their plots. Cross-spectrum is estimated using the Parzen smoothing window.
Frequency filters: procedure decomposes a time series into frequency components. It creates a new series by applying one of the following filters: low frequency, high frequency, band-pass or band-cut. For low or high frequency filter, its frequency bound is equal to the value of the Frequency parameter. For band-pass or band-cut filter, the frequency bounds are determined by the interval (Frequency - Window width, Frequency + Window width) . An option Detrend allows to detrend the time series before filtering (the trend component is added to the filtering results).

References

Farnum, N.R., Stanton, L.W., Quantitative Forecasting Methods , PWS-KENT Publishing Company, Boston, 1989.

Kendall, M.G., Stuart, A., The Advanced Theory of Statistics , Volume 3 - Design and Analysis, and time series, Second edition, Griffin, London, 1968.

Marple Jr, S.L., Digital Spectral Analysis with Applications , Prentice-Hall, Inc., 1987.