# Time Series Lab

Find the Signal in Your Time Series.

**For the Sports Statistics Edition click here.**

v1.50 is the most recent version of Time Series Lab - Dynamic Score Edition. Currently, Windows 64 bit is the only supported platform.

Make sure to read the license agreement before installing the program.

The software manual can be found here.

Time Series Lab should be cited whenever it is used in the following way: Lit, R., S.J. Koopman, and A.C. Harvey (2019-2021), Time Series Lab: https://timeserieslab.com.

# Changelog

v1.50: *2021-01-11*

- Sand graph (contribution for each explanatory variable) added to graphical output
- Spectral density added to graphical output
- Small changes to outline of GUI

v1.40: *2020-11-26*

- Bug fixed that made TSL incompatible with some computers with AMD64 processor architecture

v1.30: *2020-10-10*

- Skellam distribution added to discrete distributions
- Fixed small bug in graphical representation of in and out-of-sample region

v1.20: *2020-06-21:* Download for Windows

- New outline and more plot options on graph page
- TSL manual added to docs folder in TSL install folder
- Fixed bug that occurred when Level and AR processes were combined and Level was the init component
- Right mouse click on text output shows additional options
- Undo and redo added as options for text output on main page
- Exponential link function is now the default for Poisson and Negative Binomial distribution
- Components can now be saved as .csv as well
- Updating of score parameter is renamed from alpha to kappa to be consistent with literature

v1.10: *2020-05-07*

- On the frontpage of the program, users can choose from pre-defined models like ARMA and GARCH models
- Multiple lags of the score can be added to the model which gives the option to exactly replicate ARMA(p,q) and GARCH(p,q) models
- Screenshots of the program can be taken with Ctrl+p (main monitor)
- The Weibull distribution is added to the continuous distributions
- Forecasts can now be saved as .csv as well
- Loss function contributions per time period are saved together with forecasts

v1.03: *2020-03-26*

- The Generalised Gaussian, or General Error, distribution is added to the continuous distributions
- Parameter estimates from former model run can be used as starting values even if the model changed
- Fixed bug in standard error notation
- For exponential link functions, leverage is incorporated in the score function
- Initialisation parameters of seasonal component can be estimated together with other hyper parameters
- Added 1-step-ahead and multi-step-ahead forecasting on "Forecast page"

v1.01 / 1.02

- Fixed bug in Integrated random walk calculation
- Added additional text output in State Information section
- Fixed bug in forecast output column headers
- Added Autocorrelation function of scores to "Graph page"

v1.00

- New outline of "Model setup" and "Advanced settings" menu
- Fixed bug regarding the identification of two AR components
- Fixed inefficiency in likelihood calculation
- Added functionality where the user can select the time series axis from the loaded database
- Right mouse click on graphs for additional options on "Graphics" and "Forecasting" page
- More elaborate description of estimated model on "Main menu" page
- Functionality added to update the program without the need of manually downloading the newest install file

v0.40

- Diagnostic tests added
- Interactive forecasting where losses from several loss functions are reported for user-selected forecast windows
- Addition of probability distributions "Exponential Generalised Beta 2", "Exponential", and "Bernoulli"
- Addition of leverage effect in scale
- Tooltip messages are added to inform the user about functionalities

v0.30

- Addition of second autoregressive component
- ACF and PACF plotting functionality on "Load data" page
- Interactive model description on "Model setup" page
- Significance levels included in parameter report

v0.20

- Graphical capabilities added to the "Load data" page
- Data transformations can be applied to the data
- Addition of explanatory variables for location and scale
- User can select a range (t1-t2) from the time series to estimate